27.12.2012 Views

Wave Propagation in Linear Media | re-examined

Wave Propagation in Linear Media | re-examined

Wave Propagation in Linear Media | re-examined

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

DISSERTATION<br />

<strong>Wave</strong> <strong>Propagation</strong> <strong>in</strong> L<strong>in</strong>ear <strong>Media</strong> | <strong>re</strong>-exam<strong>in</strong>ed<br />

ausgefuhrt zum Zwecke der Erlangung des akademischen Grades e<strong>in</strong>es<br />

Doktors der technischen Wissenschaften unter Leitung von<br />

O.Univ.Prof. Dr. Fritz Paschke<br />

Institut fur Allgeme<strong>in</strong>e Elektrotechnik und Elektronik<br />

und<br />

O.Univ.Prof. Dr. Gottfried Magerl<br />

Institut fur Elektrische Me technik<br />

e<strong>in</strong>ge<strong>re</strong>icht an der Technischen Universitat Wien<br />

Fakultat fur Elektrotechnik<br />

von<br />

Thilo Sauter<br />

Matr.-Nr. 8526692<br />

Tanbruckgasse 5/29<br />

A-1120 Wien<br />

Wien, im April 1998 ................................


Thilo Sauter<br />

Technische Universitat Wien<br />

Institut fur Computertechnik<br />

Gu haustra e 27/E384<br />

A-1040 Wien<br />

Tel: +43-1-58801-3985<br />

email: sauter@ict.tuwien.ac.at


Kurzfassung<br />

Seit der Entdeckung des quantenmechanischen Tunnele ektes haben Physiker daruber diskutiert,<br />

wie lange e<strong>in</strong> Partikel braucht, um e<strong>in</strong>e Barrie<strong>re</strong> zu uberw<strong>in</strong>den, und ob diese Geschw<strong>in</strong>digkeit<br />

jene des Lichts ubersteigen kann. Neue Nahrung erhielt die Debatte <strong>in</strong> den letzten<br />

Jah<strong>re</strong>n durch die E<strong>in</strong>beziehung von evaneszenten elektromagnetischen Wellen, die mit tunnelnden<br />

Teilchen formal verwandt und dank der gro e<strong>re</strong>n Zeitkonstanten Messungen leichter<br />

zuganglich s<strong>in</strong>d. Die durchgefuhrten Experimente schienen tatsachlich " superlum<strong>in</strong>ale\ Wellenausb<strong>re</strong>itung<br />

zu bestatigen und losten e<strong>in</strong>e heftige Kontroverse aus. Dabei zeigte sich, da<br />

der Begri der Wellenausb<strong>re</strong>itung ke<strong>in</strong>eswegs klar und e<strong>in</strong>heitlich ist, was e<strong>in</strong>e e<strong>in</strong>gehende<strong>re</strong><br />

Untersuchung s<strong>in</strong>nvoll ersche<strong>in</strong>en la t.<br />

Die vorliegende Arbeit besteht aus zwei Teilen, de<strong>re</strong>n erster <strong>in</strong> e<strong>in</strong>em kurzen historischen<br />

Abri die gangigsten Geschw<strong>in</strong>digkeitsbegri e prasentiert, namlich die Phasen-, Gruppen-,<br />

Signal- und Energiegeschw<strong>in</strong>digkeit. Im Anschlu daran wird <strong>in</strong> e<strong>in</strong>er Reihe von e<strong>in</strong>dimensionalen<br />

Fallstudien das Verhalten von Wellen <strong>in</strong> l<strong>in</strong>ea<strong>re</strong>n und dispersiven Medien untersucht.<br />

Darunter fallen elektromagnetische Ubertragungsleitungen, Hohlleiter und e<strong>in</strong> verlustf<strong>re</strong>ies<br />

Plasma sowie die klassischen Stufen- und Rechteckbarrie<strong>re</strong>n der Quantenmechanik. Dazu werden<br />

jeweils zunachst die entsp<strong>re</strong>chenden Di e<strong>re</strong>ntialgleichungen fur monochromatische Wellen<br />

gelost. Allgeme<strong>in</strong>e<strong>re</strong> Losungen konnen dann mittels Fourier<strong>in</strong>tegralen konstruiert werden. Im<br />

elektromagnetischen Fall zeigt sich, da e<strong>in</strong> Signalsprung sich stets exakt mit Lichtgeschw<strong>in</strong>digkeit<br />

fortbewegt. Allerd<strong>in</strong>gs ist e<strong>in</strong>e derartige Wellenfront immer b<strong>re</strong>itbandig, woh<strong>in</strong>gegen<br />

die Spekt<strong>re</strong>n von Tunnelmoden auf den Be<strong>re</strong>ich unterhalb der G<strong>re</strong>nzf<strong>re</strong>quenz beschrankt se<strong>in</strong><br />

sollten. Diese Forderung <strong>re</strong>sultiert <strong>in</strong> glatten, gaussformigen Signalen, de<strong>re</strong>n Maximum oft<br />

als Ma fur Ausb<strong>re</strong>itungsgeschw<strong>in</strong>digkeit genommen wird. Obwohl diese De nition jener der<br />

Gruppengeschw<strong>in</strong>digkeit entspricht, ist sie im H<strong>in</strong>blick auf die Bestimmung von Tunnelzeiten<br />

problematisch, da die Barrie<strong>re</strong> als Hochpa lter wirkt und die Maxima von e<strong>in</strong>fallender und<br />

transmittierter Welle <strong>in</strong> ke<strong>in</strong>er Beziehung zue<strong>in</strong>ander stehen.<br />

Der zweite Teil der Arbeit beschaftigt sich mit der numerischen Auswertung von Fourier<strong>in</strong>tegralen<br />

im Zusammenhang mit dispersiver Wellenausb<strong>re</strong>itung. Die Quadratur solcher Integrale<br />

wird durch den unbeschrankten Integrationsbe<strong>re</strong>ich und den stark oszillie<strong>re</strong>nden Integranden<br />

sehr muhsam. Obwohl es Losungsansatze fur dieses Problem gab, ist ke<strong>in</strong>e fertige Softwa<strong>re</strong><br />

verfugbar, weshalb e<strong>in</strong>e spezialisierte Quadraturrout<strong>in</strong>e auf Basis des modernen Softwa<strong>re</strong>pakets<br />

Mathematica entwickelt wurde. Dabei wird das Integrations<strong>in</strong>tervall an den Nullstellen<br />

des Integranden zerlegt und e<strong>in</strong>e Partialsummenfolge aufgestellt, de<strong>re</strong>n G<strong>re</strong>nzwert durch<br />

Extrapolation bestimmt wird. Die Funktion der Quadraturrout<strong>in</strong>e wird anhand bekannter<br />

Testprobleme veri ziert, und abschlie end wird beispielhaft ih<strong>re</strong> Anwendung auf das Problem<br />

der Partikelst<strong>re</strong>uung an e<strong>in</strong>em Stufenpotential gezeigt.<br />

III


Abstract<br />

Ever s<strong>in</strong>ce the quantum mechanical tunnel e ect was discove<strong>re</strong>d, physicists have been discuss<strong>in</strong>g<br />

the question of how fast a tunnell<strong>in</strong>g particle actually traverses a barrier and whether<br />

this velocity can exceed that of light. In the <strong>re</strong>cent past, the debate was extended to evanescent<br />

electromagnetic waves which a<strong>re</strong> formally equivalent to tunnell<strong>in</strong>g particles and a<strong>re</strong> mo<strong>re</strong><br />

amenable to measu<strong>re</strong>ments. Experiments we<strong>re</strong> carried out and seemed to prove `superlum<strong>in</strong>al'<br />

wave propagation. The follow<strong>in</strong>g discussion brought to light that the velocity ofwave propagation<br />

is by no means a clearly de ned term | enough to merit some further <strong>in</strong>vestigation.<br />

This work is divided <strong>in</strong>to two parts. In a brief historical survey, the rst part <strong>in</strong>troduces<br />

the most common velocity concepts associated with wave propagation. These a<strong>re</strong> the phase,<br />

group, signal and energy velocities. Subsequently, anumber of one-dimensional case studies<br />

a<strong>re</strong> p<strong>re</strong>sented that <strong>in</strong>vestigate the behaviour of both electromagnetic and quantum waves and<br />

<strong>in</strong> l<strong>in</strong>ear, dispersive media. They comprise electromagnetic transmission l<strong>in</strong>es, wave guides,<br />

and the model of a lossless plasma. The quantum mechanical e ects a<strong>re</strong> exam<strong>in</strong>ed by means<br />

of the classical step and squa<strong>re</strong> potential barriers. The approach adopted is to solve the<br />

<strong>re</strong>spective di e<strong>re</strong>ntial equations for monochromatic waves and to construct Fourier <strong>in</strong>tegrals<br />

characteris<strong>in</strong>g waves with a broader spectrum. In the electromagnetic case, it is seen that a<br />

sudden signal change propagates exactly at the velocity of light. While such awavefront is<br />

always a broad-band signal, the spectra of tunnell<strong>in</strong>g waves should be con ned to the <strong>re</strong>gion<br />

below the cuto f<strong>re</strong>quency. This gives smooth, Gaussian-like functions, the peak of which<br />

is often used to measu<strong>re</strong> the propagation velocity. Although this de nition is similar to the<br />

group velocity, which is applicable for small-band signals, it is problematic when a tunell<strong>in</strong>g<br />

time is to be determ<strong>in</strong>ed. The case studies show that the barrier acts as a high-pass lter<br />

and that the peaks of the <strong>in</strong>cident and transmitted waves a<strong>re</strong> not <strong>re</strong>lated to each other.<br />

The second part of this work is devoted to the numerical evaluation of the Fourier <strong>in</strong>tegrals associated<br />

with dispersive wave propagation. What makes quadratu<strong>re</strong> particularly cumbersome<br />

<strong>in</strong> this case is the fact that the <strong>in</strong>tegration <strong>in</strong>terval is <strong>in</strong> nite and the <strong>in</strong>tegrands oscillate ir<strong>re</strong>gularly.<br />

The<strong>re</strong> have been approaches to this problem, but no <strong>re</strong>ady-to-use computer rout<strong>in</strong>e<br />

is available. Consequently, a custom quadratu<strong>re</strong> rout<strong>in</strong>e is devised us<strong>in</strong>g the modern softwa<strong>re</strong><br />

tool Mathematica. It is based on the partition of the <strong>in</strong>tegrand at its zeros. Quadratu<strong>re</strong> is<br />

carried out over the sub<strong>in</strong>tervals and the <strong>re</strong>sult<strong>in</strong>g sequence of partial sums is extrapolated<br />

to nd its limit. The rout<strong>in</strong>e is tested with benchmark problems and found to give cor<strong>re</strong>ct<br />

answers. F<strong>in</strong>ally, as an exemplary application, the scatter<strong>in</strong>g of quantum waves at a step<br />

potential barrier is p<strong>re</strong>sented and tested.<br />

V


Nullum est iam dictum, quod non sit dictum prius.<br />

Publius Te<strong>re</strong>ntius Afer, Eunuchus<br />

VII


Alles Gescheite ist schon gedacht worden, man mu nur versuchen, es noch<br />

e<strong>in</strong>mal zu denken.<br />

Johann Wolfgang von Goethe, Maximen und Re exionen<br />

VIII


P<strong>re</strong>face<br />

Our popular writers and <strong>re</strong>porters, when they have to deal with physics,<br />

<strong>in</strong>dulge <strong>in</strong> similies of all sorts; the trouble is that they leave the <strong>re</strong>ader<br />

helpless <strong>in</strong> nd<strong>in</strong>g out how far these pungent analogies cover the <strong>re</strong>al issue,<br />

and the<strong>re</strong>fo<strong>re</strong> mo<strong>re</strong> often lead him astray than enlighten him.<br />

Hermann Weyl, The Mathematical Way of Th<strong>in</strong>k<strong>in</strong>g, <strong>re</strong>pr<strong>in</strong>ted <strong>in</strong> [1]<br />

Ra<strong>re</strong>ly does theo<strong>re</strong>tical physics attract the attention of a broad public. Even if news of this<br />

branch ofscience works its way <strong>in</strong>to the papers, it turns <strong>in</strong>to marg<strong>in</strong> notes rather than eyecatch<strong>in</strong>g<br />

headl<strong>in</strong>es. Of course, the<strong>re</strong> a<strong>re</strong> the annual <strong>re</strong>view articles and featu<strong>re</strong>s celebrat<strong>in</strong>g the<br />

Nobel lau<strong>re</strong>ates, and occasionally the<strong>re</strong> a<strong>re</strong> obituaries of prom<strong>in</strong>ent physicists. These <strong>re</strong>ports,<br />

however, always h<strong>in</strong>ge on the personalities themselves, and physics plays only a subord<strong>in</strong>ate<br />

role the<strong>re</strong> | as opposed to the strik<strong>in</strong>g importance physics, as well as eng<strong>in</strong>eer<strong>in</strong>g, undeniably<br />

have <strong>in</strong> modern life. One could blame the underrat<strong>in</strong>g of physics on a general scepticism about<br />

science and technology. While this is certa<strong>in</strong>ly true, it can hardly be the only explanation.<br />

If it was, whence would all the popular articles on biology, cosmology or geophysics come?<br />

The <strong>re</strong>vival of the d<strong>in</strong>osaurs, the pros and cons of genetic manipulation, the <strong>in</strong>tricacies of the<br />

weather, the possibilities of extrater<strong>re</strong>strial life on mars or whe<strong>re</strong>ver, and naturally all shades<br />

of medic<strong>in</strong>e | the list of <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g topics wander<strong>in</strong>g through the media is long and may be<br />

extended arbitrarily. Theo<strong>re</strong>tical physics, though, hardly ever turns up.<br />

Another po<strong>in</strong>t: stories on physics seem to be <strong>re</strong>levant only if the <strong>re</strong>ported achievements have<br />

an impact on technology or | better still | on daily life. A better understand<strong>in</strong>g of quantum<br />

physics assists <strong>in</strong> the development of smaller microelectronic devices that permit the <strong>in</strong>vention<br />

of circuits with ever-<strong>in</strong>c<strong>re</strong>as<strong>in</strong>g complexity, but dec<strong>re</strong>as<strong>in</strong>g power consumption. On the other<br />

end of the scale, nuclear fusion is still believed to be a possible solution for the grow<strong>in</strong>g demand<br />

for electric power. Now, one might argue that these topics do exert an <strong>in</strong> uence on our lives,<br />

and the<strong>re</strong>fo<strong>re</strong> people have the right to be <strong>in</strong>formed about them. This eventually comes down<br />

to divid<strong>in</strong>g science <strong>in</strong> two parts; the knowledge-oriented and the application-oriented. The<br />

former is often <strong>re</strong>garded as science per se, whe<strong>re</strong>as the latter is commonly labelled as (often<br />

with a slightly derogatory undertone) eng<strong>in</strong>eer<strong>in</strong>g. I am not to discuss the gap that separates<br />

the two sides, which, by the way, was chie y dug by the scientists themselves. I just want<br />

to state that what is add<strong>re</strong>ssed to the broad public is mostly the application of physics and<br />

ra<strong>re</strong>ly me<strong>re</strong> knowledge-oriented work. This is by no means so <strong>in</strong> other elds, or is the<strong>re</strong> any<br />

IX


practical use <strong>in</strong> the popular <strong>re</strong>ports on how certa<strong>in</strong> <strong>in</strong>sects handle their <strong>re</strong>production under<br />

evolutional p<strong>re</strong>ssu<strong>re</strong>? But, on the other hand, such articles a<strong>re</strong> easy to understand for a<br />

mildly <strong>in</strong>te<strong>re</strong>sted <strong>re</strong>ader. Ba<strong>re</strong> physics is so perplex<strong>in</strong>g that few journalists da<strong>re</strong> to bother<br />

their <strong>re</strong>aders and listeners with it. Maybe this is the trivial <strong>re</strong>ason: physics just doesn't sell.<br />

In view of these arguments, it is even mo<strong>re</strong> astonish<strong>in</strong>g that with<strong>in</strong> the last two or th<strong>re</strong>e<br />

years, the p<strong>re</strong>dom<strong>in</strong>antly theo<strong>re</strong>tical debate whether or not it is possible to transmit signals<br />

faster than light has <strong>re</strong>ceived quite some publicity. Known orig<strong>in</strong>ally only to a handful of<br />

<strong>re</strong>searchers, the topic somehow was noticed by journalists and began to sp<strong>re</strong>ad <strong>in</strong> newspapers<br />

and magaz<strong>in</strong>es all over the world. 1 Yet the discussion is noth<strong>in</strong>g particularly new | it only<br />

g<strong>re</strong>w <strong>in</strong> vigour dur<strong>in</strong>g the last few years, when experiments we<strong>re</strong> conducted to support earlier<br />

theo<strong>re</strong>tical work. In fact its roots date back to the beg<strong>in</strong>n<strong>in</strong>g of the century, be<strong>in</strong>g closely<br />

l<strong>in</strong>ked to the discovery and understand<strong>in</strong>g of wave propagation, which is a fasc<strong>in</strong>at<strong>in</strong>g subject<br />

<strong>in</strong>deed.<br />

Water waves have been known and studied for a long time, but it is the electromagnetic waves<br />

that literally penetrate all aspects of modern life. They had been p<strong>re</strong>dicted by James Clerk<br />

Maxwell <strong>in</strong> his beautiful, lucid electrodynamic theory. In the 1880's, He<strong>in</strong>rich Hertz set out to<br />

prove their existence, and <strong>in</strong> 1886 he achieved his goal. 2 Hertz was the rst to generate such<br />

waves and transmit them over a distance of several met<strong>re</strong>s. He did, however, never consider<br />

any practical application of his discovery. The <strong>re</strong>cognition of its enormous value was left to<br />

others, like Popow or Ernest Rutherford. 3 But the merit for the <strong>re</strong>al b<strong>re</strong>ak-through goes to<br />

Giuglielmo Marconi, the rst modern radio eng<strong>in</strong>eer. In 1901, he succeeded <strong>in</strong> transmitt<strong>in</strong>g<br />

radio signals across the Atlantic Ocean from Cornwall to Newfoundland and, <strong>in</strong> a way, laid the<br />

foundations of modern telecommunications. It was an ambitious, expensive endeavour with<br />

a fortuitous outcome. Hadn't it been for the dispersive properties of the ionosphe<strong>re</strong>, which<br />

we<strong>re</strong> still unknown at that time, the experiment would have failed, because electromagnetic<br />

waves <strong>in</strong> f<strong>re</strong>e space follow l<strong>in</strong>ear paths, and America cannot be seen from Europe. Still, this<br />

unexpected aid does not belittle Marconi's achievement and its importance for science as well<br />

as eng<strong>in</strong>eer<strong>in</strong>g. 4<br />

Now what was the <strong>re</strong>ason that the faster-than-light-debate worked its way <strong>in</strong>to the massmedia?<br />

It can hardly be the general signi cance of waves <strong>in</strong> modern physics, for this aspect<br />

was scarcely add<strong>re</strong>ssed <strong>in</strong> the <strong>re</strong>ports. The<strong>re</strong> we<strong>re</strong> other details that made the story worth<br />

publish<strong>in</strong>g. Today, it is common knowledge that even a beam of light takes some time to<br />

travel a certa<strong>in</strong> distance. Likewise, a lightyear is well understood as a synonym for an almost<br />

<strong>in</strong>c<strong>re</strong>dible length. Only partly do we owe this awa<strong>re</strong>ness to physics lessons taught <strong>in</strong> school<br />

| science ction series <strong>in</strong> TV and c<strong>in</strong>ema have been much mo<strong>re</strong> e ective <strong>in</strong> <strong>re</strong>ach<strong>in</strong>g (and<br />

teach<strong>in</strong>g) a broad public. Actually, the famous `Star T<strong>re</strong>k' series was often quoted to illustrate<br />

1 New Scientist, 1. April 1995; Frankfurter Allgeme<strong>in</strong>e Zeitung, 7. July 1995; Die ZEIT, 21. July 1995; Bild<br />

der Wissenschaft, August 1997; pro l, 15. September 1997<br />

2 Arm<strong>in</strong> Hermann, Welt<strong>re</strong>ich der Physik, Bechtle Verlag, 1980.<br />

3 Peter Volkmann, Technikpionie<strong>re</strong>, vde-verlag, 1990.<br />

4 The <strong>re</strong>ceiver Marconi used (<strong>in</strong> pr<strong>in</strong>ciple an early semiconductor diode) soon gave rise to a scandal. As<br />

the orig<strong>in</strong> of the device <strong>re</strong>ma<strong>in</strong>ed unclear for a long time, several contemporaries accused Marconi and his<br />

collaborators of <strong>in</strong>tellectual theft. For an account of the a air, see a series of articles <strong>in</strong> the January issue of<br />

the Proceed<strong>in</strong>gs of the IEEE, 1998.<br />

X


the quest for superlum<strong>in</strong>ality and to speculate what life beyond the speed of light could be<br />

like. By contrast, it was E<strong>in</strong>ste<strong>in</strong> who postulated that the velocity of light is the utmost speed<br />

for the transmission of energy or <strong>in</strong>formation, and the ba<strong>re</strong> mention of perhaps the best-known<br />

physicist ever is su cient toevoke <strong>in</strong>te<strong>re</strong>st. Even mo<strong>re</strong> thrill<strong>in</strong>g was the fact that his legacy<br />

seemed to be at stake. His <strong>re</strong>lativistic theory, perhaps the best-known physical theory ever,<br />

seemed to be endange<strong>re</strong>d and eventually falsi ed by experimental evidence. What followed<br />

looked p<strong>re</strong>tty much like a strife between disciples and opponents of E<strong>in</strong>ste<strong>in</strong>. In fact, the<br />

enti<strong>re</strong> discussion was and still is rather circl<strong>in</strong>g around the question how a signal can be<br />

de ned. This was, however, much too <strong>in</strong>volved for many commentators, compa<strong>re</strong>d with the<br />

excit<strong>in</strong>g imag<strong>in</strong>ation of signal transport faster than light. So they often forgot to po<strong>in</strong>t out an<br />

important limitation: superlum<strong>in</strong>ality, as it is cur<strong>re</strong>ntly discussed, is con ned to ext<strong>re</strong>mely<br />

short distances that span but a few centimet<strong>re</strong>s, depend<strong>in</strong>g on the f<strong>re</strong>quency range of the<br />

waves under consideration. This c<strong>re</strong>ates an odd contrast to the notion of comp<strong>re</strong>hensive<br />

unboundedness we a<strong>re</strong> so <strong>in</strong>cl<strong>in</strong>ed to associate with the propagation of light.<br />

The p<strong>re</strong>ced<strong>in</strong>g <strong>re</strong>marks account for the somewhat `scienti cal' <strong>in</strong>g<strong>re</strong>dients of the story. But<br />

the<strong>re</strong> is also a rather cultural aspect. At p<strong>re</strong>sent, the quar<strong>re</strong>l about superlum<strong>in</strong>al wave<br />

propagation is ma<strong>in</strong>ta<strong>in</strong>ed by two parties that happen to work <strong>in</strong> Germany and the United<br />

States, <strong>re</strong>spectively. 5 Consequently, some <strong>re</strong>ports we<strong>re</strong> <strong>in</strong>terspersed with slightly patriotic<br />

(but nonetheless super uous and annoy<strong>in</strong>g) undertones. One could, at times, not escape the<br />

feel<strong>in</strong>g that the writer pa<strong>in</strong>ted the pictu<strong>re</strong> of a sports competition rather than an academic<br />

discussion. So to answer our <strong>in</strong>itial question, I believe itwas the mixtu<strong>re</strong> of popular names,<br />

scienti c rivalry, and the d<strong>re</strong>am of br<strong>in</strong>g<strong>in</strong>g science ction one step closer to <strong>re</strong>ality that made<br />

the subject so fasc<strong>in</strong>at<strong>in</strong>g for journalists and <strong>re</strong>aders alike. Who ca<strong>re</strong>s that accuracy was left<br />

beh<strong>in</strong>d <strong>in</strong> the attempt of mak<strong>in</strong>g th<strong>in</strong>gs comp<strong>re</strong>hensible? After all, physics got a mention<br />

<strong>in</strong> the media, and I was g<strong>re</strong>atly surprised how many people outside the so-called `scienti c<br />

community' took notice of it.<br />

It would be dishonest to deny that this thesis was stimulated by the discussion rag<strong>in</strong>g about<br />

the <strong>in</strong>terp<strong>re</strong>tation of the <strong>re</strong>cent experiments. But by the time the work began, the controversy<br />

still belonged to the ma<strong>in</strong> contenders, and it could not be fo<strong>re</strong>seen that the subject would<br />

sometime merit public attention. The p<strong>re</strong>sent text has several objectives. First, it is meant as<br />

asurvey of the eventful history of wave propagation, its discovery and comp<strong>re</strong>hension, with<br />

a particular emphasis on the superlum<strong>in</strong>ality issue. Furthermo<strong>re</strong>, it provides case studies on<br />

wave propagation phenomena. They a<strong>re</strong> <strong>in</strong>tended as supplements to earlier theo<strong>re</strong>tical work<br />

along these l<strong>in</strong>es | <strong>in</strong> the spirit of what Sir Karl Popper decla<strong>re</strong>d to be the co<strong>re</strong> of a scienti cal<br />

theory: that it must, <strong>in</strong> pr<strong>in</strong>ciple, be subjectable to falsi cation. 6 Now the idea of carry<strong>in</strong>g<br />

out case studies is neither <strong>in</strong>genious nor spectacular, it is as old as science itself. What<br />

has changed <strong>in</strong> the <strong>re</strong>cent past a<strong>re</strong> the ways and means that a<strong>re</strong> at hand to <strong>re</strong>ach the goal.<br />

While <strong>in</strong> former times <strong>re</strong>searchers had to use approximation methods to evaluate complex<br />

exp<strong>re</strong>ssions by hand, the ever-grow<strong>in</strong>g power of computers and their softwa<strong>re</strong> today enables<br />

us to numerically solve problems <strong>in</strong> g<strong>re</strong>ater detail and accuracy. The formulation of the<br />

5 Prof. Nimtz at the University of Cologne (http://www.uni-koeln.de/math-nat-fak/ph2/) and Prof.<br />

Chiao at the University of Berkeley, CA(http://physics1.berkeley.edu/<strong>re</strong>search/chiao/)<br />

6 Karl R. Popper, The Logic of Scienti c Discovery, 1934.<br />

XI


<strong>in</strong>vestigated examples was straightforward <strong>in</strong> terms of classical mathematics. The evaluation<br />

of the <strong>re</strong>sults, conversely, called for numerical computation. This is why a substantial part of<br />

this work is concerned with the application of a modern softwa<strong>re</strong> tool to tackle the questions<br />

aris<strong>in</strong>g from wave propagation, speci cally the computation of Fourier <strong>in</strong>tegrals.<br />

Interdiscipl<strong>in</strong>ary as this work is, many people have contributed to its successful term<strong>in</strong>ation,<br />

whose support I would like to gratefully acknowledge. I am <strong>in</strong>debted fo<strong>re</strong>most and above<br />

all to my adviser, Prof. Fritz Paschke, for the stimulus and superb guidance he provided<br />

dur<strong>in</strong>g the years of close, highly enjoyable co-operation. In fact, he <strong>in</strong>itiated the work <strong>in</strong><br />

the most unspectacular way when he came up to me one ne day and asked me for the<br />

numerical evaluation of a wave <strong>in</strong>tegral. App<strong>re</strong>ciat<strong>in</strong>g the e orts needed to cope with this<br />

task, he proposed to pursue it further and make a doctorate thesis out of it. Although I was<br />

work<strong>in</strong>g on a completely di e<strong>re</strong>nt subject at that time and wave propagation always <strong>re</strong>ma<strong>in</strong>ed<br />

a part-time occupation, I never <strong>re</strong>g<strong>re</strong>tted to have accepted his o er.<br />

The<strong>re</strong> a<strong>re</strong> still others I owe thanks to. My second adviser, Prof. Gottfried Magerl, <strong>re</strong>ad<br />

the text very ca<strong>re</strong>fully and made many suggestions that helped to make the p<strong>re</strong>sentation<br />

clea<strong>re</strong>r. The contribution of Prof. Helmut Rauch was an <strong>in</strong>di<strong>re</strong>ct yet essential one, for it was<br />

him who d<strong>re</strong>w Prof. Paschke's attention to the newly arisen discussion on superlum<strong>in</strong>ality.<br />

Also, he provided us with a very useful collection of literatu<strong>re</strong>. I app<strong>re</strong>ciated the help of<br />

Doz. Christoph Uberhuber very much, who patiently and will<strong>in</strong>gly answe<strong>re</strong>d my <strong>in</strong>sistent<br />

questions on numerics <strong>in</strong> general and quadratu<strong>re</strong> <strong>in</strong> particular. Special thanks go to my<br />

supervisor, Prof. Dietmar Dietrich, who granted me su cient f<strong>re</strong>edom to nish my work even<br />

though the subject was beyond his own eld of <strong>in</strong>te<strong>re</strong>st. I had useful discussions with many of<br />

my colleagues, but I wish to give those with Niki Kero a special mention because they left a<br />

dist<strong>in</strong>ct mark on the appearance and style of the text. I am glad that Steven Gallop underwent<br />

the tedious task of proof-<strong>re</strong>ad<strong>in</strong>g the manuscript, pitilessly cutt<strong>in</strong>g back sentences that had<br />

grown too long and demand<strong>in</strong>g <strong>re</strong>writ<strong>in</strong>g of passages when they we<strong>re</strong> unclear. F<strong>in</strong>ally, I wish<br />

to thank my wife, who bo<strong>re</strong> with me dur<strong>in</strong>g the last, st<strong>re</strong>ssful phase of writ<strong>in</strong>g down this<br />

work, for her patience and encourag<strong>in</strong>g support.<br />

XII


Contents<br />

Part I <strong>Wave</strong> propagation phenomena 1<br />

1 The many velocities of wave propagation 2<br />

1.1 Phase and group velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3<br />

1.2 A few notes on dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7<br />

1.3 Signal velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10<br />

1.4 Energy velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14<br />

1.5 Other velocity de nitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17<br />

2 Signals faster than light? 20<br />

2.1 Superlum<strong>in</strong>al wave propagation . . . . . . . . . . . . . . . . . . . . . . . . . . 21<br />

2.2 Quantum mechanical tunnell<strong>in</strong>g . . . . . . . . . . . . . . . . . . . . . . . . . . 26<br />

3 <strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es 31<br />

3.1 Model of a transmission l<strong>in</strong>e . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32<br />

3.2 Excursion: a delay l<strong>in</strong>e . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35<br />

3.3 Re ection due to term<strong>in</strong>ation mismatch . . . . . . . . . . . . . . . . . . . . . 37<br />

3.4 A simple thought experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . 39<br />

3.5 A dispersive system: the lossless plasma . . . . . . . . . . . . . . . . . . . . . 41<br />

3.6 Inhomogeneous transmission l<strong>in</strong>e . . . . . . . . . . . . . . . . . . . . . . . . . 46<br />

3.7 Turn-on e ects <strong>in</strong> a lossless plasma . . . . . . . . . . . . . . . . . . . . . . . . 48<br />

3.8 Turn-on e ects <strong>in</strong> a wave guide . . . . . . . . . . . . . . . . . . . . . . . . . . 60<br />

3.9 A Gaussian pulse <strong>in</strong> plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64<br />

XIII


4 One-dimensional quantum tunnell<strong>in</strong>g 72<br />

4.1 The potential step . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73<br />

4.2 Initial wave forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76<br />

4.3 Examples of scatter<strong>in</strong>g processes . . . . . . . . . . . . . . . . . . . . . . . . . 81<br />

4.4 The squa<strong>re</strong> barrier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93<br />

4.5 Tunnell<strong>in</strong>g time de nitions for a squa<strong>re</strong> barrier . . . . . . . . . . . . . . . . . 96<br />

4.6 Examples of tunnell<strong>in</strong>g events . . . . . . . . . . . . . . . . . . . . . . . . . . . 100<br />

Interlude <strong>Wave</strong> functions <strong>in</strong> graphical <strong>re</strong>p<strong>re</strong>sentation 119<br />

Part II Numerical aspects of wave equations 123<br />

5 Numerical quadratu<strong>re</strong> and extrapolation 124<br />

5.1 Univariate numerical quadratu<strong>re</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 125<br />

5.1.1 Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125<br />

5.1.2 Computer rout<strong>in</strong>es . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127<br />

5.2 Convergence acceleration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129<br />

6 Towards a quadratu<strong>re</strong> rout<strong>in</strong>e 132<br />

6.1 Partition<strong>in</strong>g the <strong>in</strong>tegration <strong>in</strong>terval . . . . . . . . . . . . . . . . . . . . . . . 133<br />

6.2 Choos<strong>in</strong>g the rst partition po<strong>in</strong>t . . . . . . . . . . . . . . . . . . . . . . . . . 138<br />

6.3 How to compute the rst <strong>in</strong>tegral . . . . . . . . . . . . . . . . . . . . . . . . . 141<br />

6.4 Asymptotic partition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145<br />

6.5 Considerations for a Mathematica implementation . . . . . . . . . . . . . . . 149<br />

6.6 Controll<strong>in</strong>g the accuracy of the extrapolation <strong>re</strong>sult . . . . . . . . . . . . . . . 155<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function 161<br />

7.1 User <strong>in</strong>terface of the function OscInt . . . . . . . . . . . . . . . . . . . . . . . 161<br />

7.2 Structu<strong>re</strong> of the package . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164<br />

7.3 Implementation of OscInt and <strong>re</strong>lated functions . . . . . . . . . . . . . . . . 165<br />

7.3.1 OscInt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166<br />

7.3.2 PartitionTable . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167<br />

7.3.3 PartitionOffs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168<br />

XIV


7.3.4 PartitionPo<strong>in</strong>ts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170<br />

7.3.5 PartInt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171<br />

7.3.6 OscIntControlled . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172<br />

7.4 Auxiliary functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173<br />

7.4.1 Zero computation for quadratic arguments . . . . . . . . . . . . . . . . 174<br />

7.4.2 Zero computation for hyperbolic arguments . . . . . . . . . . . . . . . 174<br />

7.4.3 Approximation error control . . . . . . . . . . . . . . . . . . . . . . . . 178<br />

7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e . . . . . . . . . . . . . . . . . . . . . . . . . . 181<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e 190<br />

8.1 P<strong>re</strong>paration of the wave <strong>in</strong>tegrals for quadratu<strong>re</strong> . . . . . . . . . . . . . . . . 191<br />

8.2 Outl<strong>in</strong>e of the program structu<strong>re</strong> . . . . . . . . . . . . . . . . . . . . . . . . . 197<br />

8.3 Implementation of the quadratu<strong>re</strong> modules . . . . . . . . . . . . . . . . . . . 200<br />

8.4 Test of the package . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205<br />

8.4.1 Numerical tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205<br />

8.4.2 Formal veri cation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211<br />

A Mathematica packages 214<br />

A.1 Numerical quadratu<strong>re</strong> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214<br />

A.2 Solutions for the step potential . . . . . . . . . . . . . . . . . . . . . . . . . . 221<br />

A.3 Solutions for the squa<strong>re</strong> barrier . . . . . . . . . . . . . . . . . . . . . . . . . . 229<br />

A.4 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232<br />

A.5 Utilities for display<strong>in</strong>g po<strong>in</strong>ts <strong>in</strong> 3D . . . . . . . . . . . . . . . . . . . . . . . . 234<br />

Bibliography 237<br />

Index 247<br />

XV


XVI


Part I<br />

<strong>Wave</strong> propagation phenomena<br />

Soon after the formulation of E<strong>in</strong>ste<strong>in</strong>'s theory of <strong>re</strong>lativity, a discussion arose whether or not<br />

certa<strong>in</strong> types of waves can propagate faster than light. These a<strong>re</strong> evanescent electromagnetic<br />

waves and tunnell<strong>in</strong>g quantum particles that a<strong>re</strong> formally equivalent to some electromagnetic<br />

cases. In <strong>re</strong>cent years, the fairly old debate g<strong>re</strong>w <strong>in</strong> <strong>in</strong>tensity when experimental <strong>re</strong>sults<br />

seemed to support `superlum<strong>in</strong>al' theories. The rst part of this work is the<strong>re</strong>fo<strong>re</strong> dedicated<br />

to the many facets of the discussion.<br />

To provide an <strong>in</strong>sight<strong>in</strong>to the subject, the rst chapters give a brief overview on the de nitions<br />

of wave propagation velocities <strong>in</strong> general and on the faster-than-light issue <strong>in</strong> particular. The<br />

variety of op<strong>in</strong>ions and theories is rather unsatisfactory from a practical po<strong>in</strong>t of view, and so<br />

the subsequent chapters p<strong>re</strong>sent anumber of one-dimensional case studies to explo<strong>re</strong> the problems<br />

further. The rst among these a<strong>re</strong> concerned with the propagation of electromagnetic<br />

waves with a special focus on evanescent modes. We shall beg<strong>in</strong> with simple monochromatic<br />

examples whe<strong>re</strong> closed exp<strong>re</strong>ssions for a propagation velocity can still be found. We then<br />

extend the monochromatic <strong>in</strong>vestigation to dispersive media and nally solve the wave equation<br />

for broad-band signals <strong>in</strong> a lossless plasma as well as <strong>in</strong> a <strong>re</strong>ctangular wave guide. These<br />

last examples will demonstrate that sudden signal changes travel at a wave front velocity no<br />

faster than the speed of light.<br />

The second portion of these case studies is dedicated to the classical quantum mechanical<br />

tunnel e ect. Like <strong>in</strong> the sections befo<strong>re</strong>, we start with formulat<strong>in</strong>g the problem for a plane<br />

wave, but immediately extend the <strong>re</strong>sults to the general case of signals with arbitrary bandwidth.<br />

We shall then explo<strong>re</strong> the scatter<strong>in</strong>g of a particle o a potential step. Subsequently,<br />

after hav<strong>in</strong>g looked <strong>in</strong>to quantum mechanical tunnell<strong>in</strong>g time de nitions for plane waves, we<br />

deal with a particle tunnell<strong>in</strong>g through a <strong>re</strong>ctangular barrier. In all these cases, we must<br />

employ numerical techniques to solve the Fourier <strong>in</strong>tegrals that constitute the solutions of the<br />

wave equations. The examples will show that for very small-band pulses, the <strong>re</strong>sults could<br />

be <strong>in</strong>terp<strong>re</strong>ted as superlum<strong>in</strong>al propagation of a wave packet. However, as the bandwidth is<br />

<strong>in</strong>c<strong>re</strong>ased, the tunnell<strong>in</strong>g time becomes even negative, which is quite an absurd <strong>re</strong>sult. So we<br />

eventually arrive at the still open question whether a signal pulse is adequately described by<br />

its peak or cent<strong>re</strong> of mass, or what actually forms the <strong>in</strong>formation be<strong>in</strong>g transmitted.<br />

1


Chapter 1<br />

1 The many velocities of wave propagation<br />

The many velocities of wave<br />

propagation<br />

When a mathematician th<strong>in</strong>ks of wave propagation, he starts by writ<strong>in</strong>g<br />

a well-known second order di e<strong>re</strong>ntial equation and discuss<strong>in</strong>g its peculiar<br />

properties. The physicist is <strong>in</strong>te<strong>re</strong>sted <strong>in</strong> these <strong>re</strong>sults, but he immediately<br />

asks some <strong>in</strong>disc<strong>re</strong>et questions about waves <strong>in</strong> a dispersive medium, when<br />

the velocity of propagation is not a constant, but strongly depends upon<br />

the f<strong>re</strong>quency. Leon Brillou<strong>in</strong> [2]<br />

Ever s<strong>in</strong>ce its discovery, the phenomenon of wave propagation has raised the question of how<br />

fast a wave actually proceeds. The answer, though fairly simple-look<strong>in</strong>g at rst sight, is far<br />

from be<strong>in</strong>g evident and has kept physicists busy for a long time. This cont<strong>in</strong>ued occupation<br />

with the subject is closely <strong>re</strong>lated to the question of the de nition of signals. It is commonly<br />

understood that a signal transfers <strong>in</strong>formation from one part of a medium to another, but this<br />

description is vague and conta<strong>in</strong>s no clueas to how the propagation velocity can be de ned.<br />

Consequently, the<strong>re</strong> we<strong>re</strong> di e<strong>re</strong>nt <strong>in</strong>terp<strong>re</strong>tations, and this lack of a p<strong>re</strong>cise de nition is<br />

mirro<strong>re</strong>d <strong>in</strong> an unexpected variety ofvelocity concepts. They a<strong>re</strong> of course consistent <strong>in</strong> that<br />

they all give the same <strong>re</strong>sult for the special case of a pulse travell<strong>in</strong>g with unalte<strong>re</strong>d shape<br />

through space and time. In the mo<strong>re</strong> <strong>re</strong>alistic cases, however, when the pulse is deformed and<br />

sp<strong>re</strong>ads out <strong>in</strong> the course of time, their <strong>re</strong>sults may di er grossly.<br />

Ow<strong>in</strong>g to the multitude of approaches, this chapter is concerned with an overview of several<br />

concepts of propagation velocity. In particular, we shall concentrate on four concepts that<br />

have become most widely known and a<strong>re</strong> <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g also from a historical po<strong>in</strong>t of view: the<br />

phase and group velocity, the signal velocity, and the energy velocity. The<strong>re</strong> have been other<br />

suggestions, too, which did not <strong>re</strong>ceive somuch attention and a<strong>re</strong> discussed the<strong>re</strong>fo<strong>re</strong> only <strong>in</strong><br />

a brief survey.<br />

2


1.1 Phase and group velocity<br />

1.1 Phase and group velocity<br />

It may seem strange, but the<strong>re</strong> is no undisputed, clear-cut de nition of what constitutes a<br />

wave. In many textbooks, speci cally those dedicated to electromagnetism, they a<strong>re</strong> <strong>in</strong>troduced<br />

as solutions of a partial di e<strong>re</strong>ntial equation called the wave equation,<br />

, 1<br />

v 2<br />

@ 2<br />

=0; (1.1)<br />

@t2 whose solution is easily found to be (r vt). It appears, though, that this de nition is far<br />

too <strong>re</strong>strictive, for many other equations can be thought of that yield similar <strong>re</strong>sults. We<br />

shall the<strong>re</strong>fo<strong>re</strong> adopt the pragmatic approach of Whitham [3] and characterise the wave by<br />

the solution di<strong>re</strong>ctly, ir<strong>re</strong>spective of whatever equation might have led to it. So we follow<br />

the <strong>in</strong>tuitive and familiar notion that a wave is any <strong>re</strong>cognisable function that at di e<strong>re</strong>nt<br />

<strong>in</strong>stances of time can be localised at di e<strong>re</strong>nt positions <strong>in</strong> space, such that the spectator has<br />

the imp<strong>re</strong>ssion of a cont<strong>in</strong>uous movement. This is exp<strong>re</strong>ssed by<br />

= (r , vt) (1.2)<br />

with an arbitrary function , a space coord<strong>in</strong>ate r and some velocity vector v. Although this<br />

is a fairly general formulation of a wave, a special case is of much mo<strong>re</strong> practical <strong>in</strong>te<strong>re</strong>st:<br />

the harmonic wave ( ) = cos( )or () = s<strong>in</strong>( ), which is | thanks to the Fourier transform<br />

| the build<strong>in</strong>g block for any other waveform <strong>in</strong> l<strong>in</strong>ear systems. For the simplest case of a<br />

one-dimensional, monochromatic wave of f<strong>re</strong>quency !, may be written as<br />

(x; t) =Ae j(kx,!t) ; (1.3)<br />

with a constant amplitude A and the wave number k. The imp<strong>re</strong>ssion of wave motion becomes<br />

obvious if we spot one po<strong>in</strong>t <strong>in</strong> the function and determ<strong>in</strong>e its trajectory. This implies that<br />

kx , !t = const : (1.4)<br />

Obviously the po<strong>in</strong>t ismov<strong>in</strong>g <strong>in</strong> positive di<strong>re</strong>ction, and we nd its velocity<br />

vp = dx<br />

dt<br />

= !<br />

k<br />

: (1.5)<br />

S<strong>in</strong>ce the propagation of the c<strong>re</strong>sts of this wave actually describes loci of constant phase, vp<br />

is called phase velocity.<br />

Remark (Notational detail) A signi cant part of this work will be on electomagnetic<br />

waves. Follow<strong>in</strong>g the notational conventions <strong>in</strong> electrodynamics, we shall use the symbol j<br />

for the imag<strong>in</strong>ary unit p ,1, as opposed to i which is common <strong>in</strong> physics and mathematics.<br />

Next we consider the classical example of two superimposed waves that di er slightly <strong>in</strong><br />

f<strong>re</strong>quency and wave number. The two waves a<strong>re</strong><br />

1 = A cos ((k , k)x , (! , !)t)<br />

2 = A cos ((k + k)x,(!+ !)t) :<br />

3<br />

(1.6)


The superposition then yields the well-known <strong>re</strong>sult<br />

1 The many velocities of wave propagation<br />

= 1 + 2 =2Acos( kx , !t) cos(kx , !t) : (1.7)<br />

The <strong>in</strong>terfer<strong>in</strong>g waves form beats or wave packets with the modulation f<strong>re</strong>quency !. The<br />

trajectories of the envelope a<strong>re</strong> now determ<strong>in</strong>ed by<br />

and thus the wave packets propagate at the velocity<br />

kx , !t = const ; (1.8)<br />

vg = !<br />

k<br />

! @!<br />

@k<br />

; (1.9)<br />

which is called group velocity. The carrier, however, moves at the phase velocity like befo<strong>re</strong>.<br />

If the two velocities di er, the p<strong>re</strong>cise shape of the wave will alter as it travels along, because<br />

the oscillations of the carrier seem to move with<strong>in</strong> the wave packets. The orig<strong>in</strong>al shape is<br />

<strong>re</strong>ta<strong>in</strong>ed if and only if group and phase velocity a<strong>re</strong> the same. If we assume that vp is a<br />

function of ! (or, alternatively, k), we can comb<strong>in</strong>e the two de nitions to<br />

vg = vp + k @vp<br />

@k<br />

; (1.10)<br />

which clearly shows that vp must not depend on k <strong>in</strong> order to avoid distortion. This condition<br />

is satis ed for transversal electromagnetic waves <strong>in</strong> f<strong>re</strong>e space, whe<strong>re</strong> both phase and group<br />

velocity a<strong>re</strong> equal to the velocity of light c. In any other medium, vp depends on the f<strong>re</strong>quency,<br />

and the medium is said to be dispersive. The <strong>re</strong>lationship between wave number and f<strong>re</strong>quency<br />

is often displayed <strong>in</strong> a dispersion diagram like the one <strong>in</strong> g. 1.1 .<br />

The dist<strong>in</strong>ction between phase and group velocity has al<strong>re</strong>ady been po<strong>in</strong>ted out by Rayleigh<br />

[4], and he himself admitted that Stokes had t<strong>re</strong>ated this subject <strong>in</strong>dependently befo<strong>re</strong> [5].<br />

Brillou<strong>in</strong> also mentions [2] that the very rst idea of group velocity dates even back to Hamilton,<br />

who published it as early as 1839.<br />

Both a monochromatic s<strong>in</strong>usoid and a periodic series of wave packets a<strong>re</strong> <strong>in</strong>adequate to transport<br />

any k<strong>in</strong>d of <strong>in</strong>formation. An <strong>in</strong>dividual pulse or any other non-periodic function, which<br />

is of far mo<strong>re</strong> practical signi cance, can be described by means of a Fourier <strong>in</strong>tegral<br />

or, alternatively,<br />

(x; t) =<br />

(x; t) =<br />

Z 1<br />

,1<br />

Z 1<br />

,1<br />

A(!) e j(k(!)x,!t) d! (1.11)<br />

A(k) e j(kx,!(k)t) dk : (1.12)<br />

An <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g case is that of a modulated signal, which is composed of an envelope 0 and<br />

a carrier with f<strong>re</strong>quency !c,<br />

(x; t) = 0(x; t) e j(kcx,!ct) : (1.13)<br />

4


1.1 Phase and group velocity<br />

!<br />

!c<br />

vp = !<br />

k<br />

kc<br />

vg = @!<br />

@k<br />

Figu<strong>re</strong> 1.1: Brillou<strong>in</strong> diagram.<br />

!(k)<br />

Us<strong>in</strong>g the Fourier <strong>re</strong>p<strong>re</strong>sentation of the signal, we can write the envelope as<br />

0(x; t) =<br />

Z 1<br />

,1<br />

A(k) e j((k,kc)x,(!(k),!c)t) dk : (1.14)<br />

The dispersion <strong>re</strong>lation !(k) may be expanded <strong>in</strong> a Taylor series about kc, and if terms of<br />

higher than l<strong>in</strong>ear order can be neglected, we nd (!(k) , !c)t = vg(k , kc)t. This nally<br />

yields<br />

0(x; t) =<br />

Z 1<br />

,1<br />

A(k) e j(k,kc)(x,vgt) dk ; (1.15)<br />

which is exactly the <strong>in</strong>itial waveform 0(x; 0), but displaced by an amount vgt. In other<br />

words: the envelope p<strong>re</strong>serves its shape and moves with the group velocity. This <strong>re</strong>sult is<br />

exact if and only if the dispersion <strong>re</strong>lation has the form of a straight l<strong>in</strong>e ! = vgk + !p [2]. In<br />

most cases, however, the dispersion <strong>re</strong>lation will have a mo<strong>re</strong> general form, and (1.15) is then<br />

only a rst-order approximation that applies to a su ciently small <strong>re</strong>gion of the spectrum<br />

cent<strong>re</strong>d at kc and !c, <strong>re</strong>spectively. If the spectrum of the envelope becomes broader, the<br />

group velocity di e<strong>re</strong>nces of the harmonic components will cause the packet to sp<strong>re</strong>ad.<br />

Remark (Modulated versus baseband signals) Note that for modulated signals a<br />

constant group velocity is a p<strong>re</strong><strong>re</strong>quisite so that the pulse shape is not deformed. For<br />

baseband signals, on the other hand, the phase velocity must be <strong>in</strong>variant with<strong>in</strong> the<br />

5<br />

k


1 The many velocities of wave propagation<br />

bandwidth of the signal. We see this by means of a similar consideration as befo<strong>re</strong>. In<br />

the Fourier <strong>re</strong>p<strong>re</strong>sentation of a baseband signal,<br />

(x; t) =<br />

Z 1<br />

A(k) e<br />

,1<br />

j(kx,!(k)t) dk ; (1.16)<br />

we <strong>re</strong>place the dispersion <strong>re</strong>lation aga<strong>in</strong> with ! = vgk + !p, which gives<br />

(x; t) =<br />

Z 1<br />

A(!) e<br />

,1<br />

j(k(x,vgt),!pt) dk : (1.17)<br />

This cor<strong>re</strong>sponds to the shifted <strong>in</strong>itial signal (x; 0) if and only if !p = 0, which <strong>in</strong> turn<br />

means that vg = vp.<br />

The conclusion to be drawn from the p<strong>re</strong>ced<strong>in</strong>g the<strong>re</strong>fo<strong>re</strong> is that the concept of group velocity<br />

is mean<strong>in</strong>gful only for narrow-band signals. Needless to emphasise that a narrow spectrum<br />

implies a large pulse duration, which is undesi<strong>re</strong>d <strong>in</strong> many applications. The notion of wave<br />

packets propagat<strong>in</strong>g at group velocity is not necessarily cor<strong>re</strong>ct <strong>in</strong> the p<strong>re</strong>sence of dispersion<br />

(see, for example, Stratton [6]).<br />

The<strong>re</strong> have also been other <strong>in</strong>terp<strong>re</strong>tations of the group velocity <strong>in</strong> the literatu<strong>re</strong>. Lighthill<br />

[7] mentions a k<strong>in</strong>ematic derivation of vg start<strong>in</strong>g from a cont<strong>in</strong>uity equation for the number<br />

of wave c<strong>re</strong>sts, @k=@t + @!=@x =0. This is a suitable approximation only if the f<strong>re</strong>quency<br />

components have al<strong>re</strong>ady been separated by dispersion. Two other approaches <strong>re</strong>gard the<br />

motion of a cent<strong>re</strong> of gravity. Smith [8] <strong>re</strong>ports a new de nition which takes vg to be the<br />

velocity of the temporal cent<strong>re</strong> of gravity of the wave packet,<br />

R 1<br />

1 ,1 t j (r;t)jdt<br />

= r R 1<br />

: (1.18)<br />

vg ,1 j (r;t)jdt<br />

Consider<strong>in</strong>g the case of a modulated signal (1.13), P<strong>re</strong>stwich [9] stays with the orig<strong>in</strong>al de -<br />

nition of the group velocity, but <strong>re</strong>lates it to the centroid (or spatial cent<strong>re</strong> of gravity) of the<br />

envelope 0(x; t),<br />

He then shows that<br />

hx(t)i =<br />

R 1<br />

,1 x 0(x; t) dx<br />

R 1<br />

: (1.19)<br />

,1 0(x; t) dx<br />

hx(t)i = hx(0)i + @!<br />

t; (1.20)<br />

@k k=k0<br />

which means that the cent<strong>re</strong> of gravity moves at the group velocity. It is of particular <strong>in</strong>te<strong>re</strong>st<br />

that <strong>in</strong> contrast to (1.15), this derivation <strong>in</strong>volves no power expansion of the dispersion<br />

<strong>re</strong>lation.<br />

Vichnevetsky [10] obta<strong>in</strong>ed an alternative derivation of the group velocity by consider<strong>in</strong>g the<br />

cent<strong>re</strong> of gravity of the energy conta<strong>in</strong>ed <strong>in</strong> a wave mov<strong>in</strong>g through a lossless medium,<br />

hx(t)i =<br />

R 1<br />

,1 x j (x; t)j2 dt<br />

R 1<br />

,1 j (x; t)j2 : (1.21)<br />

dt<br />

6


1.2 A few notes on dispersion<br />

He could then demonstrate that the motion of the wave energy can be looked at as the<br />

superposition of the spectral energy components, each mov<strong>in</strong>g at its <strong>re</strong>spective group velocity.<br />

So far, we conside<strong>re</strong>d the media to be energy conservative or non-dissipative, which implies<br />

that the dispersion <strong>re</strong>lation is a <strong>re</strong>al function. When this condition holds, the concept of group<br />

velocity is clearly de ned. As soon as dissipation or evanescence enter the stage, however,<br />

the dispersion <strong>re</strong>lation becomes complex, the imag<strong>in</strong>ary part describ<strong>in</strong>g the attenuation of<br />

the <strong>re</strong>spective monochromatic component. In such a case, the validity of the de nition of<br />

group velocity is at least questionable. For only weak absorption, Brillou<strong>in</strong> [2] argued that<br />

the group velocity may safely be determ<strong>in</strong>ed based on the <strong>re</strong>al part of the wave number,<br />

vg = @ Re !=@k. As absorption becomes mo<strong>re</strong> marked, this approach is no longer justi ed,<br />

and alternatives must be sought.<br />

1.2 A few notes on dispersion<br />

In general, the<strong>re</strong> a<strong>re</strong> two e ects that cause dispersion:<br />

Parameters of the medium may be f<strong>re</strong>quency-dependent. This accounts <strong>in</strong> particular<br />

for <strong>re</strong>sonance e ects <strong>in</strong> dielectrics. In fact the characteristics of all <strong>re</strong>al media depend<br />

on the f<strong>re</strong>quency. The only exception to this rule is vacuum.<br />

Boundary conditions may impose constra<strong>in</strong>ts on the <strong>re</strong>lation between wave number<br />

and f<strong>re</strong>quency even if no medium is p<strong>re</strong>sent atall. Atypical example a<strong>re</strong> hollow wave<br />

guides whe<strong>re</strong> waves can propagate only <strong>in</strong> certa<strong>in</strong> f<strong>re</strong>quency ranges.<br />

Let us rst explo<strong>re</strong> the e ects of f<strong>re</strong>quency-dependent parameters and consider a dielectric<br />

with losses and a dist<strong>in</strong>ct <strong>re</strong>sonance f<strong>re</strong>quency !0 caused by elastically bound polarisable<br />

electrons. The <strong>re</strong>lative dielectric constant of such a medium can be written as [2, 11]<br />

!p 2<br />

"r(!) =1,<br />

! 2 ,!0 2 +2j!<br />

; (1.22)<br />

whe<strong>re</strong> the e ects of the losses a<strong>re</strong> contracted <strong>in</strong>to a phenomenological attenuation constant<br />

. Fig. 1.2 shows the <strong>re</strong>spective curves for the lossy and lossless ( = 0) case. It is important<br />

to notice that without losses, "r is monotonically <strong>in</strong>c<strong>re</strong>as<strong>in</strong>g over the enti<strong>re</strong> range with a pole<br />

at <strong>re</strong>sonance, whe<strong>re</strong>as <strong>in</strong> the p<strong>re</strong>sence of losses, the pole degenerates to a zero at !0 and "r<br />

consequently dec<strong>re</strong>ases <strong>in</strong> the neighbourhood of this po<strong>in</strong>t. For the range whe<strong>re</strong> d"r=d! < 0,<br />

the dispersion is said to be anomalous (Jackson [11]). This is, however, not the only de nition<br />

for this term, as we shall see shortly.<br />

Remark (A word of caution) In the derivation of (1.2), the time dependence of the<br />

physical quantities was taken to be e ,j!t , which is a common assumption <strong>in</strong> physics.<br />

In electrical eng<strong>in</strong>eer<strong>in</strong>g, however, the sign of the phase function is usually <strong>re</strong>versed, so<br />

that the time factor <strong>re</strong>ads e j!t . With this sett<strong>in</strong>g, the denom<strong>in</strong>ator of (1.2) would have<br />

been ! 2 , !0 2 , 2j! . Consequently, if"ras given <strong>in</strong> (1.2) is used <strong>in</strong> standard electrical<br />

7


15<br />

10<br />

5<br />

0<br />

-5<br />

-10<br />

ε<br />

1 The many velocities of wave propagation<br />

2 4 6 8 10 12 ω<br />

Figu<strong>re</strong> 1.2: Dielectric constant "r of a <strong>re</strong>sonant dielectric with (th<strong>in</strong> l<strong>in</strong>es) and without losses. The<br />

solid l<strong>in</strong>es denote the <strong>re</strong>al part, the dotted l<strong>in</strong>e is the imag<strong>in</strong>ary part <strong>in</strong> the lossy case. In this example,<br />

the <strong>re</strong>sonance f<strong>re</strong>quency is at !0 =5.<br />

eng<strong>in</strong>eer<strong>in</strong>g formulas, the sign of the attenuation constant must be taken negative if<br />

is meant tobea loss factor. Otherwise, one might end up for <strong>in</strong>stance with a negative<br />

conductivity.<br />

We now<strong>in</strong>troduce some quantities f<strong>re</strong>quently used to characterise the propagation of electromagnetic<br />

waves <strong>in</strong> dispersive media. The <strong>in</strong>dex of <strong>re</strong>fraction <strong>re</strong>lates the phase velocity <strong>in</strong> the<br />

medium and the velocity of light <strong>in</strong> f<strong>re</strong>e space and is thus determ<strong>in</strong>ed by the <strong>re</strong>lative dielectric<br />

constant and the <strong>re</strong>lative permeability,<br />

n(!) = c<br />

vp<br />

= p "r r; (1.23)<br />

although the permeability is mostly omitted because the medium is chosen to be non-magnetic.<br />

From this equation we nd an alternative exp<strong>re</strong>ssion for the wave number,<br />

k(!) = !<br />

c<br />

which is aga<strong>in</strong> the dispersion <strong>re</strong>lation with ! as <strong>in</strong>dependent variable.<br />

With these sett<strong>in</strong>gs, the group velocity can also be given as<br />

vg(!) = 1<br />

dk<br />

d!<br />

=<br />

n(!); (1.24)<br />

1<br />

n(!)+! dn<br />

d!<br />

: (1.25)<br />

The velocities a<strong>re</strong> now complex, but only <strong>re</strong>al-valued quantities a<strong>re</strong> physically <strong>re</strong>asonable.<br />

Bear<strong>in</strong>g <strong>in</strong> m<strong>in</strong>d that the imag<strong>in</strong>ary part of the phase of a wave, Im(k(!)x , !t) =Imk(!)x<br />

8


1.2 A few notes on dispersion<br />

v/c<br />

6<br />

4<br />

2<br />

-2<br />

2 4 6 8 10 12 ω<br />

Figu<strong>re</strong> 1.3: Index of <strong>re</strong>fraction (solid l<strong>in</strong>e), phase velocity vp=c (dashed l<strong>in</strong>e), and group velocity vg=c<br />

(dotted l<strong>in</strong>e) of the <strong>re</strong>sonant dielectric <strong>in</strong> g. 1.2 .<br />

leads to an attenuation of the wave <strong>in</strong> the di<strong>re</strong>ction of energy propagation, it seems sensible<br />

to <strong>in</strong>sert only the <strong>re</strong>al part of the wave number <strong>in</strong>to the velocity de nitions. We thus obta<strong>in</strong><br />

and<br />

vp = !<br />

Re k<br />

vg = 1<br />

Re dk<br />

d!<br />

= c<br />

Re n<br />

(1.26)<br />

: (1.27)<br />

The <strong>re</strong>sults for the dispersive dielectric, scaled to the vacuum velocity of light, a<strong>re</strong> depicted<br />

<strong>in</strong> g. 1.3 together with the <strong>re</strong>fractive <strong>in</strong>dex. We note that the group velocity exhibits two<br />

a<strong>re</strong>as whe<strong>re</strong> it grows excessively, and becomes negative between the poles.<br />

This behaviour extends far beyond the <strong>re</strong>gion whe<strong>re</strong> anomalous dispersion <strong>in</strong> the above sense<br />

occurs. Consequently, many authors gave alternative de nitions. Accord<strong>in</strong>g to Sommerfeld<br />

[12] and Stratton [6], this is the <strong>re</strong>gion whe<strong>re</strong> vp < vg, which holds only for a narrow<br />

range about the poles. Jackson [11] and Schulz-DuBois [13] <strong>re</strong>qui<strong>re</strong> dn(!)=d! < 0, which<br />

is tantamount to the most common de nition of anomalous dispersion <strong>in</strong> modern literatu<strong>re</strong>,<br />

dvp=d! > 0 (Baldock and Bridgeman [14], Ramo et al. [15], Piefke [16]).<br />

Seen pu<strong>re</strong>ly mathematically, the imag<strong>in</strong>ary part of the wave number gives rise to an attenuation<br />

of the wave along the di<strong>re</strong>ction of propagation. Physically, this e ect can be traced back<br />

to two dist<strong>in</strong>ct causes:<br />

Absorption is associated with various losses <strong>in</strong> the medium like the dielectric <strong>re</strong>sonance<br />

<strong>in</strong> the above example. It is characterised by a complex wave number. Hence wave<br />

propagation is possible, but a part of the wave's energy is dissipated.<br />

9


1 The many velocities of wave propagation<br />

Evanescence, <strong>in</strong> contrast, is non-dissipative and due to total <strong>in</strong>ternal <strong>re</strong> ection. Because<br />

the wave number is pu<strong>re</strong>ly imag<strong>in</strong>ary, a propagation of waves is p<strong>re</strong>cluded. Evanescence<br />

typically stems from boundary conditions that limit wave propagation to a certa<strong>in</strong><br />

f<strong>re</strong>quency range like <strong>in</strong>wave guides.<br />

The term total <strong>in</strong>ternal <strong>re</strong> ection stems from optics whe<strong>re</strong> this e ect was rst observed for<br />

light rays obliquely <strong>in</strong>cident on the <strong>in</strong>terface of two transpa<strong>re</strong>nt media with di e<strong>re</strong>nt <strong>in</strong>dices<br />

of <strong>re</strong>fraction, provided that the angle between the <strong>in</strong>terface and the rays is smaller than a<br />

certa<strong>in</strong> limit. This sort of <strong>re</strong> ection, however, must not be confused with the partial <strong>re</strong> ections<br />

aris<strong>in</strong>g from discont<strong>in</strong>uities or parameter changes <strong>in</strong> the medium. The latter a<strong>re</strong> localised and<br />

take place only at the <strong>re</strong>spective <strong>in</strong>terface, whe<strong>re</strong>as the former is distributed over the enti<strong>re</strong><br />

evanescent <strong>re</strong>gion. This also means that the energy is <strong>re</strong> ected totally only if the evanescent<br />

<strong>re</strong>gion is su ciently | and <strong>in</strong> a strict sense <strong>in</strong> nitely | long (with allusion to optics this<br />

is f<strong>re</strong>quently called the opaque limit). If the evanescent medium, however, has nite length,<br />

then a portion of the <strong>in</strong>cident energy will appear at the far side of the barrier. In optics, this<br />

e ect has the colourful name frustrated total <strong>in</strong>ternal <strong>re</strong> ection [17, 18].<br />

In physical <strong>re</strong>ality, evanescence and absorption cannot always be disentangled. Fiber optic<br />

wave guides, for <strong>in</strong>stance, <strong>re</strong>ly on a cladd<strong>in</strong>g material that is radially evanescent for the<br />

f<strong>re</strong>quencies that make up the signal <strong>in</strong> the co<strong>re</strong>. Still, a certa<strong>in</strong> amount of loss <strong>in</strong> the cladd<strong>in</strong>g<br />

as well as <strong>in</strong> the co<strong>re</strong> is almost <strong>in</strong>evitable. The same is true for mirrors, which also absorb<br />

a small fraction of the <strong>in</strong>cident <strong>in</strong>tensity rather than <strong>re</strong> ect<strong>in</strong>g it. But perhaps the simplest<br />

example is a common LC- lter, whose components a<strong>re</strong> never ideal <strong>re</strong>actances and always<br />

entail losses.<br />

In the course of the <strong>re</strong>view of wave propagation we shall often come across the terms absorption<br />

and evanescence. To be consistent with the <strong>re</strong>spective literatu<strong>re</strong>, they we<strong>re</strong> <strong>re</strong>ta<strong>in</strong>ed as the<br />

cor<strong>re</strong>spond<strong>in</strong>g authors employed them. Nevertheless, it is to be kept <strong>in</strong> m<strong>in</strong>d that many of<br />

the given statements concern<strong>in</strong>g propagation velocity apply to either of the two.<br />

1.3 Signal velocity<br />

Throughout the decades follow<strong>in</strong>g Rayleigh's formulation of group velocity, itwas commonly<br />

believed that this was the actual velocity at which a nite signal propagates through a dispersive<br />

medium. E<strong>in</strong>ste<strong>in</strong>'s theory of <strong>re</strong>lativity, however, eventually raised considerable doubts<br />

concern<strong>in</strong>g this <strong>in</strong>terp<strong>re</strong>tation. The problem was that <strong>in</strong> absorptive media the group velocity<br />

may easily <strong>re</strong>ach an arbitrarily large value. Hence such examples we<strong>re</strong> p<strong>re</strong>sented as<br />

contradictions to the postulate that a signal transmission at a velocity g<strong>re</strong>ater than that of<br />

light <strong>in</strong> vacuum is impossible. This discussion was the motivation for the classical paper of<br />

Sommerfeld [12]. He <strong>in</strong>vestigated the propagation of a step-modulated signal,<br />

(0;t)= (t) s<strong>in</strong> !ct ; (1.28)<br />

(t) be<strong>in</strong>g the Heaviside unit step function. The medium he conside<strong>re</strong>d is described by the<br />

Lo<strong>re</strong>ntz model, which assumes that the medium consists of a large number of oscillat<strong>in</strong>g<br />

10


1.3 Signal velocity<br />

dipoles with a s<strong>in</strong>gle <strong>re</strong>sonance f<strong>re</strong>quency !0 and some damp<strong>in</strong>g coe<br />

<strong>re</strong>lation is thus given by<br />

cient . The dispersion<br />

k =<br />

s ! 2<br />

c 2<br />

!p 2<br />

1 ,<br />

! 2 , !0 2 +2j!<br />

; (1.29)<br />

!p be<strong>in</strong>g the plasma f<strong>re</strong>quency. Sommerfeld then showed that the wave front of any signal<br />

always travels at the velocity of light c.<br />

Proof The proof is based upon function theo<strong>re</strong>tical arguments. S<strong>in</strong>ce the Fourier <strong>in</strong>tegral for the<br />

propagated wave, R 1<br />

d!<br />

ej(kx,!t)<br />

,1 , does not converge if the <strong>in</strong>tegration is carried out along the <strong>re</strong>al<br />

!,!c<br />

axis due to the pole at !c, one can move the contour of <strong>in</strong>tegration to the upper half of the complex<br />

plane, such that ! 7! ! + j . The <strong>in</strong>tegral converges if the argument of the exponential function has<br />

a negative <strong>re</strong>al part at <strong>in</strong> nity. S<strong>in</strong>ce lim!!1 k = !=c, the argument at <strong>in</strong> nity is,j!(t , x=c), and<br />

its <strong>re</strong>al part is negative if and only if t , x=c < 0. The contour of <strong>in</strong>tegration may be closed with<br />

a semicircle to the right at <strong>in</strong> nity, and as this a<strong>re</strong>a conta<strong>in</strong>s no s<strong>in</strong>gularities, the <strong>in</strong>tegral is zero.<br />

Hence the medium is at <strong>re</strong>st whe<strong>re</strong>ver t , x=c < 0, which means that the wave front propagates at<br />

the velocity of light. This statement applies to all media whose dispersion <strong>re</strong>lation is characterised by<br />

kc = !j!!1.<br />

The work of Sommerfeld was complemented by a paper by Brillou<strong>in</strong> [19], who elaborated the<br />

behaviour of the signal. He used the <strong>re</strong>latively new method of saddle po<strong>in</strong>t <strong>in</strong>tegration and<br />

steepest descents to obta<strong>in</strong> an approximate evaluation of the <strong>in</strong>tegral. This method <strong>re</strong>lies on<br />

the fact that a Fourier-type <strong>in</strong>tegral of a complex variable, R e '(!) d!, is dom<strong>in</strong>ated by those<br />

a<strong>re</strong>as of the complex plane whe<strong>re</strong> the path of <strong>in</strong>tegration passes over a saddle po<strong>in</strong>t of'(!).<br />

Thus an approximate evaluation of the <strong>in</strong>tegral may be con ned to the environment of such<br />

po<strong>in</strong>ts, which a<strong>re</strong> crossed follow<strong>in</strong>g the path of steepest descent. In fact, this method is a<br />

generalisation of the method of stationary phase.<br />

Remark (Method of stationary phase) This approximation method was <strong>in</strong>troduced<br />

by Lord Kelv<strong>in</strong>. Consider a Fourier <strong>in</strong>tegral R 1<br />

,1 f(!)ej(k(!)x,!t) d!, whe<strong>re</strong> the spectral<br />

function f(!) is only slowly vary<strong>in</strong>g. If x or t a<strong>re</strong> large, then the ma<strong>in</strong> contribution to<br />

the <strong>in</strong>tegral comes from a<strong>re</strong>as whe<strong>re</strong> the argument of the exponential function does not<br />

change, i. e. k0 (!)x , t =0. Outside the vic<strong>in</strong>ity of such po<strong>in</strong>ts, the <strong>in</strong>tegrand oscillates<br />

strongly and the contribution is practically zero. It is the<strong>re</strong>fo<strong>re</strong> su cient toevaluate the<br />

<strong>in</strong>tegral around the po<strong>in</strong>ts of stationary phase, which maybe accomplished by expansion<br />

techniques (see, for example, Lighthill [7] or Wait [20]).<br />

From a pu<strong>re</strong>ly mathematical po<strong>in</strong>t of view, the evolution of the wave (x; t) is largely determ<strong>in</strong>ed<br />

by the location of the saddle po<strong>in</strong>ts, and the contour of <strong>in</strong>tegration must follow<br />

these po<strong>in</strong>ts as they move through the complex plane. This is what Brillou<strong>in</strong> did, and he<br />

nally found that the saddle po<strong>in</strong>ts | of which the<strong>re</strong> a<strong>re</strong> two pairs due to the structu<strong>re</strong> of the<br />

dispersion <strong>re</strong>lation (1.29) | cause two dist<strong>in</strong>ct so-called p<strong>re</strong>cursors to appear after the arrival<br />

of the wave front. The rst p<strong>re</strong>cursor had al<strong>re</strong>ady been identi ed by Sommerfeld by means of<br />

an asymptotic expansion of the dispersion <strong>re</strong>lation, thus valid only for small values of t , x=c<br />

or high f<strong>re</strong>quencies. The<strong>re</strong>fo<strong>re</strong>, subsequent authors often <strong>re</strong>fer<strong>re</strong>d to the rst p<strong>re</strong>cursor as<br />

Sommerfeld and to the second as Brillou<strong>in</strong> p<strong>re</strong>cursor.<br />

11


Brillou<strong>in</strong> p<strong>re</strong>cursor<br />

Sommerfeld p<strong>re</strong>cursor<br />

x<br />

c<br />

1 The many velocities of wave propagation<br />

signal arrival<br />

Figu<strong>re</strong> 1.4: Evolution of the wave as given by Brillou<strong>in</strong>.<br />

Follow<strong>in</strong>g the p<strong>re</strong>cursors, an oscillation with the f<strong>re</strong>quency !c of the <strong>in</strong>put signal evolves,<br />

and Brillou<strong>in</strong> conside<strong>re</strong>d this to be the signal. As g. 1.4 shows, the po<strong>in</strong>t whe<strong>re</strong> the signal<br />

actually arrives is anyth<strong>in</strong>g but clear to see, so he de ned the signal arrival to occur at the<br />

moment when the path of <strong>in</strong>tegration <strong>re</strong>aches the s<strong>in</strong>gularity at!=!c(the <strong>re</strong>sidue of which<br />

yields an explicit contribution to the <strong>in</strong>tegral afterwards, namely the steady-state solution).<br />

The <strong>in</strong>vestigation of this signal velocity showed that <strong>in</strong> the range of normal dispersion, i. e. far<br />

enough away from the <strong>re</strong>sonance f<strong>re</strong>quency !0, it equals the group velocity. Near the <strong>re</strong>sonance<br />

f<strong>re</strong>quency, however, the group velocity grows larger than c and even becomes negative (see also<br />

g. 1.3), whe<strong>re</strong>as the signal velocity <strong>re</strong>aches, accord<strong>in</strong>g to Brillou<strong>in</strong>, a maximum equal to c.<br />

This was an error aris<strong>in</strong>g from the fact that the approximation method was <strong>in</strong>exact p<strong>re</strong>cisely<br />

at this po<strong>in</strong>t. The mistake was spotted and cor<strong>re</strong>cted by Baerwald [21], who po<strong>in</strong>ted out that<br />

the signal velocity actually <strong>re</strong>aches a m<strong>in</strong>imum near <strong>re</strong>sonance. In a thorough analysis of the<br />

signal evolution, he found that for su ciently large values of x, the amplitude of the p<strong>re</strong>cursor<br />

may exceed that of the signal for the signal arrival time de ned by Brillou<strong>in</strong>. Consequently,<br />

Baerwald proposed that the signal arrival be the moment when the attenuation of p<strong>re</strong>cursor<br />

and stationary signal a<strong>re</strong> equal.<br />

In a <strong>re</strong>lated article [22], Baerwald discussed the general case of cont<strong>in</strong>uous dispersive systems<br />

consist<strong>in</strong>g of an alternat<strong>in</strong>g sequence of stop and pass bands. He con rmed that the wave<br />

front propagates with the velocity<br />

!<br />

vf = lim<br />

!!1 k(!)<br />

t<br />

(1.30)<br />

and <strong>re</strong>ma<strong>in</strong>s undistorted except for some attenuation <strong>in</strong> the p<strong>re</strong>sence of losses. In addition,<br />

he noticed that when the wave has travelled far enough <strong>in</strong>to the medium, it is split up <strong>in</strong>to its<br />

f<strong>re</strong>quency components that then propagate with dist<strong>in</strong>ct velocities. Later, this phenomenon<br />

was dubbed matu<strong>re</strong> dispersion (see, for <strong>in</strong>stance, Baldock and Bridgeman [14]).<br />

12


1.3 Signal velocity<br />

The arbitrar<strong>in</strong>ess that underlies this de nition of the signal velocity has caused considerable<br />

irritation and dispute over the years. The ma<strong>in</strong> objection was the di culty to apply the<br />

de nitions to any practical case. If one waits for the signal to <strong>re</strong>ach a certa<strong>in</strong> fraction of<br />

its maximum amplitude, one must know the enti<strong>re</strong> signal befo<strong>re</strong> be<strong>in</strong>g able to decide when<br />

it has arrived [8], and still one might catch the p<strong>re</strong>cursor <strong>in</strong>stead of the steady-state part.<br />

The other problem is that the amplitude of the steady-state signal may be much less than<br />

that of the transient part, so if one <strong>re</strong>qui<strong>re</strong>d the signal to exceed its p<strong>re</strong>cursor, one would<br />

literally wait fo<strong>re</strong>ver. Apart from these theo<strong>re</strong>tical arguments, Trizna and Weber [23] carried<br />

out a simulation based on an expansion technique and attempted an experimental veri cation<br />

of the signal velocity. To this end, they had to <strong>in</strong>vestigate the squa<strong>re</strong> of the wave (x; t) 2 .<br />

They observed a r<strong>in</strong>g<strong>in</strong>g e ect <strong>in</strong> accordance with the theory, but no pulse delay. Hence they<br />

concluded that a separation of p<strong>re</strong>cursor and steady state signal is not as straightforward as<br />

Brillou<strong>in</strong> suggested, and that the signal velocity de nition is not mean<strong>in</strong>gful <strong>in</strong> practice.<br />

Despite or perhaps just because of these discussions, the wave propagation <strong>in</strong> the Lo<strong>re</strong>ntz<br />

medium was the subject of cont<strong>in</strong>ued <strong>re</strong>search. Mo<strong>re</strong> than seventy years after the papers of<br />

Sommerfeld and Brillou<strong>in</strong>, Oughstun and Sherman [24] published an article that <strong>re</strong>conside<strong>re</strong>d<br />

the classical problem and provided an improved solution. They also used the saddle po<strong>in</strong>t<br />

<strong>in</strong>tegration method, but did not follow the paths of steepest descents, which allowed for a<br />

simpler contour of <strong>in</strong>tegration. In their <strong>in</strong>terp<strong>re</strong>tation of the signal velocity they followed<br />

Baerwald <strong>in</strong> that they de ned the signal arrival to occur when the <strong>re</strong>al part of the phase<br />

function along the contour of <strong>in</strong>tegration equals that of the <strong>re</strong>sidue at ! = !c. Physically, this<br />

means that befo<strong>re</strong> this po<strong>in</strong>t, the wave is dom<strong>in</strong>ated by the contribution of one of the saddle<br />

po<strong>in</strong>ts mov<strong>in</strong>g through the complex plane. After the signal arrival, the eld is dom<strong>in</strong>ated by<br />

the steady state signal, and the p<strong>re</strong>cursors a<strong>re</strong> negligible. The signal arrival is thus exactly<br />

the moment when the attenuation of the stationary oscillation is the same as that of the<br />

dom<strong>in</strong>at<strong>in</strong>g saddle po<strong>in</strong>t at that time | hence the use of the <strong>re</strong>al part of the phase function.<br />

A di e<strong>re</strong>nt <strong>in</strong>terp<strong>re</strong>tation useful for measu<strong>re</strong>ments and numerical experiments is that at this<br />

time, the wave beg<strong>in</strong>s to oscillate with the carrier f<strong>re</strong>quency !c [25, 26].<br />

The <strong>re</strong>sult of this analysis was that the rst or Sommerfeld p<strong>re</strong>cursor oscillates at a high<br />

f<strong>re</strong>quency, whe<strong>re</strong>as the second or Brillou<strong>in</strong> p<strong>re</strong>cursor is low-f<strong>re</strong>quent [27]. Whether they <strong>re</strong>ach<br />

a signi cant amplitude, and which of the two is mo<strong>re</strong> important, depends on the characteristic<br />

of the <strong>in</strong>put signal. In particular, the distortion due to the p<strong>re</strong>cursors becomes mo<strong>re</strong> seve<strong>re</strong> as<br />

the signal f<strong>re</strong>quency is shifted towards the absorption band or the rise and fall times a<strong>re</strong> made<br />

smaller (i. e. the spectrum is broadened). Additionally, for <strong>re</strong>ctangular pulses with su ciently<br />

short pulse width the p<strong>re</strong>cursors of the lead<strong>in</strong>g and trail<strong>in</strong>g edge may <strong>in</strong>terfe<strong>re</strong> [28, 29].<br />

S<strong>in</strong>ce the signal velocity has a similar de nition to that of Baerwald, its dependence on the<br />

signal f<strong>re</strong>quency is similar, too. Consequently, Oughstun also found a m<strong>in</strong>imum near <strong>re</strong>sonance.<br />

At f<strong>re</strong>quencies well above <strong>re</strong>sonance, however, the spac<strong>in</strong>g between the two p<strong>re</strong>cursors<br />

is large enough that the signal seems to b<strong>re</strong>ak up <strong>in</strong>to two parts separated by the Brillou<strong>in</strong><br />

p<strong>re</strong>cursor. The<strong>re</strong>fo<strong>re</strong>, Oughstun de ned two dist<strong>in</strong>ct velocities for these two parts (namely<br />

the p<strong>re</strong>pulse and the actual ma<strong>in</strong> signal), which adds a certa<strong>in</strong> amount of absurdity to the<br />

enti<strong>re</strong> concept. Balictsis and Oughstun extended the <strong>re</strong>search to Gaussian-shape pulses [30]<br />

and found that even such a pulse may evolve <strong>in</strong>to a pair of pulses provided that the <strong>in</strong>itial<br />

13


1 The many velocities of wave propagation<br />

pulse width is small enough (and the bandwidth thus large). At any rate, the signal velocity<br />

is di e<strong>re</strong>nt from the speed of the maximum of the envelope. Extensions to multiple-<strong>re</strong>sonance<br />

media also yielded consistent <strong>re</strong>sults [31].<br />

Remark (Signal velocity for Gaussian pulses) When the carrier f<strong>re</strong>quency is below<br />

the absorption band, such that 0 !c !0, the signal velocity is (approximately) equal<br />

to the energy velocity discussed <strong>in</strong> the next section. A similar criterion holds if !c lies<br />

above the absorption range [29, 30].<br />

The<strong>re</strong> is, at least, one detail whe<strong>re</strong> many authors a<strong>re</strong> of the same op<strong>in</strong>ion: the signal velocity<br />

and the group velocity a<strong>re</strong> practically the same for lossless systems (i. e. = 0) or systems<br />

with small losses such that the dispersion <strong>re</strong>lation has a dom<strong>in</strong>ant <strong>re</strong>al part [22, 23]. Only<br />

when losses become dom<strong>in</strong>ant as <strong>in</strong> the case of <strong>re</strong>sonant absorption, this concept b<strong>re</strong>aks down.<br />

1.4 Energy velocity<br />

While the de nition of the signal velocity sounds very clumsy and far-fetched, the<strong>re</strong> exists<br />

another velocity de nition that can hardly be surpassed <strong>in</strong> clarity and b<strong>re</strong>vity. The energy<br />

velocity is given as the ratio of the mean propagated energy P (which for electromagnetic<br />

waves is the Poynt<strong>in</strong>g vector <strong>in</strong>tegrated over the cross-section) and the mean energy W sto<strong>re</strong>d<br />

<strong>in</strong> the medium per unit length <strong>in</strong> the di<strong>re</strong>ction of propagation,<br />

ve = P<br />

W<br />

; (1.31)<br />

whe<strong>re</strong> the mean values a<strong>re</strong> obta<strong>in</strong>ed by averag<strong>in</strong>g over a period. The total energy density W<br />

<strong>in</strong>cludes not only the electric and magnetic elds, but also the potential and k<strong>in</strong>etic energy<br />

that <strong>re</strong>sides <strong>in</strong> the dipoles of the dielectric.<br />

Remark (Spatial averages) The<strong>re</strong> is also the | at least theo<strong>re</strong>tical | possibility to<br />

calculate the mean energies as averages over a wavelength (Ma<strong>in</strong>ardi [32]). This may be<br />

applicable as long as the wave is not attenuated. If it is, however, the spatial periodicity<br />

is lost, and the <strong>re</strong>spective average as such no longer exists.<br />

The orig<strong>in</strong> of this velocity de nition is somehow unclear. Appa<strong>re</strong>ntly it ente<strong>re</strong>d common<br />

eng<strong>in</strong>eer<strong>in</strong>g knowledge so successfully that hardly anyone ever ca<strong>re</strong>d about giv<strong>in</strong>g <strong>re</strong>fe<strong>re</strong>nces<br />

to whe<strong>re</strong> he got the de nition from. From the few who did, Loudon [33] attributed it to<br />

Brillou<strong>in</strong>, whe<strong>re</strong>as Brillou<strong>in</strong> [2] noted that this de nition was given much earlier by Havelock<br />

[34]. Anyhow, a similar exp<strong>re</strong>ssion appears al<strong>re</strong>ady <strong>in</strong> a paper by Rayleigh [5]. Although<br />

he conside<strong>re</strong>d the ratio of propagated and sto<strong>re</strong>d energy, he did not <strong>re</strong>gard it as a separate<br />

velocity, but rather proved it to be equal to the group velocity for the example of water waves.<br />

In this connection he mentioned Reynolds hav<strong>in</strong>g raised this question, and so it seems that<br />

the concept of energy velocity | like that of group velocity | was conceived mo<strong>re</strong> or less<br />

<strong>in</strong>dependently by either of the two [35].<br />

14


1.4 Energy velocity<br />

For electromagnetic waves, Geppert [36] showed how the velocity of energy transport can be<br />

derived from the cont<strong>in</strong>uity equation, which <strong>in</strong> a general form <strong>re</strong>ads<br />

r ( v)+ @<br />

@t<br />

+ S =0; (1.32)<br />

whe<strong>re</strong> is some physical quantity mov<strong>in</strong>g with mean velocity v and S is the density of sources<br />

(S < 0) and s<strong>in</strong>ks (S > 0) with<strong>in</strong> the medium. For energy conservative systems, this term<br />

vanishes. As second <strong>in</strong>g<strong>re</strong>dient, the Poynt<strong>in</strong>g theo<strong>re</strong>m may be written as<br />

r (E H)+ @<br />

@t<br />

"<br />

2 E2 + 2 H 2 + S =0: (1.33)<br />

In this notation, the cor<strong>re</strong>spondence between the two exp<strong>re</strong>ssions is so strik<strong>in</strong>g that the<br />

de nition of the energy velocity is almost self-evident.<br />

Remark (Energy storage <strong>in</strong> dispersive media) The second term <strong>in</strong> (1.33) is the<br />

energy density <strong>in</strong> the electromagnetic eld. Schulz-DuBois [13] divided this total energy<br />

further <strong>in</strong>to a part associated with the eld itself,<br />

We = "0<br />

2 E2 + 0<br />

2 H 2 ; (1.34)<br />

and a part describ<strong>in</strong>g the energy sto<strong>re</strong>d <strong>in</strong> the deg<strong>re</strong>es of f<strong>re</strong>edom of the dispersive medium,<br />

Ws = 1 @ ,<br />

!(" , "0)E<br />

2 @!<br />

2 + !( , 0)H 2<br />

: (1.35)<br />

The last formula is valid <strong>in</strong> a strict sense only for monochromatic waves, as was al<strong>re</strong>ady<br />

po<strong>in</strong>ted out by Borgnis [37]. It is, however, a useful approximation if the energy spectrum<br />

is narrow.<br />

Although the de nition of the energy velocity looks very pla<strong>in</strong>, the calculation of the energy<br />

density <strong>re</strong>qui<strong>re</strong>s some <strong>re</strong> ection <strong>in</strong> order not to forget a contribution. In fact, the<strong>re</strong> a<strong>re</strong> several<br />

examples of authors who did not take every possibility of energy storage <strong>in</strong>to account and<br />

obta<strong>in</strong>ed surpris<strong>in</strong>g but false <strong>re</strong>sults. One of them was Brillou<strong>in</strong> [2] himself, who for the<br />

Lo<strong>re</strong>ntz medium did not properly consider the energy sto<strong>re</strong>d <strong>in</strong> the elastic dipoles, which<br />

is particularly important at <strong>re</strong>sonance. Loudon [33] cor<strong>re</strong>cted this mistake and <strong>in</strong> addition<br />

found the energy velocity to be <strong>in</strong> good ag<strong>re</strong>ement with the signal velocity de ned by Brillou<strong>in</strong><br />

[2, 19] and Baerwald [22].<br />

A second example is the controversy between Borgnis and Kleen. Borgnis [38] <strong>in</strong>vestigated the<br />

propagation velocity of electromagnetic waves along delay l<strong>in</strong>es. The structu<strong>re</strong> he conside<strong>re</strong>d<br />

consisted of a plane conductor with slots perpendicular to the di<strong>re</strong>ction of propagation. These<br />

slots hold a certa<strong>in</strong> amount of energy that contributes to the total energy density. Borgnis<br />

did not account for this contribution and found that ve was equal to the phase velocity.<br />

Kleen and Poschl [39] discove<strong>re</strong>d the aw and published a cor<strong>re</strong>ct t<strong>re</strong>atment of the problem<br />

that <strong>re</strong>-established the well-known <strong>re</strong>sult ve = vg. Furthermo<strong>re</strong>, they provided a proof that<br />

the energy and group velocities a<strong>re</strong> identical for all lossless transmission l<strong>in</strong>es with periodic<br />

structu<strong>re</strong>.<br />

15


1 The many velocities of wave propagation<br />

The equivalence of group and energy velocity for lossless transmission l<strong>in</strong>es was of particular<br />

importance for the evolution and design of travell<strong>in</strong>g-wave tubes and the<strong>re</strong>fo<strong>re</strong> was widely used<br />

[40, 41]. In these devices, an electron beam should <strong>in</strong>teract with an electromagnetic wave.<br />

To this end, the velocity of the electrons and the phase velocity had to be approximately the<br />

same. Unfortunately, the electromagnetic wave normally travels at the speed of light, whe<strong>re</strong>as<br />

the electrons a<strong>re</strong> much slower than c. Hence delay l<strong>in</strong>es we<strong>re</strong> needed to slow down the wave<br />

and to allow for a coupl<strong>in</strong>g of the electric eld and the beam. But although the design goal<br />

was to match the electron speed and vp of the wave, the energy is actually propagated at vg.<br />

Maybe because of this practical signi cance, the group velocity is often believed to describe<br />

the motion of a wave's energy. An appeal<strong>in</strong>g argument for this identity is that the energy of<br />

the wave is localised <strong>in</strong> the wave packet. S<strong>in</strong>ce the wave packet moves with the <strong>re</strong>spective<br />

group velocity, so does the energy. This is, however, not universally true and <strong>re</strong>stricted to<br />

cases whe<strong>re</strong> dissipation is negligible. Borgnis [42] proved comparatively early the equivalence<br />

ve = vg for monochromatic waves <strong>in</strong> a dielectric with no dispersion other than that imposed<br />

by the geometrical constra<strong>in</strong>ts of a wave propagat<strong>in</strong>g along a conduct<strong>in</strong>g plane, which gives<br />

a dispersion <strong>re</strong>lation of k 2 = ! 2 " , 2 , be<strong>in</strong>g an eigenvalue of the cor<strong>re</strong>spond<strong>in</strong>g system<br />

of partial di e<strong>re</strong>ntial equations. The <strong>in</strong>vestigation of Broer [43] provided a general proof. He<br />

obta<strong>in</strong>ed the equation<br />

@E<br />

@t<br />

+ @<br />

@x (vgE) =0 (1.36)<br />

by means of the method of stationary phase, which imposes two important <strong>re</strong>strictions on the<br />

wave <strong>in</strong>tegrals <strong>in</strong> order to be applicable. First, the dispersion <strong>re</strong>lation must be a <strong>re</strong>al function<br />

| this is whe<strong>re</strong> the absence of losses comes <strong>in</strong>. Second, the waves must have separated<br />

so far that the expansion of the dispersion <strong>re</strong>lation is justi ed and the wave is described<br />

accurately enough by its local wave number. In other words, the spectrum of the wave at this<br />

po<strong>in</strong>t must be narrow. Biot [35] went one step further and showed that the identity holds<br />

also for <strong>in</strong>homogeneous media without dissipation. The parameters of the medium may be<br />

f<strong>re</strong>quency-dependent (which he called anomalous dispersion), but they must not depend on<br />

the coord<strong>in</strong>ate <strong>in</strong> the di<strong>re</strong>ction of propagation.<br />

The convenient equivalence of group velocity and energy propagation velocity is no longer<br />

applicable when the medium becomes absorb<strong>in</strong>g. For the meanwhile well-understood Lo<strong>re</strong>ntz<br />

model, Loudon [33] showed that the classical signal velocity and the energy velocity a<strong>re</strong> <strong>in</strong><br />

excellent ag<strong>re</strong>ement even <strong>in</strong> the a<strong>re</strong>a of <strong>re</strong>sonance. In the discussion of the <strong>re</strong>sults, however, he<br />

conside<strong>re</strong>d only monochromatic waves. An <strong>in</strong>vestigation of signals with broader bandwidths<br />

was carried out by Sherman and Oughstun [44]. They demonstrated that for each po<strong>in</strong>t <strong>in</strong><br />

time and space, the wave function (x; t) is dom<strong>in</strong>ated by a s<strong>in</strong>gle f<strong>re</strong>quency component. This<br />

f<strong>re</strong>quency !e is found to be determ<strong>in</strong>ed by the equation x=t = ve and | if the solution is<br />

not unique | the additional constra<strong>in</strong>t that the attenuation of the partial wave be the least<br />

possible.<br />

Remark (Inter<strong>re</strong>lation between energy and group velocity) The <strong>re</strong>lation x=t = ve<br />

is <strong>in</strong> a certa<strong>in</strong> <strong>re</strong>spect an extension of the method of stationary phase to lossy systems.<br />

16


1.5 Other velocity de nitions<br />

For negligible losses, the Fourier <strong>in</strong>tegral of (x; t) is determ<strong>in</strong>ed by f<strong>re</strong>quency components<br />

whe<strong>re</strong> the derivative ofthe phase function k(!)x , !t vanishes, which yields x=t = vg.<br />

Thus ve takes the role of vg <strong>in</strong> the p<strong>re</strong>sence of losses.<br />

In a subsequent paper, Oughstun and Shen [45] extended the analysis to multiple-<strong>re</strong>sonance<br />

media. In their most <strong>re</strong>cent publication, Sherman and Oughstun [46] provided a model that<br />

allows mo<strong>re</strong> physical <strong>in</strong>sight <strong>in</strong>to the pulse propagation than did the pu<strong>re</strong>ly mathematical<br />

t<strong>re</strong>atment with the saddle po<strong>in</strong>t <strong>in</strong>tegration. It applies if the medium is not too highly<br />

absorb<strong>in</strong>g and if only the <strong>re</strong>gion of matu<strong>re</strong> dispersion is <strong>re</strong>garded. At anygiven time and <strong>in</strong><br />

a small <strong>re</strong>gion of space, the pulse is then p<strong>re</strong>dom<strong>in</strong>antly made up of two components. One<br />

of these is always quasi-monochromatic, the other may either be also quasi-monochromatic<br />

or non-oscillatory. The two components signify the Sommerfeld and Brillou<strong>in</strong> p<strong>re</strong>cursors,<br />

<strong>re</strong>spectively, and move together with the same energy velocity. As the pulse propagates,<br />

these monochromatic components decay accord<strong>in</strong>g to their attenuation coe cients, and <strong>in</strong><br />

addition, the enti<strong>re</strong> pulse sp<strong>re</strong>ads because the partial waves travel at di e<strong>re</strong>nt velocities.<br />

Note that s<strong>in</strong>ce the authors set up the model for a Dirac delta pulse (0;t)= (t), the<strong>re</strong> is no<br />

third contribution from a steady-state solution like the one that appea<strong>re</strong>d <strong>in</strong> the <strong>in</strong>vestigation<br />

of the step-modulated <strong>in</strong>put eld.<br />

The<strong>re</strong> a<strong>re</strong> still other examples whe<strong>re</strong> ve 6= vg. Ma<strong>in</strong>ardi [32] mentions the case of a uniform<br />

transmission l<strong>in</strong>e characterised by f<strong>re</strong>quency-<strong>in</strong>dependent distributed <strong>in</strong>ductance, <strong>re</strong>sistance,<br />

capacitance, and shunt conductance. These parameters form the classical equivalent circuit<br />

of a transmission l<strong>in</strong>e we shall t<strong>re</strong>at <strong>in</strong> mo<strong>re</strong> detail <strong>in</strong> chapter 3.1. Us<strong>in</strong>g this model, it can be<br />

shown that the energy velocity equals the phase velocity. If the transmission l<strong>in</strong>e is assumed<br />

to be loss-f<strong>re</strong>e, the validity of this statement is trivial s<strong>in</strong>ce then the transmission l<strong>in</strong>e is also<br />

dispersion-f<strong>re</strong>e. If, however, losses a<strong>re</strong> taken <strong>in</strong>to account, the wave number becomes complex,<br />

and we obta<strong>in</strong> ve = != Re k = vp <strong>in</strong> accordance with the de nition of (1.26).<br />

Like all other velocity de nitions, the energy velocity was criticised, too. Smith [8] compla<strong>in</strong>ed<br />

that although the de nition is mean<strong>in</strong>gful, it gives no h<strong>in</strong>t astohow one can actually measu<strong>re</strong><br />

the velocity. In particular, unlike the signal velocity, it does not de ne a signal arrival.<br />

Concern<strong>in</strong>g laser phenomena, Schulz-DuBois [13] noted that for emissive dispersive media<br />

the energy velocity <strong>in</strong> the classical de nition can exceed the velocity of light. He the<strong>re</strong>fo<strong>re</strong><br />

proposed an alternative approach that dist<strong>in</strong>guishes the propagation of pu<strong>re</strong>ly electromagnetic<br />

energy from the energy sto<strong>re</strong>d <strong>in</strong> the <strong>re</strong>sonant medium.<br />

1.5 Other velocity de nitions<br />

The afo<strong>re</strong>mentioned th<strong>re</strong>e di e<strong>re</strong>nt velocities of propagat<strong>in</strong>g waves a<strong>re</strong> the oldest and thus<br />

classical ones. As if this had not been enough, a number of additional de nitions emerged <strong>in</strong><br />

the course of time. Many of these approaches conside<strong>re</strong>d <strong>in</strong> one way or another the movement<br />

of a moment of a distribution or, <strong>in</strong> allusion to classical dynamics, the motion of some cent<strong>re</strong><br />

of gravity. We have al<strong>re</strong>ady encounte<strong>re</strong>d one such de nition. It was discussed by Smith [8] as<br />

a proposed substitute for the group velocity and concerned the temporal cent<strong>re</strong> of gravity of<br />

the wave packet's amplitude j (x; t)j. Smith himself suggested still another alternative: the<br />

17


1 The many velocities of wave propagation<br />

centrovelocity. The<strong>re</strong>by he meant the temporal cent<strong>re</strong> of gravity of the <strong>in</strong>tensity (x; t) 2 of<br />

the wave,<br />

1<br />

vc<br />

= r<br />

R 1<br />

,1 t (r;t)2 dt<br />

R 1<br />

,1 (r;t)2 dt<br />

; (1.37)<br />

which is identical to the cent<strong>re</strong> of gravity of the energy carried by the eld. As the motion of<br />

this po<strong>in</strong>t at least partly describes what happens to the energy ow, he advocated the use of<br />

this de nition <strong>in</strong>stead of the classical energy velocity.<br />

The moments of a wave packet, which a<strong>re</strong> undoubtedly useful for characteris<strong>in</strong>g the shape<br />

of a pulse (see, for example, Baird [47]), we<strong>re</strong> also exploited by Anderson and Askne [48] to<br />

derive a pulse velocity. Their start<strong>in</strong>g po<strong>in</strong>t was the well-known l<strong>in</strong>ear approximation of the<br />

dispersion <strong>re</strong>lation, which they augmented by higher order terms <strong>in</strong> order to cope with either<br />

strongly dispersive media or large signal bandwidths. Their de nition of the propagation<br />

velocity is similar to that of Smith, except that they used the spatial distribution of energy<br />

<strong>in</strong> the envelope of the signal to obta<strong>in</strong> the moment of <strong>in</strong>ertia<br />

hx(t)i =<br />

R 1<br />

,1 0 (x; t) x 0(x; t) dx<br />

R 1<br />

,1 0 (x; t) : (1.38)<br />

0(x; t) dx<br />

The pulse velocity is simply the temporal derivative of this moment, v = @hx(t)i=@t. For the<br />

special case of a Gaussian pulse, they demonstrated that the peak of the envelope does not<br />

travel with the group velocity, but with vmax vg( )+v00 g ( )= 2 , whe<strong>re</strong> is the <strong>in</strong>itial spatial<br />

width of the pulse and = x , vgt is a comb<strong>in</strong>ed space-time parameter.<br />

The just mentioned movement ofanenvelope's maximum is a very common measu<strong>re</strong> for the<br />

pulse propagation if the pulse exhibits a pronounced peak. Obviously, the<strong>re</strong> a<strong>re</strong> two di e<strong>re</strong>nt<br />

types of peaks that can be <strong>re</strong>garded. One possibility is to determ<strong>in</strong>e the temporal maximum at<br />

a given distance and followitonitsway<strong>in</strong>to the medium. The alternative is to seek the spatial<br />

maximum for a xed time and analyse its evolution with time. In general, both trajectories<br />

will depend on the parameters of the medium and be neither identical nor simple straight<br />

l<strong>in</strong>es. Anderson et al. [49] discussed both de nitions and concluded that from a practical<br />

po<strong>in</strong>t of view, the temporal velocity is mo<strong>re</strong> <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g because it is easier to measu<strong>re</strong> <strong>in</strong> an<br />

experiment.<br />

In a way <strong>re</strong>lated to the observation of pulse peaks is the cor<strong>re</strong>lation velocity described by Bloch<br />

[50]. It is based on comput<strong>in</strong>g the cross-cor<strong>re</strong>lation of the <strong>re</strong>ceived pulse and its undistorted<br />

shape<br />

R( )= 1<br />

T<br />

Z T<br />

0<br />

(0;t) (x; t + )dt ; (1.39)<br />

whe<strong>re</strong> the wave packet is assumed to be periodic with period T , such that the computation can<br />

be accomplished by e cient techniques like the Fast Fourier Transform (FFT). The cor<strong>re</strong>lation<br />

function will have a peak at the time delay whe<strong>re</strong> the distorted wave packet <strong>re</strong>sembles most<br />

closely its <strong>in</strong>itial shape, and this delay is then used for the de nition of the velocity. It is<br />

18


1.5 Other velocity de nitions<br />

evident that this procedu<strong>re</strong> is <strong>in</strong> fact noth<strong>in</strong>g but a lter<strong>in</strong>g of the pulse, apt to make it<br />

smoother and <strong>re</strong>duce the <strong>in</strong> uence of noise. Unfortunately, it fails to give a unique <strong>re</strong>sult<br />

whenever the pulse is split <strong>in</strong>to several parts on its passage through the medium.<br />

Anyhow, it might be <strong>in</strong>deed mo<strong>re</strong> mean<strong>in</strong>gful <strong>in</strong> such a case to t<strong>re</strong>at the dist<strong>in</strong>ct parts separately<br />

than to construe an arti cial cent<strong>re</strong> of gravity, which Bloch compa<strong>re</strong>d drastically to<br />

the fact that, `if the earth we<strong>re</strong> blown to pieces by a thermonuclear device, the center of mass<br />

would still se<strong>re</strong>nely orbit the sun'. On the other hand, is the<strong>re</strong> any good <strong>re</strong>ason for t<strong>re</strong>at<strong>in</strong>g<br />

a pulse di e<strong>re</strong>ntly only because it is <strong>in</strong>cidentally torn apart by dispersion? After all, it is still<br />

one s<strong>in</strong>gle signal. These nal thoughts <strong>re</strong> ect, <strong>in</strong> essence, the natu<strong>re</strong> of the enti<strong>re</strong> discussion<br />

on wave propagation velocity.<br />

19


Chapter 2<br />

Signals faster than light?<br />

2 Signals faster than light?<br />

Wie wird es se<strong>in</strong>, wenn wir mit der Schnelligkeit des Blitzes Nachrichten<br />

uber die ganze Erde werden verb<strong>re</strong>iten konnen, wenn wir selber mit gro er<br />

Geschw<strong>in</strong>digkeit und <strong>in</strong> kurzer Zeit an die verschiedensten Stellen der Erde<br />

werden gelangen, und wenn wir mit gleicher Schnelligkeit gro e Lasten werden<br />

befordern konnen? Werden die Guter der Erde da nicht durch die<br />

Moglichkeit des leichten Austausches geme<strong>in</strong>sam werden, da allen alles<br />

zuganglich ist? Adalbert Stifter, Der Nachsommer, 1857 [51]<br />

One of the fundamental tenets of modern physics is the pr<strong>in</strong>ciple of <strong>re</strong>lativistic causality,<br />

the postulate that no energy may be transmitted at a speed g<strong>re</strong>ater than that of light <strong>in</strong><br />

f<strong>re</strong>e space. Another puzzl<strong>in</strong>g phenomenon, almost as old as E<strong>in</strong>ste<strong>in</strong>'s discovery and closely<br />

l<strong>in</strong>ked to wave propagation, is the tunnel e ect. It permits quantum particles to penetrate<br />

a barrier too high to overcome with<strong>in</strong> the scope of classical mechanics, and electromagnetic<br />

waves to cross <strong>re</strong>gions whe<strong>re</strong> they normally cannot exist. This e ect has been given much<br />

attention, and from the very beg<strong>in</strong>n<strong>in</strong>g, a central question was how much time a particle<br />

or pulse spends <strong>in</strong> the barrier, or how fast it traverses the tunnel. Many answers have<br />

been attempted. Particular <strong>in</strong> <strong>re</strong>cent years, theo<strong>re</strong>tical <strong>in</strong>vestigations as well as experimental<br />

evidence seemed to suggest that the tunnell<strong>in</strong>g process takes place <strong>in</strong> almost no time, and<br />

that the wave actually propagates faster than light if only the barrier is su ciently thick.<br />

This chapter shall illustrate that the discussion on superlum<strong>in</strong>al wave propagation <strong>re</strong>sembles<br />

closely the long-last<strong>in</strong>g debate on the velocity of wave propagation. In fact, the arguments<br />

a<strong>re</strong> <strong>in</strong> many <strong>re</strong>spects identical, and one cannot escape a feel<strong>in</strong>g of deja vu. We shall start with<br />

a <strong>re</strong>view of superlum<strong>in</strong>al aspects of electromagnetic waves, both with <strong>re</strong>spect to microwave<br />

propagation <strong>in</strong> waveguides and the mo<strong>re</strong> <strong>re</strong>cent eld of quantum optics, whe<strong>re</strong> the behaviour<br />

of <strong>in</strong>dividual photons is of importance. The second part is devoted to the `classical' quantum<br />

mechanical tunnel e ect and various approaches to describe the cor<strong>re</strong>spond<strong>in</strong>g traversal time.<br />

20


2.1 Superlum<strong>in</strong>al wave propagation<br />

2.1 Superlum<strong>in</strong>al wave propagation<br />

Causality states that the propagation of signals and energy is upper bounded by c, the velocity<br />

of light <strong>in</strong>vacuum. This does, of course, not a ect the phase velocity ofawave, and numerous<br />

thought experiments have been <strong>in</strong>vented to illustrate how vp can exceed c (perhaps the most<br />

lucid example is that of a rotat<strong>in</strong>g light source be<strong>in</strong>g observed <strong>in</strong> a su ciently large distance).<br />

Some of them have been described <strong>in</strong> an <strong>in</strong>troductory article by Rothman [52]. He <strong>re</strong>solved<br />

the puzzles <strong>in</strong> the usual way by po<strong>in</strong>t<strong>in</strong>g out the di e<strong>re</strong>nce between phase and group velocity.<br />

Furthermo<strong>re</strong>, he mentioned that <strong>in</strong> electric wave guides, the group velocity is always smaller<br />

than c due to the <strong>re</strong>lation vp vg = c 2 . This is, however, not universally true and applies only<br />

to waves governed by very simple boundary conditions <strong>in</strong> lossless media. We have al<strong>re</strong>ady<br />

mentioned that the group velocity may grow beyond the velocity of light <strong>in</strong> an absorptive<br />

medium. Such examples eventually led to Sommerfeld's proof that a wave front always travels<br />

at c, and to the formulation of the signal velocity.<br />

Remark (Phase and group velocity) We can most easily derive a criterion necessary<br />

for the <strong>re</strong>lation vp vg = c 2 to hold. If we <strong>re</strong>place vp and vg with their de nitions from the<br />

p<strong>re</strong>vious chapter and separate the variables, we obta<strong>in</strong><br />

Integration of both sides gives<br />

!d!=c 2 kdk : (2.1)<br />

! 2 = c 2 k 2 + C; (2.2)<br />

with some <strong>in</strong>tegration constant C that must be determ<strong>in</strong>ed from the boundary conditions.<br />

The dispersion <strong>re</strong>lation the<strong>re</strong>fo<strong>re</strong> is of the form ! = c p k 2 + C, which is true for hollow<br />

wave guides and a lossless plasma. The exp<strong>re</strong>ssion vp vg = c 2 is not valid, on the other<br />

hand, <strong>in</strong> delay l<strong>in</strong>es, as we shall see <strong>in</strong> the follow<strong>in</strong>g chapter. In an article by Borgnis [42],<br />

this <strong>re</strong>lation appears as vp ve = c 2 , which would be equivalent whenever the velocity of<br />

energy transport equals the group velocity. In delay l<strong>in</strong>es, the group and energy velocities<br />

a<strong>re</strong> <strong>in</strong>deed identical, and hence the above <strong>re</strong>lation is false <strong>in</strong> these cases.<br />

For some decades, all <strong>re</strong>ma<strong>in</strong>ed quiet. With the <strong>in</strong>vention of lasers and thus emissive media,<br />

experiments we<strong>re</strong> carried out, the <strong>re</strong>sults of which allowed <strong>in</strong>terp<strong>re</strong>tations of `superlum<strong>in</strong>al'<br />

pulse velocities. Anderson et al. [49] <strong>re</strong>fer to such publications. As they we<strong>re</strong> consider<strong>in</strong>g the<br />

motion of the maximum of a pulse, they also noted that this velocity can exceed c for certa<strong>in</strong><br />

parameter values <strong>in</strong> an atomic medium. At the same time, however, they conceded that <strong>in</strong><br />

this case (when the peak moves faster than the pulse front) the underly<strong>in</strong>g assumption of<br />

their analysis | i. e. that the amplitude of the pulse envelope is only slowly vary<strong>in</strong>g | will<br />

eventually be violated.<br />

Only <strong>re</strong>cently, Enders and Nimtz [53, 54] carried out microwave experiments to study an<br />

analogy of the quantum mechanical tunnel e ect | the propagation of evanescent electromagnetic<br />

waves <strong>in</strong> a wave guide. It is well known that the basic mode <strong>in</strong> a <strong>re</strong>ctangular wave<br />

guide is governed by the dispersion <strong>re</strong>lation<br />

k = 1<br />

c<br />

p ! 2 , !c 2 ; (2.3)<br />

21


d<br />

a1 a2 a1<br />

!c1 = c<br />

a1<br />

!c2 = c<br />

a2<br />

2 Signals faster than light?<br />

Figu<strong>re</strong> 2.1: <strong>Wave</strong> guide used <strong>in</strong> the Enders-Nimtz experiment. The cent<strong>re</strong> f<strong>re</strong>quency of the pulse is<br />

chosen to satisfy !c1


2.1 Superlum<strong>in</strong>al wave propagation<br />

a<br />

!c1=<br />

"r1; r1<br />

c<br />

a p "r1 r1<br />

d<br />

"r2; r2<br />

!c2 =<br />

c<br />

a p "r2 r2<br />

"r1; r1<br />

Figu<strong>re</strong> 2.2: <strong>Wave</strong> guide discussed by Mart<strong>in</strong> and Landauer. The dielectrics and the cent<strong>re</strong> f<strong>re</strong>quency<br />

of the pulse must satisfy !c1 d) is proportional to the <strong>in</strong>put pulse<br />

(x; t) / (x , d , tc;0) ; (2.6)<br />

if the pulse was <strong>in</strong>itially cent<strong>re</strong>d about x = 0. This means that the pulse seems to arrive<br />

beh<strong>in</strong>d the barrier at a time t (x , d)=c, without spend<strong>in</strong>g any time <strong>in</strong> the barrier itself.<br />

The derivation was based on a causal theory. However they appea<strong>re</strong>d to be uncerta<strong>in</strong> about<br />

the <strong>in</strong>terp<strong>re</strong>tation and called the <strong>re</strong>sult `an appa<strong>re</strong>nt violation of causality'.<br />

Soon afterwards, Hass and Busch [63] came up with a discussion of the same model. They<br />

focused on the e ect of an <strong>in</strong>c<strong>re</strong>as<strong>in</strong>g barrier thickness by consider<strong>in</strong>g only the di e<strong>re</strong>nce<br />

23


2 Signals faster than light?<br />

between two tunnell<strong>in</strong>g experiments with di e<strong>re</strong>nt barrier lengths. With a rst-order approximation<br />

of the dispersion <strong>re</strong>lation (2.3) and for narrow-band Gaussian pulses they could show<br />

that the pulse <strong>in</strong>deed needs no extra time for the additional barrier length. Apart from this<br />

veri cation of the measu<strong>re</strong>ments, they also found that any wave front does travel, at most,<br />

with the speed of light. Consider<strong>in</strong>g the analogous quantum mechanical tunnell<strong>in</strong>g problem,<br />

Azbel' [64] obta<strong>in</strong>ed the same <strong>re</strong>sult. He exam<strong>in</strong>ed the superposition of smooth wave functions<br />

and small perturbations and found the latter to be travell<strong>in</strong>g at c. Hence no violation of<br />

causality occurs. Recently, Emig [65] veri ed the <strong>re</strong>sults of Mart<strong>in</strong> and Landauer by means<br />

of computer simulation based on the solution of the Maxwell equations.<br />

Remark (Causality and dispersion) Although the concept of group velocity is usually<br />

<strong>re</strong>garded as physically mean<strong>in</strong>gless <strong>in</strong> the p<strong>re</strong>sence of absorption, it is still widely used <strong>in</strong><br />

connection with Gaussian pulses. When vg then is not smaller than c <strong>in</strong>side an absorption<br />

band (vg >c,vg=1,or vg < 0), the group velocity is simply termed abnormal, but<br />

nonetheless used further to describe the propagation of the maximum of the pulse. This<br />

habit clearly calls for an explanation concern<strong>in</strong>g the <strong>re</strong>spectation of causality. Bolda et al.<br />

[66] proved that any causal dispersive medium must have at least one f<strong>re</strong>quency whe<strong>re</strong> the<br />

group velocity becomes abnormal. In particular, this is a f<strong>re</strong>quency whe<strong>re</strong> the absorption<br />

is an absolute maximum. The proof is based <strong>in</strong> pr<strong>in</strong>ciple on the dispersion <strong>re</strong>lations<br />

formulated by Kramers and Kronig (see, for <strong>in</strong>stance, Jackson [11]) that <strong>re</strong>late the <strong>re</strong>al<br />

and imag<strong>in</strong>ary parts of the <strong>in</strong>dex of <strong>re</strong>fraction n(!) =nr(!)+jni(!), n(!) =ck(!)=!,<br />

nr(!) =1+ 2 C<br />

ni(!) =, 2! C<br />

Z 1<br />

0<br />

Z 1<br />

0<br />

! 0 ni(! 0 )<br />

! 02 ,! 2 d!0<br />

nr(! 0 ),1<br />

! 02 ,! 2 d!0 ;<br />

(2.7)<br />

whe<strong>re</strong> the <strong>in</strong>tegrals a<strong>re</strong> taken to be the Cauchy pr<strong>in</strong>cipal value. All media satisfy<strong>in</strong>g these<br />

<strong>re</strong>lations a<strong>re</strong> causal (<strong>in</strong>clud<strong>in</strong>g the Lo<strong>re</strong>ntz medium <strong>in</strong>vestigated by Sommerfeld). An even<br />

earlier t<strong>re</strong>atment was provided by Toll [67].<br />

An enlighten<strong>in</strong>g contribution to the discussion came from Ste<strong>in</strong>berg [68], who himself had<br />

<strong>in</strong>vestigated the tunnell<strong>in</strong>g of s<strong>in</strong>gle photons through opaque barriers [58]. In these experiments,<br />

too, the photon is best described by a Gaussian pulse, and its peak also appea<strong>re</strong>d at<br />

the tunnel exit sooner than the velocity of light would have permitted. The confusion about<br />

the Gaussian pulses stems largely from the fact that they a<strong>re</strong> not exactly localised as might<br />

be expected, but <strong>in</strong> fact have an <strong>in</strong> nite duration. Yet their pronounced maximum is often<br />

mistaken for the signal itself, such that its <strong>re</strong>ceipt is associated with a signal arrival. As<br />

well the emission of the peak is thought to be the signal transmission. In <strong>re</strong>ality, however,<br />

a Gaussian pulse has neither a beg<strong>in</strong>n<strong>in</strong>g nor an end (and is the<strong>re</strong>fo<strong>re</strong>, <strong>in</strong> a strict sense, <strong>in</strong><br />

itself a violation of causality). Even if the signal has only approximately Gaussian shape, it<br />

must be switched on long befo<strong>re</strong> the peak actually is generated. The turn-on po<strong>in</strong>t would be<br />

a better, if not the only choice as <strong>re</strong>fe<strong>re</strong>nce for delay calculation. This po<strong>in</strong>t, however, travels<br />

with the velocity c. An illustrative | <strong>in</strong> the spirit of Lewis Carroll | explanation of both<br />

this e ect and the quantum optical experiments was provided by Chiao et al. [69]. A critical<br />

<strong>re</strong>view of these experiments was published by Landauer [70]. Nimtz objected vigorously to<br />

24


2.1 Superlum<strong>in</strong>al wave propagation<br />

this argument <strong>in</strong>two <strong>re</strong>cent publications [71, 72]. Real signals a<strong>re</strong> always band-limited, they<br />

argued, so the<strong>re</strong> exists no actual wave front to run ahead of the pulse at the speed of light.<br />

Consequently, superlum<strong>in</strong>al transport of both signals and energy is possible.<br />

Apart from the misconception <strong>re</strong>gard<strong>in</strong>g the actual pulse width, such e ects a<strong>re</strong> well known as<br />

`pulse <strong>re</strong>shap<strong>in</strong>g'. This peculiar behaviour of the evanescent <strong>re</strong>gion can be expla<strong>in</strong>ed <strong>in</strong> a very<br />

illustrative manner [61]. Consider a Gaussian wave packet that has al<strong>re</strong>ady travelled some<br />

distance <strong>in</strong> the medium outside the tunnel. Ow<strong>in</strong>g to dispersion, the lead<strong>in</strong>g tail of the pulse<br />

will be dom<strong>in</strong>ated by higher-f<strong>re</strong>quency components that propagate faster. In addition, these<br />

components a<strong>re</strong> less attenuated when mov<strong>in</strong>g through the evanescent <strong>re</strong>gion. Hence the pulse<br />

appear<strong>in</strong>g on the far side of the barrier will be caused ma<strong>in</strong>ly by the front of the wave packet.<br />

The trail<strong>in</strong>g tail, on the other hand, consists of lower-f<strong>re</strong>quency components that a<strong>re</strong> mostly<br />

absorbed or <strong>re</strong> ected. Consequently, the peak of the transmitted pulse seems to be shifted<br />

forward <strong>in</strong> time, and <strong>in</strong> addition, its cent<strong>re</strong> f<strong>re</strong>quency has <strong>in</strong>c<strong>re</strong>ased. The overall imp<strong>re</strong>ssion is<br />

that the pulse has moved through the barrier at an abnormally g<strong>re</strong>at velocity. In the ext<strong>re</strong>me<br />

case of a medium with ga<strong>in</strong>, the velocity of the pulse may even become negative (see Chiao<br />

et al. [73, 74, 75]).<br />

Remark (Negative pulse velocities) Formally, a medium that provides ga<strong>in</strong> can be<br />

described analogous to the Lo<strong>re</strong>ntz medium with a negative oscillator st<strong>re</strong>ngth. This<br />

medium <strong>in</strong> uences the wave <strong>in</strong>versely to the e ect just described, such that the peak of<br />

the pulse at the end of the evanescent <strong>re</strong>gion leaves the medium even befo<strong>re</strong> the peak of<br />

the <strong>in</strong>cident pulse enters it. Ow<strong>in</strong>g to the fact that the wave can temporarily borrow<br />

energy from the <strong>in</strong>verted atomic states <strong>in</strong> the medium, the energy velocity <strong>in</strong> the classical<br />

sense is negative. This does, however, not <strong>in</strong>terfe<strong>re</strong> with causality, and a wave front due<br />

to a turn-on of the signal still propagates with the velocity of light [74].<br />

Another approach was pursued by Yun-p<strong>in</strong>g and Dian-l<strong>in</strong> [76], who expla<strong>in</strong>ed the <strong>re</strong>shap<strong>in</strong>g<br />

e ect with <strong>in</strong>terfe<strong>re</strong>nce between di e<strong>re</strong>nt possible paths along which aphoton can travel. If<br />

an <strong>in</strong>cident Gaussian wave 0(t) can take several paths, each characterised by a delay time<br />

i and a probability j ij2 , then the output wave can be written as<br />

(t) = X<br />

i 0(t, i): (2.8)<br />

i<br />

If the delay times do not di er too much, the orig<strong>in</strong>al shape of the pulse is approximately<br />

p<strong>re</strong>served despite the <strong>in</strong>terfe<strong>re</strong>nce, and the peak of the pulse is advanced with <strong>re</strong>spect to that<br />

of a <strong>re</strong>fe<strong>re</strong>nce pulse propagat<strong>in</strong>g through f<strong>re</strong>e space. The only problem with this formulation<br />

is that it obviously neglects the dispersion e ects of the medium. Yet, the superposition of<br />

undistorted waves yields an overall dispersive transfer function.<br />

Remark (Multipath propagation) The concept of electromagnetic waves travell<strong>in</strong>g<br />

along di e<strong>re</strong>nt paths from a sender to a <strong>re</strong>ceiver plays an important role <strong>in</strong> the modell<strong>in</strong>g<br />

of mobile communication systems. Particularly <strong>in</strong> urban a<strong>re</strong>as, the signal at the <strong>re</strong>ceiver<br />

antenna is a comb<strong>in</strong>ation of waves scatte<strong>re</strong>d from and di racted around the surround<strong>in</strong>g<br />

build<strong>in</strong>gs. Depend<strong>in</strong>g on the phase delays, the <strong>in</strong>terfe<strong>re</strong>nce can be constructive or<br />

25


2 Signals faster than light?<br />

destructive, chang<strong>in</strong>g rapidly as the <strong>re</strong>ceiver moves on. Thus the signal amplitude undergoes<br />

signi cant variations over short distances. This phenomenon is known as short-term<br />

or Rayleigh fad<strong>in</strong>g [77], and it is dispersive <strong>in</strong> natu<strong>re</strong> even though the propagation of the<br />

<strong>in</strong>dividual waves is not.<br />

Consider<strong>in</strong>g a monochromatic signal and keep<strong>in</strong>g the <strong>re</strong>ceiver xed (thus sidestepp<strong>in</strong>g the<br />

problem of a Doppler shift <strong>in</strong>troduced by amov<strong>in</strong>g <strong>re</strong>ceiver), we can describe the output<br />

signal of a multipath channel by a sum of delayed, time-shifted copies of the <strong>in</strong>put signal<br />

like <strong>in</strong> (2.8). Cor<strong>re</strong>spond<strong>in</strong>gly, the transfer function of the channel is given by [78]<br />

H(j!)= X<br />

aie ,jkli = X<br />

ai e ,j! li c = X<br />

ai e ,j! i ; (2.9)<br />

i<br />

i<br />

whe<strong>re</strong> we used the de nition of the f<strong>re</strong>quency-<strong>in</strong>dependent phase velocity vp = c = !=k.<br />

In order to be dispersive, the phase of H(j!) must depend other than l<strong>in</strong>early on !, so<br />

that the phase delay arg H(j!)=! is not constant. This is most likely the case <strong>in</strong> practice,<br />

although the li can be chosen so as to <strong>re</strong>sult <strong>in</strong> a transfer function with l<strong>in</strong>ear phase (which<br />

provides an <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g connection to the theory of l<strong>in</strong>ear FIR lters).<br />

2.2 Quantum mechanical tunnell<strong>in</strong>g<br />

Although the knowledge on electromagnetic wave propagation <strong>in</strong> uenced the development of<br />

quantum mechanics to a g<strong>re</strong>at extent, the analogy between the <strong>re</strong>spective tunnell<strong>in</strong>g phenomena<br />

was not given too much attention <strong>in</strong> the past. Only <strong>re</strong>cently, Mart<strong>in</strong> and Landauer<br />

[61] demonstrated that the cont<strong>in</strong>uity conditions for TE and TM modes at the <strong>in</strong>terfaces<br />

between di e<strong>re</strong>nt dielectrics <strong>in</strong> a uniform waveguide ( g. 2.2) a<strong>re</strong> exactly the same as the ones<br />

that must be satis ed by the wave function at the edges of a <strong>re</strong>ctangular potential barrier<br />

for the classical tunnel e ect. In the TE case and if the permeabilities of both <strong>re</strong>gions a<strong>re</strong><br />

equal, the cont<strong>in</strong>uity of the longitud<strong>in</strong>al magnetic and the transverse electric eld components<br />

cor<strong>re</strong>spond to the cont<strong>in</strong>uity of the quantum mechanical wave function. On the other hand,<br />

the cont<strong>in</strong>uity of the transverse components of the magnetic eld is equivalent to the cont<strong>in</strong>uity<br />

ofthederivative ofthe wave function <strong>in</strong> a tunnell<strong>in</strong>g problem. Thus the <strong>in</strong>vestigation<br />

of evanescent electromagnetic waves also provides <strong>in</strong>sight <strong>in</strong>to the behaviour of tunnell<strong>in</strong>g<br />

electrons.<br />

Apart from the formal aspects the<strong>re</strong> is also a good experimental <strong>re</strong>ason why this cor<strong>re</strong>spondence<br />

is <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g. Quantum mechanical measu<strong>re</strong>ments a<strong>re</strong> always <strong>in</strong>vasive and disturb the<br />

process be<strong>in</strong>g observed. This is particularly annoy<strong>in</strong>g if the arrival of a wave packet is to be<br />

measu<strong>re</strong>d just ahead of the barrier <strong>in</strong> order to determ<strong>in</strong>e its traversal time. Electromagnetic<br />

wave packets, on the contrary, consist of many photons, some of which may safely be used<br />

for a measu<strong>re</strong>ment without exert<strong>in</strong>g too g<strong>re</strong>at an <strong>in</strong> uence on the <strong>re</strong>st of the wave. This<br />

<strong>re</strong>ason<strong>in</strong>g was the departu<strong>re</strong> po<strong>in</strong>t for the experiments discussed <strong>in</strong> the p<strong>re</strong>vious section.<br />

One might expect that the equivalence of the two tunnell<strong>in</strong>g phenomena is su cient to discuss<br />

the problem of tunnell<strong>in</strong>g times, yet the exclusive consideration of evanescent waves<br />

gives only part of the pictu<strong>re</strong> of the controversy. It is nonetheless worthwile to sketch other<br />

approaches to the traversal time that orig<strong>in</strong>ated from the discussion of the `classical' tunnel<br />

26<br />

i


2.2 Quantum mechanical tunnell<strong>in</strong>g<br />

e ect. In quantum mechanics, a particle is described <strong>in</strong> terms of its cor<strong>re</strong>spond<strong>in</strong>g complex<br />

wave function (x; t), which satis es a particular wave equation, the Schrod<strong>in</strong>ger equation:<br />

~ 2<br />

2m<br />

+ j~ @<br />

@t<br />

, V =0: (2.10)<br />

In this equation, ~ = h=(2 ) is Planck's quantum of action, m is the mass of the particle, and<br />

V is a usually time-<strong>in</strong>variant potential. In the absence of transient processes the probability<br />

density j (x; t)j 2 is <strong>in</strong>dependent of time, and one can obta<strong>in</strong> a simpler version by separat<strong>in</strong>g<br />

the variables and sett<strong>in</strong>g<br />

(x; t) = (x)e ,j!t ; (2.11)<br />

which gives the time-<strong>in</strong>dependent Schrod<strong>in</strong>ger equation or energy eigenequation<br />

~ 2<br />

2m<br />

+ E ,V =0; (2.12)<br />

whe<strong>re</strong> E = !~ is the energy of the particle. Obviously, this equation has one-dimensional<br />

solutions of the form<br />

(x) =Ae<br />

j 1<br />

~ xp 2m(E,V) : (2.13)<br />

The amplitude A is a normalisation constant and chosen such that the overall probability density<br />

of the particle equals one, which means that the particle is guaranteed to be somewhe<strong>re</strong>.<br />

Hence,<br />

Z 1<br />

,1<br />

(x) (x) dx =1: (2.14)<br />

pThe case we a<strong>re</strong> <strong>in</strong>te<strong>re</strong>sted <strong>in</strong> occurs whenever E < V , so that the wave number k =<br />

2m(E , V )=~ becomes imag<strong>in</strong>ary. Then the oscillation of the stationary wave turns <strong>in</strong>to an<br />

exponential decay and we have an e ect similar to evanescence <strong>in</strong> the electromagnetic case.<br />

The simplest and classical arrangement lead<strong>in</strong>g to the tunnel e ect is a <strong>re</strong>ctangular potential<br />

barrier with<br />

V =<br />

8<br />

><<br />

>:<br />

0 if ,1


2 Signals faster than light?<br />

Remark (Scatter<strong>in</strong>g and escap<strong>in</strong>g) The potential barrier conside<strong>re</strong>d above isonly<br />

one possibility whe<strong>re</strong> tunnell<strong>in</strong>g can occur. It is a scatter<strong>in</strong>g con guration <strong>in</strong> that an<br />

<strong>in</strong>dividual particle imp<strong>in</strong>g<strong>in</strong>g on the barrier can be <strong>re</strong> ected or transmitted. A st<strong>re</strong>am of<br />

particles will the<strong>re</strong>fo<strong>re</strong> be scatte<strong>re</strong>d. The other scenario is the escape from bound states<br />

(see, for example, Landauer [79]). A particle caught <strong>in</strong> a potential well surrounded by<br />

walls with nite height can leave the well by tunnell<strong>in</strong>g through the wall. The classical<br />

problem, however, is that of scatter<strong>in</strong>g, and so we <strong>re</strong>strict the survey to the controversy<br />

on tunnell<strong>in</strong>g times <strong>in</strong> such processes.<br />

In the course of time, the<strong>re</strong> we<strong>re</strong> quite a number of survey articles on various de nitions of<br />

quantum mechanical tunnell<strong>in</strong>g times, and we shall browse he<strong>re</strong> only through the best-known<br />

approaches. Perhaps the most <strong>in</strong>tuitive one is the phase time. It is based on the method<br />

of stationary phase and thus <strong>re</strong>lated to the group velocity concept of classical waves. The<br />

motion of the wave packet is aga<strong>in</strong> de ned via the evolution of a particular po<strong>in</strong>t. Usually the<br />

maximum is chosen as <strong>re</strong>fe<strong>re</strong>nce. Hence this concept is limited to narrow-band signals with<br />

a clearly observable peak like Gaussian pulses (which <strong>in</strong> this context a<strong>re</strong> also <strong>re</strong>fer<strong>re</strong>d to as<br />

m<strong>in</strong>imum uncerta<strong>in</strong>ty pulses). For suchawave packet, Hartman [80] found that the wave does<br />

not, as had p<strong>re</strong>viously been assumed, traverse a potential barrier <strong>in</strong> no app<strong>re</strong>ciable time, but<br />

rather experiences a dist<strong>in</strong>ct delay that is <strong>in</strong> priciple <strong>in</strong>dependent of the barrier thickness. This<br />

gave rise to the al<strong>re</strong>ady mentioned superlum<strong>in</strong>ality discussions and also to the Enders-Nimtz<br />

experiment, although Hartman himself emphasised that the e ect was caused by a f<strong>re</strong>quency<br />

shift <strong>in</strong> the wave packet emerg<strong>in</strong>g from the tunnel and was thus <strong>in</strong> perfect ag<strong>re</strong>ement with<br />

causality. Fletcher [81] con rmed these observations for the even simpler case of a cosh ,1 -like<br />

wave packet. Another variant of the phase time uses the centroid of the wave function rather<br />

than its peak (Mart<strong>in</strong> and Landauer [61]). Derived from the phase time concept is that of<br />

group delay, whe<strong>re</strong> the propagation delay di e<strong>re</strong>nce between two identical wave packets is<br />

conside<strong>re</strong>d | one pass<strong>in</strong>g through a barrier, its tw<strong>in</strong> travell<strong>in</strong>g the same distance <strong>in</strong> f<strong>re</strong>e<br />

space. This approach appears chie y <strong>in</strong> quantum optics and <strong>re</strong> ects a classical experimental<br />

setup (Ste<strong>in</strong>berg and Chiao [18]).<br />

The phase time concept has been, and still is, very controversial. It has been argued that<br />

its vividness stems from a too classical understand<strong>in</strong>g of particle motion (Thornber et al.<br />

[82]) and dis<strong>re</strong>gards the wave aspect that is essential for the tunnell<strong>in</strong>g phenomenon. In fact,<br />

the wave is collapsed <strong>in</strong>to a s<strong>in</strong>gle po<strong>in</strong>t whose trajectory is traced. This is a second po<strong>in</strong>t of<br />

attack: it is by no means clear that the peak of the wave function ahead of the barrier and the<br />

one beh<strong>in</strong>d a<strong>re</strong> identical, let alone <strong>re</strong>latable <strong>in</strong> a physically sensible way. On the other side,<br />

<strong>re</strong>views like those by Coll<strong>in</strong>s et al. [83], Barker [84], or Hauge and St vneng [85] advocated<br />

the phase time as appropriate description of the tunnel e ect at least for Gaussian pulses.<br />

The other group of phase time proponents consists of <strong>re</strong>searchers deal<strong>in</strong>g with electromagnetic<br />

waves like <strong>in</strong> quantum optics, as we have al<strong>re</strong>ady seen <strong>in</strong> the p<strong>re</strong>vious section. Ranfagni et al.,<br />

for example, found the phase time most suitable to <strong>re</strong>produce their microwave experiments<br />

[55]. Nonetheless, particularly <strong>in</strong> <strong>re</strong>cent articles, the phase time as <strong>re</strong>levant tunnell<strong>in</strong>g time is<br />

enti<strong>re</strong>ly denied (Landauer and Mart<strong>in</strong> [86]) or <strong>re</strong>stricted to very special cases, i. e. whe<strong>re</strong> the<br />

trajectories of the pulse peaks that undeniably exist outside the barrier can be extrapolated<br />

<strong>in</strong>to the barrier to give a simple description of the tunnell<strong>in</strong>g event (Leavens et al. [87]).<br />

28


2.2 Quantum mechanical tunnell<strong>in</strong>g<br />

Remark (Interfe<strong>re</strong>nce <strong>in</strong> front of the barrier) The<strong>re</strong> is still another subtlety <strong>in</strong>hibit<strong>in</strong>g<br />

the straightforward assumption that the phase time gives p<strong>re</strong>cisely the time spent<br />

<strong>in</strong> the barrier by the particle. S<strong>in</strong>ce the phase time (like the group velocity) is asymptotic<br />

<strong>in</strong> natu<strong>re</strong>, it <strong>re</strong>qui<strong>re</strong>s that the wave packet conta<strong>in</strong>s only few f<strong>re</strong>quency (or energy)<br />

components, which <strong>in</strong> turn makes the packet very broad <strong>in</strong> space. When this long packet<br />

approaches the barrier, a portion of its lead<strong>in</strong>g tail will <strong>in</strong>evitably be <strong>re</strong> ected and <strong>in</strong>terfe<strong>re</strong><br />

with the ma<strong>in</strong> body of the pulse. This slows down the approach ofthe pulse, and<br />

the<strong>re</strong>fo<strong>re</strong> the extrapolated phase time consists of two parts that cannot be separated: the<br />

delay due to self-<strong>in</strong>terfe<strong>re</strong>nce and the <strong>in</strong>teraction time with the barrier itself [85].<br />

Another tunnell<strong>in</strong>g time approach is the dwell time, which is de ned as the ratio of the<br />

probability density <strong>in</strong> the barrier to the <strong>in</strong>cident ux J,<br />

D = 1<br />

J<br />

Z<br />

B<br />

j (x; t)j 2 dx : (2.17)<br />

The <strong>in</strong>tegration extends over the barrier <strong>re</strong>gion and gives the time a particle spends <strong>in</strong> the<br />

tunnel, averaged over all <strong>in</strong>com<strong>in</strong>g particles. The ux or probability density cur<strong>re</strong>nt can be<br />

derived from a cont<strong>in</strong>uity equation for the probability density and is <strong>in</strong> its most general<br />

form<br />

J = j~ ,<br />

r , r<br />

2m<br />

: (2.18)<br />

For the special case of plane waves = Ae jkx , this exp<strong>re</strong>ssion is simpli ed to J = ~k=m. In<br />

a way this de nition <strong>re</strong>sembles the velocity of energy propagation for classical waves, whe<strong>re</strong><br />

we had the ratio of transported and sto<strong>re</strong>d enery. The dwell time, however, su ers from a<br />

seve<strong>re</strong> drawback <strong>in</strong> that it cannot dist<strong>in</strong>guish whether a particle is ultimately <strong>re</strong> ected or<br />

transmitted. It has the<strong>re</strong>fo<strong>re</strong> been argued that while it is an appropriate description of the<br />

mean <strong>in</strong>teraction time of a particle with the barrier, it may not be used to characterise the<br />

actual transmission time (Buttiker [88], Coll<strong>in</strong>s et al. [83]). On the other hand, <strong>re</strong> ection<br />

and transmission a<strong>re</strong> mutually exclusive events, and hence one can at least formally de ne<br />

dist<strong>in</strong>ct <strong>re</strong> ection and transmission times R and T that must be <strong>re</strong>lated by the <strong>re</strong>spective<br />

<strong>re</strong> ection and transmission coe cients, R + T =1,<br />

D=T T+R R : (2.19)<br />

This formulation has become widely accepted, and Hauge and St vneng [85] made it a focal<br />

criterion of their <strong>re</strong>view article.<br />

Remark (Objections aga<strong>in</strong>st (2.19)) The decomposition of the dwell time is not<br />

undisputed. Landauer and Mart<strong>in</strong> [86] argued that (2.17) and (2.19) imply<br />

Z<br />

B<br />

j (x; t)j 2 dx = JT T +JR R: (2.20)<br />

This comes down to the addition of probability densities. In quantum mechanics, however,<br />

it is usual to add only complex wave functions, and the<strong>re</strong>fo<strong>re</strong> (2.19) lacks any physical<br />

mean<strong>in</strong>g.<br />

29


2 Signals faster than light?<br />

The conditional average approach of (2.19) was <strong>re</strong>cently extended by Ste<strong>in</strong>berg [89, 90] on the<br />

basis of conditional probabilities to <strong>in</strong>clude the e ects of <strong>re</strong>al-world measu<strong>re</strong>ments. To this<br />

end, the conventional de nition of the dwell time had to be extended by an imag<strong>in</strong>ary part<br />

describ<strong>in</strong>g the backaction of the measu<strong>re</strong>ment device on the tunnell<strong>in</strong>g particle. In addition,<br />

he found that the major contributions to the barrier <strong>in</strong>teraction time come from the edges of<br />

the barrier, but hardly from its cent<strong>re</strong>. A transmitted particle the<strong>re</strong>fo<strong>re</strong> spends equal amounts<br />

of time near either of the <strong>in</strong>terfaces, a <strong>re</strong> ected particle stays of course only close to the tunnel<br />

entrance.<br />

Apart from the phase time de nition the<strong>re</strong> a<strong>re</strong> several other approaches that apply the classical<br />

concept of trajectories to quantum mechanics. They a<strong>re</strong> derived from di e<strong>re</strong>nt formal ways<br />

to describe quantum mechanics, such as Bohm's trajectory formulation (Leavens and Aers<br />

[91]) or Feynman paths. The former t<strong>re</strong>ats a wave packet always like a particle <strong>in</strong> the classical<br />

sense, whe<strong>re</strong>as the latter yields complex tunnell<strong>in</strong>g times, which makes many <strong>re</strong>searches feel<br />

uneasy because <strong>re</strong>sults <strong>in</strong> quantum mechanics a<strong>re</strong> usually <strong>re</strong>al-valued quantities. An overview<br />

of the trajectory approaches was given by Landauer and Mart<strong>in</strong> [86] as well as Leavens et al.<br />

[87].<br />

The last group of tunnell<strong>in</strong>g time de nitions is that of quantum clocks. This approach associates<br />

the motion of the particle with some physical deg<strong>re</strong>e of f<strong>re</strong>edom that plays the role of<br />

a clock and can be used to measu<strong>re</strong> the time elapsed dur<strong>in</strong>g the tunnell<strong>in</strong>g event. The<strong>re</strong> have<br />

been several proposals like oscillat<strong>in</strong>g barriers, oscillat<strong>in</strong>g <strong>in</strong>cident amplitudes or the Larmor<br />

p<strong>re</strong>cession. The idea of the last one is to exploit the p<strong>re</strong>cession of the sp<strong>in</strong> of a particle <strong>in</strong><br />

a magnetic eld, provided that the eld is <strong>in</strong> nitesimally small and con ned to the barrier<br />

<strong>re</strong>gion (Buttiker [88]). The time it takes the particle to tunnel through then determ<strong>in</strong>es the<br />

rotation of the sp<strong>in</strong>. The acceptance of quantum clocks is not unanimous: they we<strong>re</strong> criticised<br />

as un<strong>re</strong>liable [85], but also strongly favou<strong>re</strong>d [86], at any rate they a<strong>re</strong> well established today.<br />

30


Chapter 3<br />

<strong>Wave</strong> propagation <strong>in</strong><br />

electromagnetic transmission l<strong>in</strong>es<br />

One cannot escape the feel<strong>in</strong>g that these mathematical formulae have an<br />

<strong>in</strong>dependent existence of their own, that they a<strong>re</strong> wiser than we a<strong>re</strong>, wiser<br />

even than their discove<strong>re</strong>rs, that we get mo<strong>re</strong> out of them than was orig<strong>in</strong>ally<br />

put <strong>in</strong>to them. He<strong>in</strong>rich Hertz, quoted <strong>in</strong> [1]<br />

As the historical <strong>re</strong>view shows, op<strong>in</strong>ions a<strong>re</strong> divided on the question of wave propagation.<br />

To ga<strong>in</strong> some further <strong>in</strong>sight, we shall explo<strong>re</strong> several examples <strong>in</strong> the sequel, rang<strong>in</strong>g from<br />

the simple case of a homogeneous transmission l<strong>in</strong>e to transient phenomena <strong>in</strong> wave guides.<br />

The ma<strong>in</strong> <strong>in</strong>te<strong>re</strong>st of the analysis lies on signals <strong>in</strong> the evanescent <strong>re</strong>gion or stop band of<br />

transmission l<strong>in</strong>es, and among the many de nitions of the propagation velocity, we shall<br />

concentrate on that of energy velocity. Not only is its de nition clear-cut and <strong>in</strong>tuitive, we<br />

shall also nd that it gives <strong>re</strong>asonable <strong>re</strong>sults whe<strong>re</strong> the concept of group velocity must fail.<br />

At the beg<strong>in</strong>n<strong>in</strong>g of the chapter, we study the homogeneous and lossless transmission l<strong>in</strong>e<br />

<strong>in</strong> the pass band and con rm the equality of group and energy velocity. We then dig<strong>re</strong>ss<br />

to explo<strong>re</strong> a classical delay l<strong>in</strong>e known from the literatu<strong>re</strong>, whe<strong>re</strong> this identity also holds.<br />

Gradually complicat<strong>in</strong>g the model, we consider evanescence <strong>in</strong> connection with a mismatched<br />

term<strong>in</strong>ation of a nitely long transmission l<strong>in</strong>e. The logical extension of this <strong>in</strong>vestigation is to<br />

consider a f<strong>re</strong>quency-dependent medium. We thus evaluate the behaviour of waves <strong>in</strong> a lossless<br />

plasma. A short section on an <strong>in</strong>homogeneous transmission l<strong>in</strong>e concludes the t<strong>re</strong>atment of<br />

monochromatic signals. A large part is then devoted to the evaluation of transients <strong>in</strong> a<br />

plasma and a <strong>re</strong>ctangular wave guide. In both cases we shall encounter the propagation of a<br />

wave front, whose velocity will be equal to the velocity of light. The <strong>re</strong>ma<strong>in</strong>der of the chapter<br />

explo<strong>re</strong>s the propagation of a Gaussian pulse <strong>in</strong> plasma. In connection with this comparatively<br />

small-band signal, we shall also touch upon the question of causality <strong>in</strong>evanescent media.<br />

31


U<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

I<br />

jX 0 dx<br />

dx<br />

jB 0 dx<br />

Figu<strong>re</strong> 3.1: Equivalent circuit for a lossless transmission l<strong>in</strong>e<br />

3.1 Model of a transmission l<strong>in</strong>e<br />

In section 1.4 we al<strong>re</strong>ady mentioned that the energy velocity equals the group velocity for lossless<br />

media and for narrow-band signals. Let us now verify this statement for a simple example.<br />

We consider a general homogeneous transmission l<strong>in</strong>e without losses. In addition, we assume<br />

an ideal term<strong>in</strong>ation with the characteristic impedance, so that we need not worry about<br />

<strong>re</strong> ections. The transmission l<strong>in</strong>e can be described by means of its well-known equivalent<br />

circuit as depicted <strong>in</strong> g. 3.1 with arbitrary distributed <strong>re</strong>actances X 0 (!) and susceptances<br />

B 0 (!). The impedances may exhibit any dependence on the f<strong>re</strong>quency provided that they<br />

satisfy Foster's theo<strong>re</strong>m [92]<br />

dX 0<br />

d!<br />

dB 0<br />

d!<br />

jX 0 j<br />

!<br />

jB 0 j<br />

!<br />

(3.1)<br />

which exp<strong>re</strong>sses the physical necessity that the energy sto<strong>re</strong>d <strong>in</strong> such an impedance is always<br />

non-negative. The di e<strong>re</strong>ntial equations for voltage and cur<strong>re</strong>nt a<strong>re</strong> obta<strong>in</strong>ed as<br />

and may be solved by sett<strong>in</strong>g<br />

dU<br />

dx = ,jX0 I<br />

dI<br />

dx = ,jB0 U<br />

U = Ue j(!t,kx)<br />

I = Ie j(!t,kx) :<br />

Remark (Time dependence of the solutions) Note that <strong>in</strong> contrast to the p<strong>re</strong>ceed<strong>in</strong>g<br />

chapters, the waves <strong>in</strong> (3.3) have been formulated with a time dependence e j!t . Throughout<br />

the case studies, we shall use this standard electrodynamical notation because it is<br />

32<br />

(3.2)<br />

(3.3)


3.1 Model of a transmission l<strong>in</strong>e<br />

the foundation of the well-known cor<strong>re</strong>spondence @<br />

@t 7! j! for stationary oscillations and<br />

consequently the basis for the equivalent circuit <strong>in</strong> g. 3.1.<br />

We then obta<strong>in</strong> easily the propagation constant<br />

and the characteristic impedance<br />

k = p X 0 B 0 (3.4)<br />

Z0 = U<br />

I =<br />

r X 0<br />

: (3.5)<br />

B0 <strong>Wave</strong> propagation is possible only if X 0 B 0 > 0, i. e. the signs of the distributed <strong>re</strong>actances<br />

and susceptances must be equal. From the dispersion <strong>re</strong>lation (3.4) we nd the <strong>re</strong>ciprocal of<br />

the group velocity<br />

1<br />

vg<br />

= dk<br />

d!<br />

= 1<br />

2k<br />

X0 dB0<br />

d!<br />

+ B0 dX0<br />

d!<br />

: (3.6)<br />

Remark (Di<strong>re</strong>ction of propagation) As can be seen from (3.6), X 0 > 0 and B 0 > 0<br />

<strong>re</strong>sult <strong>in</strong> sign vp = sign vg, so that the wave c<strong>re</strong>sts and wave packets move <strong>in</strong> the same<br />

di<strong>re</strong>ction. If we choose negative impedances, X 0 < 0 and B 0 < 0, <strong>re</strong>spectively, the phase<br />

and group velocities have opposite signs, which ischaracteristic for a backward wave.<br />

In the absence of <strong>re</strong> ection, which was our <strong>in</strong>itial assumption, the average of the propagated<br />

energy is simply determ<strong>in</strong>ed by the amplitude of the cur<strong>re</strong>nt I and the characteristic<br />

impedance,<br />

P = I2<br />

2<br />

r X 0<br />

: (3.7)<br />

B0 On the other hand, the mean sto<strong>re</strong>d energy per unit length, with U as voltage amplitude, is<br />

given by<br />

W = 1<br />

4<br />

2 dX0 2 dB0<br />

I + U<br />

d! d!<br />

I2 dX0 dB0<br />

= B0 + X0<br />

4B0 d! d!<br />

; (3.8)<br />

whe<strong>re</strong> we used the de nition of the characteristic impedance (3.5) to obta<strong>in</strong> the second exp<strong>re</strong>ssion.<br />

From (3.7) and (3.8) we see immediately that the energy velocity ve = P=W equals<br />

the group velocity (3.6).<br />

Now let us brie y exam<strong>in</strong>e the situation when the transmission l<strong>in</strong>e is characterised by a<br />

f<strong>re</strong>quency-<strong>in</strong>dependent <strong>in</strong>ductance L 0 and a capacitance C 0 per unit length. We then have<br />

X 0 = !L 0 ; B 0 = !C 0 : (3.9)<br />

33


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

Losses a<strong>re</strong> accounted for by a series <strong>re</strong>sistance R 0 andashunt conductance G 0 . The propagation<br />

constant aswell as the characteristic impedance then become complex,<br />

k = p (!L 0 , jR 0 )(!C 0 , jG 0 )= +j (3.10)<br />

Z 0 =<br />

s !L 0 , jR 0<br />

!C 0 , jG 0 = Z0 + jX0 : (3.11)<br />

With these de nitions, we nd the propagated and sto<strong>re</strong>d energies<br />

P =Re UI<br />

2<br />

W = 1<br />

4<br />

and subsequently the energy velocity<br />

= 1<br />

2 I2 Z0<br />

, I 2 L 0 + U 2 C 0 = 1<br />

4 I2 , L 0 + jZ 0j 2 C 0<br />

ve = P<br />

W =<br />

(3.12)<br />

(3.13)<br />

2Z0<br />

L0 +(Z0 2 +X0 2 : (3.14)<br />

)C0 We can now use the <strong>re</strong>lation k = Z 0(!C 0 , jG 0 ) to exp<strong>re</strong>ss Z0 <strong>in</strong> terms of and . Furthermo<strong>re</strong>,<br />

we take the squa<strong>re</strong> of (3.10) to obta<strong>in</strong> an exp<strong>re</strong>ssion for . The absolute value of Z 0<br />

is <strong>re</strong>adily found from (3.11). Tak<strong>in</strong>g all these <strong>re</strong>lations together and <strong>in</strong>sert<strong>in</strong>g them <strong>in</strong>to the<br />

exp<strong>re</strong>ssion for the energy velocity, we get after some algebra the astonish<strong>in</strong>g <strong>re</strong>sult<br />

ve = ! ; (3.15)<br />

which is the phase velocity de ned by the <strong>re</strong>al part of the complex dispersion <strong>re</strong>lation. This<br />

observation has also been <strong>re</strong>ported by Ma<strong>in</strong>ardi [32] (see section 1.4). Needless to emphasise<br />

that <strong>in</strong> the absence of losses (R 0 = G 0 = 0), the phase, group, and energy velocities a<strong>re</strong><br />

identical, and no dispersion occurs. In this connection and for the sake of completeness, we<br />

also <strong>re</strong>call the well-known condition R 0 C 0 = L 0 G 0 for a dispersion-f<strong>re</strong>e transmission l<strong>in</strong>e even<br />

<strong>in</strong> the p<strong>re</strong>sence of losses.<br />

Remark (F<strong>re</strong>quency dependence of losses) At rst sight, the propagation constant<br />

(3.10) <strong>re</strong>veals an <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g property for ! !1. With<strong>in</strong> the scope of this model, the<br />

<strong>re</strong>al part is roughly proportional to !, whe<strong>re</strong>as the imag<strong>in</strong>ary part <strong>re</strong>ma<strong>in</strong>s constant.<br />

This <strong>in</strong>dicates the existence of a dist<strong>in</strong>ct wave front velocity (see section 1.3). Thus a<br />

step <strong>in</strong> a signal may be attenuated, but the wave front itself would never be distorted.<br />

Unfortunately, the assumption of f<strong>re</strong>quency-<strong>in</strong>dependent losses a<strong>re</strong> no longer justi ed<br />

at high f<strong>re</strong>quencies. In fact, the series <strong>re</strong>sistance is <strong>in</strong>c<strong>re</strong>ased by the sk<strong>in</strong> e ect (thus<br />

R 0 / p !), whe<strong>re</strong>as the shunt conductance is deteriorated by dielectric losses (G 0 / !).<br />

Consequently, the imag<strong>in</strong>ary part will <strong>in</strong>c<strong>re</strong>ase with grow<strong>in</strong>g f<strong>re</strong>quency, and the wave<br />

front will be distorted.<br />

An analysis based on an equivalent circuit is <strong>in</strong>complete without discussion of its limits of<br />

validity. The derivation of the equivalent circuit <strong>re</strong>lies on the fact that adjacent <strong>in</strong> nitesimal<br />

34


3.2 Excursion: a delay l<strong>in</strong>e<br />

sections of the transmission l<strong>in</strong>e do not <strong>in</strong> uence each other. As the voltage and cur<strong>re</strong>nt <strong>in</strong> the<br />

l<strong>in</strong>e stand for the electric and magnetic elds, <strong>re</strong>spectively, this is tantamount to the exclusive<br />

occu<strong>re</strong>nce of TEM waves [93]. A slightly less <strong>re</strong>strictive de nition based on the alternative<br />

description of the elds by a scalar and a vector potential was given by Paschke [92]. The use<br />

of the equivalent circuit is then justi ed when the contribution of the vector potential to the<br />

electric eld is negligible compa<strong>re</strong>d with that of the scalar potential, so that the electric eld<br />

is almost curl-f<strong>re</strong>e (r E ' 0). Apart from TEM waves, whe<strong>re</strong> the <strong>re</strong>qui<strong>re</strong>ment is identically<br />

satis ed, it is a good approximation also for slow waves characterised by vp c.<br />

3.2 Excursion: a delay l<strong>in</strong>e<br />

As a second example, we <strong>re</strong>view now the model of a delay l<strong>in</strong>e shown <strong>in</strong> g. 3.2 , which was<br />

al<strong>re</strong>ady analysed by Borgnis [38] as well as Kleen and Poschl [39]. The walls of the wave<br />

guide consist of conduct<strong>in</strong>g combs with narrowly spaced teeth. For the sake of analytical<br />

simplicity, we assume that the extention <strong>in</strong> z-di<strong>re</strong>ction is <strong>in</strong> nite and all eld components <strong>in</strong><br />

this di<strong>re</strong>ction a<strong>re</strong> constant. In addition, s<strong>in</strong>ce the boundaries a<strong>re</strong> perfectly conduct<strong>in</strong>g, the<strong>re</strong><br />

is no component of the electric eld <strong>in</strong> z-di<strong>re</strong>ction. The magnetic eld, on the contrary, does<br />

exist <strong>in</strong> this di<strong>re</strong>ction only. Hence we have<br />

E z=0; H x=H y=0;<br />

@<br />

@z<br />

=0: (3.16)<br />

With these <strong>re</strong>strictions and the assumption of a l<strong>in</strong>ear and isotropic medium, the Maxwell<br />

equations <strong>re</strong>duce to (from r H =@D=@t)<br />

and (from r E=,@B=@t)<br />

@Hz @y = " @Ex @t<br />

;<br />

@Hz @x = ," @Ey @t<br />

@Ey @x , @Ex @y = , @Hz @t<br />

(3.17)<br />

: (3.18)<br />

Note that the other two Maxwell equations a<strong>re</strong> also satis ed. It follows di<strong>re</strong>ctly from our<br />

assumptions (3.16) that rH = 0, and rE = 0 can be shown by di e<strong>re</strong>ntiat<strong>in</strong>g this equation<br />

with <strong>re</strong>spect to t and those of (3.17) with <strong>re</strong>spect to x and y.<br />

The di e<strong>re</strong>ntial equations can be comb<strong>in</strong>ed to yield a wave equation for the transverse magnetic<br />

component<br />

@2H z<br />

@x2 + @2Hz @y2 = " @2H z<br />

@t2 : (3.19)<br />

Like befo<strong>re</strong>, we set H z = Hze j(!t,kx) , and with c 2 =1=( ") and = p k 2 , (!=c) 2 we nally<br />

obta<strong>in</strong> the two possible solutions for the amplitude of the magnetic eld, an antisymmetric<br />

and a symmetric mode,<br />

Hz = C s<strong>in</strong>h y ; Hz = C cosh y (3.20)<br />

35


z<br />

y<br />

x<br />

l<br />

2d<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

Figu<strong>re</strong> 3.2: Structu<strong>re</strong> of the delay l<strong>in</strong>e.<br />

as well as the cor<strong>re</strong>spond<strong>in</strong>g electric eld components<br />

Ey = kC<br />

!" s<strong>in</strong>h y ; Ey = kC<br />

!"<br />

Ex = C<br />

j!" cosh y ; Ex = C<br />

j!"<br />

cosh y ; (3.21)<br />

s<strong>in</strong>h y : (3.22)<br />

As for the boundary conditions, the conduct<strong>in</strong>g combs act like a shorted strip l<strong>in</strong>e of length<br />

l, and the<strong>re</strong>fo<strong>re</strong> the elds see at y = d the cor<strong>re</strong>spond<strong>in</strong>g well-known impedance. This<br />

structu<strong>re</strong> was quite commonly used and is known as `<strong>in</strong>ductive wall'. The boundary condition<br />

then becomes<br />

, Ex<br />

r<br />

= jX(!) =j<br />

Hz y=d<br />

" tan !l<br />

: (3.23)<br />

c<br />

With the normalisation K = kd, = !d=c, and L = l=d we nally obta<strong>in</strong> the implicit<br />

dispersion <strong>re</strong>lations for the antisymmetric mode,<br />

p p<br />

K2 , 2 coth K2 , 2 = tan( L) (3.24)<br />

and for the symmetric mode<br />

p p<br />

K2 , 2 tanh K2 , 2 = tan( L) : (3.25)<br />

The numerical evaluation of these <strong>re</strong>lations for the base band is given <strong>in</strong> g. 3.3 . Note that<br />

the boundary condition (3.23) <strong>re</strong>lies on the fact that the teeth of the combs a<strong>re</strong> <strong>in</strong> nitely close<br />

together. If this distance is not small compa<strong>re</strong>d to the wavelength, then spatial harmonics<br />

would have tobetaken <strong>in</strong>to account.<br />

For the calculation of the sto<strong>re</strong>d energy it is important to consider not only the space jyj d<br />

whe<strong>re</strong> the propagation occurs, but also the <strong>in</strong>ductive walls d jyj d + l. It was this<br />

component that Borgnis [38] dis<strong>re</strong>garded <strong>in</strong> his rst analysis and that led to the false <strong>re</strong>sult<br />

ve = vp. The average energy sto<strong>re</strong>d <strong>in</strong> the upper wall is given by<br />

W w = 1<br />

4<br />

jHzj 2<br />

36<br />

y=d<br />

dX<br />

d!<br />

(3.26)


3.3 Re ection due to term<strong>in</strong>ation mismatch<br />

1.5<br />

1.25<br />

1<br />

0.75<br />

0.5<br />

0.25<br />

Ω<br />

0 2 4 6 8 10 12 14<br />

Figu<strong>re</strong> 3.3: Dispersion <strong>re</strong>lations for the delay l<strong>in</strong>e of g. 3.2 for L =1. The solid curve cor<strong>re</strong>sponds<br />

to the symmetric low-pass mode, the dashed curve to the antisymmetric band-pass mode. The th<strong>in</strong><br />

l<strong>in</strong>es a<strong>re</strong> the l<strong>in</strong>ear approximations for low and high f<strong>re</strong>quencies. Note that the diagonal l<strong>in</strong>e marks<br />

the velocity of light, =K = 1. Thus the band-pass mode exhibits a phase velocity g<strong>re</strong>ater than c<br />

slightly above the lower f<strong>re</strong>quency limit. It is clear to see that the <strong>re</strong>lation vp vg = c 2 mentioned <strong>in</strong><br />

the p<strong>re</strong>vious chapter is not valid he<strong>re</strong>.<br />

with X(!) taken from (3.23). The proof of the identity ve = vg is then straightforward but<br />

tedious.<br />

3.3 Re ection due to term<strong>in</strong>ation mismatch<br />

If we want to exam<strong>in</strong>e evanescence, it is not very sensible to consider a steady-state excitation<br />

of the transmission l<strong>in</strong>e without <strong>re</strong> ection. In such a case, the characteristic impedance (3.5)<br />

as well as the wave number (3.4) a<strong>re</strong> imag<strong>in</strong>ary, <strong>re</strong>sult<strong>in</strong>g <strong>in</strong> total <strong>in</strong>ternal <strong>re</strong> ection, but no<br />

energy transport. The<strong>re</strong>fo<strong>re</strong> we deem it mean<strong>in</strong>gful to study the simple example of <strong>re</strong> ection<br />

due to an ohmic load. The transmission l<strong>in</strong>e is aga<strong>in</strong> described by the equivalent circuit of<br />

g. 3.1, but has now a nite length and a term<strong>in</strong>at<strong>in</strong>g <strong>re</strong>sistor ( g. 3.4).<br />

Befo<strong>re</strong> we start with evanescence, let us brie y <strong>in</strong>vestigate the behaviour of the pass band<br />

to ga<strong>in</strong> some physical <strong>in</strong>sight. If we solve the di e<strong>re</strong>ntial equations (3.2) for the boundary<br />

condition U2 = RI2 at the end of the l<strong>in</strong>e, we obta<strong>in</strong> the well-known spatial voltage and<br />

cur<strong>re</strong>nt evolution with <strong>re</strong>spect to the voltage across the load,<br />

U = U2 cos k(l , x)+j Z0<br />

R<br />

I = U2<br />

Z0<br />

Z0<br />

R<br />

s<strong>in</strong> k(l , x)<br />

cos k(l , x)+js<strong>in</strong> k(l , x) :<br />

37<br />

K<br />

(3.27)


I 0<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

Figu<strong>re</strong> 3.4: Transmission l<strong>in</strong>e with <strong>re</strong>al-valued term<strong>in</strong>ation<br />

With these <strong>re</strong>sults we can calculate the averaged propagated and sto<strong>re</strong>d energies<br />

P =Re UI<br />

2<br />

; W = UU<br />

4<br />

dB 0<br />

d!<br />

l<br />

+ II<br />

4<br />

which after some simple manipulations yields the velocity of energy transport<br />

ve =<br />

x<br />

dX 0<br />

d!<br />

R<br />

; (3.28)<br />

2k<br />

, 2 s<strong>in</strong> k(l , x)+ 2cos2 k(l , x) B0 dX0 d! + , cos2 k(l , x)+ 2s<strong>in</strong>2 k(l , x) X0 dB0 d!<br />

(3.29)<br />

with the term<strong>in</strong>ation factor = Z0=R. This complicated exp<strong>re</strong>ssion is <strong>re</strong>duced dramatically<br />

if we evaluate it for the special case of a dispersionless l<strong>in</strong>e with<br />

Then (3.29) becomes, with 1=c = p L 0 C 0 ,<br />

B 0 = !C 0 ; X 0 = !L 0 : (3.30)<br />

ve<br />

c<br />

= 2 Z0<br />

R<br />

1+ Z0<br />

R<br />

If we <strong>in</strong>troduce the <strong>re</strong> ection factor of the cur<strong>re</strong>nt<br />

r =<br />

we can <strong>re</strong>write (3.31) <strong>in</strong> a mo<strong>re</strong> plausible way,<br />

ve<br />

c<br />

Z0<br />

R<br />

Z0<br />

R<br />

, 1<br />

: (3.31)<br />

+1 ; (3.32)<br />

1 , r2<br />

= : (3.33)<br />

1+r2 This <strong>in</strong>vokes the pictu<strong>re</strong> of an overall transport velocity weighted by the power the waves<br />

carry: the <strong>in</strong>cident wave transports a certa<strong>in</strong> amount of energy per unit time, P0, whe<strong>re</strong>as the<br />

<strong>re</strong> ected wave carries a part r 2 P0 of it back. Apart from this <strong>in</strong>terp<strong>re</strong>tation, (3.33) mirrors<br />

exactly the basic idea of the energy velocity de nition: a net propagated energy per unit<br />

length divided by the total energy content of that unit length. We notice that the energy<br />

velocity isnot equal to the group velocity vg = c. In the limit of a matched term<strong>in</strong>ation r =0,<br />

however, we haveve=vg. Exp<strong>re</strong>ssion (3.33) also holds for an unterm<strong>in</strong>ated or short-circuited<br />

transmission l<strong>in</strong>e R = 1 or R = 0, thus r 2 = 1, whe<strong>re</strong> ve = 0 <strong>in</strong>dicates that an energy<br />

transport is impossible at all.<br />

38


3.4 A simple thought experiment<br />

I0<br />

Z0<br />

Figu<strong>re</strong> 3.5: Transmission l<strong>in</strong>e charged with a short <strong>in</strong>put cur<strong>re</strong>nt pulse.<br />

3.4 A simple thought experiment<br />

To judge whether (3.31) is <strong>in</strong>deed mean<strong>in</strong>gful, we explo<strong>re</strong> yet another example. Consider<br />

the term<strong>in</strong>ated transmission l<strong>in</strong>e like befo<strong>re</strong>, but with a pulse source rather than with a<br />

monochromatic excitation ( g. 3.5). Let the source emit a cur<strong>re</strong>nt pulse with a duration<br />

T l=c much shorter than the propagation time needed to <strong>re</strong>ach the end of the l<strong>in</strong>e. Then<br />

a total energy<br />

W1 = I0 2 TZ0<br />

l<br />

R<br />

(3.34)<br />

will be transmitted to the load after a propagation delay = l=c. It will, however, only partly<br />

be absorbed due to partial <strong>re</strong> ection. The <strong>re</strong> ected part will be enti<strong>re</strong>ly <strong>re</strong> ected <strong>in</strong> turn by<br />

the ideal <strong>in</strong> nite <strong>in</strong>ternal impedance of the source. The<strong>re</strong>fo<strong>re</strong>, the amount of energy that is<br />

by and by absorbed by the load can be exp<strong>re</strong>ssed by a series of contributions<br />

W (t) =I0 2 T(1 + r) 2 R , (t , )+r 2 (t,3 )+r 4 (t,5 )+::: (3.35)<br />

with the step function (t). The cur<strong>re</strong>nt (1 + r)I0 is the superposition of the <strong>in</strong>cident and<br />

<strong>re</strong> ected wave seen by the load <strong>re</strong>sistance. For t !1,the <strong>in</strong> nite series <strong>in</strong> (3.35) of course<br />

gives (3.34). The de nition of an energy velocity <strong>in</strong> this case is mo<strong>re</strong> or less arbitrary, but<br />

may <strong>in</strong> a sensible way be given by<br />

ve = l<br />

e<br />

; W ( e)<br />

W1<br />

1 , 1<br />

; (3.36)<br />

e<br />

whe<strong>re</strong> e is the <strong>in</strong>stance when a given portion of the total energy has been absorbed, the<br />

<strong>re</strong>ma<strong>in</strong>der be<strong>in</strong>g determ<strong>in</strong>ed by the base e of the natural logarithm. Thus the time-dependent<br />

function of (3.35),<br />

f(t) =<br />

1X<br />

i=0<br />

(t , (2i +1) )r 2i ; (3.37)<br />

is evaluated only to the <strong>in</strong>dex i = N whe<strong>re</strong> this condition is rst met. S<strong>in</strong>ce the summation<br />

starts with <strong>in</strong>dex i = 0, we have at this po<strong>in</strong>t N +1 terms <strong>in</strong>cluded. Know<strong>in</strong>g that<br />

limt!1 f(t) =f(1)=1=(1 , r 2 ) and with the summation rule for nite geometric series, we<br />

obta<strong>in</strong><br />

f( e)<br />

f(1) = W ( e)<br />

=1,r<br />

W1<br />

2(N+1) ; e =(2N+1) : (3.38)<br />

39


ve/c<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

2.5 5 7.5 10 12.5 15 17.5 20 Z0/R<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

ve/c<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

0<br />

0.2 0.4 0.6 0.8 1 Z0/R<br />

Figu<strong>re</strong> 3.6: Comparison of the energy velocities obta<strong>in</strong>ed for steady-state signals and short pulses.<br />

The energy velocity thus becomes<br />

ve<br />

c =<br />

1<br />

: (3.39)<br />

2N +1<br />

The <strong>re</strong>sults of (3.39) compa<strong>re</strong>d with the monochromatic case (3.31) a<strong>re</strong> shown <strong>in</strong> g. 3.6 .<br />

The cor<strong>re</strong>spondence is satisfactory for a <strong>re</strong>latively bad term<strong>in</strong>ation, i. e. when Z0=R 1 or<br />

Z0=R 1. If the term<strong>in</strong>ation is close to its optimum Z0=R 1, the energy transfer will<br />

surpass the limit (3.36) with the rst time the pulse <strong>re</strong>aches the <strong>re</strong>sistor.<br />

Remark (Relation between steady-state and pulse signal) The good cor<strong>re</strong>spondence<br />

between the curves <strong>in</strong> g. 3.6 is by no means fortuitous. It is rather due to the<br />

ca<strong>re</strong>ful choice of the energy limit <strong>in</strong> (3.36). In fact, the value 1=e for the still miss<strong>in</strong>g<br />

part of the transfer<strong>re</strong>d energy gives exactly the time constant of the cor<strong>re</strong>spond<strong>in</strong>g quasistationary<br />

model. In such an approach the e ect of the <strong>re</strong> ections is collapsed <strong>in</strong>to one<br />

s<strong>in</strong>gle parameter determ<strong>in</strong><strong>in</strong>g the overall shape of the output signal just like the charg<strong>in</strong>g<br />

of a capacitance. Thus the underly<strong>in</strong>g process of wave propagation completely disappears,<br />

and the curve is smooth, behav<strong>in</strong>g like 1,e ,t= e . The exact model that accounts for the<br />

wave propagation converges to the quasi-stationary model when the<strong>re</strong> a<strong>re</strong> many <strong>re</strong> ections<br />

or only small steps <strong>in</strong> the output signal, i. e. when the term<strong>in</strong>ation is far away from the<br />

optimum.<br />

The <strong>re</strong>ason<strong>in</strong>g just given is of course not the whole truth. After all, the smooth curve <strong>in</strong><br />

the gu<strong>re</strong> was drawn us<strong>in</strong>g (3.31), which was derived from a steady-state signal. The pulse<br />

of the example, however, can be seen as the superposition of monochromatic components.<br />

From Parseval's theo<strong>re</strong>m we know that<br />

Z 1<br />

,1<br />

ji(t)j 2 dt = 1<br />

2<br />

Z 1<br />

jI(!)j<br />

,1<br />

2 d! ; (3.40)<br />

which means that the energy propagation of the signal is determ<strong>in</strong>ed by the energy of its<br />

f<strong>re</strong>quency components. If these components travel with a f<strong>re</strong>quency-<strong>in</strong>dependent velocity<br />

ve as <strong>in</strong> our example, then so does the enti<strong>re</strong> energy of the pulse.<br />

40


3.5 A dispersive system: the lossless plasma<br />

3.5 A dispersive system: the lossless plasma<br />

After the above <strong>re</strong> ections to ensu<strong>re</strong> the plausibility of (3.29), we may apply it to dispersive<br />

and evanescent systems. Rema<strong>in</strong><strong>in</strong>g with our transmission l<strong>in</strong>e model, we assume it this time<br />

to be lled with a lossless plasma. This has no e ect on the wave type, but only on its<br />

propagation velocity. We still consider a TEM wave, and so the equivalent circuit is still<br />

valid, although c becomes f<strong>re</strong>quency-dependent. If we neglect the motion of ions and focus<br />

on the contribution of the electrons, the <strong>re</strong>lative dielectric constant is given by<br />

with the plasma f<strong>re</strong>quency [94]<br />

"r =1, !p<br />

!<br />

!p =<br />

s q 2 n0<br />

"0m<br />

2<br />

(3.41)<br />

: (3.42)<br />

In this formula, q is the elementary charge, n0 the electron density, m the electron mass and<br />

"0 the dielectric constant ofvacuum. Hence we have<br />

and the dispersion <strong>re</strong>lation<br />

X 0 =!L 0 ; B 0 = !C 0 1 , !p<br />

!<br />

k(!) = !p<br />

c<br />

s !<br />

!p<br />

2<br />

2<br />

; c = 1<br />

p L 0 C 0<br />

(3.43)<br />

,1: (3.44)<br />

We see immediately that the wavenumber becomes imag<strong>in</strong>ary <strong>in</strong> the evanescent <strong>re</strong>gion !


whe<strong>re</strong>as the energy velocity <strong>in</strong> the evanescent <strong>re</strong>gion is<br />

ve<br />

c =<br />

,1 , 2 + , !p<br />

!<br />

2 + , !p<br />

!<br />

If we chose a matched term<strong>in</strong>ation<br />

then (3.46) would <strong>re</strong>duce to<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

2<br />

, !p<br />

!<br />

2 ,1+ 2 + , !p<br />

!<br />

R = Z0 =<br />

ve<br />

c =<br />

r L 0<br />

s<br />

C 0<br />

2 , 1<br />

2 cosh<br />

1<br />

q 1 , , !p<br />

!<br />

1 , !p<br />

!<br />

2<br />

2<br />

2(X , L)<br />

r<br />

1 , !<br />

!p<br />

! : (3.47)<br />

2<br />

; (3.48)<br />

; (3.49)<br />

which is exactly the group velocity vg = d!=dk <strong>re</strong>fer<strong>re</strong>d to the speed of light. Numerical<br />

examples of the energy velocity <strong>in</strong> dependence on the position and the term<strong>in</strong>ation ratio a<strong>re</strong><br />

given <strong>in</strong> g. 3.7 for the pass band as well as <strong>in</strong> g. 3.8 for the evanescent <strong>re</strong>gion. It is worth<br />

notic<strong>in</strong>g that (3.46) and (3.47) a<strong>re</strong> cont<strong>in</strong>uous at the plasma f<strong>re</strong>quency. They do not exhibit<br />

a zero, but the nite value<br />

ve(!p)<br />

c<br />

= 2<br />

2+<br />

2 : (3.50)<br />

The gu<strong>re</strong>s clearly show that the energy velocity depends on the position along the transmission<br />

l<strong>in</strong>e. As could have been expected, the local velocity <strong>in</strong> the pass band oscillates<br />

uniformly. In the stop band, however, the velocity nearly vanishes almost everywhe<strong>re</strong>, but<br />

<strong>in</strong>c<strong>re</strong>ases sharply towards the end of the l<strong>in</strong>e with<strong>in</strong> a characteristic distance that is practically<br />

<strong>in</strong>dependent of the signal f<strong>re</strong>quency. This behaviour stems from the fact that the sto<strong>re</strong>d<br />

energy per unit length <strong>in</strong> the transmission l<strong>in</strong>e decays exponentially, whe<strong>re</strong>as the transmitted<br />

energy <strong>re</strong>ma<strong>in</strong>s constant due to the absence of losses. Consequently, the energy velocity must<br />

rise as the wave approaches the end of the l<strong>in</strong>e.<br />

From a macroscopic po<strong>in</strong>t of view, the local energy velocity is of only m<strong>in</strong>or importance |<br />

an overall energy propagation time through the transmission l<strong>in</strong>e is much mo<strong>re</strong> <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g.<br />

This can be de ned via<br />

e =<br />

Z l<br />

and we obta<strong>in</strong> an `e ective' velocity like <strong>in</strong> (3.36),<br />

0<br />

1<br />

ve<br />

ve;e = l<br />

42<br />

e<br />

dx ; (3.51)<br />

: (3.52)


3.5 A dispersive system: the lossless plasma<br />

0<br />

0.66<br />

ve/c 0.4 .4<br />

0.2 .2<br />

0<br />

100<br />

2<br />

4<br />

X<br />

6<br />

10<br />

8<br />

10<br />

1<br />

ρ<br />

0.1<br />

0.01<br />

Figu<strong>re</strong> 3.7: Evaluation of the energy velocity (3.46) <strong>in</strong> the pass band of a lossless plasma depend<strong>in</strong>g<br />

on the spatial coord<strong>in</strong>ate X and the term<strong>in</strong>ation factor . The length of the l<strong>in</strong>e is L = 10, the signal<br />

f<strong>re</strong>quency is given by !p=! =0:8. Note the partially logarithmic scale.<br />

0<br />

100<br />

2<br />

10<br />

4<br />

X<br />

6<br />

1<br />

ρ<br />

8<br />

10<br />

0.1<br />

0.01<br />

0.2<br />

0<br />

0.6<br />

0.4<br />

ve/c<br />

Figu<strong>re</strong> 3.8: Evaluation of the energy velocity (3.47) <strong>in</strong> the evanescent <strong>re</strong>gion of a lossless plasma<br />

depend<strong>in</strong>g on the spatial coord<strong>in</strong>ate X and the term<strong>in</strong>ation factor . The parameters a<strong>re</strong> L = 10 and<br />

!p=! =1:25 . Note the partially logarithmic scale.<br />

43


Straightforward <strong>in</strong>tegration yields for the pass band<br />

ve;e<br />

c =<br />

and for the stop band<br />

ve;e<br />

c =<br />

1+ 2 , , !p<br />

!<br />

,1 , 2 + , !p<br />

!<br />

2 + , !p<br />

!<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

2 + , !p<br />

!<br />

2<br />

2 1 , , !p<br />

!<br />

2<br />

2 1, 2 , , !p<br />

!<br />

, !p<br />

!<br />

2 , 1<br />

2 ,1+ 2 + , !p<br />

!<br />

2<br />

2<br />

r<br />

s<strong>in</strong> 2L<br />

!!p<br />

r<br />

2L<br />

!!p<br />

r<br />

s<strong>in</strong>h 2L 1, !<br />

!p r 2<br />

2L 1, !<br />

!p<br />

2<br />

,1<br />

2<br />

,1<br />

2<br />

(3.53)<br />

: (3.54)<br />

These velocities a<strong>re</strong> depicted <strong>in</strong> g. 3.9 and g. 3.10, <strong>re</strong>spectively. We see that if the l<strong>in</strong>e is<br />

term<strong>in</strong>ated with the characteristic impedance, = p 1 , (!p=!) 2 , the energy velocity <strong>in</strong> the<br />

pass band is <strong>in</strong>dependent ofLand equal to the group velocity. Apart from very short l<strong>in</strong>es,<br />

the velocity is virtually <strong>in</strong>dependent of the length.<br />

Remark (Velocity maximum) It is <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g to notice that for short l<strong>in</strong>es, the<br />

maximum velocity of energy propagation does not occur when the term<strong>in</strong>ation is optimal,<br />

but for larger values of . For the example <strong>in</strong> g. 3.9, the value would be m 1:28 <strong>in</strong><br />

the limit L ! 0. It is not clear whe<strong>re</strong> this behaviour comes from.<br />

A similar e ect appears for evanescence. Although a matched term<strong>in</strong>ation is not de ned<br />

<strong>in</strong> this case, one might expect the velocity maximum to be <strong>re</strong>lated to the propagat<strong>in</strong>g<br />

mode. In fact, for L ! 0 the value whe<strong>re</strong> the optimum is <strong>re</strong>ached <strong>in</strong> g. 3.10 is m 1:6,<br />

which has noth<strong>in</strong>g p whatsoever to do with any other parameter. For large values of !p=!,<br />

we nd m !p=!.<br />

The <strong>re</strong>sult for the stop band ( g. 3.10) seems very plausible. Evanescence slows down the<br />

energy transfer, and the energy velocity gradually dim<strong>in</strong>ishes. At any rate, energy transport<br />

through an evanescent <strong>re</strong>gion is possible even <strong>in</strong> the strictly monochromatic case, when we<br />

can guarantee that the spectrum of the signal is con ned to the stop band | unlike signals<br />

with a broader spectral width, whe<strong>re</strong> we must always worry about components ly<strong>in</strong>g <strong>in</strong> the<br />

pass band and thus distort<strong>in</strong>g the <strong>re</strong>sult. On the other hand, no energy can be transfer<strong>re</strong>d if<br />

the l<strong>in</strong>e is open ( ! 0) or shorted ( !1), which naturally <strong>re</strong> ects only our p<strong>re</strong>condition<br />

that the energy is to be dissipated at the term<strong>in</strong>ation <strong>re</strong>sistance.<br />

Remark (Formal equivalence to a wave guide) The <strong>re</strong>sults obta<strong>in</strong>ed for the transmission<br />

l<strong>in</strong>e with a lossless plasma a<strong>re</strong> formally identical with the H01 mode of a <strong>re</strong>ctangular<br />

wave guide with width b. The <strong>re</strong>spective boundary conditions a<strong>re</strong> that the excitation<br />

at x = 0 is monomodal and that at x = l,<br />

, Ez(l)<br />

= R; (3.55)<br />

Hy(l)<br />

when x is the di<strong>re</strong>ction of propagation. The <strong>re</strong>sults from (3.46) onwards then apply with<br />

the <strong>re</strong>placements !p 7! c=b and p L 0 =C 0 7! p =".<br />

44


3.5 A dispersive system: the lossless plasma<br />

0<br />

L<br />

2<br />

4<br />

6<br />

8<br />

10<br />

0.01<br />

0.1<br />

1<br />

ρ<br />

10<br />

0<br />

100<br />

0.6 0<br />

0.4 0. ve/c<br />

Figu<strong>re</strong> 3.9: Evaluation of the `e ective' energy velocity (3.53) <strong>in</strong> the pass band of a lossless plasma<br />

depend<strong>in</strong>g on length L and the term<strong>in</strong>ation factor . The signal f<strong>re</strong>quency is !p=! =0:8. Note the<br />

partially logarithmic scale.<br />

0.01<br />

0.6<br />

0.4<br />

ve/c<br />

0.2<br />

0<br />

0<br />

0.1<br />

2<br />

ρ<br />

1<br />

4<br />

10<br />

L<br />

100<br />

6<br />

Figu<strong>re</strong> 3.10: Evaluation of the `e ective' energy velocity (3.54) <strong>in</strong> the evanescent <strong>re</strong>gion of a lossless<br />

plasma depend<strong>in</strong>g on the length L and the term<strong>in</strong>ation factor for !p=! =1:25. Note the partially<br />

logarithmic scale.<br />

45<br />

8<br />

10<br />

0.2


3.6 Inhomogeneous transmission l<strong>in</strong>e<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

So far, the transmission l<strong>in</strong>es we <strong>in</strong>vestigated we<strong>re</strong> homogeneous, so that their parameters did<br />

not depend on the position along the l<strong>in</strong>e. We now extend our analysis to <strong>in</strong>homogeneous<br />

transmission l<strong>in</strong>es and brie y sketch the t<strong>re</strong>atment of a so-called exponential l<strong>in</strong>e. Such<br />

transmission l<strong>in</strong>es a<strong>re</strong> not only of academic <strong>in</strong>te<strong>re</strong>st | they have actually been used <strong>in</strong> the early<br />

days of <strong>in</strong>tegrated circuit design for impedance transformation [95]. The equivalent circuit<br />

of g. 3.1 and the <strong>re</strong>spective di e<strong>re</strong>ntial equations (3.2) a<strong>re</strong> still valid, but the distributed<br />

<strong>re</strong>actance and susceptance a<strong>re</strong> now position-dependent,<br />

X 0 = !L0 0 e x ; B 0 = !C0 0 e , x ; c =<br />

1<br />

p<br />

L0 0 : (3.56)<br />

0<br />

C0<br />

Insert<strong>in</strong>g solutions of the form (3.3) <strong>in</strong>to the di e<strong>re</strong>ntial equations yields a propagation constant<br />

0<br />

k = @j<br />

2<br />

s<br />

2!<br />

c<br />

1<br />

2<br />

, 1A<br />

(3.57)<br />

and a complex, position-dependent wave impedance<br />

Z 0 =<br />

0<br />

@,j 0 +<br />

2!C0<br />

With the abb<strong>re</strong>viations and normalisations<br />

= 2!<br />

c<br />

s L0 0<br />

C0 0 , 2!C0 0<br />

1<br />

2<br />

A e x : (3.58)<br />

; X = x ; L = l ; (3.59)<br />

the general solution <strong>in</strong> the pass band consists of a right- and left-go<strong>in</strong>g wave for the voltage<br />

and a cor<strong>re</strong>spond<strong>in</strong>g cur<strong>re</strong>nt<br />

U = K e X 2 (1,j p 2 ,1)+j!t +(U0 ,K)e X 2(1+j p 2 ,1)+j!t<br />

I = j2!C0 0 K=<br />

1+j p e<br />

2 ,1 ,X 2(1+j p 2 ,1)+j!t j2!C0<br />

+ 0 (U0 , K)=<br />

1 , j p 2 , 1<br />

e , X 2 (1,j p 2 ,1)+j!t<br />

(3.60)<br />

(3.61)<br />

with a complex constant K that must be determ<strong>in</strong>ed to comply with the boundary conditions.<br />

With the de nitions of the propagated energy,<br />

and sto<strong>re</strong>d energy,<br />

W = 1<br />

4<br />

P = 1<br />

Re U I ; (3.62)<br />

2<br />

I I dX0<br />

d!<br />

46<br />

+ U U dB0<br />

d!<br />

; (3.63)


3.6 Inhomogeneous transmission l<strong>in</strong>e<br />

0<br />

1<br />

0.75 75<br />

ve/c<br />

0.5 .5<br />

0.25 25<br />

0<br />

100<br />

2<br />

X<br />

4<br />

10<br />

6<br />

8<br />

1<br />

ρ<br />

0.1<br />

0.01<br />

Figu<strong>re</strong> 3.11: Evaluation of the energy velocity (3.65) <strong>in</strong> the pass band of an exponential l<strong>in</strong>e depend<strong>in</strong>g<br />

on the spatial coord<strong>in</strong>ate X and the term<strong>in</strong>ation factor . The length of the l<strong>in</strong>e is L = 8, the signal<br />

f<strong>re</strong>quency is given by 1= =0:5. Note the partially logarithmic scale.<br />

we nd for the special case of a matched term<strong>in</strong>ation the energy velocity<br />

ve<br />

c =<br />

p<br />

2 , 1<br />

; (3.64)<br />

which is exactly the group velocity we would have obta<strong>in</strong>ed from the dispersion <strong>re</strong>lation.<br />

If we can manage to term<strong>in</strong>ate the transmission l<strong>in</strong>e with a <strong>re</strong>al-valued multiple of the<br />

characteristic impedance Z0, the energy velocity with<strong>in</strong> a l<strong>in</strong>e of length l <strong>in</strong> the pass band,<br />

2 > 1, becomes after a lengthy calculation<br />

q<br />

( 2 3<br />

, 1)<br />

ve<br />

c =<br />

2<br />

2 (1 + 2 ) , 2 , ( , 1) 2 cos e X , ( 2 , 1) s<strong>in</strong> e X ; (3.65)<br />

with the abb<strong>re</strong>viations = p 2 , 1 and e X = X , L. In the stop band, whe<strong>re</strong> 2 < 1,<br />

a term<strong>in</strong>ation with the wave impedance makes no sense because no energy could then be<br />

transmitted due to evanescence, so we assume a pu<strong>re</strong>ly ohmic term<strong>in</strong>ation R. With the<br />

sett<strong>in</strong>gs = p 1 , 2 and r = R= p L0 0 =C0 0 ,we obta<strong>in</strong> the cor<strong>re</strong>spond<strong>in</strong>g energy velocity<br />

ve<br />

c =<br />

r , 1 , 2<br />

(cosh e X + s<strong>in</strong>h e X , 2 )e L + r 2 (cosh e X , s<strong>in</strong>h e X , 2 )e ,L<br />

47<br />

: (3.66)


0<br />

100000<br />

2<br />

10000<br />

X<br />

4<br />

1000<br />

6<br />

r<br />

8<br />

100<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

10<br />

1<br />

0.2<br />

0<br />

0.4<br />

ve/c<br />

Figu<strong>re</strong> 3.12: Evaluation of the energy velocity (3.66) <strong>in</strong> the stop band of an exponential l<strong>in</strong>e depend<strong>in</strong>g<br />

on the spatial coord<strong>in</strong>ate X and the term<strong>in</strong>ation factor r. The length of the l<strong>in</strong>e is L = 8, the signal<br />

f<strong>re</strong>quency is given by =0:5. Note the partially logarithmic scale.<br />

The energy velocity <strong>in</strong> the pass band is depicted <strong>in</strong> g. 3.11. Its dependence on the position<br />

along the l<strong>in</strong>e and the term<strong>in</strong>ation is very similar to that <strong>in</strong> a lossless plasma ( g. 3.7). Except<br />

for matched term<strong>in</strong>ation, = 1, spatial harmonics appear <strong>in</strong> the local velocity. The situation<br />

<strong>in</strong> the stop band, however, is a bit di e<strong>re</strong>nt, as can be seen by compar<strong>in</strong>g g. 3.12 and g. 3.8.<br />

While the local velocity still <strong>in</strong>c<strong>re</strong>ases towards the end of the l<strong>in</strong>e, its maximum | at least <strong>in</strong><br />

a broad range of term<strong>in</strong>ation factors | is no longer located di<strong>re</strong>ctly at x = l but signi cantly<br />

ahead of this po<strong>in</strong>t. Note that <strong>in</strong> this case, r = 1 has noth<strong>in</strong>g at all to do with matched<br />

term<strong>in</strong>ation. Like <strong>in</strong> the p<strong>re</strong>vious example, we could elim<strong>in</strong>ate the position dependence by<br />

<strong>in</strong>troduc<strong>in</strong>g the e ective velocity (3.51). However, this provides little additional <strong>in</strong>sight and<br />

has the<strong>re</strong>fo<strong>re</strong> been omitted.<br />

3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

In the examples <strong>in</strong>vestigated so far we <strong>re</strong>stricted ourselves p<strong>re</strong>dom<strong>in</strong>antly to monochromatic<br />

waves, which is an <strong>in</strong>structive special case and easy to analyse. On the other hand, we usually<br />

apply waves to transmit signals, and as a strictly monochromatic wave cannot carry any<br />

<strong>in</strong>formation, this case has no practical <strong>re</strong>levance. What we need <strong>in</strong>stead, is a modulation of the<br />

wave to form pulses we can identify with the <strong>in</strong>formation. A simple way to do so is to switch<br />

48


3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

a carrier oscillation on and o , which is <strong>in</strong> fact a crude amplitude modulation also known as<br />

on-o -key<strong>in</strong>g. Such apulse can then transport one bit of <strong>in</strong>formation. Ow<strong>in</strong>g to its sharply<br />

limited duration, however, its spectrum now spans a broad range of f<strong>re</strong>quencies. Ir<strong>re</strong>spective<br />

of any measu<strong>re</strong>s taken to limit the spectrum, it still widens as the pulse is shortened. Hence<br />

the narrow-band approximation quickly becomes senseless.<br />

Remark (Relation between spectral and pulse width) Like <strong>in</strong> quantum mechanics,<br />

an <strong>in</strong>equality can be given that establishes a lower bound for the product of a pulse<br />

p x 2 , x 2<br />

duration t and the width of the dom<strong>in</strong>ant peak of its spectrum !. If x =<br />

denotes the variance of the <strong>re</strong>spective <strong>in</strong>tensities (i. e. the squa<strong>re</strong>d signal value and the<br />

power spectrum), then this uncerta<strong>in</strong>ty pr<strong>in</strong>ciple <strong>re</strong>ads<br />

holds for Gaussian distributions [11].<br />

t ! 1=2. The equal sign<br />

We now turn to a fairly classical problem: the propagation of a pulsed wave <strong>in</strong> a dispersive<br />

medium. This has been scrut<strong>in</strong>ised over and over aga<strong>in</strong>, beg<strong>in</strong>n<strong>in</strong>g with Sommerfeld [12] and<br />

Brillou<strong>in</strong> [19]. Yet we shall add a new facet to the plethora of <strong>re</strong>sults.<br />

The model we exam<strong>in</strong>e is aga<strong>in</strong> that of a lossless plasma like <strong>in</strong> the p<strong>re</strong>vious section. But<br />

this time we leave it unbounded for x>0 so that we need not bother about <strong>re</strong> ections. We<br />

also adhe<strong>re</strong> to the transmission l<strong>in</strong>e model of g. 3.1 . In pr<strong>in</strong>ciple we could as well calculate<br />

the electric and magnetic elds <strong>in</strong> the plasma di<strong>re</strong>ctly, but as we a<strong>re</strong> concerned only with the<br />

basic TEM wave, we p<strong>re</strong>fer to use the convenient and proven equivalent circuit. The voltage<br />

then cor<strong>re</strong>sponds for <strong>in</strong>stance to the electric eld vector Ey and the cur<strong>re</strong>nt to the magnetic<br />

component Hz, ifxis the di<strong>re</strong>ction of propagation. All other components do not exist.<br />

The plasma shall be excited by a monochromatic cur<strong>re</strong>nt source be<strong>in</strong>g switched on at t =0<br />

and turned o an <strong>in</strong>teger multiple of periods later, at t = =2n ,<br />

I(0;t)=<br />

8<br />

><<br />

>:<br />

0 if t<br />

: (3.67)<br />

The latter <strong>re</strong>qui<strong>re</strong>ment entails no loss of generality but only facilitates the analytical t<strong>re</strong>atment.<br />

It is evident that due to the l<strong>in</strong>earity of the model, the pulse can also be described as<br />

superposition of two delayed step functions of opposite sign,<br />

I(0;t)=I0 (t) cos !0t , I0 (t , ) cos !0t ; (3.68)<br />

such that we may <strong>re</strong>strict our analysis to the step <strong>re</strong>sponse of the system. Similar <strong>in</strong>vestigations<br />

we<strong>re</strong> carried out for example by Knop [96] or Haskell and Case [97]. In contrast to<br />

the approach pursued he<strong>re</strong>, they used modulated s<strong>in</strong>e waves, whe<strong>re</strong>as we consider a cos<strong>in</strong>e<br />

wave, which has the convenient side bene t that the wave front of the signal can be identi ed<br />

without any doubt.<br />

Remark (A glance at the literatu<strong>re</strong>) The orig<strong>in</strong>al articles of Sommerfeld and Brillou<strong>in</strong><br />

we<strong>re</strong> concerned with the propagation of light through a <strong>re</strong>sonant Lo<strong>re</strong>ntz-type medium<br />

49


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

and had been of pu<strong>re</strong>ly academic <strong>in</strong>te<strong>re</strong>st at that time, as Brillou<strong>in</strong> himself admitted [2].<br />

The work on wave propagation <strong>in</strong> plasma, however, was stimulated by the <strong>re</strong>search on the<br />

properties of the ionosphe<strong>re</strong>, which plays an important role <strong>in</strong> the transmission of radio<br />

signals. A survey of this work and the dispersive e ects <strong>in</strong> the ionosphe<strong>re</strong> was given by<br />

McIntosh and Malaga [98].<br />

In the p<strong>re</strong>vious chapter we mentioned microwave experiments that we<strong>re</strong> thought to prove<br />

superlum<strong>in</strong>al wave propagation <strong>in</strong> the evanescent <strong>re</strong>gion of a wave guide or plasma that<br />

exhibits the same dispersion properties. As a follow-up to this discussion we choose the<br />

f<strong>re</strong>quency of our signal source to be below the cuto f<strong>re</strong>quency of the plasma. Of course the<br />

spectrum of the pulse will extend <strong>in</strong>to the pass band, however the spectral peak about the<br />

carrier f<strong>re</strong>quency can be con ned to the stop band.<br />

At rst, we seek the steady-state solution of the wave. With a strictly monochromatic source<br />

I(0;t)=I0cos !0t, wewould obta<strong>in</strong> a cur<strong>re</strong>nt<br />

whe<strong>re</strong> 0 is the imag<strong>in</strong>ary part of the wave number at !0,<br />

Is<br />

I0<br />

0 = !p<br />

c<br />

= e , 0x cos !0t ; (3.69)<br />

s<br />

1 , !0<br />

!p<br />

2<br />

: (3.70)<br />

Bear<strong>in</strong>g <strong>in</strong> m<strong>in</strong>d that the characteristic impedance is imag<strong>in</strong>ary, we <strong>re</strong>adily obta<strong>in</strong> the voltage<br />

from U = Z0I with the <strong>re</strong>lation Re je j!t = , s<strong>in</strong> !t,<br />

Us p<br />

I0 L0 =C0 = ,<br />

e , 0x<br />

r<br />

2<br />

!p<br />

!0<br />

, 1<br />

s<strong>in</strong> !0t : (3.71)<br />

If the signal source is never turned o , then the complete exp<strong>re</strong>ssions of voltage and cur<strong>re</strong>nt<br />

must converge to this steady-state solution for t !1. To get the e ect of turn-on, we need<br />

additional transient solutions such that the wave complies with the <strong>in</strong>itial conditions for t =0<br />

I(x; 0) = 0 ; U(x; 0) = 0 ; x>0: (3.72)<br />

The voltage <strong>in</strong> (3.71) al<strong>re</strong>ady satis es this <strong>re</strong>qui<strong>re</strong>ment, the cur<strong>re</strong>nt does not. Hence we <strong>re</strong>sort<br />

to a little trick: we add a function that cancels the cur<strong>re</strong>nt everywhe<strong>re</strong> except for the orig<strong>in</strong><br />

x = 0, whe<strong>re</strong> the wave front I0 emerges at t = 0. We construct this desi<strong>re</strong>d function by<br />

expand<strong>in</strong>g the exponential function ,e , 0x antisymmetrically <strong>in</strong>to the negative half space<br />

x


3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

Each of the partial waves that make up the function has a dist<strong>in</strong>ct f<strong>re</strong>quency, just like the<br />

steady-state solution (3.69). We can thus immediately write down the transient part of the<br />

cur<strong>re</strong>nt,<br />

It<br />

= , 2 Z 1<br />

s<strong>in</strong> x cos !( )t<br />

d : (3.74)<br />

I0<br />

0<br />

2 + 0 2<br />

Note that <strong>in</strong> (3.74) we used a spatial decomposition of the transient solution and not a<br />

temporal one as might have been expected. This is the second signi cant di e<strong>re</strong>nce of the<br />

p<strong>re</strong>sent approach as compa<strong>re</strong>d with those known from the literatu<strong>re</strong>. All available publications<br />

start from the calculation of the f<strong>re</strong>quency spectrum of the <strong>in</strong>itial conditions. By contrast,<br />

we have determ<strong>in</strong>ed the wave number spectrum of the boundary conditions. Whether one<br />

chooses one possibility or the other is essentially a matter of op<strong>in</strong>ion. In our special case, the<br />

wave number method was particularly appeal<strong>in</strong>g because the transient solution could <strong>re</strong>adily<br />

be found.<br />

We still need the transient solution of the voltage. To that end we <strong>re</strong>call the di e<strong>re</strong>ntial<br />

equations that <strong>re</strong>late cur<strong>re</strong>nt and voltage <strong>in</strong> the evanescent <strong>re</strong>gion,<br />

@U<br />

@x<br />

= ,L0@I<br />

@t<br />

@I !p<br />

= C0<br />

@x<br />

If we <strong>in</strong>sert (3.74) <strong>in</strong> the rst one, we obta<strong>in</strong><br />

Ut<br />

U0L0 = 2 Z 1<br />

0<br />

!<br />

2<br />

, 1 @U<br />

@t :<br />

(3.75)<br />

!( ) cos x s<strong>in</strong> !( )t<br />

d : (3.76)<br />

2 + 0 2<br />

We <strong>re</strong>cognise that the transient voltage also complies with the boundary conditions (3.72)<br />

and vanishes everywhe<strong>re</strong> for t =0. The dispersion <strong>re</strong>lation !( ) used <strong>in</strong> (3.74) and (3.76) is<br />

q<br />

!( )= !p 2 +( c) 2 ; (3.77)<br />

which is <strong>re</strong>adily obta<strong>in</strong>ed by <strong>in</strong>vert<strong>in</strong>g (3.44).<br />

We havenow collected all components of the wave and can put together the complete solution.<br />

For the numerical evaluation of the wave functions, however, it is advisable to dispense with<br />

the small physical constants and to use scaled variables. To this end we <strong>in</strong>troduce a normalised<br />

<strong>in</strong>tegration variable<br />

= c<br />

!p<br />

(3.78)<br />

and scale the space and time coord<strong>in</strong>ates to the wavelength and period of the plasma f<strong>re</strong>quency,<br />

T = !pt (3.79)<br />

X = !px<br />

c<br />

= !0<br />

!p<br />

51<br />

(3.80)<br />

: (3.81)


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

After some simple <strong>re</strong>arrangements, we nally obta<strong>in</strong> the complete exp<strong>re</strong>ssion for the cur<strong>re</strong>nt<br />

along the l<strong>in</strong>e<br />

I<br />

I0<br />

Likewise, we nd the voltage<br />

I0<br />

= e ,Xp1, 2<br />

cos T ,<br />

, 2 Z 1<br />

0<br />

s<strong>in</strong> X cos T p 1+ 2<br />

1+ 2 , 2<br />

U<br />

p<br />

L0 =C0 = ,e,Xp 1, 2 1<br />

p s<strong>in</strong> T +<br />

1= 2 , 1<br />

+ 2 Z 1<br />

0<br />

d :<br />

p 1+ 2 cos X s<strong>in</strong> T p 1+ 2<br />

1+ 2 , 2<br />

d :<br />

(3.82)<br />

(3.83)<br />

Befo<strong>re</strong> giv<strong>in</strong>g numerical evaluations of the above exp<strong>re</strong>ssions, we take a look at the their<br />

range of validity. The lower end of the evanescence <strong>re</strong>gion is !0 = 0, i. e. a DC excitation. It<br />

can be <strong>re</strong>adily seen by sett<strong>in</strong>g =0that this raises no particular problem, and the special<br />

case of a non-oscillat<strong>in</strong>g step source is cove<strong>re</strong>d by our model as we could have expected.<br />

Th<strong>in</strong>gs a<strong>re</strong> di e<strong>re</strong>nt for a source oscillat<strong>in</strong>g with the plasma f<strong>re</strong>quency !p, which is exactly<br />

the boundary between evanescent and propagation mode. The steady-state part of the cur<strong>re</strong>nt<br />

(3.69) <strong>re</strong>duces to the unattenuated oscillation cos !pt, and for t =0,we nd with the identity<br />

sign x = 2 R 1 s<strong>in</strong> x<br />

d that the cur<strong>re</strong>nt still complies with the boundary conditions. The<br />

0<br />

voltage, however, does not, because the steady-state solution (3.71) becomes unbounded.<br />

The deeper <strong>re</strong>ason for this is our assumption of a lossless plasma, which entails a pole of the<br />

characteristic impedance at ! = !p. We could have circumvented this particular problem<br />

by choos<strong>in</strong>g a voltage <strong>in</strong>stead of a cur<strong>re</strong>nt source, the<strong>re</strong>by forc<strong>in</strong>g the cur<strong>re</strong>nt to vanish at<br />

<strong>re</strong>sonance.<br />

Numerical evaluations of (3.82) and (3.83) a<strong>re</strong> shown <strong>in</strong> g. 3.13 for the DC case = 0, g. 3.14<br />

for = 0:2 , and g. 3.15 for = 0:8. The most strik<strong>in</strong>g property of the <strong>re</strong>sult is that the wave<br />

front <strong>in</strong>deed goes straight through the medium with a velocity ofX=T = 1 or, exp<strong>re</strong>ssed with<br />

the unscaled variables, x=t = c. Thus the plasma is completely at <strong>re</strong>st befo<strong>re</strong> the disturbance<br />

arrives with the speed of light. This nd<strong>in</strong>g is, <strong>in</strong> pr<strong>in</strong>ciple, noth<strong>in</strong>g spectacular <strong>in</strong> that it<br />

only con rms the mo<strong>re</strong> general <strong>re</strong>sults obta<strong>in</strong>ed by Sommerfeld. It emphasises, however, that<br />

even <strong>in</strong> the evanescent <strong>re</strong>gion not the fa<strong>in</strong>test evidence of superlum<strong>in</strong>al wave propagation can<br />

be found.<br />

Apart from the wave front, the<strong>re</strong> a<strong>re</strong> other similarities between the th<strong>re</strong>e cases as well. The<br />

<strong>in</strong> uence of the forced oscillation applied by the signal source dim<strong>in</strong>ishes with the attenuation<br />

distance 1= 0 = c p 1<br />

, so that further down the l<strong>in</strong>e, the wave is determ<strong>in</strong>ed exclusively<br />

!p 1, 2<br />

by the transient solutions (3.74) and (3.76). Note that the transients consist solely of f<strong>re</strong>quency<br />

components above the plasma f<strong>re</strong>quency. The evanescent <strong>re</strong>gion does not come <strong>in</strong>to play he<strong>re</strong>,<br />

which is of course me<strong>re</strong>ly a mathematical e ect and bears no physical signi cance. The<br />

f<strong>re</strong>e oscillations of the plasma far away from both the entrance (as for the cur<strong>re</strong>nt) and<br />

52


3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

30<br />

30<br />

0<br />

20<br />

0<br />

20<br />

10<br />

X<br />

10<br />

X<br />

T<br />

10<br />

T<br />

10<br />

20<br />

20<br />

30<br />

1<br />

0<br />

0<br />

0.5<br />

I/I0<br />

0<br />

30<br />

1<br />

0.5<br />

U/U0<br />

Figu<strong>re</strong> 3.13: Evolution of the cur<strong>re</strong>nt (3.82) and voltage (3.83) along a one-dimensional transmission<br />

l<strong>in</strong>e with a lossless plasma for a non-oscillat<strong>in</strong>g step as <strong>in</strong>put cur<strong>re</strong>nt ( = 0).<br />

53<br />

0


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

the wave front a<strong>re</strong> dom<strong>in</strong>ated by the plasma f<strong>re</strong>quency, which aga<strong>in</strong> is not surpris<strong>in</strong>g. This<br />

becomes most clear by <strong>in</strong>spection of the voltage at the source at low signal f<strong>re</strong>quencies,<br />

whe<strong>re</strong> the <strong>in</strong> nitely high source impedance permits U to oscillate f<strong>re</strong>ely. It shows the<strong>re</strong>fo<strong>re</strong><br />

a superposition of the high <strong>re</strong>sonance f<strong>re</strong>quency and a component determ<strong>in</strong>ed by the signal<br />

( gs. 3.13 and 3.14).<br />

Remark (Dispersion e ects) That the plasma f<strong>re</strong>quency ultimately dom<strong>in</strong>ates the<br />

f<strong>re</strong>e oscillations is not at all unexpected. The <strong>re</strong>sonance f<strong>re</strong>quency has the smallest group<br />

velocity vg = d!=d (note that we may safely use this de nition he<strong>re</strong> s<strong>in</strong>ce the wave<br />

number k = + j is pu<strong>re</strong>ly <strong>re</strong>al <strong>in</strong> the non-evanescent mode). The fastest components,<br />

on the other hand, a<strong>re</strong> the high-f<strong>re</strong>quent ones that constitute the sharp wave front. They<br />

rush through, and the slow components around !p <strong>re</strong>ma<strong>in</strong>. That the local f<strong>re</strong>quency<br />

equals exactly the plasma f<strong>re</strong>quency holds of course only <strong>in</strong> the limit t !1.<br />

The local wavelength exhibits a complementary e ect. If we take a snapshot of the wave<br />

along the l<strong>in</strong>e at some given time, we nd that a long way beh<strong>in</strong>d the wave front, the<br />

oscillation gradually fades away. He<strong>re</strong>, it is the wavenumber that tends to zero as ! ! !p,<br />

and thus the local wavelength grows <strong>in</strong> nitely.<br />

The explanations given so far have been of a rather mathematical natu<strong>re</strong>. The<strong>re</strong> is, however,<br />

also a physical <strong>in</strong>terp<strong>re</strong>tation why the wave front travels undistorted through the medium.<br />

The electrons <strong>in</strong> the plasma have a nite mass and consequently a certa<strong>in</strong> amount of <strong>in</strong>ertia.<br />

Hence they cannot <strong>re</strong>act to the arriv<strong>in</strong>g wave front <strong>in</strong>stantaneously, but only with a short<br />

delay, and the wave front at the very <strong>in</strong>stant t = x=c <strong>re</strong>ma<strong>in</strong>s una ected. This explanation<br />

was al<strong>re</strong>ady given by Sommerfeld, who <strong>in</strong> turn owed it to Voigt [2, 12].<br />

Remark (Confession of a graphical <strong>re</strong>touch) In the gu<strong>re</strong>s, the values of the waves<br />

at X = T have been set to unity. This is, admittedly, not the cor<strong>re</strong>ct numerical <strong>re</strong>sult<br />

and has been <strong>in</strong>troduced deliberatly to illustrate the actual height of the wave front. The<br />

true value, as dictated by Fourier, naturally is 0:5 , and 1 is only the left-sided limit, as<br />

will be demonstrated below.<br />

The space-time evolution diagrams have a sampl<strong>in</strong>g grid that has been chosen su ciently<br />

coarse to keep the computational e ort at a <strong>re</strong>asonable value. This impai<strong>re</strong>d particularly the<br />

a<strong>re</strong>a about the wave front, which we exam<strong>in</strong>e mo<strong>re</strong> closely now. A <strong>re</strong> ned computation of the<br />

wave front shows that it <strong>re</strong>ma<strong>in</strong>s <strong>in</strong>deed unalte<strong>re</strong>d even for large values of X and T , as depicted<br />

<strong>in</strong> g. 3.16 for the cur<strong>re</strong>nt and g. 3.17 for the voltage. The graphs show the spatial evolution<br />

of the wave for di e<strong>re</strong>nt <strong>in</strong>stances <strong>in</strong> time and <strong>re</strong>veal two <strong>re</strong>markable featu<strong>re</strong>s. First, voltage<br />

and cur<strong>re</strong>nt a<strong>re</strong> <strong>in</strong> phase for some time after X = T . This is plausible if we consider that<br />

the high f<strong>re</strong>quency components at the wave front see a <strong>re</strong>al-valued characteristic impedance.<br />

Thus any phase shift is impossible.<br />

Second, the trail<strong>in</strong>g edge of the wave front becomes steeper as it travels down the l<strong>in</strong>e. The<br />

left po<strong>in</strong>t of the time <strong>in</strong>terval <strong>in</strong> each of the gu<strong>re</strong>s <strong>in</strong>dicates to a good approximation a<br />

constant value of the wave function. It can be calculated <strong>in</strong> dependence on the X-value as<br />

54


3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

30<br />

30<br />

0<br />

20<br />

0<br />

20<br />

10<br />

X<br />

10<br />

X<br />

T<br />

10<br />

T<br />

10<br />

20<br />

20<br />

30<br />

1<br />

0.5<br />

0<br />

-0.5<br />

-1<br />

0<br />

30<br />

1<br />

0<br />

0<br />

-0.5<br />

I/I0<br />

0.5<br />

U/U0<br />

Figu<strong>re</strong> 3.14: Evolution of the cur<strong>re</strong>nt (3.82) and voltage (3.83) along a one-dimensional transmission<br />

l<strong>in</strong>e with a lossless plasma for a slowly oscillat<strong>in</strong>g <strong>in</strong>put cur<strong>re</strong>nt ( =0:2).<br />

55


30<br />

30<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

0<br />

20<br />

0<br />

20<br />

10<br />

X<br />

10<br />

X<br />

T<br />

10<br />

T<br />

10<br />

20<br />

20<br />

30<br />

1<br />

0.5<br />

0<br />

-0.5<br />

-1<br />

0<br />

30<br />

1<br />

0<br />

-1<br />

-2<br />

0<br />

I/I0<br />

U/U0<br />

Figu<strong>re</strong> 3.15: Evolution of the cur<strong>re</strong>nt (3.82) and voltage (3.83) along a one-dimensional transmission<br />

l<strong>in</strong>e with a lossless plasma for an <strong>in</strong>put cur<strong>re</strong>nt with high f<strong>re</strong>quency ( = 0:8). Note the high voltage<br />

amplitude at the entrance.<br />

56


3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

9.9 9.95 10.05 10.1<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

99.99 99.995 100.005 100.01<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

999.999 1000 1000.<br />

Figu<strong>re</strong> 3.16: <strong>Wave</strong> front of the cur<strong>re</strong>nt for = 0:5 depend<strong>in</strong>g on the spatial coord<strong>in</strong>ate.<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

9.9 9.95 10.05 10.1<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

99.99 99.995 100.005 100.01<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

999.999 1000 1000.<br />

Figu<strong>re</strong> 3.17: <strong>Wave</strong> front of the voltage for = 0:5 depend<strong>in</strong>g on the spatial coord<strong>in</strong>ate.<br />

T = X ,1=X, so that an arbitrary <strong>in</strong>itial po<strong>in</strong>t (X0;T0) follows a phenomenological trajectory<br />

given by<br />

X , T<br />

X0 , T0<br />

= T0<br />

T<br />

: (3.84)<br />

This aga<strong>in</strong> is due to dispersion. The high f<strong>re</strong>quencies move ahead, the lower ones mo<strong>re</strong> and<br />

mo<strong>re</strong> lag beh<strong>in</strong>d, such that the high-pass lter e ect becomes mo<strong>re</strong> marked. The behaviour<br />

of the wave is the<strong>re</strong>fo<strong>re</strong> locally similar to the output of a high-pass lter whose characteristic<br />

f<strong>re</strong>quency is tuned higher and higher, i. e. the needle of the output pulse becomes sharper.<br />

As last item of the exam<strong>in</strong>ation, we discuss the propagation of a <strong>re</strong>ctangular pulse through<br />

the plasma. Ow<strong>in</strong>g to the l<strong>in</strong>earity of the medium, we al<strong>re</strong>ady stated <strong>in</strong> (3.68) that such a<br />

pulse can be constructed from two delayed step functions. If the duration of the pulse is an<br />

<strong>in</strong>teger multiple n of the signal period, we can use identical step <strong>re</strong>sponses for switch-on and<br />

switch-o and need not take ca<strong>re</strong> of the additional phase shift otherwise <strong>in</strong>troduced by an<br />

arbitrary pulse length. In our scaled variables, the pulse duration then becomes<br />

T t= = 2n : (3.85)<br />

We expect that both the lead<strong>in</strong>g and the trail<strong>in</strong>g edge of the pulse cause a sharp wave front<br />

travell<strong>in</strong>g at the velocity of light. The contour plots for short pulses with low ( g. 3.18)<br />

and high ( g. 3.19) <strong>in</strong>put f<strong>re</strong>quencies con rm this suspicion. They also show how the phase<br />

trajectories of the peaks and troughs of the wave asymptotically approach the wave front.<br />

Note that if we trace the evolution of such a peak, we nd that it travels always at a velocity<br />

57


120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

0<br />

120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

0<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

0 20 40 60 80 100 120<br />

0 20 40 60 80 100 120<br />

Figu<strong>re</strong> 3.18: Evolution of the cur<strong>re</strong>nt (upper graph) and voltage (lower graph) for a short pulse with<br />

=0:2 spann<strong>in</strong>g two signal periods (abscissa: X, ord<strong>in</strong>ate: T ).<br />

58


3.7 Turn-on e ects <strong>in</strong> a lossless plasma<br />

80<br />

60<br />

40<br />

20<br />

0<br />

80<br />

60<br />

40<br />

20<br />

0<br />

0 20 40 60 80<br />

0 20 40 60 80<br />

Figu<strong>re</strong> 3.19: Evolution of the cur<strong>re</strong>nt (upper graph) and voltage (lower graph) for a short pulse with<br />

=0:8 last<strong>in</strong>g for two signal periods (abscissa: X, ord<strong>in</strong>ate: T ).<br />

59


0.5<br />

-0.5<br />

I/I0<br />

-1<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

100 125 150 175 200 225 250 275<br />

Figu<strong>re</strong> 3.20: Cur<strong>re</strong>nt due to a short pulse spann<strong>in</strong>g two signal periods (n = 2) with = 0:2 far away<br />

from the signal source (at X = 100).<br />

g<strong>re</strong>ater than that of light. Its speed is a maximum immediately beh<strong>in</strong>d the entrance and<br />

dec<strong>re</strong>ases as it approaches the wave front. This mirrors <strong>in</strong> fact the phase velocity of a locally<br />

dom<strong>in</strong>at<strong>in</strong>g f<strong>re</strong>quency. In an experiment whe<strong>re</strong> no sharp wave front is p<strong>re</strong>sent, however, a<br />

su ciently narrow-band wave packet could likely be mistaken to move at a superlum<strong>in</strong>al<br />

velocity.<br />

Fig. 3.20 nally shows what a short cur<strong>re</strong>nt pulse looks like at a considerable distance from the<br />

tunnel entrance. The twowave fronts a<strong>re</strong> clearly marked by sharp needle-like pulses. Note that<br />

after switch-o , the <strong>re</strong>siduary oscillations follow<strong>in</strong>g the two wave fronts have slightly di e<strong>re</strong>nt<br />

temporary f<strong>re</strong>quencies and consequently form wave packets due to <strong>in</strong>terfe<strong>re</strong>nce. Dispersion,<br />

however, has <strong>re</strong>nde<strong>re</strong>d the orig<strong>in</strong>al pulse nearly un<strong>re</strong>cognisable.<br />

3.8 Turn-on e ects <strong>in</strong> a wave guide<br />

As we have seen <strong>in</strong> the last section, the wave front of a TEM wave <strong>in</strong> a lossless plasma<br />

propagates at the velocity of light. With a view to the microwave experiments carried out <strong>in</strong><br />

search of superlum<strong>in</strong>ality, we st<strong>re</strong>ss that the <strong>re</strong>sults obta<strong>in</strong>ed for the unbounded plasma also<br />

hold for the TE modes of a <strong>re</strong>ctangular wave guide because of the identical dispersion <strong>re</strong>lation.<br />

While the strictly causal wave front motion may seem plausible for a pu<strong>re</strong>ly transverse wave,<br />

the<strong>re</strong> is still the open question as to how a mo<strong>re</strong> general mode <strong>re</strong>acts to a disturbance. In<br />

this context, it is of particular <strong>in</strong>te<strong>re</strong>st how a eld component <strong>in</strong> the di<strong>re</strong>ction of propagation<br />

evolves. We will thus explo<strong>re</strong> the turn-on of such a mode.<br />

The TE01 or H01 mode <strong>in</strong> a <strong>re</strong>ctangular wave guide ( g. 3.21) is characterised by a longitud<strong>in</strong>al<br />

component of the magnetic eld. The eld components perpendicular to that di<strong>re</strong>ction a<strong>re</strong><br />

60<br />

T


3.8 Turn-on e ects <strong>in</strong> a wave guide<br />

b<br />

y<br />

z<br />

x<br />

Figu<strong>re</strong> 3.21: Rectangular wave guide.<br />

constant along one axis and s<strong>in</strong>usoidal along the other, the s<strong>in</strong>e spann<strong>in</strong>g only a half period.<br />

The cuto f<strong>re</strong>quency then is<br />

!p = c<br />

;<br />

b<br />

(3.86)<br />

b be<strong>in</strong>g the width of the wave guide, which is then half the transverse wavelength of the eld.<br />

Remark (Plasma and cuto f<strong>re</strong>quencies) To emphasise the similarity to the plasma<br />

example, we <strong>re</strong>ta<strong>in</strong> the symbol !p for the cuto f<strong>re</strong>quency, although it is normally denoted<br />

by !c <strong>in</strong> the literatu<strong>re</strong>.<br />

The steady-state components of this mode a<strong>re</strong> well known (Piefke [16], Harr<strong>in</strong>gton [93]). For<br />

f<strong>re</strong>quencies above cuto , !>!p,wehave<br />

Ez=Es<strong>in</strong> y x<br />

e,j<br />

b<br />

Hx = j b! E cos y<br />

b<br />

e,j x<br />

Hy = , ! E s<strong>in</strong> y<br />

b e,j x ;<br />

(3.87)<br />

with the dispersion <strong>re</strong>lation<br />

= 1q<br />

! 2 , !p<br />

c<br />

2 : (3.88)<br />

Note that we haveomitted the explicit time dependence of the elds. In the evanescent <strong>re</strong>gion,<br />

the propagation constant becomes imag<strong>in</strong>ary,<br />

j = = 1q<br />

!p<br />

c<br />

2 , ! 2 ; (3.89)<br />

and the eld components change to<br />

Ez = E s<strong>in</strong> y x<br />

e,<br />

b<br />

Hx = j b! E cos y<br />

b<br />

e, x<br />

Hy = j ! E s<strong>in</strong> y<br />

b e, x :<br />

61<br />

(3.90)


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

What we want to study is the evolution of the elds if we switch on the transverse component<br />

of the magnetic eld at t = 0 <strong>in</strong> such a way that only the TE01 mode is excited. This<br />

assumption is quite <strong>re</strong>asonable s<strong>in</strong>ce this mode has the lowest cuto f<strong>re</strong>quency and is thus<br />

the dom<strong>in</strong>at<strong>in</strong>g mode <strong>in</strong> the wave guide [15, 93]. So the boundary condition <strong>re</strong>ads<br />

Hy(0;t)= (t)H0cos !0t s<strong>in</strong> y<br />

b<br />

: (3.91)<br />

With the substitution H0 = j 0<br />

! E and tak<strong>in</strong>g the <strong>re</strong>al parts of the complex functions, we<br />

nally obta<strong>in</strong> the steady-state elds<br />

H0<br />

Ez;s<br />

Hx;s<br />

H0<br />

Hy;s<br />

H0<br />

=<br />

p =" =<br />

r<br />

1<br />

1 , !0<br />

!p<br />

= s<strong>in</strong> y<br />

b e, 0x cos !0t<br />

r !p<br />

!0<br />

1<br />

2<br />

2<br />

, 1<br />

cos y<br />

b e, 0x cos !0t<br />

s<strong>in</strong> y<br />

b e, 0x s<strong>in</strong> !0t ;<br />

(3.92)<br />

whe<strong>re</strong> 0 = 1p<br />

!p c<br />

2 , !0 2 is the attenuation constant at the signal f<strong>re</strong>quency. Compar<strong>in</strong>g<br />

these solutions with the ones we obta<strong>in</strong>ed for the cur<strong>re</strong>nt (3.69) and voltage (3.71) <strong>in</strong> the<br />

plasma transmission l<strong>in</strong>e, we notice a strik<strong>in</strong>g similarity: the transverse component of the<br />

magnetic eld, Hx, cor<strong>re</strong>sponds to the cur<strong>re</strong>nt <strong>in</strong> the transmission l<strong>in</strong>e, and the electric eld<br />

to the negative voltage. As the <strong>in</strong>itial conditions a<strong>re</strong> also the same, we can apply the same<br />

method we used successfully <strong>in</strong> the last section to obta<strong>in</strong> the transient parts of the<br />

With the familiar scaled variables<br />

elds.<br />

= c<br />

!p<br />

; T = !pt; X= !px<br />

c<br />

; = !0<br />

!p<br />

; Y = y<br />

b<br />

we can immediately write down the complete solutions for the transverse elds,<br />

H0<br />

Hy<br />

H0<br />

= s<strong>in</strong> Y<br />

Ez p = s<strong>in</strong> Y<br />

="<br />

h e ,X p 1, 2<br />

, 2 Z 1<br />

0<br />

cos T ,<br />

s<strong>in</strong> ( X) cos T p 1+ 2<br />

1+ 2 , 2<br />

h<br />

,X<br />

e p 1, 2 1<br />

p s<strong>in</strong> T ,<br />

1= 2 , 1<br />

, 2 Z 1<br />

0<br />

p 1+ 2 cos ( X) s<strong>in</strong> T p 1+ 2<br />

62<br />

1+ 2 , 2<br />

d<br />

i<br />

; (3.93)<br />

d<br />

i :<br />

(3.94)<br />

(3.95)


3.8 Turn-on e ects <strong>in</strong> a wave guide<br />

Hx<br />

4<br />

2<br />

0<br />

-2<br />

10<br />

-4<br />

0<br />

7.5<br />

2.5<br />

X<br />

5<br />

5<br />

T<br />

2.5<br />

0<br />

7.5<br />

Figu<strong>re</strong> 3.22: Evolution of the longitud<strong>in</strong>al component (3.96) of the magnetic eld for aTE01 wave.<br />

The parameters a<strong>re</strong> = 0:8 and Y =1=4.<br />

The longitud<strong>in</strong>al magnetic component is best calculated from the condition that the magnetic<br />

eld be source-f<strong>re</strong>e, @Hx=@x = ,@Hy=@y, which yields after some simple manipulations<br />

h Hx<br />

,X<br />

= cos Y e<br />

H0<br />

p 1, 2 1<br />

p cos T ,<br />

1 , 2<br />

, 2 Z 1 cos ( X) cos T p 1+ 2<br />

1+ 2 , 2<br />

(3.96)<br />

i<br />

d :<br />

0<br />

We know al<strong>re</strong>ady how the transverse components behave, so we need not <strong>re</strong>produce the <strong>re</strong>sults<br />

of the last section. The longitud<strong>in</strong>al component of the magnetic eld, however, is new. A<br />

numerical evaluation is shown <strong>in</strong> g. 3.22 . We see that the wave front ofHx, too, propagates<br />

with c. Furthermo<strong>re</strong>, at t = x=c, the longitud<strong>in</strong>al component vanishes. Thus the front<br />

of the wave travell<strong>in</strong>g along the wave guide after turn-on consists exclusively of transverse<br />

components, i. e. the mode acts exactly like a TEM mode at this <strong>in</strong>stant. This becomes even<br />

mo<strong>re</strong> imp<strong>re</strong>ssive if we look at the magnetic eld l<strong>in</strong>es of the wave ( g. 3.23). The example<br />

demonstrates clearly that even waves with eld components <strong>in</strong> propagation di<strong>re</strong>ction do not<br />

propagate faster than light.<br />

Remark (Optical cuto f<strong>re</strong>quencies) Accord<strong>in</strong>g to (3.86), the cuto f<strong>re</strong>quency of a<br />

wave guide is <strong>in</strong>versely proportional to its width or | <strong>in</strong> the case of a cyl<strong>in</strong>drical guide<br />

| its diameter. This imposes an <strong>in</strong>convenient limit on the m<strong>in</strong>iaturisation of <strong>in</strong>tegrated<br />

optical devices. The<strong>re</strong> is, however, a way to circumvent the problem. Takahara et al. [99]<br />

proposed to use metallic wave guides at optical f<strong>re</strong>quencies. At these high f<strong>re</strong>quencies, the<br />

metal actually exhibits a negative dielectric constant, so that alow-pass TM mode can<br />

63<br />

10


0.8<br />

0.6<br />

0.4<br />

0.2<br />

Y<br />

1<br />

0<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

0 1 2 3 4 5<br />

Figu<strong>re</strong> 3.23: Snapshot of the magnetic eld of a TE01 wave. The parameters a<strong>re</strong> = 0:8 and T =4:87 .<br />

propagate without a lower cuto f<strong>re</strong>quency. In fact, this is a surface wave at the <strong>in</strong>terface<br />

between metal and `normal' dielectric, and the<strong>re</strong>fo<strong>re</strong> the geometrical limitations a<strong>re</strong> no<br />

longer <strong>re</strong>levant. Unfortunately, this mode is associated with a rather high attenuation,<br />

but for small-distance applications, this might be tolerable (see also Paschke [100]).<br />

3.9 A Gaussian pulse <strong>in</strong> plasma<br />

To round o the <strong>re</strong>sults of the p<strong>re</strong>vious sections, we now <strong>in</strong>vestigate the behaviour of a<br />

Gaussian pulse <strong>in</strong> a lossless plasma. The model we use is still the transmission l<strong>in</strong>e of section<br />

3.5 with the dispersion <strong>re</strong>lation<br />

k(!) = !p<br />

c<br />

s !<br />

!p<br />

2<br />

,1: (3.97)<br />

The boundary condition, i. e. the pulse applied at the <strong>in</strong>terface x = 0, is given by<br />

u(0;t)=U0e , t,t 0 2<br />

Re e j!0t ; (3.98)<br />

whe<strong>re</strong> t0 denotes the position of the peak and is the standard deviation of the exponential<br />

distribution. Ow<strong>in</strong>g to the l<strong>in</strong>earity of the system, we mayuse the complex notation and take<br />

the <strong>re</strong>al part of the exp<strong>re</strong>ssions if necessary. The voltage at any po<strong>in</strong>t <strong>in</strong>side the plasma is<br />

then given <strong>in</strong> the well-known manner as the Fourier <strong>in</strong>tegral<br />

u(x; t) =Re<br />

Z 1<br />

,1<br />

A(!) e j(!t,k(!)x) d! : (3.99)<br />

Note that s<strong>in</strong>ce the pulse has an <strong>in</strong> nite duration, the<strong>re</strong> is no <strong>in</strong>itial condition to comply<br />

with. Consequently, we cannot apply a trick as <strong>in</strong> the turn-on case of section 3.7 to make the<br />

evaluation easier, and we have to compute the wave <strong>in</strong>tegral di<strong>re</strong>ctly.<br />

The spectrum A(!) of the <strong>in</strong>itial pulse can easily be calculated, which gives<br />

A(!) = 1<br />

2<br />

Z 1<br />

,1<br />

u(0;t) e ,j!t dt = p e<br />

2 , (!,! 0 )<br />

2<br />

64<br />

2<br />

e ,jt0(!,!0) : (3.100)<br />

X


3.9 A Gaussian pulse <strong>in</strong> plasma<br />

Like always, we <strong>in</strong>troduce a normalised <strong>in</strong>tegration variable<br />

and normalised parameters<br />

= !<br />

!p<br />

T = !pt ; T0=!pt0; =!p ; X= !px<br />

c<br />

; = !0<br />

!p<br />

(3.101)<br />

: (3.102)<br />

With these new variables, we can write the voltage at any position along the l<strong>in</strong>e as<br />

U(X; T )=U0 2 p<br />

Z 1<br />

,1<br />

( , )<br />

, 2 e<br />

2<br />

e ,jT0( , ) e j( T,K( )X) d ; (3.103)<br />

with the normalised dispersion <strong>re</strong>lation K( )= p 2 ,1. As for the sign of the squa<strong>re</strong> root,<br />

we must keep <strong>in</strong> m<strong>in</strong>d that s<strong>in</strong>ce we <strong>re</strong>gard only right-go<strong>in</strong>g waves, the negative branch applies<br />

to negative f<strong>re</strong>quencies. In the evanescent <strong>re</strong>gion, we must also choose the negative solution<br />

of the root <strong>in</strong> order to obta<strong>in</strong> attenuation along the l<strong>in</strong>e. So <strong>in</strong> the th<strong>re</strong>e parts of the range of<br />

<strong>in</strong>tegration, the dispersion <strong>re</strong>lation is given by<br />

K( )=<br />

8<br />

><<br />

>:<br />

, p 2 ,1 if 1<br />

: (3.104)<br />

Remark (Transfer function of the plasma) Look<strong>in</strong>g at the wave <strong>in</strong>tegral from a<br />

system theo<strong>re</strong>tical po<strong>in</strong>t of view, we obta<strong>in</strong> still another <strong>re</strong>ason<strong>in</strong>g for the choice of the<br />

sign of the dispersion <strong>re</strong>lation. We can th<strong>in</strong>k of the wave <strong>in</strong>tegral (3.99) as the <strong>re</strong>sponse<br />

of a system to a signal u(0;t),<br />

u(x; t) =<br />

Z 1<br />

whe<strong>re</strong> the transfer function of the system is<br />

A(!)H(!) e<br />

,1<br />

j!t d! ; (3.105)<br />

H(!) =e ,jk(!)x : (3.106)<br />

It is well known that a physically mean<strong>in</strong>gful system must give a pu<strong>re</strong>ly <strong>re</strong>al <strong>re</strong>sponse<br />

when excited with a pu<strong>re</strong>ly <strong>re</strong>al signal. Accord<strong>in</strong>g to the basic properties of the Fourier<br />

transform, this can only be guaranteed if the <strong>re</strong>al and the imag<strong>in</strong>ary parts of the transfer<br />

function a<strong>re</strong> even and odd functions <strong>in</strong> !, <strong>re</strong>spectively. It is easy to see that the de nition<br />

of the dispersion <strong>re</strong>lation given above satis es these <strong>re</strong>qui<strong>re</strong>ments. With<strong>in</strong> the pass<br />

band, the imag<strong>in</strong>ary part of H(!) is essentially the s<strong>in</strong>e of an odd argument function<br />

and thus also odd, whe<strong>re</strong>as the <strong>re</strong>al part is the cos<strong>in</strong>e of an odd function and thus even.<br />

In the evanescent <strong>re</strong>gion, the imag<strong>in</strong>ary part of the transfer function vanishes, and the<br />

exponentially decay<strong>in</strong>g <strong>re</strong>al part <strong>re</strong>ma<strong>in</strong>s even.<br />

The cur<strong>re</strong>nt <strong>in</strong> the transmission l<strong>in</strong>e could be calculated by <strong>in</strong>sert<strong>in</strong>g the voltage <strong>in</strong>to the<br />

di e<strong>re</strong>ntial equations (3.75) and solv<strong>in</strong>g for I, which gives<br />

i(x; t) =L 0<br />

Z 1<br />

A(!)<br />

,1<br />

k(!)<br />

! ej(!t,k(!)x) d! : (3.107)<br />

65


0.75<br />

0.5<br />

0.25<br />

-0.25<br />

-0.5<br />

-0.75<br />

U/U0<br />

1<br />

50 100 150 200 T<br />

3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

|A|<br />

4<br />

3<br />

2<br />

1<br />

-2 -1 1 2 η<br />

Figu<strong>re</strong> 3.24: Initial pulse and its spectrum jA( )j for the parameter values T0 = 100 and = 17. The<br />

carrier f<strong>re</strong>quency is = 0:5 <strong>in</strong> these particular graphs.<br />

Unfortunately, a pole turns up at ! =0,p<strong>re</strong>vent<strong>in</strong>g straightforward <strong>in</strong>tegration. Tak<strong>in</strong>g the<br />

Cauchy pr<strong>in</strong>cipal value of the <strong>in</strong>tegral would solve the problem, but as we do not need the<br />

cur<strong>re</strong>nt to ga<strong>in</strong> some general <strong>in</strong>sight <strong>in</strong>to the propagation of a Gaussian wave, we leave it<br />

aside.<br />

The exp<strong>re</strong>ssion for the voltage (3.103) is valid <strong>in</strong> both the stop and pass bands. We can thus<br />

compute the voltage <strong>in</strong>side the plasma for a given pulse with vary<strong>in</strong>g carrier f<strong>re</strong>quencies. The<br />

<strong>in</strong>itial shape and its spectrum a<strong>re</strong> shown <strong>in</strong> g. 3.24. If the cuto f<strong>re</strong>quency of the plasma<br />

is assumed to be !p = 10 10 9 s ,1 (a value roughly cor<strong>re</strong>spond<strong>in</strong>g to Nimtz' microwave<br />

experiments), then the pulse width is about 6 ns, which is fairly short. Accord<strong>in</strong>gly, the<br />

spectrum is quite wide, and so we must expect a small portion of the pulse to be classically<br />

transmitted even though the major part of the eld orig<strong>in</strong>ates from the evanescent <strong>re</strong>gion.<br />

The rst case we exam<strong>in</strong>e is the base-band signal with =0. The <strong>re</strong>sult given <strong>in</strong> g. 3.25<br />

is not very spectacular s<strong>in</strong>ce the spectrum is practically con ned to the range below cuto .<br />

Thus the<strong>re</strong> a<strong>re</strong> no propagat<strong>in</strong>g waves, and all we can see is the exponential decay along the<br />

spatial coord<strong>in</strong>ate. In contrast to the follow<strong>in</strong>g gu<strong>re</strong>s, this one shows no modulation, and<br />

so the vanish<strong>in</strong>g voltage outside the actual pulse also denotes the least possible value of the<br />

eld. The<strong>re</strong>fo<strong>re</strong> the a<strong>re</strong>a surround<strong>in</strong>g the contour of the pulse is shaded <strong>in</strong> black.<br />

When we use a carrier with = 0:5, the behaviour of the wave is much mo<strong>re</strong> <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g.<br />

Fig. 3.26 shows rst of all a dom<strong>in</strong>at<strong>in</strong>g evanescent signal that <strong>re</strong>sembles the unmodulated case<br />

t<strong>re</strong>ated befo<strong>re</strong>. It is particularly noteworthy that the carrier f<strong>re</strong>quency <strong>in</strong>c<strong>re</strong>ases gradually<br />

with X. This is manifested <strong>in</strong> the graph by the narrow<strong>in</strong>g of the black and white stripes<br />

and can be expla<strong>in</strong>ed as a pulse <strong>re</strong>shap<strong>in</strong>g e ect. S<strong>in</strong>ce the dispersion <strong>in</strong> the plasma causes<br />

higher f<strong>re</strong>quency components to be less attenuated than lower ones, the cent<strong>re</strong> f<strong>re</strong>quency of<br />

the spectrum is slightly raised. Note that although it looks as if the pulse became shorter with<br />

<strong>in</strong>c<strong>re</strong>as<strong>in</strong>g X, this is not the case. It is only the attenuation <strong>in</strong> comb<strong>in</strong>ation with the limited<br />

plott<strong>in</strong>g range that c<strong>re</strong>ates this deceptive imp<strong>re</strong>ssion. The overall shape of the pulse <strong>re</strong>ma<strong>in</strong>s<br />

mo<strong>re</strong> or less the same, albeit with <strong>re</strong>duced amplitude. Di<strong>re</strong>ct comparison of the unmodulated<br />

and the p<strong>re</strong>sent <strong>re</strong>sults <strong>re</strong>veal another detail: the plot ranges of the two graphs a<strong>re</strong> identical,<br />

66


3.9 A Gaussian pulse <strong>in</strong> plasma<br />

250<br />

200<br />

150<br />

100<br />

50<br />

0<br />

0 10 20 30 40 50<br />

Figu<strong>re</strong> 3.25: Evolution of the voltage for = 0. The abscissa denotes X, the ord<strong>in</strong>ate T .<br />

250<br />

200<br />

150<br />

100<br />

50<br />

0<br />

0 10 20 30 40 50<br />

Figu<strong>re</strong> 3.26: Evolution of the voltage for = 0:5. The abscissa denotes X, the ord<strong>in</strong>ate T .<br />

67


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

which enables one to see that the modulated pulse penetrates deeper <strong>in</strong>to the medium than<br />

the base-band signal. This is <strong>in</strong> accordance with the dispersion <strong>re</strong>lation and the <strong>re</strong>spective<br />

attenuation coe cient.<br />

The bulk of the signal energy is concentrated with<strong>in</strong> the parabola-shaped <strong>re</strong>gion <strong>in</strong> the timespace<br />

graph. But we also <strong>re</strong>cognise a small fraction of the pulse runn<strong>in</strong>g away down the<br />

l<strong>in</strong>e when the major part of the signal is al<strong>re</strong>ady dim<strong>in</strong>ish<strong>in</strong>g. The high oscillation f<strong>re</strong>quency<br />

of this low but very broad pulse (twice the carrier f<strong>re</strong>quency) shows that it consists of the<br />

spectral components just above cuto . In g. 3.24, which shows exactly the spectrum of this<br />

particular case, the spectrum seems to be su ciently below cuto , but evidently it is not. We<br />

can roughly calculate the f<strong>re</strong>quency that dom<strong>in</strong>ates this part of the signal by measur<strong>in</strong>g the<br />

slope X= T parallel to the peaks and troughs visible <strong>in</strong> the graph. They give an estimate<br />

for the phase velocity, while the l<strong>in</strong>es connect<strong>in</strong>g the outer ends of the ripples a<strong>re</strong> <strong>re</strong>lated to<br />

the group velocity. Us<strong>in</strong>g the de nition of the dispersion <strong>re</strong>lation (3.97), we <strong>re</strong>adily nd the<br />

normalised exp<strong>re</strong>ssions for the phase velocity,<br />

and the group velocity,<br />

vp<br />

c =<br />

r 2<br />

vg<br />

c =<br />

r<br />

1 , 1<br />

2<br />

2 , 1 ; (3.108)<br />

; (3.109)<br />

<strong>re</strong>spectively. With these two equations, which di<strong>re</strong>ctly cor<strong>re</strong>spond to the appropriately measu<strong>re</strong>d<br />

slope X= T , we nally obta<strong>in</strong> an <strong>in</strong>terval for the carrier f<strong>re</strong>quency, 2 [1; 1:02].<br />

The left bound of the <strong>in</strong>terval is obta<strong>in</strong>ed from the left end of the propagat<strong>in</strong>g wave close<br />

to X = 0 whe<strong>re</strong> the tangents to the ripples a<strong>re</strong> horizontal and the ends of the valleys lie<br />

on a vertical l<strong>in</strong>e. The same consideration applied to the right end of the wave gives the<br />

upper limit of the <strong>in</strong>terval. It is clear, though, that the data we can gather from the graph<br />

to nd this limit is di<strong>re</strong>ctly dependent on the <strong>re</strong>solution we choose for the voltage amplitude.<br />

In <strong>re</strong>ality, the <strong>in</strong>terval is as unlimited as the spectrum of the pulse, and only the numerical<br />

accuracy imposes constra<strong>in</strong>ts on the <strong>re</strong>liability of the <strong>re</strong>sults (which <strong>in</strong>deed p<strong>re</strong>vented us from<br />

discover<strong>in</strong>g the same e ect <strong>in</strong> the base-band pulse). One must of course not overlook the fact<br />

that the amplitude of the propagat<strong>in</strong>g part is n<strong>in</strong>e orders of magnitude smaller than the ma<strong>in</strong><br />

pulse. The<strong>re</strong>fo<strong>re</strong> it is likely to be missed <strong>in</strong> an experiment.<br />

The third example ( g. 3.27) shows a pulse whe<strong>re</strong> the carrier f<strong>re</strong>quency equals the cuto f<strong>re</strong>quency<br />

( = 1). He<strong>re</strong>, half the spectrum lies <strong>in</strong> the pass band. Thus the pulse is propagated,<br />

but it is seve<strong>re</strong>ly distorted. The <strong>re</strong>ason for this behaviour is evidently the dispersion <strong>re</strong>lation.<br />

A moderate pulse distortion <strong>re</strong>qui<strong>re</strong>s k(!) to be approximately l<strong>in</strong>ear <strong>in</strong> the f<strong>re</strong>quency range<br />

that makes up the dom<strong>in</strong>at<strong>in</strong>g part of the spectrum. About cuto , this condition is de -<br />

nitely not met, and we end up with a mixtu<strong>re</strong> of spectral components hav<strong>in</strong>g grossly di e<strong>re</strong>nt<br />

propagation velocities.<br />

In g. 3.28, = 1:5 was chosen for the carrier. As we could have expected, this gives a<br />

pulse propagat<strong>in</strong>g straight through the medium. This <strong>re</strong>sult can now be used to verify the<br />

68


3.9 A Gaussian pulse <strong>in</strong> plasma<br />

250<br />

200<br />

150<br />

100<br />

50<br />

0<br />

0 10 20 30 40 50<br />

Figu<strong>re</strong> 3.27: Evolution of the voltage for = 1. The abscissa denotes X, the ord<strong>in</strong>ate T .<br />

250<br />

200<br />

150<br />

100<br />

50<br />

0<br />

0 10 20 30 40 50<br />

Figu<strong>re</strong> 3.28: Evolution of the voltage for = 1:5. The abscissa denotes X, the ord<strong>in</strong>ate T .<br />

69


3<strong>Wave</strong> propagation <strong>in</strong> electromagnetic transmission l<strong>in</strong>es<br />

classical understand<strong>in</strong>g of the phase and group velocities. The slope of the stripes yields the<br />

phase velocity, and the contour of the enti<strong>re</strong> pulse (most easily traced at the edge whe<strong>re</strong> they<br />

fade <strong>in</strong>to the g<strong>re</strong>y surround<strong>in</strong>gs) de nes the group velocity. Obviously, these two velocities<br />

a<strong>re</strong> di e<strong>re</strong>nt, which empirically a rms the theo<strong>re</strong>tical nd<strong>in</strong>gs of section 1.1: <strong>in</strong> a dispersive<br />

medium, the envelope of a pulse propagates at a dist<strong>in</strong>ct velocity with the tra<strong>in</strong> of carrier<br />

oscillations mov<strong>in</strong>g <strong>in</strong>depently underneath its overall shape. As for the determ<strong>in</strong>ation of<br />

from the graph, both (3.108) and (3.109) give the cor<strong>re</strong>ct answers.<br />

Remark (Dispersion e ects) In a dispersive medium, we should of course be able<br />

to observe a broaden<strong>in</strong>g of the pulse. We a<strong>re</strong>, however, at a loss to do so <strong>in</strong> this case<br />

because the <strong>in</strong>itial pulse is too wide and the distance too short for this e ect to become<br />

visible. But if we su ciently <strong>re</strong>duced the plotted voltage range (by some n<strong>in</strong>e orders of<br />

magnitude), we would encounter an e ect similar to that <strong>in</strong> g. 3.26: namely, a small<br />

portion of the wave travell<strong>in</strong>g, if at all, at very low speed. Aga<strong>in</strong>, this is caused by the<br />

small f<strong>re</strong>quency range immediately above cuto whe<strong>re</strong> the group velocity is hardly larger<br />

than zero. In contrast, the e ect of the evanescent components (which a<strong>re</strong> p<strong>re</strong>sent also <strong>in</strong><br />

this case) is not noticeable at all.<br />

In accordance with the theory, the examples show that if the spectrum of a pulse is bundled<br />

below cuto , the<strong>re</strong> is no phase shift, and the pulse decays exponentially without allow<strong>in</strong>g for<br />

wave propagation. The amount of attenuation is determ<strong>in</strong>ed by the cent<strong>re</strong> f<strong>re</strong>quency of the<br />

pulse, !0, so that we can estimate the voltage amplitude by<br />

U(X) / e ,Xp 1, 2<br />

: (3.110)<br />

If <strong>in</strong> a data transmission system we decided, for the sake of proper signal detection, to<br />

tolerate an attenuation of 10 ,4 (80 dB), we could calculate the admissible length of the<br />

l<strong>in</strong>e. For = 0:5 this would give X = 10:64, whe<strong>re</strong>as with = 0:8, we could cover a<br />

normalised distance of X =15:35. The actual length of the transmission l<strong>in</strong>e can be obta<strong>in</strong>ed<br />

from the de nition of the scaled space coord<strong>in</strong>ate, x = Xc=!p. In a plasma with a cuto<br />

f<strong>re</strong>quency of 10 GHz, X = 10 cor<strong>re</strong>sponds to an actual distance of x = 4:77 cm. So if we<br />

we<strong>re</strong> to use signals with superlum<strong>in</strong>al group velocities (as they a<strong>re</strong> sometimes called) for data<br />

transmission, we could cover but a few centimet<strong>re</strong>s. This, however, makes their usefulness for<br />

practical applications rather questionable.<br />

The maximum of a band-limited pulse can be transmitted at a velocity g<strong>re</strong>ater than that of<br />

light, the<strong>re</strong> is no doubt about it. But does this mean that <strong>in</strong>formation can also be transmitted<br />

superlum<strong>in</strong>ally? The <strong>re</strong>sponse of a l<strong>in</strong>ear system to an <strong>in</strong>put signal can be written as a<br />

convolution <strong>in</strong>tegral,<br />

u(x; t) =<br />

Z 1<br />

,1<br />

of the <strong>in</strong>put u(0;t) and the transfer function<br />

g(x; t) = 1<br />

2<br />

u(0; ) g(x; t , ) d ; (3.111)<br />

Z 1<br />

e<br />

,1<br />

,jk(!)x e j!t d! : (3.112)<br />

70


3.9 A Gaussian pulse <strong>in</strong> plasma<br />

Note that while communication eng<strong>in</strong>eers talk of transfer functions (or, likewise, impulse<br />

<strong>re</strong>sponses), theo<strong>re</strong>tical physicists f<strong>re</strong>quently call g(x; t) a G<strong>re</strong>en's function. In either case, it<br />

is said to be causal if it is identically zero for t


Chapter 4<br />

One-dimensional quantum<br />

tunnell<strong>in</strong>g<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Das Best<strong>re</strong>ben, der Dispersion Herr zu werden, leuchtet mir fur langsame<br />

Wellen e<strong>in</strong>, das Ubrige ist mir aus den Andeutungen nicht klar geworden.<br />

Das Au allende ist, wie viel man mit klassischer Mechanik befriedigend<br />

machen kann. Wenn es nur e<strong>in</strong>mal gelange, das Pr<strong>in</strong>zipielle an den Quanten<br />

e<strong>in</strong>igermassen zu erkla<strong>re</strong>n! Aber me<strong>in</strong>e Ho nung, das zu erleben, wird<br />

immer kle<strong>in</strong>er.<br />

Albert E<strong>in</strong>ste<strong>in</strong> <strong>in</strong> a letter to Arnold Sommerfeld (1. 2. 1918 [103])<br />

Hav<strong>in</strong>g exhausted the question of electromagnetic wave propagation we now turn to the<br />

second a<strong>re</strong>a whe<strong>re</strong> tunnell<strong>in</strong>g occurs and whe<strong>re</strong>, after all, the term `tunnell<strong>in</strong>g' was co<strong>in</strong>ed.<br />

To start with, however, we shall <strong>re</strong>strict our <strong>in</strong>vestigation to the comparatively simple step<br />

potential barrier, whe<strong>re</strong> we have to cope with only one <strong>in</strong>terface. In a second step, we <strong>re</strong>gard<br />

tunnell<strong>in</strong>g <strong>in</strong> its orig<strong>in</strong>al mean<strong>in</strong>g as the motion of a particle through a classically impenetrable<br />

barrier. In the simplest form of a <strong>re</strong>ctangular potential wall, such an <strong>in</strong>vestigation <strong>re</strong>qui<strong>re</strong>s<br />

the <strong>in</strong>clusion of two <strong>in</strong>terfaces, which <strong>re</strong>nders the t<strong>re</strong>atment mo<strong>re</strong> complicated albeit by no<br />

means impossible. In both cases, we study how the particle evolves <strong>in</strong> time and space upon<br />

imp<strong>in</strong>g<strong>in</strong>g on the barrier.<br />

The chapter beg<strong>in</strong>s with the well-known formulation of the scatter<strong>in</strong>g process consist<strong>in</strong>g of an<br />

<strong>in</strong>cident f<strong>re</strong>e particle be<strong>in</strong>g partly <strong>re</strong> ected and partly penetrat<strong>in</strong>g the obstacle. The solutions<br />

will be given as Fourier <strong>in</strong>tegrals and primarily be <strong>in</strong>dependent of the actual <strong>in</strong>itial condition.<br />

We shall then explo<strong>re</strong> several possibilities of <strong>in</strong>itial wave forms or particle shapes that nally<br />

a<strong>re</strong> used to obta<strong>in</strong> numerical <strong>re</strong>sults of the wave propagation both <strong>in</strong>side and outside the<br />

barrier. The squa<strong>re</strong> potential barrier will be conside<strong>re</strong>d next, and we shall brie y exam<strong>in</strong>e the<br />

variety of tunnell<strong>in</strong>g time de nitions known from the literatu<strong>re</strong> for monochromatic waves. The<br />

conclud<strong>in</strong>g numerical evaluation of tunnell<strong>in</strong>g events is <strong>re</strong>stricted to Gaussian wave packets.<br />

72


4.1 The potential step<br />

4.1 The potential step<br />

Be<strong>in</strong>g <strong>in</strong>te<strong>re</strong>sted chie y <strong>in</strong> the dynamical evolution of a wave <strong>in</strong>side an evanescent barrier, we<br />

assume the barrier to be <strong>in</strong> nitely extended <strong>in</strong>to the right half space. The potential function<br />

is the<strong>re</strong>fo<strong>re</strong><br />

(<br />

0 if x


4 One-dimensional quantum tunnell<strong>in</strong>g<br />

spectrum <strong>in</strong> wave number space, A( ), each of the monochromatic components is itself a<br />

steady-state solution of Schrod<strong>in</strong>ger's equation, and the superposition of all partial waves<br />

then yields the desi<strong>re</strong>d evolution of a s<strong>in</strong>gle wave. The general form of this solution is<br />

(x; t) =<br />

Z 1<br />

C A( ) e<br />

,1<br />

j( x,!( )t) d ; (4.7)<br />

whe<strong>re</strong> C and also have tobechosen accord<strong>in</strong>g to the <strong>re</strong>spective <strong>re</strong>gion. The <strong>in</strong>cident wave<br />

then <strong>re</strong>ads<br />

<strong>in</strong>c =<br />

Z 1<br />

,1<br />

j x,j<br />

~<br />

A( ) e 2m 2t d : (4.8)<br />

The formulation of the <strong>re</strong> ected part is not so straightforward and <strong>re</strong>qui<strong>re</strong>s a moment's<br />

consideration. The <strong>re</strong> ection coe cient conta<strong>in</strong>s the propagation constants of both the <strong>re</strong>gion<br />

outside the tunnel and the tunnel itself. The l<strong>in</strong>k between the <strong>re</strong>spective dispersion <strong>re</strong>lations,<br />

however, is provided by the f<strong>re</strong>quency ! that alone <strong>re</strong>ma<strong>in</strong>s <strong>in</strong>variant as the monochromatic<br />

wavelets penetrate <strong>in</strong>to the barrier, whe<strong>re</strong>as the propagation constant changes. The<strong>re</strong>fo<strong>re</strong> we<br />

must not mix up 1 and 2. Instead, we <strong>re</strong>ta<strong>in</strong> 1 as and exp<strong>re</strong>ss 2 by means of (4.6)<br />

as 2 2 = 2 , 2m!p=~. Consequently, the <strong>in</strong>tegration range falls <strong>in</strong>to th<strong>re</strong>e dist<strong>in</strong>ct <strong>in</strong>tervals<br />

determ<strong>in</strong>ed by<br />

2 =<br />

8<br />

><<br />

>:<br />

,<br />

q 2 , 2m!p<br />

q 2m!p<br />

~ if <br />

~<br />

q<br />

2m!p<br />

~<br />

q<br />

2m!p<br />

~<br />

q<br />

2m!p<br />

~<br />

q 2m!p<br />

~<br />

: (4.9)<br />

Remark (Transformation of ) The use of the negative root<strong>in</strong>thetransformation<br />

becomes obvious if we call to m<strong>in</strong>d that the dispersion <strong>re</strong>lations a<strong>re</strong> even functions <strong>in</strong> and<br />

separated <strong>in</strong> ord<strong>in</strong>ate di<strong>re</strong>ction only by the cuto f<strong>re</strong>quency !p. Hence the transformation<br />

rule must be even, too, which gives 2 = sign p 2 , 2m!p=~ outside the evanescent<br />

<strong>re</strong>gion.<br />

With (4.9) and (4.4), we nally obta<strong>in</strong> the th<strong>re</strong>e parts of the <strong>re</strong> ected wave,<br />

<strong>re</strong>f =<br />

Z ,<br />

+<br />

,1<br />

q 2m!p<br />

~<br />

Z q 2m!p<br />

~<br />

q 2m!p<br />

, ~<br />

Z 1<br />

+ q 2m!p<br />

~<br />

A( ) +<br />

A( )<br />

,<br />

q<br />

2 2m!p<br />

, ~ q<br />

2 2m!p<br />

, ~<br />

q<br />

2m!p<br />

, j ~ , 2<br />

q<br />

2m!p<br />

+ j ~ , 2<br />

A( ) ,<br />

q<br />

2 2m!p<br />

, ~<br />

+<br />

q<br />

2 2m!p<br />

, ~<br />

74<br />

,j x,j<br />

~<br />

e 2m 2t d +<br />

,j x,j<br />

~<br />

e 2m 2t d +<br />

,j x,j<br />

~<br />

e 2m 2t d :<br />

(4.10)


4.1 The potential step<br />

Inside the barrier, we must aga<strong>in</strong> apply (4.9), although one might perhaps argue that <strong>in</strong> this<br />

<strong>re</strong>gion the orig<strong>in</strong>al 2 would have been mo<strong>re</strong> appropriate. This would <strong>in</strong> fact be true if the<br />

spectrum of the <strong>in</strong>itial wave, A( ), had been given <strong>in</strong> terms of the wave numbers <strong>in</strong>side the<br />

tunnel. But as it is, the outset of our <strong>in</strong>vestigation is a wave strictly con ned to the outside of<br />

the barrier, and so we cannot but describe it <strong>in</strong> that wave number space. Thus the transmitted<br />

part of the wave becomes<br />

tun =<br />

Z ,<br />

+<br />

,1<br />

q 2m!p<br />

~<br />

Z q 2m!p<br />

~<br />

q 2m!p<br />

, ~<br />

A( )<br />

A( )<br />

Z 1<br />

+ q A( )<br />

2m!p<br />

~<br />

,<br />

+<br />

2<br />

q 2 , 2m!p<br />

~<br />

+ j<br />

2<br />

q 2m!p<br />

~ , 2<br />

2<br />

q 2 , 2m!p<br />

~<br />

e ,jx<br />

e jx<br />

q<br />

2, 2m!p<br />

~ ,j ~<br />

2m 2 t d +<br />

e ,x<br />

q 2m!p<br />

~ , 2 ,j ~<br />

2m 2t d +<br />

q<br />

2, 2m!p<br />

~ ,j ~<br />

2m 2t d :<br />

(4.11)<br />

As usual, we <strong>in</strong>troduce normalised variables to ease the numerical evaluation,<br />

r<br />

~ !p<br />

c =<br />

2m ; T = !p t; X= !p<br />

x; (4.12)<br />

c<br />

as well as a new <strong>in</strong>tegration variable<br />

=<br />

s ~<br />

2m!p<br />

= c<br />

!p<br />

: (4.13)<br />

Note that <strong>in</strong> this context, c has noth<strong>in</strong>g at all to do with the velocity of light. It is noth<strong>in</strong>g<br />

butavariable that <strong>in</strong>cidentally has the physical dimension of a velocity. The <strong>in</strong>dividual wave<br />

functions can be <strong>re</strong>written as<br />

<strong>in</strong>c = !p<br />

c<br />

<strong>re</strong>f = !p<br />

c<br />

+ !p<br />

c<br />

+ !p<br />

c<br />

Z 1<br />

A( ) e<br />

,1<br />

jX ,j 2T d ; (4.14)<br />

Z ,1<br />

,1<br />

Z 1<br />

,1<br />

Z 1<br />

1<br />

A( ) + p 2 , 1<br />

, p ,jX ,jT 2<br />

e d +<br />

2 , 1<br />

A( ) , jp1 , 2<br />

+ j p ,jT 2<br />

e,jX d +<br />

1 , 2<br />

A( ) , p 2 , 1<br />

+ p ,jX ,jT 2<br />

e d ;<br />

2 , 1<br />

75<br />

(4.15)


tun = !p<br />

c<br />

+ !p<br />

c<br />

+ !p<br />

c<br />

Z ,1<br />

,1<br />

Z 1<br />

,1<br />

Z 1<br />

1<br />

A( )<br />

A( )<br />

A( )<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

2<br />

, p e<br />

2 , 1 ,jX<br />

p<br />

2,1,jT 2<br />

d +<br />

2<br />

+ j p 1 ,<br />

2 e,X<br />

p 1, 2 ,jT 2<br />

d +<br />

2<br />

+ p e<br />

2 , 1 jX<br />

p<br />

2,1,jT 2<br />

d :<br />

(4.16)<br />

These equations describe the complete scatter<strong>in</strong>g process <strong>in</strong>dependently of the <strong>in</strong>itial wave.<br />

Remark (Energy and momentum space) From electromagnetic waves we a<strong>re</strong> used to<br />

decompos<strong>in</strong>g wave packets either <strong>in</strong>to a f<strong>re</strong>quency or wave number spectrum. Throughout<br />

the <strong>re</strong>ma<strong>in</strong>der of this chapter, we shall adhe<strong>re</strong> to these terms also for quantum mechanical<br />

wave packets. It should be noted, however, that quantum physicists mo<strong>re</strong> commonly talk<br />

about energy and momentum spectra. This seems a bit confus<strong>in</strong>g, but <strong>in</strong> fact is equivalent<br />

to our term<strong>in</strong>ology. The energy space cor<strong>re</strong>sponds to the f<strong>re</strong>quency spectrum because of<br />

the <strong>re</strong>lation E = ~!. The momentum <strong>in</strong> quantum mechanics is de ned as p = ~ and<br />

the<strong>re</strong>fo<strong>re</strong> cor<strong>re</strong>sponds to our usual wave number space. So, apart from the factor ~, both<br />

views a<strong>re</strong> identical.<br />

4.2 Initial wave forms<br />

The<strong>re</strong> a<strong>re</strong> two mutually exclusive choices for the <strong>in</strong>itial conditions. First, we can <strong>re</strong>strict the<br />

location of the wave packet to the space outside the barrier. In this case, the spectrum will<br />

<strong>in</strong>evitably conta<strong>in</strong> components with an energy above the barrier. Conversely, we can as well<br />

limit the bandwidth of the wave packet to energies below the barrier, but then the wave will<br />

not be bounded <strong>in</strong> space and the<strong>re</strong>fo<strong>re</strong> extend <strong>in</strong>to the barrier from the beg<strong>in</strong>n<strong>in</strong>g. We deem<br />

this mo<strong>re</strong> undesirable and thus choose the rst possibility.<br />

We now calculate the spectra of th<strong>re</strong>e di e<strong>re</strong>nt wave forms to be used as <strong>in</strong>itial conditions<br />

<strong>in</strong> a scatter<strong>in</strong>g process. The position of the wave packets is a snapshot of the very moment<br />

when they <strong>re</strong>ach the barrier.<br />

The rst and simplest one is a <strong>re</strong>ctangular pulse. It is given by<br />

(<br />

0 if x


4.2 Initial wave forms<br />

-60 -50 -40 -30 -20 -10<br />

1<br />

0.5<br />

-0.5<br />

-1<br />

Ψ0<br />

X<br />

0.15<br />

0.125<br />

0.1<br />

0.075<br />

0.05<br />

0.025<br />

A<br />

-2 -1 1 2 ξ<br />

-2 -1 1 2 ξ<br />

Figu<strong>re</strong> 4.1: <strong>Wave</strong> function and spectrum of a <strong>re</strong>ctangular pulse for k = 6 and = 0:8. The left graph<br />

shows the <strong>re</strong>al part of the wave function (th<strong>in</strong> l<strong>in</strong>e) and the cor<strong>re</strong>spond<strong>in</strong>g probability density (thick<br />

l<strong>in</strong>e). The right graph gives the absolute value jA( ) l ,1=2 j of the wave number spectrum (4.22). The<br />

evanescent part is the dark g<strong>re</strong>y a<strong>re</strong>a, and the peak of the spectrum lies at = 1=2 .<br />

which immediately gives<br />

A<strong>re</strong>ct( )= p l 1<br />

2<br />

s<strong>in</strong>( 0 , ) l<br />

2<br />

( 0 , ) l<br />

e<br />

2<br />

,j( 0, ) l 2 : (4.19)<br />

We p<strong>re</strong>fer, however, to write the spectrum with the normalised variable (4.13). To this end,<br />

we add the additional de nitions<br />

= !0<br />

!p<br />

; 2 k = 0 l: (4.20)<br />

The length l of the orig<strong>in</strong>al wave packet is thus given <strong>in</strong> a multiple k of the wavelength of<br />

the modulation f<strong>re</strong>quency. From the dispersion <strong>re</strong>lation (4.6) and (4.12) we nd the identity<br />

!p=c = 0= p , which together with (4.13) yields<br />

l =2 kp : (4.21)<br />

If we furthermo<strong>re</strong> write the s<strong>in</strong>e function <strong>in</strong> its Euler form, we obta<strong>in</strong> the normalised spectrum<br />

of the <strong>re</strong>ctangular wave packet,<br />

A <strong>re</strong>ct( )<br />

p l = j<br />

4 2 k<br />

1<br />

1 , 1<br />

p<br />

,j2 k 1, p<br />

1<br />

e<br />

, 1 : (4.22)<br />

Fig. 4.1 shows both the <strong>re</strong>ctangular wave packet and its spectrum (4.22). Note that <strong>in</strong> the<br />

scaled notation, the length l changes to 2 k= p and the phase function 0 x becomes p X.<br />

The second <strong>in</strong>itial wave packet we consider has a triangular or tent shape,<br />

0 (x) =<br />

8<br />

><<br />

>:<br />

0 if x


-60 -50 -40 -30 -20 -10<br />

3<br />

2<br />

1<br />

-1<br />

Ψ0<br />

X<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

A<br />

0.14<br />

0.12<br />

0.1<br />

0.08<br />

0.06<br />

0.04<br />

0.02<br />

-2 -1 1 2 ξ<br />

-2 -1 1 2 ξ<br />

Figu<strong>re</strong> 4.2: Triangular pulse with k = 6 and = 0:8. The left graph shows the <strong>re</strong>al part of the<br />

wave function (th<strong>in</strong> l<strong>in</strong>e) and the cor<strong>re</strong>spond<strong>in</strong>g probability density (thick l<strong>in</strong>e). The right graph gives<br />

the absolute value jA( ) l ,1=2 j of the wave number spectrum (4.25), peaked about = 1=2 . The<br />

evanescent <strong>re</strong>gion is the <strong>in</strong>terval [,1; 1].<br />

Like befo<strong>re</strong>, the factor p 3=l is due to the normalisation R 1<br />

easily found to be<br />

or, <strong>in</strong> scaled variables,<br />

A tria ( )<br />

p l<br />

p<br />

3<br />

=<br />

4 3k2 Atria( )= p p<br />

3<br />

l<br />

4<br />

1<br />

1 , 1<br />

p<br />

,1 0 0<br />

dx =1. The spectrum is<br />

s<strong>in</strong>( 0 , ) l<br />

4<br />

( 0 , ) l<br />

! 2<br />

e<br />

4<br />

,j( 0, ) l 2 (4.24)<br />

,j k 1, p<br />

1<br />

2e 2<br />

The wave packet and its spectrum a<strong>re</strong> shown <strong>in</strong> g. 4.2 .<br />

,j2 k 1, p<br />

1<br />

, e<br />

, 1 : (4.25)<br />

The last example is also the most popular <strong>in</strong> the literatu<strong>re</strong>: the Gaussian wave packet. If we<br />

start with a normal distribution N( ; 2 ) for the probability density and choose = ,l=2,<br />

= l=(2n), we get the wave function<br />

0 (x) =<br />

s<br />

2n<br />

l p 2 e,n2 ( x l +1 2) 2<br />

e j 0x<br />

: (4.26)<br />

As the envelope function is not truncated at x =0,ca<strong>re</strong> must be taken to set n su ciently<br />

large so that the wave function almost vanishes <strong>in</strong>side the barrier. The spectrum of this wave<br />

packet is simply<br />

Agauss ( )= p 1<br />

lp<br />

p e<br />

2 n 2 , ( 0<br />

78<br />

, )l<br />

2n<br />

2<br />

,j( 0, ) l 2 (4.27)


4.2 Initial wave forms<br />

-60 -50 -40 -30 -20 -10<br />

3<br />

2<br />

1<br />

-1<br />

Ψ0<br />

X<br />

0.12<br />

0.1<br />

0.08<br />

0.06<br />

0.04<br />

0.02<br />

A<br />

-2 -1 1 2 ξ<br />

-2 -1 1 2 ξ<br />

Figu<strong>re</strong> 4.3: Gaussian pulse with k =6, =0:8 , and n =4. The left graph shows the <strong>re</strong>al part of the<br />

wave function (th<strong>in</strong> l<strong>in</strong>e) and the cor<strong>re</strong>spond<strong>in</strong>g probability density (thick l<strong>in</strong>e). The right graph gives<br />

the absolute value jA( ) l ,1=2 j of the wave number spectrum (4.28). The maximum of the spectrum<br />

is at = 1=2 , and the evanescent part is colou<strong>re</strong>d dark g<strong>re</strong>y.<br />

and with the use of our scaled variables, <strong>re</strong>spectively,<br />

A gauss ( )<br />

p l<br />

=<br />

1<br />

p 2 n p 2 e<br />

,<br />

0<br />

@ 2 k p 1<br />

,1<br />

2n<br />

1<br />

A 2<br />

,j k 1, 1<br />

p<br />

Fig. 4.3 shows aga<strong>in</strong> the <strong>in</strong>itial wave packet and the cor<strong>re</strong>spond<strong>in</strong>g spectrum.<br />

: (4.28)<br />

The gu<strong>re</strong>s shown above demonstrate imp<strong>re</strong>ssively how large the portion of spectral components<br />

outside the evanescent <strong>re</strong>gion can be even though the modulation f<strong>re</strong>quency is selected<br />

from the forbidden band. The <strong>re</strong>ason is that the pulses a<strong>re</strong> all very short, i. e. they span only<br />

very few wavelengths of the carrier wave and the<strong>re</strong>fo<strong>re</strong> have a broad spectrum. This applies<br />

even to the Gaussian wave that otherwise exhibits a spectrum rapidly dec<strong>re</strong>as<strong>in</strong>g at either<br />

side of the cent<strong>re</strong> f<strong>re</strong>quency or wave number.<br />

The <strong>in</strong>spection of the spectra naturally raises the question as to what percentage of the whole<br />

wave probability density actually lies <strong>in</strong> the evanescent <strong>re</strong>gion. This quantity can easily be<br />

calculated from the spectral probability density jA( )j 2 <strong>in</strong> the <strong>in</strong>terval [,1; 1] <strong>re</strong>fer<strong>re</strong>d to the<br />

total probability as given by R 1<br />

,1 jA( )j2 d . Not surpris<strong>in</strong>gly, the total probability is the same<br />

for all th<strong>re</strong>e waveforms,<br />

Pt = lp<br />

4k<br />

2 ; (4.29)<br />

which of course stems from the normalisation of the probability density.<br />

Remark (Total probability) In pr<strong>in</strong>ciple, the<strong>re</strong> is an alternative and mo<strong>re</strong> convenient<br />

way to calculate the probability of the wave packet. Instead of <strong>in</strong>tegrat<strong>in</strong>g the absolute<br />

values of the spectra (4.22), (4.25), and (4.28), we can as well use the spatial waveform.<br />

79


0.1<br />

0.001<br />

0.00001<br />

-7<br />

1. 10<br />

-9<br />

1. 10<br />

0.1<br />

0.001<br />

0.00001<br />

-7<br />

1. 10<br />

-9<br />

1. 10<br />

ΔΡ<br />

ΔΡ<br />

0.2 0.4 0.6 0.8 1 Ω<br />

0.2 0.4 0.6 0.8 1 Ω<br />

0.1<br />

0.001<br />

0.00001<br />

-7<br />

1. 10<br />

-9<br />

1. 10<br />

0.1<br />

0.001<br />

0.00001<br />

-7<br />

1. 10<br />

-9<br />

1. 10<br />

ΔΡ<br />

ΔΡ<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

0.2 0.4 0.6 0.8 1 Ω<br />

0.2 0.4 0.6 0.8 1 Ω<br />

Figu<strong>re</strong> 4.4: Fraction of the probability contribution from the pass-band spectral components (4.32)<br />

for <strong>re</strong>ctangular (solid l<strong>in</strong>e), triangular (dotted l<strong>in</strong>e), and Gaussian (dashed l<strong>in</strong>e) <strong>in</strong>itial wave forms and<br />

short waves. The parameters a<strong>re</strong> k = 6 (upper left diagram), k = 20 (upper right diagram), k = 100<br />

(lower left diagram), and k = 1000 (lower right diagram), <strong>re</strong>spectively. The Gaussian wave packet was<br />

computed for a wide (n = 4, right dashed l<strong>in</strong>e <strong>in</strong> each pictu<strong>re</strong>) and a narrow (n= 10, left dashed l<strong>in</strong>e)<br />

example.<br />

For the Fourier transform pair (4.7) and (4.18), Parseval's theo<strong>re</strong>m <strong>re</strong>ads<br />

Z 1<br />

,1<br />

j (X)j 2 dX =2<br />

Z 1<br />

,1<br />

jA( )j 2 d : (4.30)<br />

Ow<strong>in</strong>g to normalisation, these <strong>in</strong>tegrals should give unity. The <strong>re</strong>ason why (4.29) still<br />

di ers from this value is simply the use of the scaled variables, whe<strong>re</strong> the factor p l could<br />

not be transformed. If we had calculated the probability from the orig<strong>in</strong>al spectra A( ),<br />

we would have obta<strong>in</strong>ed unity as expected. For the ratio of evanescent contribution and<br />

total probability we a<strong>re</strong> <strong>in</strong>te<strong>re</strong>sted <strong>in</strong>, the constant factor is of course ir<strong>re</strong>levant and cancels<br />

out.<br />

We now calculate the probability contribution from the spectral components <strong>in</strong> the evanescent<br />

<strong>re</strong>gion,<br />

Pe =<br />

Z 1<br />

,1<br />

jA( )j 2 d ; (4.31)<br />

80


4.3 Examples of scatter<strong>in</strong>g processes<br />

and determ<strong>in</strong>e the fraction Pe=Pt for the various <strong>in</strong>ital waveforms. To plot this ratio, however,<br />

it is mo<strong>re</strong> suitable to <strong>re</strong>gard the complementary quantity, i. e. the part of the probability that<br />

still orig<strong>in</strong>ates from the pass band,<br />

P =1, Pe<br />

Pt<br />

: (4.32)<br />

Evaluations of this <strong>re</strong>lation a<strong>re</strong> depicted <strong>in</strong> g. 4.4 <strong>in</strong> dependence on the cent<strong>re</strong> f<strong>re</strong>quency and<br />

for di e<strong>re</strong>nt lengths of the wave packet.<br />

The plots <strong>re</strong>veal that even for waves compris<strong>in</strong>g a large number of carrier wavelengths, the<br />

pass band still contributes a signi cant part of the total probability ifwe start with a squa<strong>re</strong><br />

wave. The situation is much better for triangular waves that even surpass narrow Gaussian<br />

wave packets for f<strong>re</strong>quencies slighty below cuto .<br />

4.3 Examples of scatter<strong>in</strong>g processes<br />

In this section we nally exam<strong>in</strong>e several examples of scatter<strong>in</strong>g events of electrons imp<strong>in</strong>g<strong>in</strong>g<br />

on a step potential barrier. The partial waves outside and <strong>in</strong>side the barrier a<strong>re</strong> de ned by<br />

the Fourier <strong>in</strong>tegrals (4.14) { (4.16). The appropriate wave number spectra of the <strong>in</strong>itial wave<br />

forms a<strong>re</strong> taken from section 4.2, and the <strong>in</strong>tegrals a<strong>re</strong> evaluated numerically. The plots always<br />

show the probability density function P = j j depend<strong>in</strong>g on the normalised variables of<br />

space and time, X and T .<br />

Let us beg<strong>in</strong> with an <strong>in</strong>itially <strong>re</strong>ctangular wave like <strong>in</strong> g. 4.1. As we have al<strong>re</strong>ady seen, a large<br />

portion of its spectral components lies <strong>in</strong> the pass band, which makes such a wave a rather<br />

unfortunate choice for an <strong>in</strong>vestigation of the tunnel e ect. It is, however, useful to study the<br />

e ects of partial transmission of the high-f<strong>re</strong>quency components. But befo<strong>re</strong> we do so, let us<br />

brie y consider the wave packet as such. Ow<strong>in</strong>g to its wide spectrum, we must expect that it<br />

will soon lose the orig<strong>in</strong>al shape and take on a smoother form. This suspicion can easily be<br />

con rmed by evaluat<strong>in</strong>g the <strong>in</strong>cident wave(4.14) not only to the left of the barrier, but also<br />

<strong>in</strong> the right half space, whe<strong>re</strong> it cor<strong>re</strong>sponds to a particle mov<strong>in</strong>g f<strong>re</strong>ely without obstacle. The<br />

<strong>re</strong> ected and transmitted parts of the wave a<strong>re</strong> the<strong>re</strong>fo<strong>re</strong> not of concern <strong>in</strong> this case. Fig. 4.5<br />

shows <strong>in</strong>deed that dispersion quickly changes the shape of the wave packet from the <strong>in</strong>itial<br />

<strong>re</strong>ctangle to a broad pulse with a marked peak and a number of small sidelobes.<br />

Remark (Plot ranges) In the sequel, the th<strong>re</strong>e-dimensional plots of the probability<br />

density outside the barrier a<strong>re</strong> always accompanied by a contour plot of the same function.<br />

To allow the perception of as many details as possible, the plots often deliberately do not<br />

show the full range of function values. This is particularly important for the contour plots<br />

to avoid an exceed<strong>in</strong>gly large number of contour l<strong>in</strong>es. The bright white holes <strong>in</strong> these<br />

plots thus <strong>in</strong>dicate the a<strong>re</strong>as that have been clipped.<br />

Now let this particle collide with the step potential. The plot of the probability density<br />

outside the barrier ( g. 4.6) shows the <strong>in</strong>cident wave packet mov<strong>in</strong>g towards the barrier while<br />

al<strong>re</strong>ady los<strong>in</strong>g its shape. The actual scatter<strong>in</strong>g event is characterised by strong <strong>in</strong>terfe<strong>re</strong>nce<br />

81


30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

0<br />

-20<br />

0<br />

30<br />

X<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

20<br />

20<br />

T<br />

10<br />

-20 -10 0 10 20 30 40<br />

0<br />

0<br />

40<br />

Figu<strong>re</strong> 4.5: Evolution of the probability density P = j j of a f<strong>re</strong>ely mov<strong>in</strong>g wave packet (4.14) with<br />

<strong>in</strong>itially <strong>re</strong>ctangular shape for the parameters k =3and =0:8.<br />

82<br />

1.5<br />

1<br />

0.5<br />

P


4.3 Examples of scatter<strong>in</strong>g processes<br />

40<br />

30<br />

20<br />

10<br />

-40<br />

-30<br />

40<br />

30<br />

-20<br />

X<br />

T<br />

20<br />

-10<br />

10<br />

0<br />

-40 -30 -20 -10 0<br />

0<br />

2<br />

1<br />

3<br />

0<br />

0<br />

Figu<strong>re</strong> 4.6: Evolution of the probability density P = j j of a scatter<strong>in</strong>g wave packet with <strong>in</strong>itially<br />

<strong>re</strong>ctangular shape outside the barrier. The parameters a<strong>re</strong> k = 3 and = 0:8.<br />

83<br />

4<br />

P


30<br />

1<br />

0.88<br />

0.6 .6<br />

P 0.4 .4<br />

0.2 .2<br />

0<br />

0<br />

20<br />

10<br />

X<br />

10<br />

20<br />

T<br />

0<br />

30<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Figu<strong>re</strong> 4.7: Evolution of the probability density P = j j of a scatter<strong>in</strong>g wave packet with <strong>in</strong>itially<br />

<strong>re</strong>ctangular shape with<strong>in</strong> the barrier. The parameters a<strong>re</strong> k =3and =0:8. The spectral components<br />

with higher energy than the edge of the barrier propagate further.<br />

of the emerg<strong>in</strong>g <strong>re</strong> ected wave and the trail<strong>in</strong>g tail of the still arriv<strong>in</strong>g wave. The ultimately<br />

<strong>re</strong> ected part exhibits a smooth shape bear<strong>in</strong>g no <strong>re</strong>semblance to the orig<strong>in</strong>al <strong>re</strong>ctangle.<br />

Inside the barrier ( g. 4.7), we notice the rapid decay of the wave function orig<strong>in</strong>at<strong>in</strong>g from<br />

the evanescent spectral components. The c<strong>re</strong>sts and troughs runn<strong>in</strong>g not <strong>in</strong> parallel to the<br />

axes, however, a<strong>re</strong> patterns characteristic for propagat<strong>in</strong>g waves. We also notice very small<br />

but ir<strong>re</strong>gular ripples for times immediately after the wave beg<strong>in</strong>s to penetrate the barrier.<br />

These a<strong>re</strong> <strong>in</strong> fact the spectral components of the pass band that a<strong>re</strong> be<strong>in</strong>g transmitted. For<br />

about T < 10, the computation grid is too coarse to <strong>re</strong>solve the high-f<strong>re</strong>quency oscillations.<br />

We shall have a closer look at them later.<br />

Remark (Actual physical dimensions) At this po<strong>in</strong>t it is sensible give some <strong>re</strong>alworld<br />

values for the normalised variables X = !px=c and T = !pt we use to scale the<br />

pictu<strong>re</strong>s. In terms of the barrier energy V = !p~ and with c = p !p~=(2m), they <strong>re</strong>ad<br />

T = V<br />

~ t; X=<br />

r<br />

2mV<br />

~ 2<br />

: (4.33)<br />

If we assume the e ective mass of the imp<strong>in</strong>g<strong>in</strong>g electron to be 0:063 m0 with the <strong>re</strong>st<br />

mass m0 =9:12 10 ,31 kg, we obta<strong>in</strong> for two widely used barrier energies [83] the values<br />

given <strong>in</strong> table 4.1.<br />

84<br />

40


4.3 Examples of scatter<strong>in</strong>g processes<br />

P<br />

30<br />

2<br />

1.55<br />

1<br />

0.5 .5<br />

0<br />

0<br />

20<br />

10<br />

X<br />

10<br />

20<br />

T<br />

0<br />

30<br />

Figu<strong>re</strong> 4.8: Evolution of the probability density P = j j of a triangular wave packet with<strong>in</strong> the<br />

barrier. The parameters a<strong>re</strong> k =3and =0:8.<br />

40<br />

barrier energy T =1b= X=1b=<br />

0:3eV=4:8 10 ,20 J t =2:194 fs x =1:42 nm = 14:2A<br />

0:1eV=1:6 10 ,20 J t =6:582 fs x =2:46 nm = 24:6A<br />

Table 4.1: Actual dimensions for typical barriers.<br />

We now move on to triangular electrons, the spectrum of which is far mo<strong>re</strong> well-behaved<br />

than that of a <strong>re</strong>ctangular particle. For the rst example we use the same parameters as<br />

<strong>in</strong> the p<strong>re</strong>vious case. The behaviour outside the barrier ( g. 4.9) is essentially the same as<br />

befo<strong>re</strong>, but the wave forms a<strong>re</strong> much smoother now. The sharp peak of the <strong>in</strong>itial wave<br />

packet naturally disappears after a short time, and the <strong>re</strong> ected wave al<strong>re</strong>ady <strong>re</strong>sembles a<br />

Gaussian distribution. The same observation can be made for the part that penetrates the<br />

barrier ( g. 4.8). S<strong>in</strong>ce the parameters we<strong>re</strong> chosen such that a fairly large contribution still<br />

comes from the pass band, we <strong>re</strong>cognise a soft c<strong>re</strong>st mov<strong>in</strong>g ahead. Actually, the smoothlook<strong>in</strong>g<br />

wave function is superposed with small high-f<strong>re</strong>quent ripples that, like <strong>in</strong> the p<strong>re</strong>vious<br />

example, cannot be <strong>re</strong>solved <strong>in</strong> this plot.<br />

The wave packets conside<strong>re</strong>d so far we<strong>re</strong> ext<strong>re</strong>mely short (i. e. compris<strong>in</strong>g very few wavelengths<br />

of the carrier), which consequently <strong>re</strong>sulted <strong>in</strong> a broad spectrum. Let us now turn to the mo<strong>re</strong><br />

85


P<br />

0<br />

6<br />

4<br />

2<br />

0<br />

-40<br />

10<br />

40<br />

30<br />

20<br />

10<br />

T<br />

20<br />

-30<br />

30<br />

40<br />

-20<br />

X<br />

-10<br />

0<br />

-40 -30 -20 -10 0<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Figu<strong>re</strong> 4.9: Evolution of the probability density P = j j of a scatter<strong>in</strong>g triangular wave packet<br />

outside the barrier. The parameters a<strong>re</strong> k =3and =0:8.<br />

86<br />

0


4.3 Examples of scatter<strong>in</strong>g processes<br />

P<br />

8<br />

6<br />

4<br />

0<br />

2<br />

0<br />

-150<br />

50<br />

175<br />

150<br />

125<br />

100<br />

75<br />

50<br />

25<br />

0<br />

T<br />

100<br />

-100<br />

150<br />

X<br />

-50<br />

-140 -120 -100 -80 -60 -40 -20 0<br />

Figu<strong>re</strong> 4.10: Evolution of the probability density P = j j of a longer triangular wave packet outside<br />

the barrier. The parameters a<strong>re</strong> k =10and =0:2.<br />

87<br />

0


P<br />

0<br />

2<br />

1.5<br />

1<br />

0.5<br />

0<br />

0<br />

1<br />

50<br />

2<br />

X<br />

T<br />

100<br />

3<br />

150<br />

4<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Figu<strong>re</strong> 4.11: Evolution of the probability density P = j j of a longer triangular wave packet with<strong>in</strong><br />

the barrier. The parameters a<strong>re</strong> k = 10 and = 0:2.<br />

<strong>re</strong>alistic case of a longer electron with triangular shape. In addition, we choose a carrier wave<br />

with lower energy. The contribution of the pass band to the total probability density <strong>in</strong>the<br />

<strong>in</strong>itial wave form is then low enough that we may expect `proper' penetration of the wave<br />

<strong>in</strong>to the barrier. This <strong>re</strong>duces the disturb<strong>in</strong>g e ects of the transmitted components, although<br />

we will of course never <strong>re</strong>ally get rid of them. S<strong>in</strong>ce the wave spans mo<strong>re</strong> wavelengths of the<br />

carrier, <strong>in</strong>terfe<strong>re</strong>nce between <strong>in</strong>cident and <strong>re</strong> ected part generates mo<strong>re</strong> peaks, as can be seen<br />

from g. 4.10 .<br />

Inside the barrier, the probability density is largely determ<strong>in</strong>ed by evanescent components,<br />

so that the overall shape of the function decays exponentially ( g. 4.11). However, a closer<br />

<strong>in</strong>spection of the probability would aga<strong>in</strong> <strong>re</strong>veal high-f<strong>re</strong>quency ripples cover<strong>in</strong>g the surface at<br />

least for small values of T . Ow<strong>in</strong>g to the choice parameters, they a<strong>re</strong> very small <strong>in</strong> comparison<br />

with the evanescent part, but nonetheless become prom<strong>in</strong>ent for large values of X, aswe shall<br />

see shortly.<br />

Remark (Evolution of the peak) The contour plots of the scatter<strong>in</strong>g events give<br />

a bright illustration of the di culties associated with the phase time concept. In this<br />

approach, the trajectory of the peak of the wave packet determ<strong>in</strong>es the tunnell<strong>in</strong>g time.<br />

But as the plots show, it is impossible to identify the peak <strong>in</strong> front of the barrier because<br />

of the <strong>in</strong>terfe<strong>re</strong>nce of <strong>in</strong>cident and <strong>re</strong> ected wave. Hence the trajectory can only be<br />

extrapolated from the evolution of the wave packet far away from the barrier whe<strong>re</strong><br />

<strong>in</strong>terfe<strong>re</strong>nce has not yet distorted its shape (see also the <strong>re</strong>spective <strong>re</strong>mark <strong>in</strong> section 2.2).<br />

The last example shows a Gaussian wave packet with a narrow spectrum. The parameters<br />

a<strong>re</strong> the same as those used <strong>in</strong> g. 4.4 , so that the fraction of the pass-band components is<br />

given by the left graph <strong>in</strong> g. 4.4 . The plots of the scatter<strong>in</strong>g process <strong>in</strong> front ofthebarrier<br />

88


4.3 Examples of scatter<strong>in</strong>g processes<br />

P<br />

10<br />

7.5<br />

5<br />

2.5<br />

0<br />

-100<br />

140<br />

120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

-75<br />

-50<br />

X<br />

-25<br />

0<br />

-100 -80 -60 -40 -20 0<br />

Figu<strong>re</strong> 4.12: Evolution of the probability density P = j j of a scatter<strong>in</strong>g Gaussian wave packet<br />

outside the barrier. The parameters a<strong>re</strong> k =6,n=4,and =0:2.<br />

89<br />

0<br />

0<br />

50<br />

100<br />

150<br />

T


P<br />

2<br />

1.5<br />

1<br />

0.5<br />

0<br />

0<br />

2<br />

X<br />

4<br />

0<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Figu<strong>re</strong> 4.13: Evolution of the probability density P = j j of a Gaussian wave packet with<strong>in</strong> the<br />

barrier. The parameters a<strong>re</strong> k =6,n=4,and =0:2.<br />

( g. 4.12) provide no new or unexpected <strong>in</strong>formation, except that this time the shape of the<br />

wave packet seems to <strong>re</strong>ma<strong>in</strong> una ected throughout the whole event. Also <strong>in</strong>side the barrier<br />

( g. 4.13), the shape of the probability density looks Gaussian, and like befo<strong>re</strong>, the<strong>re</strong> a<strong>re</strong> no<br />

over-the-edge components that keep on propagat<strong>in</strong>g.<br />

Remark (Trajectory of the maximum) Suppose we would, at every po<strong>in</strong>t whe<strong>re</strong><br />

X > 0, determ<strong>in</strong>e when the maximum of the wave arrives. From g. 4.13, we could<br />

expect this <strong>in</strong>stant to be the same everywhe<strong>re</strong> <strong>in</strong>dependent of the position along the<br />

barrier. Astonish<strong>in</strong>gly enough, this is not the case, and we nd that the farther we shift<br />

the po<strong>in</strong>t of observation <strong>in</strong>to the barrier, the earlier the peak is detected. Accord<strong>in</strong>gly,<br />

the enti<strong>re</strong> pulse becomes narrower, apart of course from the exponentially dim<strong>in</strong>ish<strong>in</strong>g<br />

amplitude. While this <strong>re</strong>sult looks rather weird at rst sight, the<strong>re</strong> seems to be a plausible<br />

explanation: the front ofthe pulse, as mentioned al<strong>re</strong>ady <strong>in</strong> section 2.1, consists of the<br />

higher-f<strong>re</strong>quency components of the spectrum, i. e. those with a high E = !~. If they a<strong>re</strong><br />

evanescent, which we p<strong>re</strong>sume he<strong>re</strong>, they experience no phase shift as they penetrate <strong>in</strong>to<br />

the barrier. Their stationary penetration depth, however, is given by the <strong>re</strong>ciprocal of the<br />

propagation constant = j 1p<br />

~ 2m(V , E) and tends to <strong>in</strong> nity as the E approaches the<br />

barrier energy V . Hence the higher-f<strong>re</strong>quency components <strong>re</strong>ach deeper <strong>in</strong>to the barrier<br />

than the low-f<strong>re</strong>quency components that make up the tail of the pulse, and this is why<br />

the pulse seems to be shorter <strong>in</strong> a g<strong>re</strong>ater distance.<br />

The grid chosen for the computation of the p<strong>re</strong>ced<strong>in</strong>g th<strong>re</strong>e-dimensional pictu<strong>re</strong>s does of course<br />

not allow for the <strong>re</strong>solution of the ne structu<strong>re</strong>s caused by the high-f<strong>re</strong>quency components<br />

of the waves. It is the<strong>re</strong>fo<strong>re</strong> worthwhile to <strong>re</strong>-exam<strong>in</strong>e the evolution of the waves <strong>in</strong>side the<br />

barrier at selected positions, namely di<strong>re</strong>ctly at the <strong>in</strong>terface (X = 0) and su ciently far<br />

down the l<strong>in</strong>e (X = 10) whe<strong>re</strong> the evanescent parts of the spectra should have vanished<br />

compa<strong>re</strong>d with the transmitted components. In contrast to the p<strong>re</strong>vious examples, however,<br />

90<br />

50<br />

T<br />

100<br />

150


4.3 Examples of scatter<strong>in</strong>g processes<br />

2.5<br />

2<br />

1.5<br />

1<br />

0.5<br />

P<br />

0 20 40 60 80 100 120 140 T<br />

2.5<br />

2<br />

1.5<br />

1<br />

0.5<br />

P<br />

0 20 40 60 80 100 120 140 T<br />

Figu<strong>re</strong> 4.14: Pulse shapes produced by several short wave packets at the <strong>in</strong>terface of the barrier. The<br />

parameter a<strong>re</strong> k = 6 and = 0:2. The left graph shows the <strong>re</strong>ctangular (solid l<strong>in</strong>e) and triangular<br />

(dotted l<strong>in</strong>e) packet. On the right side, the <strong>re</strong>sults of a narrow (n= 10, solid l<strong>in</strong>e) and a wide (n =4,<br />

dotted l<strong>in</strong>e) Gaussian packet a<strong>re</strong> given.<br />

0.014<br />

0.012<br />

0.01<br />

0.008<br />

0.006<br />

0.004<br />

0.002<br />

P<br />

-6<br />

8. 10<br />

-6<br />

6. 10<br />

-6<br />

4. 10<br />

-6<br />

2. 10<br />

P<br />

20 40 60 80 100 120 140 T<br />

20 40 60 80 100 120 140 T<br />

0.0001<br />

0.00008<br />

0.00006<br />

0.00004<br />

0.00002<br />

-8<br />

5. 10<br />

-8<br />

4. 10<br />

-8<br />

3. 10<br />

-8<br />

2. 10<br />

-8<br />

1. 10<br />

P<br />

P<br />

20 40 60 80 100 120 140 T<br />

20 40 60 80 100 120 140 T<br />

Figu<strong>re</strong> 4.15: Pulse shapes produced by several short wave packets at X = 10 <strong>in</strong>side the barrier. The<br />

parameter a<strong>re</strong> k =6and =0:2. The graphs show the <strong>re</strong>ctangular (upper left), triangular (upper<br />

right), narrow Gaussian (n = 10, lower left), and wide Gaussian (n =4,lower right) packet.<br />

91


5<br />

4<br />

3<br />

2<br />

1<br />

P<br />

0 100 200 300 400 T<br />

5<br />

4<br />

3<br />

2<br />

1<br />

P<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

0 100 200 300 400 T<br />

Figu<strong>re</strong> 4.16: Pulse shapes produced by several long wave packets at the <strong>in</strong>terface of the barrier. The<br />

parameter a<strong>re</strong> k = 20 and = 0:2. The left graph shows the <strong>re</strong>ctangular (solid l<strong>in</strong>e) and triangular<br />

(dotted l<strong>in</strong>e) packet. In the right graph, the <strong>re</strong>sults of a narrow (n= 10, solid l<strong>in</strong>e) and a wide (n =4,<br />

dotted l<strong>in</strong>e) Gaussian packet a<strong>re</strong> given.<br />

0.012<br />

0.01<br />

0.008<br />

0.006<br />

0.004<br />

0.002<br />

P<br />

100 200 300 400 T<br />

-6<br />

6. 10<br />

-6<br />

5. 10<br />

-6<br />

4. 10<br />

-6<br />

3. 10<br />

-6<br />

2. 10<br />

-6<br />

1. 10<br />

P<br />

100 200 300 400 T<br />

P<br />

-7<br />

1. 10<br />

-8<br />

8. 10<br />

-8<br />

6. 10<br />

-8<br />

4. 10<br />

-8<br />

2. 10<br />

100 200 300 400 T<br />

Figu<strong>re</strong> 4.17: Pulse shapes produced by several long wave packets at X = 10 <strong>in</strong>side the barrier. The<br />

parameter a<strong>re</strong> k =20and =0:2. The graphs show the <strong>re</strong>ctangular (left), triangular (middle), and<br />

Gaussian (right) packet. In the latter, the <strong>re</strong>sults for a narrow (n= 10, solid l<strong>in</strong>e) and wide (n =4,<br />

dotted l<strong>in</strong>e) packet a<strong>re</strong> displayed.<br />

we nowchoose the parameters so as to obta<strong>in</strong> as narrow a spectrum as possible. To permit a<br />

di<strong>re</strong>ct comparison of the spectra, we aga<strong>in</strong> take the parameter values of g. 4.4 together with<br />

alow-f<strong>re</strong>quency carrier ( = 0:2).<br />

Fig. 4.14 shows the pulses of a short electron (k = 6) at the <strong>in</strong>terface of the barrier. Not<br />

surpris<strong>in</strong>gly, the <strong>re</strong>ctangular wave packet exhibits strong oscillations. As a di<strong>re</strong>ct consequence<br />

of dispersion, the temporal f<strong>re</strong>quency of these oscillations dec<strong>re</strong>ases rapidly <strong>in</strong> the course of<br />

time. On the other hand the triangular pulse, though hav<strong>in</strong>g also spectral components <strong>in</strong> the<br />

pass band, shows no such oscillations. The graph also <strong>re</strong>veals that the Gaussian wave packets<br />

do not produce Gaussian pulses, but wave forms with a <strong>re</strong>markable skewness. This e ect is<br />

mo<strong>re</strong> marked if the wave packet is narrow and the<strong>re</strong>fo<strong>re</strong> has a broad spectrum, as expla<strong>in</strong>ed<br />

above.<br />

Inside the barrier ( g. 4.15), the wave forms undergo signi cant changes. For the <strong>re</strong>ctangular<br />

wave packet, the high-f<strong>re</strong>quency components that can propagate <strong>in</strong>to the barrier become<br />

mo<strong>re</strong> pronounced. This also goes for the triangular packet, which had looked so smooth<br />

at the entrance. The<strong>re</strong> the evanescent parts had obviously dom<strong>in</strong>ated the shape, but now<br />

92


4.4 The squa<strong>re</strong> barrier<br />

they have vanished, and only the over-the-barrier contributions <strong>re</strong>ma<strong>in</strong>. Even the narrow<br />

Gaussian pulse b<strong>re</strong>aks up <strong>in</strong>to dist<strong>in</strong>ct oscillations. Only the orig<strong>in</strong>ally wide Gaussian wave<br />

packet <strong>re</strong>ta<strong>in</strong>s its shape, of course with a dramatically <strong>re</strong>duced amplitude. So it is just this<br />

last case whe<strong>re</strong> we can assume the spectrum to have been narrow enough and su ciently<br />

below the top of the barrier. Note also that the amplitude of the narrow pulse is about<br />

two orders of magnitude larger than that of the wide pulse, which undeniably identi es the<br />

high-f<strong>re</strong>quency components.<br />

The last examples explo<strong>re</strong> a longer wave packet (k = 20) at the same positions. At the<br />

<strong>in</strong>terface ( g. 4.16), the situation has not changed very much. The<strong>re</strong> a<strong>re</strong> still oscillations,<br />

although with a lower amplitude, <strong>in</strong> the <strong>re</strong>ctangular pulse, and its shape can now be identi ed<br />

much easier than befo<strong>re</strong>. The skewness of the Gaussian pulses is much smaller now, accord<strong>in</strong>g<br />

to the narrower spectrum. With<strong>in</strong> the barrier ( g. 4.17), it is <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g to notice that the<br />

rst peak of the <strong>re</strong>ctangular wave packet has about the same height as for the short electron<br />

<strong>in</strong> g. 4.15. This demonstrates once mo<strong>re</strong> that the number of carrier wavelengths the particle<br />

embraces has no g<strong>re</strong>at <strong>in</strong> uence on the fraction of pass-band spectral components, as can<br />

also be seen from g. 4.4 . The oscillations that <strong>re</strong>ma<strong>in</strong> from the triangular wave packet,<br />

however, have become smaller as compa<strong>re</strong>d with the short particle befo<strong>re</strong>. But the most<br />

strik<strong>in</strong>g di e<strong>re</strong>nce is the behaviour of the narrow Gaussian pulse, whose spectrum this time<br />

is also enti<strong>re</strong>ly below the barrier level. Consequently, the <strong>re</strong>lation between the amplitudes<br />

<strong>in</strong>side the barrier is roughly the same as at the <strong>in</strong>terface.<br />

4.4 The squa<strong>re</strong> barrier<br />

The <strong>in</strong>vestigation of the penetration of an electron <strong>in</strong>to a step potential barrier provided<br />

al<strong>re</strong>ady some <strong>in</strong>sight <strong>in</strong>to the behaviour of evanescent waves <strong>in</strong> quantum mechanics. It did<br />

not, however, give an answer to the question of the tunnell<strong>in</strong>g time of a particle, because an<br />

<strong>in</strong> nitely extended step barrier is of course always opaque and permits no tunnell<strong>in</strong>g through.<br />

Neither can the <strong>re</strong>sults of the step potential be easily extrapolated to a squa<strong>re</strong> barrier due<br />

to the p<strong>re</strong>sence of a second <strong>in</strong>terface that will <strong>in</strong>evitably cause <strong>re</strong> ections and <strong>in</strong>terfe<strong>re</strong>nce.<br />

The<strong>re</strong> is thus no choice other than to exam<strong>in</strong>e also the case of a squa<strong>re</strong> potential barrier<br />

V (x) =<br />

8<br />

><<br />

>:<br />

0 if xd<br />

: (4.34)<br />

S<strong>in</strong>ce the<strong>re</strong> a<strong>re</strong> now th<strong>re</strong>e dist<strong>in</strong>ct <strong>re</strong>gions, we have th<strong>re</strong>e waves to determ<strong>in</strong>e. To the left of<br />

the barrier, the<strong>re</strong> is the <strong>in</strong>cident and a <strong>re</strong> ected part,<br />

1 = C 1 e ,j!t e j 1x + C2 e ,j!t e ,j 1x : (4.35)<br />

Inside the barrier, also a right- and a left-go<strong>in</strong>g solution will exist,<br />

2 = C 3 e ,j!t e j 2x + C4 e ,j!t e ,j 2x ; (4.36)<br />

93


4 One-dimensional quantum tunnell<strong>in</strong>g<br />

whe<strong>re</strong>as beh<strong>in</strong>d the barrier, only the transmitted part of the wave will <strong>re</strong>ma<strong>in</strong>,<br />

3 = C 5 e ,j!t e j 1x : (4.37)<br />

To obta<strong>in</strong> the coe cients, we set C 1 = 1 and establish the cont<strong>in</strong>uity conditions of the wave<br />

functions and their spatial derivatives at the <strong>in</strong>terfaces x = 0 and x = d, <strong>re</strong>spectively,<br />

C 1 + C 2 = C 3 + C 4<br />

1C 1 , 1C 2 = 2C 3 , 2C 4<br />

C 3 e j 2d + C4 e ,j 2d = C5 e j 1d<br />

2C 3 e j 2d , 2C 4 e ,j 2d = 1C 5 e j 1d :<br />

Solv<strong>in</strong>g this set of equations, we nd the coe cients<br />

j( 2<br />

C2 =<br />

2 , 1 2 ) s<strong>in</strong> 2d<br />

2 1 2 cos 2d , j( 1 2 + 2 2 ) s<strong>in</strong> 2d<br />

( 1 2 + 1 2 )e ,j 2d<br />

C3 =<br />

2 1 2 cos 2d , j( 1 2 + 2 2 ) s<strong>in</strong> 2d<br />

( 1 2 , 1<br />

C4 =<br />

2 )ej 2d<br />

2 1 2 cos 2d , j( 1 2 + 2 2 ) s<strong>in</strong> 2d<br />

C 5 =<br />

2 1 2e ,j 1d<br />

2 1 2 cos 2d , j( 1 2 + 2 2 ) s<strong>in</strong> 2d :<br />

(4.38)<br />

(4.39)<br />

If we furthermo<strong>re</strong> exp<strong>re</strong>ss the propagation constant <strong>in</strong>side the barrier, 2, bythe one outside as<br />

we did <strong>in</strong> (4.9) and use our normalised variables (4.12) and (4.13) together with the normalised<br />

barrier thickness D = !p<br />

d, the coe c cients become<br />

C 2 = ,<br />

C 3 =<br />

C 4 =<br />

C 5 =<br />

j s<strong>in</strong> D p 2 , 1<br />

2 p 2 , 1 cos D p 2 , 1 , j(2 2 , 1) s<strong>in</strong> D p 2 , 1<br />

( p p<br />

2 2 ,jD 2,1<br />

, 1+ )e<br />

2 p 2 , 1 cos D p 2 , 1 , j(2 2 , 1) s<strong>in</strong> D p 2 , 1<br />

( p p<br />

2 2 jD 2,1<br />

, 1 , )e<br />

2 p 2 , 1 cos D p 2 , 1 , j(2 2 , 1) s<strong>in</strong> D p 2 , 1<br />

2 p 2 , 1e ,j D<br />

2 p 2 , 1 cos D p 2 , 1 , j(2 2 , 1) s<strong>in</strong> D p 2 , 1 :<br />

94<br />

(4.40)


4.4 The squa<strong>re</strong> barrier<br />

1<br />

0.8<br />

0.6<br />

0.4<br />

0.2<br />

|C5|<br />

0.5 1 1.5 2<br />

Figu<strong>re</strong> 4.18: Transmission coe cient jC 5j depend<strong>in</strong>g on the normalised propagation constant for<br />

several values of the barrier thickness D (dotted l<strong>in</strong>e: 1, dashed l<strong>in</strong>e: 5, solid l<strong>in</strong>e: 10). Note the<br />

<strong>re</strong>lation 2 = !=!p.<br />

As for the signs of the squa<strong>re</strong> roots, the same considerations as <strong>in</strong> section 4.1 apply, so<br />

that for < ,1, the negative value of the squa<strong>re</strong> roots must be taken, whe<strong>re</strong>as > 1<br />

cor<strong>re</strong>sponds to the positive solution. Note that the wave functions <strong>in</strong>side the barrier we<strong>re</strong><br />

de ned for the transmissive case. If the energy of the monochromatic wave is below V0, then<br />

p 2 , 1 7! j p 1 , 2 , and hyperbolic functions take the places of the circular functions.<br />

The transmission coe cient, jC 5j, is shown <strong>in</strong> g. 4.18 <strong>in</strong> dependence of the barrier thickness.<br />

Not surpris<strong>in</strong>gly, the thicker the barrier, the closer the f<strong>re</strong>quency of an evanescent wave,<br />

!, must be to the cuto f<strong>re</strong>quency, !p, <strong>in</strong> order to allow a noticeable fraction to tunnel<br />

through. Above the top of the barrier ( >1), the transmission coe cient exhibits oscillations<br />

depend<strong>in</strong>g on as well as D, and only very high-f<strong>re</strong>quency waves a<strong>re</strong> transmitted enti<strong>re</strong>ly.<br />

With the coe cients for the <strong>in</strong>dividual parts of the wave, we can formulate the wave function<br />

<strong>in</strong> the th<strong>re</strong>e <strong>re</strong>gions as Fourier <strong>in</strong>tegrals with an arbitrary spectrum of the <strong>in</strong>itial wave, A( ),<br />

1 = !p<br />

c<br />

Z 1<br />

A( ) e<br />

,1<br />

jX ,j 2T d ,<br />

Z 1<br />

!p<br />

c ,1<br />

A( ) j s<strong>in</strong> Dp 2 , 1<br />

N( )<br />

95<br />

e ,jX ,j 2 T d ;<br />

ξ<br />

(4.41)


2 = !p<br />

c<br />

3 = !p<br />

c<br />

!p<br />

c<br />

Z 1<br />

,1<br />

Z 1<br />

Z 1<br />

,1<br />

,1<br />

A( )<br />

A( )<br />

p 2 , 1+ 2<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

e<br />

N( )<br />

j(X,D)<br />

p<br />

2,1,jT 2<br />

d +<br />

p<br />

2 2 , 1 ,<br />

N( )<br />

A( ) 2 p 2 , 1<br />

N( )<br />

e ,j(X,D)<br />

p<br />

2,1,jT 2<br />

d ;<br />

(4.42)<br />

2<br />

j(X,D) ,jT<br />

e d ; (4.43)<br />

whe<strong>re</strong> we set N( )=2 p 2 ,1 cos D p 2 , 1 , j(2 2 , 1) s<strong>in</strong> D p 2 , 1.<br />

4.5 Tunnell<strong>in</strong>g time de nitions for a squa<strong>re</strong> barrier<br />

Befo<strong>re</strong> look<strong>in</strong>g at the time-dependent evolution of wave packets imp<strong>in</strong>g<strong>in</strong>g on the squa<strong>re</strong><br />

barrier of the p<strong>re</strong>vious section, we brie y discuss some well-known approaches to compute<br />

the tunnell<strong>in</strong>g time through such anobstacle. They all apply to monochromatic waves, but<br />

have also been used to describe the behaviour of small-band pulses, <strong>in</strong> particular Gaussian<br />

wave packets.<br />

For a start, we consider the time it takes for a f<strong>re</strong>e particle to travel a distance d. This can<br />

easily be found from the dwell time de nition (2.17), which with the monochromatic wave<br />

= C1ej( x,!t) and the propagation constant = p 2m~!=~ yields<br />

f = d<br />

2<br />

r<br />

2m<br />

: (4.44)<br />

~!<br />

We could obta<strong>in</strong> exactly the same <strong>re</strong>sult with the application of the group velocity de nition<br />

and the `classical' argument that the propagation time is the distance d divided by the group<br />

velocity 1=vg = p m=(2~!). With the normalised variables (4.12), this time becomes<br />

f = d<br />

p : (4.45)<br />

2c<br />

In the sequel, we shall use this undisturbed propagation delay of a f<strong>re</strong>e electron as a <strong>re</strong>fe<strong>re</strong>nce<br />

for other concepts.<br />

With a consideration similar to the afo<strong>re</strong>mentioned, we can derive a simple rst-cut approximation<br />

to the barrier traversal time. Suppos<strong>in</strong>g that the particle moves f<strong>re</strong>ely <strong>in</strong>side a<br />

potential barrier with the <strong>re</strong>sidual energy E , V0, we <strong>re</strong>place ! <strong>in</strong> the above exp<strong>re</strong>ssion by<br />

! , !p and get the so-called semi-classical time [83, 86],<br />

d<br />

s =<br />

2c p : (4.46)<br />

, 1<br />

Note that for waves with energy below the top of the barrier, this time is imag<strong>in</strong>ary, a fact<br />

that has widely been critisised. Refer<strong>re</strong>d to the f<strong>re</strong>e particle, the semi-classical time <strong>re</strong>ads<br />

s<br />

f<br />

=<br />

r<br />

96<br />

: (4.47)<br />

, 1


4.5 Tunnell<strong>in</strong>g time de nitions for a squa<strong>re</strong> barrier<br />

The de nition of the dwell time has al<strong>re</strong>ady been given <strong>in</strong> (2.17). For a <strong>re</strong>ctangular barrier,<br />

this time is found to be [83, 88]<br />

D = mk<br />

~<br />

2 d( 2 , k2 )+k0 2 s<strong>in</strong>h 2 d<br />

4k2 2 + k0 4 s<strong>in</strong>h 2 ; (4.48)<br />

d<br />

whe<strong>re</strong> <strong>in</strong> our notation k 2 =2m!=~, k0 2 =2m!p=~, and 2 = k0 2 , k 2 . Us<strong>in</strong>g scaled variables<br />

and D = d!p=c, the exp<strong>re</strong>ssion can be written as<br />

D<br />

f<br />

=2<br />

1,2 + s<strong>in</strong>h 2Dp 1,<br />

2D p 1,<br />

4 (1 , ) + s<strong>in</strong>h 2 D p 1 ,<br />

: (4.49)<br />

Buttiker [88] derived a traversal time based on the idea of a quantum clock. This traversal<br />

time is determ<strong>in</strong>ed by the p<strong>re</strong>cession of the sp<strong>in</strong> a particle experiences <strong>in</strong> a small magnetic<br />

eld con ned to the barrier. This Buttiker time (or Buttiker-Landauer time, as it was called<br />

<strong>in</strong> the survey of Hauge and St vneng [85]), consists of two components,<br />

with the dwell time d given by (4.48) and the Larmor time<br />

z =<br />

mk0 2<br />

~ 2<br />

In normalised notation, this <strong>re</strong>ads<br />

z<br />

f<br />

=<br />

r<br />

1 ,<br />

B = p d 2 + z 2 ; (4.50)<br />

( 2 , k2 ) s<strong>in</strong>h 2 d + 1<br />

2 dk0 2 s<strong>in</strong>h 2 d<br />

4k2 2 + k0 4 s<strong>in</strong>h 2 : (4.51)<br />

d<br />

(1 , 2 ) s<strong>in</strong>h2 D p 1,<br />

D p 1,<br />

+ 1<br />

2 s<strong>in</strong>h 2Dp 1 ,<br />

4 (1 , ) + s<strong>in</strong>h 2 D p 1 ,<br />

: (4.52)<br />

The last one of the tunnell<strong>in</strong>g time de nitions conside<strong>re</strong>d he<strong>re</strong> is the phase time, which<br />

essentially describes the movement of the peak of a wave packet. For our squa<strong>re</strong> barrier, it is<br />

given <strong>in</strong> the literatu<strong>re</strong> [83] as<br />

p = m<br />

~k<br />

which can nally be <strong>re</strong>written as<br />

p<br />

f<br />

=2<br />

2 dk2 ( 2 , k2 )+k0 2 s<strong>in</strong>h 2 d<br />

4k2 2 + k0 4 s<strong>in</strong>h 2 ; (4.53)<br />

d<br />

(1 , 2 )+ s<strong>in</strong>h 2Dp 1,<br />

2D p 1,<br />

4 (1 , ) + s<strong>in</strong>h 2 D p 1 ,<br />

: (4.54)<br />

Prior to compar<strong>in</strong>g these four traversal times for two selected cases, we add a fth based<br />

on the dwell time de nition. We <strong>re</strong>call that the dwell time was de ned as the ratio of the<br />

probability density <strong>in</strong> the barrier and the <strong>in</strong>cident ux, <strong>in</strong>tegrated over the barrier <strong>re</strong>gion.<br />

The <strong>in</strong>cident ux, however, conta<strong>in</strong>s also a portion that will be <strong>re</strong> ected from the barrier, and<br />

it is by no means evident why this <strong>re</strong> ected wave should be taken <strong>in</strong>to account when a time<br />

97


τ<br />

8<br />

7<br />

6<br />

5<br />

4<br />

3<br />

2<br />

1<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

0 0.5 1 1.5 2 Ω<br />

Figu<strong>re</strong> 4.19: Tunnell<strong>in</strong>g times <strong>re</strong>fer<strong>re</strong>d to the propagation time of a f<strong>re</strong>e particle, f , for a thick barrier<br />

(D = 10). The curves denote the phase time p (solid l<strong>in</strong>e), dwell time d (dotted l<strong>in</strong>e), Buttiker-<br />

Landauer time B (dashed l<strong>in</strong>e), the absolute value of the semi-classical time j sj (dot-dashed l<strong>in</strong>e),<br />

and the e ective dwell time e (thick solid l<strong>in</strong>e).<br />

measu<strong>re</strong> for the traversal through the barrier is sought. The<strong>re</strong>fo<strong>re</strong> we try a di e<strong>re</strong>nt approach<br />

and use only the wave function <strong>in</strong>side the barrier for the computation of the ux.<br />

In pr<strong>in</strong>ciple, we follow the same path as <strong>in</strong> the calculation of the electromagnetic energy<br />

velocity <strong>in</strong> section 3.5 . The local velocity of the wave then, is the ratio between the ux<br />

J = j~ ,<br />

r , r : (4.55)<br />

2m<br />

and the probability density averaged over a period <strong>in</strong> time,<br />

v = J<br />

(4.56)<br />

Inside the barrier, we must use 2 from (4.36). After a lengthy calculation, we end up with<br />

the <strong>re</strong>ciprocal of the local velocity, a local transmission time<br />

1<br />

v =<br />

s 2m<br />

~(!p , !)<br />

2 + 2<br />

8<br />

e ,2 (x,d) + e 2 (x,d) +2<br />

2 , 2<br />

2 + 2<br />

; (4.57)<br />

with = 1 = p p 2m!=~ as the propagation constant outside the barrier and = ,j 2 =<br />

2m(!p , !)=~ as the one with<strong>in</strong>, <strong>re</strong>spectively. To obta<strong>in</strong> the e ective overall traversal time,<br />

98


4.5 Tunnell<strong>in</strong>g time de nitions for a squa<strong>re</strong> barrier<br />

τ<br />

8<br />

7<br />

6<br />

5<br />

4<br />

3<br />

2<br />

1<br />

0 0.5 1 1.5 2 Ω<br />

Figu<strong>re</strong> 4.20: Tunnell<strong>in</strong>g times <strong>re</strong>fer<strong>re</strong>d to the propagation time of a f<strong>re</strong>e particle, f , for a th<strong>in</strong> barrier<br />

(D = 3). The curves denote the phase time p (solid l<strong>in</strong>e), dwell time d (dotted l<strong>in</strong>e), Buttiker-<br />

Landauer time B (dashed l<strong>in</strong>e), the absolute value of the semi-classical time j sj (dot-dashed l<strong>in</strong>e),<br />

and the e ective dwell time e (thick solid l<strong>in</strong>e).<br />

we <strong>in</strong>tegrate (4.57) over the barrier,<br />

Integration is straightforward, and with our scaled variables we nd<br />

e<br />

f<br />

=<br />

1<br />

2(1 , )<br />

e =<br />

Z d<br />

0<br />

1<br />

dx : (4.58)<br />

v<br />

s<strong>in</strong>h 2D p 1 ,<br />

2D p 1 ,<br />

+1,2 : (4.59)<br />

Fig. 4.19 shows a comparison of the <strong>in</strong>dividual de nitions for a thick barrier. For waves<br />

above the top of the barrier, > 1, all approaches mo<strong>re</strong> or less give the same <strong>re</strong>sults. The<br />

dwell time, phase time, and the Buttiker-Landauer time a<strong>re</strong> nearly identical and exhibit an<br />

oscillatory behaviour, whe<strong>re</strong>as the semi-classical time is someth<strong>in</strong>g like the mean value of<br />

the oscillations and our e ective dwell time gives an upper bound. For evanescent waves, all<br />

de nitions except e <strong>re</strong>sult <strong>in</strong> traversal times that a<strong>re</strong> shorter than the propagation time of a<br />

f<strong>re</strong>e particle ( = f < 1) at least for low-f<strong>re</strong>quency waves. The e ective dwell time <strong>re</strong>aches its<br />

maximum at = 0, which seems <strong>re</strong>asonable from the particle po<strong>in</strong>t of view: if the energy is<br />

low, it takes longer to penetrate the barrier. Note also the pole of the semi-classical time at<br />

=1. For a th<strong>in</strong> barrier ( g. 4.20), the di e<strong>re</strong>nce between evanescent and transmitted waves<br />

is less marked, which is of course due to the fact that the transmission coe cient for th<strong>in</strong><br />

barriers is su ciently high also <strong>in</strong> the <strong>re</strong>gion below the barrier top. Likewise, the maximum<br />

of our e ective dwell time is much smaller.<br />

99


4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Remark (Absolute scal<strong>in</strong>g) In the literatu<strong>re</strong>, diagrams like <strong>in</strong> g. 4.19 and g. 4.20<br />

often appear to compa<strong>re</strong> various approaches. Unlike our gu<strong>re</strong>s, however, they give the<br />

tunnell<strong>in</strong>g time usually di<strong>re</strong>ctly <strong>in</strong> seconds or scaled to some xed value. Buttiker, for<br />

example, used f = d<br />

q<br />

2m<br />

<strong>in</strong>stead of (4.45) and furthermo<strong>re</strong> took 2 ~!p<br />

p !=!p as abscissa<br />

variable.<br />

The considerations lead<strong>in</strong>g to the de nition of (4.59) a<strong>re</strong> not enti<strong>re</strong>ly new. Moura and Albuquerque<br />

[104] described the same idea <strong>in</strong> a short note <strong>in</strong> 1990, and Leavens and Aers [91]<br />

obta<strong>in</strong>ed the <strong>re</strong>sult by <strong>in</strong>vestigat<strong>in</strong>g Bohm trajectories. They also po<strong>in</strong>ted out that this approach,<br />

too, is associated with a comb<strong>in</strong>ation of a transmitted and <strong>re</strong> ected wave | this time<br />

aris<strong>in</strong>g from the <strong>re</strong>ar <strong>in</strong>terface of the barrier. This was <strong>re</strong>ason enough for Landauer [86] to<br />

call <strong>in</strong>to question the physical usefulness of the Bohm trajectories.<br />

4.6 Examples of tunnell<strong>in</strong>g events<br />

The barrier traversal times discussed <strong>in</strong> the last section a<strong>re</strong> valid <strong>in</strong> a strict sense only for<br />

monochromatic waves, which isnotavery practical case. Hence we follow the same procedu<strong>re</strong><br />

we successfully applied <strong>in</strong> p<strong>re</strong>vious chapters and <strong>in</strong>vestigate a few examples of wave packets<br />

imp<strong>in</strong>g<strong>in</strong>g on <strong>re</strong>ctangular barriers of di e<strong>re</strong>nt height and width. The goal is to determ<strong>in</strong>e<br />

the behaviour of the transmitted part of the wave, if the<strong>re</strong> is any, and possibly establish a<br />

connection to the theo<strong>re</strong>tically found tunnell<strong>in</strong>g times. As we have seen <strong>in</strong> the scatter<strong>in</strong>g<br />

experiments with the step potential, discont<strong>in</strong>uities like edges or sharp peaks <strong>in</strong> the orig<strong>in</strong>al<br />

wave function a<strong>re</strong> smoothed out by dispersion after a short time. The<strong>re</strong>fo<strong>re</strong> we limit the<br />

analysis to Gaussian wave packets, which o er the additional advantage of be<strong>in</strong>g easier to t<strong>re</strong>at<br />

numerically. Even mo<strong>re</strong> important is the fact that if we want to study pu<strong>re</strong> tunnell<strong>in</strong>g without<br />

the disturb<strong>in</strong>g e ects of over-the-barrier components, we must ensu<strong>re</strong> a proper bandwidth<br />

limitation. Out of the th<strong>re</strong>e possible <strong>in</strong>itial wave forms we proposed <strong>in</strong> section 4.2, the<br />

Gaussian wave is the only one to comply su ciently with this <strong>re</strong>qui<strong>re</strong>ment.<br />

The departu<strong>re</strong> po<strong>in</strong>t for the calculations a<strong>re</strong> the wave functions <strong>in</strong> the th<strong>re</strong>e <strong>re</strong>gions (4.41)<br />

{ (4.43) together with the <strong>in</strong>itial spectrum (4.28). We evaluate the <strong>in</strong>tegrals numerically<br />

and plot the <strong>re</strong>sults. Like <strong>in</strong> the <strong>in</strong>vestigation of scatter<strong>in</strong>g, the probability density will be<br />

depicted <strong>in</strong> a th<strong>re</strong>e-dimensional pictu<strong>re</strong> as well as a cor<strong>re</strong>spond<strong>in</strong>g contour plot that allows a<br />

better perception of the paths the <strong>in</strong>dividual waves move along. In addition, we determ<strong>in</strong>e<br />

the trajectories of the peak of the wave packets <strong>in</strong> order to ga<strong>in</strong> <strong>in</strong>sight <strong>in</strong>to the controversial<br />

question whether they can be ascribed a physical mean<strong>in</strong>g to or not.<br />

At rst, we consider a high-energy electron <strong>in</strong>teract<strong>in</strong>g with a wide barrier. The parameters<br />

of the <strong>in</strong>itial wave packet a<strong>re</strong> the same as <strong>in</strong> the rst scatter<strong>in</strong>g example of section 4.3, and<br />

g. 4.21 shows the <strong>re</strong>sults. The edges of the barrier a<strong>re</strong> marked with l<strong>in</strong>es on the faces of the<br />

plotted a<strong>re</strong>a. In addition, the th<strong>re</strong>e-dimensional pictu<strong>re</strong> was rotated so that the <strong>re</strong>gion outside<br />

the barrier is on the right side of the plot. The behaviour <strong>in</strong> front of the barrier is similar<br />

to the scatter<strong>in</strong>g case, the<strong>re</strong> is the well-known <strong>in</strong>terfe<strong>re</strong>nce between <strong>in</strong>com<strong>in</strong>g and <strong>re</strong> ected<br />

waves. Beh<strong>in</strong>d the barrier, we notice a fa<strong>in</strong>t wave emerg<strong>in</strong>g from the tunnel. Mo<strong>re</strong> <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g,<br />

however, is the wave function <strong>in</strong>side the barrier itself. As the <strong>in</strong>cident wavepenetrates <strong>in</strong>to<br />

100


4.6 Examples of tunnell<strong>in</strong>g events<br />

P<br />

2<br />

40<br />

1<br />

0<br />

40<br />

40<br />

30<br />

20<br />

10<br />

30<br />

20<br />

T<br />

20<br />

10<br />

0<br />

X<br />

0<br />

-20<br />

-40<br />

0<br />

-40 -20 0 20 40<br />

Figu<strong>re</strong> 4.21: Evolution of the probability density P = j j of a Gaussian wave packet with the<br />

parameters k =6,n= 4, and = 0:8 imp<strong>in</strong>g<strong>in</strong>g on a barrier of thickness D = 10.<br />

101


4 One-dimensional quantum tunnell<strong>in</strong>g<br />

the tunnel and bounces back aga<strong>in</strong>, a fraction of it seems to get caught <strong>in</strong> the barrier, runn<strong>in</strong>g<br />

back and forth between the <strong>in</strong>terfaces while decay<strong>in</strong>g not too quickly. This peculiar e ect<br />

has noth<strong>in</strong>g to do with evanescent waves but stems from the ord<strong>in</strong>arily transmitted parts<br />

of the spectrum. It is just the common observation made for propagat<strong>in</strong>g waves that every<br />

change <strong>in</strong> the medium <strong>in</strong>evitably causes <strong>re</strong> ections. Look<strong>in</strong>g ca<strong>re</strong>fully, we even discover a<br />

small after-runner of the <strong>re</strong> ected wave (at T = 40 and X ,20 <strong>in</strong> the th<strong>re</strong>e-dimensional<br />

plot), which can be traced back to the <strong>in</strong>stant when the wave <strong>in</strong>side the barrier hits the left<br />

<strong>in</strong>terface (at X = 0) for the rst time.<br />

To <strong>in</strong>vestigate this behaviour <strong>in</strong> g<strong>re</strong>ater detail, we <strong>re</strong>gard an even thicker barrier and the same<br />

electron as befo<strong>re</strong> (as <strong>in</strong> the close-up of g. 4.22). In this case, the bounc<strong>in</strong>g of the wave <strong>in</strong>side<br />

the barrier is mo<strong>re</strong> marked, and each time it <strong>re</strong>aches an <strong>in</strong>terface, a small wave is <strong>re</strong>leased to<br />

the outside. This way, apart from the widen<strong>in</strong>g of the pulse due to dispersion, the <strong>in</strong>ner wave<br />

gradually fades away.<br />

Return<strong>in</strong>g to the orig<strong>in</strong>al example, we exam<strong>in</strong>e the trajectory of the peak of the wave packet<br />

( g. 4.23). To determ<strong>in</strong>e this path, we havetwo possibilities to identify the maximum (see also<br />

the discussion <strong>in</strong> section 1.5): a temporal and a spatial one. He<strong>re</strong> we consider the temporal<br />

maximum, i. e. the moment <strong>in</strong> time for a given position <strong>in</strong> space when the observed eld <strong>re</strong>aches<br />

its maximum amplitude. We <strong>re</strong>cognise the l<strong>in</strong>ear motion of the <strong>in</strong>cident and transmitted wave,<br />

as well as the not surpris<strong>in</strong>g fact that a de nition of the peak's trajectory is not mean<strong>in</strong>gful<br />

<strong>in</strong> the <strong>in</strong>terfe<strong>re</strong>nce zone <strong>in</strong> front of the barrier. Inte<strong>re</strong>st<strong>in</strong>gly enough, the trajectory of the<br />

transmitted wave is mo<strong>re</strong> strongly <strong>in</strong>cl<strong>in</strong>ed towards the axis than the <strong>in</strong>cident wave packet,<br />

mean<strong>in</strong>g that the packet to the right of the barrier runs faster. Aga<strong>in</strong>, this is largely due to<br />

the spectral components that can propagate over the barrier, whe<strong>re</strong>as the evanescent parts a<strong>re</strong><br />

enti<strong>re</strong>ly attenuated by the thick obstacle. Hence the low-f<strong>re</strong>quency components a<strong>re</strong> lte<strong>re</strong>d<br />

out, and the wave packet emanat<strong>in</strong>g from the far side of the barrier now has a spectrum that<br />

is narrower, but cent<strong>re</strong>d about a higher f<strong>re</strong>quency than befo<strong>re</strong>. He<strong>re</strong> and <strong>in</strong> the subsequent<br />

examples, the actual cent<strong>re</strong> f<strong>re</strong>quencies for both the <strong>in</strong>com<strong>in</strong>g and transmitted wave packets<br />

a<strong>re</strong> computed from the trajectories and given <strong>in</strong> the captions of the <strong>re</strong>spective gu<strong>re</strong>s. In the<br />

cur<strong>re</strong>nt example, this cent<strong>re</strong> f<strong>re</strong>quency is even above the cuto f<strong>re</strong>quency of the barrier. The<br />

propagation of the high-f<strong>re</strong>quency part is also <strong>re</strong>sponsible for the s-like shape of the trajectory<br />

with<strong>in</strong> the barrier and the o set between the extrapolated trajectories of the <strong>in</strong>cident and<br />

transmitted waves.<br />

It is not di cult to calculate the actual cent<strong>re</strong> f<strong>re</strong>quency from the trajectories. S<strong>in</strong>ce the<br />

wave packets a<strong>re</strong> smooth and well-behaved, we may safely take the view of the phase time<br />

supporters and assume that the wave proceeds at the group velocity, which wehavefound to<br />

be<br />

r<br />

2~!<br />

vg = :<br />

m<br />

(4.60)<br />

It is easy to t a straight l<strong>in</strong>e <strong>in</strong>to the space-time po<strong>in</strong>ts of the trajectory outside the barrier<br />

and away from the <strong>in</strong>terfe<strong>re</strong>nce <strong>re</strong>gion, and from the normalised variables, we obta<strong>in</strong> di<strong>re</strong>ctly<br />

102


4.6 Examples of tunnell<strong>in</strong>g events<br />

200<br />

150<br />

100<br />

50<br />

30<br />

20<br />

0<br />

50<br />

10<br />

X<br />

T<br />

100<br />

0<br />

150<br />

0<br />

-10 0 10 20 30<br />

200<br />

0<br />

-10<br />

0.1<br />

0.075<br />

P<br />

0.05<br />

0.025<br />

Figu<strong>re</strong> 4.22: Probability density P = j j ofawave packet with the parameters k =6,n= 4, and<br />

=0:8 <strong>in</strong>side a barrier of thickness D = 20.<br />

103


30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

T<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

-10 0 10 20 30<br />

Figu<strong>re</strong> 4.23: Trajectory of the temporal peak of a wave packet with the parameters k =6,n= 4, and<br />

=0:8<strong>in</strong>teract<strong>in</strong>g with a barrier of thickness D = 10. The e ective cent<strong>re</strong> f<strong>re</strong>quency of the <strong>in</strong>cident<br />

wave is ci 0:8181, for the transmitted wave itis ct 1:11.<br />

a <strong>re</strong>lation to the carrier f<strong>re</strong>quency !c=!p = c,<br />

T<br />

X<br />

c t 1<br />

= =<br />

vg t 2 p : (4.61)<br />

c<br />

Remark (Calculation of the cent<strong>re</strong> f<strong>re</strong>quency) For the <strong>in</strong>cident wave, the e ective<br />

cent<strong>re</strong> f<strong>re</strong>quency c should of course be equal to the carrier f<strong>re</strong>quency selected for the<br />

<strong>in</strong>itial wave. The value obta<strong>in</strong>ed from the trajectory, however, is slightly di e<strong>re</strong>nt, and<br />

experiments show that for a trajectory consist<strong>in</strong>g of spatial maxima, the cor<strong>re</strong>spondence<br />

would have been better. This is not surpris<strong>in</strong>g as the <strong>in</strong>itial wave packet was de ned <strong>in</strong><br />

space rather than <strong>in</strong> time. Though be<strong>in</strong>g fairly close together, the spatial and temporal<br />

maxima a<strong>re</strong> still slightly di e<strong>re</strong>nt.<br />

It should also be noted that the cent<strong>re</strong> f<strong>re</strong>quency calculated from the trajectory is not the<br />

peak f<strong>re</strong>quency of the spectrum, as one could suspect. Accord<strong>in</strong>g to (4.43), the spectrum<br />

of the <strong>in</strong>itial pulse is lte<strong>re</strong>d through the transmission coe cient, which <strong>in</strong>troduces a<br />

skewness to the orig<strong>in</strong>ally symmetric distribution. In the particular case of = 0:8 , the<br />

spectrum is even oscillatory, because of the <strong>re</strong>sonance peaks of the transmission coe cient<br />

above cuto (see also g. 4.18).<br />

Figs. 4.24 and 4.25 show the situation for the unalte<strong>re</strong>d <strong>in</strong>itial wave packet, but with a dec<strong>re</strong>as<strong>in</strong>g<br />

width of the barrier. The e ects the obstacle exerts on the wave <strong>re</strong>ma<strong>in</strong> <strong>in</strong> pr<strong>in</strong>ciple<br />

the same as befo<strong>re</strong>, with one g<strong>re</strong>at exception: the<strong>re</strong> is no <strong>re</strong>sidual wave <strong>in</strong>side the barrier after<br />

104<br />

X


4.6 Examples of tunnell<strong>in</strong>g events<br />

-40<br />

40<br />

30<br />

20<br />

10<br />

-20<br />

0<br />

X<br />

40<br />

30<br />

20<br />

T<br />

20<br />

0<br />

-40 -20 0 20 40<br />

10<br />

0<br />

1<br />

0<br />

40<br />

Figu<strong>re</strong> 4.24: Probability density P = j j ofawave packet with the parameters k =6,n= 4, and<br />

=0:8<strong>in</strong>teract<strong>in</strong>g with a medium-sized barrier (D = 5).<br />

105<br />

2<br />

P


P<br />

0<br />

3<br />

2<br />

1<br />

10<br />

0<br />

-40<br />

T<br />

40<br />

30<br />

20<br />

10<br />

20<br />

30<br />

-20<br />

40<br />

0<br />

X<br />

20<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

40<br />

0<br />

-40 -20 0 20 40<br />

Figu<strong>re</strong> 4.25: Probability density P = j j of a wave packet with the parameters k =6,n= 4, and<br />

=0:8<strong>in</strong>teract<strong>in</strong>g with a th<strong>in</strong> barrier (D = 1).<br />

106


4.6 Examples of tunnell<strong>in</strong>g events<br />

the <strong>in</strong>teraction took place. Seem<strong>in</strong>gly, the barrier is too th<strong>in</strong> for a fully developed wave packet<br />

to evolve. As we could have expected, the transmitted wave grows stronger as the barrier size<br />

is <strong>re</strong>duced, and for the th<strong>in</strong> obstacle even surpasses the <strong>re</strong> ected wave <strong>in</strong> magnitude. Not at<br />

all unexpected comes the form of the trajectories of the two wave packets ( gs. 4.26 and 4.27,<br />

<strong>re</strong>spectively). In the example of the medium-sized obstacle, the trajectory of the transmitted<br />

wave packet is still mo<strong>re</strong> <strong>in</strong>cl<strong>in</strong>ed than that of the imp<strong>in</strong>g<strong>in</strong>g wave, albeit not to such an extent<br />

as with the thick barrier above. For the th<strong>in</strong> barrier, on the other hand, this e ect is hardly<br />

noticeable at all, which con rms the <strong>in</strong>tuitive suspicion that a th<strong>in</strong> barrier does less harm to<br />

the spectrum of a pulse s<strong>in</strong>ce the evanescent components have a better chance of tunnell<strong>in</strong>g<br />

through.<br />

Apart from the di e<strong>re</strong>nt <strong>in</strong>cl<strong>in</strong>ations of the trajectories, the<strong>re</strong> is another <strong>in</strong>structive detail<br />

the pictu<strong>re</strong>s <strong>re</strong>veal. Suppose we <strong>re</strong>gard only the <strong>in</strong>terfaces of the barrier <strong>in</strong> search of a<br />

plausible way to de ne a tunnell<strong>in</strong>g time. Intuitively, we could measu<strong>re</strong> the delay between<br />

the appearance of the wave packet's peak at the front and <strong>re</strong>ar edge. Such an approach is,<br />

however, deceptive <strong>in</strong> that it dis<strong>re</strong>gards the way the peak at the front <strong>in</strong>terface is generated.<br />

In fact, it is not the peak of the <strong>in</strong>cident wave packet but it orig<strong>in</strong>ates rather <strong>in</strong>cidentally<br />

from the superposition of <strong>in</strong>com<strong>in</strong>g and <strong>re</strong> ected wave. Actually, it <strong>re</strong>aches the barrier later<br />

than the f<strong>re</strong>e wave packet would without the p<strong>re</strong>sence of the obstacle. For the spectator at<br />

the front edge of the barrier, it looks as though the approach of the packet is slowed down by<br />

<strong>in</strong>terfe<strong>re</strong>nce. The<strong>re</strong>fo<strong>re</strong> the trajectory, if at all, is mean<strong>in</strong>gful only <strong>in</strong> an asymptotic sense if<br />

extrapolated from the undisturbed <strong>re</strong>gion to the barrier.<br />

We now explo<strong>re</strong> a di e<strong>re</strong>nt <strong>in</strong>itial wave packet, namely one with a spectrum su ciently<br />

con ned to the evanescent <strong>re</strong>gion. S<strong>in</strong>ce we can expect genu<strong>in</strong>e tunnell<strong>in</strong>g this time, we must<br />

choose a rather th<strong>in</strong> barrier <strong>in</strong> order to obta<strong>in</strong> a <strong>re</strong>cognisable wave packet beh<strong>in</strong>d the tunnel.<br />

Indeed, g. 4.28 shows that a medium-sized obstacle al<strong>re</strong>ady <strong>re</strong> ects almost the complete<br />

<strong>in</strong>cident wave, and practically noth<strong>in</strong>g <strong>re</strong>aches the far side of the barrier. The situation is<br />

naturally better for a th<strong>in</strong> barrier, as g. 4.29 demonstrates. It is clear, though, that the lower<br />

the cent<strong>re</strong> f<strong>re</strong>quency of the wave is, the narrower the barrier must be <strong>in</strong> order to obta<strong>in</strong> a<br />

noticeable tunnel e ect.<br />

The <strong>in</strong>spection of the trajectories of the wave packets outside the barrier also yields noth<strong>in</strong>g<br />

unexpected. Like befo<strong>re</strong>, the path of the wave leav<strong>in</strong>g the far side of the barrier is mo<strong>re</strong><br />

<strong>in</strong>cl<strong>in</strong>ed towards the axis, which emphasises once mo<strong>re</strong> the high-pass behaviour of the barrier.<br />

Consequently, this e ect is mo<strong>re</strong> pronounced if the barrier is wide. So far, the <strong>re</strong>sults qualitatively<br />

ag<strong>re</strong>e with those obta<strong>in</strong>ed for the higher-f<strong>re</strong>quency wave packet t<strong>re</strong>ated befo<strong>re</strong>. The<strong>re</strong><br />

is, however, a signi cant di e<strong>re</strong>nce: from a comparatively thick barrier (like <strong>in</strong> g. 4.30), the<br />

peak of the tunnell<strong>in</strong>g wave packet emanates befo<strong>re</strong> the peak of a f<strong>re</strong>ely mov<strong>in</strong>g packet would<br />

have <strong>re</strong>ached the front of the barrier. Accord<strong>in</strong>gly, the trajectory even runs backwards <strong>in</strong> time<br />

<strong>in</strong>side the barrier. This phenomenon is still mo<strong>re</strong> imp<strong>re</strong>ssive for thicker barriers ( gs. 4.32 and<br />

4.33). At the rst super cial exam<strong>in</strong>ation, this could be taken for a violation of causality,<br />

because it <strong>in</strong>deed looks like turn<strong>in</strong>g the clock back while the electron is tunnell<strong>in</strong>g through<br />

the wall. Tak<strong>in</strong>g one step further, we could as well see this <strong>re</strong>sult as a proof for a wave packet<br />

travell<strong>in</strong>g faster than the speed of light | although the velocity of light actually never showed<br />

up throughout the enti<strong>re</strong> derivation of the tunnel e ect.<br />

107


25<br />

20<br />

15<br />

10<br />

5<br />

T<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

-10 0 10 20 30<br />

Figu<strong>re</strong> 4.26: Trajectory of the temporal peak of a wave packet with the parameters k =6,n= 4, and<br />

=0:8<strong>in</strong>teract<strong>in</strong>g with a medium-sized barrier (D = 5). The <strong>in</strong>cident wave is actually cent<strong>re</strong>d about<br />

ci 0:8181, the transmitted wave about ct 0:996.<br />

25<br />

20<br />

15<br />

10<br />

5<br />

T<br />

-10 0 10 20 30<br />

Figu<strong>re</strong> 4.27: Trajectory of the temporal peak of a wave packet with the parameters k =6,n= 4, and<br />

=0:8<strong>in</strong>teract<strong>in</strong>g with a th<strong>in</strong> barrier (D = 1). The e ective cent<strong>re</strong> f<strong>re</strong>quencies a<strong>re</strong> ci 0:8181 for<br />

the <strong>in</strong>cident and ct 0:8293 for the transmitted wave.<br />

108<br />

X<br />

X


4.6 Examples of tunnell<strong>in</strong>g events<br />

P<br />

2<br />

0<br />

1<br />

0<br />

140<br />

120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

50<br />

T<br />

-50<br />

100<br />

X<br />

0<br />

50<br />

0<br />

-80 -60 -40 -20 0 20 40 60<br />

Figu<strong>re</strong> 4.28: Probability density P = j j ofawave packet with the parameters k =6,n= 4, and<br />

=0:2<strong>in</strong>teract<strong>in</strong>g with a medium-sized barrier (D = 3).<br />

109


P<br />

2<br />

0<br />

1<br />

0<br />

140<br />

120<br />

100<br />

80<br />

60<br />

40<br />

20<br />

50<br />

T<br />

-50<br />

100<br />

X<br />

0<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

50<br />

0<br />

-80 -60 -40 -20 0 20 40 60<br />

Figu<strong>re</strong> 4.29: Probability density P = j j of a wave packet with the parameters k =6,n= 4, and<br />

=0:2<strong>in</strong>teract<strong>in</strong>g with a th<strong>in</strong> barrier (D = 1).<br />

110


4.6 Examples of tunnell<strong>in</strong>g events<br />

80<br />

60<br />

40<br />

20<br />

T<br />

-30 -20 -10 0 10 20 30<br />

Figu<strong>re</strong> 4.30: Trajectory of the temporal peak of a wave packet with the parameters k = 6, n = 4,<br />

and =0:2<strong>in</strong>teract<strong>in</strong>g with a medium-sized barrier (D = 3). The e ective cent<strong>re</strong> f<strong>re</strong>quencies a<strong>re</strong><br />

ci 0:2045 for the <strong>in</strong>cident and ct 0:217 for the transmitted wave packet.<br />

80<br />

60<br />

40<br />

20<br />

T<br />

-30 -20 -10 0 10 20 30<br />

Figu<strong>re</strong> 4.31: Trajectory of the temporal peak of a wave packet with the parameters k = 6, n = 4,<br />

and = 0:2 <strong>in</strong>teract<strong>in</strong>g with a th<strong>in</strong> barrier (D = 1). The <strong>in</strong>cident wave isactually cent<strong>re</strong>d about<br />

ci 0:2045, the transmitted wave about ct 0:2103.<br />

111<br />

X<br />

X


4 One-dimensional quantum tunnell<strong>in</strong>g<br />

Th<strong>in</strong>k<strong>in</strong>g the <strong>re</strong>sults over aga<strong>in</strong>, we <strong>re</strong>cognise that the afo<strong>re</strong>mentioned speculations have no<br />

foundation. The wave packet seems to be advanced, but as a matter of fact, it is not. Rather,<br />

we have one mo<strong>re</strong> example of pulse <strong>re</strong>shap<strong>in</strong>g, an e ect we encounte<strong>re</strong>d al<strong>re</strong>ady <strong>in</strong> section<br />

2.1. It is the high-f<strong>re</strong>quency components <strong>in</strong> the wave packet that arrive earlier at the barrier<br />

than the <strong>re</strong>st and <strong>in</strong> addition have a g<strong>re</strong>ater tunnell<strong>in</strong>g probability (or penetration depth). So<br />

it is these spectral components that constitute the packet beh<strong>in</strong>d the barrier, which wehave<br />

seen from the <strong>in</strong>c<strong>re</strong>ased group velocity. Hence the peak is not conserved <strong>in</strong> the tunnell<strong>in</strong>g<br />

process. The fact that we could draw a trajectory <strong>in</strong>side the barrier stems from the de nition<br />

of the maximum we decided to observe. Had we not taken the temporal maximum of the<br />

wave function, but the spatial counterpart, we would not have been able to nd a peak with<strong>in</strong><br />

the barrier at all, due to the exponential decay.<br />

The<strong>re</strong> is still another aspect of the trajectory <strong>in</strong>terp<strong>re</strong>tation: if the barrier is wide enough that<br />

the e ects mentioned above a<strong>re</strong> signi cant, the maximum of the probability density beyond<br />

the wall becomes vanish<strong>in</strong>gly small. So the trajectory covers the important fact that we a<strong>re</strong><br />

trad<strong>in</strong>g propagation velocity for tunnell<strong>in</strong>g probability or, <strong>in</strong> other words and from a practical<br />

po<strong>in</strong>t of view, signal speed for signal energy.<br />

Though be<strong>in</strong>g con ned to the evanescent <strong>re</strong>gion, the spectrum of the <strong>in</strong>itial pulse <strong>in</strong> the<br />

p<strong>re</strong>vious examples still had a considerable width orig<strong>in</strong>at<strong>in</strong>g from the fact that we <strong>in</strong>vestigated<br />

very `short' electrons with k = 6. This is of course no <strong>re</strong>ason to raise doubts about the<br />

cor<strong>re</strong>ctness and validity of the <strong>re</strong>sults, <strong>in</strong> particular the nd<strong>in</strong>g that pulse <strong>re</strong>shap<strong>in</strong>g would<br />

allow for negative tunnell<strong>in</strong>g times, if the peak of the wave packet is taken as a <strong>re</strong>fe<strong>re</strong>nce po<strong>in</strong>t<br />

for measur<strong>in</strong>g the propagation. Yet it is sensible to consider the behaviour of electrons with<br />

a narrower spectrum. We can expect that <strong>in</strong> such a case, the acceleration of the transmitted<br />

wave packet will be less prom<strong>in</strong>ent because the spectrum rapidly dec<strong>re</strong>ases to both sides of<br />

the orig<strong>in</strong>al cent<strong>re</strong>, leav<strong>in</strong>g little room for a f<strong>re</strong>quency shift. Consequently, the backward-shift<br />

<strong>in</strong> time of the trajectory <strong>in</strong>side the barrier will occur to a lesser extent, so that eventually<br />

the tunnell<strong>in</strong>g time will <strong>re</strong>ach a small but positive value. Indeed, g. 4.34 satis es these<br />

expectations. The extrapolated trajectory of the <strong>in</strong>cident wave packet is nearly parallel to<br />

that of the transmitted wave, and the small velocity di e<strong>re</strong>nce is noticeable only <strong>in</strong> the<br />

numerical determ<strong>in</strong>ation of the trajectory ascent.<br />

Let us now try to establish a connection to the various tunnell<strong>in</strong>g time de nition of the last<br />

section. Rema<strong>in</strong><strong>in</strong>g with the trajectory approach, we must nd unambiguous de nitions of the<br />

paths the wave packets move along. Beh<strong>in</strong>d the barrier, this is easy because the trajecory is<br />

l<strong>in</strong>ear. Immediately <strong>in</strong> front of the barrier, however, <strong>in</strong>terfe<strong>re</strong>nce impairs a clear perception of<br />

the maximum. Obviously we must x the trajectory to po<strong>in</strong>ts far away from the barrier whe<strong>re</strong><br />

the <strong>in</strong>cident wave is still undisturbed. We have seen al<strong>re</strong>ady that the velocity of the wave<br />

packet can safely be assumed to be the group velocity of its carrier f<strong>re</strong>quency, and we know<br />

of course the exact <strong>in</strong>itial position x0 of the peak at t =0. Wecan then calculate the time<br />

the wave packet takes to arrive at the barrier | it is simply the propagation time <strong>re</strong>qui<strong>re</strong>d<br />

to cover the distance x0 if no barrier is p<strong>re</strong>sent at all. The moment when the transmitted<br />

peak emanates from the <strong>re</strong>ar <strong>in</strong>terface can be computed from the trajectory like befo<strong>re</strong>, and<br />

subtract<strong>in</strong>g the two times we obta<strong>in</strong> what can be <strong>re</strong>garded as the tunnell<strong>in</strong>g time. This is the<br />

approach pursued also by Coll<strong>in</strong>s et al. [83].<br />

112


4.6 Examples of tunnell<strong>in</strong>g events<br />

80<br />

60<br />

40<br />

20<br />

T<br />

-30 -20 -10 0 10 20 30<br />

Figu<strong>re</strong> 4.32: Trajectory of the temporal peak of a wave packet with the parameters k =6,n= 4, and<br />

=0:2imp<strong>in</strong>g<strong>in</strong>g on a thick barrier (D = 10). The e ective cent<strong>re</strong> f<strong>re</strong>quencies a<strong>re</strong><br />

0:23385.<br />

ci 0:2045 and<br />

ct<br />

80<br />

60<br />

40<br />

20<br />

T<br />

-20 0 20 40<br />

Figu<strong>re</strong> 4.33: Trajectory of the temporal peak of a wave packet with the parameters k =6,n= 4, and<br />

=0:2 imp<strong>in</strong>g<strong>in</strong>g on a very wide barrier (D = 20). The e ective cent<strong>re</strong> f<strong>re</strong>quencies a<strong>re</strong> ci 0:2045<br />

and ct 0:2617.<br />

113<br />

X<br />

X


200<br />

150<br />

100<br />

50<br />

T<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

-125 -100 -75 -50 -25 0 25<br />

Figu<strong>re</strong> 4.34: Trajectory of the temporal peak of a wave packet with the parameters k = 20, n =4,<br />

and = 0:2 imp<strong>in</strong>g<strong>in</strong>g on a thick barrier (D = 10). The e ective cent<strong>re</strong> f<strong>re</strong>quencies a<strong>re</strong> ci 0:2004<br />

and ct 0:2028.<br />

From the wave form de nitions <strong>in</strong> section 4.2, we nd the <strong>in</strong>itial position of the peak,<br />

X0 = k<br />

p : (4.62)<br />

The time the wave packet needs to travel this distance is <strong>re</strong>adily obta<strong>in</strong>ed from the propagation<br />

velocity of the f<strong>re</strong>e electron (4.45)<br />

T<strong>in</strong>c = X0 k<br />

p =<br />

2 2<br />

: (4.63)<br />

S<strong>in</strong>ce we want to explo<strong>re</strong> the dependence of the tunnell<strong>in</strong>g time on the carrier f<strong>re</strong>quency, we<br />

must ensu<strong>re</strong> that the spectrum has equal width for all values of . The dom<strong>in</strong>at<strong>in</strong>g factor <strong>in</strong><br />

the wave number spectrum (4.28) is e ,( k=n ( =p ,1)) 2<br />

, whichs yields the <strong>re</strong>lation<br />

k<br />

p = ; (4.64)<br />

n<br />

with some arbitrary, non-negative constant . The width of the <strong>in</strong>itial wave packet, n, is held<br />

constant, is our <strong>in</strong>dependent variable, so we must set k = n p = <strong>in</strong> the spectrum and<br />

get the nal exp<strong>re</strong>ssion for the time of f<strong>re</strong>e motion,<br />

T<strong>in</strong>c = n<br />

p : (4.65)<br />

2<br />

114<br />

X


4.6 Examples of tunnell<strong>in</strong>g events<br />

7<br />

6<br />

5<br />

4<br />

3<br />

2<br />

1<br />

τ<br />

0.5 1 1.5 2 2.5 3 Ω<br />

8<br />

6<br />

4<br />

τ<br />

0.5 1 1.5 2 2.5 3 Ω<br />

Figu<strong>re</strong> 4.35: Numerically determ<strong>in</strong>ed tunnell<strong>in</strong>g times compa<strong>re</strong>d with the phase time theory for th<strong>in</strong><br />

(D = 1, left graph) and medium-sized (D = 5, right graph) barriers and a small-band Gaussian<br />

electron ( = 100, n = 4).<br />

20<br />

15<br />

10<br />

5<br />

-5<br />

-10<br />

τ<br />

0.5 1 1.5 2 2.5 3 Ω<br />

20<br />

15<br />

10<br />

5<br />

τ<br />

0.5 1 1.5 2 2.5 3 Ω<br />

Figu<strong>re</strong> 4.36: Numerically determ<strong>in</strong>ed tunnell<strong>in</strong>g times compa<strong>re</strong>d with the phase time theory for a thick<br />

barrier (D = 10), but electrons of di e<strong>re</strong>nt bandwidths (left graph: =10:54 b= k = 6 at = 0:2,<br />

right graph: = 35:12 b= k = 20 at = 0:2, n = 4 <strong>in</strong> both cases). The left plot also shows the<br />

semi-classical time (dashed l<strong>in</strong>e).<br />

Fig. 4.35 shows the <strong>re</strong>sults for a small-band electron and two di e<strong>re</strong>nt barrier widths. Out of all<br />

theo<strong>re</strong>tical tunnell<strong>in</strong>g time approaches <strong>in</strong> section 4.5, the phase time suits best the numerical<br />

<strong>re</strong>sults. This is not <strong>re</strong>ally surpris<strong>in</strong>g, because the way we set up the numerical computation of<br />

the tunnell<strong>in</strong>g time mimics exactly the basic idea of the phase time. We follow the motion of<br />

the wave packet's maximum, and furthermo<strong>re</strong> the spectrum is sharply peaked about a s<strong>in</strong>gle<br />

wave number. Note that for low <strong>in</strong>cident energies, the phase time p<strong>re</strong>diction and the numerical<br />

<strong>re</strong>sults disag<strong>re</strong>e. Coll<strong>in</strong>s et al. attributed this behaviour to the transmission coe cient they<br />

thought to be exponentially vary<strong>in</strong>g at low energies | an argument we cannot follow after<br />

<strong>in</strong>spection of g. 4.18, which shows the transmission coe cient to be particularly smooth for<br />

small = p !=!p. Rather, we can identify the <strong>in</strong>evitable pulse <strong>re</strong>shap<strong>in</strong>g phenomenon as<br />

the <strong>re</strong>ason of this disc<strong>re</strong>pancy. Ow<strong>in</strong>g to our assumptions, the spectral width of the wave<br />

packet is held constant ir<strong>re</strong>spective of its cent<strong>re</strong> f<strong>re</strong>quency. Hence the extent to which the<br />

f<strong>re</strong>quency can be shifted dur<strong>in</strong>g the tunnell<strong>in</strong>g event is also xed or at least upper-bounded.<br />

115


17.5<br />

15<br />

12.5<br />

10<br />

7.5<br />

5<br />

2.5<br />

τ<br />

0.5 1 1.5 2 2.5 3 Ω<br />

35<br />

30<br />

25<br />

20<br />

15<br />

10<br />

5<br />

τ<br />

4 One-dimensional quantum tunnell<strong>in</strong>g<br />

0.5 1 1.5 2 2.5 3 Ω<br />

Figu<strong>re</strong> 4.37: Numerically determ<strong>in</strong>ed tunnell<strong>in</strong>g times compa<strong>re</strong>d with the phase time theory for a thick<br />

barrier (left graph: D = 10, right graph: D = 20) and small-band electrons ( = 100, n = 4).<br />

The f<strong>re</strong>quency shift may be small, but it always occurs. The <strong>re</strong>lative variation of the group<br />

velocity, on the other hand, is much larger for very small energies because the group velocity<br />

approaches zero <strong>in</strong> the DC-limit, and so the acceleration e ect the wave packet experiences<br />

is larger, too.<br />

Remark (Trajectories <strong>in</strong>side the barrier) The above explanation becomes even<br />

mo<strong>re</strong> elucidat<strong>in</strong>g if we consider how the trajectory <strong>in</strong>side the barrier emerges. We al<strong>re</strong>ady<br />

po<strong>in</strong>ted out (see section 4.3) that it is the <strong>re</strong>sult of the successive arrival of the wave<br />

packet's components, all travell<strong>in</strong>g at dist<strong>in</strong>ct velocities. The rst to arrive a<strong>re</strong> the highenergy<br />

parts that also have a higher tunnell<strong>in</strong>g probability. Hence comes the <strong>in</strong>cl<strong>in</strong>ation<br />

of the trajectory towards the axis. The steady-state tunnell<strong>in</strong>g probability cor<strong>re</strong>sponds to<br />

the stationary penetration depth, which naturally is very small for low energies, so that<br />

its <strong>re</strong>lative <strong>in</strong>c<strong>re</strong>ase due to a xed f<strong>re</strong>quency shift is much larger for low than for high<br />

energies.<br />

The considerations given above also anticipate what will happen if we choose a broader<br />

spectrum for the <strong>in</strong>cident wavepacket. Fig. 4.36 shows the <strong>re</strong>sults for a thick barrier, whe<strong>re</strong><br />

the e ect is even mo<strong>re</strong> pronounced. The left graph was drawn with the same parameters we<br />

chose for calculat<strong>in</strong>g the trajectories of the low-energy electrons earlier <strong>in</strong> this section. We<br />

<strong>re</strong>cognise that for . 0:5, the numerically determ<strong>in</strong>ed tunnell<strong>in</strong>g time becomes negative,<br />

which is <strong>in</strong> l<strong>in</strong>e with the observations we made when comput<strong>in</strong>g the trajectories. Likewise,<br />

the<strong>re</strong> is no <strong>re</strong>semblance at all between the numerical <strong>re</strong>sults and the phase time theory |<br />

on the contrary, they a<strong>re</strong> rather close to the simple semi-classical time <strong>in</strong> that they do not<br />

exhibit any <strong>re</strong>sonance peaks above cuto and have a maximum exactly at = 1, <strong>in</strong> contrast<br />

to all other de nitions.<br />

The situation changes if the bandwidth of the imp<strong>in</strong>g<strong>in</strong>g electron is <strong>re</strong>duced like <strong>in</strong> the right<br />

pictu<strong>re</strong> <strong>in</strong> g. 4.36, whe<strong>re</strong> the parameters of the trajectory <strong>in</strong> g. 4.34 we<strong>re</strong> used. F<strong>in</strong>ally, <strong>in</strong><br />

g. 4.37 we selected aga<strong>in</strong> an <strong>in</strong>cident wavepacket with a `properly' narrow bandwidth and<br />

obta<strong>in</strong> the expected excellent ag<strong>re</strong>ement between numerics and the phase time p<strong>re</strong>diction.<br />

116


4.6 Examples of tunnell<strong>in</strong>g events<br />

The two graphs also <strong>re</strong>veal that the deviation between the two curves deteriorates as the<br />

barrier thickness is <strong>in</strong>c<strong>re</strong>ased.<br />

The conclusions to be drawn from the above a<strong>re</strong>:<br />

The barrier acts like a high-pass lter favour<strong>in</strong>g the high-energy components of the<br />

<strong>in</strong>cident wave packet. Thus the transmitted portion of the wave experiences a f<strong>re</strong>quency<br />

skew<strong>in</strong>g and a shift of its e ective cent<strong>re</strong> f<strong>re</strong>quency to higher values. This pulse <strong>re</strong>shap<strong>in</strong>g<br />

e ect occurs always, no matter how narrow-band the wave packet orig<strong>in</strong>ally was, and<br />

it is mo<strong>re</strong> marked for low-energy waves.<br />

The phase time theory is <strong>in</strong> good ag<strong>re</strong>ement with numerically found tunnell<strong>in</strong>g times,<br />

provided that the wave packet exhibits a dom<strong>in</strong>at<strong>in</strong>g peak that can be traced and<br />

taken as a measu<strong>re</strong> for the propagation. Furthermo<strong>re</strong>, the wave packet must also be<br />

very narrow-band so that the group velocity approximation for the propagation velocity<br />

holds. The spectrum need not be con ned to the evanescent <strong>re</strong>gion, and it may also<br />

extend <strong>in</strong>to the pass band.<br />

The phase time approximation becomes worse for very low energies of the <strong>in</strong>cident<br />

electron due to the <strong>re</strong>latively seve<strong>re</strong>r e ect of pulse <strong>re</strong>shap<strong>in</strong>g <strong>in</strong> this <strong>re</strong>gion.<br />

If the <strong>in</strong>itial wave packet is not small-band, the phase time gives no appropriate description<br />

of the tunnell<strong>in</strong>g time except for very large energies <strong>in</strong> the pass band. If the<br />

tunnell<strong>in</strong>g time is still based on the evolution of the peak of the wave packet, even<br />

negative delay times may be obta<strong>in</strong>ed. It is questionable whether the de nition of the<br />

tunnell<strong>in</strong>g time is still sensible then.<br />

As for the faster-than-light debate, we st<strong>re</strong>ss that the fo<strong>re</strong>go<strong>in</strong>g exam<strong>in</strong>ation was based<br />

on the non-<strong>re</strong>lativistic Schrod<strong>in</strong>ger equation. In the enti<strong>re</strong> chapter, the velocity of light<br />

appea<strong>re</strong>d not a s<strong>in</strong>gle time. So <strong>in</strong> order to judge how fast the particles or waves actually<br />

travel, we would need to take the theory of <strong>re</strong>lativity <strong>in</strong>to account. Note also that <strong>in</strong><br />

contrast to the electromagnetic case, the quantum mechanical wave function has no nite<br />

wave front velocity. The <strong>re</strong>ason for this di e<strong>re</strong>nce is the quadratic dispersion <strong>re</strong>lation<br />

of quantum waves, whe<strong>re</strong>as those of wave guides and lossless plasmas a<strong>re</strong> hyperbolic.<br />

This exposes the limitation of the analogy between undersized wave guides and squa<strong>re</strong><br />

barriers: it is valid as long as the <strong>re</strong>spective di e<strong>re</strong>ntial equations a<strong>re</strong> <strong>re</strong>garded. As one<br />

looks <strong>in</strong>to their solutions, however, it b<strong>re</strong>aks down.<br />

117


118<br />

4 One-dimensional quantum tunnell<strong>in</strong>g


Interlude<br />

<strong>Wave</strong> functions <strong>in</strong> graphical<br />

<strong>re</strong>p<strong>re</strong>sentation<br />

<strong>Wave</strong> functions a<strong>re</strong> complicated th<strong>in</strong>gs. They a<strong>re</strong> most easily formulated as Fourier <strong>in</strong>tegrals<br />

| provided, of course, we can nd the Fourier transform of some <strong>in</strong>itial state of the wave<br />

and have a closed exp<strong>re</strong>ssion for the dispersion <strong>re</strong>lation. In many cases, these a<strong>re</strong> only<br />

m<strong>in</strong>or hurdles, and the formulation of the wave <strong>in</strong>tegral is <strong>re</strong>adily found. But this is al<strong>re</strong>ady<br />

whe<strong>re</strong> simplicity ends and troubles start, for the evaluation of the <strong>in</strong>tegral is under most<br />

circumstances anyth<strong>in</strong>g but straightforward. The<strong>re</strong> a<strong>re</strong> ra<strong>re</strong> cases whe<strong>re</strong> the <strong>in</strong>itial wave or<br />

the dispersion <strong>re</strong>lation a<strong>re</strong> so well-behaved that the <strong>in</strong>verse Fourier transform can be calculated<br />

right away or after application of some sophisticated mathematical tricks. Sometimes, special<br />

values for x or t also make <strong>in</strong>tegration feasible. However, normally the wave functions balk<br />

at analytical evaluation.<br />

On the other hand, it is clearly desirable and valuable to be able to calculate the wave<br />

<strong>in</strong>tegrals <strong>in</strong> order to depict at least snapshots of the wave and to underl<strong>in</strong>e analytical nd<strong>in</strong>gs<br />

<strong>in</strong> a graphical way. The early <strong>in</strong>vestigators of wave propagation we<strong>re</strong> hard-p<strong>re</strong>ssed to tackle<br />

this problem and had no choice other than to <strong>re</strong>sort to approximation techniques. For the<br />

exam<strong>in</strong>ation of the wave front and the rst p<strong>re</strong>cursor <strong>in</strong> a Lo<strong>re</strong>ntz medium, Sommerfeld [12]<br />

used a high-f<strong>re</strong>quency expansion of the dispersion <strong>re</strong>lation, which then led to <strong>in</strong>tegratable<br />

exp<strong>re</strong>ssions. This approximation, however, was valid only about the wave front whe<strong>re</strong> highf<strong>re</strong>quency<br />

components a<strong>re</strong> dom<strong>in</strong>at<strong>in</strong>g. Brillou<strong>in</strong> [19] and Baerwald [21, 22], who looked <strong>in</strong>to<br />

the evaluation of the signal beh<strong>in</strong>d the wave front, employed the method of saddle po<strong>in</strong>t<br />

<strong>in</strong>tegration whe<strong>re</strong> they made use of two approximations: the rst one be<strong>in</strong>g the fact that<br />

the <strong>re</strong>levant contributions to the wave <strong>in</strong>tegral come from the vic<strong>in</strong>ities of the saddle po<strong>in</strong>ts,<br />

and the second one be<strong>in</strong>g the approximation of the <strong>in</strong>tegrand about these saddle po<strong>in</strong>ts. By<br />

us<strong>in</strong>g these techniques, the authors could <strong>re</strong>duce the <strong>in</strong>tegrals to comb<strong>in</strong>ations of well-known<br />

functions and eventually obta<strong>in</strong> graphical <strong>re</strong>p<strong>re</strong>sentations.<br />

With the advent of electronic comput<strong>in</strong>g mach<strong>in</strong>es, the problems with the evaluation of wave<br />

<strong>in</strong>tegrals we<strong>re</strong> alleviated. First, computers allow us to make mo<strong>re</strong> accurate plots of given<br />

functions s<strong>in</strong>ce they can calculate function values at a much larger number of po<strong>in</strong>ts much<br />

faster than it was possible with the tedious and error-prone look-up methods us<strong>in</strong>g mathematical<br />

tables. Second and mo<strong>re</strong> importantly, numerical methods a<strong>re</strong> available to evaluate<br />

<strong>in</strong>tegrals that would be impossible be by hand calculations. Still, numerics a<strong>re</strong> no general<br />

<strong>re</strong>medy for all problems raised by the <strong>in</strong>vestigation of wave functions. Actually, the di culty<br />

of nd<strong>in</strong>g a closed form for the <strong>in</strong>tegral at all is only transformed <strong>in</strong>to a question of comput<strong>in</strong>g<br />

119


Interlude<br />

power and numerical accuracy, both of which can <strong>in</strong> turn be hard to cope with. Consequently,<br />

particularly <strong>in</strong> the early days of computers, approximation and expansion methods we<strong>re</strong> still<br />

widely employed to obta<strong>in</strong> <strong>re</strong>asonable <strong>re</strong>sults with<strong>in</strong> <strong>re</strong>asonable time.<br />

Scann<strong>in</strong>g the literatu<strong>re</strong>, we nd two pr<strong>in</strong>cipal approaches to t<strong>re</strong>at the wave propagation problem<br />

by means of numerical comput<strong>in</strong>g. The rst is to formulate the wave <strong>in</strong>tegral and then<br />

apply numerics to evaluate the exp<strong>re</strong>ssion. As mentioned befo<strong>re</strong>, the <strong>in</strong>tegrand is often simplied<br />

or transformed via series expansion to facilitate <strong>in</strong>tegration either <strong>in</strong> terms of convergence<br />

or accuracy. This way the normally <strong>in</strong> nite <strong>in</strong>tegrals may f<strong>re</strong>quently be truncated at some<br />

nite value without <strong>in</strong>troduc<strong>in</strong>g too large an error. The second approach is to dispense with<br />

the wave <strong>in</strong>tegral itself and start from the di e<strong>re</strong>ntial equation together with the appropriate<br />

<strong>in</strong>itial and boundary conditions describ<strong>in</strong>g the propagation phenomenon. This usually<br />

<strong>in</strong>volves some sort of disc<strong>re</strong>tisation of the di e<strong>re</strong>ntial equation on a properly chosen grid <strong>in</strong><br />

order to nd a numerical solution. Such a strategy was for example pursued by Weiland [105],<br />

who used it calculate electromagnetic elds <strong>in</strong> complex geometric structu<strong>re</strong>s. He divided the<br />

a<strong>re</strong>a of <strong>in</strong>te<strong>re</strong>st <strong>in</strong>to small elementary cells of essentially arbitrary shape. Out of the many<br />

possible disc<strong>re</strong>tisations, he selected one that uses the <strong>in</strong>tegral form of the Maxwell equations<br />

such that at the boundaries between the cells, the tangential component oftheelectric eld<br />

and the perpendicular component of the magnetic eld a<strong>re</strong> cont<strong>in</strong>uous (which is not necessarily<br />

the case if other disc<strong>re</strong>tisation schemes a<strong>re</strong> used). Solv<strong>in</strong>g Maxwell's equations is then<br />

<strong>re</strong>duced to the easier task of solv<strong>in</strong>g matrix equations.<br />

The alternative approach of evaluat<strong>in</strong>g the wave <strong>in</strong>tegrals numerically has been used by a<br />

larger number of authors | at least as far as electromagnetic wave propagation is concerned.<br />

At the beg<strong>in</strong>n<strong>in</strong>g of the computer era, Haskell and Case [97] did not <strong>re</strong>ally evaluate the <strong>in</strong>tegral<br />

di<strong>re</strong>ctly, but used the familiar method of saddle-po<strong>in</strong>t <strong>in</strong>tegration to obta<strong>in</strong> an approximation<br />

for large values of x. This strategy was stimulated by the observation that most solutions at<br />

that time we<strong>re</strong> suitable only for small propagation distances, and they ended up with a solution<br />

consist<strong>in</strong>g of F<strong>re</strong>snel <strong>in</strong>tegrals that could be computed numerically. As for the question which<br />

part of the <strong>in</strong>tegrand is to be expanded, a variety of aswers we<strong>re</strong> attempted. Knop [96], for<br />

example, expanded the s<strong>in</strong>e function de n<strong>in</strong>g the <strong>in</strong>put signal of the transmission l<strong>in</strong>e <strong>in</strong>to a<br />

series of Bessel functions and then applied the <strong>in</strong>verse Laplace transform. Trizna and Weber<br />

[23], on the contrary, expanded the enti<strong>re</strong> <strong>in</strong>tegrand also <strong>in</strong>to a series of Bessel functions<br />

that could be <strong>in</strong>tegrated numerically. Mo<strong>re</strong> <strong>re</strong>cently, Wyns et al. [25] used <strong>in</strong>verse Laplace<br />

transform, but prior to that, approximated the factor e st by a function which then could<br />

be t<strong>re</strong>ated by series expansion. As a <strong>re</strong>sult, they could <strong>in</strong>tegrate the simpli ed functions<br />

analytically and employ a truncated summation to achieve the desi<strong>re</strong>d accuracy. Albanese et<br />

al. [106] cleverly sidestepped the enti<strong>re</strong> problem of Fourier transform by <strong>re</strong>gard<strong>in</strong>g a <strong>re</strong>petitive<br />

<strong>in</strong>put pulse so that the Fourier <strong>in</strong>tegral degenerated to the much simpler Fourier sum. Bolda<br />

et al. [74], as the last and most <strong>re</strong>cent example, used the modern method of Fast Fourier<br />

Transform (FFT) to compute the propagation of a Gaussian pulse <strong>in</strong> an <strong>in</strong>verted medium.<br />

The term `wave propagation' is deeply l<strong>in</strong>ked to a strong and colourful imp<strong>re</strong>ssion of a lm-like<br />

sequence of <strong>in</strong>dividual pictu<strong>re</strong>s show<strong>in</strong>g the motion of a wave packet. It should thus not come<br />

as a surprise if some authors took up this idea and attempted to generate such lms <strong>in</strong> order<br />

to br<strong>in</strong>g the term `motion' to optical life. Indeed, the<strong>re</strong> have been such approaches <strong>in</strong> the<br />

120


<strong>Wave</strong> functions <strong>in</strong> graphical <strong>re</strong>p<strong>re</strong>sentation<br />

past. It is <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g to notice that they we<strong>re</strong> chie y <strong>re</strong>stricted to quantum mechanical wave<br />

packets. Seem<strong>in</strong>gly, the imag<strong>in</strong>ation of quantum particles mov<strong>in</strong>g like waves was much mo<strong>re</strong><br />

demand<strong>in</strong>g and called for a visualisation rather than the analogous problem of electromagnetic<br />

wave propagation. The<strong>re</strong> a<strong>re</strong> quite a number of examples of computer animation of wave<br />

packets, the rst and nowadays almost legendary one be<strong>in</strong>g that of Goldberg et al. [107],<br />

who analysed the scatter<strong>in</strong>g of a Gaussian wave packet from potential barriers and wells<br />

as early as 1967. To obta<strong>in</strong> the wave functions they did not formulate the wave <strong>in</strong>tegrals,<br />

but <strong>in</strong>stead solved the Schrod<strong>in</strong>ger equation numerically. The method they used is the wellknown<br />

Crank-Nicholson algorithm that essentially is based on an equidistant disc<strong>re</strong>tisation<br />

of the di e<strong>re</strong>ntial equation. This <strong>in</strong> turn leads to a set of <strong>re</strong>cursive equations, and to get<br />

<strong>in</strong>itial values for the <strong>re</strong>cursion, the enti<strong>re</strong> system is placed <strong>in</strong> a box with <strong>in</strong> nitely high walls.<br />

Clearly the walls must be su ciently far away from the <strong>re</strong>gion of <strong>in</strong>te<strong>re</strong>st to avoid <strong>re</strong> ections<br />

dur<strong>in</strong>g the conside<strong>re</strong>d time span. In addition, the <strong>re</strong>solution must be chosen ne enough to<br />

follow the oscillations of the wave function.<br />

The Crank-Nicholson scheme has become the method of choice for many other authors, too.<br />

Coll<strong>in</strong>s et al. [108, 83] did not shoot movies with it, but used it to simulate the evolution of<br />

a Gaussian wave packet <strong>in</strong> order to compa<strong>re</strong> their <strong>re</strong>view of tunnell<strong>in</strong>g time de nitions with<br />

numerical <strong>re</strong>sults. They <strong>re</strong>ported that the disc<strong>re</strong>tisation of the discont<strong>in</strong>uous squa<strong>re</strong> barrier<br />

potential pro le was crucial for the success of the method and had a signi cant <strong>in</strong> uence on the<br />

<strong>re</strong>sults. In the <strong>re</strong>cent past, computers have become powerful enough that this method can even<br />

be employed to generate on-l<strong>in</strong>e animations of wave motion, like <strong>in</strong> the Mathematica-based<br />

example of Robb [109], who comb<strong>in</strong>ed the convenience of a modern mathematical softwa<strong>re</strong><br />

environment with the computational power of a `workhorse' written <strong>in</strong> C. A softwa<strong>re</strong> package<br />

solely dedicated to quantum physics is the program of Hiller et al. [110], which besides many<br />

other functions to explo<strong>re</strong> quantum mechanics, also conta<strong>in</strong>s a section whe<strong>re</strong> the motion of a<br />

wave packet through a given potential structu<strong>re</strong> can be observed.<br />

Somehow di e<strong>re</strong>nt to the afo<strong>re</strong>mentioned examples is that by Merrill [111], who <strong>in</strong> the early<br />

days of computers p<strong>re</strong>sented a simple BASIC program to study the propagation of waves <strong>in</strong> a<br />

dispersive medium. Intended | like all other animation approaches | primarily for students,<br />

it <strong>re</strong>qui<strong>re</strong>d the <strong>in</strong>itial wave form and the dispersion <strong>re</strong>lation to be supplied by the user. By<br />

use of the trapezoidal rule, it then computed the spectrum of the <strong>in</strong>itial wave form as well as<br />

the <strong>in</strong>verse Fourier transform of the wave at some later time.<br />

Try<strong>in</strong>g to p<strong>re</strong>sent mov<strong>in</strong>g pictu<strong>re</strong>s <strong>in</strong> a written work is a practically futile attempt due to<br />

the obvious lack of motion. The<strong>re</strong> is only one conceivable situation whe<strong>re</strong> someth<strong>in</strong>g like a<br />

contiuously ow<strong>in</strong>g time enters an otherwise static book | when a <strong>re</strong>ader quickly thumbs<br />

through the pages without tak<strong>in</strong>g the time to <strong>re</strong>ad through them. While clearly be<strong>in</strong>g undesirable<br />

from the author's po<strong>in</strong>t of view, this opens a subtle possibility to still convey some<br />

<strong>in</strong>formation, even <strong>in</strong>formation a thorough <strong>re</strong>ader would hardly ever <strong>re</strong>cognise. Hence the<strong>re</strong><br />

a<strong>re</strong> two `thumb movies' <strong>in</strong>cluded <strong>in</strong> the outer top corners of the pages. The sequence on the<br />

odd pages shows a Gaussian wave packet tunnell<strong>in</strong>g through a squa<strong>re</strong> barrier like <strong>in</strong> section<br />

4.6. The parameters | for the sake of completeness | a<strong>re</strong> n =4,k=6, =0:8, and D =4.<br />

The displayed spatial range is [,40; 40], and time runs from T = 0 to approximately T = 42.<br />

The barrier has been portrayed as a <strong>re</strong>ctangle, however, its height has no physical signi cance<br />

121


Interlude<br />

whatsoever. Consequently the <strong>in</strong>terfe<strong>re</strong>nce peaks <strong>in</strong> front of the <strong>in</strong>terface can grow higher<br />

than its top without swash<strong>in</strong>g over it.<br />

The even pages depict the evolution of an <strong>in</strong>itially triangular wave packet imp<strong>in</strong>g<strong>in</strong>g on a<br />

step barrier like <strong>in</strong> section 4.3. The parameters of this wave packet a<strong>re</strong> k =3and =0:8.<br />

The spatial <strong>re</strong>gion of <strong>in</strong>te<strong>re</strong>st is [,40; 40], and like befo<strong>re</strong>, the time <strong>in</strong>terval is [0; 42]. He<strong>re</strong>,<br />

too, the height of the barrier is shown only for graphical <strong>re</strong>asons and has no actual mean<strong>in</strong>g.<br />

In order to provide a spectacular <strong>re</strong>sult, the wave number spectrum was chosen so as to<br />

comprise considerable pass-band components as well, which is why we can see a small and<br />

rapidly widen<strong>in</strong>g wave runn<strong>in</strong>g further <strong>in</strong>to the barrier.<br />

122


Part II<br />

Numerical aspects of wave<br />

equations<br />

In the rst part, we focused on the physical mean<strong>in</strong>gs of energy velocity and wave propagation.<br />

To this end, we took the numerical solutions of the wave equations for granted and left aside<br />

the particular problems associated with these computations. It is now time to make up for this<br />

de ciency and demonstrate how we obta<strong>in</strong>ed the numerical answers to our physical questions.<br />

The second part of this work is the<strong>re</strong>fo<strong>re</strong> dedicated to the computation of <strong>in</strong> nite <strong>in</strong>tegrals<br />

by means of numerical algorithms.<br />

In the follow<strong>in</strong>g chapter we shall rst give anoverview of the basics of this branch ofnumerical<br />

mathematics and various approaches that have been published over the years. Unfortunately,<br />

we shall discover that none of the commonly available softwa<strong>re</strong> algorithms is suited to help<br />

us solve our problems. Consequently, the next chapters deal with general thoughts on the<br />

particular di culties and nally the implementation of a suitable quadratu<strong>re</strong> strategy to<br />

compute the type of wave <strong>in</strong>tegrals we a<strong>re</strong> <strong>in</strong>te<strong>re</strong>sted <strong>in</strong>. Although the implementation of the<br />

quadratu<strong>re</strong> rout<strong>in</strong>e itself is application-oriented, it is still <strong>in</strong>dependent of the actual examples<br />

we exam<strong>in</strong>ed <strong>in</strong> the rst part. The<strong>re</strong>fo<strong>re</strong>, we shall also p<strong>re</strong>sent how the general functions a<strong>re</strong><br />

applied to the problem of the tunnell<strong>in</strong>g particle.<br />

The programm<strong>in</strong>g language used for the actual implementation of the functions was Mathematica<br />

| despite FORTRAN's still dom<strong>in</strong>at<strong>in</strong>g role <strong>in</strong> the world of numerical mathematics.<br />

The<strong>re</strong> we<strong>re</strong> several <strong>re</strong>asons for this choice: rstly, Mathematica o ers a high-level functional<br />

programm<strong>in</strong>g environment which makes it ideally suited to the prototyp<strong>in</strong>g of algorithms.<br />

Secondly, it can perform symbolic as well as arbitrary-p<strong>re</strong>cision numerical calculations. Although<br />

quadratu<strong>re</strong> functions naturally place the ma<strong>in</strong> emphasis on numerical computation,<br />

the symbolic capabilities a<strong>re</strong> also needed for auxiliary functions and the test<strong>in</strong>g of the implementation.<br />

An additional advantage is that | <strong>in</strong> contrast to FORTRAN | Mathematica<br />

can cope with complex-valued <strong>in</strong>tegrands <strong>in</strong> arbitrary p<strong>re</strong>cision, which allows a very convenient<br />

programm<strong>in</strong>g of certa<strong>in</strong> wave <strong>in</strong>tegrals. F<strong>in</strong>ally, the visualisation of the <strong>re</strong>sults <strong>re</strong>qui<strong>re</strong>s<br />

no extra e orts and is <strong>re</strong>adily accomplished from with<strong>in</strong> Mathematica. The cumbersome<br />

and time-consum<strong>in</strong>g computations could of course be implemented <strong>in</strong> the signi cantly faster<br />

FORTRAN at a later date, but this was beyond the scope of this work.<br />

123


Chapter 5<br />

Numerical quadratu<strong>re</strong> and<br />

extrapolation<br />

5 Numerical quadratu<strong>re</strong> and extrapolation<br />

Someone has <strong>re</strong>cently de ned an applied mathematician as an <strong>in</strong>dividual<br />

enclosed <strong>in</strong> a small o ce and engaged <strong>in</strong> the study of mathematical problems<br />

which <strong>in</strong>te<strong>re</strong>st him personally; he waits for someone to stick his head<br />

<strong>in</strong> the door and <strong>in</strong>troduce himself by say<strong>in</strong>g, \I've got a problem." Usually<br />

the person com<strong>in</strong>g for help may beaphycicist, eng<strong>in</strong>eer, meteorologist,<br />

statistician, or chemist who has suddenly <strong>re</strong>ached a po<strong>in</strong>t <strong>in</strong> his <strong>in</strong>vestigation<br />

whe<strong>re</strong> he encounters a mathematical problem call<strong>in</strong>g for an unusual<br />

or nonstandard technique for its solution. [...] The range of mathematical<br />

topics from which queries may arise is all-<strong>in</strong>clusive. However, a topic which<br />

arises f<strong>re</strong>quently enough to merit some discussion is one which particularizes<br />

the statement \I've got a problem," to \I've got an <strong>in</strong>tegral."<br />

Milton Abramowitz [112]<br />

In the course of the case studies <strong>in</strong> the rst part, we encounte<strong>re</strong>d exclusively solutions <strong>in</strong> the<br />

form of <strong>in</strong> nite wave <strong>in</strong>tegrals. Unfortunately, these functions cannot be <strong>in</strong>tegrated analytically,<br />

and the<strong>re</strong>fo<strong>re</strong> we must <strong>re</strong>sort to numerical quadratu<strong>re</strong>. The term numerical quadratu<strong>re</strong><br />

is generally used <strong>in</strong> numerical mathematics to dist<strong>in</strong>guish the process of nd<strong>in</strong>g a numerical<br />

value for a de nite <strong>in</strong>tegral from the numerical solution of di e<strong>re</strong>ntial equations, which is then<br />

called numerical <strong>in</strong>tegration.<br />

The objective of this chapter is to give an overview of <strong>re</strong>levant aspects of this part of numerics.<br />

In particular, these a<strong>re</strong> univariate quadratu<strong>re</strong> and the problem of convergence or<br />

series acceleration, which is almost <strong>in</strong>evitable if one tries to compute <strong>in</strong>tegrals over an <strong>in</strong> nite<br />

range. Many of the p<strong>re</strong>sented algorithms have been <strong>in</strong>corporated <strong>in</strong> specialised computer rout<strong>in</strong>es<br />

that provide <strong>re</strong>ady-to-use solutions. Particular emphasis is placed upon these computer<br />

rout<strong>in</strong>es and their applicability to the class of <strong>in</strong>tegrals we a<strong>re</strong> <strong>in</strong>te<strong>re</strong>sted <strong>in</strong>.<br />

124


5.1 Univariate numerical quadratu<strong>re</strong><br />

5.1 Univariate numerical quadratu<strong>re</strong><br />

All e cient quadratu<strong>re</strong> methods approximate the <strong>in</strong>tegrand by elementary functions (polynomials)<br />

whose <strong>in</strong>tegrals can be determ<strong>in</strong>ed analytically. This is fairly simple if the <strong>in</strong>tegrand is<br />

smooth. Oscillatory functions, however, a<strong>re</strong> di cult to approximate and <strong>re</strong>qui<strong>re</strong> either higher<br />

order polynomials or a subdivision of the <strong>in</strong>tegration <strong>in</strong>terval. Th<strong>in</strong>gs get even worse if the<br />

<strong>in</strong>terval is <strong>in</strong> nite, like it is for our wave <strong>in</strong>tegrals. In this section we shall thus give a brief<br />

survey of both published algorithms and available computer rout<strong>in</strong>es that tackle the problem<br />

of univariate quadratu<strong>re</strong> of <strong>in</strong> nitely oscillat<strong>in</strong>g functions.<br />

5.1.1 Algorithms<br />

Many authors have conside<strong>re</strong>d nite <strong>in</strong>tegrals of the form<br />

I =<br />

Z b<br />

a<br />

f(x) e jq(x) dx : (5.1)<br />

Such an approach may be applicable to <strong>in</strong> nite <strong>in</strong>tegration <strong>in</strong>tervals if the <strong>in</strong>tegrand decays<br />

fast enough (exponentially) that a truncation is justi ed, so a look at some of the methods<br />

is <strong>re</strong>asonable (see also [113]). Lev<strong>in</strong> [114] pursues the idea that if f we<strong>re</strong> of the form f(x) =<br />

jq 0 (x)p(x)+p 0 (x), then the <strong>in</strong>tegral could be evaluated di<strong>re</strong>ctly. Thus the <strong>in</strong>tegration problem<br />

is transformed <strong>in</strong>to the problem of solv<strong>in</strong>g a di e<strong>re</strong>ntial equation.<br />

Evans [115] subdivides the <strong>in</strong>tegral<br />

I =<br />

NX<br />

i=1<br />

Z a+ih<br />

a+(i,1)h<br />

f(x) s<strong>in</strong><br />

cos<br />

!q(x)dx (5.2)<br />

and approximates both f(x) and q(x) <strong>in</strong> each sub<strong>in</strong>terval by l<strong>in</strong>ear or quadratic forms, <strong>re</strong>spectively.<br />

These can be <strong>in</strong>tegrated analytically, which is equivalent to the trapezoidal and<br />

Simpson's rules for general quadratu<strong>re</strong>. The quadratic approximation <strong>in</strong> fact dates back to<br />

a <strong>re</strong>markably early idea of Filon [116]. An enti<strong>re</strong>ly di e<strong>re</strong>nt approach is that of Eh<strong>re</strong>nmark<br />

[117] who proposes a th<strong>re</strong>e-po<strong>in</strong>t formula with weights chosen such that the formula is exact<br />

for the functions x, s<strong>in</strong> kx, and cos kx. This method is fairly easy to implement and, accord<strong>in</strong>g<br />

to Evans [115], lends itself to an extension to a higher number of po<strong>in</strong>ts.<br />

Xu and Mal [118] compute wave number <strong>in</strong>tegrals of the form R b<br />

f(x) cos(rx) dx by a modi ed<br />

a<br />

Clenshaw-Curtis scheme whe<strong>re</strong> they approximate only f(x) by Chebyshev polynomials. To<br />

achieve the desi<strong>re</strong>d accuracy, they turn to adaptive <strong>in</strong>terval subdivision if the given maximum<br />

order polynomial approximation is not su cient. For certa<strong>in</strong> <strong>in</strong>tegrands, specialised complexplane<br />

techniques as descibed by Davies [119] may be employed. By transform<strong>in</strong>g the contour<br />

of <strong>in</strong>tegration, the <strong>in</strong>tegral can be evaluated easily.<br />

If the <strong>in</strong>tegration <strong>in</strong>terval is <strong>in</strong> nite and the decay of the <strong>in</strong>tegrand is such that the <strong>in</strong>tegral<br />

converges too slowly, a common technique is to partition the <strong>in</strong>tegral and to extrapolate from<br />

a nite series of sub<strong>in</strong>tervals to its limit. All proposed methods <strong>re</strong>qui<strong>re</strong> some knowledge of<br />

125


5 Numerical quadratu<strong>re</strong> and extrapolation<br />

the <strong>in</strong>tegrand, especially about the distribution of its zeros. Some of the follow<strong>in</strong>g methods<br />

a<strong>re</strong> also discussed <strong>in</strong> the classic book of Davis and Rab<strong>in</strong>owitz [113] and the survey paper of<br />

Rab<strong>in</strong>owitz [120].<br />

Lyness [121] conside<strong>re</strong>s <strong>in</strong>tegrals of the type R 1<br />

f(x) dx with f(x) slowly decay<strong>in</strong>g and oscil-<br />

a<br />

lat<strong>in</strong>g <strong>in</strong> sign. The zeros a<strong>re</strong> <strong>re</strong>qui<strong>re</strong>d to be asymptotically equidistant like <strong>in</strong>f(x)=g(x)j(x)<br />

whe<strong>re</strong> j(x) is a circular or Bessel function and g(x) is positive over the whole <strong>in</strong>tegration<br />

<strong>in</strong>terval. Under these circumstances, the <strong>in</strong>tegral can be <strong>re</strong>-written as<br />

I =<br />

1X<br />

j=0<br />

Z xj+1<br />

xj<br />

f(x) dx =<br />

1X<br />

j=0<br />

wj ; (5.3)<br />

whe<strong>re</strong> the sub<strong>in</strong>terval length xj+1,xj equals the asymptotic distance between successive zeros<br />

of the <strong>in</strong>tegrand. The <strong>re</strong>sult<strong>in</strong>g series is then accelerated by simply calculat<strong>in</strong>g the average<br />

of two adjacent elements, which isvery e cient as long as the sequence wj is alternat<strong>in</strong>g <strong>in</strong><br />

sign. This idea was later used by Lyness and H<strong>in</strong>es [122] as the basis of a quadratu<strong>re</strong> rout<strong>in</strong>e,<br />

and it was further extended by Espelid and Overholt [123]. They suppose the oscillat<strong>in</strong>g<br />

behaviour of the <strong>in</strong>tegrand f(x) stems from a periodic function that is antisymmetric and<br />

posesses a constant period. Like Lyness, they partition the <strong>in</strong>terval <strong>in</strong>to half periods to c<strong>re</strong>ate<br />

an oscillat<strong>in</strong>g series amenable to extrapolation. However, they also observe the quadratu<strong>re</strong><br />

error <strong>in</strong> the sub<strong>in</strong>tervals and <strong>re</strong> ne the partition if necessary.<br />

It is exactly the Lyness partition that Haider and Liu use to calculate Fourier or Bessel<br />

transforms over nite <strong>in</strong>tegrals [124]. They call it `partitioned' Gaussian method, however,<br />

they make no attempt to adapt it to <strong>in</strong> nite <strong>in</strong>tegrals, s<strong>in</strong>ce the <strong>in</strong>tegrands they consider<br />

decay exponentially and the<strong>re</strong>fo<strong>re</strong> may be safely truncated at some large value of x. For<br />

functions with an <strong>in</strong> nite number of zeros, they admit that the method will fail.<br />

The methods discussed so far a<strong>re</strong> useful only if the <strong>in</strong>tegrand is at least asymptotically periodic.<br />

If, however, the distance between subsequent zeros vanishes as the <strong>in</strong>tegration variable<br />

approaches <strong>in</strong> nity, other algorithms a<strong>re</strong> <strong>re</strong>qui<strong>re</strong>d. Probably the most versatile ideas that a<strong>re</strong><br />

applicable to a wide range of even ir<strong>re</strong>gularly oscillat<strong>in</strong>g <strong>in</strong>tegrands have been developed by<br />

Sidi <strong>in</strong> a number of papers.<br />

In [125] Sidi <strong>in</strong>troduces a set of extrapolation methods (namely, the D- and D-transformations)<br />

e<br />

to compute <strong>in</strong>tegrals of the form R 1<br />

a f(x) dx by evaluat<strong>in</strong>g a number of <strong>in</strong>tegrals R xl f(x) dx,<br />

a<br />

whe<strong>re</strong> the xl a<strong>re</strong> chosen to be subsequent zeros of the f(x) with a


5.1 Univariate numerical quadratu<strong>re</strong><br />

Po<strong>in</strong>ca<strong>re</strong>-type asymptotic expansions<br />

(x) x<br />

1X<br />

i=0<br />

i<br />

: (5.5)<br />

xi Speci cally, (x) may be written as (x) = (x)+ (x), whe<strong>re</strong> its polynomial part (x) isof<br />

deg<strong>re</strong>e and (x) P 1 i=0 +i=x i as x !1. The xl a<strong>re</strong> taken to be the zeros of u , (x) ,so<br />

they approach the zeros of f(x) only asymptotically. The W -algorithm <strong>in</strong> comb<strong>in</strong>ation with<br />

a Clenshaw-Curtis rule was used by Hasegawa and Torii [127] to compute <strong>in</strong> nite <strong>in</strong>tegrals<br />

with circular functions as oscillat<strong>in</strong>g factors.<br />

R xl<br />

a f(x) dx by (xl) = R xl+1<br />

xl<br />

In an even later work [128], Sidi <strong>re</strong>turns to the ideas of Lyness and <strong>re</strong>places the <strong>in</strong>tegrals<br />

f(x) dx, thus form<strong>in</strong>g a sequence of <strong>in</strong>tegrals between successive<br />

zeros of u , (x) , whose members a<strong>re</strong> at least for large xl alternat<strong>in</strong>g <strong>in</strong> sign. Nonetheless, he<br />

uses a modi cation of his W -algorithm to accelerate the convergence of the <strong>re</strong>sult<strong>in</strong>g series<br />

P1i=,1 (xi), whe<strong>re</strong> (x,1) = Rx0 f(x)dx denotes the <strong>in</strong>tegral from the lower <strong>in</strong>terval limit<br />

a<br />

to the rst zero that is g<strong>re</strong>ater than a. In contrast to the orig<strong>in</strong>al W -transformation, the<br />

modi ed version needs only analysis of the phase (x) of oscillations, whe<strong>re</strong>as <strong>in</strong>formation<br />

on the amplitude is not <strong>re</strong>qui<strong>re</strong>d. The choice of the partition po<strong>in</strong>ts xl is simple, too, as it<br />

<strong>in</strong>volves noth<strong>in</strong>g mo<strong>re</strong> than the determ<strong>in</strong>ation of the largest zero of a given polynomial.<br />

As a variant to Sidi's method, Eh<strong>re</strong>nmark [129] proposes not to use subdivision, but aga<strong>in</strong><br />

the sequence of truncated <strong>in</strong>tegrals F (xn) = Rxn f(x)dx. However, he chooses not the zeros<br />

a<br />

of f(x) as xn, but seeks those values of x at which the truncated <strong>in</strong>tegral equals its limit<br />

limx!1 F (x).<br />

Another approach by Khanh [130] is based on the di<strong>re</strong>ct evaluation of the tail of the <strong>in</strong>tegral.<br />

It is applicable to <strong>in</strong>tegrands that satisfy either the l<strong>in</strong>ear di e<strong>re</strong>ntial equation of rst order<br />

f(x) =p(x)f 0 (x)orthat of second order f(x) =p(x)f 0 (x)+q(x)f 00 (x) with f(x), p(x), and<br />

q(x) hav<strong>in</strong>g asymptotic expansions at <strong>in</strong> nity. Then the tail R 1<br />

f(x) dx may be approximated<br />

x<br />

by a l<strong>in</strong>ear comb<strong>in</strong>ation of these expansions and their derivatives. The error can be controlled<br />

by choos<strong>in</strong>g the po<strong>in</strong>t whe<strong>re</strong> the tail is to be evaluated and the number of <strong>in</strong>volved terms<br />

su ciently large. The <strong>re</strong>ma<strong>in</strong><strong>in</strong>g nite <strong>in</strong>tegral R x<br />

f(x) dx has to be calculated us<strong>in</strong>g some<br />

a<br />

other method.<br />

5.1.2 Computer rout<strong>in</strong>es<br />

Compa<strong>re</strong>d to the large number of scienti c papers that have add<strong>re</strong>ssed the problem of <strong>in</strong> nitely<br />

oscillat<strong>in</strong>g <strong>in</strong>tegrals, the<strong>re</strong> a<strong>re</strong> astonish<strong>in</strong>gly few <strong>re</strong>ady-to-use computer programs available.<br />

Many of them a<strong>re</strong> public doma<strong>in</strong> softwa<strong>re</strong> and a<strong>re</strong> conta<strong>in</strong>ed <strong>in</strong> the comp<strong>re</strong>hensive library<br />

Netlib that is accessible via the World-Wide-Web under the add<strong>re</strong>ss http://www.netlib.org.<br />

Another entry po<strong>in</strong>t for seek<strong>in</strong>g computer rout<strong>in</strong>es is the GAMS (Guide to Available Mathematical<br />

Softwa<strong>re</strong>) that is ma<strong>in</strong>ta<strong>in</strong>ed by the US National Institute of Standards and Technology<br />

at http://gams.nist.gov and provides a classi cation scheme allow<strong>in</strong>g fast <strong>re</strong>trieval<br />

of appropriate rout<strong>in</strong>es. For other softwa<strong>re</strong> <strong>re</strong>sources on the Internet, see also the book by<br />

Uberhuber [131].<br />

127


5 Numerical quadratu<strong>re</strong> and extrapolation<br />

Search<strong>in</strong>g these <strong>in</strong>dices, one nds essentially th<strong>re</strong>e di e<strong>re</strong>nt quadratu<strong>re</strong> rout<strong>in</strong>es, all written<br />

<strong>in</strong> FORTRAN:<br />

OSCINT is part of the sublibrary Toms (toms/639) <strong>in</strong> Netlib and was written by<br />

Lyness and H<strong>in</strong>es [122]. It is aimed at <strong>in</strong>tegrands that may be separated <strong>in</strong>to an oscillat<strong>in</strong>g<br />

factor with an at least asymptotically constant period (i. e. circular or Bessel<br />

functions) and an ultimately positive factor. The user has to specify the function to<br />

be <strong>in</strong>tegrated, its (ultimate) period, and a start<strong>in</strong>g po<strong>in</strong>t from which onwards the extrapolation<br />

is performed. The algorithm then carries out a sequence of nite <strong>in</strong>terval<br />

quadratu<strong>re</strong>s, each over an <strong>in</strong>tegral hav<strong>in</strong>g the length of a half period, and accelerates<br />

the <strong>re</strong>sult<strong>in</strong>g sequence of partial sums by means of the Euler transformation [113]. By<br />

default the quadratu<strong>re</strong> is performed us<strong>in</strong>g the trapezoidal rule with a number of sampl<strong>in</strong>g<br />

po<strong>in</strong>ts given by the user. The quadratu<strong>re</strong> error is not controlled, and the user<br />

may as well <strong>re</strong>place the trapezoidal by a mo<strong>re</strong> sophisticated algorithm. The rout<strong>in</strong>e<br />

uses double p<strong>re</strong>cision arithmetic.<br />

DQAINF by Espelid and Overholt [123] can be found <strong>in</strong> the sublibrary Numeralgo<br />

(numeralgo/na6). They extend the approach of Lyness and H<strong>in</strong>es <strong>in</strong> that the <strong>in</strong>tegrand<br />

may be a l<strong>in</strong>ear comb<strong>in</strong>ation of periodic functions p(x), all with the same (ultimate)<br />

period, thus f(x) = Pk i=1 pi(x)gi(x). The gi(x) a<strong>re</strong> supposed to have similar asymptotic<br />

expansions as x ! 1, gi(x) = x , P1 j=0 c (i)<br />

j =xj , with > 0. The user must specify<br />

basically the period of the functions, the rst subdivision po<strong>in</strong>t, and of course an error<br />

tolerance. Then the algorithm computes <strong>in</strong>tegral approximations over half-period<br />

<strong>in</strong>tervals us<strong>in</strong>g a 21-node Gauss rule and estimates the true <strong>in</strong>tegral value by means<br />

of extrapolation. In addition, the quadratu<strong>re</strong> error <strong>in</strong> the <strong>in</strong>tervals is estimated and<br />

if necessary, the <strong>in</strong>terval is subdivided <strong>in</strong>to th<strong>re</strong>e equal parts. The<strong>re</strong> a<strong>re</strong> both s<strong>in</strong>gle<br />

(SQAINF) and double p<strong>re</strong>cision versions available.<br />

DQAWF is a rout<strong>in</strong>e of the Quadpack package [132], which is <strong>in</strong>cluded <strong>in</strong> many<br />

sublibraries of Netlib (Slatec, Cmlib). This double p<strong>re</strong>cision rout<strong>in</strong>e has also a s<strong>in</strong>gle<br />

p<strong>re</strong>cision counterpart (QAWF) and is designed for calculat<strong>in</strong>g Fourier transforms. Thus<br />

the <strong>in</strong>tegrand needs to be of the form f(x) s<strong>in</strong> !x or f(x) cos !x. With f(x) and ! given<br />

by the user, the algorithm <strong>in</strong>tegrates between the zeros of the oscillat<strong>in</strong>g factor. The<br />

quadratu<strong>re</strong> is based on a Clenshaw-Curtis rule <strong>in</strong> comb<strong>in</strong>ation with a globally adaptive<br />

subdivision strategy, so the<strong>re</strong> is virtually no limitation imposed on the function to<br />

be <strong>in</strong>tegrated. The convergence of the <strong>re</strong>sult<strong>in</strong>g series is then accelerated us<strong>in</strong>g the<br />

-algorithm [113, 133].<br />

Variants of this rout<strong>in</strong>e with slight modi cations may also be found <strong>in</strong> proprietary<br />

libraries. In the NAG library (The Numerical Algorithms Group) it is called D01ASF<br />

[134], <strong>in</strong> the IMSL (International Mathematical and Statistical Libraries) [135] it appears<br />

as DQDAWF.<br />

All these subrout<strong>in</strong>es a<strong>re</strong> best suited to <strong>in</strong>tegrands with a constant period as x !1. If this is<br />

not the case, like for s<strong>in</strong> (f(x)) with f(x) hav<strong>in</strong>g a polynomial part with deg<strong>re</strong>e g<strong>re</strong>ater than<br />

128


5.2 Convergence acceleration<br />

one, these rout<strong>in</strong>es cannot be applied di<strong>re</strong>ctly. It may be possible to extract an oscillatory<br />

factor with constant period <strong>in</strong> order to obta<strong>in</strong>, say, s<strong>in</strong>( ~ f(x)) cos !x such that <strong>in</strong>tegration<br />

can be carried out over <strong>in</strong>dividual `wave packets'. It is obvious that this <strong>re</strong>qui<strong>re</strong>s the use<br />

of a rout<strong>in</strong>e with an adaptive subdivision strategy like those from the Quadpack package.<br />

Depend<strong>in</strong>g on the parameters, however, the number of zeros between the zeros of the periodic<br />

envelope may be very high, which <strong>in</strong> turn <strong>in</strong>c<strong>re</strong>ases the number of subdivisions needed and<br />

the execution time of the program. For comparison, the example of wave propagation <strong>in</strong> the<br />

transmission l<strong>in</strong>e lled with plasma (section 3.7) was computed with the DQAWF subrout<strong>in</strong>e.<br />

Although it worked very well (and fast) <strong>in</strong> general, it failed to produce cor<strong>re</strong>ct <strong>re</strong>sults for<br />

special cases like the boundary.<br />

As po<strong>in</strong>ted out befo<strong>re</strong>, Sidi <strong>in</strong>vestigated ir<strong>re</strong>gularly oscillat<strong>in</strong>g functions [128], which would<br />

suit our purposes exactly. Unfortunately, the<strong>re</strong> is no implementation of his work available <strong>in</strong><br />

a standard subrout<strong>in</strong>e library. Thus the conclusions of this section a<strong>re</strong>:<br />

For the computation of <strong>in</strong> nitely oscillat<strong>in</strong>g <strong>in</strong>tegrals, a comb<strong>in</strong>ation of subdivid<strong>in</strong>g the<br />

<strong>in</strong>tegration range and extrapolat<strong>in</strong>g the sequence of partial sums is the method of choice.<br />

Although many authors have add<strong>re</strong>ssed this topic, the<strong>re</strong> is no <strong>re</strong>ady-to-use softwa<strong>re</strong><br />

package available for the case of ir<strong>re</strong>gular oscillations.<br />

5.2 Convergence acceleration<br />

One of the basic ideas <strong>in</strong> comput<strong>in</strong>g the value of an <strong>in</strong> nite <strong>in</strong>tegral is to de ne a sequence that<br />

converges to this value. What <strong>re</strong>ma<strong>in</strong>s is the problem of nd<strong>in</strong>g this limit numerically with<br />

su cient accuracy. This section will <strong>in</strong>troduce both the concept of series acceleration and<br />

the notation used throughout subsequent chapters. Furthermo<strong>re</strong>, we shall p<strong>re</strong>sent two widely<br />

used algorithms that a<strong>re</strong> <strong>re</strong>levant for the implementation of our own quadratu<strong>re</strong> strategy <strong>in</strong><br />

Mathematica.<br />

Suppose that a sequence fSng is known to converge and only a nite set of its members is<br />

available, then any member of it (usually the last one) may be taken as a rough estimate for<br />

the limit S. If, however, the convergence is very slow, the computational e ort to determ<strong>in</strong>e<br />

enough sequence members to achieve a <strong>re</strong>asonably accurate <strong>re</strong>sult may be too high. In such<br />

cases, alternatives a<strong>re</strong> sought to accelerate the convergence of the sequence. A very simple<br />

possibility is to form the average of two consecutive sequence members<br />

Tn = Sn + Sn+1<br />

2<br />

; (5.6)<br />

which is particularly suitable for alternat<strong>in</strong>g sequences (see, for example, Lyness [121] or Boyd<br />

[136] for an extension of this scheme). Such a sequence transformation T : fSng 7!fTngmust<br />

have two properties <strong>in</strong> order to be of practical use [137, 138]:<br />

It must be <strong>re</strong>gular, that is, fTng must converge to the same limit as fSng.<br />

fTng must converge faster than fSng, which is exp<strong>re</strong>ssed by limn!1 Tn,S<br />

Sn,S =0.<br />

129


5 Numerical quadratu<strong>re</strong> and extrapolation<br />

Apart from l<strong>in</strong>ear transformations (like the Euler transformation [139]), nonl<strong>in</strong>ear transformations<br />

a<strong>re</strong> widely used for convergence acceleration. One of them is Aitken's 2 process<br />

with the transformation rule [113, 137]<br />

Tn = SnSn+2 , S 2 n+1<br />

Sn+2 , 2 Sn+1 + Sn<br />

(Sn+1 , Sn) 2<br />

= Sn ,<br />

; (5.7)<br />

Sn+2 , 2 Sn+1 + Sn<br />

the second exp<strong>re</strong>ssion be<strong>in</strong>g p<strong>re</strong>ferable for implementation due to its better numerical stability.<br />

This transformation is known to accelerate sequences with l<strong>in</strong>ear convergence, that is [138,<br />

139],<br />

Sn+1 , S<br />

lim<br />

n!1 Sn , S<br />

and is exact (i. e. Tn = S 8n) for sequences of the form<br />

= ; 62 f1; 0g (5.8)<br />

Sn = S + a n ; a 6= 0; j j


5.2 Convergence acceleration<br />

(1)<br />

,1 =0<br />

(2)<br />

,1 =0<br />

(3)<br />

,1 =0<br />

(4)<br />

,1 =0<br />

(5)<br />

,1 =0<br />

.<br />

(0)<br />

0<br />

(1)<br />

0<br />

(2)<br />

0<br />

(3)<br />

0<br />

(4)<br />

0<br />

= S0<br />

= S1<br />

= S2<br />

= S3<br />

= S4<br />

.<br />

(0)<br />

1<br />

(1)<br />

1<br />

(2)<br />

1<br />

(3)<br />

1<br />

.<br />

(0)<br />

2 = S e(2) 0<br />

(1)<br />

2 = S e(2) 1<br />

(2)<br />

2 = S e(2) 2<br />

.<br />

(0)<br />

3<br />

(1)<br />

3<br />

.<br />

(0)<br />

4 = S e(4) 0<br />

Figu<strong>re</strong> 5.1: The -array. Only the values (n)<br />

2k <strong>in</strong> the even columns a<strong>re</strong> valid estimates for the sequence<br />

limit S.<br />

then the -algorithm gives the exact <strong>re</strong>sult (0)<br />

2p<br />

of the orig<strong>in</strong>al sequence a<strong>re</strong> <strong>re</strong>qui<strong>re</strong>d.<br />

.<br />

. ..<br />

= S [113], which means that 2p + 1 elements<br />

In any case, a sequence transformation may be seen as a functional <strong>re</strong>lation between a member<br />

of the transformed sequence and a certa<strong>in</strong> subset of the orig<strong>in</strong>al sequence. S<strong>in</strong>ce our goal is<br />

to obta<strong>in</strong> a better estimate for the limit of the sequence, we normally desi<strong>re</strong> only one <strong>re</strong>sult<br />

from the function and not a transformed sequence. Thus we can write<br />

Tn = e S (k)<br />

n = F (Sn;Sn+1 :::Sn+k) ; (5.13)<br />

with k +1 be<strong>in</strong>g the number of sequence members <strong>in</strong>volved <strong>in</strong> the transformation and n<br />

the start<strong>in</strong>g <strong>in</strong>dex. If the orig<strong>in</strong>al sequence comprised N members, then we would typically<br />

compute S e(k) N,k or S e(N) 0 as the (hopefully) best approximation to the exact limit S.<br />

131


Chapter 6<br />

Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

\Can you do Addition?" the White Queen asked. \What's one and one<br />

and one and one and one and one and one and one and one and one?"<br />

\I don't know," said Alice. \I lost count."<br />

\She can't do Addition," the Red Queen <strong>in</strong>terrupted. \Can you do Subtraction?<br />

Take n<strong>in</strong>e from eight."<br />

\N<strong>in</strong>e from eight I can't, you know," Alice <strong>re</strong>plied very <strong>re</strong>adily: \but| "<br />

\She can't do Subtraction," said the White Queen.<br />

Lewis Carroll, Through the look<strong>in</strong>g glass<br />

With the discourag<strong>in</strong>g <strong>re</strong>sult of the search for an o -the-shelf program to solve our problems,<br />

the<strong>re</strong> is noth<strong>in</strong>g left but to write a dedicated quadratu<strong>re</strong> rout<strong>in</strong>e that suits our needs. This<br />

is, however, not at all a straightforward task, and we have to avoid ca<strong>re</strong>fully a number of<br />

pitfalls associated with the <strong>in</strong> nitely oscillat<strong>in</strong>g <strong>in</strong>tegrals we wish to compute. This chapter is<br />

thus devoted to some practical considerations that emerge <strong>in</strong> the course of the development<br />

of such a quadratu<strong>re</strong> rout<strong>in</strong>e. While it seems obvious that a comb<strong>in</strong>ation of partition<strong>in</strong>g and<br />

extrapolation is best suited to this purpose, the<strong>re</strong> a<strong>re</strong> some detailed questions yet to solve.<br />

The most <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g one is certa<strong>in</strong>ly the choice of the partition itself, and we shall the<strong>re</strong>fo<strong>re</strong><br />

discuss which po<strong>in</strong>ts a<strong>re</strong> most suitable for this subdivision procedu<strong>re</strong> | the zeros, ext<strong>re</strong>ma,<br />

or any other po<strong>in</strong>ts of the <strong>in</strong>tegrand. A rather implementation-speci c question is how the<br />

partial <strong>in</strong>tegrals a<strong>re</strong> to be computed e ciently. We shall also <strong>in</strong>vestigate a strategy how to<br />

control the accuracy of the computation.<br />

Throughout the subsequent sections, we shall p<strong>re</strong>sent many examples programmed <strong>in</strong> Mathematica<br />

to illustrate the various di culties that need to be conside<strong>re</strong>d. To understand these<br />

pieces of code, a certa<strong>in</strong> familiarity with Mathematica or any other functional programm<strong>in</strong>g<br />

language is favourable, but not imperative. S<strong>in</strong>ce most of the used functions a<strong>re</strong> selfexpla<strong>in</strong><strong>in</strong>g,<br />

we shall give further explanations only whe<strong>re</strong> they a<strong>re</strong> deemed necessary.<br />

132


6.1 Partition<strong>in</strong>g the <strong>in</strong>tegration <strong>in</strong>terval<br />

6.1 Partition<strong>in</strong>g the <strong>in</strong>tegration <strong>in</strong>terval<br />

Once we know that an <strong>in</strong> nite <strong>in</strong>tegral exists, we can theo<strong>re</strong>tically choose any arbitrary<br />

partition of the <strong>in</strong>terval to calculate the sequence of partial sums. In practice, however, we<br />

a<strong>re</strong> better o to select the subdivision po<strong>in</strong>ts such that the <strong>re</strong>sult<strong>in</strong>g sequence is amenable<br />

to convergence acceleration. A ca<strong>re</strong>ful choice is al<strong>re</strong>ady half the guarantee for a trustworthy<br />

<strong>re</strong>sult. Let an example illustrate this pitfall.<br />

We consider the well-known <strong>in</strong>tegral<br />

Z 1<br />

0<br />

s<strong>in</strong> x<br />

x dx = : (6.1)<br />

2<br />

We subdivide the <strong>in</strong>terval at the zeros of the s<strong>in</strong>e and calculate the rst 100 elements of<br />

the sequence of partial sums. To this end, we rst evaluate the contributions of the <strong>in</strong>tegrals<br />

Ii = R xi+1<br />

f(x) dx between the partition po<strong>in</strong>ts and then accumulate them to yield the partial<br />

xi<br />

sums Si = Pi k=0 Ii. We nally plot this alternat<strong>in</strong>g sequence.<br />

In[1]:= sequ = Table[NIntegrate[S<strong>in</strong>[x]/x,fx,i Pi,(i+1) Pig],<br />

fi,0,100g];<br />

partial = FoldList[Plus,0,sequ];<br />

1.59<br />

1.58<br />

1.57<br />

1.56<br />

1.55<br />

1.54<br />

z1 = ListPlot[partial,PlotStyle->fPo<strong>in</strong>tSize[0.012]g];<br />

20 40 60 80 100<br />

In[2]:= SequenceLimit[partial] - N[Pi/2]<br />

Out[2]= -15<br />

-1.77636 10<br />

133


6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Remark (Cumulative sums) Us<strong>in</strong>g the function FoldList is a very elegant and fast<br />

way to compute cumulative sums of a given list. FoldList[Plus,x,fa,b,cg] <strong>re</strong>turns<br />

fx,x+a,x+a+b,x+a+b+cg [140]. Normally, x will be set to zero. If the lead<strong>in</strong>g zero element<br />

of the <strong>re</strong>sult might disturb subsequent calculations, it can be omitted with Drop[list,1].<br />

The numerically found limit of this sequence is exact with<strong>in</strong> the mach<strong>in</strong>e p<strong>re</strong>cision of 16 digits.<br />

Now we <strong>in</strong>advertently skip every other zero <strong>in</strong> the calculation of the series and <strong>in</strong>tegrate<br />

over full <strong>in</strong>stead of half periods. Consequently the series no longer alternates, which has a<br />

catastrophic e ect on the determ<strong>in</strong>ation of the limit.<br />

In[3]:= sequ = Table[NIntegrate[S<strong>in</strong>[x]/x,fx,2 i Pi,2 (i+1) Pig],<br />

fi,0,100g];<br />

partial = FoldList[Plus,0,sequ];<br />

1.5675<br />

1.565<br />

1.5625<br />

1.56<br />

1.5575<br />

1.555<br />

1.5525<br />

z2 = ListPlot[partial,PlotStyle->fPo<strong>in</strong>tSize[0.012]g];<br />

20 40 60 80 100<br />

In[4]:= SequenceLimit[partial] - N[Pi/2]<br />

Out[4]= -0.00025589<br />

Remark (Improvement of the <strong>re</strong>sult) The <strong>re</strong>sult may neither be improved by <strong>in</strong>c<strong>re</strong>as<strong>in</strong>g<br />

the p<strong>re</strong>cision of the numerical computations nor by <strong>in</strong>clud<strong>in</strong>g a larger number of<br />

sequence members <strong>in</strong> the extrapolation.<br />

The example may be conside<strong>re</strong>d rather pathological, but we can con rm the di e<strong>re</strong>nces between<br />

half- and full-period <strong>in</strong>tervals also for other logarithmically converg<strong>in</strong>g <strong>in</strong>tegrals like<br />

134


6.1 Partition<strong>in</strong>g the <strong>in</strong>tegration <strong>in</strong>terval<br />

R 1<br />

s<strong>in</strong> x 0 2 dx =<br />

functions like<br />

p =2<br />

2<br />

or R 1<br />

0<br />

s<strong>in</strong> x2 x dx = . For exponentially decay<strong>in</strong>g <strong>in</strong>tegrands or even<br />

4<br />

s<strong>in</strong> x<br />

(1+x) p ; p>2, the di e<strong>re</strong>nce <strong>in</strong> the extrapolation error vanishes. We conclude<br />

that the -algorithm (which is actually beh<strong>in</strong>d SequenceLimit) yields far mo<strong>re</strong> <strong>re</strong>liable <strong>re</strong>sults<br />

if it is applied to alternat<strong>in</strong>g series.<br />

This experience is <strong>in</strong> l<strong>in</strong>e with a number of articles that have been published over the years.<br />

Smith and Ford [141] found this algorithm to be particularly useful for such series. Bell and<br />

Phillips [142] proved that Aitken's 2 process is well-conditionend for all cases whe<strong>re</strong> the<br />

errors en = Sn , S oscillate <strong>in</strong> sign. Wynn [143] showed the stability of the -algorithm<br />

for oscillat<strong>in</strong>g sequences. Tailo<strong>re</strong>d to his (forward averag<strong>in</strong>g) acceleration algorithm, Lyness<br />

[122] also mentions the problems associated with <strong>in</strong>terval subdivision like a wrong asymptotic<br />

distance between the consecutive zeros. As for monotonic sequences, Smith and Ford [141]<br />

tested several convergence acceleration algorithms on logarithmically converg<strong>in</strong>g series like<br />

P 11 1=n 2 and found that the -algorithm failed on all of them.<br />

Remark (Possible alternative) An alternative would have been to use the function<br />

EulerSum conta<strong>in</strong>ed <strong>in</strong> the Standard Package NumericalMath`NLimit` that uses an iterated<br />

Euler transformation [144]. However, the <strong>re</strong>sults a<strong>re</strong> practically the same as with<br />

the use of SequenceLimit. In addition, the option Terms that speci es how may terms<br />

a<strong>re</strong> explicitly summed up befo<strong>re</strong> extrapolation has to be set to a higher value than the default<br />

to achieve the desi<strong>re</strong>d accuracy even for the favourable case. Inc<strong>re</strong>as<strong>in</strong>g ExtraTerms,<br />

which is the number of sequence members used <strong>in</strong> the extrapolation process, can <strong>re</strong>sult <strong>in</strong><br />

complete garbage <strong>in</strong> the case of full-period <strong>in</strong>tervals.<br />

The second possibility for the choice of the partition po<strong>in</strong>ts is to take the ext<strong>re</strong>ma of the<br />

oscillat<strong>in</strong>g <strong>in</strong>tegrand. Lyness discove<strong>re</strong>d this to be superior to subdivision at the zeros, but<br />

only by a <strong>re</strong>latively small marg<strong>in</strong>. Espelid and Overholt [123] found that it improves the<br />

convergence by an order of magnitude. In fact it <strong>re</strong>duces the possible truncation error as<br />

the positive and negative contributions with<strong>in</strong> a sub<strong>in</strong>terval tend to cancel each other and<br />

the partial <strong>in</strong>tegrals the<strong>re</strong>fo<strong>re</strong> a<strong>re</strong> much smaller. This e ect is somehow comparable to the<br />

transformation of an alternat<strong>in</strong>g series by simply calculat<strong>in</strong>g the mean value of two adjacent<br />

members. We apply this strategy to our example by simply shift<strong>in</strong>g the <strong>in</strong>tegration limits by<br />

] separately.<br />

2 and <strong>in</strong>clud<strong>in</strong>g the <strong>in</strong>tegral over [0; 2<br />

In[5]:= sequ = Table[NIntegrate[S<strong>in</strong>[x]/x,fx,i Pi + Pi/2,(i+1) Pi + Pi/2g],<br />

fi,0,100g];<br />

partial = FoldList[Plus,NIntegrate[S<strong>in</strong>[x]/x,fx,0,Pi/2g],sequ];<br />

SequenceLimit[partial] - N[Pi/2]<br />

Out[5]= -14<br />

-5.79536 10<br />

It is not surpris<strong>in</strong>g that the <strong>in</strong>tegration over half periods is as accurate as befo<strong>re</strong>. However,<br />

the experiment with full periods now gives a better <strong>re</strong>sult.<br />

135


1.575<br />

1.57<br />

1.565<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

20 40 60 80 100<br />

Figu<strong>re</strong> 6.1: Sequences of partial sums for R 1 s<strong>in</strong> x<br />

0 x dx. The partition po<strong>in</strong>ts a<strong>re</strong> the zeros ( ) or ext<strong>re</strong>ma<br />

( ) of the <strong>in</strong>tegrand. The <strong>in</strong>ner sequences a<strong>re</strong> obta<strong>in</strong>ed when the <strong>in</strong>tegration range for the sub<strong>in</strong>tervals<br />

is a full period.<br />

In[6]:= sequ = Table[NIntegrate[S<strong>in</strong>[x]/x,<br />

fx,2 i Pi + Pi/2,2 (i+1) Pi + Pi/2g],<br />

fi,0,100g];<br />

partial = FoldList[Plus,NIntegrate[S<strong>in</strong>[x]/x,fx,0,Pi/2g],sequ];<br />

SequenceLimit[partial] - N[Pi/2]<br />

Out[6]= -7<br />

-2.13857 10<br />

Fig. 6.1 shows the sequences of partial sums calculated for this example and <strong>re</strong>veals that the<br />

ext<strong>re</strong>ma a<strong>re</strong> <strong>in</strong>deed mo<strong>re</strong> suitable as partition po<strong>in</strong>ts because the errors jSn , Sj dec<strong>re</strong>ase like<br />

1=n 2 <strong>in</strong> this case. When the zeros a<strong>re</strong> used to subdivide the <strong>in</strong>tegration range, the errors<br />

behave like 1=n, which gives only half the speed of convergence.<br />

Proof We shall rst derive an estimate for the magnitude of the partial <strong>in</strong>tegrals. The second meanvalue<br />

theo<strong>re</strong>m states that if two functions f(x) and g(x) a<strong>re</strong> cont<strong>in</strong>uous on [a; b] and g(x) is also<br />

monotonic, then the<strong>re</strong> exists a po<strong>in</strong>t 2 [a; b] such that<br />

Z b<br />

a<br />

Z<br />

f(x) g(x) dx = g(a+)<br />

a<br />

Z b<br />

f(x) dx + g(b,) f(x) dx : (6.2)<br />

We set f(x) = s<strong>in</strong> x, g(x) =1=x and choose the <strong>in</strong>tegration <strong>in</strong>terval [(n + ) ; (n +1+ ) ], whe<strong>re</strong><br />

136


6.1 Partition<strong>in</strong>g the <strong>in</strong>tegration <strong>in</strong>terval<br />

0


6.2 Choos<strong>in</strong>g the rst partition po<strong>in</strong>t<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

When us<strong>in</strong>g an extrapolation method to determ<strong>in</strong>e the value of an <strong>in</strong>tegral, we face the<br />

problem that we must conclude from a nite sequence of partial sums to its limit. While it<br />

is normally easy to judge the existence of an <strong>in</strong>de nite <strong>in</strong>tegral, it takes some considerations<br />

to nd the right start<strong>in</strong>g po<strong>in</strong>t for the sequence be<strong>in</strong>g subject to the extrapolation algorithm.<br />

If the choice was wrong, the algorithm may <strong>in</strong> some cases <strong>re</strong>turn a warn<strong>in</strong>g message if it<br />

encounters convergence problems. However, the check<strong>in</strong>g capabilities a<strong>re</strong> limited and strongly<br />

data dependent, and most extrapolation algorithms even nd nite <strong>re</strong>sults for divergent series.<br />

Hence the <strong>re</strong>sponsibility to be ca<strong>re</strong>ful always <strong>re</strong>sts with the user.<br />

To illustrate the possible di culties, we <strong>re</strong>gard a simple example. Consider the <strong>in</strong>tegral<br />

I =<br />

Z 1<br />

a<br />

1<br />

x +1 s<strong>in</strong> (x , xm) 2 dx : (6.8)<br />

We want toevaluate it through extrapolation and choose the partition po<strong>in</strong>ts to be the zeros<br />

of the s<strong>in</strong>-function. For the sake of simplicity, we assume that the lower <strong>in</strong>tegration limit a<br />

be the smallest positive zero. As we need the sequence of partial sums, we must de ne a<br />

function that <strong>re</strong>turns all zeros of the <strong>in</strong>tegrand <strong>in</strong> ascend<strong>in</strong>g order, start<strong>in</strong>g with <strong>in</strong>dex 0. To<br />

that end, it is essential to dist<strong>in</strong>guish the two branches of the parabolic argument (x,xm) 2 .<br />

It is easy to see that the<strong>re</strong> a<strong>re</strong> o = bxm 2 = c zeros between the lower limit of <strong>in</strong>tegration and<br />

the m<strong>in</strong>imum of the argument function at x = xm, so this value can be used as an <strong>in</strong>dex o set<br />

for the calculation of the zeros.<br />

In Mathematica notation, the <strong>in</strong>tegrand and the function <strong>re</strong>turn<strong>in</strong>g the k-th positive zero<br />

xk a<strong>re</strong> given below. The position of the ext<strong>re</strong>mum xm and the <strong>in</strong>dex o set o a<strong>re</strong> global<br />

parameters.<br />

In[7]:= f[x_] := 1/(x+1) S<strong>in</strong>[(x-xm)^2];<br />

zero[k_] := If[k >= o, xm + Sqrt[(k-o) Pi],<br />

xm - Sqrt[(o-k) Pi]];<br />

Now we calculate the rst 100 elements of the sequence of partial sums with the zeros used<br />

as subdivision po<strong>in</strong>ts.<br />

In[8]:= xm = 15;<br />

o = N[Floor[xm^2/Pi]];<br />

zmax = 100;<br />

sequ = Table[NIntegrate[f[x],fx,zero[i],zero[i+1]g,<br />

Method->GaussKronrod],fi,0,zmaxg];<br />

partial = FoldList[Plus,0,sequ];<br />

ListPlot[partial,PlotStyle->fPo<strong>in</strong>tSize[0.006]g];<br />

138


6.2 Choos<strong>in</strong>g the rst partition po<strong>in</strong>t<br />

0.12<br />

0.1<br />

0.08<br />

0.06<br />

0.04<br />

0.02<br />

20 40 60 80 100<br />

The surpris<strong>in</strong>g step <strong>in</strong> the sequence stems from the m<strong>in</strong>imum of the argument function that<br />

causes the distance of successive zeros to be enlarged <strong>in</strong> its neighbourhood, which <strong>in</strong> turn<br />

<strong>re</strong>sults <strong>in</strong> exceed<strong>in</strong>gly large partial <strong>in</strong>tegrals. For the rst 50 elements, this e ect is hidden<br />

1<br />

by the dom<strong>in</strong>at<strong>in</strong>g factor <strong>in</strong> the <strong>in</strong>tegrand and the sequence seems to converge. Indeed,<br />

x+1<br />

the extrapolation algorithm <strong>re</strong>turns a valid limit without any warn<strong>in</strong>g when we feed it with<br />

the sequence elements 20 to 30.<br />

In[9]:= SequenceLimit[Take[partial,f20,30g]]<br />

Out[9]= 0.0313787<br />

We get a <strong>re</strong>sult even if we use the portion of the sequence that seems to diverge as we approach<br />

the ext<strong>re</strong>mum of the argument function.<br />

In[10]:= SequenceLimit[Take[partial,f60,70g]]<br />

Out[10]= 0.0313787<br />

Only if we start the extrapolation beh<strong>in</strong>d the step, we obta<strong>in</strong> the cor<strong>re</strong>ct limit and the <strong>re</strong>al<br />

value of the <strong>in</strong>tegral.<br />

In[11]:= SequenceLimit[Take[partial,f80,90g]]<br />

Out[11]= 0.109863<br />

Remark (Possible warn<strong>in</strong>gs) If we had <strong>re</strong>placed the factor x+1 by x2 , we would<br />

+1<br />

have <strong>re</strong>ceived at least a warn<strong>in</strong>g that the algorithm suspected the <strong>re</strong>sults to be <strong>in</strong>cor<strong>re</strong>ct <strong>in</strong><br />

the rst two cases (the <strong>re</strong>turned values would have been `cor<strong>re</strong>ct' anyhow). However, one<br />

cannot <strong>re</strong>ly completely on such messages as the<strong>re</strong> a<strong>re</strong> other <strong>in</strong>tegrands whe<strong>re</strong> the <strong>re</strong>sults<br />

a<strong>re</strong> de nitely right despite these warn<strong>in</strong>gs.<br />

139<br />

1<br />

x


6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

The conclusion to be drawn from this example is that the rst partition po<strong>in</strong>t must be beyond<br />

any ext<strong>re</strong>mum or even po<strong>in</strong>t of <strong>in</strong> exion of the argument function. The most rigorous way<br />

to ensu<strong>re</strong> this is to <strong>re</strong>qui<strong>re</strong> that the argument function and its derivatives be monotonic for<br />

x>x0. From this po<strong>in</strong>t of view, many authors seem to be rather sloppy <strong>in</strong> the choice of the<br />

partition.<br />

Sidi [128], for example, takes x0 to be the rst zero of the circular function s<strong>in</strong> ( (x)) that is<br />

g<strong>re</strong>ater than the lower <strong>in</strong>tegration limit a. Thus x0 satis es the polynomial equation (x) =q<br />

for some <strong>in</strong>teger q. The consecutive partition po<strong>in</strong>ts xi ; i>0 a<strong>re</strong> then determ<strong>in</strong>ed to be the<br />

largest positive solution of (x) =(q+l) . As can be seen from our example for large values<br />

of xm, this <strong>re</strong>sults <strong>in</strong> a large second partial <strong>in</strong>tegral I1. A solution would be to determ<strong>in</strong>e<br />

also x0 as largest root of (x) =q and not only as the zero closest to a. But then aga<strong>in</strong> we<br />

would obta<strong>in</strong> an unnecessarily large partial <strong>in</strong>terval (this time I0).<br />

Even worse is the fact that although the proposed method guarantees that the xi a<strong>re</strong> of<br />

ascend<strong>in</strong>g order, it does by no means ensu<strong>re</strong> that problems like the afo<strong>re</strong>mentioned a<strong>re</strong> avoided.<br />

If the<strong>re</strong> a<strong>re</strong>, for example, ext<strong>re</strong>ma at large ord<strong>in</strong>ate values, they might not be comprised <strong>in</strong><br />

the sequence of the partial <strong>in</strong>tegrals because the series seems to converge earlier and the<br />

evaluation is truncated consequently.<br />

Espelid and Overholt [123] also <strong>re</strong>qui<strong>re</strong> the rst partition po<strong>in</strong>t to be supplied by the user.<br />

They can of course provide no <strong>in</strong>formation other than to choose it small enough to keep the<br />

computational costs for the rst <strong>in</strong>terval low and large enough that the asymptotic properties<br />

of the functions a<strong>re</strong> satis ed.<br />

The<strong>re</strong> is still another side to the story: the non-oscillat<strong>in</strong>g factor of the <strong>in</strong>tegrand ought tobe<br />

smooth compa<strong>re</strong>d to the oscillat<strong>in</strong>g part, otherwise the sequence of partial sums may aga<strong>in</strong><br />

show discont<strong>in</strong>uities. Not only do discont<strong>in</strong>uous factors like the step function provoke such<br />

problems, also rather <strong>in</strong>nocent functions like e ,x2 or tanh(x) can hamper the straightforward<br />

use of extrapolation. The<strong>re</strong>fo<strong>re</strong> these parts of the <strong>in</strong>tegrand, too, should be monotonic for<br />

x>x0, as should their derivatives. This aspect is also ra<strong>re</strong>ly add<strong>re</strong>ssed <strong>in</strong> the literatu<strong>re</strong>.<br />

Sidi [128] does not consider the properties of the modulat<strong>in</strong>g factor at all, he just states<br />

that it must be non-oscillat<strong>in</strong>g at x ! 1 and have an asymptotic expansion such that<br />

the enti<strong>re</strong> function is <strong>in</strong>tegrable at <strong>in</strong> nity. Lyness [121] brie y discusses the problems of<br />

unexpected peaks <strong>in</strong> the <strong>in</strong>tegrand that a<strong>re</strong> too far out to be <strong>in</strong>cluded <strong>in</strong> the sequence used<br />

for extrapolation. Consequently, for his quadratu<strong>re</strong> rout<strong>in</strong>e [122], he lets the user decide whe<strong>re</strong><br />

to start the extrapolation.<br />

Remark (Connection to physics) We have al<strong>re</strong>ady encounte<strong>re</strong>d the fact that wave<br />

<strong>in</strong>tegrals a<strong>re</strong> dom<strong>in</strong>ated by a rather small <strong>re</strong>gion of the <strong>in</strong>tegration range. In fact, the<br />

method of stationary phase is based upon this nd<strong>in</strong>g. The mathematical formulation<br />

that an ext<strong>re</strong>mum of the argument function yields a large partial <strong>in</strong>tegral is equivalent<br />

to the concept of local wave number and f<strong>re</strong>quency, both of which a<strong>re</strong> determ<strong>in</strong>ed by the<br />

ext<strong>re</strong>ma of the wave's phase.<br />

140


6.3 How to compute the rst <strong>in</strong>tegral<br />

6.3 How to compute the rst <strong>in</strong>tegral<br />

While the <strong>in</strong>tegrals Ii = R xi+1<br />

f(x) dx to the right of the rst partition po<strong>in</strong>t x0 a<strong>re</strong> computed<br />

xi<br />

between successive zeros and the<strong>re</strong>fo<strong>re</strong> have smooth <strong>in</strong>tegrands, the <strong>in</strong>tegration range of the<br />

rst <strong>in</strong>tegral I0 = R x0<br />

f(x) dx can comprise mo<strong>re</strong> than just two zeros of the <strong>in</strong>tegrand. In<br />

a<br />

fact, the <strong>in</strong>tegrand can be either smooth or strongly oscillat<strong>in</strong>g, depend<strong>in</strong>g on the parameters<br />

of the <strong>in</strong>tegrand. For an automatic quadratu<strong>re</strong> rout<strong>in</strong>e, however, we need a strategy that can<br />

cope with a broad variety of functions.<br />

For the numerical evaluation of de nite <strong>in</strong>tegrals, Mathematica o ers th<strong>re</strong>e di e<strong>re</strong>nt methods<br />

that can be selected with the <strong>re</strong>spective choice for the option Method of the function<br />

NIntegrate. The algorithms may be found <strong>in</strong> any textbook (for example Davis [113] or<br />

Uberhuber [131]), the Mathematica-speci c details a<strong>re</strong> t<strong>re</strong>ated by Keiper [145].<br />

GaussKronrod <strong>re</strong>fers to an adaptive Gauss-Kronrod quadratu<strong>re</strong> and is the default value<br />

for Method. Ow<strong>in</strong>g to its simplicity, itiswell suited to smooth <strong>in</strong>tegrands.<br />

DoubleExponential selects an algorithm (DE) that transforms the <strong>in</strong>tegrand prior to<br />

quadratu<strong>re</strong>. This method gives particularly good <strong>re</strong>sults with oscillat<strong>in</strong>g <strong>in</strong>tegrands.<br />

Trapezoidal uses the well-known trapezoidal rule. It is useful for periodic <strong>in</strong>tegrands<br />

whe<strong>re</strong> the quadratu<strong>re</strong> <strong>in</strong>terval is exactly one period. Apart from this case its performance<br />

is rather poor.<br />

All methods estimate the error <strong>in</strong> the approximation. For the trapezoidal and the DE rule,<br />

which e ectively uses the trapezoidal rule after the variable transformation, the stepsize is<br />

halved if the error is too large. This way the p<strong>re</strong>vious evaluations of the <strong>in</strong>tegrand may<br />

be <strong>in</strong>corporated <strong>in</strong> the <strong>re</strong> ned computation. The Gauss-Kronrod rule, on the other hand,<br />

is adaptive <strong>in</strong> that it <strong>re</strong>cursively subdivides the <strong>in</strong>tegration <strong>in</strong>terval if necessary and is thus<br />

able to concentrate its e orts on the <strong>re</strong>gions whe<strong>re</strong> the error is <strong>re</strong>ally high. The<strong>re</strong> a<strong>re</strong> two<br />

important parameters that control these subdivision procedu<strong>re</strong>s:<br />

P<strong>re</strong>cisionGoal speci es how many digits of p<strong>re</strong>cision the <strong>re</strong>sult of the computation<br />

should have. E ectively, this sets the limit for the <strong>re</strong>lative error. The default value<br />

is Automatic, which yields a p<strong>re</strong>cision goal with ten digits less than the cur<strong>re</strong>nt sett<strong>in</strong>g<br />

of Work<strong>in</strong>gP<strong>re</strong>cision (which <strong>in</strong> turn defaults to the mach<strong>in</strong>e-dependent constant<br />

$Mach<strong>in</strong>eP<strong>re</strong>cision). The sett<strong>in</strong>g of P<strong>re</strong>cisionGoal is by no means a guarantee that<br />

the <strong>re</strong>sult will satisfy this <strong>re</strong>qui<strong>re</strong>ment.<br />

AccuracyGoal sets the number of digits to the right of the decimal po<strong>in</strong>t that should<br />

be cor<strong>re</strong>ct <strong>in</strong> the <strong>re</strong>sult and thus speci es the allowed absolute error <strong>in</strong> the numerical<br />

rout<strong>in</strong>e. The default sett<strong>in</strong>g is Inf<strong>in</strong>ity, which means that this criterion shall not be<br />

used. As with P<strong>re</strong>cisionGoal, the actual <strong>re</strong>sult may have less digits of accuracy than<br />

desi<strong>re</strong>d.<br />

The terms p<strong>re</strong>cision and accuracy have very dist<strong>in</strong>ct mean<strong>in</strong>gs <strong>in</strong> Mathematica, other than <strong>in</strong><br />

common use. The di e<strong>re</strong>nce is best illustrated by an example.<br />

141


6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Example 6.3.1 The function P<strong>re</strong>cision <strong>re</strong>turns the number of digits a oat<strong>in</strong>g-po<strong>in</strong>t number<br />

posesses. For exact numbers, such as <strong>in</strong>tegers or fractions of <strong>in</strong>tegers, it <strong>re</strong>turns Inf<strong>in</strong>ity. For<br />

ord<strong>in</strong>ary mach<strong>in</strong>e p<strong>re</strong>cision numbers, the <strong>re</strong>sult is $Mach<strong>in</strong>eP<strong>re</strong>cision, even if the number has less<br />

digits [146].<br />

In[12]:= P<strong>re</strong>cision[1234.5678901234567890]<br />

Out[12]= 19<br />

In[13]:= P<strong>re</strong>cision[1234.56]<br />

Out[13]= 16<br />

The function Accuracy, on the other hand, <strong>re</strong>fers to the digits to the right of the decimal po<strong>in</strong>t. Like<br />

befo<strong>re</strong>, it <strong>re</strong>turns Inf<strong>in</strong>ity if the number is exact, and assumes mach<strong>in</strong>e p<strong>re</strong>cision numbers if the<br />

argument has less than | <strong>in</strong> our case | 16 digits.<br />

In[14]:= Accuracy[1234.5678901234567890]<br />

Out[14]= 16<br />

In[15]:= Accuracy[1234.56]<br />

Out[15]= 13<br />

In fact, both functions signify the error bounds of a oat<strong>in</strong>g-po<strong>in</strong>t number <strong>in</strong> Mathematica. The<br />

accuracy of a number x then is the absolute error bound log j j while the p<strong>re</strong>cision means the<br />

<strong>re</strong>lative error bound log jx= j [144]. The maximum error clearly depends on the work<strong>in</strong>g p<strong>re</strong>cision.<br />

Both values a<strong>re</strong> rounded to the nea<strong>re</strong>st <strong>in</strong>teger and thus look like numbers of signi cant digits although<br />

they can be <strong>in</strong>cor<strong>re</strong>ct by one digit due to the round<strong>in</strong>g.<br />

All other parameters of NIntegrate a<strong>re</strong> not of primary importance for our purpose he<strong>re</strong> and<br />

will be expla<strong>in</strong>ed as needed <strong>in</strong> the follow<strong>in</strong>g chapters.<br />

We will now explo<strong>re</strong> the two di e<strong>re</strong>nt quadratu<strong>re</strong> algorithms as they a<strong>re</strong> applied to the rst<br />

partial <strong>in</strong>tegral over the <strong>in</strong>terval [a; x0] between the left <strong>in</strong>tegration limit and the rst partition<br />

po<strong>in</strong>t. Let us <strong>re</strong>turn to the example (6.8) of the p<strong>re</strong>vious section. If we set xm = 40, we obta<strong>in</strong><br />

an <strong>in</strong>dex o set for the step <strong>in</strong> the sequence of partial sums o = 509 such that we select our rst<br />

partition po<strong>in</strong>t to be the 510-th zero. We then want to know how long it takes to compute<br />

this <strong>in</strong>tegral <strong>in</strong> one.<br />

In[16]:= Tim<strong>in</strong>g[<br />

NIntegrate[f[x],fx,zero[0],zero[510]g,MaxRecursion->10,<br />

Method->GaussKronrod]<br />

]<br />

142


6.3 How to compute the rst <strong>in</strong>tegral<br />

GaussKronrod DoubleExponential Trapezoidal<br />

xm =40<br />

R z510<br />

0 f(x)dx 103.81 20.59 2968.07<br />

P 509<br />

k=0<br />

R z6<br />

0<br />

xm =4<br />

P 5<br />

k=0<br />

R zk+1<br />

z k<br />

f(x)dx 20.05 169.56 3119.32<br />

f(x)dx 0.61 0.82 47.78<br />

R zk+1<br />

z k<br />

f(x)dx 0.38 1.59 36.63<br />

Table 6.1: Comput<strong>in</strong>g times <strong>in</strong> seconds for di e<strong>re</strong>nt quadratu<strong>re</strong> algorithms.<br />

Out[16]= f103.81 Second, 0.0491901g<br />

Remark (Recursion depth) The function Tim<strong>in</strong>g <strong>re</strong>turns the evaluation time as well<br />

as the <strong>re</strong>sult. Note that because the <strong>in</strong>tegration <strong>in</strong>terval spans many zeros of f(x), we<br />

must <strong>in</strong>c<strong>re</strong>ase the maximum number of <strong>re</strong>cursive subdivisions Mathematica tries. The<br />

accord<strong>in</strong>g option is MaxRecursion. Otherwise the evaluation would term<strong>in</strong>ate with an<br />

error message and most likely a wrong <strong>re</strong>sult.<br />

Another possibility istosubdivide the <strong>in</strong>tegration range at the zeros of the <strong>in</strong>tegrand | as<br />

we would do for the <strong>in</strong>terval to the right of the rst partition po<strong>in</strong>t | and to compute the<br />

sum of these partial <strong>in</strong>tegrals. The computation is now much faster because the <strong>in</strong>dividual<br />

<strong>in</strong>tegrals a<strong>re</strong> smooth and we thus bene t from the simplicity of the Gauss-Kronrod formula.<br />

In[17]:= Tim<strong>in</strong>g[<br />

Sum[NIntegrate[f[x],fx,zero[i],zero[i+1]g,<br />

Method->GaussKronrod],fi,0,509g]<br />

]<br />

Out[17]= f20.05 Second, 0.0491901g<br />

We can carry out the same computations for the double exponential quadratu<strong>re</strong> formula and<br />

for a di e<strong>re</strong>nt value of xm = 4, which yields o = 5. The various <strong>re</strong>sults a<strong>re</strong> summarised <strong>in</strong><br />

tab. 6.1 . We <strong>re</strong>cognise rst of all that the trapezoidal rule <strong>in</strong>deed has a poor performance.<br />

The double exponential algorithm seems to be a suitable choice, even if the transformation of<br />

the <strong>in</strong>tegration variable takes some time, which slows down the computation <strong>in</strong> comparison<br />

to the Gauss-Kronrod method if the<strong>re</strong> a<strong>re</strong> only few zeros <strong>in</strong> the <strong>in</strong>tegration <strong>in</strong>terval. The<br />

DE rule, on the other hand, needs no evaluation of any zeros, which can be a signi cant<br />

advantage if the zeros cannot be found explicitly. In such a case, a subdivision strategy us<strong>in</strong>g<br />

Gauss-Kronrod quadratu<strong>re</strong> would be <strong>in</strong>ferior as well. To demonstrate this, we assume that<br />

we have no <strong>in</strong>formation about the zeros of the <strong>in</strong>tegrand <strong>in</strong> (6.8). Instead, we use equidistant<br />

po<strong>in</strong>ts to partition the <strong>in</strong>terval and determ<strong>in</strong>e the total comput<strong>in</strong>g time depend<strong>in</strong>g on the<br />

number of sub<strong>in</strong>tervals (aga<strong>in</strong> with xm = 40).<br />

143


100<br />

90<br />

80<br />

70<br />

60<br />

50<br />

40<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

200 400 600 800 1000 1200 1400<br />

Figu<strong>re</strong> 6.2: Comput<strong>in</strong>g time for a Gauss-Kronrod rule depend<strong>in</strong>g on the number of sub<strong>in</strong>tervals for<br />

the <strong>in</strong>tegrand <strong>in</strong> (6.8) with xm = 40.<br />

In[18]:= timtab = Table[fsteps,First[<br />

Tim<strong>in</strong>g[<br />

m<strong>in</strong> = zero[0];<br />

max = zero[510];<br />

Sum[NIntegrate[f[x],fx,m<strong>in</strong> + i (max-m<strong>in</strong>)/steps,<br />

m<strong>in</strong> + (i+1) (max-m<strong>in</strong>)/stepsg,<br />

Method->GaussKronrod,MaxRecursion->15],<br />

fi,0,steps-1g]<br />

]<br />

]/Secondg,fsteps,1,1401,10g];<br />

Remark (Implementation) To speed up the computation, we evaluate the <strong>in</strong>terval<br />

bounds only once and save them as constants. The list is made up of the number of<br />

<strong>in</strong>tervals and the comput<strong>in</strong>g time for each trial, which is obta<strong>in</strong>ed by tak<strong>in</strong>g the rst<br />

element of the list <strong>re</strong>turned by the function Tim<strong>in</strong>g. To be able to plot this list, we<br />

must rst divide each evaluation time by the constant Second. The actual <strong>re</strong>sults of the<br />

<strong>in</strong>dividual computations a<strong>re</strong> not sto<strong>re</strong>d separately as they a<strong>re</strong> all cor<strong>re</strong>ct.<br />

The <strong>re</strong>sults <strong>in</strong> g. 6.2 show that with only a few subdivisions, the overall comput<strong>in</strong>g time<br />

can be dramatically dec<strong>re</strong>ased. The m<strong>in</strong>imum, however, is still above 30 seconds, which is<br />

signi cantly higher than the values <strong>in</strong> tab. 6.1 . The equidistant partition is also the <strong>re</strong>ason<br />

why it takes about 800 sub<strong>in</strong>tervals to achieve the m<strong>in</strong>imum although the<strong>re</strong> a<strong>re</strong> only 510<br />

zeros <strong>in</strong> the <strong>in</strong>tegration range. The l<strong>in</strong>ear slope beyond the m<strong>in</strong>imum stems from the time-<br />

144


6.4 Asymptotic partition<br />

consum<strong>in</strong>g p<strong>re</strong>paration of the <strong>in</strong>ternal sampl<strong>in</strong>g po<strong>in</strong>ts and weights needed for the Gauss-<br />

Kronrod formula.<br />

The<strong>re</strong> a<strong>re</strong> two conclusions to be drawn from these experiments with <strong>re</strong>spect to the implementation<br />

of numerical quadratu<strong>re</strong> for an automatic algorithm:<br />

The rst <strong>in</strong>tegral between the lower <strong>in</strong>tegration bound and the rst partition po<strong>in</strong>t is<br />

best evaluated with the sett<strong>in</strong>g Method->DoubleExponential, which yields an optimum<br />

performance <strong>in</strong> case of a strongly oscillat<strong>in</strong>g <strong>in</strong>tegrand <strong>in</strong> the <strong>in</strong>terval and is still<br />

<strong>re</strong>asonable <strong>in</strong> all other cases.<br />

The subsequent <strong>in</strong>tegrals that a<strong>re</strong> then used for extrapolation ought to be computed<br />

with Method->GaussKronrod, because he<strong>re</strong> the <strong>in</strong>tegrand is smooth enough that the<br />

simplicity of the Gauss-Kronrod rule will be an advantage.<br />

6.4 Asymptotic partition<br />

The zeros or ext<strong>re</strong>ma of the <strong>in</strong>tegrand a<strong>re</strong> a perfect choice for the de nition of the sub<strong>in</strong>tervals<br />

and the best one as far as the suitability for series acceleration is concerned. In terms of<br />

comput<strong>in</strong>g time, this is true only as long as they a<strong>re</strong> easily computable. If the oscillat<strong>in</strong>g<br />

factor is su ciently complex so that these po<strong>in</strong>ts can only be determ<strong>in</strong>ed numerically, itmay<br />

be mo<strong>re</strong> e cient tolook for a di e<strong>re</strong>nt partition<strong>in</strong>g strategy. A good idea is to consider the<br />

<strong>in</strong>tegrand's behaviour at <strong>in</strong> nity and choose a partition that approaches the zeros as x !1.<br />

This way we make su<strong>re</strong> that the <strong>re</strong>sult<strong>in</strong>g series is at least ultimately alternat<strong>in</strong>g and can be<br />

t<strong>re</strong>ated by a convergence acceleration algorithm. Mo<strong>re</strong> formally, ifwehavean<strong>in</strong>tegral of the<br />

k<strong>in</strong>d<br />

I =<br />

Z 1<br />

a<br />

(x) e j (x) dx ; (6.9)<br />

whe<strong>re</strong> the argument function (x) has an asymptotic expansion as x !1<br />

(x)=x p<br />

with p be<strong>in</strong>g a non-negative <strong>in</strong>teger, we can take its polynomial part<br />

(x) =x p<br />

1X<br />

i=0<br />

pX<br />

i=0<br />

ai<br />

; (6.10)<br />

xi ai<br />

x i<br />

(6.11)<br />

to calculate the subdivision po<strong>in</strong>ts, which maybemuch easier because we need to solve only<br />

a polynomial equation. This strategy has been pursued by Sidi [128].<br />

Aga<strong>in</strong> we p<strong>re</strong>sent an example to make the application of the basic idea clear. The <strong>in</strong>tegral<br />

Z 1<br />

x<br />

I =<br />

x2 p<br />

s<strong>in</strong> x2 , a2 dx = (6.12)<br />

, a2 2<br />

a<br />

145


is most easily obta<strong>in</strong>ed from R 1<br />

0<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

s<strong>in</strong> y<br />

y dy with the substitution y = p x 2 , a 2 . For the purpose<br />

of demonstration, we assume that we cannot calculate the zeros of the <strong>in</strong>tegrand di<strong>re</strong>ctly.<br />

Instead, we know that for x a, p x 2 , a 2 behaves like x, which is exactly the polynomial<br />

part (6.11) of its expansion (6.10). So we compute the partition po<strong>in</strong>ts as if the argument<br />

of the s<strong>in</strong>-function was x, and we plot the sequence of partial sums for the parameter value<br />

a =50 . Ow<strong>in</strong>g to the s<strong>in</strong>gularity of the <strong>in</strong>tegrand at x = a, we determ<strong>in</strong>e the value of the<br />

rst sub<strong>in</strong>terval analytically and <strong>in</strong>clude it explicitly <strong>in</strong> the sequence of the partial sums.<br />

In[19]:= a = 50 Pi;<br />

firstval = Integrate[S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,50 Pi,51 Pig]<br />

Out[19]= S<strong>in</strong>Integral[Sqrt[101] Pi]<br />

In[20]:= sequ = Table[NIntegrate[S<strong>in</strong>[Sqrt[x^2 - a^2]] x/(x^2 - a^2),<br />

fx,i Pi,(i+1) Pig],<br />

fi,51,300g];<br />

partial = FoldList[Plus,N[firstval],sequ];<br />

1.576<br />

1.574<br />

1.572<br />

1.568<br />

1.566<br />

z1 = ListPlot[partial,PlotStyle->Po<strong>in</strong>tSize[0.006]];<br />

50 100 150 200 250<br />

In[21]:= SequenceLimit[partial] - N[Pi/2]<br />

Out[21]= -14<br />

1.59872 10<br />

We emphasise that <strong>in</strong> this case the computation of the ext<strong>re</strong>ma <strong>in</strong>stead of the zeros yields no<br />

better <strong>re</strong>sult s<strong>in</strong>ce the partial <strong>in</strong>tegrals <strong>in</strong> either case cover both negative and positive portions<br />

146


6.4 Asymptotic partition<br />

of the <strong>in</strong>tegrand. The terms `zero' and `ext<strong>re</strong>mum' lose their mean<strong>in</strong>gs as they now <strong>re</strong>fer to<br />

the approximation rather than to the <strong>re</strong>al <strong>in</strong>tegrand. Thus we would actually bene t from<br />

the e ect we discussed <strong>in</strong> the last section only for x !1.<br />

In the computation of the limit we used the complete sequence of partial sums with almost 250<br />

members. Normally, wewould not beg<strong>in</strong> with that large a number because their determ<strong>in</strong>ation<br />

could be time consum<strong>in</strong>g. As the rst few sequence members look fairly ir<strong>re</strong>gular, the burn<strong>in</strong>g<br />

question is how many wemust take <strong>in</strong> order to obta<strong>in</strong> a <strong>re</strong>liable <strong>re</strong>sult for the limit and how<br />

we can improve the accuracy if needed. This obviously <strong>re</strong>qui<strong>re</strong>s that mo<strong>re</strong> sequence members<br />

be taken <strong>in</strong>to account, however, the<strong>re</strong> a<strong>re</strong> two possibilities to do so:<br />

1. We can <strong>in</strong>c<strong>re</strong>ase the number of sequence members used <strong>in</strong> the extrapolation, that is, we<br />

compute limk!1 e S (k)<br />

n with n = 0 if we start with the very rst member.<br />

2. We leave the length k of the subset subject to extrapolation xed and start the extrapolation<br />

at a member with a higher <strong>in</strong>dex, which islimn!1 e S (k)<br />

n .<br />

Both limit<strong>in</strong>g processes have not <strong>re</strong>ceived much attention <strong>in</strong> the literatu<strong>re</strong>. Sidi [126] found<br />

that for his W -transformation, method 1 gives better <strong>re</strong>sults and is mo<strong>re</strong> e cient.<br />

We now exam<strong>in</strong>e these two possibilities brie y for our example. For the rst one, we take<br />

the rst k members of the sequence to determ<strong>in</strong>e the limit and plot the absolute value of the<br />

approximation error e S (k)<br />

n , S depend<strong>in</strong>g on k.<br />

In[22]:= approxerr1 = Table[Abs[SequenceLimit[Take[partial,f1,kg]] - N[Pi/2]],<br />

fk,10,240g];<br />

LogListPlot[approxerr1,PlotStyle->Po<strong>in</strong>tSize[0.006],PlotRange->All];<br />

0.001<br />

-6<br />

1. 10<br />

-9<br />

1. 10<br />

-12<br />

1. 10<br />

0 50 100 150 200<br />

147


6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

For the second version we take only twelve consecutive sequence members to compute the<br />

limit but start at di e<strong>re</strong>nt positions n. Aga<strong>in</strong> we plot the approximation error as a function<br />

of the <strong>in</strong>dex value whe<strong>re</strong> extrapolation starts.<br />

In[23]:= approxerr2 = Table[Abs[SequenceLimit[Take[partial,fn,n+12g]] - N[Pi/2]],<br />

fn,230g];<br />

LogListPlot[approxerr2,PlotStyle->Po<strong>in</strong>tSize[0.006],PlotRange->All];<br />

0.001<br />

-6<br />

1. 10<br />

-9<br />

1. 10<br />

-12<br />

1. 10<br />

0 50 100 150 200<br />

Remark (Logarithmic plots) The function LogListPlot is part of the standard<br />

package Graphics`Graphics` that has to be loaded prior to evaluat<strong>in</strong>g the p<strong>re</strong>vious<br />

commands. The <strong>re</strong>ason why we take exactly twelve members of the sequence to determ<strong>in</strong>e<br />

the limit <strong>in</strong> the second trial is simply that this is the default value of the option<br />

NSumExtraTerms used <strong>in</strong> the function NSum when <strong>in</strong> nite sums a<strong>re</strong> computed by extrapolation.<br />

We notice that the <strong>re</strong>sults of the rst experiment look mo<strong>re</strong> <strong>re</strong>liable for high numbers of<br />

<strong>in</strong>cluded sequence members, whe<strong>re</strong>as the<strong>re</strong> a<strong>re</strong> s<strong>in</strong>gular po<strong>in</strong>ts with surpris<strong>in</strong>gly high errors <strong>in</strong><br />

the second trial even for large n. On the other hand, the second process is signi cantly faster<br />

s<strong>in</strong>ce it uses fewer values for the extrapolation. By comb<strong>in</strong><strong>in</strong>g these two processes one could<br />

nd a <strong>re</strong>asonable compromise between speed and accuracy. Other possibilities for improv<strong>in</strong>g<br />

the <strong>re</strong>sults a<strong>re</strong> discussed <strong>in</strong> the follow<strong>in</strong>g section.<br />

S<strong>in</strong>ce we normally do not know the actual limit of the sequence of partial sums, we would<br />

like toderive some criterion how tochoose the parameters of the extrapolation. We quickly<br />

see that the approximation error (x) , (x) is not very useful for this purpose because<br />

it neglects the periodicity of the s<strong>in</strong>-function. A better approach is to <strong>re</strong>gard the angular<br />

velocities<br />

@ (x)<br />

@x<br />

@ (x)<br />

and @x , <strong>re</strong>spectively. The di e<strong>re</strong>nce between the argument function and its<br />

148


6.5 Considerations for a Mathematica implementation<br />

polynomial approximation causes the sequence of partial sums to take on the form of dist<strong>in</strong>ct<br />

`wave packets' <strong>in</strong> the rhythm of the beat f<strong>re</strong>quency @ ,<br />

@x (x) , (x) . A <strong>re</strong>asonable criterion<br />

could thus be to seek the abscissa value whe<strong>re</strong> such awavepacket comprises a given number<br />

of k elements, that is<br />

0 (x)<br />

0 (x) , 0 (x) = k: (6.13)<br />

The<strong>re</strong> is, unfortunately, no general guidel<strong>in</strong>e how to choose k | the larger it is, the better<br />

the <strong>re</strong>sult will be.<br />

6.5 Considerations for a Mathematica implementation<br />

Let us now put the considerations of the last section <strong>in</strong>to a context that is closer to an<br />

implementation <strong>in</strong> Mathematica. To evaluate an <strong>in</strong> nite sum by means of extrapolation,<br />

we normally use the built-<strong>in</strong> function NSum with the option Method->SequenceLimit. This<br />

function <strong>in</strong>ternally performs exactly the same operations as we did <strong>in</strong> the example above |<br />

rst it computes a sequence of partial sums and then it passes an appropriate portion of this<br />

list to the function SequenceLimit for extrapolation. The selection of this sublist is controlled<br />

by the two parameters NSumTerms and NSumExtraTerms, <strong>re</strong>spectively. The extrapolation<br />

process itself may be <strong>in</strong> uenced by the option WynnDeg<strong>re</strong>e.<br />

NSumTerms is the number n of terms that a<strong>re</strong> summed up explicitly befo<strong>re</strong> the extrapolation<br />

is performed.<br />

NSumExtraTerms is the number k of terms that a<strong>re</strong> used <strong>in</strong> the extrapolation.<br />

WynnDeg<strong>re</strong>e sets someth<strong>in</strong>g like the `deg<strong>re</strong>es of f<strong>re</strong>edom' for the extrapolation [144]. It is<br />

also the only option to SequenceLimit. If its value is 1, Aitken's 2 algorithm is used,<br />

for values g<strong>re</strong>ater than one the -algorithm is <strong>in</strong>voked. Apart from this dist<strong>in</strong>ction, the<br />

actual function of this parameter is not documented further, but it seem<strong>in</strong>gly speci es<br />

how far the -array ( g. 5.1) is computed.<br />

Remark (Sequence extrapolation) Accord<strong>in</strong>g to Keiper [144], the basic algorithm of<br />

SequenceLimit transforms the sequence <strong>in</strong>to a sequence of length n,2. In the triangular<br />

scheme that is often used to illustrate the data dependencies of the -algorithm, this is<br />

equivalent to the computation of the two columns to the right of the start<strong>in</strong>g sequence.<br />

As we have seen <strong>in</strong> section 5.2, the very rst execution of this algorithm yields the same<br />

<strong>re</strong>sult as the 2 algorithm. The basic transformation is <strong>re</strong>peated WynnDeg<strong>re</strong>e times. If<br />

the <strong>re</strong>sult<strong>in</strong>g sequence still has mo<strong>re</strong> than two elements, the whole process starts aga<strong>in</strong><br />

until the outcome is too short or | <strong>in</strong> terms of the -array |until the triangle is completed.<br />

One could suspect that if the -array iscompleted anyhow, the <strong>re</strong>sult ought to<br />

be <strong>in</strong>dependent of the parameter WynnDeg<strong>re</strong>e, but this is not true. The di e<strong>re</strong>nce is that<br />

the column to the left of the start<strong>in</strong>g sequence always consists of zeros, so for each time<br />

the algorithm is <strong>re</strong>started <strong>in</strong> the middle of the -array, the <strong>re</strong>spective column to the left<br />

of the p<strong>re</strong>vious <strong>re</strong>sult is <strong>re</strong>placed with zeros, which naturally a ects the nal outcome.<br />

149


0.001<br />

-6<br />

1. 10<br />

-9<br />

1. 10<br />

-12<br />

1. 10<br />

-15<br />

1. 10<br />

0 50 100 150 200<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Figu<strong>re</strong> 6.3: Extrapolation error e S (k)<br />

n , S for n = 0 and vary<strong>in</strong>g k, computed with the 2 algorithm<br />

( ) and the -algorithm ( ).<br />

The<strong>re</strong> is one exception to the afo<strong>re</strong>mentioned operation pr<strong>in</strong>ciple of the sequence transformation:<br />

If WynnDeg<strong>re</strong>e is set to Inf<strong>in</strong>ity, as many transformations as possible a<strong>re</strong> carried<br />

out <strong>in</strong> a row without <strong>re</strong>start<strong>in</strong>g the process. This is now the case whe<strong>re</strong> the -array is <strong>re</strong>ally<br />

computed from left to right, which matches the descriptions of the -algorithm known<br />

from the literatu<strong>re</strong>.<br />

It is worthwile to <strong>re</strong>view the <strong>re</strong>sults of the p<strong>re</strong>vious example (6.12) with a view to the extrapolation<br />

algorithm used. Be<strong>in</strong>g <strong>in</strong>di e<strong>re</strong>nt to this detail <strong>in</strong> the last chapter, we left the<br />

sett<strong>in</strong>gs of SequenceLimit at their defaults, which selected the 2 algorithm. If we choose<br />

the -algorithm (for example with WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity) and carry out the computation<br />

anew, we can compa<strong>re</strong> both algorithms di<strong>re</strong>ctly. Fig. 6.3 shows this comparison for the case<br />

whe<strong>re</strong> the number of terms subject to extrapolation a<strong>re</strong> varied, and g. 6.4 the alternative<br />

with a xed number of extrapolation terms and a di e<strong>re</strong>nt start<strong>in</strong>g <strong>in</strong>dex. In either case, the<br />

-algorithm yields better <strong>re</strong>sults for a given number of sequence members.<br />

Apart from the algorithm used, which is likely to be chosen once and for all <strong>in</strong> a series of<br />

computations, the<strong>re</strong> a<strong>re</strong> <strong>in</strong> pr<strong>in</strong>ciple two knobs for the ne tun<strong>in</strong>g of the extrapolation <strong>re</strong>sults.<br />

So far, we only twiddled either one or the other, explor<strong>in</strong>g the boundaries of this parameter<br />

space. Yet it might be possible to nd a good compromise for the values of n and k, and<br />

the<strong>re</strong>fo<strong>re</strong> we want to study their e ects <strong>in</strong> some detail now. We aga<strong>in</strong> take the <strong>in</strong>tegral (6.12),<br />

this time with a = 20, and vary both NSumTerms and NSumExtraTerms (he<strong>re</strong> denoted as ns<br />

and ne). We compute this array for di e<strong>re</strong>nt values of WynnDeg<strong>re</strong>e and plot the logarithm to<br />

the base 10 of the approximation error, which is essentially the number of cor<strong>re</strong>ct digits.<br />

150


6.5 Considerations for a Mathematica implementation<br />

0.001<br />

-6<br />

1. 10<br />

-9<br />

1. 10<br />

-12<br />

1. 10<br />

-15<br />

1. 10<br />

0 50 100 150 200<br />

Figu<strong>re</strong> 6.4: Extrapolation error e S (k)<br />

n , S for k = 12 and vary<strong>in</strong>g n, computed with the 2 algorithm<br />

( ) and the -algorithm ( ).<br />

In[24]:= a = 20 Pi;<br />

firstval = N[Integrate[S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,20 Pi,21 Pig],40];<br />

seq = Table[NIntegrate[S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,i Pi,(i+1) Pig],<br />

fi,21,350g];<br />

partial = FoldList[Plus,firstval,seq];<br />

errtab = Table[fne,ns,Log[10,Abs[SetP<strong>re</strong>cision[<br />

SequenceLimit[Take[partial,fns+1,ns+ne+1g],<br />

WynnDeg<strong>re</strong>e->1],40]-N[Pi/2,40]]]g,<br />

fne,6,50,1g,fns,0,60,1g];<br />

The <strong>re</strong>sults for WynnDeg<strong>re</strong>e->1 a<strong>re</strong> shown <strong>in</strong> g. 6.5 . It is <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g to notice that if only a<br />

small number of terms a<strong>re</strong> used for extrapolation, peaks appear whe<strong>re</strong> the extrapolation error<br />

becomes unexpectedly high. These peaks seem to be <strong>re</strong>lated to the a<strong>re</strong>as of the sequence<br />

whe<strong>re</strong> the alternation is not very pronounced | the `nodes' of its wave-like envelope. A<br />

contour plot of the <strong>re</strong>sults ( g. 6.6) <strong>re</strong>veals that these peaks follow the l<strong>in</strong>es 2 ne + ns = const.<br />

This is not surpris<strong>in</strong>g if we <strong>re</strong>member that each pass of the sequence transformation <strong>re</strong>duces<br />

the length of the sequence by two. In the end the<strong>re</strong> may be either one or two members left<br />

depend<strong>in</strong>g on whether we started with an odd (2m +1) or even (2m +2) number of sequence<br />

members. Both <strong>re</strong>sults | after exactly the same number m of transformations | a<strong>re</strong> not<br />

supposed to di er very much. If su cient terms a<strong>re</strong> used for the extrapolation or if it is<br />

started whe<strong>re</strong> the wave packets a<strong>re</strong> broad enough (for low ne but high ns), then we obta<strong>in</strong><br />

151


6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

the best <strong>re</strong>sults that a<strong>re</strong> <strong>in</strong> uenced chie y by the implementation details and the associated<br />

numerical e ects. This <strong>re</strong>gion is characterised by extrapolation errors that a<strong>re</strong> constant for<br />

ne + ns = const and thus depend on the total number of sequence terms <strong>re</strong>gardless of how<br />

many a<strong>re</strong> actually used for extrapolation.<br />

The contour plot ( g. 6.6) also shows that it is di cult to decide whe<strong>re</strong> to beg<strong>in</strong> the extrapolation<br />

if not too many terms a<strong>re</strong> to be used. For 40 terms, for example, the optimum lies<br />

somewhe<strong>re</strong> at ne = 15, ns = 25.<br />

The <strong>re</strong>sults we obta<strong>in</strong> when choos<strong>in</strong>g the -algorithm with WynnDeg<strong>re</strong>e->2 a<strong>re</strong> shown <strong>in</strong><br />

g. 6.7. The <strong>in</strong>convenient peaks have disappea<strong>re</strong>d (which is the same for all even values of<br />

WynnDeg<strong>re</strong>e, but not for the odd ones), yet it is hard to nd optimum values for ne and ns.<br />

The sett<strong>in</strong>g WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity yields the best extrapolation <strong>re</strong>sults ( g. 6.8). It is still<br />

true that the<strong>re</strong> a<strong>re</strong> pairs of sequences that give the same estimates for the limit S e(2m+1) n and<br />

eS (2m+2)<br />

n+1 , <strong>re</strong>spectively, which can be seen from the bands runn<strong>in</strong>g <strong>in</strong> parallel to the ns-axis.<br />

The <strong>re</strong>spective contour plot (6.9) shows that the best way to improve the extrapolation <strong>re</strong>sult<br />

is to <strong>in</strong>c<strong>re</strong>ase ne, perhaps with a few terms summed up explicitly. We have thus con rmed<br />

the suspicion that the limit<strong>in</strong>g process 1 from the last section is <strong>in</strong>deed superior.<br />

For comparison we compute the extrapolation error if we do not use approximate partition,<br />

but take the zeros of the <strong>in</strong>tegrand as partition po<strong>in</strong>ts. Just the determ<strong>in</strong>ation of the partial<br />

<strong>in</strong>tegrals is di e<strong>re</strong>nt. The <strong>re</strong>st of the evaluation <strong>re</strong>ma<strong>in</strong>s the same as with the asymptotic<br />

partition befo<strong>re</strong>.<br />

In[25]:= a = 20 Pi;<br />

firstval = N[Integrate[S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,a,Sqrt[Pi^2+a^2]g],40];<br />

seq = Table[NIntegrate[S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,Sqrt[(i Pi)^2+a^2],Sqrt[((i+1) Pi)^2+a^2]g],<br />

fi,1,300g];<br />

We nd that the <strong>re</strong>sults a<strong>re</strong> far better and practically <strong>in</strong>dependent of the value we choose for<br />

WynnDeg<strong>re</strong>e.<br />

We thus conclude that when implement<strong>in</strong>g an automatic quadratu<strong>re</strong> rout<strong>in</strong>e with Mathematica,<br />

we ought to consider the follow<strong>in</strong>g po<strong>in</strong>ts:<br />

The sett<strong>in</strong>g WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity gives the best <strong>re</strong>sults for asymptotic partition and<br />

<strong>re</strong>asonably good ones for standard (zero) partition, so it is a suitable choice for a default<br />

value of this option.<br />

To <strong>in</strong>c<strong>re</strong>ase the accuracy of the extrapolation <strong>re</strong>sult, the number of terms <strong>in</strong>cluded <strong>in</strong><br />

the extrapolation (NSumExtraTerms) should be <strong>in</strong>c<strong>re</strong>ased, whe<strong>re</strong>as NSumTerms may be<br />

left unchanged.<br />

Whenever feasible, the zeros of the <strong>in</strong>tegrand should be chosen as the partition po<strong>in</strong>ts.<br />

This yields mo<strong>re</strong> accurate <strong>re</strong>sults with fewer terms <strong>in</strong>volved.<br />

152


6.5 Considerations for a Mathematica implementation<br />

0<br />

-5<br />

-10<br />

-15<br />

0<br />

20<br />

ne<br />

40<br />

Figu<strong>re</strong> 6.5: Negative number of cor<strong>re</strong>ct digits, log e S (k)<br />

n , S , for vary<strong>in</strong>g n (ns) and k (ne), computed<br />

with the 2 algorithm (WynnDeg<strong>re</strong>e->1).<br />

60<br />

50<br />

40<br />

30<br />

20<br />

10<br />

0<br />

0<br />

20<br />

10 20 30 40 50 60<br />

Figu<strong>re</strong> 6.6: Contour plot of g. 6.5, abscissa: ne, ord<strong>in</strong>ate: ns.<br />

153<br />

40<br />

ns<br />

60


0<br />

-5<br />

-10<br />

-15<br />

0<br />

20<br />

ne<br />

40<br />

0<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Figu<strong>re</strong> 6.7: Negative number of cor<strong>re</strong>ct digits, log e S (k)<br />

n , S , for vary<strong>in</strong>g n (ns) and k (ne), computed<br />

with the -algorithm (WynnDeg<strong>re</strong>e->2).<br />

-5<br />

-10<br />

-15<br />

10<br />

20<br />

30<br />

ne<br />

40<br />

50<br />

Figu<strong>re</strong> 6.8: Negative number of cor<strong>re</strong>ct digits, log e S (k)<br />

n , S , for vary<strong>in</strong>g n (ns) and k (ne), computed<br />

with the -algorithm (WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity).<br />

154<br />

0<br />

20<br />

20<br />

40<br />

ns<br />

40<br />

ns<br />

60<br />

60


6.6 Controll<strong>in</strong>g the accuracy of the extrapolation <strong>re</strong>sult<br />

60<br />

50<br />

40<br />

30<br />

20<br />

10<br />

0<br />

10 20 30 40 50 60<br />

Figu<strong>re</strong> 6.9: Contour plot of g. 6.8, abscissa: ne, ord<strong>in</strong>ate: ns.<br />

6.6 Controll<strong>in</strong>g the accuracy of the extrapolation <strong>re</strong>sult<br />

As we have seen <strong>in</strong> the last sections, we can improve the <strong>re</strong>sult of the extrapolation under<br />

most circumstances by <strong>in</strong>c<strong>re</strong>as<strong>in</strong>g the number of sequence elements be<strong>in</strong>g subject to the<br />

extrapolation. The seve<strong>re</strong> problem that still <strong>re</strong>ma<strong>in</strong>s is that this is only a qualitative nd<strong>in</strong>g<br />

and that we have no guidel<strong>in</strong>e whatsoever as to how this parameter is to be chosen. From a<br />

user's po<strong>in</strong>t of view, the exact number of sequence members necessary to achieve a certa<strong>in</strong><br />

accuray is enti<strong>re</strong>ly ir<strong>re</strong>levant, provided that the goal is <strong>re</strong>ached. On the contrary, it is much<br />

mo<strong>re</strong> convenient to let the user specify only the desi<strong>re</strong>d accuracy and to leave the choice of<br />

the other parameters to the algorithm, which then has to be somehow adaptive.<br />

If we want to adapt only the parameters of a basic numerical algorithm, then we <strong>in</strong> turn need<br />

a meta algorithm that starts from a given parameter set, calls the basic algorithm, judges the<br />

<strong>re</strong>sults and changes the parameters. This procedu<strong>re</strong> must be <strong>re</strong>peated iteratively until either<br />

the outcome meets the <strong>re</strong>qui<strong>re</strong>ments or a maximum number of iterations is <strong>re</strong>ached | just<br />

to p<strong>re</strong>vent the algorithm from be<strong>in</strong>g lost <strong>in</strong> an <strong>in</strong> nite loop or wast<strong>in</strong>g p<strong>re</strong>cious comput<strong>in</strong>g<br />

time. From the p<strong>re</strong>vious sections, we can easily conclude what the th<strong>re</strong>e major elements of<br />

such a meta algorithm must look like:<br />

1. The basic algorithm consists of the computation of the sequence of partial sums for a<br />

given <strong>in</strong>tegrand together with the extrapolation. It <strong>re</strong>turns an estimate of the <strong>in</strong>tegral.<br />

2. The parameters that can be <strong>in</strong> uenced di<strong>re</strong>ctly by the meta algorithm a<strong>re</strong> the number<br />

155


-5<br />

-7.5<br />

-10<br />

-12.5<br />

10<br />

20<br />

30<br />

ne<br />

40<br />

50<br />

0<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Figu<strong>re</strong> 6.10: Negative number of cor<strong>re</strong>ct digits, log e S (k)<br />

n , S , for vary<strong>in</strong>g n (ns) and k (ne), computed<br />

with the 2 algorithm (WynnDeg<strong>re</strong>e->1) and the zeros of the <strong>in</strong>tegrand as partition po<strong>in</strong>ts.<br />

of sequence members used <strong>in</strong> the extrapolation as well as the p<strong>re</strong>cision of the <strong>in</strong>ternal<br />

computations. We have al<strong>re</strong>ady noticed that it is most promis<strong>in</strong>g to use the enti<strong>re</strong><br />

sequence for this purpose rather than to shift the start<strong>in</strong>g po<strong>in</strong>t for the extrapolation<br />

to higher <strong>in</strong>dex values.<br />

3. The parameter to be controlled is the accuracy of the <strong>re</strong>sult or | <strong>in</strong> other words<br />

| its error. We thus need a function that gives us an a posteriori estimate of the<br />

approximation error.<br />

A simple way to obta<strong>in</strong> an error estimate <strong>in</strong> Mathematica is to use the functions Accuracy<br />

and P<strong>re</strong>cision on the <strong>re</strong>sult of a numerical computation. The former <strong>re</strong>turns the number of<br />

signi cant digits to the right of the decimal po<strong>in</strong>t and the<strong>re</strong>fo<strong>re</strong> is a measu<strong>re</strong> for the absolute<br />

error. The latter gives the total number of signi cant digits <strong>in</strong> a oat<strong>in</strong>g-po<strong>in</strong>t number. For<br />

very small numbers, the zeros between the decimal po<strong>in</strong>t and the rst non-zero digit a<strong>re</strong> not<br />

counted, thus the p<strong>re</strong>cision of a number denotes its <strong>re</strong>lative error (see also the explanations<br />

<strong>in</strong> example 6.3.1).<br />

Unfortunately, these two functions work only with variable-p<strong>re</strong>cision arithmetic, which cannot<br />

make use of the fast hardwa<strong>re</strong>-based mach<strong>in</strong>e-p<strong>re</strong>cision operations. Instead, it is implemented<br />

<strong>in</strong> softwa<strong>re</strong> to keep track of the uncerta<strong>in</strong>ty associated with every number <strong>in</strong> the calculation,<br />

which makes it much slower than mach<strong>in</strong>e-p<strong>re</strong>cision arithmetic. For this <strong>re</strong>ason, when Mathematica<br />

comes across a mach<strong>in</strong>e number <strong>in</strong> the course of the computation, all subsequent<br />

156<br />

20<br />

40<br />

ns<br />

60


6.6 Controll<strong>in</strong>g the accuracy of the extrapolation <strong>re</strong>sult<br />

operations a<strong>re</strong> carried out <strong>in</strong> mach<strong>in</strong>e-p<strong>re</strong>cision, s<strong>in</strong>ce it makes no sense to ma<strong>in</strong>ta<strong>in</strong> a high<br />

p<strong>re</strong>cision if one of the operands <strong>in</strong>volves an error of unknown magnitude [144]. The<strong>re</strong>fo<strong>re</strong>, the<br />

user must be ext<strong>re</strong>mely ca<strong>re</strong>ful not to <strong>in</strong>advertently <strong>in</strong>troduce a mach<strong>in</strong>e number somewhe<strong>re</strong><br />

<strong>in</strong> the calculation.<br />

After this <strong>in</strong>troduction, we exam<strong>in</strong>e an example similar to (6.12). The only di e<strong>re</strong>nce is a<br />

constant factor that makes f(x) 1 to demonstrate the di e<strong>re</strong>nce between p<strong>re</strong>cision and<br />

accuracy of the extrapolation <strong>re</strong>sults.<br />

Example 6.6.1 We consider<br />

I =<br />

Z 1<br />

a<br />

10 10<br />

x<br />

x2 p<br />

s<strong>in</strong> x2 , a2 dx = 10<br />

, a2 2 10<br />

with a = 50. The code needed to generate the sequence of partial sums is slightly di e<strong>re</strong>nt now.<br />

In[26]:= a = 50 Pi;<br />

firstval = N[Integrate[10^10 S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,50 Pi,51 Pig],40];<br />

seq = Table[NIntegrate[10^10 S<strong>in</strong>[Sqrt[x^2 - a^2]]/(x^2 - a^2) x,<br />

fx,i Pi,(i+1) Pig,Work<strong>in</strong>gP<strong>re</strong>cision->27],<br />

fi,51,300g];<br />

partial = FoldList[Plus,firstval,seq];<br />

trueval = N[10^10 Pi/2,40];<br />

(6.14)<br />

For the numerical quadratu<strong>re</strong> rout<strong>in</strong>e NIntegrate, we set Work<strong>in</strong>gP<strong>re</strong>cision->27. S<strong>in</strong>ce the default<br />

sett<strong>in</strong>g for P<strong>re</strong>cisionGoal is ten digits less than Work<strong>in</strong>gP<strong>re</strong>cision,we can expect the series members<br />

to be accurate to about 17 digits or perhaps even mo<strong>re</strong> (<strong>in</strong> fact, Accuracy[seq] <strong>re</strong>turns the worst<br />

case value 19). All exact exp<strong>re</strong>ssions a<strong>re</strong> converted to arbitrary-p<strong>re</strong>cision numbers with the function<br />

N[x,p<strong>re</strong>c], <strong>in</strong> contrast to the ord<strong>in</strong>ary N[x], which yields mach<strong>in</strong>e-p<strong>re</strong>cision numbers.<br />

To compa<strong>re</strong> the extrapolation <strong>re</strong>sults di<strong>re</strong>ctly with the <strong>re</strong>spective accuracy estimated by Mathematica,<br />

we calculate the logarithm of the actual extrapolation error and <strong>in</strong>vert it such that we obta<strong>in</strong> essentially<br />

the number of cor<strong>re</strong>ct decimal digits for each trial. Cor<strong>re</strong>spond<strong>in</strong>g to the <strong>re</strong>sults of the last chapter,<br />

we set WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity. The follow<strong>in</strong>g statements compute the tables for the enti<strong>re</strong> sequence<br />

be<strong>in</strong>g subject to extrapolation. The <strong>re</strong>sults a<strong>re</strong> shown <strong>in</strong> g. 6.11.<br />

In[27]:= err = Table[fne,-Log[10,Abs[SequenceLimit[Take[partial,ne],<br />

WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity]-trueval]]g,<br />

fne,6,Length[partial]g];<br />

acc = Table[fne,Accuracy[SequenceLimit[Take[partial,ne],<br />

WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity]]g,<br />

fne,6,Length[partial]g];<br />

Example 6.6.2 For the above <strong>in</strong>tegral (6.14) we now compute the <strong>re</strong>lative approximation errors<br />

and the <strong>re</strong>spective estimates obta<strong>in</strong>ed by the function P<strong>re</strong>cision. Fig. 6.12 shows the <strong>re</strong>sults.<br />

157


7.5<br />

5<br />

2.5<br />

-2.5<br />

-5<br />

-7.5<br />

Figu<strong>re</strong> 6.11: Absolute extrapolation error , log e S (k)<br />

0<br />

of Accuracy[x] ( ).<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

50 100 150 200 250<br />

, S for vary<strong>in</strong>g k ( ) and the <strong>re</strong>spective <strong>re</strong>sults<br />

In[28]:= err = Table[fne,-Log[10,Abs[SequenceLimit[Take[partial,ne],<br />

WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity]/trueval - 1]]g,<br />

fne,6,Length[partial]g];<br />

acc = Table[fne,P<strong>re</strong>cision[SequenceLimit[Take[partial,ne],<br />

WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity]]g,<br />

fne,6,Length[partial]g];<br />

The examples make su ciently clear that the absolute error is not adequate as the only<br />

criterion for term<strong>in</strong>at<strong>in</strong>g the iteration. When the true value of the <strong>in</strong>tegral is very large, the<br />

absolute error can safely be large as well, even if it a ects digits to the left of the decimal<br />

po<strong>in</strong>t. On the other hand, for very small <strong>in</strong>tegrals, the absolute error must be very small,<br />

too, <strong>in</strong> order to obta<strong>in</strong> <strong>re</strong>asonable <strong>re</strong>sults. Consequently, the selection of the tolerable error<br />

limit <strong>re</strong>qui<strong>re</strong>s knowledge of the a priori unknown value of the <strong>in</strong>tegral, which is impractical<br />

for an automatic quadratu<strong>re</strong> rout<strong>in</strong>e that is supposed to be user-friendly.<br />

A better solution is to use the <strong>re</strong>lative error estimate as criterion for error control. In the<br />

example, we could have achieved a p<strong>re</strong>cision of 12 (i. e. a <strong>re</strong>lative error of 10 ,12 ) ir<strong>re</strong>spective<br />

of the constant factor of the <strong>in</strong>tegrand. The only problem that might emerge is that of an<br />

<strong>in</strong>tegral very close to zero, <strong>in</strong> which case the desi<strong>re</strong>d <strong>re</strong>lative error could not be <strong>re</strong>ached for<br />

numerical <strong>re</strong>asons. In this case, the speci cation of an absolute error could be a mo<strong>re</strong> feasible<br />

alternative.<br />

So far, we always assumed that the parameter to be changed by the meta algorithm is the<br />

158


6.6 Controll<strong>in</strong>g the accuracy of the extrapolation <strong>re</strong>sult<br />

17.5<br />

15<br />

12.5<br />

10<br />

7.5<br />

2.5<br />

Figu<strong>re</strong> 6.12: Relative extrapolation error , log e S (k)<br />

0 ,S<br />

S<br />

P<strong>re</strong>cision[x] ( ).<br />

50 100 150 200 250<br />

for vary<strong>in</strong>g k ( ) and the <strong>re</strong>spective <strong>re</strong>sults of<br />

number of sequence members <strong>in</strong>cluded <strong>in</strong> the extrapolation. One might argue that alter<strong>in</strong>g the<br />

sett<strong>in</strong>g for Work<strong>in</strong>gP<strong>re</strong>cision should have a similar e ect. The follow<strong>in</strong>g example explo<strong>re</strong>s<br />

this possibility.<br />

Example 6.6.3 The <strong>in</strong>tegral (6.12)<br />

I =<br />

Z 1<br />

a<br />

x<br />

x2 p<br />

s<strong>in</strong> x2 , a2dx =<br />

, a2 2<br />

shall be evaluated for a = 50 with di e<strong>re</strong>nt values for Work<strong>in</strong>gP<strong>re</strong>cision and with WynnDeg<strong>re</strong>e-><br />

Inf<strong>in</strong>ity. As g. 6.13 shows, a high p<strong>re</strong>cision of the <strong>in</strong>ternal calculations naturally allows for a higher<br />

p<strong>re</strong>cision of the <strong>re</strong>sult, but this is only true if enough sequence members a<strong>re</strong> taken <strong>in</strong>to account. Befo<strong>re</strong><br />

the <strong>re</strong>spective saturation is <strong>re</strong>ached, the sett<strong>in</strong>g of the <strong>in</strong>ternal p<strong>re</strong>cision has almost no e ect on the<br />

p<strong>re</strong>cision of the <strong>re</strong>sult. But the<strong>re</strong> is an even mo<strong>re</strong> important aspect to this parameter: If the meta<br />

algorithm was to change the Work<strong>in</strong>gP<strong>re</strong>cision, the enti<strong>re</strong> sequence of partial sums would have tobe<br />

<strong>re</strong>-computed for each iteration, which isvery time-consum<strong>in</strong>g. If, on the other hand, the number of<br />

sequence elements t<strong>re</strong>ated with SequenceLimit is alte<strong>re</strong>d, then only a few new members need to be<br />

added <strong>in</strong> each cycle. For these <strong>re</strong>asons, the value of the <strong>in</strong>ternal p<strong>re</strong>cision <strong>re</strong>ma<strong>in</strong>s xed throughout<br />

the computation. However, it must be set such that the <strong>re</strong>qui<strong>re</strong>d p<strong>re</strong>cision can be <strong>re</strong>ached at all.<br />

159


25<br />

20<br />

15<br />

10<br />

5<br />

6Towards a quadratu<strong>re</strong> rout<strong>in</strong>e<br />

20 40 60 80 100 120<br />

Figu<strong>re</strong> 6.13: Absolute extrapolation error , log e S (k)<br />

0<br />

, S depend<strong>in</strong>g on the number of sequence mem-<br />

bers be<strong>in</strong>g subject to extrapolation for Work<strong>in</strong>gP<strong>re</strong>cision->20 ( ), 30 ( ), and 40 ( ).<br />

The conclusions of this section concern<strong>in</strong>g the p<strong>re</strong>cision control of an automatic quadratu<strong>re</strong><br />

rout<strong>in</strong>e a<strong>re</strong> as follows:<br />

The parameter that is controlled iteratively by the meta algorithm is the length of the<br />

sequence of partial sums.<br />

The criterion for the term<strong>in</strong>ation of the iteration loop is the achievement of either a<br />

given absolute or <strong>re</strong>lative error. This is the same concept that Mathematica uses for<br />

adaptive numerical quadratu<strong>re</strong>.<br />

The default criterion must be that of the <strong>re</strong>lative error. A speci cation of a m<strong>in</strong>imum<br />

absolute error must <strong>in</strong> any case be based upon a speci c knowledge of the quadratu<strong>re</strong><br />

problem.<br />

The work<strong>in</strong>g p<strong>re</strong>cision must be high enough that the error <strong>re</strong>qui<strong>re</strong>ments can be met.<br />

160


Chapter 7<br />

Mathematica implementation of a<br />

quadratu<strong>re</strong> function<br />

Les mathematiciens n'etudient pas des objets, mais des <strong>re</strong>lations ent<strong>re</strong> les<br />

objets; il leur est donc <strong>in</strong>di e<strong>re</strong>nt de <strong>re</strong>mplacer ces objets par d'aut<strong>re</strong>s,<br />

pourvu que les <strong>re</strong>lations ne changent pas. La matie<strong>re</strong> ne leur importe pas,<br />

la forme seule les <strong>in</strong>te<strong>re</strong>sse. Henry Po<strong>in</strong>ca<strong>re</strong>, quoted <strong>in</strong> [1]<br />

Based on the nd<strong>in</strong>gs of the p<strong>re</strong>vious sections, this chapter describes how an automatic<br />

quadratu<strong>re</strong> rout<strong>in</strong>e for an extrapolation strategy can be implemented <strong>in</strong> Mathematica. We<br />

beg<strong>in</strong> the discussion from a user's po<strong>in</strong>t of view with the options that control the operation of<br />

the function. The <strong>in</strong>ternal structu<strong>re</strong> of the function is another topic of <strong>in</strong>te<strong>re</strong>st, and nally we<br />

shall po<strong>in</strong>t out some implementation details. The enti<strong>re</strong> source code of the package is listed<br />

<strong>in</strong> the appendix. In the last section of the chapter we shall apply the function to examples<br />

known from the literatu<strong>re</strong> to verify the cor<strong>re</strong>ct operation.<br />

7.1 User <strong>in</strong>terface of the function OscInt<br />

The top level function of the package has the same name as the whole package, OscInt. Its<br />

basic syntax is<br />

OscInt[f<strong>in</strong>t, fzero, a, opts],<br />

and its purpose is to nd the numerical value of the <strong>in</strong>tegral<br />

I =<br />

Z 1<br />

a<br />

(x) u( (x)) dx ; (7.1)<br />

whe<strong>re</strong> u( ) is e j , e ,j or any l<strong>in</strong>ear comb<strong>in</strong>ation of both, <strong>in</strong> particular s<strong>in</strong>( ) and cos( ). The<br />

<strong>in</strong>dividual <strong>in</strong>put parameters of OscInt have the follow<strong>in</strong>g mean<strong>in</strong>gs:<br />

161


7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

f<strong>in</strong>t is the <strong>in</strong>tegrand function (x) u( (x)), which can be de ned <strong>in</strong> two ways. One is<br />

s<strong>in</strong> x<br />

to use a so-called pu<strong>re</strong> function, e. g. S<strong>in</strong>[#]/#& for x (see also the open<strong>in</strong>g <strong>re</strong>mark<br />

<strong>in</strong> section 7.5). If the <strong>in</strong>tegrand is decla<strong>re</strong>d explicitly, then it must be of the form f[x]<br />

or, if the<strong>re</strong> a<strong>re</strong> several other parameters, f[a,b,c][x]. In this case, only the name of<br />

the function together with the additional parameters (f[a,b,c]) has to be passed to<br />

OscInt.<br />

fzero is the function be<strong>in</strong>g used to determ<strong>in</strong>e the partition po<strong>in</strong>ts fx0;x1;x2;:::g. Depend<strong>in</strong>g<br />

on the option FunctionType it is understood either as an explicit enumeration<br />

of the partition po<strong>in</strong>ts x (n); n 0 or as the argument function (x) (or an approximation<br />

(x)) of the oscillat<strong>in</strong>g factor of the <strong>in</strong>tegrand, which is then used to compute<br />

the subdivision po<strong>in</strong>ts xn numerically. In the latter case, the function must be di e<strong>re</strong>ntiable.<br />

In the former case, it is the user's <strong>re</strong>sponsibility to ensu<strong>re</strong> that fzero <strong>re</strong>turns<br />

po<strong>in</strong>ts <strong>in</strong> ascend<strong>in</strong>g order that lie exclusively to the right of the rightmost ext<strong>re</strong>mum or<br />

<strong>in</strong> exion po<strong>in</strong>t of the argument function. It is, however, not necessary to see to it that<br />

x (0) > a, as the rout<strong>in</strong>e will seek the rst x (k) > a itself. Like f<strong>in</strong>t, the function<br />

fzero can be given either as a pu<strong>re</strong> function or <strong>in</strong> explicit notation.<br />

a is the lower <strong>in</strong>tegration limit.<br />

opts a<strong>re</strong> options the user can set to take <strong>in</strong> uence on the computation. They a<strong>re</strong><br />

expla<strong>in</strong>ed below.<br />

Normally, the partition po<strong>in</strong>ts a<strong>re</strong> chosen such that they lie <strong>in</strong> a <strong>re</strong>gion whe<strong>re</strong> (x) and its<br />

derivative 0 (x) a<strong>re</strong> monotonic, <strong>in</strong> order to avoid the problems discussed <strong>in</strong> section 6.2. If an<br />

enumeration function is passed to OscInt, this function must be written accord<strong>in</strong>gly. If the<br />

orig<strong>in</strong>al (x) is used as fzero, then z0 is selected to be larger than the largest <strong>re</strong>al part of the<br />

solutions of 0 (x) =0. In some cases, however, it might be mo<strong>re</strong> convenient tochoose the rst<br />

subdivision po<strong>in</strong>t accord<strong>in</strong>g to a di e<strong>re</strong>nt criterion | particularly if we use an approximation<br />

(x). Thus the<strong>re</strong> is a second way to <strong>in</strong>voke OscInt with an explicit speci cation of a lower<br />

subdivision limit a0:<br />

OscInt[f<strong>in</strong>t, fzero, fa, a0g, opts].<br />

Apart from the new parameter, the mean<strong>in</strong>gs of the other <strong>in</strong>put values <strong>re</strong>ma<strong>in</strong> unchanged.<br />

The partition po<strong>in</strong>ts a<strong>re</strong> now chosen such that x (k) >a0; k K(a0), or x0 >a0, depend<strong>in</strong>g<br />

on the <strong>in</strong>terp<strong>re</strong>tation of fzero.<br />

The<strong>re</strong> is a variety of options <strong>in</strong>dispensable to control the evaluation of OscInt. Some of them<br />

a<strong>re</strong> standard Mathematica options that a<strong>re</strong> used also for other numerical functions, others<br />

have been newly <strong>in</strong>troduced.<br />

WynnDeg<strong>re</strong>e is the standard option for the computation of the limit of the sequence of<br />

partial sums. Its default value is Inf<strong>in</strong>ity.<br />

162


7.1 User <strong>in</strong>terface of the function OscInt<br />

NSumTerms is taken from NSum although this function is not used with<strong>in</strong> OscInt. Its<br />

mean<strong>in</strong>g, however, is the same | it speci es how many elements of the sequence of<br />

partial sums a<strong>re</strong> not <strong>in</strong>cluded <strong>in</strong> the extrapolation. The default value is 10.<br />

NSumExtraTerms is used like <strong>in</strong> NSum to p<strong>re</strong>scribe the number of sequence members<br />

subject to extrapolation (default value 12). Together with NSumTerms, it de nes how<br />

many elements of the sequence of partial sums have to be computed <strong>in</strong> total.<br />

M<strong>in</strong>Recursion is a standard option used for all numerical <strong>in</strong>tegrations that speci es<br />

how many levels of <strong>re</strong>cursion a<strong>re</strong> at least <strong>re</strong>qui<strong>re</strong>d befo<strong>re</strong> tests for convergence start. Its<br />

default value is 0.<br />

MaxRecursion sets the maximum depth for <strong>re</strong>cursive algorithms and a ects both the<br />

numerical quadratu<strong>re</strong> with NIntegrate and the numerical solv<strong>in</strong>g of equations with<br />

F<strong>in</strong>dRoot. For the latter, its value is passed to the option MaxIterations. This<br />

change of name was necessary for <strong>re</strong>asons of consistency, because OscInt uses the option<br />

MaxIterations <strong>in</strong> a di e<strong>re</strong>nt way. The default value is MaxRecursion->20.<br />

AccuracyGoal is used to specify the absolute error limit for NIntegrate and defaults<br />

to Inf<strong>in</strong>ity (i. e. it is not used as a criterion for the quadratu<strong>re</strong>).<br />

P<strong>re</strong>cisionGoal sets the <strong>re</strong>lative error limit for the numerical quadratu<strong>re</strong> and like <strong>in</strong><br />

NIntegrate has the default value Automatic, which means that it is set to ten digits<br />

less than Work<strong>in</strong>gP<strong>re</strong>cision.<br />

PartitionAccuracy is <strong>in</strong> fact the sett<strong>in</strong>g for AccuracyGoal used <strong>in</strong> F<strong>in</strong>dRoot to determ<strong>in</strong>e<br />

the partition po<strong>in</strong>ts. With the change of name this parameter can be set <strong>in</strong>dependently<br />

of the AccuracyGoal of NIntegrate. The default value is PartitionAccuracy<br />

->Automatic.<br />

AccuracyControl sets the upper limit for the absolute error <strong>in</strong> the extrapolation <strong>re</strong>sult<br />

<strong>in</strong> that is speci es the digits of accuracy <strong>re</strong>qui<strong>re</strong>d. Its default value is Inf<strong>in</strong>ity, which<br />

means that it is not used as a criterion for term<strong>in</strong>at<strong>in</strong>g the iteration. If it is used, ca<strong>re</strong><br />

must be taken to set the work<strong>in</strong>g p<strong>re</strong>cision high enough that the desi<strong>re</strong>d accuracy can<br />

be <strong>re</strong>ached.<br />

P<strong>re</strong>cisionControl gives the upper limit for the <strong>re</strong>lative error <strong>in</strong> the extrapolation <strong>re</strong>sult<br />

and is <strong>re</strong>commended for controll<strong>in</strong>g the iteration. With its default value Inf<strong>in</strong>ity, itis<br />

normally not used. If p<strong>re</strong>cision or accuracy control is enabled, the iterative procedu<strong>re</strong><br />

cont<strong>in</strong>ues to <strong>in</strong>c<strong>re</strong>ase the length of the sequence of partial sums until either the absolute<br />

or <strong>re</strong>lative error goal is achieved or the iteration count <strong>re</strong>aches MaxIterations.<br />

MaxIterations speci es how many improvements of the extrapolation a<strong>re</strong> tried befo<strong>re</strong><br />

OscInt term<strong>in</strong>ates with an error message. The default value is 6.<br />

IterationLength denotes how many elements a<strong>re</strong> added to the sequence of partial sums<br />

with each iteration. It defaults to 4.<br />

163


7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

Work<strong>in</strong>gP<strong>re</strong>cision is the standard option to set the p<strong>re</strong>cision of the <strong>in</strong>ternal calculations,<br />

with $Mach<strong>in</strong>eP<strong>re</strong>cision as default value. Normally it may be set by the user<br />

to any arbitrary value (between the bounds $M<strong>in</strong>P<strong>re</strong>cision and $MaxP<strong>re</strong>cision, of<br />

course). If accuracy or p<strong>re</strong>cision control a<strong>re</strong> used, then its t<strong>re</strong>atment is slightly diffe<strong>re</strong>nt,<br />

because these functions need high-p<strong>re</strong>cision arithmetic which <strong>in</strong> turn <strong>re</strong>qui<strong>re</strong>s a<br />

work<strong>in</strong>g p<strong>re</strong>cision that is higher than the mach<strong>in</strong>e p<strong>re</strong>cision. Thus Work<strong>in</strong>gP<strong>re</strong>cision<br />

is set to the maximum of $Mach<strong>in</strong>eP<strong>re</strong>cision plus one digit, the p<strong>re</strong>cision probably<br />

speci ed by the user, and the value of P<strong>re</strong>cisionControl plus 13 digits | provided<br />

that p<strong>re</strong>cision control is used at all. The value of AccuracyControl cannot be used to<br />

set the work<strong>in</strong>g p<strong>re</strong>cision s<strong>in</strong>ce this would <strong>re</strong>qui<strong>re</strong> one to know the order of magnitude<br />

of the <strong>in</strong>tegral befo<strong>re</strong>hand.<br />

FunctionType 2 fS<strong>in</strong>Argument, CosArgument, ZeroListg selects how the function<br />

fzero is to be <strong>in</strong>terp<strong>re</strong>ted. ZeroList means that the function <strong>re</strong>turns consecutive<br />

partition po<strong>in</strong>ts. The other possibilities specify the circular function whose zeros a<strong>re</strong><br />

computed for the use as subdivision po<strong>in</strong>ts. For S<strong>in</strong>Argument, which is also the default<br />

value, the equation (x) = k is solved, for CosArgument, it is the equation<br />

(x) = k + =2. If the oscillat<strong>in</strong>g factor u( ) is a pu<strong>re</strong> circular function, then this<br />

option additionally allows one to choose the ext<strong>re</strong>ma rather than the zeros as partition<br />

po<strong>in</strong>ts.<br />

The default options of OscInt cause the function to run at mach<strong>in</strong>e p<strong>re</strong>cision and with the<br />

control loop for the p<strong>re</strong>cision of the <strong>re</strong>sult disabled. The second argument is <strong>in</strong> this case<br />

<strong>re</strong>garded as the argument (x) of the oscillation. These default sett<strong>in</strong>gs make up the fastest<br />

and most user-friendly con guration.<br />

7.2 Structu<strong>re</strong> of the package<br />

The function OscInt is actually noth<strong>in</strong>g but a driver that calls other functions depend<strong>in</strong>g on<br />

the sett<strong>in</strong>gs of various options. Fig. 7.1 shows the <strong>re</strong>spective functions and their dependencies.<br />

The most important dist<strong>in</strong>ction <strong>in</strong>side the package is that between the use of error control<br />

for the extrapolation and the pla<strong>in</strong> computation of the <strong>in</strong>tegral without any feedback loop.<br />

Depend<strong>in</strong>g on this decision, OscInt <strong>in</strong>vokes the follow<strong>in</strong>g functions:<br />

PartitionTable computes a list of subdivision po<strong>in</strong>ts with a given length, start<strong>in</strong>g at<br />

a suitable po<strong>in</strong>t above the lower <strong>in</strong>tegration limit. In fact it only calls the functions<br />

PartitionOffs and PartitionPo<strong>in</strong>ts with a <strong>re</strong>duced set of <strong>in</strong>put parameters.<br />

PartInt takes as <strong>in</strong>put a set of subdivision po<strong>in</strong>ts and computes the <strong>in</strong>tegral by extrapolation.<br />

Like PartitionTable, itis used only for straightforward quadratu<strong>re</strong> without<br />

error control.<br />

OscIntControlled is the version of OscInt used with error control. It comprises a<br />

control loop that monitors the p<strong>re</strong>cision of the extrapolation <strong>re</strong>sult and <strong>in</strong>c<strong>re</strong>ases the<br />

length of the sequence of partial sums if needed.<br />

164


7.3 Implementation of OscInt and <strong>re</strong>lated functions<br />

OscInt<br />

PartInt PartitionTable<br />

OscIntControlled<br />

PartitionPo<strong>in</strong>ts PartitionOffs<br />

Figu<strong>re</strong> 7.1: The package structu<strong>re</strong> of OscInt<br />

Auxiliary<br />

Functions<br />

PartitionOffs computes an <strong>in</strong>dex o set value for the list of partition po<strong>in</strong>ts. This<br />

o set is then used <strong>in</strong> PartitionPo<strong>in</strong>ts.<br />

PartitionPo<strong>in</strong>ts <strong>re</strong>turns a list of subdivision po<strong>in</strong>ts with given length and the possibility<br />

to specify a start<strong>in</strong>g <strong>in</strong>dex. The <strong>in</strong>dex o set obta<strong>in</strong>ed from PartitionOffs adjusts<br />

the number<strong>in</strong>g such that the po<strong>in</strong>t with <strong>in</strong>dex zero always is the lowest possible partition<br />

po<strong>in</strong>t, with both the lower <strong>in</strong>tegration limit a or a manually set rst partition po<strong>in</strong>t a0<br />

taken <strong>in</strong>to account. The<strong>re</strong>fo<strong>re</strong>, x (0) > maxfa; a0g or x0 > maxfa; a0g, depend<strong>in</strong>g on<br />

the function be<strong>in</strong>g used to compute the subdivision po<strong>in</strong>ts.<br />

The afo<strong>re</strong>mentioned functions a<strong>re</strong> also di<strong>re</strong>ctly accessible from outside the package. Their<br />

syntax and implementation details a<strong>re</strong> dealt with <strong>in</strong> the next section. Apart from these<br />

functions that constitute the build<strong>in</strong>g blocks for the quadratu<strong>re</strong> rout<strong>in</strong>e, a few auxiliary<br />

function a<strong>re</strong> also <strong>in</strong>cluded <strong>in</strong> the package. They have noth<strong>in</strong>g to do with the quadratu<strong>re</strong><br />

itself, but a<strong>re</strong> useful for several special applications. The names and dependencies of these<br />

functions a<strong>re</strong> summarised <strong>in</strong> g. 7.2.<br />

7.3 Implementation of OscInt and <strong>re</strong>lated functions<br />

This section discusses the submodules of the quadratu<strong>re</strong> rout<strong>in</strong>e and provides <strong>in</strong>formation on<br />

the actual implementation. However, as we only want to outl<strong>in</strong>e the pr<strong>in</strong>ciples of operation,<br />

we shall concentrate on the most important details and avoid bor<strong>in</strong>g source code list<strong>in</strong>gs.<br />

Also, we shall not concern ourselves with the formal way of de n<strong>in</strong>g a Mathematica package,<br />

as this is cove<strong>re</strong>d extensively <strong>in</strong> books like those by Wolfram [140] and Maeder [146] or <strong>in</strong><br />

tutorial works like the book by Shaw and Tigg [147].<br />

165


7.3.1 OscInt<br />

ApproxLimGeneric<br />

AsymptoticExpand<br />

PolynomialDeg<strong>re</strong>e<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

Approximation error control Zero computation<br />

ApproxLimQuad<br />

ApproxLimHyp<br />

QuadOffset<br />

QuadZero<br />

HypOffset<br />

HypZero<br />

Figu<strong>re</strong> 7.2: The auxiliary functions of OscInt<br />

HypOffsetExact<br />

HypZeroExact<br />

S<strong>in</strong>ce the<strong>re</strong> a<strong>re</strong> two di e<strong>re</strong>nt list of arguments that can be passed to OscInt (see also section<br />

7.1), we must specify two di e<strong>re</strong>nt rules that match these possibilities. The mo<strong>re</strong> general set<br />

of arguments is the one whe<strong>re</strong> the rst partition po<strong>in</strong>t is p<strong>re</strong>set by the user. The case whe<strong>re</strong><br />

only the lower <strong>in</strong>tegration limit is given can be easily extended to the general rule by sett<strong>in</strong>g<br />

a0 = a. The<strong>re</strong>fo<strong>re</strong> only the body of the general rule needs to conta<strong>in</strong> the enti<strong>re</strong> code, whe<strong>re</strong>as<br />

the body of the mo<strong>re</strong> special case simply calls OscInt aga<strong>in</strong> with the proper parameters,<br />

which is generally conside<strong>re</strong>d good programm<strong>in</strong>g style.<br />

OscInt[f<strong>in</strong>t_, fzero_, fa_, a0_g, opts___Rule] :=<br />

Module[fparttab,lowlim,p<strong>re</strong>c,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.foptsg/.Options[OscInt],<br />

nt = NSumTerms/. foptsg/.Options[OscInt],<br />

ne = NSumExtraTerms/. foptsg/.Options[OscInt],<br />

ac = AccuracyControl/. foptsg/.Options[OscInt],<br />

pc = P<strong>re</strong>cisionControl/.foptsg/.Options[OscInt]g,<br />

If[ac === Inf<strong>in</strong>ity && pc === Inf<strong>in</strong>ity,<br />

lowlim = If[a == a0, a, N[Max[a,a0],2 wp]];<br />

parttab = PartitionTable[fzero,lowlim,ne+nt+1,opts];<br />

PartInt[f<strong>in</strong>t,parttab,a,opts],<br />

p<strong>re</strong>c = If[pc === Inf<strong>in</strong>ity,<br />

Max[wp,$Mach<strong>in</strong>eP<strong>re</strong>cision+1],<br />

Max[wp,$Mach<strong>in</strong>eP<strong>re</strong>cision+1,pc+13]];<br />

OscIntControlled[f<strong>in</strong>t,fzero,a,Work<strong>in</strong>gP<strong>re</strong>cision->p<strong>re</strong>c,opts]]<br />

];<br />

OscInt[f<strong>in</strong>t_, fzero_, a_, opts___Rule] :=<br />

OscInt[f<strong>in</strong>t,fzero,fa,ag,opts];<br />

Remark (Multiple de nitions) Note that the order of the two de nitions with<strong>in</strong> the<br />

package is essential, because Mathematica searches the list from the top to the bottom<br />

166


7.3 Implementation of OscInt and <strong>re</strong>lated functions<br />

<strong>in</strong> order to nd a rule that matches a given set of arguments. If the two de nitions we<strong>re</strong><br />

<strong>in</strong>terchanged, a list fb,b0g would match the pattern a , and consequently the variable a<br />

would erroneously be assigned the value fb,b0g.<br />

The function rst extracts the values of a couple of <strong>re</strong>levant options and then checks whether<br />

error control is <strong>re</strong>qui<strong>re</strong>d. If both AccuracyControl and P<strong>re</strong>cisionControl equal Inf<strong>in</strong>ity,<br />

then no error control is used, and OscInt performs a straightforward computation. The lower<br />

limit of the list of subdivision po<strong>in</strong>ts is set either to a or maxfa; a0g, the latter be<strong>in</strong>g calculated<br />

numerically at a higher p<strong>re</strong>cision, because the function Max works only with numbers but not<br />

with symbols.<br />

Remark (Options and defaults) The lter<strong>in</strong>g of the options is worth a separate<br />

mention [146]. It is accomplished by a double <strong>re</strong>placement rule like for <strong>in</strong>stance <strong>in</strong> wp<br />

= Work<strong>in</strong>gP<strong>re</strong>cision/.foptsg/.Options[OscInt], which works as follows: If the user<br />

speci es a value Work<strong>in</strong>gP<strong>re</strong>cision->p<strong>re</strong>c for this option, then it is passed to the function<br />

<strong>in</strong> the argument opts. The statement is now evaluated from the left to the right, and<br />

the rst <strong>re</strong>placement evaluates to p<strong>re</strong>c. The second substitution fails because the list<br />

of default options obta<strong>in</strong>ed by Options[OscInt] conta<strong>in</strong>s of course no rule p<strong>re</strong>c->x. If,<br />

on the other hand, the function is called without any option, then the rst <strong>re</strong>placement<br />

cannot be evaluated. In this case, the second rule matches and wp is set to the default<br />

value.<br />

The number of partition po<strong>in</strong>ts is obta<strong>in</strong>ed from NSumTerms plus NSumExtraTerms plus one,<br />

so that the sequence of partial sums has the cor<strong>re</strong>ct length for extrapolation. With the list<br />

of subdivision po<strong>in</strong>ts, PartInt is called to compute the actual <strong>in</strong>tegral.<br />

If error control is to be used, a <strong>re</strong>asonable value for the Work<strong>in</strong>gP<strong>re</strong>cision is calculated<br />

from the mach<strong>in</strong>e p<strong>re</strong>cision plus one, the parameter wp that might have been set by the user<br />

(or equals the default value otherwise), and the P<strong>re</strong>cisionControl parameter if applicable.<br />

This work<strong>in</strong>g p<strong>re</strong>cision, which might di er from the user's speci cation, is then used <strong>in</strong> the<br />

call of the function OscIntControlled. Note that this particular option is placed befo<strong>re</strong><br />

the other options opts, which causes the new sett<strong>in</strong>g for Work<strong>in</strong>gP<strong>re</strong>cision to override any<br />

other speci cation for this option, because Mathematica takes the rst value it encounters<br />

and dis<strong>re</strong>gards all subsequent ones.<br />

7.3.2 PartitionTable<br />

This function actually only calls PartitionOffs and PartitionPo<strong>in</strong>ts, but is nonetheless<br />

useful for the computation of a list of subdivision po<strong>in</strong>ts. Its syntax is<br />

PartitionTable[zerof, a, n, opts]<br />

with the same <strong>in</strong>put arguments as OscInt except for n, which denotes the number of partition<br />

po<strong>in</strong>ts to be computed. The options a<strong>re</strong> the same as for OscInt.<br />

167


7.3.3 PartitionOffs<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

The aim of the function PartitionOffs is to nd an <strong>in</strong>dex o set for the subsequent computation<br />

of the partition po<strong>in</strong>ts such that we can assume that the po<strong>in</strong>t with <strong>in</strong>dex zero is the<br />

smallest possible one | ir<strong>re</strong>spective of the lower <strong>in</strong>tegration limit or any particular shape of<br />

the argument function (x). The function call is<br />

PartitionOffs[zerof, a, opts],<br />

with zerof as function de n<strong>in</strong>g either the argument (x) oranenumerat<strong>in</strong>g function for the<br />

partition po<strong>in</strong>ts x (n). The parameter a is not necessarily the lower <strong>in</strong>tegration limit but<br />

can also be any arbitrarily chosen abscissa po<strong>in</strong>t meant asalower bound for the sequence of<br />

partition po<strong>in</strong>ts. The options a<strong>re</strong> the same as for OscInt.<br />

How the function is evaluated chie y depends on the option FunctionType that speci es how<br />

zerof is to be <strong>in</strong>terp<strong>re</strong>ted. We shall rst discuss the case whe<strong>re</strong> FunctionType->ZeroList,<br />

which means that zerof di<strong>re</strong>ctly <strong>re</strong>turns the subdivision po<strong>in</strong>ts x (n); n 0. This function<br />

has to be de ned follow<strong>in</strong>g one important rule: the rst po<strong>in</strong>t it <strong>re</strong>turns must be larger than<br />

the position of any ext<strong>re</strong>mum of (x) or its derivatives, so both (x) and its derivatives<br />

must be monotonic for x x (0). Then we must simply nd the smallest <strong>in</strong>dex o 0<br />

whe<strong>re</strong> x (o) a. This operation is performed by the follow<strong>in</strong>g statement, whe<strong>re</strong> the equation<br />

x (na) =ais solved numerically and the <strong>re</strong>sult is set to o = dnae. For this purpose, zerof[n]<br />

must be able to accept <strong>in</strong>put parameters of type <strong>re</strong>al although the value it <strong>re</strong>turns makes sense<br />

only for <strong>in</strong>teger <strong>in</strong>put values. The option MaxIterations is set to the value of the OscInt<br />

option MaxRecursion, AccuracyGoal <strong>in</strong> this particular case equals to PartitionAccuracy.<br />

If[N[zerof[0]] < N[a],<br />

Ceil<strong>in</strong>g[x/.F<strong>in</strong>dRoot[zerof[x] == a,fx,0,1g,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->pa,<br />

MaxIterations->ma]],<br />

0]<br />

Remark (Implementation) To determ<strong>in</strong>e whether x (0)


7.3 Implementation of OscInt and <strong>re</strong>lated functions<br />

the complete list of solutions. We then <strong>re</strong>gard only the <strong>re</strong>al part of the possibly conjugate<br />

complex solutions and <strong>re</strong>turn the largest one, provided it is still larger than the lower limit a.<br />

Thus we haveo= maxfxo;ag.<br />

ext<strong>re</strong>ma = NSolve[zerof'[x] == 0,x,Work<strong>in</strong>gP<strong>re</strong>cision->wp];<br />

If[Length[ext<strong>re</strong>ma] == 0 || ext<strong>re</strong>ma === ffgg, a,<br />

Max[Re[x/.Last[ext<strong>re</strong>ma]],a]]<br />

Remark (Ext<strong>re</strong>ma) To nd the ext<strong>re</strong>ma of (x), we cannot use F<strong>in</strong>dRoot s<strong>in</strong>ce we do<br />

not know whe<strong>re</strong> to start the iteration. Hence we cannot ensu<strong>re</strong> that we obta<strong>in</strong> the largest<br />

solution. This is why wehave to use the slower NSolve. From the code it is also clear that<br />

the function zerof must be di e<strong>re</strong>ntiable. The odd-look<strong>in</strong>g test for ext<strong>re</strong>ma === ffgg<br />

is noth<strong>in</strong>g but the condition 0 (x) 0, because this structu<strong>re</strong> is what NSolve <strong>re</strong>turns to<br />

<strong>in</strong>dicate that the equation is satis ed for all x.<br />

Example 7.3.1 Consider the argument function<br />

(x) =x(x,1+j)(x,1,j)(x,4+j)(x,4,j)<br />

=x , x 4 ,10x 3 +35x 2 ,50x +34 ;<br />

which has no ext<strong>re</strong>mum at all, as the plot shows. Consequently, the rst derivative has two pairs of<br />

conjugate complex zeros, and PartitionOffs <strong>in</strong>deed <strong>re</strong>turns the larger <strong>re</strong>al part of the two.<br />

In[1]:= f[x_] := x (x-1+I) (x-1-I) (x-4+I) (x-4-I)<br />

In[2]:= Plot[f[x],fx,-0.5,4.5g];<br />

60<br />

40<br />

20<br />

-20<br />

1 2 3 4<br />

In[3]:= NSolve[f'[x]==0,x]<br />

Out[3]= ffx -> 0.740373 - 0.294381 Ig,fx -> 0.740373 + 0.294381 Ig,<br />

fx -> 3.25963 - 0.294381 Ig,fx -> 3.25963 + 0.294381 Igg<br />

In[4]:= PartitionOffs[f,0]<br />

Out[4]= 3.25963<br />

169


7.3.4 PartitionPo<strong>in</strong>ts<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

This function <strong>re</strong>turns a list of subdivision po<strong>in</strong>ts <strong>in</strong> ascend<strong>in</strong>g order. The way to call it is<br />

PartitionPo<strong>in</strong>ts[zerof, a, n, offs, i<strong>re</strong>l, opts],<br />

zerof be<strong>in</strong>g the well-known function (x) orx (n). The argument a denotes the lower bound<br />

of the sequence, n is the desi<strong>re</strong>d number of po<strong>in</strong>ts. The variable offs is the o set o obta<strong>in</strong>ed<br />

from PartitionOffs. This function could have been merged <strong>in</strong>to PartitionPo<strong>in</strong>ts (and <strong>in</strong><br />

fact had been <strong>in</strong> a former version of PartitionTable), but as for <strong>in</strong>stance OscIntControlled<br />

makes multiple calls to PartitionPo<strong>in</strong>ts to gradually build up the table of subdivision po<strong>in</strong>ts,<br />

is was deemed mo<strong>re</strong> e cient to split these two functional blocks. The parameter i<strong>re</strong>l denotes<br />

the <strong>in</strong>dex r 0 of the rst po<strong>in</strong>t and can be used to compute subdivision tables <strong>in</strong> two or<br />

mo<strong>re</strong> portions.<br />

Like PartitionOffs, this function also behaves di e<strong>re</strong>ntly for di e<strong>re</strong>nt values of the option<br />

FunctionType. The simplest case is when zerof <strong>re</strong>turns the partition po<strong>in</strong>ts themselves<br />

(FunctionType->ZeroList). Then we just need to build up the table fx (r);x (r+<br />

1);::: ;x (r+n,1)g.<br />

Th<strong>in</strong>gs a<strong>re</strong> mo<strong>re</strong> complicated when (x) is given di<strong>re</strong>ctly (FunctionType->S<strong>in</strong>Argument or<br />

CosArgument). Then the evaluation depends on the behaviour of (x) at x ! 1. If the<br />

limit is de nite, then the <strong>in</strong>tegrand is not <strong>in</strong> nitely oscillat<strong>in</strong>g, and a partition extrapolation<br />

quadratu<strong>re</strong> is not applicable at all. In this case, we <strong>re</strong>turn Inf<strong>in</strong>ity so that the rst partial<br />

<strong>in</strong>tegral al<strong>re</strong>ady covers the enti<strong>re</strong> range. For an <strong>in</strong>de nite limit, we must dist<strong>in</strong>guish between<br />

<strong>in</strong>c<strong>re</strong>as<strong>in</strong>g and dec<strong>re</strong>as<strong>in</strong>g functions and nally obta<strong>in</strong> the partition po<strong>in</strong>t from the follow<strong>in</strong>g<br />

equations:<br />

zi :<br />

8<br />

><<br />

>:<br />

xi = 1 if ,1Inf<strong>in</strong>ity]]],<br />

Table[Inf<strong>in</strong>ity,fk,0,n-1g],<br />

N[zerof'[offs+1]] > 0,<br />

koffs = N[Ceil<strong>in</strong>g[zerof[offs]/Pi],2 wp];<br />

Re[Table[x/.F<strong>in</strong>dRoot[zerof[x] == (k+koffs+cfoffs) Pi,<br />

fx,offs,offs+1g],<br />

fk,i<strong>re</strong>l,i<strong>re</strong>l+n-1g]],<br />

True,<br />

170<br />

(7.2)


7.3 Implementation of OscInt and <strong>re</strong>lated functions<br />

],<br />

koffs = N[Floor[zerof[offs]/Pi],2 wp];<br />

Re[Table[x/.F<strong>in</strong>dRoot[zerof[x] == (-k+koffs-cfoffs) Pi,<br />

fx,offs,offs+1g],<br />

fk,i<strong>re</strong>l,i<strong>re</strong>l+n-1g]]<br />

Remark (F<strong>in</strong>d<strong>in</strong>g the limit) The test for a de nite limit of (x) was implemented<br />

di<strong>re</strong>ctly with the built-<strong>in</strong> function Limit. This works well enough for any k<strong>in</strong>d of polynomial<br />

and rational functions, but usually gives problems with trancendental functions,<br />

which, however, a<strong>re</strong> not supposed to be part of (x) anyhow. The outcome of Limit is<br />

then converted to a mach<strong>in</strong>e number and exam<strong>in</strong>ed whether it is a number or not to lter<br />

out the value Inf<strong>in</strong>ity.<br />

Whether (x) is <strong>in</strong>c<strong>re</strong>as<strong>in</strong>g or dec<strong>re</strong>as<strong>in</strong>g can be identi ed with the sign of 0 (x) for x>o.<br />

Accord<strong>in</strong>gly, the table of subdivision po<strong>in</strong>ts is computed with F<strong>in</strong>dRoot, which is aga<strong>in</strong><br />

straightforward because the <strong>re</strong>spective equations have only one solution above the start<strong>in</strong>g<br />

po<strong>in</strong>ts o and o +1. Although these solutions must be <strong>re</strong>al-valued, the<strong>re</strong> a<strong>re</strong> often spurious<br />

imag<strong>in</strong>ary parts left, which a<strong>re</strong> elim<strong>in</strong>ated by tak<strong>in</strong>g only the <strong>re</strong>al part of the <strong>re</strong>sult<strong>in</strong>g<br />

table.<br />

7.3.5 PartInt<br />

This function actually computes the <strong>in</strong>tegral R 1<br />

f(x) dx based on a set of n subdivision po<strong>in</strong>ts<br />

a<br />

fx0;x1;::: ;xn,1g; xi a by means of extrapolation. Its syntax is<br />

PartInt[f, zerotab, a, opts],<br />

whe<strong>re</strong> f is the <strong>in</strong>tegrand function (de ned either as pu<strong>re</strong> function or like f[x]), zerotab<br />

is the list of partition po<strong>in</strong>ts, and a is the lower <strong>in</strong>tegration limit. The essential part of the<br />

function body is listed below without most of the option assignments. The rst statement isa<br />

test that should always fail under normal conditions, because the functions that generate the<br />

subdivision po<strong>in</strong>ts al<strong>re</strong>ady make su<strong>re</strong> that the rst po<strong>in</strong>t x0 a. If this condition, however, is<br />

not satis ed, then all members g<strong>re</strong>ater than a a<strong>re</strong> lte<strong>re</strong>d out for further use, and a warn<strong>in</strong>g<br />

message is issued to <strong>in</strong>form the user that only fewer sequence members than <strong>in</strong>tended will be<br />

computed.<br />

Remark (Select<strong>in</strong>g members of lists) The lter<strong>in</strong>g of the partition po<strong>in</strong>ts is accomplished<br />

with the function Select[list, crit], which tests each element of a list with a<br />

given criterion and picks out those for which the test yields True.<br />

The value of the <strong>in</strong>tegral consists of two parts. The rst is the de nite <strong>in</strong>tegral R x0<br />

f(x) dx<br />

a<br />

(if, of course, x0 6= a) that is computed with the double exponential rule accord<strong>in</strong>g to the<br />

<strong>re</strong>sults of chapter 6.3. The second contribution is the extrapolation <strong>re</strong>sult that is obta<strong>in</strong>ed<br />

<strong>in</strong> the well-known manner by comput<strong>in</strong>g a sequence of partial <strong>in</strong>tegrals R xi+1<br />

f(x) dx and<br />

xi<br />

nd<strong>in</strong>g the numerical limit of this series. If the rst partition po<strong>in</strong>t equals Inf<strong>in</strong>ity, then<br />

this extrapolation is skipped. This featu<strong>re</strong> is imperative <strong>in</strong> order to <strong>in</strong>clude all special cases<br />

of parameter-dependent <strong>in</strong>tegrands that lose their <strong>in</strong> nite oscillations under certa<strong>in</strong> circumstances.<br />

171


7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

If[N[zeros[[1]]] < N[a],<br />

zeros = Select[zeros,(N[#] >= N[a])&];<br />

Message[OscInt::lowfirst,Length[zeros]]];<br />

If[N[zeros[[1]]] == N[a], 0,<br />

NIntegrate[f[x],fx,a,zeros[[1]]g,<br />

Method->DoubleExponential]] +<br />

If[N[zeros[[1]]] == Inf<strong>in</strong>ity, 0,<br />

seqtab = Table[NIntegrate[f[x],fx,zeros[[i]],zeros[[i+1]]g,<br />

Method->GaussKronrod],<br />

fi,1,Length[zeros]-1g];<br />

SequenceLimit[Take[FoldList[Plus,0,seqtab],-ne],WynnDeg<strong>re</strong>e->wd]]<br />

7.3.6 OscIntControlled<br />

This function is probably the most <strong>in</strong>tricate of the package. It is basically a copy ofPartInt<br />

extended by a control loop. The user <strong>in</strong>terface is exactly the same as that of OscInt, it is<br />

<strong>in</strong>voked by<br />

OscIntControlled[f, fzero, a, opts].<br />

The only di e<strong>re</strong>nce <strong>in</strong> the behaviour towards the user is that the sett<strong>in</strong>g of Work<strong>in</strong>gP<strong>re</strong>cision<br />

is not touched <strong>in</strong> any way, which means that the user himself must choose a <strong>re</strong>asonable value<br />

(<strong>in</strong> particular one g<strong>re</strong>ater than the mach<strong>in</strong>e p<strong>re</strong>cision). The meta algorithm that is used is<br />

shown <strong>in</strong> g. 7.3. This algorithm is of course applied only if error control is enabled, otherwise<br />

the function works much like PartInt.<br />

compute <strong>in</strong>itial set of partition po<strong>in</strong>ts fx0;x1;::: ;xng<br />

compute rst partial <strong>in</strong>tegral R x0<br />

a f(x) dx<br />

compute partial <strong>in</strong>tegrals R xi+1<br />

f(x) dx<br />

xi<br />

compute sequence limit<br />

iterations := 1<br />

while not <strong>re</strong>ached p<strong>re</strong>cision or accuracy or maximum iterations do<br />

compute il additional partition po<strong>in</strong>ts<br />

compute additional partial <strong>in</strong>tegrals<br />

iterations := iterations + 1<br />

compute sequence limit<br />

check error conditions<br />

Figu<strong>re</strong> 7.3: The meta algorithm for OscIntControlled<br />

The portion of the function body that implements the control loop is given below. Note<br />

that the table of partition po<strong>in</strong>ts parttab is not augmented <strong>in</strong> each pass, but only the last<br />

element is kept befo<strong>re</strong> the new po<strong>in</strong>ts a<strong>re</strong> added. For the sake of simplicity, the variable ne<br />

that <strong>in</strong>itially holds the value of NSumExtraTerms is <strong>in</strong>c<strong>re</strong>ased by the number il of additional<br />

sequence members di<strong>re</strong>ctly.<br />

172


7.4 Auxiliary functions<br />

While[iterations < it && Accuracy[<strong>in</strong>tlim] < ac && P<strong>re</strong>cision[<strong>in</strong>tlim] < pc,<br />

parttab = Jo<strong>in</strong>[Last[parttab],<br />

PartitionPo<strong>in</strong>ts[fzero,a,il,offs,ne+nt+1,opts]];<br />

seqtab = Jo<strong>in</strong>[seqtab,<br />

Table[NIntegrate[f[x],x,parttab[[i]],parttab[[i+1]],<br />

Method->GaussKronrod],<br />

i,1,Length[parttab]-1]];<br />

ne = ne + il;<br />

iterations++;<br />

<strong>in</strong>tlim = SequenceLimit[Take[FoldList[Plus,firstval,seqtab],<br />

-ne],WynnDeg<strong>re</strong>e->wd]<br />

];<br />

Example 7.3.2 We want to compute R 1<br />

s<strong>in</strong> x<br />

x<br />

dx to the p<strong>re</strong>cision of 10 signi cant digits. Note that<br />

0<br />

the <strong>in</strong>tegrand as well as the argument a<strong>re</strong> given as pu<strong>re</strong> functions. If we choose the work<strong>in</strong>g p<strong>re</strong>cision<br />

too low, we hit the iteration limit.<br />

In[5]:= OscIntControlled[S<strong>in</strong>[#]/#&,#&,0,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->17,P<strong>re</strong>cisionControl->10]<br />

OscInt::accfail:<br />

OscInt failed to achieve the desi<strong>re</strong>d accuracy or p<strong>re</strong>cision after 6<br />

iterations. Inc<strong>re</strong>ase MaxIterations or Work<strong>in</strong>gP<strong>re</strong>cision (cur<strong>re</strong>ntly 17).<br />

Out[5]= 1.570798<br />

If we <strong>in</strong>c<strong>re</strong>ase the work<strong>in</strong>g p<strong>re</strong>cision, we obta<strong>in</strong> the desi<strong>re</strong>d <strong>re</strong>sult.<br />

In[6]:= OscIntControlled[S<strong>in</strong>[#]/#&,#&,0,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->20,P<strong>re</strong>cisionControl->10]<br />

Out[6]= 1.570796327<br />

7.4 Auxiliary functions<br />

In addition to the functions that a<strong>re</strong> needed for pu<strong>re</strong> quadratu<strong>re</strong>, a number of support<strong>in</strong>g<br />

rout<strong>in</strong>es have been added to the package. These functions facilitate the use of the quadratu<strong>re</strong><br />

programs <strong>in</strong> that they provide <strong>in</strong>formation <strong>re</strong>qui<strong>re</strong>d by the quadratu<strong>re</strong> algorithm.<br />

A set of such functions <strong>re</strong>turns the zeros fx0;x1;x2;:::g of s<strong>in</strong> f(x) for quadratic and hyperbolic<br />

arguments f(x). In order to be suitable as partition po<strong>in</strong>ts for the quadratu<strong>re</strong>, they<br />

a<strong>re</strong> de ned such that the rst zero x0 always lies to the right of the ext<strong>re</strong>mum of f(x). It is<br />

thus necessary to nd solutions for the equation f(x) =(k+o) , whe<strong>re</strong> k 0 and the o set<br />

o = f(xe)= is chosen so that f(x) is monotonic for x>xe.<br />

173


7.4.1 Zero computation for quadratic arguments<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

Let f(x) be a quadratic polynomial. We thus need to solve the equation<br />

ax 2 + bx + c =(k+o) ; k 0 (7.3)<br />

for arbitrary coe cients and for the right branch of the parabola. Depend<strong>in</strong>g on the coe -<br />

cients, we nd the solutions<br />

x =<br />

8<br />

><<br />

>:<br />

p<br />

,b+ b2 ,4a(c,(k+o) )<br />

2a<br />

,b, p b 2 ,4a(c,(,k+o) )<br />

2a<br />

(k+o) ,c<br />

b<br />

(,k+o) ,c<br />

b<br />

if a>0<br />

if a0<br />

if a =0, b<<br />

>:<br />

d 4ac,b2<br />

4a e if a>0<br />

b 4ac,b2<br />

4a c if a


7.4 Auxiliary functions<br />

L<strong>in</strong>ear approximation<br />

For x 0, we can approximate (7.6) by<br />

with the solutions<br />

(a + b)x + d =(k+o) ; k 0 (7.7)<br />

x =<br />

8<br />

><<br />

>:<br />

(k+o) ,d<br />

a+b<br />

(,k+o) ,d<br />

a+b<br />

if a + b>0<br />

if a + b<<br />

>:<br />

f 0 (x) =a+<br />

q<br />

+ a2c (7.8)<br />

x<br />

p x 2 +c =0; (7.9)<br />

b2 ,a2 if jbj > jaj; a b jaj; a b>0:<br />

With these values we can now solve (7.7) for o and get<br />

o =<br />

8<br />

><<br />

>:<br />

d (a+b)xe+d e if jbj > jaj; b>0<br />

b (a+b)xe+d c if jbj > jaj; b0<br />

b d c if jbj jaj; a+b


Exact solution<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

For the exact solution of (7.6) we must dist<strong>in</strong>guish several cases: when f(x) is monotonically<br />

dec<strong>re</strong>as<strong>in</strong>g (a + b0) or below (b<<br />

>:<br />

a((k+o) ,d), p b2 ((a2 ,b2 )c+((k+o) ,d) 2 )<br />

a2 ,b2 p<br />

a((k+o) ,d)+ b2 ((a2 ,b2 )c+((k+o) ,d) 2 )<br />

a2 ,b2 p<br />

a((,k+o) ,d)+ b2 ((a2 ,b2 )c+((,k+o) ,d) 2 )<br />

a2 ,b2 a((,k+o) ,d), p b 2 ((a 2 ,b 2 )c+((,k+o) ,d) 2 )<br />

a 2 ,b 2<br />

((k+o) ,d) 2 ,a 2 c<br />

2a((k+o) ,d)<br />

if a + b>0; b>0<br />

if a + b>0; b<<br />

>:<br />

b d,p (b 2 ,a 2 )c c if jbj > jaj; b0<br />

b d+bp c c if jbj jaj; a+b


7.4 Auxiliary functions<br />

For the sake of comput<strong>in</strong>g e ciency, two dist<strong>in</strong>ct functions have been de ned for each case.<br />

The one that computes the <strong>in</strong>dex o set <strong>re</strong>qui<strong>re</strong>s noth<strong>in</strong>g but the coe cients as an <strong>in</strong>put,<br />

whe<strong>re</strong>as the other also needs the p<strong>re</strong>-computed o set value and the <strong>in</strong>dex of the subdivision<br />

po<strong>in</strong>t.<br />

HypOffsetExact[a_,b_,c_,d_] := Which[<br />

(Abs[b] > Abs[a]) && (c >= 0) && b > 0,<br />

Ceil<strong>in</strong>g[(Sqrt[c (b^2-a^2)] + d)/Pi],<br />

(Abs[b] > Abs[a]) && (c >= 0) && b < 0,<br />

Floor[(-Sqrt[c (b^2-a^2)] + d)/Pi],<br />

(a+b) > 0, Ceil<strong>in</strong>g[(b Sqrt[c] + d)/Pi],<br />

(a+b) < 0, Floor[(b Sqrt[c] + d)/Pi],<br />

True, 0];<br />

HypZeroExact[a_,b_,c_,d_,o_][k_] := Which[<br />

a == b && a != 0,<br />

(((k+o) Pi - d)^2 - a^2 c)/(2 a ((k+o) Pi - d)),<br />

(a+b) > 0,<br />

(a (( k+o) Pi-d) -<br />

Sign[b] Sqrt[b^2 ((a^2-b^2) c + (( k+o) Pi-d)^2)])/(a^2-b^2),<br />

(a+b) < 0,<br />

(a ((-k+o) Pi-d) +<br />

Sign[b] Sqrt[b^2 ((a^2-b^2) c + ((-k+o) Pi-d)^2)])/(a^2-b^2),<br />

True, Inf<strong>in</strong>ity];<br />

In the example, the use of the signum function allowed to contract the cases <strong>in</strong> (7.12) that<br />

di er only <strong>in</strong> the sign of b. Although this br<strong>in</strong>gs no advantage <strong>in</strong> terms of execution time, it<br />

<strong>re</strong>nders the code mo<strong>re</strong> compact.<br />

Remark (Implementation) The Which[test1, exp1, test2, exp2,...] statement<br />

that is used to implement the functions consists of a sequence of conditional statements<br />

and exp<strong>re</strong>ssions. The tests a<strong>re</strong> evaluated <strong>in</strong> turn until one of them yields True. In this<br />

case, the follow<strong>in</strong>g exp<strong>re</strong>ssion is evaluated, and the <strong>re</strong>sult is <strong>re</strong>turned. The tests the<strong>re</strong>fo<strong>re</strong><br />

need not be mutually exclusive, but ca<strong>re</strong> must be taken to arrange them <strong>in</strong> the cor<strong>re</strong>ct<br />

order. It is also advisable to <strong>in</strong>clude a default value with a p<strong>re</strong>ceed<strong>in</strong>g True statement as<br />

last element to ensu<strong>re</strong> that the function always <strong>re</strong>turns a <strong>re</strong>sult.<br />

It is important that the coe cients used <strong>in</strong> the function call be numbers and not symbolic<br />

exp<strong>re</strong>ssions. The <strong>re</strong>ason is that the conditional statements could not be evaluated when used<br />

with symbolic exp<strong>re</strong>ssions unless their values a<strong>re</strong> computed explicitly. 2 Pi > Pi is the<strong>re</strong>fo<strong>re</strong><br />

left unevaluated, whe<strong>re</strong>as N[2 Pi] > N[Pi] calculates the numerical values of both sides of<br />

the exp<strong>re</strong>ssion prior to compar<strong>in</strong>g them and yields the cor<strong>re</strong>ct answer. To make the functions<br />

versatile with <strong>re</strong>spect to the numerical p<strong>re</strong>cision, such a conversion has not been <strong>in</strong>cluded <strong>in</strong><br />

the functions.<br />

177


7.4.3 Approximation error control<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

Another auxilliary function implements the criterion of equation (6.13),<br />

0 (x)<br />

0 (x) , 0 (x) = k;<br />

which has been given as a quality measu<strong>re</strong> for asymptotic partition. It <strong>re</strong>lates the argument<br />

function and its polynomial approximation by means of their <strong>re</strong>spective oscillation f<strong>re</strong>quencies.<br />

The di e<strong>re</strong>nce between these two cause the actual partition po<strong>in</strong>ts to deviate from the ideal<br />

phase <strong>in</strong>variant positions, which <strong>in</strong> turn produces wavepacket-like pictu<strong>re</strong>s <strong>in</strong> the sequence of<br />

partial sums. Hence the above formula hence gives the length of such awave packet depend<strong>in</strong>g<br />

on the abscissa value. We would like, however, to p<strong>re</strong>scribe k and get the cor<strong>re</strong>spond<strong>in</strong>g x as<br />

a <strong>re</strong>sult <strong>in</strong> order to know whe<strong>re</strong> to start the extrapolation. The functions needed to this end<br />

exploit the symbolical calculation capabilities of Mathematica.<br />

To nd the polynomial approximation of a given function f(x) for x ! 1, we de ne two<br />

small rout<strong>in</strong>es that determ<strong>in</strong>e the polynomial deg<strong>re</strong>e of f(x) and then compute the asymptotic<br />

expansion to the cor<strong>re</strong>ct order.<br />

PolynomialDeg<strong>re</strong>e[f_,x_Symbol] :=<br />

Length[CoefficientList[Series[f,fx,Inf<strong>in</strong>ity,1g],x]]-1;<br />

AsymptoticExpand[f_,x_Symbol] :=<br />

Normal[Series[f,fx,Inf<strong>in</strong>ity,PolynomialDeg<strong>re</strong>e[f,x]g]];<br />

Remark (Series expansion) The function Series[f, x, x0, n] generates a power series<br />

expansion for a given function to the order n about the po<strong>in</strong>t x0. In PolynomialDeg<strong>re</strong>e<br />

we use it just to determ<strong>in</strong>e the highest order terms, that is why we set n = 1. With<br />

CoefficientList we then obta<strong>in</strong> a list of the coe cients of the <strong>re</strong>sult<strong>in</strong>g polynomial<br />

and, by <strong>re</strong>gard<strong>in</strong>g the length of this list, the deg<strong>re</strong>e of the polynomial.<br />

In AsymptoticExpand, we aga<strong>in</strong> calculate the power series, but now to the cor<strong>re</strong>ct order.<br />

The command Normal nally converts the series expansion to a normal exp<strong>re</strong>ssion by<br />

cutt<strong>in</strong>g o the error term that is also <strong>re</strong>turned by Series.<br />

Example 7.4.1 Consider the function f(x) =x+ p x 4 +x 3 ,x+1. The series expansion calculated<br />

to nd the deg<strong>re</strong>e of the polynomial <strong>in</strong>cludes only the highest order terms and an error term.<br />

In[7]:= Series[x+Sqrt[x^4+x^3-x+1],fx,Inf<strong>in</strong>ity,1g]<br />

Out[7]= 2 3 x 1 0<br />

x + --- + O[-]<br />

2 x<br />

The length of the list of coe cients (start<strong>in</strong>g with x 0 ) gives the desi<strong>re</strong>d value.<br />

In[8]:= CoefficientList[%,x]<br />

178


7.4 Auxiliary functions<br />

Out[8]= 3<br />

f0, -, 1g<br />

2<br />

The cor<strong>re</strong>ct asymptotic expansion comprises all terms <strong>in</strong> x down to x 0 , but of course no error term.<br />

In[9]:= AsymptoticExpand[x+Sqrt[x^4+x^3-x+1],x]<br />

Out[9]= 1 3 x 2<br />

-(-) + --- + x<br />

8 2<br />

These support<strong>in</strong>g functions a<strong>re</strong> used <strong>in</strong>side a mo<strong>re</strong> <strong>in</strong>tricate module that rst determ<strong>in</strong>es the<br />

asymptotic expansion, then solves the equation given above and <strong>re</strong>turns the largest <strong>re</strong>al-valued<br />

solution. The<strong>re</strong> a<strong>re</strong> several cases that lead to error or warn<strong>in</strong>g messages:<br />

As the asymptotic expansion must be of the form (x) =xpP1ai i=0 xi , p be<strong>in</strong>g a nonnegative<br />

<strong>in</strong>teger, all expansions with other than <strong>in</strong>teger exponents a<strong>re</strong> <strong>re</strong>jected (for <strong>in</strong>stance<br />

(x) = p x3 ).<br />

If no solution exists, an error is issued.<br />

The existence of mo<strong>re</strong> than one solution is <strong>re</strong>ported to the user, and the largest solution<br />

is <strong>re</strong>turned.<br />

The function ApproxLimGeneric takes as <strong>in</strong>puts the argument function that is to be approximated<br />

and the number of subdivision po<strong>in</strong>ts that a<strong>re</strong> sought with<strong>in</strong> one wavepacket.<br />

ApproxLimGeneric[f_,goal_] :=<br />

Module[fx,fapprox,g,coeffs,<strong>re</strong>spos,<strong>re</strong>sneg,<strong>re</strong>sg,<br />

fapprox = AsymptoticExpand[f[x],x];<br />

coeffs = CoefficientList[fapprox,x];<br />

If[D[coeffs,x] =!= Table[0,fLength[coeffs]g],<br />

Message[ApproxLimGeneric::nopoly],<br />

g = D[fapprox,x] / D[fapprox - f[x],x];<br />

<strong>re</strong>spos = Select[NSolve[g == goal,x],(Im[x/.#] == 0)&];<br />

<strong>re</strong>sneg = Select[NSolve[g == -goal,x],(Im[x/.#] == 0)&];<br />

<strong>re</strong>s = Select[Jo<strong>in</strong>[<strong>re</strong>spos,<strong>re</strong>sneg],Positive[x/.#]&];<br />

If[Length[<strong>re</strong>s] == 0,<br />

Message[ApproxLimGeneric::nosolution],<br />

If[Length[<strong>re</strong>s] > 1,<br />

Message[ApproxLimGeneric::nounique,Length[<strong>re</strong>s],<br />

Map[x/.#&,<strong>re</strong>s]]];<br />

x/.Last[Sort[<strong>re</strong>s]]]]<br />

]<br />

179


7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

Remark (Implementation) To check whether the exponential expansion comprises<br />

only terms with <strong>in</strong>teger exponents, we simply form the derivative of the list of coe cients.<br />

S<strong>in</strong>ce terms with non-<strong>in</strong>teger exponents a<strong>re</strong> listed together with the constant terms, the<br />

derivative <strong>in</strong>such cases is di e<strong>re</strong>nt from zero. Note that to check this, we must test the<br />

two exp<strong>re</strong>ssions for symbolic identity with the <strong>re</strong>lational operator =!=, because a simple<br />

test for equality with != will not be evaluated if the variables <strong>in</strong> the symbolic exp<strong>re</strong>ssion<br />

D[f, x] a<strong>re</strong> not assigned any values.<br />

S<strong>in</strong>ce it is not possible to numerically solve equations that <strong>in</strong>volve functions like Abs,<br />

NSolve must be <strong>in</strong>voked twice with both the positive and negative possibility for the<br />

variable goal. The <strong>re</strong>sults a<strong>re</strong> lte<strong>re</strong>d such that only the <strong>re</strong>al and positive solutions a<strong>re</strong><br />

<strong>re</strong>ta<strong>in</strong>ed. As NSolve <strong>re</strong>turns a list of <strong>re</strong>placement rules fx->x1, x->x2 ...g, wemust<br />

apply the <strong>re</strong>placement operator /. to the elements to enable test<strong>in</strong>g for a vanish<strong>in</strong>g<br />

imag<strong>in</strong>ary part or the like. The operation of the Select command has been described <strong>in</strong><br />

section 7.3.5 .<br />

Example 7.4.2 The next two examples show the error check<strong>in</strong>g mechanisms. The function <strong>in</strong> the<br />

rst one cannot be exp<strong>re</strong>ssed as apower series with <strong>in</strong>teger exponents, whe<strong>re</strong>as the second one has<br />

multiple solutions. Note that the functions a<strong>re</strong> speci ed as pu<strong>re</strong> functions that need no separate<br />

declaration.<br />

In[10]:= ApproxLimGeneric[Sqrt[#^3]&,10]<br />

ApproxLimGeneric::nopoly:<br />

Function has no polynomial expansion with <strong>in</strong>teger exponents.<br />

In[11]:= ApproxLimGeneric[(#^2-20 Sqrt[#^2+1])&,20]<br />

ApproxLimGeneric::nounique:<br />

The<strong>re</strong> a<strong>re</strong> 3 <strong>re</strong>al solutions f3.98475, 8.68827, 10.8449g,<br />

he<strong>re</strong> is the largest.<br />

Out[11]= 10.8449<br />

The<strong>re</strong> a<strong>re</strong> also two special cases of this module <strong>in</strong>cluded <strong>in</strong> the package: ApproxLimQuad<br />

for parabola-like functions ax 2 + b p x 2 + c + d and ApproxLimHyp for functions of the form<br />

ax + b p x 2 + c + d. They can be found together with the code list<strong>in</strong>g of the package <strong>in</strong> the<br />

appendix.<br />

180


7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

We now want to apply our quadratu<strong>re</strong> rout<strong>in</strong>e to some of the test functions that a<strong>re</strong> commonly<br />

used <strong>in</strong> the literatu<strong>re</strong>. The objective of such benchmarks usually is to exam<strong>in</strong>e how<br />

many evaluations of the <strong>in</strong>tegrand function a given algorithm needs to <strong>re</strong>turn a <strong>re</strong>sult of the<br />

<strong>re</strong>qui<strong>re</strong>d accuracy. The number of function evaluations is <strong>in</strong> turn a measu<strong>re</strong> for the e ectiveness<br />

of compet<strong>in</strong>g algorithms. In the case of our quadratu<strong>re</strong> rout<strong>in</strong>e, however, we obta<strong>in</strong> no<br />

<strong>in</strong>formation how often Mathematica needs to evaluate the <strong>in</strong>tegrand for its <strong>in</strong>ternal computations.<br />

All we can test is the accuracy of the <strong>re</strong>sult depend<strong>in</strong>g on the sett<strong>in</strong>gs of the algorithm<br />

parameters. So the ma<strong>in</strong> goal of this section is to verify that the algorithm works cor<strong>re</strong>ctly<br />

rather than to compa<strong>re</strong> it with other rout<strong>in</strong>es.<br />

As the execution time of the rout<strong>in</strong>e is not our ma<strong>in</strong> concern he<strong>re</strong>, we do not use explicit<br />

enumeration functions that de ne the partition po<strong>in</strong>ts for the quadratu<strong>re</strong>. We just specify<br />

the argument function of the oscillat<strong>in</strong>g factors of the <strong>in</strong>tegrand and leave the determ<strong>in</strong>ation<br />

of the appropriate subdivision of the <strong>in</strong>tegration range to Mathematica.<br />

Remark (Pu<strong>re</strong> functions) To avoid unnecessary declarations, we use pu<strong>re</strong> functions<br />

[140] to specify the functions that a<strong>re</strong> needed to compute the partition. In such a function<br />

the places whe<strong>re</strong> the <strong>in</strong>dependent variables a<strong>re</strong> to be <strong>in</strong>serted upon evaluation a<strong>re</strong> marked<br />

with slots (#, ##, #1, ::: ), the function itself is de ned as pu<strong>re</strong> by a trail<strong>in</strong>g ampersand (&).<br />

So while we would de ne a quadratic polynomial <strong>in</strong> full notation as p[x_] := x^2 + 3 x,<br />

we write the pu<strong>re</strong> function as (#^2 + 3 #)&. This concept enables us to write very<br />

compact code, but it is useful only if the function is not <strong>re</strong>qui<strong>re</strong>d to have a name.<br />

Prior to test<strong>in</strong>g we must load the package.<br />

In[12]:=


Out[14]= 0.537450389063711<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

The <strong>re</strong>sult is cor<strong>re</strong>ct to 13 digits, although Mathematica <strong>in</strong>itially considers it less <strong>re</strong>liable. The di e<strong>re</strong>nce<br />

<strong>in</strong> the last two digits is due to roundo errors. We can still improve the <strong>re</strong>sult by us<strong>in</strong>g a higher<br />

work<strong>in</strong>g p<strong>re</strong>cision and <strong>in</strong>clud<strong>in</strong>g mo<strong>re</strong> terms <strong>in</strong> the extrapolation.<br />

In[15]:= OscInt[f,#&,0,Work<strong>in</strong>gP<strong>re</strong>cision->30,NSumExtraTerms->25]<br />

Out[15]= 0.53745038906373283<br />

It is <strong>in</strong>te<strong>re</strong>st<strong>in</strong>g to notice that the last digit of the `exact' value given by Eh<strong>re</strong>nmark seems to be wrong,<br />

for if we let Mathematica compute the <strong>in</strong>tegral analytically, we obta<strong>in</strong> a slightly di e<strong>re</strong>nt <strong>re</strong>sult.<br />

In[16]:= Integrate[f[x],fx,0,Inf<strong>in</strong>ityg]<br />

Out[16]= Pi BesselI[0, 2] 3 3<br />

---------------- - 2 HypergeometricPFQ[f1g, f-, -g, 1]<br />

2 2 2<br />

In[17]:= N[%,20]<br />

Out[17]= 0.5374503890637328029<br />

Example 7.5.2<br />

Z 1<br />

In[18]:= f[x_] := S<strong>in</strong>[x]/Sqrt[x];<br />

OscInt[f,#&,0] - N[Sqrt[Pi/2],20]<br />

Out[18]= -12<br />

5.05929 10<br />

0<br />

r<br />

s<strong>in</strong> x<br />

p dx =<br />

x 2<br />

Aga<strong>in</strong> this <strong>re</strong>sult could be improved with a higher work<strong>in</strong>g p<strong>re</strong>cision, albeit at the expense of an<br />

<strong>in</strong>c<strong>re</strong>ased comput<strong>in</strong>g time.<br />

In[19]:= OscInt[f,#&,0,NSumTerms->30,NSumExtraTerms->20,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->34] - N[Sqrt[Pi/2],30]<br />

Out[19]= -19<br />

0. 10<br />

The next <strong>in</strong>tegrals have been used by Espelid and Overholt as benchmarks [123]. S<strong>in</strong>ce their<br />

algorithm was tailo<strong>re</strong>d to <strong>re</strong>gularly oscillat<strong>in</strong>g <strong>in</strong>tegrands, the arguments of the s<strong>in</strong>-functions<br />

a<strong>re</strong> at least asymptotically l<strong>in</strong>ear.<br />

182


7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Example 7.5.3<br />

Z 1<br />

In[20]:= f[x_] := S<strong>in</strong>[x]/Sqrt[1+x];<br />

SetP<strong>re</strong>cision[OscInt[f,#&,0],16]<br />

Out[20]= 0.8095254817472858<br />

Example 7.5.4<br />

0<br />

Z 1<br />

1<br />

s<strong>in</strong> x<br />

p 1+x dx =0:809525481747<br />

s<strong>in</strong>(x + 1<br />

x )<br />

p dx =0:232948197094<br />

x<br />

In[21]:= f[x_] := S<strong>in</strong>[x+1/x]/Sqrt[x];<br />

SetP<strong>re</strong>cision[OscInt[f,(#+1/#)&,1],16]<br />

Out[21]= 0.23294819709403225<br />

Example 7.5.5<br />

Z 1<br />

1<br />

s<strong>in</strong>(x + 1 p ) x<br />

p dx =0:0416328516893<br />

x<br />

In[22]:= f[x_] := S<strong>in</strong>[x+1/Sqrt[x]]/Sqrt[x];<br />

SetP<strong>re</strong>cision[OscInt[f,(#+1/Sqrt[#])&,1],16]<br />

Out[22]= 0.04163285168937769<br />

In the last two examples it would have also been possible to use only the l<strong>in</strong>ear part of the<br />

argument function for the determ<strong>in</strong>ation of the partition po<strong>in</strong>ts by substitut<strong>in</strong>g the pu<strong>re</strong><br />

function #& for the <strong>re</strong>spective arguments <strong>in</strong> the function call. The <strong>re</strong>sults would <strong>in</strong> fact have<br />

been the same.<br />

Some other functions with l<strong>in</strong>early grow<strong>in</strong>g numbers of oscillations have been exam<strong>in</strong>ed by<br />

Hasegawa and Torii [127].<br />

Example 7.5.6<br />

Z 1<br />

0<br />

e ,x cos xdx = 1<br />

2<br />

183


In[23]:= f[x_] := Exp[-x] Cos[x];<br />

OscInt[f,#&,0]<br />

Out[23]= 0.5<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

In this case we do not even have the slightest approximation error, as can be shown by the attempt<br />

to <strong>re</strong>veal further digits of the <strong>re</strong>sult.<br />

In[24]:= SetP<strong>re</strong>cision[%,40]<br />

Out[24]= 0.5<br />

Example 7.5.7<br />

Z 1<br />

0<br />

x<br />

x 2 +1 s<strong>in</strong> xdx = 2e<br />

In[25]:= f[x_] := x/(x^2+1) S<strong>in</strong>[x];<br />

SetP<strong>re</strong>cision[OscInt[f,#&,0],16] - N[Pi/(2 E),20]<br />

Out[25]= -14<br />

-1.93 10<br />

Example 7.5.8<br />

Z 1<br />

0<br />

cos x<br />

p x 2 +1 dx = K0(1)<br />

In[26]:= f[x_] := Cos[x]/Sqrt[x^2+1];<br />

SetP<strong>re</strong>cision[OscInt[f,#&,0],16] - N[BesselK[0,1]]<br />

Out[26]= -13<br />

-1.12188 10<br />

Example 7.5.9<br />

Z 1<br />

0<br />

ln x2 +4<br />

x2 +1 cos !x dx = , ,! ,2!<br />

e , e<br />

!<br />

In[27]:= f[x_] := Log[(x^2+4)/(x^2+1)] Cos[omega x];<br />

omega = 5;<br />

SetP<strong>re</strong>cision[OscInt[f,omega #&,0],16] -<br />

N[(Exp[-omega]-Exp[-2 omega]) Pi/omega,20]<br />

184


7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Out[27]= -15<br />

-8.471 10<br />

The next test functions we<strong>re</strong> taken from Sidi [128].<br />

Example 7.5.10<br />

Z 1<br />

0<br />

s<strong>in</strong> x<br />

x<br />

s<strong>in</strong>h x<br />

10<br />

s<strong>in</strong>h 2x dx = arctan tan<br />

10<br />

1<br />

4<br />

tanh 10<br />

4<br />

In[28]:= f[x_] := S<strong>in</strong>[x]/x S<strong>in</strong>h[1/10 x]/S<strong>in</strong>h[2/10 x];<br />

OscInt[f,#&,0] - N[ArcTan[Tan[1/4 Pi] Tanh[10 Pi/4]]]<br />

Out[28]= -13<br />

-2.72449 10<br />

Inc<strong>re</strong>as<strong>in</strong>g both Work<strong>in</strong>gP<strong>re</strong>cision and NSumExtraTerms,we obta<strong>in</strong> a <strong>re</strong>sult with mo<strong>re</strong> cor<strong>re</strong>ct digits.<br />

Although Sidi gives the exact value of the <strong>in</strong>tegral as 0.785398012695720765, the last two digits di er<br />

<strong>in</strong> the <strong>re</strong>sult found by Mathematica as well as <strong>in</strong> the numerical evaluation of the analytical solution.<br />

In[29]:= OscInt[f,#&,0,Work<strong>in</strong>gP<strong>re</strong>cision->34,NSumExtraTerms->30]<br />

Out[29]= 0.78539801269572077061037<br />

In[30]:= N[ArcTan[Tan[1/4 Pi] Tanh[10 Pi/4]],20]<br />

Out[30]= 0.78539801269572077061<br />

Example 7.5.11<br />

Z 1<br />

In[31]:= f[x_] := S<strong>in</strong>[Pi/2 x^2];<br />

OscInt[f,Pi/2 #^2&,0] - 0.5<br />

Out[31]= -14<br />

-2.08167 10<br />

0<br />

s<strong>in</strong> x2<br />

2<br />

185<br />

dx = 1<br />

2


Example 7.5.12<br />

Z 1<br />

0<br />

s<strong>in</strong> x 2<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

cos x2<br />

4<br />

dx<br />

x2 = e, , 1<br />

4 p 2<br />

This <strong>in</strong>tegral, taken from an earlier work of Sidi [126], is di e<strong>re</strong>nt from the others exam<strong>in</strong>ed so far<br />

<strong>in</strong> that the <strong>in</strong>tegrand has an <strong>in</strong> nite number of oscillations both for x ! 1 and x ! 0. Thus it<br />

cannot be t<strong>re</strong>ated <strong>in</strong> a straightforward manner like the <strong>in</strong>tegrals befo<strong>re</strong> and we split the <strong>in</strong>tegration<br />

<strong>in</strong>terval at the po<strong>in</strong>t x =1. The right part I1 = R 1<br />

s<strong>in</strong> 0 x2 cos x2 dx<br />

4 x2 <strong>re</strong>ma<strong>in</strong>s unchanged, whe<strong>re</strong>as we<br />

transform the left part by achange of variable x ! 1=x to obta<strong>in</strong> a second <strong>in</strong>tegral <strong>in</strong> our standard<br />

form I2 = R 1<br />

s<strong>in</strong>( x 1 2 ) cos 4x2 dx.<br />

In[32]:= f1[x_] := S<strong>in</strong>[Pi/x^2] Cos[Pi x^2/4]/x^2;<br />

f2[x_] := S<strong>in</strong>[Pi x^2] Cos[Pi/(x^2 4)];<br />

OscInt[f1,Pi/4 #^2&,1] + OscInt[f2,Pi #^2&,1] -<br />

N[(Exp[-Pi] - 1)/(4 Sqrt[2])]<br />

Out[32]= -14<br />

4.54636 10<br />

The next <strong>in</strong>tegrals a<strong>re</strong> taken from the tables of Abramowitz and Stegun [148] and <strong>in</strong>volve<br />

Bessel functions that a<strong>re</strong> known to behave like cos x for large values of x. Thus the partition<br />

po<strong>in</strong>ts may be computed by a l<strong>in</strong>ear function.<br />

Example 7.5.13<br />

In[33]:= f[x_] := BesselJ[0,x];<br />

OscInt[f,#&,0] - 1<br />

Out[33]= -14<br />

-4.74065 10<br />

Example 7.5.14<br />

Z 1<br />

0<br />

Z 1<br />

0<br />

J0(x) dx =1<br />

cos xK0(x)dx =1<br />

In[34]:= f[x_] := Cos[x] BesselK[0,x];<br />

OscInt[f,#&,0] - N[Pi/(2 Sqrt[2])]<br />

Out[34]= -14<br />

-2.84217 10<br />

186


7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

The follow<strong>in</strong>g <strong>in</strong>tegrals can be found <strong>in</strong> the tables of Grobner and Hof<strong>re</strong>iter [149]. Some of<br />

them have complex-valued <strong>in</strong>tegrands to show that the quadratu<strong>re</strong> rout<strong>in</strong>e is able to t<strong>re</strong>at<br />

complex functions as well.<br />

Example 7.5.15<br />

Z 1<br />

0<br />

x ,<br />

e 10 s<strong>in</strong> p xdx = 10<br />

2<br />

p 10 e , 10 4<br />

In[35]:= f[x_] := Exp[-x/10] S<strong>in</strong>[Sqrt[x]];<br />

OscInt[f,Sqrt[#]&,0] - N[5 Sqrt[10 Pi] Exp[-10/4]]<br />

Out[35]= -11<br />

2.87295 10<br />

Example 7.5.16<br />

Z 1<br />

,1<br />

e j(x2 +x) dx = p e j( 4 , 1 4 )<br />

In order to be able to compute this <strong>in</strong>tegral, we divide it at the orig<strong>in</strong> and transform the left half<br />

onto the positive <strong>re</strong>al axis. We the<strong>re</strong>by obta<strong>in</strong> two <strong>in</strong>tegrals <strong>in</strong> our standard notation R 1<br />

0 ej(x2 +x) dx +<br />

R 1<br />

0 ej(x2 ,x) dx. The <strong>re</strong>sult is accurate only to n<strong>in</strong>e digits if we leave the options at their default values.<br />

A larger number of terms, however, signi cantly improves the <strong>re</strong>sult.<br />

In[36]:= f1[x_] := Exp[I (x^2 + x)];<br />

f2[x_] := Exp[I (x^2 - x)];<br />

OscInt[f1,(#^2+#)&,0,NSumExtraTerms->30] +<br />

OscInt[f2,(#^2-#)&,0,NSumExtraTerms->30] -<br />

N[Sqrt[Pi] Exp[I (Pi/4 - 1/4)],20]<br />

Out[36]= -13 -13<br />

1.42331 10 - 1.4494 10 I<br />

Example 7.5.17<br />

Z 1<br />

0<br />

1 j(x,<br />

e x ) dx<br />

r<br />

p =(1+j)<br />

x 2 e,2<br />

S<strong>in</strong>ce the <strong>in</strong>tegrand <strong>in</strong> this case, too, has an <strong>in</strong> nite number of zeros as it approaches the lower<br />

<strong>in</strong>tegration limit, we aga<strong>in</strong> apply the successful strategy of divid<strong>in</strong>g the <strong>in</strong>tegration <strong>in</strong>terval at the<br />

po<strong>in</strong>t x = 1. We map the <strong>in</strong>terval [0; 1] to the semi-<strong>in</strong> nite range [1; 1[ by the transformation<br />

x ! 1=x and obta<strong>in</strong> the two <strong>in</strong>tegrals R 1<br />

0 ej(x, 1 x ) dx<br />

px and R 1<br />

0 ej( 1 x ,x) p x<br />

x 2 dx that a<strong>re</strong> amenable to our<br />

quadratu<strong>re</strong> rout<strong>in</strong>e.<br />

187


In[37]:= f1[x_] := Exp[I (x - 1/x)]/Sqrt[x];<br />

f2[x_] := Exp[I (1/x - x)] Sqrt[x]/x^2;<br />

OscInt[f1,(#-1/#)&,1] + OscInt[f2,(1/#-#)&,1] -<br />

N[(1+I) Sqrt[Pi/2] Exp[-2]]<br />

Out[37]= -13 -13<br />

9.02889 10 + 2.14218 10 I<br />

7 Mathematica implementation of a quadratu<strong>re</strong> function<br />

Instead of us<strong>in</strong>g the exact argument functions x , 1=x and 1=x , x to determ<strong>in</strong>e the partition po<strong>in</strong>ts<br />

we could as well take their polynomial parts x and ,x, <strong>re</strong>spectively, which yields even slightly better<br />

<strong>re</strong>sults.<br />

In[38]:= OscInt[f1,#&,1] + OscInt[f2,-#&,1] -<br />

N[(1+I) Sqrt[Pi/2] Exp[-2]]<br />

Out[38]= -15 -14<br />

2.52576 10 + 3.71925 10 I<br />

Example 7.5.18<br />

Z 1<br />

0<br />

s<strong>in</strong> 3 x<br />

x 2<br />

In[39]:= f[x_] := S<strong>in</strong>[x]^3/x^2;<br />

OscInt[f,#&,0] - N[3/4 Log[3]]<br />

Out[39]= -14<br />

-7.39409 10<br />

Example 7.5.19<br />

Z 1<br />

0<br />

s<strong>in</strong> x , x cos x<br />

x 3<br />

In[40]:= f[x_] := (S<strong>in</strong>[x] - x Cos[x])/x^3;<br />

OscInt[f,#&,0] - N[Pi/4]<br />

Out[40]= -13<br />

-2.498 10<br />

Example 7.5.20<br />

Z 1<br />

0<br />

s<strong>in</strong> x 2 , 1<br />

x 2<br />

dx = 3<br />

ln 3<br />

4<br />

dx = 4<br />

dx = 1<br />

r<br />

2 2 e,2<br />

Divid<strong>in</strong>g the range of <strong>in</strong>tegration and with the change of variable x2 ! 1=x2 , we obta<strong>in</strong> the two<br />

2 dx<br />

<strong>in</strong>tegrals R 1<br />

1 s<strong>in</strong> , x 2 , 1<br />

x 2<br />

dx and R 1<br />

1 s<strong>in</strong> , 1<br />

x 2 , x<br />

188<br />

x 2 , <strong>re</strong>spectively.


7.5 Test of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

In[41]:= f1[x_] := S<strong>in</strong>[x^2 - 1/x^2];<br />

f2[x_] := S<strong>in</strong>[1/x^2 - x^2]/x^2;<br />

OscInt[f1,(#^2-1/#^2)&,1] + OscInt[f2,(1/#^2-#^2)&,1] -<br />

N[Sqrt[Pi/2] Exp[-2]/2]<br />

Out[41]= -13<br />

-1.51323 10<br />

With only the polynomial parts of the argument functions and an <strong>in</strong>c<strong>re</strong>ased number of sequence<br />

members for the extrapolation, we obta<strong>in</strong> a comparable <strong>re</strong>sult.<br />

In[42]:= OscInt[f1,(#^2)&,1,NSumExtraTerms->20] +<br />

OscInt[f2,(-#^2)&,1,NSumExtraTerms->20] -<br />

N[Sqrt[Pi/2] Exp[-2]/2]<br />

Out[42]= -13<br />

-1.78635 10<br />

The last <strong>in</strong>tegral is aga<strong>in</strong> taken from Sidi [128] and shows a limitation of our rout<strong>in</strong>e.<br />

Example 7.5.21<br />

Z 1<br />

0<br />

J0(x) J1(x) dx<br />

x<br />

S<strong>in</strong>ce the Bessel functions for large x behave like cos x, wechoose the l<strong>in</strong>ear function to compute the<br />

partition po<strong>in</strong>ts. Still, the <strong>re</strong>sult of the quadratu<strong>re</strong> is rather poor and any improvement needs a large<br />

number of terms to be <strong>in</strong>cluded <strong>in</strong> the summation. But even for several thousand terms the accuracy<br />

is limited to n<strong>in</strong>e digits.<br />

In[43]:= f[x_] := BesselJ[0,x] BesselJ[1,x]/x;<br />

OscInt[f,#&,0] - N[2/Pi]<br />

Out[43]= -7<br />

-2.31225 10<br />

We have seen <strong>in</strong> this section that our quadratu<strong>re</strong> rout<strong>in</strong>e is versatile and can be applied<br />

successfully to a wide variety of <strong>in</strong>tegrals with ir<strong>re</strong>gularly oscillat<strong>in</strong>g behaviour. In some<br />

cases such as<strong>in</strong>tegration <strong>in</strong>tervals that a<strong>re</strong> <strong>in</strong> nite at both ends, the <strong>in</strong>tegral must be divided<br />

and transformed befo<strong>re</strong> it can be t<strong>re</strong>ated with the algorithm. We now can nd a proper<br />

formulation for the wave <strong>in</strong>tegrals to <strong>re</strong>nder them amenable to our rout<strong>in</strong>e.<br />

189<br />

= 2


Chapter 8<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Application of the quadratu<strong>re</strong><br />

rout<strong>in</strong>e<br />

Newton Verstehen Sie etwas von Elektrizitat, Richard?<br />

Inspektor Ich b<strong>in</strong> ke<strong>in</strong> Physiker.<br />

Newton Ich verstehe auch wenig davon. Ich stelle nur aufgrund von<br />

Naturbeobachtungen e<strong>in</strong>e Theorie daruber auf. Diese Theorie sch<strong>re</strong>ibe ich<br />

<strong>in</strong> der Sprache der Mathematik nieder und erhalte meh<strong>re</strong><strong>re</strong> Formeln. Dann<br />

kommen die Techniker. Sie kummern sich nur noch um die Formeln. Sie<br />

gehen mit der Elektrizitat um wie der Zuhalter mit der Dirne. Sie nutzen<br />

sie aus. Sie stellen Masch<strong>in</strong>en her, und brauchbar ist e<strong>in</strong>e Masch<strong>in</strong>e erst<br />

dann, wenn sie von der Erkenntnis unabhangig geworden ist, die zu ih<strong>re</strong>r<br />

Er ndung fuhrte. Friedrich Dur<strong>re</strong>nmatt, Die Physiker<br />

Hav<strong>in</strong>g a quadratu<strong>re</strong> rout<strong>in</strong>e at hand that is capable of deal<strong>in</strong>g with ir<strong>re</strong>gularly oscillat<strong>in</strong>g<br />

functions, we can now apply it to the wave <strong>in</strong>tegrals of the rst part of this work | as an<br />

example, we choose the tunnel<strong>in</strong>g particle. To this end, a little bit of <strong>re</strong>writ<strong>in</strong>g is necessary<br />

to permit an e cient computation. In this chapter, we shall develop a Mathematica package<br />

compris<strong>in</strong>g all functions necessary to compute the wave function <strong>in</strong>side and outside the tunnel<br />

as well as provid<strong>in</strong>g a simple user <strong>in</strong>terface. F<strong>in</strong>ally, we shall test the implementation both<br />

numerically and analytically.<br />

The<strong>re</strong> is a good <strong>re</strong>ason why weselect the tunnel e ect to demonstrate what the quadratu<strong>re</strong><br />

rout<strong>in</strong>e is capable of: from all physical models we discussed <strong>in</strong> the rst part, this is by far the<br />

most <strong>in</strong>tricate one. While we mighthave succeeded with standard algorithms <strong>in</strong> all other cases,<br />

this one <strong>re</strong>ally calls for our non-standard approach. But not only a<strong>re</strong> the <strong>in</strong>tegrals di cult to<br />

compute, the variety of functions also <strong>re</strong>sults <strong>in</strong> a complex structu<strong>re</strong> of the package. Hav<strong>in</strong>g<br />

understood the implementation of this example, one can easily deduce the others.<br />

190


8.1 P<strong>re</strong>paration of the wave <strong>in</strong>tegrals for quadratu<strong>re</strong><br />

8.1 P<strong>re</strong>paration of the wave <strong>in</strong>tegrals for quadratu<strong>re</strong><br />

We <strong>re</strong>call from section 4.1 the <strong>in</strong>tegrals describ<strong>in</strong>g the wave function <strong>in</strong> the di e<strong>re</strong>nt <strong>re</strong>gions<br />

of a step potential barrier:<br />

<strong>in</strong>c = !p<br />

c<br />

<strong>re</strong>f = !p<br />

c<br />

+ !p<br />

c<br />

+ !p<br />

c<br />

tun = !p<br />

c<br />

+ !p<br />

c<br />

+ !p<br />

c<br />

Z 1<br />

A( ) e<br />

,1<br />

jX ,j 2T d ; (8.1)<br />

Z ,1<br />

,1<br />

Z 1<br />

,1<br />

Z 1<br />

1<br />

Z ,1<br />

,1<br />

Z 1<br />

,1<br />

Z 1<br />

1<br />

A( ) + p 2 , 1<br />

, p ,jX ,jT 2<br />

e d +<br />

2 , 1<br />

A( ) , jp1 , 2<br />

+ j p 2 ,jT<br />

e,jX d +<br />

1 , 2<br />

A( ) , p 2 , 1<br />

+ p ,jX ,jT 2<br />

e d ;<br />

2 , 1<br />

A( )<br />

A( )<br />

A( )<br />

2<br />

, p e<br />

2 , 1 ,jX<br />

p<br />

2,1,jT 2<br />

d +<br />

2<br />

+ j p 1 ,<br />

2 e,X<br />

p 1, 2 ,jT 2<br />

d +<br />

2<br />

+ p e<br />

2 , 1 jX<br />

p<br />

2,1,jT 2<br />

d :<br />

In section 4.2, we found the spectra of the <strong>in</strong>itial wave forms,<br />

A <strong>re</strong>ct ( )<br />

p l = j<br />

4 2 k<br />

A tria ( )<br />

p l<br />

A gauss( )<br />

p l<br />

p<br />

3<br />

=<br />

4 3k2 =<br />

1<br />

1 , 1<br />

p<br />

1<br />

1 , 1<br />

p<br />

1<br />

p 2 n p 2 e<br />

,<br />

0<br />

,j2 k 1, p1<br />

e<br />

,j k 1, p<br />

1<br />

2e 2<br />

@ 2 k p 1<br />

,1<br />

2n<br />

191<br />

1<br />

A 2<br />

(8.2)<br />

(8.3)<br />

, 1 ; (8.4)<br />

,j k 1, 1<br />

p<br />

,j2 k 1, p<br />

1<br />

, e<br />

, 1 ; (8.5)<br />

: (8.6)


8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

If we take a closer look at the <strong>in</strong>tegrands, we <strong>re</strong>cognise that they all have a similar structu<strong>re</strong>.<br />

The <strong>in</strong>tegrands of the wave functions have the general form<br />

f( )=CrA( )cr( )e jpr( ) ; (8.7)<br />

with a constant Cr, a coe cient function cr( ) that is de ned by the transmission and <strong>re</strong>ection<br />

coe cients and a function pr( ) describ<strong>in</strong>g the phase of the <strong>in</strong>tegrand. All these<br />

functions depend on the <strong>re</strong>gion whe<strong>re</strong> the wave function is evaluated and on the di<strong>re</strong>ction<br />

of propagation, <strong>re</strong>spectively. In addition, they might be slightly di e<strong>re</strong>nt for positive and<br />

negative wavenumbers because of the dispersion <strong>re</strong>lations.<br />

For the step potential barrier, these functions a<strong>re</strong> given below. Note that for the evanescent<br />

part of the spectrum <strong>in</strong> the <strong>in</strong>terval ,1 < 1ifwe take the pr<strong>in</strong>cipal value of the complex root. The<br />

constant Cr = !c=c is <strong>in</strong>dependent of both the <strong>re</strong>gion and the di<strong>re</strong>ction of propagation, so we<br />

dis<strong>re</strong>gard it <strong>in</strong> the sequel.<br />

c<strong>in</strong>c( )=1 (8.8)<br />

p<strong>in</strong>c( )=X , T 2<br />

c <strong>re</strong>f( )=<br />

8<br />

><<br />

>:<br />

p<br />

+ 2,1<br />

, p 2 ,1<br />

, p 2 ,1<br />

p<br />

+ 2,1<br />

p <strong>re</strong>f( )=,X , T 2<br />

ctun( )=<br />

ptun( )=<br />

8<br />

><<br />

>:<br />

8<br />

<<br />

:<br />

2<br />

, p 2 ,1<br />

2 p<br />

+ 2,1<br />

if ,1<br />

if ,1<br />

,X p 2 ,1,T 2 if ,1<br />

(8.9)<br />

(8.10)<br />

(8.11)<br />

(8.12)<br />

(8.13)<br />

The spectral functions, too, can be written <strong>in</strong> a uniform manner as the sum of exponential<br />

functions,<br />

A( )=Csss( )<br />

mX<br />

i=1<br />

as;i e jgs;i( ) ; (8.14)<br />

whe<strong>re</strong> Cs is some waveform-dependent constant and ss is a function that de nes the shape of<br />

the spectrum. Tak<strong>in</strong>g equations (8.7) and (8.14) together, we obta<strong>in</strong> the complete structu<strong>re</strong><br />

of the wave <strong>in</strong>tegrands,<br />

f( )=Csss( )cr( )<br />

mX<br />

i=1<br />

192<br />

as;i e j(gs;i( )+pr( )) : (8.15)


8.1 P<strong>re</strong>paration of the wave <strong>in</strong>tegrals for quadratu<strong>re</strong><br />

Now we a<strong>re</strong> <strong>re</strong>ady to formulate the <strong>in</strong>tegrals <strong>in</strong> an appropriate way for the implementation<br />

with Mathematica.<br />

The equations (8.1) to (8.6) still permit no straightforward application of our standard quadratu<strong>re</strong><br />

rout<strong>in</strong>e. Remember<strong>in</strong>g that we can extrapolate only to +1, we must make a variable<br />

transformation ! , rst for the <strong>in</strong>tegrals over the negative <strong>re</strong>al axis. Hav<strong>in</strong>g done so,<br />

we nd that the coe cient functions for the <strong>in</strong>cident, <strong>re</strong> ected, and transmitted part of the<br />

wave now become unique for both <strong>in</strong>tegrals, whe<strong>re</strong>as the exponential arguments as well as the<br />

coe cient functions of the wave spectra now a<strong>re</strong> di e<strong>re</strong>nt <strong>in</strong> the two <strong>in</strong>tegrals. The coe cient<br />

functions (8.8), (8.10), and (8.12) thus <strong>re</strong>duce to their positive branches<br />

c<strong>in</strong>c( )=1<br />

c<strong>re</strong>f( )= ,p 2 ,1<br />

+ p 2 ,1<br />

ctun( )=<br />

2<br />

+ p :<br />

2 ,1<br />

For the triangular wave packet, we can comb<strong>in</strong>e (8.5) and (8.1 { 8.3) and nd<br />

p<br />

3<br />

Ctria =<br />

2k 2p<br />

s +<br />

1<br />

tria ( )=<br />

1, 1 p<br />

s , tria ( )=<br />

1<br />

1+ 1<br />

p<br />

2<br />

2<br />

(8.16)<br />

(8.17)<br />

(8.18)<br />

With these de nitions we can put the parts together to eventually obta<strong>in</strong> the nal form of<br />

the <strong>in</strong>tegrals that is suitable for quadratu<strong>re</strong>:<br />

<strong>in</strong>c,tria<br />

p l = Ctria<br />

+ Ctria<br />

+ Ctria<br />

Z 1<br />

1<br />

Z 1<br />

1<br />

Z 1<br />

,1<br />

s +<br />

tria ( ) 2ej(,T 2 + k 1 p +X , k)<br />

,<br />

, e j(,T 2 +2 k 1<br />

p +X ,2 k) , e j(,T 2 +X ) d +<br />

s , tria ( ) 2ej(,T 2 , k 1<br />

p ,X , k) ,<br />

, e j(,T 2 ,2 k 1<br />

p ,X ,2 k) , e j(,T 2 ,X ) d +<br />

s +<br />

tria ( ) 2ej(,T 2 + k 1<br />

p +X , k)<br />

,<br />

, e j(,T 2 +2 k 1<br />

p +X ,2 k) , e j(,T 2 +X ) d<br />

193<br />

(8.19)


ef,tria<br />

trans,tria<br />

evan,tria<br />

p l = Ctria<br />

+ Ctria<br />

+ Ctria<br />

p l = Ctria<br />

+ Ctria<br />

p l = Ctria<br />

Z 1<br />

1<br />

Z 1<br />

1<br />

Z 1<br />

,1<br />

Z 1<br />

1<br />

Z 1<br />

1<br />

Z 1<br />

,1<br />

s +<br />

tria ( ) c<strong>re</strong>f( ) 2e j(,T 2 + k 1<br />

p ,X , k)<br />

,<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

, e j(,T 2 +2 k 1<br />

p ,X ,2 k) , e j(,T 2 +X ) d +<br />

s , tria ( ) c<strong>re</strong>f( ) 2e j(,T 2 , k 1<br />

p +X , k) ,<br />

, e j(,T 2 ,2 k 1<br />

p +X ,2 k) , e j(,T 2 +X ) d +<br />

s +<br />

tria ( ) c<strong>re</strong>f( ) 2e j(,T 2 + k 1<br />

p ,X , k)<br />

,<br />

, e j(,T 2 +2 k 1<br />

p ,X ,2 k) , e j(,T 2 ,X ) d<br />

s +<br />

tria ( ) ctun( ) 2e j(,T 2 + k 1<br />

p<br />

p +X 2,1, k)<br />

,<br />

, e j(,T 2 +2 k 1<br />

p +X<br />

s , tria ( ) ctun( ) 2e j(,T 2 , k 1<br />

p ,X<br />

p 2,1,2 k) , e j(,T 2 +X ) d +<br />

p 2,1, k) ,<br />

, e j(,T 2 ,2 k 1<br />

p<br />

p ,X 2,1,2 k) j(,T<br />

, e 2 ,X )<br />

d<br />

s +<br />

tria ( ) ctun( ) 2e j(,T 2 + k 1<br />

p<br />

p +X 2,1, k)<br />

,<br />

, e j(,T 2 +2 k 1<br />

p +X<br />

p 2,1,2 k) , e j(,T 2 +X ) d :<br />

(8.20)<br />

(8.21)<br />

(8.22)<br />

We dist<strong>in</strong>guish four di e<strong>re</strong>nt parts of the wave function. Inside the tunnel we have the<br />

evanescent and the transmitted portion, <strong>re</strong>spectively. Outside the tunnel, we separate the<br />

<strong>in</strong>cident from the <strong>re</strong> ected part. For these two parts a division of the <strong>in</strong>tegration range <strong>in</strong>to<br />

th<strong>re</strong>e parts | by analogy with the tunnel | has no physical <strong>re</strong>ason. Still we stick to this<br />

structu<strong>re</strong> because we then need only two modules for all parts of the wave function.<br />

As the <strong>in</strong>tegrands consist of th<strong>re</strong>e dist<strong>in</strong>ct terms, the <strong>in</strong>de nite <strong>in</strong>tegrals a<strong>re</strong> split <strong>in</strong> order<br />

to be amenable to the quadratu<strong>re</strong> function. The<strong>re</strong>fo<strong>re</strong>, the computation of the transmitted<br />

portion of the wave needs six <strong>in</strong>dividual quadratu<strong>re</strong> calls. Obviously the oscillatory parts of<br />

the <strong>in</strong>tegrands a<strong>re</strong> all very similar. Inside the tunnel, they have the structu<strong>re</strong><br />

otun( )=e j(a2 p<br />

+b +c 2,1+d)<br />

; (8.23)<br />

on the outside they look like<br />

oout( )=e j(a 2 +b +c +d) ; (8.24)<br />

with only the signs of the parameters chang<strong>in</strong>g. This notice is important for the implementation<br />

s<strong>in</strong>ce it allows us to compute the coe cients <strong>in</strong>side the module for all function calls and<br />

to change only their signs as they a<strong>re</strong> actually passed to the quadratu<strong>re</strong> function.<br />

194


8.1 P<strong>re</strong>paration of the wave <strong>in</strong>tegrals for quadratu<strong>re</strong><br />

For <strong>re</strong>ctangular <strong>in</strong>itial waves, we nd from (8.4) and (8.1 { 8.3)<br />

and<br />

<strong>in</strong>c,<strong>re</strong>ct<br />

<strong>re</strong>f,<strong>re</strong>ct<br />

trans,<strong>re</strong>ct<br />

evan,<strong>re</strong>ct<br />

p l = C<strong>re</strong>ct<br />

+ C<strong>re</strong>ct<br />

+ C<strong>re</strong>ct<br />

p l = C<strong>re</strong>ct<br />

+ C<strong>re</strong>ct<br />

+ C<strong>re</strong>ct<br />

p l = C<strong>re</strong>ct<br />

+ C<strong>re</strong>ct<br />

p l = C<strong>re</strong>ct<br />

Z 1<br />

1 Z 1<br />

1 Z 1<br />

,1<br />

Z 1<br />

1 Z 1<br />

1 Z 1<br />

,1<br />

Z 1<br />

1<br />

Z 1<br />

1<br />

Z 1<br />

,1<br />

C<strong>re</strong>ct =<br />

j<br />

p<br />

2<br />

(8.25)<br />

s +<br />

1<br />

<strong>re</strong>ct ( )=<br />

1, 1 p<br />

(8.26)<br />

s , <strong>re</strong>ct ( )=<br />

1<br />

1+ 1<br />

p<br />

s +<br />

h j(,T<br />

<strong>re</strong>ct ( ) e 2 +2 k 1 p +X ,2 k) j(,T<br />

, e 2 i<br />

+X )<br />

d +<br />

s , <strong>re</strong>ct ( )<br />

s +<br />

<strong>re</strong>ct ( )<br />

h j(,T<br />

e 2 ,2 k 1<br />

p ,X ,2 k) j(,T<br />

, e 2 i<br />

,X )<br />

d +<br />

h j(,T<br />

e 2 +2 k 1<br />

p +X ,2 k) j(,T<br />

, e 2 i<br />

+X )<br />

d<br />

s +<br />

<strong>re</strong>ct ( ) c h j(,T<br />

<strong>re</strong>f( ) e 2 +2 k 1 p ,X ,2 k) j(,T<br />

, e 2 i<br />

,X )<br />

d +<br />

h j(,T<br />

e 2 ,2 k 1 p +X ,2 k) j(,T<br />

, e 2 i<br />

+X )<br />

d +<br />

s , <strong>re</strong>ct ( ) c <strong>re</strong>f( )<br />

s +<br />

<strong>re</strong>ct ( ) c <strong>re</strong>f( )<br />

h j(,T<br />

e 2 +2 k 1<br />

p ,X ,2 k) j(,T<br />

, e 2 i<br />

,X )<br />

d<br />

s +<br />

<strong>re</strong>ct ( ) ctun( ) e j(,T 2 +2 k 1<br />

p<br />

p +X 2,1,2 k)<br />

, e j(,T 2 +X ) d +<br />

s , <strong>re</strong>ct ( ) ctun( ) e j(,T 2 ,2 k 1<br />

p<br />

p ,X 2,1,2 k)<br />

, e j(,T 2 ,X ) d<br />

s +<br />

<strong>re</strong>ct ( ) ctun( ) e j(,T 2 +2 k 1<br />

p<br />

p +X 2,1,2 k)<br />

, e j(,T 2 +X ) d :<br />

(8.27)<br />

(8.28)<br />

(8.29)<br />

(8.30)<br />

(8.31)<br />

The basic structu<strong>re</strong> of the <strong>in</strong>tegrals is the same as befo<strong>re</strong>, however <strong>in</strong> this case the computation<br />

of the transmitted wave needs only four calls of the quadratu<strong>re</strong> rout<strong>in</strong>e. Like with the<br />

triangular wave, it is su cient towrite one driver module for all parts of the wave function<br />

and alter only the parameters <strong>in</strong> the quadratu<strong>re</strong> call.<br />

195


Likewise, we collect the de nitions for a Gaussian wave,<br />

p<br />

2 k<br />

Cgauss = p p<br />

n 2<br />

s +<br />

, k<br />

n<br />

gauss( )=e<br />

s , k<br />

gauss ( )=e, n<br />

1<br />

p ,1<br />

1 , p ,1<br />

to write the nal form of the <strong>in</strong>tegrals<br />

p Z 1<br />

l = Cgauss s +<br />

gauss( ) e j(,T 2 + k 1 p +X , k)<br />

d +<br />

<strong>in</strong>c,gauss<br />

<strong>re</strong>f,gauss<br />

trans,gauss<br />

evan,gauss<br />

+ Cgauss<br />

+ Cgauss<br />

p l = Cgauss<br />

+ Cgauss<br />

+ Cgauss<br />

p l = Cgauss<br />

+ Cgauss<br />

p l = Cgauss<br />

1 Z 1<br />

1 Z 1<br />

,1<br />

Z 1<br />

1 Z 1<br />

1 Z 1<br />

,1<br />

Z 1<br />

1 Z 1<br />

1<br />

Z 1<br />

,1<br />

s , gauss ( ) ej(,T 2 , k 1<br />

p ,X , k) d +<br />

s +<br />

gauss ( ) ej(,T 2 + k 1<br />

p +X , k)<br />

d<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

s + gauss ( ) c <strong>re</strong>f( ) e j(,T 2 + k 1<br />

p ,X , k) d +<br />

s , gauss( ) c <strong>re</strong>f( ) e j(,T 2 , k 1<br />

p +X , k) d +<br />

s + gauss( ) c <strong>re</strong>f( ) e j(,T 2 + k 1<br />

p ,X , k) d<br />

2<br />

2<br />

(8.32)<br />

(8.33)<br />

; (8.34)<br />

s + gauss ( ) ctun( ) e j(,T 2 + k 1 p<br />

p<br />

+X 2,1, k)<br />

d +<br />

s , gauss ( ) ctun( ) e j(,T 2 , k 1 p<br />

p<br />

,X 2,1, k)<br />

d<br />

(8.35)<br />

(8.36)<br />

(8.37)<br />

s + gauss ( ) ctun( ) e j(,T 2 + k 1<br />

p<br />

p +X 2,1, k)<br />

d : (8.38)<br />

The computation of the <strong>in</strong>de nite <strong>in</strong>tegrals for a Gaussian wave <strong>re</strong>qui<strong>re</strong>s only two calls of the<br />

quadratu<strong>re</strong> rout<strong>in</strong>e. Furthermo<strong>re</strong>, <strong>in</strong> this special case, the<strong>re</strong> is an alternative to the extrapolat<strong>in</strong>g<br />

quadratu<strong>re</strong> strategy: as the <strong>in</strong>tegrands decay exponentially, we can safely truncate<br />

them at appropriate values of the <strong>in</strong>tegration variable and <strong>in</strong>tegrate the <strong>re</strong>ma<strong>in</strong><strong>in</strong>g de nite<br />

<strong>in</strong>tegral <strong>in</strong> a conventional way. A safe value for the truncation is whe<strong>re</strong> the shap<strong>in</strong>g functions<br />

Cgauss sgauss( ) become smaller than the work<strong>in</strong>g p<strong>re</strong>cision p, for example<br />

such that the truncation po<strong>in</strong>ts a<strong>re</strong><br />

trunc = p<br />

Cgauss sgauss 10 ,(p+1) ; (8.39)<br />

1<br />

r<br />

196<br />

ln Cgauss<br />

10 ,(p+1)<br />

n<br />

k<br />

!<br />

(8.40)


8.2 Outl<strong>in</strong>e of the program structu<strong>re</strong><br />

8.2 Outl<strong>in</strong>e of the program structu<strong>re</strong><br />

S<strong>in</strong>ce we have to cope with a large variety of similar functions, we must adopt a modular<br />

approach for the <strong>in</strong>dividual subrout<strong>in</strong>es. This means that the <strong>in</strong>tegrands a<strong>re</strong> comb<strong>in</strong>ed from<br />

several function modules follow<strong>in</strong>g the afo<strong>re</strong>mentioned structu<strong>re</strong>, which is the only possibility<br />

to ensu<strong>re</strong> consistency with<strong>in</strong> the package and to provide comparatively easy extendability. A<br />

<strong>re</strong>asonable implementation strategy <strong>in</strong>volves th<strong>re</strong>e dist<strong>in</strong>ct hierarchical levels.<br />

Driver modules provide the basic <strong>in</strong>terface to the user. The most user-friendly of<br />

them expects noth<strong>in</strong>g mo<strong>re</strong> than the parameters <strong>re</strong>qui<strong>re</strong>d to compute the wave function<br />

and an option de n<strong>in</strong>g the shape of the <strong>in</strong>itial waveform. Depend<strong>in</strong>g on the spatial<br />

coord<strong>in</strong>ate it then <strong>re</strong>turns either the wave function <strong>in</strong>side the tunnel or the sum of the<br />

<strong>in</strong>cident and <strong>re</strong> ected parts on the outside. These partial solutions a<strong>re</strong> computed <strong>in</strong><br />

turn by other driver modules that can also be accessed di<strong>re</strong>ctly. All modules must take<br />

<strong>in</strong>to account that di e<strong>re</strong>nt waveforms may need di e<strong>re</strong>nt numbers of parameters (such<br />

as the Gaussian wave packet, which <strong>re</strong>qui<strong>re</strong>s an additional geometry parameter for the<br />

variance of the distribution).<br />

Quadratu<strong>re</strong> modules a<strong>re</strong> dist<strong>in</strong>ct for all waveforms. The <strong>re</strong>ason for this is that the<br />

number of exponential functions <strong>in</strong> the sum of equation (8.15) as well as the <strong>re</strong>spective<br />

coe cients as;i a<strong>re</strong> di e<strong>re</strong>nt for each possible waveform. Furthermo<strong>re</strong>, some of<br />

the <strong>in</strong> nite <strong>in</strong>tegrals can be computed <strong>in</strong> di e<strong>re</strong>nt ways. So these modules a<strong>re</strong> typically<br />

implemented as waveform-dependent templates that need the mo<strong>re</strong> general part<br />

ss( ) cr( ) of the <strong>in</strong>tegrand as <strong>in</strong>put parameter. The they add the speci c functions<br />

P m<br />

i=1 as;i e j(gs;i( )+pr( )) and call the actual quadratu<strong>re</strong> rout<strong>in</strong>e with the appropriate<br />

options.<br />

The only exception to this scheme is the computation of the evanescent part of the<br />

<strong>in</strong>tegrals. As this <strong>in</strong>volves only a s<strong>in</strong>gle de nite <strong>in</strong>tegral, the<strong>re</strong> is no po<strong>in</strong>t <strong>in</strong> divid<strong>in</strong>g the<br />

<strong>in</strong>tegrands like we do for the <strong>in</strong> nite <strong>in</strong>tegrals <strong>in</strong> order to meet the formal <strong>re</strong>qui<strong>re</strong>ments<br />

of our quadratu<strong>re</strong> strategy. Hence the<strong>re</strong> is only one template for this task, and the<br />

whole <strong>in</strong>tegrand function is passed to this rout<strong>in</strong>e from the call<strong>in</strong>g module. Because all<br />

these rout<strong>in</strong>es a<strong>re</strong> highly specialised, they a<strong>re</strong> not made available to the user.<br />

Function modules a<strong>re</strong> the actual build<strong>in</strong>g blocks of the <strong>in</strong>tegrands. The<strong>re</strong> a<strong>re</strong> atomic<br />

functions describ<strong>in</strong>g the shape- and <strong>re</strong>gion-dependent coe cients ss( ), cr( )aswell as<br />

the constants Cs. These elements a<strong>re</strong> then comb<strong>in</strong>ed to form the <strong>in</strong>tegrands for the two<br />

<strong>in</strong>de nite transmissive parts and the evanescent section of the <strong>in</strong>tegration range. These<br />

functions, too, a<strong>re</strong> <strong>in</strong>ternal to the package and cannot be accessed by the user.<br />

For the triangular and Gaussian wave packet, the gradient of the wave function <strong>in</strong>side the<br />

tunnel is also of <strong>in</strong>te<strong>re</strong>st for the computation of the energy density and nally the energy<br />

velocity. These functions have not been <strong>in</strong>cluded <strong>in</strong> the general framework, <strong>in</strong> that they<br />

cannot be accessed from the same driver rout<strong>in</strong>e as the wave functions themselves. Although<br />

the gradient functions sha<strong>re</strong> the quadratu<strong>re</strong> modules of the pla<strong>in</strong> wave computation, they use<br />

separate driver and, of course, function modules.<br />

197


8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Know<strong>in</strong>g that we will end up with a plethora of very similar functions, we establish a consistent<br />

nam<strong>in</strong>g convention to avoid confusion and to <strong>re</strong>duce the risk of programm<strong>in</strong>g errors. All<br />

function names a<strong>re</strong> thus composed of dist<strong>in</strong>ct elements to dist<strong>in</strong>guish them from each other.<br />

A seve<strong>re</strong> obstacle to the legibility of the names is the fact that Mathematica permits no<br />

undersco<strong>re</strong>s with<strong>in</strong> a function's name because the undersco<strong>re</strong> is <strong>re</strong>served to designate patterns.<br />

So we have no possibility toseparate the tokens that make up a function name as <strong>in</strong> other<br />

programm<strong>in</strong>g languages.<br />

We start with the de nition of a number of tokens, each of which describes a part of the<br />

purpose the function is <strong>in</strong>tended for.<br />

Form 2 fRect, Tria, Gaussg denotes the shape of the <strong>in</strong>itial wavepacket and applies<br />

to all waveform-dependent functions.<br />

Region 2fInc, Ref, Out, Trans, Evan, Tung dist<strong>in</strong>guishes the <strong>re</strong>gions a function may<br />

be applicable to. Some modules depend on the di<strong>re</strong>ction of the wave propagation or<br />

whether the evanescent or the transmitted part of the <strong>in</strong>tegral is <strong>re</strong>garded. Others<br />

simply a<strong>re</strong> di e<strong>re</strong>nt <strong>in</strong>side and outside the barrier.<br />

FunctionType 2 fConst, Shape, Oscg denotes constants, shap<strong>in</strong>g functions for the<br />

spectra of the waves, and the oscillatory factors of the <strong>in</strong>tegrands.<br />

Polarity 2 fNULL, Pos, Negg is used to dist<strong>in</strong>guish <strong>in</strong>tegrands de ned for either the<br />

positive or negative <strong>re</strong>al axis. The empty element NULL applies to all functions that<br />

need no dist<strong>in</strong>ction between the two cases.<br />

Gradient 2fNULL, Gradg signi es whether the function is used for the computation of<br />

the wave function or its gradient. In the latter case, the name <strong>in</strong>cludes the additional<br />

token Grad.<br />

These tokens a<strong>re</strong> used to compose the various function modules. Which tokens a<strong>re</strong> actually<br />

comb<strong>in</strong>ed and what values a<strong>re</strong> mean<strong>in</strong>gful depends on the class of modules under consideration.<br />

Region-dependent factors have the form kRegionjFunctionTypek. For example,<br />

RefCoeff is the coe cient function for the <strong>re</strong> ected wave, TunOsc means the oscillatory<br />

factor of the <strong>in</strong>tegrands <strong>in</strong> the tunnel.<br />

Integrands and waveform-dependent factors have the mo<strong>re</strong> complicated structu<strong>re</strong><br />

kFormjGradientjfFunctionType, RegiongjPolarityk. Examples a<strong>re</strong> GaussConst for a<br />

constant factor, RectShapeNeg for a factor of an <strong>in</strong>tegrand and nally TriaGradEvan<br />

or GaussRefNeg for the enti<strong>re</strong> <strong>in</strong>tegrands.<br />

Quadratu<strong>re</strong> templates a<strong>re</strong> rather heterogenous and have the nam<strong>in</strong>g structu<strong>re</strong> kfNULL,<br />

FormgjfEvan, TransgjfNULL, Exact, TruncgjTempk, as for <strong>in</strong>stance TriaTransExactTemp<br />

or GaussEvanTruncTemp. The elements Exact and Trunc <strong>re</strong>fer to alternative comput<strong>in</strong>g<br />

strategies for the non-Gaussian and Gaussian wave packets, <strong>re</strong>spectively.<br />

198


8.2 Outl<strong>in</strong>e of the program structu<strong>re</strong><br />

Driver modules a<strong>re</strong> named accord<strong>in</strong>g to the scheme kPhijfNULL, Inc, Ref, Trans,<br />

Evangk, depend<strong>in</strong>g on whether the function depends on the <strong>re</strong>gion and the propagation<br />

di<strong>re</strong>ction or not.<br />

Two options, Shape and PPo<strong>in</strong>ts, a<strong>re</strong> provided to control the execution of the user accessible<br />

modules. The allowed values of the latter depend on the waveform.<br />

Shape 2fRect, Tria, Gaussg de nes the shape of the wave packet under consideration.<br />

The default value is Tria.<br />

PPo<strong>in</strong>ts 2 fApproximate, Zeros, Truncateg selects the way the <strong>in</strong>tegrals a<strong>re</strong> evaluated.<br />

Which one is applicable depends on the waveform. For triangular and <strong>re</strong>ctangular<br />

waves <strong>in</strong>side the tunnel, whe<strong>re</strong> the exponential functions have no polynomial arguments,<br />

the partition po<strong>in</strong>ts can be determ<strong>in</strong>ed either us<strong>in</strong>g a polynomial approximation or by<br />

comput<strong>in</strong>g the zeros of the <strong>in</strong>tegrand numerically. Outside the tunnel, the partition<br />

po<strong>in</strong>ts a<strong>re</strong> exactly the zeros, anyhow, and this option is ir<strong>re</strong>levant.<br />

For the Gaussian wave packet, on the contrary, it is possible not to use extrapolation<br />

but to compute the <strong>in</strong>tegrals di<strong>re</strong>ctly by truncat<strong>in</strong>g them at <strong>re</strong>asonably large abscissa<br />

values. This is possible both <strong>in</strong>side and outside the tunnel and also for the evanescent<br />

part of the wave. For the other shapes, the option is igno<strong>re</strong>d for the evaluation of the<br />

de nite <strong>in</strong>tegrals. The default value of this option is Approximate.<br />

Beside these two options all other options of the quadratu<strong>re</strong> function OscInt can be used.<br />

They take, however, no <strong>in</strong> uence on the evaluation of the modules <strong>in</strong> the package but a<strong>re</strong><br />

simply passed to the quadratu<strong>re</strong> rout<strong>in</strong>e.<br />

The list of arguments that have to be passed to the driver module comprises the spatial and<br />

time coord<strong>in</strong>ate X and T , the ratio between cut-o f<strong>re</strong>quency and the signal f<strong>re</strong>quency<br />

and the number of wavelengths k that the <strong>in</strong>itial wave packet spans. For Gaussian waves,<br />

the geometry parameter n is needed, otherwise an error message will be generated. For all<br />

other shapes, this additional parameter is not necessary and will be igno<strong>re</strong>d. Follow<strong>in</strong>g these<br />

arguments, options may be given, like <strong>in</strong> the examples below.<br />

Phi[x,t,w,k,Shape->Rect]<br />

PhiInc[x,t,w,k,n,Shape->Gauss,PPo<strong>in</strong>ts->Truncate]<br />

PhiRef[x,t,w,k,Shape->Tria,NSumTerms->30]<br />

PhiEvan[x,t,w,k,Shape->Rect,Work<strong>in</strong>gP<strong>re</strong>cision->25]<br />

PhiTrans[x,t,w,k,Shape->Tria,PPo<strong>in</strong>ts->Zeros]<br />

PhiGrad[x,t,w,k,n,Shape->Gauss]<br />

PhiGradTrans[x,t,w,k,Shape->Tria]<br />

PhiGradEvan[x,t,w,k,Shape->Tria]<br />

These examples a<strong>re</strong> a comp<strong>re</strong>hensive list of all available functions of the package.<br />

199


8.3 Implementation of the quadratu<strong>re</strong> modules<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Now that we have formulated all <strong>in</strong>tegrals <strong>in</strong> a way that suits our needs, we can write the<br />

quadratu<strong>re</strong> modules. S<strong>in</strong>ce they a<strong>re</strong> all based on the same structu<strong>re</strong>, we shall discuss he<strong>re</strong><br />

only the example of the <strong>re</strong>ctangular wave packet and a few functions that a<strong>re</strong> <strong>re</strong>lated to it. All<br />

other modules a<strong>re</strong> very similar <strong>in</strong> their structu<strong>re</strong> and can be understood without di culties<br />

from the list<strong>in</strong>g of the package <strong>in</strong> the appendix.<br />

The atomic function modules we use to de ne the <strong>in</strong>tegrands a<strong>re</strong> the transmission and <strong>re</strong>ection<br />

coe cients and the oscillatory factors of the <strong>in</strong>tegrands with<strong>in</strong> (TunOsc) and outside<br />

(OutOsc) the tunnel. In addition, we need the argument function of the oscillation <strong>in</strong>side<br />

the tunnel for the optional exact determ<strong>in</strong>ation of the <strong>in</strong>tegrand's zeros. These functions a<strong>re</strong><br />

<strong>in</strong>dependent of the waveform.<br />

TunCoeff[xi_] := 2 xi/(xi + Sqrt[xi^2-1]);<br />

RefCoeff[xi_] := (xi - Sqrt[xi^2-1])/(xi + Sqrt[xi^2-1]);<br />

TunOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + b xi + c Sqrt[xi^2 - 1] + d)];<br />

OutOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + (b+c) xi + d)];<br />

PhiArg[a_,b_,c_,d_][xi_] := a xi^2 + b xi + c Sqrt[xi^2 - 1] + d;<br />

Other basic functions like those describ<strong>in</strong>g the shape of the spectrum a<strong>re</strong> di e<strong>re</strong>nt for each<br />

waveform. Note that <strong>in</strong> the implementation we use the variable w for the parameter .<br />

RectConst[w_,k_] := I / (Sqrt[w] 2 Pi);<br />

RectShapePos[w_,xi_] := 1/(1-xi/Sqrt[w]);<br />

RectShapeNeg[w_,xi_] := 1/(1+xi/Sqrt[w]);<br />

With these build<strong>in</strong>g blocks we de ne the actual <strong>in</strong>tegrands that a<strong>re</strong> to be used <strong>in</strong>side the<br />

quadratu<strong>re</strong> rout<strong>in</strong>es. As we saw while formulat<strong>in</strong>g the <strong>re</strong>spective equations, we need two<br />

dist<strong>in</strong>ct functions for the positive and negative <strong>in</strong>de nite <strong>in</strong>tegrals. The <strong>in</strong>tegrand for the<br />

evanescent <strong>re</strong>gion is augmented by the coe cients and oscillatory factors that stem from the<br />

shape of the spectrum. The<strong>re</strong>fo<strong>re</strong> we can compute the <strong>in</strong>tegral with only one s<strong>in</strong>gle quadratu<strong>re</strong><br />

call. Ow<strong>in</strong>g to the particular structu<strong>re</strong> of the oscillatory factors, we need to pass only values<br />

for the parameters a and b to the quadratu<strong>re</strong> rout<strong>in</strong>e and must set b and d to zero <strong>in</strong> this<br />

case.<br />

RectTransPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * RectShapePos[w,xi] * TunOsc[a,b,c,d,xi];<br />

RectTransNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * RectShapeNeg[w,xi] * TunOsc[a,b,c,d,xi];<br />

RectEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RectTransPos[a,b,c,d,w,k][xi] *<br />

(-1 + Exp[I 2 (Pi k/Sqrt[w] xi - Pi k)]);<br />

The constant RectConst is not part of the <strong>in</strong>tegrand because it is useless to compute it<br />

with each function evaluation. Instead, the constant is multiplied to the <strong>re</strong>sult of the enti<strong>re</strong><br />

200


8.3 Implementation of the quadratu<strong>re</strong> modules<br />

quadratu<strong>re</strong>. The<strong>re</strong> a<strong>re</strong> of course other function declarations necessary for the other <strong>re</strong>gions<br />

(<strong>in</strong>cident and <strong>re</strong> ected waves). However, they look very much the same as the mentionend<br />

<strong>in</strong>tegrands and thus a<strong>re</strong> not given he<strong>re</strong>.<br />

The quadratu<strong>re</strong> modules consist of two parts, the computation of the parameters <strong>re</strong>qui<strong>re</strong>d for<br />

the function calls and the quadratu<strong>re</strong> itself. Note that the parameters b and d a<strong>re</strong> evaluated<br />

numerically. In the case of b this is necessary to ensu<strong>re</strong> the cor<strong>re</strong>ct operation of the functions<br />

QuadOffset and QuadZero that comprise conditional statements and a<strong>re</strong> not p<strong>re</strong>pa<strong>re</strong>d to<br />

take symbolical arguments. On the other hand, it improves the performance of the module<br />

because it <strong>re</strong>duces the computation time by up to ten percent (depend<strong>in</strong>g on the actual<br />

parameter values). For the same <strong>re</strong>ason, the o set parameter of the function enumerat<strong>in</strong>g the<br />

zeros is calculated befo<strong>re</strong> the quadratu<strong>re</strong> starts. To p<strong>re</strong>vent <strong>in</strong>accuracies due to the numerical<br />

evaluation of the parameters, they a<strong>re</strong> calculated to a p<strong>re</strong>cision two digits higher than the<br />

Work<strong>in</strong>gP<strong>re</strong>cision.<br />

Remark (Version peculiarities) In the orig<strong>in</strong>al version of the package the o set values<br />

we<strong>re</strong> not computed numerically as shown <strong>in</strong> the example given below. The function<br />

QuadOffset thus <strong>re</strong>turned <strong>re</strong>sults like Ceil<strong>in</strong>g[const/Pi] that we<strong>re</strong> left unevaluated<br />

because Mathematica t<strong>re</strong>ats Pi as a symbol rather than as a number. This worked<br />

well most of the time, however, for particular sets of <strong>in</strong>put parameters the quadratu<strong>re</strong><br />

function would never term<strong>in</strong>ate. This peculiar behaviour seem<strong>in</strong>gly depended on the<br />

softwa<strong>re</strong> version (2.2) but not on the platform, which was veri ed on PCs as well as on<br />

SUN and HP workstations. Neither version 2.1 nor the new version 3.0 caused similar<br />

troubles. The problem could be overcome only by chang<strong>in</strong>g the <strong>re</strong>sults of QuadOffset<br />

<strong>in</strong>to numbers, which <strong>in</strong> turn provides another slight performance improvement.<br />

The quadratu<strong>re</strong> function OscInt is called four times with the arguments of the <strong>in</strong>tegrands set<br />

to the appropriate values taken from (8.30). In this connection it is important to notice that<br />

ow<strong>in</strong>g to the uniform structu<strong>re</strong> of the <strong>in</strong>tegrand functions we need to pass only the name of the<br />

function to the template, the arguments a<strong>re</strong> then added <strong>in</strong>side the module. The<strong>re</strong>fo<strong>re</strong> we can<br />

use the template for all parts of the wave. The <strong>re</strong>sults of the <strong>in</strong>dividual function calls a<strong>re</strong> then<br />

summed up with the coe cients given <strong>in</strong> (8.30) and multiplied with the waveform-dependent<br />

constant.<br />

RectTransTemp[fpos_, fneg_, x_, t_, w_, k_, opts___Rule] :=<br />

Module[fa,b,c,d,o11,o12,o21,o22,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.foptsg/.Options[OscInt]g,<br />

a = -t;<br />

b = N[2 Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-2 Pi k,wp+2];<br />

o11 = N[QuadOffset[a, c,0],wp+2];<br />

o12 = N[QuadOffset[a,-c,0],wp+2];<br />

o21 = N[QuadOffset[a, b+c,d],wp+2];<br />

o22 = N[QuadOffset[a,-b-c,d],wp+2];<br />

(-(OscInt[fpos[a, 0, c,0,w,k],<br />

QuadZero[a, c,0,o11],1,FunctionType->ZeroList,opts] +<br />

201


8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

OscInt[fneg[a, 0,-c,0,w,k],<br />

QuadZero[a, -c,0,o12],1,FunctionType->ZeroList,opts]) +<br />

(OscInt[fpos[a, b, c,d,w,k],<br />

QuadZero[a, b +c,d,o21],1,FunctionType->ZeroList,opts] +<br />

OscInt[fneg[a,-b,-c,d,w,k],<br />

QuadZero[a,-b -c,d,o22],1,FunctionType->ZeroList,opts])<br />

) RectConst[w,k]<br />

];<br />

The second possibility to compute this <strong>in</strong>tegral is not to use the approximate partition<strong>in</strong>g<br />

strategy but employ the exact zeros for this purpose. S<strong>in</strong>ce we do not need this module for<br />

any other function than the one <strong>in</strong>side the tunnel, we need not write it as a template. So we<br />

pass only the parameters of the wave to the module but keep the names of the <strong>in</strong>tegrands<br />

xed. This time we call the quadratu<strong>re</strong> rout<strong>in</strong>e without a value for the option FunctionType<br />

because it defaults to Argument which is exactly what we want <strong>in</strong> order to compute the zeros<br />

numerically. The <strong>re</strong>st of the module is the same as the p<strong>re</strong>viously discussed version.<br />

RectTransExact[x_, t_, w_, k_, opts___Rule] :=<br />

Module[fa,b,c,d,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.foptsg/.Options[OscInt]g,<br />

a = -t;<br />

b = N[2 Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-2 Pi k,wp+2];<br />

(-(OscInt[RectTransPos[a, 0, c,0,w,k],<br />

PhiArg[a, 0, c,0],1,opts] +<br />

OscInt[RectTransNeg[a, 0,-c,0,w,k],<br />

PhiArg[a, 0,-c,0],1,opts]) +<br />

(OscInt[RectTransPos[a, b, c,d,w,k],<br />

PhiArg[a, b, c,d],1,opts] +<br />

OscInt[RectTransNeg[a,-b,-c,d,w,k],<br />

PhiArg[a,-b,-c,d],1,opts])<br />

) RectConst[w,k]<br />

];<br />

The template for the calculation of the evanescent part of the wave is di e<strong>re</strong>nt. First, we<br />

do not split the <strong>in</strong>tegrand and thus have only one call of the <strong>in</strong>tegration rout<strong>in</strong>e. Second,<br />

as we use the built-<strong>in</strong> Mathematica function NIntegrate, we must set a number of options<br />

accord<strong>in</strong>g to the values given by the user. As method for the evaluation of this de nite <strong>in</strong>tegral<br />

we choose aga<strong>in</strong> DoubleExponential. A third di e<strong>re</strong>nce is that this module can be used for<br />

all waveforms and the<strong>re</strong>fo<strong>re</strong> the waveform-dependent constant must be passed to the module<br />

as an argument.<br />

EvanTemp[f_, const_, opts___Rule] :=<br />

Module[fwp = Work<strong>in</strong>gP<strong>re</strong>cision/.foptsg/.Options[OscInt],<br />

ag = AccuracyGoal/. foptsg/.Options[OscInt],<br />

202


8.3 Implementation of the quadratu<strong>re</strong> modules<br />

pg = P<strong>re</strong>cisionGoal/. foptsg/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. foptsg/.Options[OscInt],<br />

ma = MaxRecursion/. foptsg/.Options[OscInt]g,<br />

NIntegrate[f[xi],fxi,-1,1g,<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma] const<br />

];<br />

The functions discussed so far a<strong>re</strong> <strong>in</strong>ternal only and not visible to the user | <strong>in</strong> contrast<br />

to the driver functions. These a<strong>re</strong> used to provide a consistent user <strong>in</strong>terface that is easy to<br />

control. For the transmitted part of the wave function <strong>in</strong> the tunnel, the module PhiTrans,<br />

for example, looks at the user speci ed values of the options Shape and PPo<strong>in</strong>ts or their<br />

default values, <strong>re</strong>spectively, to call the appropriate quadratu<strong>re</strong> module. Furthermo<strong>re</strong>, warn<strong>in</strong>g<br />

and error messages a<strong>re</strong> generated if an option has an unkown or <strong>in</strong>appropriate value. The<br />

argument n is a geometrical parameter that applies only to the Gaussian waveform. Thus<br />

we must provide an alternative way to call the module whe<strong>re</strong> n is not speci ed. This is the<br />

second rule given for the function name, which is applicable for <strong>re</strong>ctangular and triangular<br />

waves, but issues an error when called with Shape->Gauss. In the other cases, it calls the<br />

ma<strong>in</strong> function with n set to the dummy value one. The message bodies a<strong>re</strong> decla<strong>re</strong>d <strong>in</strong> the<br />

package header.<br />

PhiTrans[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[fsh = Shape/. foptsg/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.foptsg/.Options[Phi]g,<br />

Switch[sh,<br />

Rect,<br />

Switch[pt,<br />

Approximate, RectTransTemp[RectTransPos,<br />

RectTransNeg,x,t,w,k,opts],<br />

Zeros, RectTransExact[x,t,w,k,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

Tria,<br />

Switch[pt,<br />

Approximate, TriaTransTemp[TriaTransPos,<br />

TriaTransNeg,x,t,w,k,opts],<br />

Zeros, TriaTransExactTemp[TriaTransPos,<br />

TriaTransNeg,x,t,w,k,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, GaussTransTemp[GaussTransPos,<br />

GaussTransNeg,x,t,w,k,n,opts],<br />

Truncate, GaussTransTruncTemp[GaussTransPos,<br />

203


]<br />

];<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

GaussTransNeg,x,t,w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

PhiTrans[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.foptsg/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

PhiTrans[x,t,w,k,1,opts]];<br />

Remark (Input check<strong>in</strong>g) Apart from check<strong>in</strong>g the option values, the function provides<br />

other tests of the user's <strong>in</strong>put, too. The parameter n, for example must be g<strong>re</strong>ater than<br />

zero, which is exp<strong>re</strong>ssed with the conditional pattern match<strong>in</strong>g statement n ?Positive.<br />

The<strong>re</strong>fo<strong>re</strong> the function evaluates only if n satis es this <strong>re</strong>qui<strong>re</strong>ment. Otherwise no error<br />

message is generated, but the function is left unevaluated. The same is done with the<br />

additional arguments that match the pattern opts Rule only if they have the syntax of<br />

a rule, that is if they look like lhs->rhs. However, this does not encompass a validity<br />

check for the option names or their values.<br />

The modules PhiTrans and PhiEvan compute the wave function <strong>in</strong>side the barrier, PhiInc and<br />

PhiRef <strong>re</strong>turn the <strong>in</strong>cident and <strong>re</strong> ected wave outside the barrier. A still mo<strong>re</strong> comfortable<br />

driver is the function Phi that decides (based upon the spatial coord<strong>in</strong>ate given by the user)<br />

which module to call. For the boundary case X =0it takes the functions that a<strong>re</strong> de ned<br />

for the <strong>in</strong>terior of the barrier (which must, of course, yield the same <strong>re</strong>sults as those for the<br />

outside). This function also traps the somewhat <strong>in</strong>convenient case X = T = 0 for <strong>re</strong>ctangular<br />

waves and sets the <strong>re</strong>turn value to the exactly known <strong>re</strong>sult 1=2.<br />

Phi[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Which[t == 0 && x == 0 && Shape/.foptsg/.Options[Phi] === Rect, 0.5,<br />

t < 0, 0,<br />

x >= 0, PhiEvan[x,t,w,k,n,opts] + PhiTrans[x,t,w,k,n,opts],<br />

x < 0, PhiInc[x,t,w,k,n,opts] + PhiRef[x,t,w,k,n]];<br />

Phi[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.foptsg/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

Phi[x,t,w,k,1,opts]];<br />

All the driver functions deal<strong>in</strong>g with the wave function <strong>in</strong>side the barrier also exist <strong>in</strong> a version<br />

that computes the gradient of the wave function, namely PhiGradTrans, PhiGradEvan, and<br />

PhiGrad. They a<strong>re</strong> implemented only for triangular and Gaussian waves. For a <strong>re</strong>ctangular<br />

shape, the <strong>in</strong>tegral would diverge for the boundary and for T =0.<br />

204


8.4 Test of the package<br />

8.4 Test of the package<br />

We now have to verify that the functions <strong>in</strong> our package a<strong>re</strong> fault f<strong>re</strong>e and yield the desi<strong>re</strong>d<br />

<strong>re</strong>sults. The<strong>re</strong> a<strong>re</strong> th<strong>re</strong>e possibilities to do so. First we can check that the <strong>in</strong>itial values a<strong>re</strong><br />

cor<strong>re</strong>ct, namely that the <strong>in</strong>ital wave imp<strong>in</strong>g<strong>in</strong>g on the barrier is <strong>in</strong>deed <strong>re</strong>ctangular, triangular,<br />

or of Gaussian shape outside the barrier and zero <strong>in</strong>side the tunnel at time zero:<br />

(X; T ) T =0 =<br />

( 0<br />

if X 0<br />

0 if X > 0 :<br />

(8.41)<br />

Second, the boundary conditions must be satis ed and the wave function as well as its space<br />

derivatives must be cont<strong>in</strong>uous at the edge of the tunnel,<br />

lim<br />

X!,0<br />

(X; T )= lim<br />

X!+0<br />

(X; T) = (X; T)jX=0 ; 8T (8.42)<br />

@ (X; T ) @ (X; T)<br />

lim<br />

= lim<br />

=<br />

X!,0 @X X!+0 @X<br />

@ (X; T)<br />

@X jX=0 ; 8T (8.43)<br />

The third test is me<strong>re</strong>ly to verify the di e<strong>re</strong>nt implementations of the <strong>in</strong>de nite <strong>in</strong>tegration,<br />

because the <strong>re</strong>sults for all possible values of the option PPo<strong>in</strong>ts must be identical with<strong>in</strong> the<br />

numerical p<strong>re</strong>cision.<br />

8.4.1 Numerical tests<br />

To verify the boundary conditions, we chose a wave packet spann<strong>in</strong>g ten wavelengths with<br />

=0:1, set the time coord<strong>in</strong>ate to zero and compute a list of po<strong>in</strong>ts along the spatial axis<br />

to achieve a <strong>re</strong>asonable <strong>re</strong>solution. We beg<strong>in</strong> with a <strong>re</strong>ctangular and triangular wave. Note<br />

the use of the option Shape to set the appropriate waveform.<br />

Remark (Extract<strong>in</strong>g <strong>re</strong>al parts) S<strong>in</strong>ce the wave function is complex-valued, we plot<br />

only its <strong>re</strong>al part. This is done with the rule fx ,y g:>fx,Re[y]g that <strong>re</strong>places the<br />

second element of each po<strong>in</strong>t (which is essentially the wave function for the <strong>re</strong>spective<br />

x-coord<strong>in</strong>ate) with its <strong>re</strong>al part.<br />

Example 8.4.1<br />

In[1]:= t = 0;<br />

w = 0.1;<br />

k = 10;<br />

<strong>in</strong>itial<strong>re</strong>ct = Table[fx,Phi[x,t,w,k,Shape->Rect]g,fx,-230,10,1.25g];<br />

ListPlot[<strong>in</strong>itial<strong>re</strong>ct/.fx_,y_g:>fx,Re[y]g,PlotJo<strong>in</strong>ed->True];<br />

205


-200 -150 -100 -50<br />

Example 8.4.2<br />

1<br />

0.5<br />

-0.5<br />

-1<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

In[2]:= <strong>in</strong>itialtria = Table[fx,Phi[x,t,w,k,Shape->Tria]g,<br />

fx,-230,10,1.25g];<br />

ListPlot[<strong>in</strong>itialtria/.fx_,y_g:>fx,Re[y]g,PlotJo<strong>in</strong>ed->True];<br />

-200 -150 -100 -50<br />

1.5<br />

1<br />

0.5<br />

-0.5<br />

-1<br />

-1.5<br />

For the Gaussian wave, we need the additional parameter describ<strong>in</strong>g the variance of the<br />

probability distribution. We set it to one third of the distance between the maximum and<br />

the edge of the barrier.<br />

Example 8.4.3<br />

In[3]:= n = 6;<br />

<strong>in</strong>itialgauss = Table[fx,Phi[x,t,w,k,n,Shape->Gauss,<br />

PPo<strong>in</strong>ts->Truncate]g,fx,-230,10,1.25g];<br />

ListPlot[<strong>in</strong>itialgauss/.fx_,y_g:>fx,Re[y]g,PlotJo<strong>in</strong>ed->True,<br />

PlotRange->All];<br />

206


8.4 Test of the package<br />

-200 -150 -100 -50<br />

2<br />

1<br />

-1<br />

-2<br />

The examples show that the <strong>in</strong>itial conditions a<strong>re</strong> met. In order to test if the boundary<br />

conditions a<strong>re</strong> satis ed, we sum up the parts of the wave function on either side of the barrier<br />

edge and compute their di e<strong>re</strong>nce for a number of po<strong>in</strong>ts along the time axis. We then plot<br />

the absolute value j (,0;T), (+0;T)j of this di e<strong>re</strong>nce. Let us start with the triangular<br />

wave.<br />

Example 8.4.4<br />

In[4]:= x = 0;<br />

w = 0.2;<br />

k = 5;<br />

boundary = Table[ft,PhiInc[x,t,w,k,Shape->Tria]+<br />

PhiRef[x,t,w,k,Shape->Tria]-<br />

(PhiTrans[x,t,w,k,Shape->Tria]+<br />

PhiEvan[x,t,w,k,Shape->Tria])g,ft,0,20g];<br />

ListPlot[boundary/.fx_,y_g:>fx,Abs[y]g,PlotJo<strong>in</strong>ed->True,<br />

PlotRange->All];<br />

-7<br />

6. 10<br />

-7<br />

5. 10<br />

-7<br />

4. 10<br />

-7<br />

3. 10<br />

-7<br />

2. 10<br />

-7<br />

1. 10<br />

5 10 15 20<br />

207


8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

Th<strong>in</strong>gs a<strong>re</strong> di e<strong>re</strong>nt with the <strong>re</strong>ctangular wave packet, which is to be t<strong>re</strong>ated next. Exam<strong>in</strong><strong>in</strong>g<br />

this case, we see that the di e<strong>re</strong>nce at the orig<strong>in</strong> is ext<strong>re</strong>mely high. This is no surprise because<br />

at this po<strong>in</strong>t wemust not split the <strong>in</strong>tegrand like we did <strong>in</strong> (4.22). The <strong>re</strong>ason is that one of<br />

the <strong>re</strong>sult<strong>in</strong>g <strong>in</strong>tegrals does not converge, and this is why we ought to exclude the orig<strong>in</strong> from<br />

the computation (which <strong>in</strong> fact has been done for the mo<strong>re</strong> general driver function Phi).<br />

Example 8.4.5<br />

In[5]:= boundary = Table[ft,PhiInc[x,t,w,k,Shape->Rect]+<br />

PhiRef[x,t,w,k,Shape->Rect]-<br />

(PhiTrans[x,t,w,k,Shape->Rect]+<br />

PhiEvan[x,t,w,k,Shape->Rect])g,ft,0,20g];<br />

ListPlot[boundary/.fx_,y_g:>fx,Abs[y]g,PlotJo<strong>in</strong>ed->True,<br />

PlotRange->All];<br />

0.02<br />

0.015<br />

0.01<br />

0.005<br />

5 10 15 20<br />

The best congruence is found for the Gaussian wave, whe<strong>re</strong> we have virtually no di e<strong>re</strong>nce at<br />

all (this is why Mathematica chooses a wide plot range on the y-axis).<br />

Example 8.4.6<br />

In[6]:= n = 6;<br />

boundary = Table[ft,PhiInc[x,t,w,k,n,Shape->Gauss]+<br />

PhiRef[x,t,w,k,n,Shape->Gauss]-<br />

(PhiTrans[x,t,w,k,n,Shape->Gauss]+<br />

PhiEvan[x,t,w,k,n,Shape->Gauss])g,ft,0,20g];<br />

ListPlot[boundary/.fx_,y_g:>fx,Abs[y]g,PlotJo<strong>in</strong>ed->True,<br />

PlotRange->All];<br />

208


8.4 Test of the package<br />

0.4<br />

0.2<br />

-0.2<br />

-0.4<br />

5 10 15 20<br />

As for the di e<strong>re</strong>nt implementations of the evaluation of the <strong>in</strong>de nite <strong>in</strong>tegrals, we compa<strong>re</strong><br />

the <strong>re</strong>sults they yield as well as their comput<strong>in</strong>g time. Not surpris<strong>in</strong>gly, we nd that it takes<br />

longer to use the zeros as partition po<strong>in</strong>ts and that for Gaussian waves the truncation of the<br />

<strong>in</strong>tegrals is much faster.<br />

Example 8.4.7<br />

In[7]:= x = 1;<br />

t = 4;<br />

w = 0.1;<br />

k = 2;<br />

Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Tria]]]<br />

Out[7]= f21.696 Second, 0.0717104 - 0.0469158 Ig<br />

In[8]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Tria,PPo<strong>in</strong>ts->Zeros]]]<br />

Out[8]= f29.056 Second, 0.0717104 - 0.0469158 Ig<br />

In[9]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Rect]]]<br />

Out[9]= f18.785 Second, 0.0188638 - 0.264976 Ig<br />

In[10]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Rect,PPo<strong>in</strong>ts->Zeros]]]<br />

Out[10]= f22.903 Second, 0.0188638 - 0.264976 Ig<br />

In[11]:= n = 6;<br />

Tim<strong>in</strong>g[N[Phi[x,t,w,k,n,Shape->Gauss]]]<br />

209


Out[11]= f19.773 Second, -0.00034339 + 0.00140152 Ig<br />

In[12]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,n,Shape->Gauss,PPo<strong>in</strong>ts->Truncate]]]<br />

Out[12]= f4.998 Second, -0.00034339 + 0.00140152 Ig<br />

Example 8.4.8<br />

In[13]:= x = 1;<br />

t = 4;<br />

w = 0.1;<br />

k = 20;<br />

Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Tria]]]<br />

Out[13]= f284.017 Second, 0.00688579 - 0.00506005 Ig<br />

In[14]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Tria,PPo<strong>in</strong>ts->Zeros]]]<br />

Out[14]= f289.894 Second, 0.00688579 - 0.00506005 Ig<br />

In[15]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Rect]]]<br />

Out[15]= f218.381 Second, -0.00471446 - 0.282477 Ig<br />

In[16]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,Shape->Rect,PPo<strong>in</strong>ts->Zeros]]]<br />

Out[16]= f222.116 Second, -0.00471446 - 0.282477 Ig<br />

In[17]:= n = 6;<br />

Tim<strong>in</strong>g[N[Phi[x,t,w,k,n,Shape->Gauss]]]<br />

Out[17]= f138.467 Second, 0.0000199533 - 0.0000757562 Ig<br />

In[18]:= Tim<strong>in</strong>g[N[Phi[x,t,w,k,n,Shape->Gauss,PPo<strong>in</strong>ts->Truncate]]]<br />

Out[18]= f1.373 Second, 0.0000199533 - 0.0000757562 Ig<br />

210<br />

8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e


8.4 Test of the package<br />

8.4.2 Formal veri cation<br />

Apart from these numerical examples, the<strong>re</strong> is also a rather formal way to check the implementation.<br />

S<strong>in</strong>ce the wave function is <strong>in</strong> pr<strong>in</strong>ciple a Fourier <strong>in</strong>tegral over the space of wave<br />

numbers , each of the partial waves that make up the whole solution is itself a solution of<br />

Schrod<strong>in</strong>ger's equation. In addition, the partial waves on both sides of the tunnel for a given<br />

wave number must satisfy the boundary conditions. Thus we can use the <strong>in</strong>tegrands <strong>in</strong>stead<br />

of the enti<strong>re</strong> <strong>in</strong>tegrals for veri cation. Although the conditions must be met ir<strong>re</strong>spective of the<br />

wave number, we must, of course, keep <strong>in</strong> m<strong>in</strong>d that we have<strong>in</strong>troduced di e<strong>re</strong>nt de nitions<br />

for di e<strong>re</strong>nt ranges of .<br />

To perform the checks we must rst decla<strong>re</strong> the function modules from which the <strong>in</strong>tegrands<br />

a<strong>re</strong> composed because they a<strong>re</strong> not visible from outside the Mathematica package. As this is<br />

quite a lengthy <strong>in</strong>put, we only quote a few de nitions.<br />

In[19]:= (*-- Functions <strong>in</strong>dependent of the waveform --*)<br />

TunCoeff[xi_] := 2 xi/(xi + Sqrt[xi^2-1]);<br />

TunOsc[a_,b_,c_,d_,xi_] :=<br />

Exp[I (a xi^2 + b xi + c Sqrt[xi^2 - 1] + d)];<br />

RefCoeff[xi_] := (xi - Sqrt[xi^2-1])/(xi + Sqrt[xi^2-1]);<br />

OutOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + (b+c) xi + d)];<br />

(*-- Triangular Shape <strong>in</strong>side the tunnel --*)<br />

TriaConst[w_,k_] := Sqrt[3/w] / (2 k Pi^2);<br />

TriaShapePos[w_,xi_] := 1/(1-xi/Sqrt[w])^2;<br />

In[20]:= TriaShapeNeg[w_,xi_] := 1/(1+xi/Sqrt[w])^2;<br />

TriaTransPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * TriaShapePos[w,xi] * TunOsc[a,b,c,d,xi];<br />

TriaTransNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * TriaShapeNeg[w,xi] * TunOsc[a,b,c,d,xi];<br />

TriaEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaTransPos[a,b,c,d,w,k][xi] *<br />

(-1 + 2 Exp[I (Pi k/Sqrt[w] xi - Pi k)] -<br />

Exp[2 I (Pi k/Sqrt[w] xi - Pi k)]);<br />

(*-- and so on for the <strong>re</strong>st of the shapes --*)<br />

To start with the veri cation of the boundary conditions, we rst check that the waves<br />

<strong>in</strong>side and outside the tunnel have the same value and the same derivative at the edge. The<br />

arguments a to d a<strong>re</strong> the coe cients of the oscillatory factor. Accord<strong>in</strong>g to, for <strong>in</strong>stance, (8.30)<br />

we know that a = ,t and c = x for waves propagat<strong>in</strong>g <strong>in</strong> positive or negative di<strong>re</strong>ctions,<br />

211


8 Application of the quadratu<strong>re</strong> rout<strong>in</strong>e<br />

<strong>re</strong>spectively. The parameters b and d depend on the shape of the <strong>in</strong>itial waveform and a<strong>re</strong><br />

ir<strong>re</strong>levant as long as they a<strong>re</strong> constant for all terms.<br />

Example 8.4.9 Consider<strong>in</strong>g the evanescent parts of a <strong>re</strong>ctangular wave packet, we form the di e<strong>re</strong>nce<br />

of the waves outside and <strong>in</strong>side the barrier, <strong>re</strong>place the spatial coord<strong>in</strong>ate by its boundary value<br />

and simplify the <strong>re</strong>sult. As expected, we obta<strong>in</strong> zero <strong>in</strong>dependent of the parameters.<br />

Remark (Clear<strong>in</strong>g de nitions) To avoid <strong>in</strong>terfe<strong>re</strong>nce with the de nitions made for<br />

the numerical examples, we must rst clear the values we have assigned to x and t, for<br />

example by us<strong>in</strong>g the command Remove[x,t], prior to execut<strong>in</strong>g the symbolic operations<br />

<strong>in</strong> the sequel.<br />

In[21]:= (RectIncEvan[-t,b,x,d,w,k][xi] + RectRefEvan[-t,b,-x,d,w,k][xi] -<br />

RectEvan[-t,b,x,d,w,k][xi]) /.x->0 //Together<br />

Out[21]= 0<br />

We do the same with the derivatives at the edge of the barrier.<br />

In[22]:= (D[RectIncEvan[-t,b,x,d,w,k][xi] + RectRefEvan[-t,b,-x,d,w,k][xi] -<br />

RectEvan[-t,b,x,d,w,k][xi], x]) /.x->0 //Together<br />

Out[22]= 0<br />

We can carry out these tests for all parts of the <strong>in</strong>tegrands and for all di e<strong>re</strong>nt wavefoms to<br />

nd that the de nitions of the <strong>in</strong>tegrands meet the boundary conditions.<br />

To see whether the <strong>in</strong>tegrands <strong>re</strong>ally a<strong>re</strong> solutions of Schrod<strong>in</strong>ger's equation, we rst write<br />

the di e<strong>re</strong>ntial equation <strong>in</strong> our normalised variables, which is<br />

, @2 @<br />

, j<br />

@X2 @T<br />

+ V (X) =0 V =<br />

( 0 if X < 0<br />

1 if X 0<br />

: (8.44)<br />

For convenience, we de ne a di e<strong>re</strong>ntial operator that we can apply to the <strong>in</strong>tegrands. The<br />

potential V must be passed as an argument and depends on the <strong>re</strong>gion whe<strong>re</strong> the di e<strong>re</strong>ntial<br />

equation is evaluated.<br />

In[23]:= Schroed<strong>in</strong>ger[V_] := (-D[#,fx,2g] + V # - I D[#,t])&<br />

Example 8.4.10 Wecheck if the transmitted part of the <strong>re</strong>ctangular wave for positivewavenumbers<br />

satis es the di e<strong>re</strong>ntial equation <strong>in</strong>side the tunnel. Note that the potential is set to one. Collect<strong>in</strong>g<br />

all the terms of the <strong>re</strong>sult together we nd the expected answer.<br />

In[24]:= Schroed<strong>in</strong>ger[1][RectTransPos[-t,b,x,d,w,k][xi]] //Together<br />

212


8.4 Test of the package<br />

Out[24]= 0<br />

Example 8.4.11 Outside the barrier we must comb<strong>in</strong>e the <strong>in</strong>cident and <strong>re</strong> ected waves to obta<strong>in</strong><br />

the desi<strong>re</strong>d <strong>re</strong>sult. He<strong>re</strong> the potential vanishes of course.<br />

In[25]:= Schroed<strong>in</strong>ger[0][TriaIncNeg[-t,b,x,d,w,k][xi] +<br />

TriaRefNeg[-t,b,-x,d,w,k][xi]] //Together<br />

Out[25]= 0<br />

Perform<strong>in</strong>g these tests on all di e<strong>re</strong>nt functions, we see that they satisfy the <strong>re</strong>qui<strong>re</strong>ments.<br />

However, with this formal veri cation we can only ensu<strong>re</strong> the cor<strong>re</strong>ct implementation of the<br />

<strong>in</strong>tegrand functions. We cannot check if those parts of the <strong>in</strong>tegrals that comprise several<br />

dist<strong>in</strong>ct terms (i. e. all transmitted portions of the waves) a<strong>re</strong> put together without error. Nor<br />

can we prove the cor<strong>re</strong>ctness of the quadratu<strong>re</strong> modules di<strong>re</strong>ctly. This must be done with<br />

numerical experiments like the ones above. But tak<strong>in</strong>g together the <strong>re</strong>sults of both formal<br />

and numerical veri cation, we conclude that the implementation is right.<br />

213


Appendix A<br />

Mathematica packages<br />

A Mathematica packages<br />

The appendix conta<strong>in</strong>s source code list<strong>in</strong>gs of the most important Mathematica packages that<br />

we<strong>re</strong> needed to carry out the numerical evaluation of the wave <strong>in</strong>tegrals <strong>in</strong> the rst part of<br />

this work. The last section describes a number of utilities that made life easier with the<br />

th<strong>re</strong>e-dimensional pictu<strong>re</strong>s of propagat<strong>in</strong>g waves.<br />

A.1 Numerical quadratu<strong>re</strong><br />

The package OscInt was written to compute semi-<strong>in</strong> nite <strong>in</strong>tegrals of strongly oscillat<strong>in</strong>g<br />

<strong>in</strong>tegrands. The user <strong>in</strong>terface and implementation a<strong>re</strong> described <strong>in</strong> detail <strong>in</strong> chapter 7.<br />

(* Copyright: Copyright 1997, Institute of Computertechnology, *)<br />

(* Vienna University of Technology *)<br />

(*:Version: Mathematica 2.2.3 *)<br />

(*:Title: OscInt *)<br />

(*:Revision: 1.2 *)<br />

(*:Author: Thilo Sauter *)<br />

(*:Keywords: Semi-<strong>in</strong>f<strong>in</strong>ite Quadratu<strong>re</strong>, Very Oscillat<strong>in</strong>g Integrands,<br />

Series Acceleration *)<br />

(*:Requi<strong>re</strong>ments: None. *)<br />

(*:Warn<strong>in</strong>gs: None so far *)<br />

(*:Summary: This package provides a set of functions for the numerical<br />

quadratu<strong>re</strong> of univariate oscillat<strong>in</strong>g <strong>in</strong>tegrands over the<br />

semi-<strong>in</strong>f<strong>in</strong>te range {a,Inf<strong>in</strong>ity}. The oscillations of the<br />

<strong>in</strong>tegrand may be ir<strong>re</strong>gular and need not show an ultimately<br />

constant period. The <strong>in</strong>tegrand must be of the form<br />

f(x) u(g(x)) with u(x) be<strong>in</strong>g S<strong>in</strong>[x], Cos[x] or - mo<strong>re</strong><br />

generally - Exp[I x]. The argument function g(x) must have<br />

an asymptotic expansion at <strong>in</strong>f<strong>in</strong>ity.<br />

*)<br />

(*:History: 27-12-1996 ZerosInBetween extended to cope with special cases<br />

25-01-1997 PartInt, PartitionTable added, OscInt <strong>re</strong>written<br />

24-03-1997 OscIntControlled, PartitionOffs, PartitionPo<strong>in</strong>ts<br />

added to allow accuracy control<br />

25-03-1997 Check of the limit of zerof[x] added to functions<br />

PartitionTable and PartitionPo<strong>in</strong>ts to detect<br />

argument functions with f<strong>in</strong>ite limit (and thus<br />

f<strong>in</strong>ite set of zeros)<br />

*)<br />

Beg<strong>in</strong>Package["OscInt`"]<br />

PolynomialDeg<strong>re</strong>e::usage =<br />

"PolynomialDeg<strong>re</strong>e[f,x] <strong>re</strong>turns the deg<strong>re</strong>e of the polynomial<br />

asymptotic expansion of f <strong>re</strong>gard<strong>in</strong>g the variable x at Inf<strong>in</strong>ity. If f<br />

is a pu<strong>re</strong> function or is def<strong>in</strong>ed as f[...][x] then x need not be<br />

specified."<br />

214


A.1 Numerical quadratu<strong>re</strong><br />

AsymptoticExpand::usage =<br />

"AsymptoticExpand[f,x] <strong>re</strong>turns the polynomial part of the<br />

asymptotic expansion of f <strong>re</strong>gard<strong>in</strong>g the variable x at Inf<strong>in</strong>ity. If f<br />

is a pu<strong>re</strong> function or is def<strong>in</strong>ed as f[...][x] then x need not be<br />

specified."<br />

ApproxLimGeneric::usage =<br />

"ApproxLimGeneric[f,goal] is used to give a figu<strong>re</strong> of merit for the quality<br />

of a polynomial approximation to an argument f of a circular function. It <strong>re</strong>turns<br />

the abscissa value whe<strong>re</strong> the ratio of the circular velocity of the approximation<br />

polynomial and the approximation error, <strong>re</strong>spectively, <strong>re</strong>aches a given goal (the<br />

larger the better). This is equivalent to the number of zeros the approximat<strong>in</strong>g<br />

function has with<strong>in</strong> one half period of the f<strong>re</strong>quency def<strong>in</strong>ed by the diffe<strong>re</strong>nce<br />

between f and its approximation."<br />

ApproxLimQuad::usage =<br />

"ApproxLimHyp[a,b,c,goal] is a special case of ApproxLimGeneric for the<br />

function a x^2 + b Sqrt[x^2 + c] + d and its approximation a x^2 + b x + d."<br />

ApproxLimL<strong>in</strong>ear::usage =<br />

"ApproxLimHyp[a,b,c,goal] is a special case of ApproxLimGeneric for the<br />

function a x + b Sqrt[x^2 + c] + d and its approximation (a + b) x + d."<br />

QuadZero::usage =<br />

"QuadZero[a,b,c,offs][k] <strong>re</strong>turns the k-th root of the quadratic equation<br />

a x^2 + b x + c == (k+offs)*Pi for ascend<strong>in</strong>g x (the branch to the right of the<br />

ext<strong>re</strong>mum). The parameter offs has to be determ<strong>in</strong>ed such that the 0-th solution is<br />

al<strong>re</strong>ady larger than the abscissa value of the ext<strong>re</strong>mum (see QuadOffset)."<br />

QuadOffset::usage =<br />

"QuadOffset[a,b,c] <strong>re</strong>turns the offset value used <strong>in</strong> QuadZero, which is<br />

essentially the ord<strong>in</strong>ate value of the ext<strong>re</strong>mum divided by Pi."<br />

HypZero::usage =<br />

"HypZero[a,b,c,d,offs][k] <strong>re</strong>turns the k-th root of the quadratic equation<br />

(a+b) x + d == (k+offs)*Pi, which is the polynomial approximation of<br />

a x + b Sqrt[x^2 + c] + d at Inf<strong>in</strong>ity for ascend<strong>in</strong>g x (the branch to the right of the<br />

ext<strong>re</strong>mum). The parameter offs has to be determ<strong>in</strong>ed such that the 0-th solution is<br />

al<strong>re</strong>ady larger than the abscissa value of the ext<strong>re</strong>mum (see HypOffset)."<br />

HypOffset::usage =<br />

"HypOffset[a,b,c,d] <strong>re</strong>turns the offset value used <strong>in</strong> HypZero, which is<br />

essentially the ord<strong>in</strong>ate value of the ext<strong>re</strong>mum or bend<strong>in</strong>g po<strong>in</strong>t divided by Pi."<br />

HypZeroExact::usage =<br />

"HypZeroExact[a,b,c,d,offs][k] <strong>re</strong>turns the k-th root of the equation<br />

a x + b Sqrt[x^2 + c] + d == (k+offs)*Pi for ascend<strong>in</strong>g x (the branch to the right of the<br />

ext<strong>re</strong>mum). The parameter offs has to be determ<strong>in</strong>ed such that the 0-th solution is<br />

al<strong>re</strong>ady larger than the abscissa value of the ext<strong>re</strong>mum (see HypOffsetExact)."<br />

HypOffsetExact::usage =<br />

"HypOffsetExact[a,b,c,d] <strong>re</strong>turns the offset value used <strong>in</strong> HypZeroExact, which is <strong>in</strong><br />

pr<strong>in</strong>ciple the ord<strong>in</strong>ate value of the ext<strong>re</strong>mum or bend<strong>in</strong>g po<strong>in</strong>t divided by Pi."<br />

ZerosInBetween::usage =<br />

"ZerosInBetween[f,a,b] <strong>re</strong>turns the number of solutions of the function<br />

g[x] == k*Pi for x <strong>in</strong> (a,b). The function f[k] is <strong>re</strong>qui<strong>re</strong>d to give the k-th<br />

solution of g[x] == k*Pi. If f[0] lies <strong>in</strong> the <strong>in</strong>terval (a,b) only the part<br />

to the right of f[0] is <strong>re</strong>garded. With this function, the parameter NSumTerms<br />

of NSum can be determ<strong>in</strong>ed."<br />

HypApproxError::usage =<br />

"HypApproxError[b,c,goal] <strong>re</strong>turns the abscissa value whe<strong>re</strong> the error between<br />

the function b*Sqrt[x^2 + c] and its approximation b*x <strong>re</strong>aches a given value."<br />

PartInt::usage =<br />

"PartInt[f,partpo<strong>in</strong>ts,a,(opts)] computes the <strong>in</strong>def<strong>in</strong>ite <strong>in</strong>tegral of a function<br />

f over the range (a,Inf<strong>in</strong>ity) us<strong>in</strong>g a partition-extrapolation method when the<br />

partition po<strong>in</strong>ts partpo<strong>in</strong>ts a<strong>re</strong> explicitly given. The first <strong>in</strong>terval (a,p1) is<br />

obta<strong>in</strong>ed us<strong>in</strong>g the double-exponential method, the others by a Gauss-Kronrod formula.<br />

The partial <strong>in</strong>tegrals between the partition po<strong>in</strong>ts a<strong>re</strong> summed up and passed to<br />

SequenceLimit to determ<strong>in</strong>e the value of the <strong>in</strong>tegral. If the first member of the<br />

list of partition po<strong>in</strong>ts <strong>in</strong> Inf<strong>in</strong>ity, the no extrapolation is performed, and the<br />

<strong>in</strong>tegral is enti<strong>re</strong>ly calculated with the double-exponential rule. Note that the<br />

partition po<strong>in</strong>ts must be of ascend<strong>in</strong>g order and chosen such that extrapolation<br />

is possible (<strong>in</strong> case of an oscillat<strong>in</strong>g <strong>in</strong>tegrand, they must be to the right of<br />

the rightmost ext<strong>re</strong>mum or saddle po<strong>in</strong>t of the argument function)."<br />

PartitionTable::usage =<br />

"PartitionTable[f,a,n,(opts)] <strong>re</strong>turns a list of partition po<strong>in</strong>ts suitable for<br />

PartInt. Depend<strong>in</strong>g on the option FunctionType the function f is <strong>in</strong>terp<strong>re</strong>ted either<br />

as argument function of S<strong>in</strong>[f[x]] or as a function giv<strong>in</strong>g the k-th zero of S<strong>in</strong>[arg[x]].<br />

In the first case, the first n solutions of f[x] == K*Pi to the right of the lower<br />

215


<strong>in</strong>tegration limit a or the rightmost ext<strong>re</strong>mum of f, whichever is g<strong>re</strong>ater, a<strong>re</strong><br />

calculated. In the second case, the first n function values of f to the right of a<br />

a<strong>re</strong> sought. Note that <strong>in</strong> this case, f is supposed to <strong>re</strong>turn values only to the<br />

right of the rightmost ext<strong>re</strong>mum. In any case, f must be of the form f[...][x]."<br />

OscInt::usage =<br />

"OscInt[f,fpartition,a,(opts)] <strong>re</strong>turns the univariate <strong>in</strong>tegral of an oscillat<strong>in</strong>g<br />

function f over the range (a,Inf<strong>in</strong>ity) by means of a partition extrapolation strategy.<br />

The function f is <strong>re</strong>qui<strong>re</strong>d to of the form f[...][x], x be<strong>in</strong>g the<br />

<strong>in</strong>tegration variable. Quadratu<strong>re</strong> is carried out between successive partition po<strong>in</strong>ts<br />

def<strong>in</strong>ed by the function fpartition[...][x] that either <strong>re</strong>turns the zeros of the<br />

<strong>in</strong>tegrand or the argument of the oscillat<strong>in</strong>g part. The sequence of partition<br />

po<strong>in</strong>ts is computed by the function PartitionTable, the actual <strong>in</strong>tegration is carried<br />

out with the function PartInt. The first partition po<strong>in</strong>t can be given explicitly<br />

with OscInt[f,fpartition,{a,a0},(opts)]. The options a<strong>re</strong> passed to PartitionTable and<br />

PartInt."<br />

OscIntControlled::usage =<br />

"OscIntControlled[f,fpartition,a,(opts)] works like OscInt with the additional<br />

featu<strong>re</strong> that the accuracy of the <strong>re</strong>sult may be specified <strong>in</strong> advance."<br />

FunctionType::usage =<br />

"FunctionType is an option for PartitionTable (and OscInt) that specifies how<br />

the function passed to PartitionTable is to be <strong>in</strong>terp<strong>re</strong>ted.<br />

FunctionType -> ZeroList means that the function <strong>re</strong>turns consecutive zeros of the<br />

<strong>in</strong>tegrand. FunctionType -> S<strong>in</strong>Argument or CosArgument means that it is the argument<br />

function of the <strong>re</strong>spective cyclic function."<br />

ZeroList::usage =<br />

"ZeroList is a value to FunctionType."<br />

S<strong>in</strong>Argument::usage = CosArgument::usage =<br />

"Argument is a value to FunctionType."<br />

AccuracyControl::usage =<br />

"AccuracyControl specifies whether the accuracy of the <strong>re</strong>sult of OscInt should<br />

be controlled. With AccuracyControl -> n, at most MaxIterations a<strong>re</strong> performed<br />

to achieve the number of n cor<strong>re</strong>ct digits. With AccuracyControl -> False, no<br />

check<strong>in</strong>g is done. The control loop uses the built-<strong>in</strong> function Accuracy."<br />

IterationLength::usage =<br />

"IterationLength specifies the number of terms that a<strong>re</strong> added to the sequence<br />

of partial sums with each iteration."<br />

ApproxLimGeneric::nounique =<br />

"The<strong>re</strong> a<strong>re</strong> `1` <strong>re</strong>al solutions `2`, he<strong>re</strong> is the largest.";<br />

ApproxLimGeneric::nosolution =<br />

"The<strong>re</strong> is no positive <strong>re</strong>al valued solution.";<br />

QuadZero::noquad = "Function is not a quadratic polynomial";<br />

ApproxLimQuad::noapprox = "No approximation needed";<br />

ApproxLimL<strong>in</strong>ear::noosc = "Function not <strong>in</strong>f<strong>in</strong>itely oscillat<strong>in</strong>g";<br />

ZerosInBetween::outside = "Function has no zeros <strong>in</strong>side the <strong>in</strong>terval";<br />

OscInt::lowfirst =<br />

"First partition po<strong>in</strong>t is smaller than lower limit, only<br />

`1` po<strong>in</strong>ts a<strong>re</strong> g<strong>re</strong>ater. This may cause problems.";<br />

OscInt::badparam =<br />

"\"`1`\" is no valid parameter for AccuracyControl.";<br />

OscInt::accfail =<br />

"OscInt failed to achieve the desi<strong>re</strong>d accuracy after `1` iterations. Inc<strong>re</strong>ase<br />

MaxIterations or lower AccuracyControl.";<br />

PartitionTable::badparam =<br />

"\"`1`\" is no valid parameter for FunctionType.";<br />

Beg<strong>in</strong>["`Private`"]<br />

Options[OscInt] = {Work<strong>in</strong>gP<strong>re</strong>cision->$Mach<strong>in</strong>eP<strong>re</strong>cision,<br />

AccuracyGoal->Inf<strong>in</strong>ity,<br />

P<strong>re</strong>cisionGoal->Automatic,<br />

M<strong>in</strong>Recursion->0,<br />

MaxRecursion->20,<br />

NSumTerms->10,<br />

NSumExtraTerms->12,<br />

216<br />

A Mathematica packages


A.1 Numerical quadratu<strong>re</strong><br />

WynnDeg<strong>re</strong>e->Inf<strong>in</strong>ity,<br />

MaxIterations->5,<br />

AccuracyControl->False,<br />

IterationLength->2};<br />

Options[PartInt] = Options[OscInt];<br />

Options[PartitionTable] =<br />

{Work<strong>in</strong>gP<strong>re</strong>cision->$Mach<strong>in</strong>eP<strong>re</strong>cision,<br />

AccuracyGoal->Automatic,<br />

MaxRecursion->20,<br />

FunctionType->S<strong>in</strong>Argument};<br />

PolynomialDeg<strong>re</strong>e[f_] :=<br />

Length[CoefficientList[Series[f[x],{x,Inf<strong>in</strong>ity,1}],x]]-1;<br />

PolynomialDeg<strong>re</strong>e[f_,x_Symbol] :=<br />

Length[CoefficientList[Series[f,{x,Inf<strong>in</strong>ity,1}],x]]-1;<br />

AsymptoticExpand[f_] :=<br />

Normal[Series[f[x],{x,Inf<strong>in</strong>ity,PolynomialDeg<strong>re</strong>e[f]}]];<br />

AsymptoticExpand[f_,x_Symbol] :=<br />

Normal[Series[f,{x,Inf<strong>in</strong>ity,PolynomialDeg<strong>re</strong>e[f,x]}]];<br />

ApproxLimGeneric[f_,goal_] :=<br />

Module[{fapprox,<strong>re</strong>spos,<strong>re</strong>sneg,<strong>re</strong>s},<br />

fapprox = AsymptoticExpand[f];<br />

<strong>re</strong>spos = Select[<br />

NSolve[Evaluate[<br />

D[fapprox,x] /<br />

D[fapprox - f[x],x]] == goal,x],(Im[x/.#] == 0)&];<br />

<strong>re</strong>sneg = Select[<br />

NSolve[Evaluate[<br />

D[fapprox,x] /<br />

D[fapprox - f[x],x]] == -goal,x],(Im[x/.#] == 0)&];<br />

<strong>re</strong>s = Select[Jo<strong>in</strong>[<strong>re</strong>spos,<strong>re</strong>sneg],Positive[x/.#]&];<br />

If[Length[<strong>re</strong>s] == 0,<br />

Message[ApproxLimGeneric::nosolution],<br />

If[Length[<strong>re</strong>s] > 1,<br />

Message[ApproxLimGeneric::nounique,Length[<strong>re</strong>s],<strong>re</strong>s]];<br />

x/.Last[<strong>re</strong>s]]<br />

]<br />

ApproxLimQuad[a_,b_,c_,d_,goal_] := Which[<br />

a != 0 && c != 0 && d != 0,<br />

x/.Flatten[Last[<br />

NSolve[(2 a x + b + c) /<br />

(c (1-x/Sqrt[x^2+d])) == Sign[a c d] goal,x]]],<br />

a == 0 && c != 0 && d != 0,<br />

ApproxLimL<strong>in</strong>ear[b,c,d,goal],<br />

True,<br />

Message[ApproxLimQuad::noapprox];0]<br />

ApproxLimL<strong>in</strong>ear[a_,b_,c_,goal_] := Which[<br />

a + b == 0,<br />

Message[ApproxLimL<strong>in</strong>ear::noosc];0,<br />

Abs[(a + b) / b] > goal, 0,<br />

True,<br />

x/.Flatten[<br />

NSolve[(a + b) /<br />

(b (1-x/Sqrt[x^2+c])) == Sign[(a+b) b c] goal,x]]];<br />

QuadOffset[a_,b_,c_] := Which[<br />

a == 0, Ceil<strong>in</strong>g[c/Pi],<br />

a > 0,-Floor[(b^2 - 4 a c)/(4 a Pi)],<br />

True, -Ceil<strong>in</strong>g[(b^2 - 4 a c)/(4 a Pi)]];<br />

QuadZero[a_,b_,c_,o_][k_] := Which[<br />

a > 0, (-b+Sqrt[b^2-4 a (c-( k+o) Pi)])/(2 a),<br />

a < 0, (-b-Sqrt[b^2-4 a (c-(-k+o) Pi)])/(2 a),<br />

a == 0 && b > 0, (( k+o) Pi - c)/b,<br />

a == 0 && b < 0, ((-k+o) Pi - c)/b,<br />

True, k];<br />

HypApproxError[b_,c_,goal_] :=<br />

x/.Last[Solve[b (Sqrt[x^2 + c]-x) == Sign[c b] goal,x]];<br />

ZerosInBetween[zerof_,a_,b_] :=<br />

Module[{m<strong>in</strong>=M<strong>in</strong>[a,b], max=Max[a,b]},<br />

Which[N[zerof[0]] < m<strong>in</strong>,<br />

Ceil<strong>in</strong>g[i/.F<strong>in</strong>dRoot[zerof[i] == max,{i,0,1}]] -<br />

Ceil<strong>in</strong>g[i/.F<strong>in</strong>dRoot[zerof[i] == m<strong>in</strong>,{i,0,1}]],<br />

217


];<br />

N[zerof[0]] < max,<br />

Ceil<strong>in</strong>g[i/.F<strong>in</strong>dRoot[zerof[i] == max,{i,0,1}]],<br />

True,<br />

Message[ZerosInBetween::outside];0]<br />

HypOffset[a_,b_,c_,d_] := Which[<br />

(Abs[b] > Abs[a]) && (c >= 0) && (a+b) > 0,<br />

Ceil<strong>in</strong>g[((a+b) (-Sign[a b] Sqrt[a^2 c/(b^2-a^2)]) + d)/Pi],<br />

(Abs[b] > Abs[a]) && (c >= 0) && (a+b) < 0,<br />

Floor[((a+b) (-Sign[a b] Sqrt[a^2 c/(b^2-a^2)]) + d)/Pi],<br />

(a+b) > 0, Ceil<strong>in</strong>g[d/Pi],<br />

(a+b) < 0, Floor[d/Pi],<br />

True, 0];<br />

HypZero[a_,b_,c_,d_,o_][k_] := Which[<br />

(a+b) > 0, (( k + o) Pi - d)/(a + b),<br />

(a+b) < 0, ((-k + o) Pi - d)/(a + b),<br />

True, Inf<strong>in</strong>ity];<br />

HypOffsetExact[a_,b_,c_,d_] := Which[<br />

(Abs[b] > Abs[a]) && (c >= 0) && (a+b) > 0,<br />

Ceil<strong>in</strong>g[((a x + b Sqrt[x^2+c] + d)/Pi)/.<br />

x->(-Sign[a b] Sqrt[a^2 c/(b^2-a^2)])],<br />

(Abs[b] > Abs[a]) && (c >= 0) && (a+b) < 0,<br />

Floor[((a x + b Sqrt[x^2+c] + d)/Pi)/.<br />

x->(-Sign[a b] Sqrt[a^2 c/(b^2-a^2)])],<br />

(a+b) > 0, Ceil<strong>in</strong>g[(b Sqrt[c] + d)/Pi],<br />

(a+b) < 0, Floor[(b Sqrt[c] + d)/Pi],<br />

True, 0];<br />

HypZeroExact[a_,b_,c_,d_,o_][k_] := Which[<br />

(a+b) > 0 && (Abs[b] > Abs[a]),<br />

(a (( k+o) Pi-d) - Sqrt[b^2 ((a^2-b^2) c + (( k+o) Pi-d)^2)])/(a^2-b^2),<br />

(a+b) < 0 && (Abs[b] > Abs[a]),<br />

(a ((-k+o) Pi-d) - Sqrt[b^2 ((a^2-b^2) c + ((-k+o) Pi-d)^2)])/(a^2-b^2),<br />

(a+b) < 0 && (Abs[b] < Abs[a]),<br />

(a ((-k+o) Pi-d) - Sign[a b] Sqrt[b^2 ((a^2-b^2) c + ((-k+o) Pi-d)^2)])/(a^2-b^2),<br />

(a+b) > 0 && (Abs[b] < Abs[a]),<br />

(a (( k+o) Pi-d) - Sign[a b] Sqrt[b^2 ((a^2-b^2) c + (( k+o) Pi-d)^2)])/(a^2-b^2),<br />

a == b && N[(2 a ((k+o) Pi - d))] != 0,<br />

(((k+o) Pi - d)^2 - a^2 c)/(2 a ((k+o) Pi - d)),<br />

True, Inf<strong>in</strong>ity];<br />

PartInt[f_, zerotab_List, a_, opts___Rule] :=<br />

Module[{zeros = zerotab, seqtab,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

ma = MaxRecursion/. {opts}/.Options[OscInt],<br />

nt = NSumTerms/. {opts}/.Options[OscInt],<br />

ne = NSumExtraTerms/. {opts}/.Options[OscInt],<br />

wd = WynnDeg<strong>re</strong>e/. {opts}/.Options[OscInt]},<br />

If[N[zeros[[1]]] < N[a],<br />

zeros = Select[zeros,(N[#] >= a)&];<br />

Message[OscInt::lowfirst,Length[zeros]]];<br />

If[N[zeros[[1]]] == N[a], 0,<br />

NIntegrate[f[x],{x,a,zeros[[1]]},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma]] +<br />

If[N[zeros[[1]]] == Inf<strong>in</strong>ity, 0,<br />

seqtab = Table[NIntegrate[f[x],{x,zeros[[i]],zeros[[i+1]]},<br />

Method->GaussKronrod,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma],<br />

{i,1,Length[zeros]-1}];<br />

SequenceLimit[Take[FoldList[Plus,0,seqtab],{nt+1,nt+ne}],<br />

WynnDeg<strong>re</strong>e->wd]]<br />

];<br />

PartitionTable[zerof_, a_, n_, opts___Rule] :=<br />

Module[{ofs=0,x,xoffs,koffs,ext<strong>re</strong>ma,cfoffs=0,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[PartitionTable],<br />

ag = AccuracyGoal/. {opts}/.Options[PartitionTable],<br />

ma = MaxRecursion/. {opts}/.Options[PartitionTable],<br />

ft = FunctionType/. {opts}/.Options[PartitionTable]},<br />

218<br />

A Mathematica packages


A.1 Numerical quadratu<strong>re</strong><br />

Which[ft === ZeroList,<br />

If[N[zerof[0]] < N[a],<br />

ofs = Ceil<strong>in</strong>g[x/.F<strong>in</strong>dRoot[zerof[x] == a,{x,0,1},<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

MaxIterations->ma]]];<br />

Table[zerof[x],{x,ofs,ofs+n-1}],<br />

ft === S<strong>in</strong>Argument || ft === CosArgument,<br />

If[ft === CosArgument, cfoffs=1/2];<br />

ext<strong>re</strong>ma = Select[NSolve[zerof'[x] == 0,x,Work<strong>in</strong>gP<strong>re</strong>cision->wp],<br />

(Im[x/.#] == 0)&];<br />

xoffs = If[Length[ext<strong>re</strong>ma] == 0, a,<br />

Max[x/.Last[ext<strong>re</strong>ma],a]];<br />

Which[NumberQ[Limit[zerof[x],x->Inf<strong>in</strong>ity]],<br />

Table[Inf<strong>in</strong>ity,{k,0,n-1}],<br />

N[zerof'[xoffs+1]] > 0,<br />

koffs = N[Ceil<strong>in</strong>g[zerof[xoffs]/Pi],2 wp];<br />

Re[Table[x/.F<strong>in</strong>dRoot[zerof[x] == (k+koffs+cfoffs) Pi,<br />

{x,xoffs,xoffs+1},<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

MaxIterations->ma],<br />

{k,0,n-1}]],<br />

True,<br />

koffs = N[Floor[zerof[xoffs]/Pi],2 wp];<br />

Re[Table[x/.F<strong>in</strong>dRoot[zerof[x] == (-k+koffs-cfoffs) Pi,<br />

{x,xoffs,xoffs+1},<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

MaxIterations->ma],<br />

{k,0,n-1}]]<br />

],<br />

True,<br />

Message[PartitionTable::badparam,ft]]<br />

];<br />

PartitionOffs[zerof_, a_, opts___Rule] :=<br />

Module[{x,ext<strong>re</strong>ma,cfoffs=0,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[PartitionTable],<br />

ag = AccuracyGoal/. {opts}/.Options[PartitionTable],<br />

ma = MaxRecursion/. {opts}/.Options[PartitionTable],<br />

ft = FunctionType/. {opts}/.Options[PartitionTable]},<br />

Which[ft === ZeroList,<br />

If[N[zerof[0]] < N[a],<br />

Ceil<strong>in</strong>g[x/.F<strong>in</strong>dRoot[zerof[x] == a,{x,0,1},<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

MaxIterations->ma]],<br />

0],<br />

ft === S<strong>in</strong>Argument || ft === CosArgument,<br />

If[ft === CosArgument, cfoffs=1/2];<br />

ext<strong>re</strong>ma = Select[NSolve[zerof'[x] == 0,x,Work<strong>in</strong>gP<strong>re</strong>cision->wp],<br />

(Im[x/.#] == 0)&];<br />

If[Length[ext<strong>re</strong>ma] == 0, a,<br />

Max[x/.Last[ext<strong>re</strong>ma],a]],<br />

True,<br />

Message[PartitionTable::badparam,ft]]<br />

];<br />

PartitionPo<strong>in</strong>ts[zerof_, a_, n_, offs_, i<strong>re</strong>l_, opts___Rule] :=<br />

Module[{x,koffs,cfoffs=0,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[PartitionTable],<br />

ag = AccuracyGoal/. {opts}/.Options[PartitionTable],<br />

ma = MaxRecursion/. {opts}/.Options[PartitionTable],<br />

ft = FunctionType/. {opts}/.Options[PartitionTable]},<br />

Which[ft === ZeroList,<br />

Table[zerof[x],{x,offs+i<strong>re</strong>l,offs+i<strong>re</strong>l+n-1}],<br />

ft === S<strong>in</strong>Argument || ft === CosArgument,<br />

If[ft === CosArgument, cfoffs=1/2];<br />

Which[NumberQ[Limit[zerof[x],x->Inf<strong>in</strong>ity]],<br />

Table[Inf<strong>in</strong>ity,{k,0,n-1}],<br />

N[zerof'[offs+1]] > 0,<br />

koffs = N[Ceil<strong>in</strong>g[zerof[offs]/Pi],2 wp];<br />

Re[Table[x/.F<strong>in</strong>dRoot[zerof[x] == (k+koffs+cfoffs) Pi,<br />

{x,offs,offs+1},<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

MaxIterations->ma],<br />

{k,i<strong>re</strong>l,i<strong>re</strong>l+n-1}]],<br />

True,<br />

koffs = N[Floor[zerof[offs]/Pi],2 wp];<br />

Re[Table[x/.F<strong>in</strong>dRoot[zerof[x] == (-k+koffs-cfoffs) Pi,<br />

{x,offs,offs+1},<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

219


];<br />

AccuracyGoal->ag,<br />

MaxIterations->ma],<br />

{k,i<strong>re</strong>l,i<strong>re</strong>l+n-1}]]<br />

],<br />

True,<br />

Message[PartitionTable::badparam,ft]]<br />

OscIntControlled[f_, fzero_, a_, opts___Rule] :=<br />

Module[{seqtab,parttab,iterations=1,offs,firstval,<strong>in</strong>tlim,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

ma = MaxRecursion/. {opts}/.Options[OscInt],<br />

nt = NSumTerms/. {opts}/.Options[OscInt],<br />

ne = NSumExtraTerms/. {opts}/.Options[OscInt],<br />

wd = WynnDeg<strong>re</strong>e/. {opts}/.Options[OscInt],<br />

it = MaxIterations/. {opts}/.Options[OscInt],<br />

ac = AccuracyControl/. {opts}/.Options[OscInt],<br />

il = IterationLength/. {opts}/.Options[OscInt]},<br />

offs = PartitionOffs[zerof,a,opts];<br />

parttab = PartitionPo<strong>in</strong>ts[fzero,a,ne+nt+1,offs,0,opts];<br />

If[N[parttab[[1]]] < a,<br />

parttab = Select[parttab,(N[#] >= a)&];<br />

Message[OscInt::lowfirst,Length[parttab]]];<br />

firstval = If[N[parttab[[1]]] == N[a], 0,<br />

NIntegrate[f[x],{x,a,parttab[[1]]},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma]];<br />

If[N[parttab[[1]]] =!= Inf<strong>in</strong>ity,<br />

seqtab = Table[NIntegrate[f[x],{x,parttab[[i]],parttab[[i+1]]},<br />

Method->GaussKronrod,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma],<br />

{i,1,Length[parttab]-1}]];<br />

Which[N[parttab[[1]]] == Inf<strong>in</strong>ity,<br />

firstval,<br />

ac === False,<br />

SequenceLimit[Take[FoldList[Plus,firstval,seqtab],<br />

{nt+1,nt+ne}],WynnDeg<strong>re</strong>e->wd],<br />

IntegerQ[ac],<br />

<strong>in</strong>tlim = SequenceLimit[Take[FoldList[Plus,firstval,seqtab],<br />

{nt+1,nt+ne}],WynnDeg<strong>re</strong>e->wd];<br />

While[iterations < it && Accuracy[<strong>in</strong>tlim] < ac,<br />

parttab = Jo<strong>in</strong>[{Last[parttab]},<br />

PartitionPo<strong>in</strong>ts[fzero,a,il,offs,ne+nt+1,opts]];<br />

seqtab = Jo<strong>in</strong>[seqtab,<br />

Table[NIntegrate[f[x],{x,parttab[[i]],parttab[[i+1]]},<br />

Method->GaussKronrod,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma],<br />

{i,1,Length[parttab]-1}]];<br />

ne = ne + il;<br />

iterations++;<br />

<strong>in</strong>tlim = SequenceLimit[Take[FoldList[Plus,firstval,seqtab],<br />

{nt+1,nt+ne}],WynnDeg<strong>re</strong>e->wd]<br />

];<br />

If[Accuracy[<strong>in</strong>tlim] < ac, Message[OscInt::accfail,iterations]];<br />

<strong>in</strong>tlim,<br />

True,<br />

Message[OscInt::badparam,ac]]<br />

];<br />

OscInt[f<strong>in</strong>t_, fzero_, {a_, a0_}, opts___Rule] :=<br />

Module[{parttab,lowlim,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

nt = NSumTerms/. {opts}/.Options[OscInt],<br />

ne = NSumExtraTerms/. {opts}/.Options[OscInt]},<br />

lowlim = If[a == a0, a, N[Max[a,a0],2 wp]];<br />

parttab = PartitionTable[fzero,lowlim,ne+nt+1,opts];<br />

PartInt[f<strong>in</strong>t,parttab,a,opts]<br />

];<br />

220<br />

A Mathematica packages


A.2 Solutions for the step potential<br />

OscInt[f<strong>in</strong>t_, fzero_, a_, opts___Rule] :=<br />

OscInt[f<strong>in</strong>t,fzero,{a,a},opts]<br />

End[]<br />

SetAttributes[PolynomialDeg<strong>re</strong>e, ReadProtected]<br />

SetAttributes[AsymptoticExpand, ReadProtected]<br />

SetAttributes[ApproxLimGeneric, ReadProtected]<br />

SetAttributes[ApproxLimQuad, ReadProtected]<br />

SetAttributes[ApproxLimL<strong>in</strong>ear, ReadProtected]<br />

SetAttributes[QuadZero, ReadProtected]<br />

SetAttributes[QuadOffset, ReadProtected]<br />

SetAttributes[HypZero, ReadProtected]<br />

SetAttributes[HypOffset, ReadProtected]<br />

SetAttributes[HypZeroExact, ReadProtected]<br />

SetAttributes[HypOffsetExact, ReadProtected]<br />

SetAttributes[ZerosInBetween, ReadProtected]<br />

SetAttributes[HypApproxError, ReadProtected]<br />

SetAttributes[PartInt, ReadProtected]<br />

SetAttributes[PartitionTable, ReadProtected]<br />

SetAttributes[PartitionOffs, ReadProtected]<br />

SetAttributes[PartitionPo<strong>in</strong>ts, ReadProtected]<br />

SetAttributes[OscInt, ReadProtected]<br />

SetAttributes[OscIntControlled, ReadProtected]<br />

(*<br />

Protect[PolynomialDeg<strong>re</strong>e, AsymptoticExpand, ApproxLimGeneric]<br />

Protect[ApproxLimQuad, ApproxLimL<strong>in</strong>ear]<br />

Protect[QuadZero, QuadOffset]<br />

Protect[HypZero, HypOffset]<br />

Protect[HypZeroExact, HypOffsetExact]<br />

Protect[ZerosInBetween, HypApproxError]<br />

Protect[PartInt, PartitionTable, PartitionOffs, PartitionPo<strong>in</strong>ts]<br />

Protect[OscInt, OscIntControlled]<br />

*)<br />

EndPackage[]<br />

A.2 Solutions for the step potential<br />

The package Tunnel computes the solutions of the Schrod<strong>in</strong>ger equation for the step potential<br />

barrier discussed <strong>in</strong> section 4.3. Its structu<strong>re</strong> is expla<strong>in</strong>ed <strong>in</strong> chapter 8. For the calculation of<br />

the <strong>in</strong>tegrals, the package OscInt is used.<br />

(* Copyright: Copyright 1996, Institute of Computertechnology, *)<br />

(* Vienna University of Technology *)<br />

(*:Version: Mathematica 2.2.3 *)<br />

(*:Title: Tunnel *)<br />

(*:Author: Thilo Sauter *)<br />

(*:Keywords: Tunnel Effect *)<br />

(*:Requi<strong>re</strong>ments: None. *)<br />

(*:Warn<strong>in</strong>gs: None so far *)<br />

(*:Packages: OscInt *)<br />

(*:Summary: This package conta<strong>in</strong>s functions for the time-dependent solutions<br />

of Schroed<strong>in</strong>ger's equation. The exact solution has been evaluated<br />

for wave packet imp<strong>in</strong>g<strong>in</strong>g on a given step potential barrier. The<br />

<strong>in</strong>itial shape of the wave packet can be either <strong>re</strong>ctangular,<br />

triangular, or Gaussian. The solution of the diffe<strong>re</strong>ntial<br />

equation consists of Fourier <strong>in</strong>tegrals and must be evaluated<br />

numerically. A seve<strong>re</strong> obstacle to a straightforward <strong>in</strong>tegration<br />

a<strong>re</strong> the heavily oscillat<strong>in</strong>g <strong>in</strong>tegrands. Thus <strong>in</strong>tegration is<br />

carried out by us<strong>in</strong>g the functions <strong>in</strong> the package OscInt<br />

*)<br />

(*:History: 15-11-1996 written<br />

25-01-1997 templates added<br />

02-02-1997 new nam<strong>in</strong>g convention for functions<br />

17-02-1997 truncation quadratu<strong>re</strong> of Gaussian waves added<br />

*)<br />

Beg<strong>in</strong>Package["Tunnel`", "OscInt`"]<br />

Phi::usage =<br />

"Phi[x,t,w,k,(n),(opts)] <strong>re</strong>turns the wave function at a given coord<strong>in</strong>ate<br />

<strong>in</strong> space and time. The shape of the <strong>in</strong>cident wave is selected with the option<br />

Shape. The step barrier is supposed to beg<strong>in</strong> at x=0. Depend<strong>in</strong>g on the spatial<br />

221


coord<strong>in</strong>ate the functions either <strong>re</strong>turns the sum of <strong>in</strong>cident and <strong>re</strong>flected wave<br />

or the portion of the wave <strong>in</strong>side the tunnel. The parameter w is the ratio of<br />

the carrier f<strong>re</strong>quency of the <strong>in</strong>cident wave and the characteristic f<strong>re</strong>quency of<br />

the barrier. For tunnell<strong>in</strong>g, 0


A.2 Solutions for the step potential<br />

(*-- Functions <strong>in</strong>dependent of the waveform --*)<br />

TunCoeff[xi_] := 2 xi/(xi + Sqrt[xi^2-1]);<br />

TunOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + b xi + c Sqrt[xi^2 - 1] + d)];<br />

RefCoeff[xi_] := (xi - Sqrt[xi^2-1])/(xi + Sqrt[xi^2-1]);<br />

OutOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + (b+c) xi + d)];<br />

PhiArg[a_,b_,c_,d_][xi_] := a xi^2 + b xi + c Sqrt[xi^2 - 1] + d;<br />

GaussTruncLims[w_,k_,n_,wp_] :=<br />

{N[Sqrt[w] (1-n/(Pi k) *<br />

Sqrt[Log[Sqrt[2 Pi] k/Sqrt[n w Sqrt[2 Pi]]/10^-(wp+1)]]), wp+2],<br />

N[Sqrt[w] (1+n/(Pi k) *<br />

Sqrt[Log[Sqrt[2 Pi] k/Sqrt[n w Sqrt[2 Pi]]/10^-(wp+1)]]), wp+2]};<br />

(*-- Triangular Shape <strong>in</strong>side the tunnel --*)<br />

TriaConst[w_,k_] := Sqrt[3/w] / (2 k Pi^2);<br />

TriaShapePos[w_,xi_] := 1/(1-xi/Sqrt[w])^2;<br />

TriaShapeNeg[w_,xi_] := 1/(1+xi/Sqrt[w])^2;<br />

TriaTransPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * TriaShapePos[w,xi] * TunOsc[a,b,c,d,xi];<br />

TriaTransNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * TriaShapeNeg[w,xi] * TunOsc[a,b,c,d,xi];<br />

TriaGradTransPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * TriaShapePos[w,xi] * TunOsc[a,b,c,d,xi] *<br />

I Sqrt[xi^2 - 1];<br />

TriaGradTransNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * TriaShapeNeg[w,xi] * TunOsc[a,b,c,d,xi] *<br />

(-I Sqrt[xi^2 - 1]);<br />

TriaEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaTransPos[a,b,c,d,w,k][xi] *<br />

(-1 + 2 Exp[I (Pi k/Sqrt[w] xi - Pi k)] -<br />

Exp[2 I (Pi k/Sqrt[w] xi - Pi k)]);<br />

TriaGradEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaGradTransPos[a,b,c,d,w,k][xi] *<br />

(-1 + 2 Exp[I (Pi k/Sqrt[w] xi - Pi k)] -<br />

Exp[2 I (Pi k/Sqrt[w] xi - Pi k)]);<br />

(*-- Triangular Shape outside the tunnel --*)<br />

TriaRefPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RefCoeff[xi] * TriaShapePos[w,xi] * OutOsc[a,b,c,d,xi];<br />

TriaRefNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RefCoeff[xi] * TriaShapeNeg[w,xi] * OutOsc[a,b,c,d,xi];<br />

TriaRefEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaRefPos[a,b,c,d,w,k][xi] *<br />

(-1 + 2 Exp[I (Pi k/Sqrt[w] xi - Pi k)] -<br />

Exp[2 I (Pi k/Sqrt[w] xi - Pi k)]);<br />

TriaIncPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaShapePos[w,xi] * OutOsc[a,b,c,d,xi];<br />

TriaIncNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaShapeNeg[w,xi] * OutOsc[a,b,c,d,xi];<br />

TriaIncEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TriaIncPos[a,b,c,d,w,k][xi] *<br />

(-1 + 2 Exp[I (Pi k/Sqrt[w] xi - Pi k)] -<br />

Exp[2 I (Pi k/Sqrt[w] xi - Pi k)]);<br />

(*-- Rectangular Shape <strong>in</strong>side the tunnel --*)<br />

RectConst[w_,k_] := I / (Sqrt[w] 2 Pi);<br />

RectShapePos[w_,xi_] := 1/(1-xi/Sqrt[w]);<br />

RectShapeNeg[w_,xi_] := 1/(1+xi/Sqrt[w]);<br />

RectTransPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * RectShapePos[w,xi] * TunOsc[a,b,c,d,xi];<br />

RectTransNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

TunCoeff[xi] * RectShapeNeg[w,xi] * TunOsc[a,b,c,d,xi];<br />

RectEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RectTransPos[a,b,c,d,w,k][xi] *<br />

(-1 + Exp[I 2 (Pi k/Sqrt[w] xi - Pi k)]);<br />

(*-- Rectangular Shape outside the tunnel --*)<br />

223


RectRefPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RefCoeff[xi] * RectShapePos[w,xi] * OutOsc[a,b,c,d,xi];<br />

RectRefNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RefCoeff[xi] * RectShapeNeg[w,xi] * OutOsc[a,b,c,d,xi];<br />

RectRefEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RectRefPos[a,b,c,d,w,k][xi] *<br />

(-1 + Exp[I 2 (Pi k/Sqrt[w] xi - Pi k)]);<br />

RectIncPos[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RectShapePos[w,xi] * OutOsc[a,b,c,d,xi];<br />

RectIncNeg[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RectShapeNeg[w,xi] * OutOsc[a,b,c,d,xi];<br />

RectIncEvan[a_,b_,c_,d_,w_,k_][xi_] :=<br />

RectIncPos[a,b,c,d,w,k][xi] *<br />

(-1 + Exp[I 2 (Pi k/Sqrt[w] xi - Pi k)]);<br />

(*-- Exponential Shape <strong>in</strong>side the tunnel --*)<br />

GaussConst[w_,k_,n_] := k Sqrt[2 Pi / (w n Sqrt[2 Pi])];<br />

GaussShapePos[w_,k_,n_,xi_] := Exp[-(Pi k (xi/Sqrt[w] - 1)/n)^2];<br />

GaussShapeNeg[w_,k_,n_,xi_] := Exp[-(Pi k (-xi/Sqrt[w] - 1)/n)^2];<br />

GaussTransPos[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

TunCoeff[xi] * GaussShapePos[w,k,n,xi] * TunOsc[a,b,c,d,xi];<br />

GaussTransNeg[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

TunCoeff[xi] * GaussShapeNeg[w,k,n,xi] * TunOsc[a,b,c,d,xi];<br />

GaussGradTransPos[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

TunCoeff[xi] * GaussShapePos[w,k,n,xi] * TunOsc[a,b,c,d,xi] *<br />

I Sqrt[xi^2 - 1];<br />

GaussGradTransNeg[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

TunCoeff[xi] * GaussShapeNeg[w,k,n,xi] * TunOsc[a,b,c,d,xi] *<br />

(-I Sqrt[xi^2 - 1]);<br />

GaussEvan[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

GaussTransPos[a,b,c,d,w,k,n][xi];<br />

GaussGradEvan[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

GaussGradTransPos[a,b,c,d,w,k,n][xi];<br />

(*-- Exponential Shape outside the tunnel --*)<br />

GaussRefPos[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

RefCoeff[xi] * GaussShapePos[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussRefNeg[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

RefCoeff[xi] * GaussShapeNeg[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussRefEvan[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

GaussRefPos[a,b,c,d,w,k,n][xi];<br />

GaussIncPos[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

GaussShapePos[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussIncNeg[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

GaussShapeNeg[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussIncEvan[a_,b_,c_,d_,w_,k_,n_][xi_] :=<br />

GaussIncPos[a,b,c,d,w,k,n][xi];<br />

(*-- Templates for wave computations depend<strong>in</strong>g on the shape --*)<br />

TriaTransTemp[fpos_, fneg_, x_, t_, w_, k_, opts___Rule] :=<br />

Module[{a,b,c,d,o11,o12,o21,o22,o31,o32,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt]},<br />

a = -t;<br />

b = N[Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-Pi k,wp+2];<br />

o11 = QuadOffset[a, c,0];<br />

o12 = QuadOffset[a,-c,0];<br />

o21 = QuadOffset[a, b+c,d];<br />

o22 = QuadOffset[a,-b-c,d];<br />

o31 = QuadOffset[a, 2 b + c,2 d];<br />

o32 = QuadOffset[a,-2 b - c,2 d];<br />

(-(OscInt[fpos[a, 0, c,0,w,k],<br />

QuadZero[a, c,0,o11],1,FunctionType->ZeroList,opts] +<br />

224<br />

A Mathematica packages


A.2 Solutions for the step potential<br />

OscInt[fneg[a, 0,-c,0,w,k],<br />

QuadZero[a, -c,0,o12],1,FunctionType->ZeroList,opts]) +<br />

(OscInt[fpos[a, b, c,d,w,k],<br />

QuadZero[a, b +c,d,o21],1,FunctionType->ZeroList,opts] +<br />

OscInt[fneg[a,-b,-c,d,w,k],<br />

QuadZero[a,-b -c,d,o22],1,FunctionType->ZeroList,opts])*2 -<br />

(OscInt[fpos[a, 2 b, c,2 d,w,k],<br />

QuadZero[a, 2 b +c,2 d,o31],1,FunctionType->ZeroList,opts] +<br />

OscInt[fneg[a,-2 b,-c,2 d,w,k],<br />

QuadZero[a,-2 b -c,2 d,o32],1,FunctionType->ZeroList,opts])<br />

) TriaConst[w,k]<br />

];<br />

TriaTransExactTemp[fpos_, fneg_, x_, t_, w_, k_, opts___Rule] :=<br />

Module[{a,b,c,d,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt]},<br />

a = -t;<br />

b = N[Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-Pi k];<br />

(-(OscInt[fpos[a, 0, c,0,w,k],<br />

PhiArg[a, 0, c,0],1,opts] +<br />

OscInt[fneg[a, 0,-c,0,w,k],<br />

PhiArg[a, 0,-c,0],1,opts]) +<br />

(OscInt[fpos[a, b, c,d,w,k],<br />

PhiArg[a, b, c,d],1,opts] +<br />

OscInt[fneg[a,-b,-c,d,w,k],<br />

PhiArg[a,-b,-c,d],1,opts])*2 -<br />

(OscInt[fpos[a, 2 b, c,2 d,w,k],<br />

PhiArg[a, 2 b, c,2 d],1,opts] +<br />

OscInt[fneg[a,-2 b,-c,2 d,w,k],<br />

PhiArg[a,-2 b,-c,2 d],1,opts])<br />

) TriaConst[w,k]<br />

];<br />

EvanTemp[f_, const_, opts___Rule] :=<br />

Module[{wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

ma = MaxRecursion/. {opts}/.Options[OscInt]},<br />

NIntegrate[f[xi],{xi,-1,1},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma] const<br />

];<br />

RectTransTemp[fpos_, fneg_, x_, t_, w_, k_, opts___Rule] :=<br />

Module[{a,b,c,d,o11,o12,o21,o22,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt]},<br />

a = -t;<br />

b = N[2 Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-2 Pi k,wp+2];<br />

o11 = N[QuadOffset[a, c,0],wp+2];<br />

o12 = N[QuadOffset[a,-c,0],wp+2];<br />

o21 = N[QuadOffset[a, b+c,d],wp+2];<br />

o22 = N[QuadOffset[a,-b-c,d],wp+2];<br />

(-(OscInt[fpos[a, 0, c,0,w,k],<br />

QuadZero[a, c,0,o11],1,FunctionType->ZeroList,opts] +<br />

OscInt[fneg[a, 0,-c,0,w,k],<br />

QuadZero[a, -c,0,o12],1,FunctionType->ZeroList,opts]) +<br />

(OscInt[fpos[a, b, c,d,w,k],<br />

QuadZero[a, b +c,d,o21],1,FunctionType->ZeroList,opts] +<br />

OscInt[fneg[a,-b,-c,d,w,k],<br />

QuadZero[a,-b -c,d,o22],1,FunctionType->ZeroList,opts])<br />

) RectConst[w,k]<br />

];<br />

RectTransExact[x_, t_, w_, k_, opts___Rule] :=<br />

Module[{a,b,c,d,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt]},<br />

a = -t;<br />

b = N[2 Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-2 Pi k,wp+2];<br />

(-(OscInt[RectTransPos[a, 0, c,0,w,k],<br />

PhiArg[a, 0, c,0],1,opts] +<br />

OscInt[RectTransNeg[a, 0,-c,0,w,k],<br />

PhiArg[a, 0,-c,0],1,opts]) +<br />

(OscInt[RectTransPos[a, b, c,d,w,k],<br />

225


PhiArg[a, b, c,d],1,opts] +<br />

OscInt[RectTransNeg[a,-b,-c,d,w,k],<br />

PhiArg[a,-b,-c,d],1,opts])<br />

) RectConst[w,k]<br />

];<br />

GaussTransTemp[fpos_, fneg_, x_, t_, w_, k_, n_, opts___Rule] :=<br />

Module[{a,b,c,d,o1,o2,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt]},<br />

a = -t;<br />

b = N[Pi k/Sqrt[w],wp+2];<br />

c = x;<br />

d = N[-Pi k,wp+2];<br />

o1 = QuadOffset[a, b+c,d];<br />

o2 = QuadOffset[a,-b-c,d];<br />

(OscInt[fpos[a, b, c,d,w,k,n],<br />

QuadZero[a, b +c,d,o1],1,FunctionType->ZeroList,opts] +<br />

OscInt[fneg[a,-b,-c,d,w,k,n],<br />

QuadZero[a,-b -c,d,o2],1,FunctionType->ZeroList,opts]<br />

) GaussConst[w,k,n]<br />

];<br />

GaussTransTruncTemp[fpos_, fneg_, x_, t_, w_, k_, n_, opts___Rule] :=<br />

Module[{wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

ma = MaxRecursion/. {opts}/.Options[OscInt],<br />

limpos, limneg},<br />

{limneg,limpos} = GaussTruncLims[w,k,n,wp];<br />

(If[limpos DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma]] +<br />

If[limneg >= -1, 0,<br />

NIntegrate[fneg[-t,-Pi k/Sqrt[w],-x,-Pi k,w,k,n][xi],<br />

{xi,1,-limneg},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma]]<br />

) GaussConst[w,k,n]<br />

];<br />

GaussEvanTruncTemp[f_, w_, k_, n_, opts___Rule] :=<br />

Module[{limpos,limneg,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

ma = MaxRecursion/. {opts}/.Options[OscInt]},<br />

{limneg,limpos} = GaussTruncLims[w,k,n,wp];<br />

limneg = Max[limneg,-1];<br />

limpos = M<strong>in</strong>[limpos,1];<br />

NIntegrate[f[xi],{xi,limneg,limpos},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma] GaussConst[w,k,n]<br />

];<br />

(*-- Driver functions for the computation of the wave <strong>in</strong>tegrals --*)<br />

PhiTrans[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[{sh = Shape/. {opts}/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.{opts}/.Options[Phi]},<br />

Switch[sh,<br />

Rect,<br />

Switch[pt,<br />

Approximate, RectTransTemp[RectTransPos,RectTransNeg,x,t,w,k,opts],<br />

Zeros, RectTransExact[x,t,w,k,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

Tria,<br />

Switch[pt,<br />

226<br />

A Mathematica packages


A.2 Solutions for the step potential<br />

]<br />

];<br />

Approximate, TriaTransTemp[TriaTransPos,TriaTransNeg,x,t,w,k,opts],<br />

Zeros, TriaTransExactTemp[TriaTransPos,TriaTransNeg,x,t,w,k,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, GaussTransTemp[GaussTransPos,GaussTransNeg,x,t,w,k,n,opts],<br />

Truncate, GaussTransTruncTemp[GaussTransPos,GaussTransNeg,x,t,w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

PhiTrans[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

PhiTrans[x,t,w,k,1,opts]];<br />

PhiGradTrans[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[{sh = Shape/. {opts}/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.{opts}/.Options[Phi]},<br />

Switch[sh,<br />

Tria,<br />

Switch[pt,<br />

Approximate, TriaTransTemp[TriaGradTransPos,TriaGradTransNeg,x,t,w,k,opts],<br />

Zeros, TriaTransExactTemp[TriaGradTransPos,TriaGradTransNeg,x,t,w,k,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, GaussTransTemp[GaussGradTransPos,GaussGradTransNeg,x,t,w,k,n,opts],<br />

Truncate, GaussTransTruncTemp[GaussGradTransPos,GaussGradTransNeg,x,t,w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

]<br />

];<br />

PhiGradTrans[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

PhiGradTrans[x,t,w,k,1,opts]];<br />

PhiEvan[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[{sh = Shape/. {opts}/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.{opts}/.Options[Phi]},<br />

Switch[sh,<br />

Rect,<br />

EvanTemp[RectEvan[-t,0,x,0,w,k],RectConst[w,k],opts],<br />

Tria,<br />

EvanTemp[TriaEvan[-t,0,x,0,w,k],TriaConst[w,k],opts],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, EvanTemp[GaussEvan[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n],<br />

GaussConst[w,k,n],opts],<br />

Truncate, GaussEvanTruncTemp[GaussEvan[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n],<br />

w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

]<br />

];<br />

PhiEvan[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[PhiTransmit::miss<strong>in</strong>gval],<br />

PhiEvan[x,t,w,k,1,opts]];<br />

PhiGradEvan[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[{sh = Shape/. {opts}/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.{opts}/.Options[Phi]},<br />

Switch[sh,<br />

Tria,<br />

EvanTemp[TriaGradEvan[-t,0,x,0,w,k],TriaConst[w,k],opts],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, EvanTemp[GaussGradEvan[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n],<br />

GaussConst[w,k,n],opts],<br />

Truncate, GaussEvanTruncTemp[GaussGradEvan[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n],<br />

w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

]<br />

227


];<br />

PhiGradEvan[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

PhiGradEvan[x,t,w,k,1,opts]];<br />

(*-- outside the tunnel --*)<br />

(* Note that <strong>in</strong> this case the spatial coord<strong>in</strong>ate is <strong>in</strong>verted <strong>in</strong> the function calls<br />

because of the <strong>re</strong>verse motion of the <strong>re</strong>flected wave. *)<br />

PhiRef[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[{sh = Shape/. {opts}/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.{opts}/.Options[Phi]},<br />

Switch[sh,<br />

Rect,<br />

RectTransTemp[RectRefPos,RectRefNeg,-x,t,w,k,opts] +<br />

EvanTemp[RectRefEvan[-t,0,-x,0,w,k],RectConst[w,k],opts],<br />

Tria,<br />

TriaTransTemp[TriaRefPos,TriaRefNeg,-x,t,w,k,opts] +<br />

EvanTemp[TriaRefEvan[-t,0,-x,0,w,k],TriaConst[w,k],opts],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, GaussTransTemp[GaussRefPos,GaussRefNeg,-x,t,w,k,n,opts] +<br />

EvanTemp[GaussRefEvan[-t,Pi k/Sqrt[w],-x,-Pi k,w,k,n],<br />

GaussConst[w,k,n],opts],<br />

Truncate, GaussTransTruncTemp[GaussRefPos,GaussRefNeg,-x,t,w,k,n,opts] +<br />

GaussEvanTruncTemp[GaussRefEvan[-t,Pi k/Sqrt[w],-x,-Pi k,w,k,n],<br />

w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

]<br />

];<br />

PhiRef[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

PhiRef[x,t,w,k,1,opts]];<br />

PhiInc[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Module[{sh = Shape/. {opts}/.Options[Phi],<br />

pt = PPo<strong>in</strong>ts/.{opts}/.Options[Phi]},<br />

Switch[sh,<br />

Rect,<br />

RectTransTemp[RectIncPos,RectIncNeg,x,t,w,k,opts] +<br />

EvanTemp[RectIncEvan[-t,0,x,0,w,k],RectConst[w,k],opts],<br />

Tria,<br />

TriaTransTemp[TriaIncPos,TriaIncNeg,x,t,w,k,opts] +<br />

EvanTemp[TriaIncEvan[-t,0,x,0,w,k],TriaConst[w,k],opts],<br />

Gauss,<br />

Switch[pt,<br />

Approximate, GaussTransTemp[GaussIncPos,GaussIncNeg,x,t,w,k,n,opts] +<br />

EvanTemp[GaussIncEvan[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n],<br />

GaussConst[w,k,n],opts],<br />

Truncate, GaussTransTruncTemp[GaussIncPos,GaussIncNeg,x,t,w,k,n,opts] +<br />

GaussEvanTruncTemp[GaussIncEvan[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n],<br />

w,k,n,opts],<br />

_, Message[Phi::<strong>in</strong>validpart,pt]],<br />

_,<br />

Message[Phi::<strong>in</strong>validshape,sh]<br />

]<br />

];<br />

PhiInc[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

PhiInc[x,t,w,k,1,opts]];<br />

Phi[x_, t_, w_, k_, n_?Positive, opts___Rule] :=<br />

Which[t == 0 && x == 0 && (Shape/.{opts}/.Options[Phi]) === Rect, 0.5,<br />

t < 0, 0,<br />

x >= 0, PhiEvan[x,t,w,k,n,opts] + PhiTrans[x,t,w,k,n,opts],<br />

x < 0, PhiInc[x,t,w,k,n,opts] + PhiRef[x,t,w,k,n,opts]];<br />

Phi[x_, t_, w_, k_, opts___Rule] :=<br />

If[(Shape/.{opts}/.Options[Phi]) === Gauss,<br />

Message[Phi::miss<strong>in</strong>gval],<br />

Phi[x,t,w,k,1,opts]];<br />

PhiGrad[x_, t_, w_, k_, n_?Positive,opts___Rule] :=<br />

If[x < 0 || t < 0 || x == 0 && t == 0, 0,<br />

PhiGradEvan[x,t,w,k,n,opts] + PhiGradTrans[x,t,w,k,n,opts]];<br />

228<br />

A Mathematica packages


A.3 Solutions for the squa<strong>re</strong> barrier<br />

PhiGrad[x_, t_, w_, k_, opts___Rule] :=<br />

If[x < 0 || t < 0 || x == 0 && t == 0, 0,<br />

PhiGradEvan[x,t,w,k,opts] + PhiGradTrans[x,t,w,k,opts]];<br />

End[]<br />

SetAttributes[PhiTrans, ReadProtected]<br />

SetAttributes[PhiGradTrans, ReadProtected]<br />

SetAttributes[PhiEvan, ReadProtected]<br />

SetAttributes[PhiGradEvan, ReadProtected]<br />

SetAttributes[PhiRef, ReadProtected]<br />

SetAttributes[PhiInc, ReadProtected]<br />

SetAttributes[Phi, ReadProtected]<br />

SetAttributes[PhiGrad, ReadProtected]<br />

(*<br />

Protect[PhiTrans, PhiGradTrans, PhiEvan, PhiGradEvan]<br />

Protect[PhiRef, PhiInc, Phi, PhiGrad]<br />

*)<br />

EndPackage[]<br />

A.3 Solutions for the squa<strong>re</strong> barrier<br />

The package Gauss was written to compute the solutions of the Schrod<strong>in</strong>ger equation for the<br />

tunnell<strong>in</strong>g of an electron through a squa<strong>re</strong> barrier (section 4.6). The <strong>in</strong>itial shape of the wave<br />

packet is Gaussian. The details of the implementation a<strong>re</strong> not t<strong>re</strong>ated <strong>in</strong> p<strong>re</strong>vious chapters,<br />

however the structu<strong>re</strong> is identical to that of Tunnel.<br />

(* Copyright: Copyright 1997, Institute of Computertechnology, *)<br />

(* Vienna University of Technology *)<br />

(*:Version: Mathematica 2.2.3 *)<br />

(*:Title: Gauss *)<br />

(*:Author: Thilo Sauter *)<br />

(*:Keywords: Tunnel Effect *)<br />

(*:Requi<strong>re</strong>ments: None. *)<br />

(*:Warn<strong>in</strong>gs: None so far *)<br />

(*:Packages: OscInt *)<br />

(*:Summary: This package computes the solutions of the time-dependent<br />

Schroed<strong>in</strong>ger equation for an <strong>in</strong>itially Gaussian wave packet<br />

tunnell<strong>in</strong>g through a squa<strong>re</strong> barrier. It is based on the same<br />

structu<strong>re</strong> as the package Tunnel. The wave <strong>in</strong>tegrals a<strong>re</strong><br />

computed by truncation of the <strong>in</strong>f<strong>in</strong>ite <strong>in</strong>tegration range.<br />

*)<br />

(*:History: 31-10-1997 written<br />

*)<br />

Beg<strong>in</strong>Package["Gauss`", "OscInt`"]<br />

Phi::usage =<br />

"Phi[x,t,w,k,n,l,(opts)] <strong>re</strong>turns the wave function of an <strong>in</strong>itially Gaussian<br />

wave packet at a given coord<strong>in</strong>ate <strong>in</strong> space and time. The squa<strong>re</strong> barrier is<br />

supposed to span the <strong>in</strong>terval x=0 to x=l. Depend<strong>in</strong>g on the spatial<br />

coord<strong>in</strong>ate the function either <strong>re</strong>turns the sum of <strong>in</strong>cident and <strong>re</strong>flected wave,<br />

the portion of the wave <strong>in</strong>side the tunnel, or the transmitted wave. The parameter<br />

w is the ratio of the carrier f<strong>re</strong>quency of the <strong>in</strong>cident wave and the<br />

characteristic f<strong>re</strong>quency of the barrier. For tunnell<strong>in</strong>g, 0


PhiFro::usage =<br />

"PhiFro[x,t,w,k,n,l,(opts)] <strong>re</strong>turns the left-go<strong>in</strong>g part <strong>in</strong> the barrier.<br />

For further <strong>in</strong>formation see Phi"<br />

PhiTun::usage =<br />

"PhiTun[x,t,w,k,n,l,(opts)] <strong>re</strong>turns the wave <strong>in</strong>side the barrier. For<br />

further <strong>in</strong>formation see Phi"<br />

Beg<strong>in</strong>["`Private`"]<br />

(*-- Functions <strong>in</strong>dependent of the waveform --*)<br />

RefCoeffPos[xi_,l_] := -I S<strong>in</strong>[l Sqrt[xi^2-1]]/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] - I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]]);<br />

RefCoeffNeg[xi_,l_] := I S<strong>in</strong>[l Sqrt[xi^2-1]]/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] + I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]]);<br />

ToCoeffPos[xi_,l_] := (xi Sqrt[xi^2-1] + xi^2)/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] - I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]])<br />

ToCoeffNeg[xi_,l_] := (xi Sqrt[xi^2-1] + xi^2)/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] + I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]])<br />

FroCoeffPos[xi_,l_] := (xi Sqrt[xi^2-1] - xi^2)/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] - I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]])<br />

FroCoeffNeg[xi_,l_] := (xi Sqrt[xi^2-1] - xi^2)/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] + I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]])<br />

TransCoeffPos[xi_,l_] := 2 xi Sqrt[xi^2-1]/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] - I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]])<br />

TransCoeffNeg[xi_,l_] := 2 xi Sqrt[xi^2-1]/<br />

(2 xi Sqrt[xi^2-1] Cos[l Sqrt[xi^2-1]] + I (2 xi^2-1) S<strong>in</strong>[l Sqrt[xi^2-1]])<br />

TunOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + b xi + c Sqrt[xi^2 - 1] + d)];<br />

OutOsc[a_,b_,c_,d_,xi_] := Exp[I (a xi^2 + (b+c) xi + d)];<br />

GaussTruncLims[w_,k_,n_,wp_] :=<br />

{N[Sqrt[w] (1-n/(Pi k) *<br />

Sqrt[Log[Sqrt[2 Pi] k/Sqrt[n w Sqrt[2 Pi]]/10^-(wp+1)]]), wp+2],<br />

N[Sqrt[w] (1+n/(Pi k) *<br />

Sqrt[Log[Sqrt[2 Pi] k/Sqrt[n w Sqrt[2 Pi]]/10^-(wp+1)]]), wp+2]};<br />

(*-- <strong>Wave</strong> packet shape --*)<br />

GaussConst[w_,k_,n_] := k Sqrt[2 Pi / (w n Sqrt[2 Pi])];<br />

GaussShapePos[w_,k_,n_,xi_] := Exp[-(Pi k (xi/Sqrt[w] - 1)/n)^2];<br />

GaussShapeNeg[w_,k_,n_,xi_] := Exp[-(Pi k (-xi/Sqrt[w] - 1)/n)^2];<br />

GaussIncPos[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

GaussShapePos[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussIncNeg[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

GaussShapeNeg[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussRefPos[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

RefCoeffPos[xi,l] * GaussShapePos[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussRefNeg[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

RefCoeffNeg[xi,l] * GaussShapeNeg[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussToPos[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

ToCoeffPos[xi,l] * GaussShapePos[w,k,n,xi] * TunOsc[a,b,c,d,xi];<br />

GaussToNeg[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

ToCoeffNeg[xi,l] * GaussShapeNeg[w,k,n,xi] * TunOsc[a,b,c,d,xi];<br />

GaussFroPos[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

FroCoeffPos[xi,l] * GaussShapePos[w,k,n,xi] * TunOsc[a,b,c,d,xi];<br />

GaussFroNeg[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

FroCoeffNeg[xi,l] * GaussShapeNeg[w,k,n,xi] * TunOsc[a,b,c,d,xi];<br />

GaussTransPos[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

TransCoeffPos[xi,l] * GaussShapePos[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

GaussTransNeg[a_,b_,c_,d_,w_,k_,n_,l_][xi_] :=<br />

TransCoeffNeg[xi,l] * GaussShapeNeg[w,k,n,xi] * OutOsc[a,b,c,d,xi];<br />

(*-- Templates for wave computations --*)<br />

GaussPassTruncTemp[fpos_, fneg_, x_, t_, w_, k_, n_, l_, opts___Rule] :=<br />

Module[{wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

230<br />

A Mathematica packages


A.3 Solutions for the squa<strong>re</strong> barrier<br />

ma = MaxRecursion/. {opts}/.Options[OscInt],<br />

limpos, limneg},<br />

{limneg,limpos} = GaussTruncLims[w,k,n,wp];<br />

(If[limpos DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma]] +<br />

If[limneg >= -1, 0,<br />

NIntegrate[fneg[-t,-Pi k/Sqrt[w],-x,-Pi k,w,k,n,l][xi],<br />

{xi,1,-limneg},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma]]<br />

) GaussConst[w,k,n]<br />

];<br />

GaussEvanTruncTemp[f_, x_, t_, w_, k_, n_, l_, opts___Rule] :=<br />

Module[{limpos,limneg,<br />

wp = Work<strong>in</strong>gP<strong>re</strong>cision/.{opts}/.Options[OscInt],<br />

ag = AccuracyGoal/. {opts}/.Options[OscInt],<br />

pg = P<strong>re</strong>cisionGoal/. {opts}/.Options[OscInt],<br />

mi = M<strong>in</strong>Recursion/. {opts}/.Options[OscInt],<br />

ma = MaxRecursion/. {opts}/.Options[OscInt]},<br />

{limneg,limpos} = GaussTruncLims[w,k,n,wp];<br />

limneg = Max[limneg,-1];<br />

limpos = M<strong>in</strong>[limpos,1];<br />

NIntegrate[f[-t,Pi k/Sqrt[w],x,-Pi k,w,k,n,l][xi],<br />

{xi,limneg,limpos},<br />

Method->DoubleExponential,<br />

Work<strong>in</strong>gP<strong>re</strong>cision->wp,<br />

AccuracyGoal->ag,<br />

P<strong>re</strong>cisionGoal->pg,<br />

M<strong>in</strong>Recursion->mi,<br />

MaxRecursion->ma] GaussConst[w,k,n]<br />

];<br />

(*-- Driver functions for the computation of the wave <strong>in</strong>tegrals --*)<br />

(* Note that for the <strong>re</strong>flected wave outside and the left-go<strong>in</strong>g part<br />

<strong>in</strong>side the tunnel spatial coord<strong>in</strong>ate is <strong>in</strong>verted <strong>in</strong> the function calls<br />

because of the <strong>re</strong>verse motion of the partial waves. *)<br />

(* Note also that the spatial coord<strong>in</strong>ate for the <strong>re</strong>gions <strong>in</strong>side and<br />

beh<strong>in</strong>d the barrier is transformed to x-l. *)<br />

PhiInc[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

GaussPassTruncTemp[GaussIncPos,GaussIncNeg,x,t,w,k,n,l,opts] +<br />

GaussEvanTruncTemp[GaussIncPos,x,t,w,k,n,l,opts];<br />

PhiRef[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

GaussPassTruncTemp[GaussRefPos,GaussRefNeg,-x,t,w,k,n,l,opts] +<br />

GaussEvanTruncTemp[GaussRefPos,-x,t,w,k,n,l,opts];<br />

PhiTo[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

GaussPassTruncTemp[GaussToPos,GaussToNeg,x-l,t,w,k,n,l,opts] +<br />

GaussEvanTruncTemp[GaussToPos,x-l,t,w,k,n,l,opts];<br />

PhiFro[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

GaussPassTruncTemp[GaussFroPos,GaussFroNeg,-x+l,t,w,k,n,l,opts] +<br />

GaussEvanTruncTemp[GaussFroPos,-x+l,t,w,k,n,l,opts];<br />

PhiTrans[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

GaussPassTruncTemp[GaussTransPos,GaussTransNeg,x-l,t,w,k,n,l,opts] +<br />

GaussEvanTruncTemp[GaussTransPos,x-l,t,w,k,n,l,opts];<br />

PhiTun[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

PhiTo[x,t,w,k,n,l,opts] + PhiFro[x,t,w,k,n,l,opts];<br />

Phi[x_, t_, w_, k_, n_?Positive, l_, opts___Rule] :=<br />

Which[t < 0, 0,<br />

x >= l, PhiTrans[x,t,w,k,n,l,opts],<br />

0 < x && x < l, PhiTun[x,t,w,k,n,l,opts],<br />

x


SetAttributes[PhiInc, ReadProtected]<br />

SetAttributes[PhiRef, ReadProtected]<br />

SetAttributes[PhiTo, ReadProtected]<br />

SetAttributes[PhiFro, ReadProtected]<br />

SetAttributes[PhiTun, ReadProtected]<br />

SetAttributes[PhiTrans, ReadProtected]<br />

SetAttributes[Phi, ReadProtected]<br />

(*<br />

Protect[PhiInc, PhiRef, PhiTo, PhiFro]<br />

Protect[PhiTun, PhiTrans, Phi]<br />

*)<br />

EndPackage[]<br />

A.4 Electromagnetic waves<br />

A Mathematica packages<br />

The short package ElMag computes the solutions of the wave equation for a transmission l<strong>in</strong>e<br />

lled with a lossless plasma or, alternatively, the propagation of a wave through an unbounded<br />

plasma (section 3.7). The boundary condition is given by a cos-like cur<strong>re</strong>nt be<strong>in</strong>g switched<br />

on at t = 0 and switched o at some <strong>in</strong>tegral number of periods later.<br />

(* Copyright: Copyright 1996, Institute of Computertechnology, *)<br />

(* Vienna University of Technology *)<br />

(*:Version: Mathematica 2.2.3 *)<br />

(*:Title: ElMag *)<br />

(*:Author: Thilo Sauter *)<br />

(*:Keywords: Eletromagnetic <strong>Wave</strong>s, Plasma <strong>Wave</strong>s, Evanescent <strong>Wave</strong>s *)<br />

(*:Requi<strong>re</strong>ments: None. *)<br />

(*:Warn<strong>in</strong>gs: None so far *)<br />

(*:Packages: OscInt *)<br />

(*:Summary: This application specific package conta<strong>in</strong>s functions for<br />

the calculation of wave propagation <strong>in</strong> one dimensional<br />

transmission l<strong>in</strong>es filled with a lossless plasma. The model<br />

is that the wave is excited by a s<strong>in</strong>usoidal cur<strong>re</strong>nt source<br />

switched on at t=0. Voltage and cur<strong>re</strong>nt along the l<strong>in</strong>e may<br />

be calculated as a function of time, distance, and the ratio<br />

of the plasma f<strong>re</strong>quency and the signal f<strong>re</strong>quency.<br />

*)<br />

Beg<strong>in</strong>Package["ElMag`", "OscInt`"]<br />

CurrStat::usage =<br />

"CurrStat[x,t,omega] <strong>re</strong>turns the stationary solution of the wave equation<br />

for the cur<strong>re</strong>nt. See also Curr."<br />

CurrTrans::usage =<br />

"CurrTrans[x,t,omega,(opts)] <strong>re</strong>turns the stationary solution of the wave<br />

equation for the cur<strong>re</strong>nt. The options opts a<strong>re</strong> passed to the quadratu<strong>re</strong><br />

function OscInt. For further explanations see also Curr."<br />

Curr::usage =<br />

"Curr[x,t,omega,(opts)] gives the complete solution of the wave equation<br />

for the cur<strong>re</strong>nt <strong>in</strong> a one dimensional, <strong>in</strong>f<strong>in</strong>itely long transmission l<strong>in</strong>e<br />

filled with lossless plasma. Excitation of the wave is accomplished by a<br />

s<strong>in</strong>usoidal cur<strong>re</strong>nt source at the entrance of the l<strong>in</strong>e (x=0) be<strong>in</strong>g switched<br />

on at t=0. Evanescence is p<strong>re</strong>sumed, so the signal f<strong>re</strong>quency must be lower<br />

than the plasma f<strong>re</strong>quency. The excitation f<strong>re</strong>quency is specified by the<br />

parameter omega which denotes the ratio between signal and plasma f<strong>re</strong>quency,<br />

hence 0


A.4 Electromagnetic waves<br />

function OscInt. For further explanations see also Volt."<br />

Volt::usage =<br />

"Volt[x,t,omega,(opts)] gives the complete solution of the wave equation<br />

for the voltage <strong>in</strong> a one dimensional, <strong>in</strong>f<strong>in</strong>itely long transmission l<strong>in</strong>e<br />

filled with lossless plasma. Excitation of the wave is accomplished by a<br />

s<strong>in</strong>usoidal cur<strong>re</strong>nt source at the entrance of the l<strong>in</strong>e (x=0) be<strong>in</strong>g switched<br />

on at t=0. Evanescence is p<strong>re</strong>sumed, so the signal f<strong>re</strong>quency must be lower<br />

than the plasma f<strong>re</strong>quency. The excitation f<strong>re</strong>quency is specified by the<br />

parameter omega which denotes the ratio between signal and plasma f<strong>re</strong>quency,<br />

hence 0


(*<br />

Protect[CurrStat, CurrTrans, Curr, CurrPack]<br />

Protect[VoltStat, VoltTrans, Volt, VoltPack]<br />

*)<br />

EndPackage[]<br />

A.5 Utilities for display<strong>in</strong>g po<strong>in</strong>ts <strong>in</strong> 3D<br />

A Mathematica packages<br />

The Mathematica package Tools3D conta<strong>in</strong>s a number of functions that a<strong>re</strong> needed for the<br />

graphical <strong>re</strong>p<strong>re</strong>sentation of lists of po<strong>in</strong>ts <strong>in</strong> R 3 . Such lists a<strong>re</strong> typically the outcome of surface<br />

computations over a <strong>re</strong>ctangular grid be<strong>in</strong>g obta<strong>in</strong>ed with commands like Table[fx,y,f(x,y)g,<br />

fx,xm<strong>in</strong>,xmax,dxg, fy,ym<strong>in</strong>,ymax,dyg]. The<strong>re</strong>fo<strong>re</strong> all functions <strong>in</strong> this package assume<br />

that the data is given as two-dimensional lists of po<strong>in</strong>ts, like<br />

fffx11,y11,z11g,fx12,y12,z12g, ::: ,fx1n,y1n,z1ngg,<br />

ffx21,y21,z21g,fx22,y22,z22g, ::: ,fx2n,y2n,z2ngg,<br />

.<br />

.<br />

.<br />

.<br />

.<br />

.<br />

.<br />

ffxm1,ym1,zm1g,fxm2,ym2,zm2g, ::: ,fxmn,ymn,zmnggg.<br />

This matrix must be <strong>re</strong>ctangular, so each sublist must have an equal number of members.<br />

Some functions check this property with the statement<br />

Apply[SameQ,Map[Length,t]],<br />

which rst counts the number of elements <strong>in</strong> each row of the list t (by mapp<strong>in</strong>g the operation<br />

Length onto each member of the list; note that these members a<strong>re</strong> aga<strong>in</strong> lists) and then<br />

evaluates if all entries <strong>in</strong> the <strong>re</strong>sult<strong>in</strong>g list a<strong>re</strong> identical. The outcome of this statement is<br />

boolean | if it is False, error messages a<strong>re</strong> generated.<br />

The functions of the package have the follow<strong>in</strong>g goals:<br />

OrderGrid takes an unsorted, at list of th<strong>re</strong>e-dimensional po<strong>in</strong>ts fxi,yi,zig as <strong>in</strong>put<br />

and arranges them <strong>in</strong> the <strong>re</strong>qui<strong>re</strong>d two-dimensional array of po<strong>in</strong>ts. To this end, the<br />

x-values of all po<strong>in</strong>ts ly<strong>in</strong>g on one grid l<strong>in</strong>e must be undisturbed such that they can be<br />

grouped together.<br />

L<strong>in</strong>k3DList is useful whenever a surface is computed <strong>in</strong> mo<strong>re</strong> than one chunk and the<br />

<strong>re</strong>sults a<strong>re</strong> sto<strong>re</strong>d <strong>in</strong> di e<strong>re</strong>nt les. The function <strong>re</strong>qui<strong>re</strong>s a list with the lenames, the<br />

name for the output le, and optionally a path specify<strong>in</strong>g whe<strong>re</strong> the les a<strong>re</strong> located.<br />

The function then checks whether the output le al<strong>re</strong>ady exists | and if so issues a<br />

warn<strong>in</strong>g | befo<strong>re</strong> merg<strong>in</strong>g and sort<strong>in</strong>g the les. The function call is L<strong>in</strong>k3DList[fFFileA",FileB",<br />

::: g,OutFile"].<br />

ReSize <strong>re</strong>arranges a two-dimensional list by tak<strong>in</strong>g only every nth element. Depend<strong>in</strong>g<br />

on whether the <strong>re</strong>duction should be uniform or not, the function can be called with<br />

ReSize[list,n] or ReSize[list,m,n] for leav<strong>in</strong>g every mth row and nth column.<br />

234


A.5 Utilities for display<strong>in</strong>g po<strong>in</strong>ts <strong>in</strong> 3D<br />

Disp3DList displays surfaces consist<strong>in</strong>g of th<strong>re</strong>e-dimensional po<strong>in</strong>ts by call<strong>in</strong>g the function<br />

ListSurfacePlot3D, which is part of the standard package Graphics`Graphics3D`.<br />

Disp3DContour generates a contour plot of the list of po<strong>in</strong>ts. In this case, the <strong>re</strong>gularity<br />

of the list is checked for the convenience of the user. The <strong>re</strong>ason is that the built-<strong>in</strong><br />

function ListContourPlot would simply not evaluate (without a warn<strong>in</strong>g) if the list<br />

passed to it was not <strong>re</strong>gular.<br />

Movie takes the surface and <strong>re</strong>turns cuts <strong>in</strong> parallel to the x-z plane by <strong>in</strong>c<strong>re</strong>as<strong>in</strong>g the<br />

y-coord<strong>in</strong>ate. If no special value for PlotRange is given, this option is set so as to cover<br />

the ext<strong>re</strong>ma, be<strong>in</strong>g <strong>in</strong>variant for all pictu<strong>re</strong>s. The user can override this default with<br />

another range. Note that <strong>in</strong> order to allow this, the user's options must be written<br />

befo<strong>re</strong> the default values <strong>in</strong> the call of the function ListPlot.<br />

(* Copyright: Copyright 1994, CAD-Div., IGEE, Vienna Univ. of Techn. *)<br />

(*:Version: Mathematica 2.2.1 *)<br />

(*:Title: 3DTools *)<br />

(*:Author: Thilo Sauter *)<br />

(*:Keywords: *)<br />

(*:Requi<strong>re</strong>ments: None. *)<br />

(*:Warn<strong>in</strong>gs: The Functions def<strong>in</strong>ed he<strong>re</strong><strong>in</strong> <strong>re</strong>qui<strong>re</strong> the 3D-Lists to<br />

lie on a <strong>re</strong>ctangular x/y grid *)<br />

(*:Summary: This package conta<strong>in</strong>s functions useful for manag<strong>in</strong>g and<br />

display<strong>in</strong>g lists of 3D po<strong>in</strong>ts. The <strong>re</strong>order<strong>in</strong>g function<br />

OrderGrid is based on the package ListToArray by Jason F.<br />

Harris.<br />

*)<br />

Beg<strong>in</strong>Package["Tools3D`","Graphics`Graphics3D`"]<br />

OrderGrid::usage =<br />

"Takes a list {{x1,y1,z1},{x2,y2,z2}...} and orders it to the<br />

two-dimensional form {{{x1,y11,z11},{x1,y12,z12},..},{{x2,y21,z21},..},..}<br />

of 3D-po<strong>in</strong>ts."<br />

ReSize::usage =<br />

"ReSize[list,n] keeps only every nth row and column of a two-dimensional<br />

list. ReSize[list,{m,n}] takes every mth row and nth column."<br />

L<strong>in</strong>k3DList::usage =<br />

"L<strong>in</strong>k3DList[Names,Output,(Path)] needs a list of filenames, a name for<br />

the output file and an optional path whe<strong>re</strong> to look for the files. The files<br />

must be a two-dimensional list of 3D-po<strong>in</strong>ts, the output file has the same<br />

form (generated by OrderGrid)."<br />

Disp3DList::usage =<br />

"Display a surface consist<strong>in</strong>g of po<strong>in</strong>ts {x,y,z} over a <strong>re</strong>ctangular<br />

x/y grid (two-dimensional list of 3D-po<strong>in</strong>ts)"<br />

Disp3DContour::usage =<br />

"Disp3DContour[list,(opts)] produces a contour plot of a list of po<strong>in</strong>ts {x,y,z}<br />

over a <strong>re</strong>ctangular and equidistant x/y grid (two-dimensional list of 3D-po<strong>in</strong>ts)"<br />

Movie::usage =<br />

"Movie[list,(opts)] produces an animation of a <strong>re</strong>ctangular list of po<strong>in</strong>ts<br />

{x,y,z} with {x,z}-Frames and <strong>in</strong>c<strong>re</strong>mented y-value"<br />

L<strong>in</strong>k3DList::fewlists = "At least two files needed";<br />

L<strong>in</strong>k3DList::<strong>re</strong>ggrid = "Warn<strong>in</strong>g: Grid not <strong>re</strong>gular!";<br />

Disp3DContour::<strong>re</strong>ggrid = "Grid not <strong>re</strong>gular!";<br />

Beg<strong>in</strong>["`Private`"]<br />

OrderGrid[a_List] := Module[{xvals,asort},<br />

asort = Sort[a];<br />

xvals = Union[Flatten[Cases[asort,{x_,y_,z_}->x]]];<br />

Map[Cases[asort,{#,y_,z_}]&, xvals]];<br />

TakeOther[l_List,n_] := Map[First[#]&,Partition[l,n]];<br />

ReSize[l_List,{n_,m_}] := TakeOther[Map[TakeOther[#,m]&,l],n];<br />

235


ReSize[l_List,n_] := ReSize[l,{n,n}];<br />

L<strong>in</strong>k3DList[names_,outname_,path__:""] := Module[{dir,mergelist={}},<br />

dir = FileNames[names,path];<br />

If[Length[dir] < 2, Message[L<strong>in</strong>k3DList::fewlists],<br />

If[FileNames[outname,path] == {} ||<br />

InputStr<strong>in</strong>g["Output file exists! Overwrite (y/n)?"] == "y",<br />

Pr<strong>in</strong>t["Merg<strong>in</strong>g ",Length[dir]," lists"];<br />

mergelist = OrderGrid[Flatten[Map[Get,dir],2]];<br />

If[!Apply[SameQ,Map[Length,mergelist]],<br />

Message[L<strong>in</strong>k3DList::<strong>re</strong>ggrid]];<br />

If[path != "", SetDi<strong>re</strong>ctory[path]];<br />

Put[mergelist,outname];<br />

If[path != "", ResetDi<strong>re</strong>ctory[]]<br />

]];<br />

];<br />

Disp3DList[t_List, opts___] := ListSurfacePlot3D[t,opts,BoxRatios->{1,1,0.4},<br />

Axes->True];<br />

Disp3DContour[t_List, opts___] := If[!Apply[SameQ,Map[Length,t]],<br />

Message[Disp3DContour::<strong>re</strong>ggrid],<br />

ListContourPlot[Transpose[t]/.{x_,y_,z_}:>z,<br />

MeshRange->{{t[[1,1,1]],t[[Length[t],1,1]]},<br />

{t[[1,1,2]],t[[1,Length[t[[1]]],2]]}},opts]];<br />

Movie[list_List, opts___] := Module[{m<strong>in</strong>z,maxz,optlist},<br />

{m<strong>in</strong>z,maxz} = Apply[{M<strong>in</strong>[#],Max[#]}&,<br />

{Transpose[Flatten[list,1]][[3]]}];<br />

Map[ListPlot[#/.{x_,y_,z_}:>{x,z},<br />

opts,PlotJo<strong>in</strong>ed->True,PlotRange->{All,{m<strong>in</strong>z,maxz}}]&,<br />

Transpose[list]]<br />

];<br />

End[]<br />

SetAttributes[OrderGrid, ReadProtected]<br />

SetAttributes[ReSize, ReadProtected]<br />

SetAttributes[L<strong>in</strong>k3DList, ReadProtected]<br />

SetAttributes[Movie, ReadProtected]<br />

Protect[OrderGrid, ReSize, L<strong>in</strong>k3DList, Movie]<br />

EndPackage[]<br />

236<br />

A Mathematica packages


Bibliography<br />

Bibliography<br />

[1] James R. Newman, editor. The world of mathematics. Simon and Schuster, New York,<br />

1956.<br />

[2] Leon Brioull<strong>in</strong>. <strong>Wave</strong> <strong>Propagation</strong> and Group Velocity. Academic P<strong>re</strong>ss, New York,<br />

1960.<br />

[3] Gerald B. Whitham. L<strong>in</strong>ear and nonl<strong>in</strong>ear waves. John Wiley & Sons, New York, 1976.<br />

[4] Lord Rayleigh. The Theory of Sound. Dover Publications, New York, 1945, rst published<br />

<strong>in</strong> 1877.<br />

[5] Lord Rayleigh. On prog<strong>re</strong>ssive waves. In The Theory of Sound, pages 475{480. Dover<br />

Publications, New York, 1945, rst published <strong>in</strong> 1877.<br />

[6] Julius Adams Stratton. Electromagnetic Theory. McGraw-Hill, New York, 1941.<br />

[7] M. J. Lighthill. Group Velocity. J. Inst. Maths Applics, 1:1{28, 1965.<br />

[8] Richard L. Smith. The Velocities of Light. American Journal of Physics, 38(8):978{984,<br />

1970.<br />

[9] W. V. P<strong>re</strong>stwich. P<strong>re</strong>cise de nition of group velocity. American Journal of Physics,<br />

43(9):832{833, 1975.<br />

[10] Robert Vichnevetsky. Group velocity <strong>in</strong> general waveforms. Mathematics and Computers<br />

<strong>in</strong> Simulation, 27:327{331, 1985.<br />

[11] John David Jackson. Klassische Elektrodynamik. Walter de Gruyter, Berl<strong>in</strong>, second<br />

edition, 1982.<br />

[12] A. Sommerfeld. Uber die Fortp anzung des Lichtes <strong>in</strong> dispergie<strong>re</strong>nden Medien. Annalen<br />

der Physik, 44(10):177{202, 1914.<br />

[13] E. O. Schulz-DuBois. Energy Transport Velocity of Electromagnetic <strong>Propagation</strong> <strong>in</strong><br />

Dispersive <strong>Media</strong>. Proceed<strong>in</strong>gs of the IEEE, 57(10):1748{1757, 1969.<br />

[14] G. R. Baldock and T. Bridgeman. The Mathematical Theory of <strong>Wave</strong> Motion. Ellis<br />

Horwood, Chichester, 1981.<br />

237


Bibliography<br />

[15] Simon Ramo, John R. Wh<strong>in</strong>nery, and Theodo<strong>re</strong> van Duzer. Fields and <strong>Wave</strong>s <strong>in</strong> Communication<br />

Electronics. John Wiley & Sons, New York, 1984.<br />

[16] Gerhard Piefke. Feldtheorie. Bibliographisches Institut, Mannheim, 1971.<br />

[17] S. Zhu, A. W. Yu, D. Hawley, and R. Roy. Frustrated total <strong>in</strong>ternal <strong>re</strong> ection: A<br />

demonstration and <strong>re</strong>view. American Journal of Physics, 54:601{607, 1986.<br />

[18] Aephraim M. Ste<strong>in</strong>berg and Raymond Y. Chiao. Tunnel<strong>in</strong>g delay times <strong>in</strong> one and two<br />

dimensions. Physical Review A, 49(5):3283{3295, 1994.<br />

[19] L. Brillou<strong>in</strong>. Uber die Fortp anzung des Lichtes <strong>in</strong> dispergie<strong>re</strong>nden Medien. Annalen<br />

der Physik, 44(10):203{240, 1914.<br />

[20] James R. Wait. <strong>Propagation</strong> of Pulses <strong>in</strong> Dispersive <strong>Media</strong>. Radio Science J.Res. NBS,<br />

69D(11):1387{1401, 1965.<br />

[21] Hans Georg Baerwald. Uber die Fortp anzung von Signalen <strong>in</strong> dispergie<strong>re</strong>nden Systemen.<br />

Annalen der Physik, 7:731{760, 1930.<br />

[22] Hans Georg Baerwald. Uber die Fortp anzung von Signalen <strong>in</strong> dispergie<strong>re</strong>nden Systemen.<br />

Annalen der Physik, 6:295{369, 1930.<br />

[23] Dennis B. Trizna and Thomas A. Weber. Brillou<strong>in</strong> <strong>re</strong>visited: Signal velocity de nition<br />

for pulse propagation <strong>in</strong> a medium with <strong>re</strong>sonant anomalous dispersion. Radio Science,<br />

17(5):1169{1180, 1982.<br />

[24] Kurt Edmund Oughstun and George C. Sherman. <strong>Propagation</strong> of electromagnetic pulses<br />

<strong>in</strong> a l<strong>in</strong>ear dispersive medium with absorption (the Lo<strong>re</strong>ntz medium). J. Opt. Soc. Am.<br />

B, 5(4):817{849, 1988.<br />

[25] Philippe Wyns, Daniel P. Foty, and Kurt E. Oughstun. Numerical analysis of the<br />

p<strong>re</strong>cursor elds <strong>in</strong> l<strong>in</strong>ear dispersive pulse propagation. J. Opt. Soc. Am. A, 6(9):1421{<br />

1429, 1989.<br />

[26] Kurt E. Oughstun, Philippe Wyns, and Daniel P. Foty. Numerical determ<strong>in</strong>ation of the<br />

signal velocity <strong>in</strong> dispersive pulse propagation. J. Opt. Soc. Am. A, 6(9):1430{1440,<br />

1989.<br />

[27] Kurt Edmund Oughstun and George C. Sherman. Uniform asymptotic description<br />

of electromagnetic pulse propagation of electromagnetic pulses <strong>in</strong> a l<strong>in</strong>ear dispersive<br />

medium with absorption (the Lo<strong>re</strong>ntz medium). J. Opt. Soc. Am. A, 6(9):1394{1420,<br />

1989.<br />

[28] Kurt Edmund Oughstun and George C. Sherman. Uniform asymptotic description of<br />

ultrashort <strong>re</strong>ctangular optical pulse propagation <strong>in</strong> a l<strong>in</strong>ear, causally dispersive medium.<br />

Physical Review A, 41(11):6090{6113, 1990.<br />

238


Bibliography<br />

[29] Kurt Edmund Oughstun. Pulse <strong>Propagation</strong> <strong>in</strong> a L<strong>in</strong>ear, Causally Dispersive medium.<br />

Proceed<strong>in</strong>gs of the IEEE, 79(10):1379{1390, 1991.<br />

[30] Constant<strong>in</strong>os M. Balictsis and Kurt Edmund Oughstun. Uniform asymptotic description<br />

of ultrashort Gaussian-pulse propagation <strong>in</strong> a causal, dispersive medium. Physical<br />

Review E, 47(5):3645{3669, 1990.<br />

[31] Shioupyn Shen and Kurt Edmund Oughstun. Dispersive pulse propagation <strong>in</strong> a double<strong>re</strong>sonance<br />

Lo<strong>re</strong>ntz medium. J. Opt. Soc. Am. B, 6(5):948{963, 1989.<br />

[32] Francesco Ma<strong>in</strong>ardi. Energy velocity for hyperbolic dispersive waves. <strong>Wave</strong> Motion,<br />

9:201{208, 1985.<br />

[33] R. Loudon. The propagation of electromagnetic energy through an absorb<strong>in</strong>g medium.<br />

J. Phys. A: Gen. Phys., 3:233{245, 1970.<br />

[34] T. H. Havelock. The <strong>Propagation</strong> of Disturbances <strong>in</strong> Dispersive <strong>Media</strong>, volume 17 of<br />

Cambridge Tract <strong>in</strong> Math. and Phys. Cambridge University P<strong>re</strong>ss, London, 1914.<br />

[35] M. A. Biot. General Theo<strong>re</strong>ms on the Equivalence of Group Velocity and Energy<br />

Transport. Physical Review, 105(4):1129{1137, 1957.<br />

[36] Donovan V. Geppert. Energy-Transport Velocity <strong>in</strong> Electromagnetic <strong>Wave</strong>s. Proceed<strong>in</strong>gs<br />

of the IEEE, 53(11):1790, 1965.<br />

[37] Fritz Borgnis. Zur elektromagnetischen Energiedichte <strong>in</strong> Medien mit Dispersion.<br />

Zeitschrift fur Physik, 159:1{6, 1960.<br />

[38] Fritz Borgnis. Elektrische Wellen auf Verzogerungsleitungen. Elektrotechnische Zeitschrift<br />

A, 79(11):383{385, 1958.<br />

[39] Werner Kleen and Klaus Poschl. Gruppengeschw<strong>in</strong>digkeit und Energiegeschw<strong>in</strong>digkeit<br />

elektromagnetischer Wellen <strong>in</strong> dampfungsf<strong>re</strong>ien Leitungen. Archiv der Elektrischen<br />

Ubertragung, 12(12):567{569, 1958.<br />

[40] Werner Kleen. E<strong>in</strong>fuhrung <strong>in</strong> die Mikrowellen-Elektronik. S. Hirzel Verlag, Stuttgart,<br />

1952.<br />

[41] Werner Kleen, Johannes Labus, and Klaus Poschl. Raumladungswellen. Ergebnisse der<br />

exakten Naturwissenschaften, 29:208{274, 1956.<br />

[42] F. Borgnis. Die Fortp anzungsgeschw<strong>in</strong>digkeit der Energie monochromatischer elektromagnetischer<br />

Wellen <strong>in</strong> dielektrischen Medien. Zeitschrift fur Physik, 117:642{650,<br />

1941.<br />

[43] L. J. F. Broer. On the propagation of energy <strong>in</strong> l<strong>in</strong>ear conservative waves. Applied<br />

scienti c Research, A 2:329{344, 1950.<br />

239


Bibliography<br />

[44] George C. Sherman and Kurt Edmund Oughstun. Description of Pulse Dynamics <strong>in</strong><br />

Lo<strong>re</strong>ntz <strong>Media</strong> <strong>in</strong> Terms of the Energy Velocity and Attenuation of Time-Harmonic<br />

<strong>Wave</strong>s. Physical Review Letters, 47(20):1451{1454, 1981.<br />

[45] Kurt Edmund Oughstun and Shioupyn Shen. Velocity of energy transport for a timeharmonic<br />

eld <strong>in</strong> a multiple-<strong>re</strong>sonance Lo<strong>re</strong>ntz medium. J. Opt. Soc. Am. B, 5(11):2395{<br />

2398, 1988.<br />

[46] George C. Sherman and Kurt Edmund Oughstun. Energy-velocity description of pulse<br />

propagation <strong>in</strong> absorb<strong>in</strong>g, dispersive dielectrics. J. Opt. Soc. Am. B, 12(2):229{247,<br />

1995.<br />

[47] L. C. Baird. Moments of a <strong>Wave</strong> Packet. American Journal of Physics, 40:327{329,<br />

1972.<br />

[48] Dan G. Anderson and Jan I. H. Askne. <strong>Wave</strong> Packets <strong>in</strong> Strongly Dispersive <strong>Media</strong>.<br />

Proceed<strong>in</strong>gs of the IEEE, 62(11):1518{1523, 1974.<br />

[49] D. Anderson, J. Askne, and M. Lisak. <strong>Wave</strong> packets <strong>in</strong> an absorptive and strongly<br />

dispersive medium. Physical Review A, 12(4):1546{1552, 1975.<br />

[50] S. C. Bloch. Eighth velocity of light. American Journal of Physics, 45(6):538{549, 1977.<br />

[51] Not only is this a <strong>re</strong>markable anticipation of data transmission at the speed of<br />

light(n<strong>in</strong>g), it is even mo<strong>re</strong> astonish<strong>in</strong>g <strong>in</strong> that it was written long befo<strong>re</strong> the existence<br />

of electromagnetic waves was proved by Hertz <strong>in</strong> 1886. Stifter was, however, very<br />

well awa<strong>re</strong> of electromagnetism, and the<strong>re</strong> a<strong>re</strong> many passages <strong>in</strong> his oeuv<strong>re</strong> that underl<strong>in</strong>e<br />

this <strong>in</strong>te<strong>re</strong>st [150]. Note also the subtle description <strong>in</strong> this citation of what makes<br />

up our contemporary `global village'.<br />

[52] Milton A. Rothman. Th<strong>in</strong>gs that go faster than light. Scienti c American, pages<br />

142{152, July 1960.<br />

[53] A. Enders and G. Nimtz. Evanescent-mode propagation and quantum tunnel<strong>in</strong>g. Physical<br />

Review E, 48(1):632{634, 1993.<br />

[54] A. Enders and G. Nimtz. Photonic-tunnel<strong>in</strong>g experiments. Physical Review B,<br />

47(15):9605{9609, 1993.<br />

[55] A. Ranfagni, D. Mugnai, P. Fabeni, and G. P. Pazzi. Optical-tunnel<strong>in</strong>g time measu<strong>re</strong>s:<br />

a microwave model. Physica B, 175:283{286, 1991.<br />

[56] G. Nimtz. Instantanes Tunneln. Phys. Bl., 49(12):1119{1120, 1993.<br />

[57] P. Thoma and T. Weiland. Wie <strong>re</strong>al ist das Instantane Tunneln? Phys. Bl., 50(4):313,<br />

360{361, 1994.<br />

[58] A. M. Ste<strong>in</strong>berg, P. G. Kwiat, and R. Y. Chiao. Measu<strong>re</strong>ment of the S<strong>in</strong>gle-Photon<br />

Tunnel<strong>in</strong>g Time. Physical Review Letters, 71(5):708{711, 1993.<br />

240


Bibliography<br />

[59] Ch. Spielmann, R. Szipocs, A. St<strong>in</strong>gl, and F. Krausz. Tunnel<strong>in</strong>g of Optical Pulses<br />

through Photonic Band Gaps. Physical Review Letters, 73(17):2308{2311, 1994.<br />

[60] A. Ranfagni, P. Fabeni, G. P. Pazzi, and D. Mugnai. Anomalous pulse delay <strong>in</strong> microwave<br />

propagation: A plausible connection to the tunnel<strong>in</strong>g time. Physical Review E,<br />

48(2):1453{1460, 1993.<br />

[61] Th. Mart<strong>in</strong> and R. Landauer. Time delay ofevanescent electromagnetic waves and the<br />

analogy to particle tunnel<strong>in</strong>g. Physical Review A, 45(4):2611{2617, 1992.<br />

[62] J. M. Deutch and F. E. Low. Barrier Penetration and Superlum<strong>in</strong>al Velocity. Annals<br />

of Physics, 228:184{202, 1993.<br />

[63] Klaus Hass and Paul Busch. Causality of superlum<strong>in</strong>al barrier traversal. Physics Letters<br />

A, 185:9{13, 1994.<br />

[64] Mark Ya. Azbel'. Superlum<strong>in</strong>al Velocity, Tunnel<strong>in</strong>g Traversal Time and Causality. Solid<br />

State Communications, 91(6):439{441, 1994.<br />

[65] Thorsten Emig. <strong>Propagation</strong> of an electromagnetic pulse through a waveguide with a<br />

barrier: A time doma<strong>in</strong> solution with<strong>in</strong> classical electrodynamics. Physical Review E,<br />

54(5):5780{5787, 1993.<br />

[66] Eric L. Bolda, Raymond Y. Chiao, and John C. Garrison. Two theo<strong>re</strong>ms for the group<br />

velocity <strong>in</strong> dispersive media. Physical Review A, 48(5):3890{3893, 1993.<br />

[67] John S. Toll. Causality and the Dispersion Relation: Logical Foundations. Physical<br />

Review, 104(6):1760{1770, 1956.<br />

[68] Aephraim M. Ste<strong>in</strong>berg. Comment on \Photonic tunnel<strong>in</strong>g times". J. Phys. I France,<br />

4:1813{1816, 1994.<br />

[69] Raymond Y. Chiao, Paul G. Kwiat, and Aephraim Ste<strong>in</strong>berg. Faster Than Light?<br />

Scienti c American, pages 52{60, August 1993.<br />

[70] Rolf Landauer. Light faster than light? Natu<strong>re</strong>, 365:692{693, 1993.<br />

[71] Gunter Nimtz and W<strong>in</strong>fried Heitmann. Superlum<strong>in</strong>al Photonic Tunnel<strong>in</strong>g and Quantum<br />

Electronics. Prog<strong>re</strong>ss <strong>in</strong> Quantum Electronics, 21(2):81{108, 1997.<br />

[72] Gunter Nimtz. Schneller als das Licht? Physik <strong>in</strong> unse<strong>re</strong>r Zeit, 28(5):214{218, 1997.<br />

[73] Raymond Y. Chiao. Superlum<strong>in</strong>al (but causal) propagation of wave packets <strong>in</strong> transpa<strong>re</strong>nt<br />

media with <strong>in</strong>verted atomic populations. Physical Review A, 48(1):R34{R37,<br />

1993.<br />

[74] Eric L. Bolda, John C. Garrison, and Raymond Y. Chiao. Optical pulse propagation at<br />

negative group velocities due to a nearby ga<strong>in</strong> l<strong>in</strong>e. Physical Review A, 49(4):2938{2947,<br />

1994.<br />

241


Bibliography<br />

[75] Aephraim M. Ste<strong>in</strong>berg and Raymond Y. Chiao. Dispersionless, highly superlum<strong>in</strong>al<br />

propagation <strong>in</strong> a medium with a ga<strong>in</strong> doublet. Physical Review A, 49(3):2071{2075,<br />

1994.<br />

[76] Wang Yun-p<strong>in</strong>g and Zhang Dian-l<strong>in</strong>. Reshap<strong>in</strong>g, path uncerta<strong>in</strong>ty, and superlum<strong>in</strong>al<br />

travell<strong>in</strong>g. Physical Review A, 52(4):2597{2600, 1995.<br />

[77] Vijay K. Garg and Joseph E. Wilkes, editors. Wi<strong>re</strong>less and personal communication<br />

systems. P<strong>re</strong>ntice-Hall, Upper Saddle River, NJ, 1996.<br />

[78] W. D. Rummler, R. P. Coutts, and M. L<strong>in</strong>iger. Multipath Fad<strong>in</strong>g Channel Models for<br />

Microwave Digital Radio. IEEE Communications Magaz<strong>in</strong>e, 24(11):30{42, 1986.<br />

[79] Rolf Landauer. Barrier traversal time. Natu<strong>re</strong>, 341:567{568, 1989.<br />

[80] Thomas E. Hartman. Tunnel<strong>in</strong>g of a <strong>Wave</strong> Packet. Journal of Applied Physics,<br />

33(12):3427{3433, 1962.<br />

[81] J. R. Fletcher. Time delay <strong>in</strong> tunnell<strong>in</strong>g through a potential barrier. J. Phys. C: Solid<br />

State Phys., 18:L55{L59, 1985.<br />

[82] K. K. Thornber, Thomas C. McGill, and C. A. Mead. The Tunnel<strong>in</strong>g Time of an<br />

Electron. Journal of Applied Physics, 38(5):2384{2385, 1967.<br />

[83] S. Coll<strong>in</strong>s, David Lowe, and J. R. Barker. The quantum mechanical tunnell<strong>in</strong>g time<br />

problem|<strong>re</strong>visited. J. Phys. C: Solid State Phys., 20:6213{6232, 1987.<br />

[84] John R. Barker. Quantum Theory of Hot Electron Tunnell<strong>in</strong>g <strong>in</strong> Microstructu<strong>re</strong>s. Physica,<br />

134B:22{31, 1985.<br />

[85] E. H. Hauge and J. A. St vneng. Tunnel<strong>in</strong>g times: a critical <strong>re</strong>view. Reviews of Modern<br />

Physics, 61(4):917{936, 1989.<br />

[86] R. Landauer. Barrier <strong>in</strong>teraction time <strong>in</strong> tunnel<strong>in</strong>g. Reviews of Modern Physics,<br />

66(1):217{228, 1994.<br />

[87] J. R. Barker, S. Brouard, V. Gasparian, G. Iannaccone, A. P. Jauho, C. R. Leavens,<br />

J. G. Muga, R. Sala, and D. Sokolovski. First European Workshop on Tunnell<strong>in</strong>g Times.<br />

Phantoms Newsletter, pages 5{11, October 1994.<br />

[88] M. Buttiker. Larmor p<strong>re</strong>cession and the traversal time for tunnel<strong>in</strong>g. Physical Review<br />

B, 27(10):6178{6188, 1983.<br />

[89] Aephraim M. Ste<strong>in</strong>berg. How Much Time Does a Tunnel<strong>in</strong>g Particle Spend <strong>in</strong> the<br />

Barrier Region? Physical Review Letters, 74(13):2405{2409, 1995.<br />

[90] Aephraim M. Ste<strong>in</strong>berg. Conditional probabilities <strong>in</strong> quantum theory and the tunnel<strong>in</strong>gtime<br />

controversy. Physical Review A, 52(1):32{42, 1995.<br />

242


Bibliography<br />

[91] C. R. Leavens and G. C. Aers. Bohm Trajectories and the Tunnel<strong>in</strong>g Time Problem.<br />

In Roland Wiesendanger and Hans-Joachim Guntherodt, editors, Scann<strong>in</strong>g Tunnel<strong>in</strong>g<br />

Microscopy III, pages 105{140. Spr<strong>in</strong>ger, Berl<strong>in</strong>, 1996.<br />

[92] Fritz Paschke. Modellbildung. lectu<strong>re</strong> notes, 1993.<br />

[93] Roger F. Harr<strong>in</strong>gton. Time-harmonic Electromagnetic Fields. McGraw-Hill, New York,<br />

1961.<br />

[94] Paul Lorra<strong>in</strong> and Dale R. Corson. Electromagnetic Fields and <strong>Wave</strong>s. W. H. F<strong>re</strong>eman,<br />

San Francisco, 1970.<br />

[95] Arnulf We<strong>in</strong>z<strong>in</strong>ger. Die Exponentialleitung als Impedanztransformator und ih<strong>re</strong> Realisierung<br />

als Microstrip-Leitung. Diploma thesis, Technische Hochschule Wien, 1971.<br />

[96] Charles M. Knop. Pulsed Electromagnetic <strong>Wave</strong> <strong>Propagation</strong> <strong>in</strong> Dispersive <strong>Media</strong>.<br />

IEEE Transactions on Antennas and <strong>Propagation</strong>, 12:494{496, 1964.<br />

[97] Richard E. Haskell and Carl T. Case. Transient Signal <strong>Propagation</strong> <strong>in</strong> Lossless, Isotropic<br />

Plasma. IEEE Transactions on Antennas and <strong>Propagation</strong>, 15(3):458{464, 1967.<br />

[98] R. E. McIntosh and Alfonso Malaga. Time dispersion of electromagnetic pulses by the<br />

ionosphe<strong>re</strong>. Radio Science, 15(3):645{654, 1980.<br />

[99] Junichi Takanaka, Suguru Yamagishi, Hiroaki Taki, Akihiro Morimoto, and Tetsuro<br />

Kobayashi. Guid<strong>in</strong>g of a one-dimensional optical beam with nanometer diameter. Optics<br />

Letters, 22(7):475{477, 1997.<br />

[100] Fritz Paschke. Zum Verhalten der St<strong>re</strong>ifenleitung bei Submikrometerabstanden. lectu<strong>re</strong><br />

notes, 1997.<br />

[101] G. Diener. Superlum<strong>in</strong>al group velocities and <strong>in</strong>formation transfer. Physics Letters A,<br />

223:327{331, 1996.<br />

[102] W. Heitmann and G. Nimtz. On causality proofs of superlum<strong>in</strong>al barrier traversal of<br />

f<strong>re</strong>quency band limited wave packets. Physics Letters A, 196:154{158, 1994.<br />

[103] Arm<strong>in</strong> Hermann, editor. Albert E<strong>in</strong>ste<strong>in</strong> / Arnold Sommerfeld, Briefwechsel. Schwabe<br />

& Co, Basel, 1968.<br />

[104] Marco A. de Moura and Douglas F. de Albuquerque. Remarks on the traversal time <strong>in</strong><br />

a tunnel<strong>in</strong>g process. Solid State Communications, 74(5):353{354, 1990.<br />

[105] Thomas Weiland. Die Disk<strong>re</strong>tisierung der Maxwell-Gleichungen. Phys. Bl., 42(7):191{<br />

201, 1986.<br />

[106] Richard Albanese, John Penn, and Richard Med<strong>in</strong>a. Short-rise-time microwave pulse<br />

propagation through dispersive biological media. J. Opt. Soc. Am. A, 6(9):1441{1446,<br />

1989.<br />

243


Bibliography<br />

[107] Abraham Goldberg, Harry M. Schey, and Judah L. Schwartz. Computer-Generated<br />

Motion Pictu<strong>re</strong>s of One-Dimensional Quantum-Mechanical Transmission and Re ection<br />

Phenomena. American Journal of Physics, 35(3):177{186, 1967.<br />

[108] S. Coll<strong>in</strong>s, David Lowe, and J. R. Barker. A dynamic analysis of <strong>re</strong>sonant tunnell<strong>in</strong>g.<br />

J. Phys. C: Solid State Phys., 20:6233{6243, 1987.<br />

[109] Terry Robb. Schroed<strong>in</strong>ger's Equation. Mathematica notebook available on MathSource,<br />

1994. URL:http://www.mathsource.com/cgi-b<strong>in</strong>/MathSource/Enhancements/<br />

Interfac<strong>in</strong>g/InterCall/0204-657.<br />

[110] John R. Hiller, Ian D. Johnston, and Daniel F. Styer. Quantum Mechanics Simulations,<br />

The Consortium for Upper-Level Physics Softwa<strong>re</strong>. John Wiley & Sons, New York, 1995.<br />

[111] John R. Merrill. <strong>Wave</strong>s <strong>in</strong> Dispersive <strong>Media</strong>: Another Use of Computers <strong>in</strong> Introductory<br />

Physics. American Journal of Physics, 39:539{544, 1971.<br />

[112] Milton Abramowitz. On the Practical Evaluation of Integrals. SIAM Journal of Applied<br />

Mathematics, 2:20{35, 1954.<br />

[113] Philip J. Davis and Philip Rab<strong>in</strong>owitz. Methods of numerical <strong>in</strong>tegration. Academic<br />

P<strong>re</strong>ss, Inc., Orlando, Florida, second edition, 1984.<br />

[114] David Lev<strong>in</strong>. Procedu<strong>re</strong>s for comput<strong>in</strong>g one and two dimensional <strong>in</strong>tegrals of functions<br />

with rapid ir<strong>re</strong>gular oscillations. Mathematics of Computation, 38(158):531{538, 1982.<br />

[115] G. A. Evans. Two robust methods for ir<strong>re</strong>gular oscillatory <strong>in</strong>tegrals over a nite range.<br />

Applied Numerical Mathematics, 14(4):383{395, 1994.<br />

[116] L. N. G. Filon. On a quadratu<strong>re</strong> formula for trigonometric <strong>in</strong>tegrals. Proc. Roy. Soc.<br />

Ed<strong>in</strong>burgh, 49:38{47, 1928.<br />

[117] Ulf Torsten Eh<strong>re</strong>nmark. A th<strong>re</strong>e-po<strong>in</strong>t formula for numerical quadratu<strong>re</strong> of oscillatory<br />

<strong>in</strong>tegrands with variable f<strong>re</strong>quency. Journal of Computational and Applied Mathematics,<br />

21:87{99, 1988.<br />

[118] Pei-Cheng Xu and A. K. Mal. An adaptive <strong>in</strong>tegration scheme for ir<strong>re</strong>gularly oscillatory<br />

functions. <strong>Wave</strong> Motion, 7:235{243, 1985.<br />

[119] K. T. R. Davies. Complex-plane methods for evaluat<strong>in</strong>g <strong>in</strong>tegrals with highly oscillatory<br />

<strong>in</strong>tegrands. Journal of Computational Physics, 80(2):498{505, 1989.<br />

[120] Philip Rab<strong>in</strong>owitz. Extrapolation methods <strong>in</strong> numerical <strong>in</strong>tegration. Numerical Algorithms,<br />

3:17{28, 1992.<br />

[121] J. N. Lyness. Integrat<strong>in</strong>g some <strong>in</strong> nite oscillat<strong>in</strong>g tails. Journal of Computational and<br />

Applied Mathematics, 12 & 13:109{117, 1985.<br />

[122] James Lyness and Gwendolen H<strong>in</strong>es. To Integrate Some In nite Oscillat<strong>in</strong>g Tails. ACM<br />

Transactions on Mathematical Softwa<strong>re</strong>, 12(1):24{25, 1986.<br />

244


Bibliography<br />

[123] T. O. Espelid and K. J. Overholt. DQAINF: An algorithm for automatic <strong>in</strong>tegration of<br />

<strong>in</strong> nite oscillat<strong>in</strong>g tails. Numerical Algorithms, 8(2):83{101, 1994.<br />

[124] Q. Haider and L. C. Liu. Fourier or Bessel transformations of highly oscillatory functions.<br />

J. Phys. A: Math. Gen., 25(24):6755{6760, 1992.<br />

[125] Avram Sidi. Extrapolation methods for oscillatory <strong>in</strong> nite <strong>in</strong>tegrals. J. Inst. Maths<br />

Applics, 26:1{20, 1980.<br />

[126] Avram Sidi. The Numerical Evaluation of Very Oscillatory In nite Integrals by Extrapolation.<br />

Mathematics of Computation, 38(158):517{529, 1982.<br />

[127] Takemitsu Hasegawa and Tatsuo Torii. Inde nite <strong>in</strong>tegration of oscillatory functions by<br />

the Chebyshev series expansion. Journal of Computational and Applied Mathematics,<br />

17:21{29, 1987.<br />

[128] Avram Sidi. A user-friendly extrapolation method for oscillatory <strong>in</strong> nite <strong>in</strong>tegrals.<br />

Mathematics of Computation, 51(183):249{266, 1988.<br />

[129] Ulf Torsten Eh<strong>re</strong>nmark. A note on an extension of extrapolative techniques for a class<br />

of <strong>in</strong> nite oscillatory <strong>in</strong>tegrals. BIT, 30(1):152{155, 1990.<br />

[130] Bui Doan Khanh. A numerical evaluation of very oscillatory <strong>in</strong> nite <strong>in</strong>tegrals by asymptotics.<br />

Applied Mathematics and Computation, 40(2):117{124, 1990.<br />

[131] Christoph Uberhuber. Computer-Numerik. Spr<strong>in</strong>ger, Berl<strong>in</strong>, 1995.<br />

[132] R. Piessens, E. de Doncker-Kapenga, C. W. Uberhuber, and D. K. Kahaner. QUAD-<br />

PACK, A Subrout<strong>in</strong>e Package for Automatic Integration, volume 1 of Series <strong>in</strong> Computational<br />

Math. Spr<strong>in</strong>ger, Berl<strong>in</strong>, 1983.<br />

[133] P. Wynn. On a device for comput<strong>in</strong>g the em (Sn) transformation. Math. Comp., 10:91{<br />

96, 1956.<br />

[134] Numerical Algorithms Group, NAG O ce, Oxford. NAG Libary, Mark 17, 1996. URL:<br />

http://www.nag.co.uk.<br />

[135] International Mathematical and Statistical Libraries, Inc., Houston, TX. IMSL Library,<br />

V3.0, 1996. URL: http://www.vni.com/adt.dir/adt.html.<br />

[136] John P. Boyd. A Lag-Averaged Genralization of Euler's Method for Accelerat<strong>in</strong>g Series.<br />

Applied Mathematics and Computation, 72:143{166, 1995.<br />

[137] Claude B<strong>re</strong>z<strong>in</strong>ski and Michela Redivo Zaglia. Extrapolation methods: theory and practice.<br />

North-Holland, Amsterdam, 1991.<br />

[138] C. B<strong>re</strong>z<strong>in</strong>ski and M. Redivo-Zaglia. Extrapolation methods. Applied Numerical Mathematics,<br />

15(2):123{131, 1994.<br />

245


Bibliography<br />

[139] R. Riedel. E<strong>in</strong>fuhrung <strong>in</strong> die Theorie der Konvergenzbeschleunigung fur Reihen. Wissenschaftliche<br />

Beitrage, Mart<strong>in</strong>-Luther-Universitat Halle-Wittenburg, 46:43{61, 1988.<br />

[140] Stephen Wolfram. Mathematica, A System for Do<strong>in</strong>g Mathematics by Computer.<br />

Addison-Wesley, Read<strong>in</strong>g, MA, second edition, 1991.<br />

[141] David A. Smith and William F. Ford. Acceleration of l<strong>in</strong>ear and logarithmic convergence.<br />

J. SIAM Numer. Anal., 16(2):223{240, 1979.<br />

[142] G. E. Bell and G. M. Phillips. Aitken acceleration of some alternat<strong>in</strong>g series. BIT,<br />

24:70{77, 1984.<br />

[143] P. Wynn. On the convergence and stability of the epsilon algorithm. J. SIAM Numer.<br />

Anal., 3(1):91{122, 1966.<br />

[144] Jerry Keiper. Numerical Computation II. In Repr<strong>in</strong>t from The Mathematica Confe<strong>re</strong>nce,<br />

Boston, MA, June 1992, URL: http://www.mathsource.com/cgi-b<strong>in</strong>/MathSource/<br />

Enhancements/Numerical/0204-028, 1992. Wolfram Research, Inc.<br />

[145] Jerry Keiper. The N functions of Mathematica. In Repr<strong>in</strong>t from The Mathematica<br />

Confe<strong>re</strong>nce, Boston, MA, June 1992, URL: http://www.wolfram.com/cgi-b<strong>in</strong>/<br />

MathSource/General/Tutorials/Numerical/0203-948, 1992. Wolfram Research, Inc.<br />

[146] Roman Maeder. Programm<strong>in</strong>g <strong>in</strong> Mathematica. Addison-Wesley, Read<strong>in</strong>g, MA, second<br />

edition, 1991.<br />

[147] William T. Shaw and Jason Tigg. Applied Mathematica, Gett<strong>in</strong>g Started, Gett<strong>in</strong>g It<br />

Done. Addison-Wesley, Read<strong>in</strong>g, MA, 1994.<br />

[148] Milton Abramowitz and I<strong>re</strong>ne A. Stegun. Handbook of Mathematical Functions. Dover<br />

Publications, New York, n<strong>in</strong>th edition, 1972.<br />

[149] Wolfgang Grobner and Nikolaus Hof<strong>re</strong>iter. Integraltafel. Spr<strong>in</strong>ger, Wien, fourth edition,<br />

1966.<br />

[150] Fritz Paschke. Walter Schottkys erste Arbeitsgebiete. Siemens Forsch.- und Entwickl.-<br />

Ber., 15(6):287{290, 1986.<br />

246


Index<br />

Index<br />

absorption, 7, 9<br />

accuracy<br />

with<strong>in</strong> Mathematica, 141, 156<br />

approximation<br />

of <strong>in</strong>tegrands, 125<br />

approximation error, 137, 147, 150<br />

estimation, 156, 178<br />

asymptotic expansion, 127, 128, 145, 178<br />

Buttiker time, 97<br />

band-limited signal, 25, 71, 100<br />

Bohm trajectory, 30, 100<br />

causality, 21, 71, 107<br />

and dispersion, 24<br />

and Gaussian pulses, 24<br />

centroid, 6, 23, 28<br />

centrovelocity, 18<br />

Clenshaw-Curtis rule, 125, 128<br />

cont<strong>in</strong>uity equation, 15<br />

convergence<br />

l<strong>in</strong>ear, 130<br />

convergence acceleration, 126, 129<br />

convolution <strong>in</strong>tegral, 70<br />

cor<strong>re</strong>lation velocity, 18<br />

Crank-Nicholson algorithm, 121<br />

cuto f<strong>re</strong>quency<br />

optical, 63<br />

plasma, 50, 66<br />

quantum mechanical, 73, 95<br />

wave guide, 22, 61<br />

2 algorithm, 130, 149<br />

<strong>re</strong>liability, 135<br />

delay l<strong>in</strong>e, 15, 21, 35<br />

dispersion, 4, 7<br />

and causality, 24<br />

247<br />

anomalous, 7, 9, 16<br />

dispersion <strong>re</strong>lation, 8, 24<br />

at negative f<strong>re</strong>quencies, 65<br />

at negative wavenumbers, 74, 192<br />

delay l<strong>in</strong>e, 36<br />

exponential l<strong>in</strong>e, 46<br />

l<strong>in</strong>ear approximation, 5, 68<br />

Lo<strong>re</strong>ntz medium, 11<br />

plasma, 41, 51, 64<br />

quantum mechanical, 73, 117<br />

transmission l<strong>in</strong>e, 33<br />

wave guide, 21, 61<br />

distortion, 5, 68<br />

double exponential rule, 141, 145, 171<br />

dwell time, 29, 96, 97<br />

e ective, 98<br />

-algorithm, 128, 130, 149<br />

<strong>re</strong>liability, 135<br />

energy<br />

propagated, 33, 38, 46<br />

sto<strong>re</strong>d, 15, 33, 36, 38, 46<br />

energy velocity, 14, 22, 29, 98<br />

connection to group velocity, 16<br />

e ective, 42<br />

electromagnetic wave, 15<br />

<strong>in</strong> the pass band, 41, 44, 47<br />

<strong>in</strong> the stop band, 42, 44, 47<br />

negative, 25<br />

of a cur<strong>re</strong>nt pulse, 39<br />

term<strong>in</strong>ated transmission l<strong>in</strong>e, 38<br />

envelope, 5, 18, 70<br />

error<br />

absolute vs. <strong>re</strong>lative, 142, 158<br />

Euler transformation, 128, 130, 135<br />

evanescence, 10


<strong>in</strong> a plasma, 41<br />

<strong>in</strong> wave guides, 22, 61<br />

quantum mechanical, 27, 73<br />

exponential l<strong>in</strong>e, 46<br />

extrapolation<br />

adaptive accuracy control, 155, 172<br />

<strong>in</strong>c<strong>re</strong>as<strong>in</strong>g the accuracy, 147, 152<br />

Mathematica implementation, 149<br />

of sequences, 125<br />

performance comparison, 150<br />

Fast Fourier Transform, 18, 120<br />

Feynman path, 30<br />

ux, 29, 98<br />

Fourier <strong>in</strong>tegral, 4, 11, 50, 64, 73, 211<br />

approximate solutions, 119<br />

numerical evaluation, 120<br />

f<strong>re</strong>e particle, 81, 96, 114<br />

Gauss rule, 128<br />

Gauss-Kronrod rule, 141, 145<br />

Gaussian pulse, 18, 64<br />

G<strong>re</strong>en's function, 71<br />

group delay, 28<br />

group velocity, 4, 8, 22, 28, 68, 96, 102, 112<br />

abnormal, 24<br />

connection to phase velocity, 21<br />

<strong>in</strong> a plasma, 42, 54<br />

k<strong>in</strong>ematic derivation, 6<br />

transmission l<strong>in</strong>e, 33<br />

high-pass lter e ect, 57, 102, 117<br />

<strong>in</strong>dex of <strong>re</strong>fraction, 8, 24<br />

<strong>in</strong>ductive wall, 36<br />

<strong>in</strong>formation transmission, 24, 70<br />

ionosphe<strong>re</strong><br />

dispersive e ects, 50<br />

Laplace transform, 120<br />

Larmor time, 97<br />

light cone, 71<br />

Lo<strong>re</strong>ntz medium, 10, 24<br />

matu<strong>re</strong> dispersion, 12<br />

248<br />

modulation, 4, 48<br />

monochromatic wave, 3, 27, 96<br />

multipath propagation, 25<br />

on-o -key<strong>in</strong>g, 49<br />

oscillat<strong>in</strong>g <strong>in</strong>tegrand, 125, 141<br />

Index<br />

partial sums<br />

sequence of, 133, 138, 146<br />

partition<br />

asymptotic, 145, 175<br />

of <strong>in</strong>tegrals, 126, 133<br />

start<strong>in</strong>g po<strong>in</strong>t, 138, 168<br />

zeros vs. ext<strong>re</strong>ma, 135, 146<br />

phase time, 28, 97, 102, 115<br />

phase velocity, 3, 17, 34, 60, 68<br />

connection to group velocity, 21<br />

plasma<br />

lossless, 41, 48, 64<br />

plasma f<strong>re</strong>quency, 41, 54<br />

p<strong>re</strong>cision<br />

arbitrary, 156<br />

with<strong>in</strong> Mathematica, 141, 156<br />

p<strong>re</strong>cursor, 11, 17<br />

probability density, 27, 79<br />

pulse peak, 24<br />

temporal vs. spatial, 18, 102, 104<br />

trajectory, 28, 88, 100, 107, 112<br />

pulse <strong>re</strong>shap<strong>in</strong>g, 25, 66, 71, 90, 112, 115<br />

quadratu<strong>re</strong>, 125<br />

computer rout<strong>in</strong>es, 127<br />

performance comparison, 141<br />

quadratu<strong>re</strong> module, 214<br />

application example, 197, 221<br />

Mathematica implementation, 161<br />

options, 162, 167<br />

partition strategy, 162, 170<br />

program structu<strong>re</strong>, 164<br />

test examples, 181<br />

test of application example, 205<br />

quantum clock, 30, 97<br />

<strong>re</strong> ection coe cient, 29, 73, 192<br />

<strong>re</strong>sonance, 7


Index<br />

saddle po<strong>in</strong>t <strong>in</strong>tegration, 11, 119<br />

scatter<strong>in</strong>g<br />

examples, 81, 221<br />

vs. escap<strong>in</strong>g, 28<br />

Schrod<strong>in</strong>ger equation, 27, 212<br />

time-<strong>in</strong>dependent, 27, 73<br />

semi-classical time, 96<br />

sequence<br />

alternat<strong>in</strong>g, 126, 129, 133<br />

speed of convergence, 137<br />

transformation, 129<br />

Shanks transformation, 130<br />

signal<br />

small-band, 6, 28, 96<br />

signal velocity, 10<br />

for Gaussian pulses, 14<br />

for <strong>re</strong>ctangular pulses, 13<br />

measu<strong>re</strong>ment of, 13<br />

Simpson rule, 125<br />

spectrum<br />

broad, 49, 85<br />

energy and momentum, 76<br />

Gaussian pulse, 64<br />

Gaussian wave packet, 78<br />

narrow, 6, 16, 22, 92, 112, 116<br />

<strong>re</strong>ctangular wave packet, 76<br />

triangular wave packet, 77<br />

wave number vs. f<strong>re</strong>quency, 51<br />

squa<strong>re</strong> barrier, 27, 93, 121<br />

stationary phase<br />

method of, 11, 16, 28, 140<br />

step barrier, 73, 81, 122, 191<br />

physical dimensions, 84<br />

subdivision<br />

adaptive, 128, 141<br />

e ect on convergence acceleration, 133<br />

superlum<strong>in</strong>al velocity, 21, 70<br />

experiments, 22<br />

TE mode, 60<br />

TEM wave, 35, 41, 49, 60, 63<br />

term<strong>in</strong>ation, 37<br />

total <strong>in</strong>ternal <strong>re</strong> ection, 10<br />

transfer function, 65, 70<br />

249<br />

transmission coe cient, 29, 73, 95, 104, 192<br />

transmission l<strong>in</strong>e, 17, 32<br />

characteristic impedance, 33, 42, 44, 46<br />

dispersion-f<strong>re</strong>e, 34<br />

equivalence to a wave guide, 44<br />

equivalent circuit, 32, 49<br />

f<strong>re</strong>quency dependence of losses, 34<br />

<strong>in</strong>homogeneous, 46<br />

trapezoidal rule, 121, 125, 128, 141<br />

travell<strong>in</strong>g-wave tube, 16<br />

truncation<br />

of <strong>in</strong>tegrals, 125, 196<br />

tunnel e ect, 21, 26<br />

analogy to wave guides, 26, 117<br />

examples, 100, 229<br />

tunnell<strong>in</strong>g time, 28, 96, 112<br />

negative, 107, 116<br />

uncerta<strong>in</strong>ty pr<strong>in</strong>ciple, 49<br />

W -transformation, 126, 147<br />

wave equation, 3<br />

wave front velocity, 11, 12, 24, 34, 41, 52,<br />

57, 63, 71<br />

wave function<br />

normalisation, 27<br />

wave guide, 7, 10, 21, 44, 60<br />

wave number, 3, 8, 50, 54<br />

quantum mechanical, 27<br />

wave packet, 4, 16, 76, 129, 149<br />

cent<strong>re</strong> of gravity, 6, 17, 23<br />

Gaussian, 88, 96, 100, 121, 196<br />

moments of a, 18<br />

<strong>re</strong>ctangular, 81, 195<br />

triangular, 85, 122, 193


250


Curriculum vitae<br />

Dipl.-Ing. Thilo Sauter<br />

3. 9. 1967 gebo<strong>re</strong>n <strong>in</strong> Biberach/Ri , Deutschland<br />

1973 { 1977 Besuch der Volksschule <strong>in</strong> Biberach/Ri<br />

September 1977 Ubersiedlung nach Oster<strong>re</strong>ich<br />

1977 { 1979 Besuch der Hauptschule Miem<strong>in</strong>g, Tirol<br />

1979 { 1985 Besuch des <strong>re</strong>alistischen Bundesymnasiums St. Johann/Pongau<br />

3. 6. 1985 Matura mit Auszeichnung<br />

Oktober 1985 Beg<strong>in</strong>n des Studiums der Elektrotechnik, Studienzweig Industrielle Elektronik<br />

und Regelungstechnik an der Technischen Universitat Wien<br />

19. 3. 1992 Abschlu des Studiums mit Auszeichnung, Diplomarbeit XR-III-Bus<br />

und XR-III-Bus-Controller uber die Entwicklung e<strong>in</strong>es fehlertoleranten<br />

Sensor-Aktor-Feldbusses<br />

1992 { 1996 Tatigkeit als Vertragsassistent am Institut fur Allgeme<strong>in</strong>e Elektrotechnik<br />

und Elektronik der TU Wien<br />

Aufbau und Bet<strong>re</strong>uung von Laborubungen zu den Themen Aufbau von<br />

Operationsverstarkern sowie Paramete<strong>re</strong>xtraktion zur Modellierung von<br />

MOS-Transisto<strong>re</strong>n<br />

1992 { 1993 Realisierung e<strong>in</strong>es fehlertoleranten Feldbus-Prototyps auf der Basis von<br />

programmierba<strong>re</strong>n Logikbauste<strong>in</strong>en<br />

ab 1993 Durchfuhrung von Industrie- und Forschungsprojekten auf den Gebieten<br />

Entwurf gedruckter Schaltungen<br />

Entwurf programmierba<strong>re</strong>r <strong>in</strong>tegrierter Schaltungen<br />

Codierung digitaler Signale<br />

ab 1994 Entwurf analoger <strong>in</strong>tegrierter Schaltungen<br />

Beg<strong>in</strong>n der Dissertation<br />

ab 1996 Universitatsassistent am Institut fur Computertechnik der TU Wien<br />

Leiter des Kompetenzzentrums fur Feldbussysteme<br />

Lehrauftrag zum Thema Entwurf <strong>in</strong>tegrierter Schaltungen<br />

251

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!