12.07.2015 Views

1 Studies in the History of Statistics and Probability ... - Sheynin, Oscar

1 Studies in the History of Statistics and Probability ... - Sheynin, Oscar

1 Studies in the History of Statistics and Probability ... - Sheynin, Oscar

SHOW MORE
SHOW LESS

Create successful ePaper yourself

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

<strong>of</strong> <strong>the</strong> ma<strong>the</strong>matical model <strong>of</strong> <strong>the</strong> Brownian motion allow<strong>in</strong>g todeterm<strong>in</strong>e precisely <strong>the</strong> coefficient <strong>of</strong> diffusion given observations <strong>of</strong> ahowever small <strong>in</strong>terval <strong>of</strong> a realization <strong>of</strong> that motion. He who believesthat this is <strong>in</strong>deed true for a physical Brownian motion will be wrong.Here is ano<strong>the</strong>r such example. Any broadcast<strong>in</strong>g station istransmitt<strong>in</strong>g over a waveb<strong>and</strong> <strong>of</strong> restricted width. If a radio signal isconsidered as a stochastic process, its spectrum will be conta<strong>in</strong>ed <strong>in</strong>that f<strong>in</strong>ite <strong>in</strong>terval. And <strong>the</strong>re exists a ma<strong>the</strong>matical <strong>the</strong>orem stat<strong>in</strong>gthat with probability 1 a realization ξ i <strong>of</strong> such a process is an analyticalfunction <strong>of</strong> t. Consequently, after listen<strong>in</strong>g for any however short<strong>in</strong>terval <strong>of</strong> time, we may unambiguously establish what was <strong>and</strong> whatwill be broadcast, an obviously absurd conclusion.It is certa<strong>in</strong>ly easy to <strong>in</strong>dicate <strong>the</strong> mistake here. First, a broadcast isnot a stochastic process s<strong>in</strong>ce it is not an element <strong>of</strong> some statisticalensemble; second, an analytical function can be reconstructed given itsvalues on any <strong>in</strong>terval only if <strong>the</strong>y are given absolutely preciselywhich is impossible for a function <strong>of</strong> a cont<strong>in</strong>uous variable. Even as<strong>in</strong>gle number can not be written down precisely, much less a totality<strong>of</strong> an <strong>in</strong>f<strong>in</strong>itely many numbers. Third, a radio signal is not an analyticalfunction <strong>of</strong> time because <strong>in</strong> <strong>the</strong> 19 th century <strong>the</strong>re were no broadcast<strong>in</strong>gstations whereas an analytical function vanish<strong>in</strong>g on some <strong>in</strong>tervalvanishes everywhere.A digression about <strong>the</strong> concept <strong>of</strong> function <strong>in</strong> ma<strong>the</strong>matics is <strong>in</strong>order here 7 . At <strong>the</strong> emergence <strong>of</strong> ma<strong>the</strong>matical analysis it was usuallyunderstood as a formula determ<strong>in</strong><strong>in</strong>g a dependence y = y(x). And allfunctions except at a few po<strong>in</strong>ts were cont<strong>in</strong>uous <strong>and</strong> differentiable.The problem concern<strong>in</strong>g <strong>the</strong> pro<strong>of</strong> <strong>of</strong> differentiability did not evenexist. Later, however, <strong>in</strong> <strong>the</strong> 19 th century an idea was established that afunction is simply a relation between <strong>the</strong> sets <strong>of</strong> values <strong>of</strong> <strong>the</strong> argumentx <strong>and</strong> <strong>the</strong> function y = y(x). It is usually dem<strong>and</strong>ed that exactly onevalue <strong>of</strong> y corresponded to each value <strong>of</strong> x, but that <strong>the</strong> <strong>in</strong>verse was notnecessarily true. And <strong>the</strong>re was no cause for an arbitrarycorrespondence y = y(x) to be cont<strong>in</strong>uous or differentiable.It is ra<strong>the</strong>r difficult but <strong>the</strong>refore <strong>in</strong>terest<strong>in</strong>g to provide an example<strong>of</strong> a cont<strong>in</strong>uous but nowhere differentiable function. The first suchexample was due to Weierstrass, later o<strong>the</strong>r <strong>and</strong> more simple exampleswere discovered. Such objects proved very <strong>in</strong>terest<strong>in</strong>g forma<strong>the</strong>maticians <strong>and</strong> to <strong>the</strong>m <strong>the</strong>ir attention had been to a large extentswung. For us, it is especially <strong>in</strong>terest<strong>in</strong>g that ma<strong>the</strong>maticalconsiderations concern<strong>in</strong>g <strong>the</strong> <strong>the</strong>ory <strong>of</strong> stochastic processes lead to<strong>the</strong> realization <strong>of</strong> many such processes which should be recognized ascont<strong>in</strong>uous but not differentiable functions (or functions only twice,say, differentiable with a cont<strong>in</strong>uous but not anymore differentiablesecond derivative).This was <strong>in</strong>deed joyful because it apparently proved that nondifferentiablefunctions <strong>in</strong>deed existed <strong>in</strong> nature. However, we wish tocast a shadow on that joy: it is absolutely absurd to believe that such afunction can be experimentally observed. Such a realization <strong>of</strong> astochastic process can not be given ei<strong>the</strong>r by a formula, or a table, agraph, or an algorithm <strong>of</strong> calculation. When consider<strong>in</strong>g it <strong>in</strong>deed real,exactly known at all <strong>of</strong> its po<strong>in</strong>ts, we will be able to come to absurd67

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!