08.09.2014 Views

A comparative study of models for predation and parasitism

A comparative study of models for predation and parasitism

A comparative study of models for predation and parasitism

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

A COMPARATIVE STUDY OF MODELS FOR PREDATION<br />

AND<br />

PARASITISM<br />

by<br />

T. ROYAMA<br />

Canadian Forestry Service, P. O. Box 4000,<br />

Fredericton, New Brunswick, Canada<br />

TABLE OF CONTENTS<br />

Page<br />

1. Introduction 1<br />

2. A brief inquiry into the role <strong>of</strong><br />

<strong>models</strong> in scientific inference 2<br />

3. A background theory <strong>of</strong> the<br />

structure <strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> 9<br />

4. The existing <strong>models</strong> 20<br />

a) The LOTKA-VOLTERRA model 21<br />

b) The NICHOLSON-BAILEY model 24<br />

c) HOLLiNG'S disc equation 31<br />

d) The IVLEV-GAusE equation 35<br />

e) ROYAMA'S model <strong>of</strong> r<strong>and</strong>om<br />

searching <strong>and</strong> probability <strong>of</strong><br />

r<strong>and</strong>om encounters 42<br />

f) WATT'S equation 55<br />

g) The T~o~esoN-SwoY equations<br />

<strong>for</strong> <strong>parasitism</strong> 61<br />

h) The HASSELL-VARLSY model <strong>of</strong><br />

social interference in parasites 66<br />

i) A geometric model <strong>for</strong> social<br />

interaction among parasites<br />

(this <strong>study</strong>) 70<br />

Appendix to w 4i. Is the concept<br />

<strong>of</strong> 'area <strong>of</strong> discovery' useful in<br />

studies <strong>of</strong> <strong>predation</strong> <strong>and</strong><br />

<strong>parasitism</strong> ? 74<br />

j) HOLLING'S hunger model 76<br />

5. Discussion <strong>and</strong> conclusions 78<br />

6. Summary 84<br />

7. Acknowledgements 85<br />

8. References 85<br />

Appendix 1. The pro<strong>of</strong> <strong>of</strong> LI=I/2RX 88<br />

Appendix 2. The pro<strong>of</strong> <strong>of</strong><br />

Lz- ~ (I/2~-,XR)/UX - Re- ~rR~'X 88<br />

I_e-~R~X<br />

Appendix 3. The pro<strong>of</strong> <strong>of</strong> eq. (4i. 6) 89<br />

Appendix 4. List <strong>of</strong> symbols 90<br />

1. INTRODUCTION<br />

ELTON (1935), in his review <strong>of</strong> the works <strong>of</strong> the great biomathematician, the late<br />

ALFRED J. LOTKA, wrote:<br />

"When LOTKA published his first notes on this subject in 1920, animal ecology<br />

had entered on a new phase, though we are probably only now beginning to see the<br />

importance <strong>of</strong> it." However, "like most mathematicians, he takes the hopeful biologist<br />

to the edge <strong>of</strong> a pond, points out that a good swim will help his work, <strong>and</strong> then<br />

pushes him in <strong>and</strong> leaves him to drown."<br />

When NICHOLSON <strong>and</strong> BAILEY (1935) published their theoretical paper, SMITH<br />

(1939) predicted that "in that admirable work by NICHOLSON <strong>and</strong> BAILEY, 'The<br />

Balance <strong>of</strong> Animal Populations', will be found enough population problems to keep<br />

several laboratories busy <strong>for</strong> the next twenty years."<br />

Shortly after the publication <strong>of</strong> these theoretical works, there arose among ecolo-


gists a storm <strong>of</strong> controversy which has lasted <strong>for</strong> more than 20 years ~nd hss not<br />

yet subsided. Much <strong>of</strong> the dispute has been based on varying degrees <strong>of</strong> mutual<br />

misunderst<strong>and</strong>ing, <strong>and</strong> many innocent students <strong>of</strong> natural history have perhaps been<br />

drowned. Nevertheless, theoretical approaches <strong>and</strong> mathematical concepts still play<br />

an important role in animal population ecology, chiefly <strong>for</strong> the following reason. In<br />

the <strong>study</strong> <strong>of</strong> population processes, what we can observe is an integrated complex <strong>of</strong><br />

factors. But the elemental components <strong>and</strong> their interactions are not always apparent<br />

<strong>and</strong> indeed may even be impossible to detect by ordinary observation. Wherever<br />

unobservables are involved, they must be detected through reasoning by analogy.<br />

For the past decade, biomathematics <strong>and</strong> statistics have become increasingly more<br />

sophisticated, while the students <strong>of</strong> natural history have by observation accumulated<br />

a vast amount <strong>of</strong> in<strong>for</strong>mation on animal behaviour. Yet, there seems to be an increasing<br />

gap between the two approaches. My aim is to bridge this gap, <strong>and</strong> the scope<br />

<strong>of</strong> this paper is to show what the existing theories on <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> really<br />

mean, <strong>and</strong> what role these theories would, or would not, play in leading to an underst<strong>and</strong>ing<br />

<strong>of</strong> <strong>predation</strong> processes in relation to population dynamics.<br />

Be<strong>for</strong>e the existing theories are critically reviewed, I shall discuss the role <strong>of</strong><br />

<strong>models</strong> in general terms <strong>and</strong> then consider some basic attributes <strong>of</strong> <strong>predation</strong> <strong>and</strong><br />

<strong>parasitism</strong>. In fact, these two sections are the result <strong>of</strong> my <strong>comparative</strong> <strong>study</strong> <strong>of</strong> the<br />

existing <strong>models</strong> rather than a starting point. Nevertheless, this part <strong>of</strong> the conclusion<br />

is presented first because the argument can then be more readily followed.<br />

2. A BRIEF INQUIRY INTO THE ROLE OF MODELS IN SCIENTIFIC INFERENCE<br />

To make a critical <strong>study</strong> <strong>of</strong> existing <strong>predation</strong> (<strong>parasitism</strong>) <strong>models</strong>, we need to<br />

have a clear idea <strong>of</strong> what a 'model' is. However, the concept <strong>of</strong> <strong>models</strong> in science<br />

varies from one case to another, depending on what is aimed at in each individual<br />

case. Thus it is perhaps better to examine the past use <strong>of</strong> the word rather than to<br />

begin with an attempt to define it.<br />

The word 'model' has been used more or less synonymously with: an assumption;<br />

a hypothesis; a proposition; a theory; a law; or even a mere mathematical equation.<br />

A typical example <strong>and</strong> positive justification <strong>for</strong> this broad usage <strong>of</strong> the word is found<br />

in WALKER (1963, p. 4) :<br />

"The word model in a particular sentence may refer to one or more <strong>of</strong> many<br />

related aspects <strong>of</strong> the general notion. Thus cortical model refers to the model as it<br />

is recorded in the structure <strong>and</strong> arrangement <strong>of</strong> molecules in the memory banks <strong>of</strong><br />

the brain. Conceptual model refers to the mental picture <strong>of</strong> the model that is intro-<br />

spectively present when one thinks about the model. This picture probably corresponds<br />

to some scanning process over the appropriate memory banks. The verbal model<br />

consists <strong>of</strong> the spoken or written description <strong>of</strong> the model. The postulational model<br />

is a certain type <strong>of</strong> verbal model that consists <strong>of</strong> a list <strong>of</strong> the postulates <strong>of</strong> the model.<br />

The geometrical model refers to the diagrams or drawings that are used to describe


the model. The mathematical model refers to the equations or other relationships<br />

that provide the quantitative predictions <strong>of</strong> the model. The material model is the<br />

arrangement <strong>and</strong> interactions <strong>of</strong> fundamental particles, their fields <strong>and</strong> aggregates.<br />

When a writer refers to the 'Bohr model <strong>of</strong> the hydrogen atom', he may have in<br />

mind any or all <strong>of</strong> these aspects; the reader must select the aspects appropriate to<br />

the context."<br />

WALKER'S broad usage <strong>of</strong> the word includes, in later chapters <strong>of</strong> his book, the<br />

MENDELIAN law <strong>of</strong> inheritance <strong>and</strong> the DARWINIAN theory <strong>of</strong> natural selection as<br />

<strong>models</strong>. Clearly, the word 'model' in WALKER'S sense is used to categorize similar,<br />

but distinctly different notions in a single, convenient, descriptive term. This catego.<br />

rization may at times be needed in scientific communication, but I would rather use<br />

the word in a much restricted sense in this paper, in order to emphasize the role <strong>of</strong><br />

a certain type <strong>of</strong> model that distinguishes itself from other similar notions, e.g.<br />

'assumptions', 'hypotheses', 'theories', or 'descriptions' <strong>of</strong> laws <strong>and</strong> rules.<br />

A dictionary (e. g. The WEBSTER's Third International Dictionary, 1968) treats<br />

the word as a synonym <strong>for</strong> : Example, Pattern, Exemplar, Paradigm, Ideal, etc.; or it<br />

is something perfect <strong>of</strong> its kind. The dictionary also states that a model is: a thing<br />

that serves as a pattern or source <strong>of</strong> inspiration <strong>for</strong> an artist or writer; or an analogy<br />

used to help visualize, <strong>of</strong>ten in a simplified way, something that cannot be observed<br />

directly. The last definition is particularly important <strong>and</strong> most relevant to my investigation.<br />

WALKER further stated that "the main purpose <strong>of</strong> a model is to make predictions",<br />

<strong>and</strong> that "if a mathematical model predicts future events accurately, there is no necessity<br />

<strong>for</strong> any interpretation or visualization <strong>of</strong> the process described by the equation."<br />

These statements should be interpreted with caution, however. If they are taken<br />

literally, it might be concluded that the purpose <strong>of</strong> an assumption, hypothesis, theory,<br />

etc. is to predict but not to aid underst<strong>and</strong>ing <strong>of</strong> natural order: that is to say,<br />

WALKER'S statement might be taken erroneously as synonymous with 'the aim <strong>of</strong><br />

science is predictions'. TOULMIN (1961) pointed out that the ancient Babylonian<br />

astronomers who predicted the motion <strong>of</strong> stars amazingly accurately by arithmetic<br />

means failed to underst<strong>and</strong> underlying mechanisms, while the Ionian philosophers'<br />

crude model <strong>of</strong> the universe eventually led to 20 th-century physics.<br />

The underst<strong>and</strong>ing <strong>of</strong> natural order is achieved through the <strong>for</strong>mation <strong>of</strong> new<br />

concepts. SCHON (1967) emphasized the role <strong>of</strong> metaphor in the <strong>for</strong>mation <strong>of</strong> a new<br />

concept, through which a novel idea or discovery was made. He maintained that the<br />

new concept would emerge by shifting already existing concepts to a new situation<br />

by metaphor. That is to say, the old concepts, shifted to the new situation by metaphor,<br />

are <strong>models</strong> <strong>for</strong> the new concept. If we use the word 'model' in a similar way,<br />

<strong>and</strong> I am inclined to do so here, then some <strong>of</strong> WALKER'S examples should be excluded.<br />

For example, the DARWINIAN theory can be a model if its principle is shifted


<strong>and</strong> applied, say, to social phenomena, but as long as the theory remains in the domain<br />

<strong>of</strong> organic evolution I do not call it a model. Similarly, while the MENDELIAN law<br />

itself is by the same token not a model <strong>of</strong> M~NDELIAN inheritance, its principle can<br />

be demonstrated by a model in which equal numbers <strong>of</strong> red <strong>and</strong> white balls in a jar<br />

are sampled at r<strong>and</strong>om, two balls at a time. The statistical expectation <strong>of</strong> the propor-<br />

tion <strong>of</strong> white-white pairs is one-quarter <strong>of</strong> the total number <strong>of</strong> pairs drawn. This<br />

model can be stated by a simple mathematical equation, <strong>and</strong> the equation is just the<br />

means <strong>of</strong> statement. The equation as a statement <strong>of</strong> the model can at the same time<br />

be a statement <strong>of</strong> the law, since the symbolic expression <strong>for</strong> both the model <strong>and</strong> the<br />

law takes the same <strong>for</strong>m. By means <strong>of</strong> the balls-in-a-jar model, however, the empirical<br />

law found by MENDEL becomes underst<strong>and</strong>able <strong>and</strong> intelligible, <strong>and</strong> the model leads<br />

to the postulation <strong>of</strong> particulate inheritance--a hypothesis; the reliability <strong>of</strong> this hypo-<br />

thesis is tested in an organized way against further observations until it emerges as<br />

a biological theory <strong>and</strong> principle.<br />

We should distinguish, however, between an equation as a general, symbolic<br />

method <strong>of</strong> statement, <strong>and</strong> one as a mathematical operation as a means <strong>of</strong> reasoning. In<br />

a model as simple as the balls-in-a-jar example, the mathematical probability <strong>of</strong> a pair<br />

<strong>of</strong> one kind, say white-white, may be obtained intuitively <strong>and</strong> correctly (i. e. a priori),<br />

whereas <strong>of</strong>ten a more complicated mathematical operation is required to draw a con-<br />

clusion. An equation that states a result <strong>of</strong> inferences should there<strong>for</strong>e be distinguished<br />

from an equation which is adopted as a convenient description <strong>of</strong> an empirical law,<br />

such as a polynomial equation obtained in curve-fitting by the least squares method.<br />

The latter is generally not the statement <strong>of</strong> a model nor a hypothesis;it is merely<br />

one casual <strong>and</strong> tentative way, among many others, <strong>of</strong> describing what has been<br />

observed, although it may at times play a certain role in the <strong>for</strong>mulation <strong>of</strong> ideas,<br />

as a tentative part <strong>of</strong> a model.<br />

In a very few cases, an empirical law can be stated accurately by a simple math-<br />

ematical equation in which the value <strong>of</strong> every coefficient involved is clearly defined,<br />

but without underst<strong>and</strong>ing. A typical example <strong>of</strong> this is NEWTON'S gravitational law,<br />

i.e. the <strong>for</strong>ce <strong>of</strong> gravitational interaction is proportional to the product <strong>of</strong> the masses<br />

<strong>of</strong> two interacting bodies <strong>and</strong> inversely proportional to the square <strong>of</strong> the distance<br />

between them. This is an accurate statement.<br />

It should be noticed, however, that<br />

even this accurate statement <strong>of</strong> the universal law had no rational explanatory model<br />

behind it until the early 20 th century when EINSTEIN explained it in his relativistic<br />

theory <strong>of</strong> gravity (GAMoW 1962).<br />

A similar example is seen in the logistic law in demography first <strong>for</strong>mulated by<br />

VERHURST (1838), which was later generalized to population growth in other animal<br />

species by PEARL (1927). The VERHURST-PEARL logistic law assumes that the instantaneous<br />

rate <strong>of</strong> increase per animal is proportional to the still unutilized opportunity<br />

<strong>for</strong> growth, <strong>and</strong> is expressed in the well-known mathematical equation. But no positive<br />

rationalization <strong>of</strong> the assumption has been made;there is no rational relationship


etween the assumption (CHAPMAN's (1931) concept <strong>of</strong> 'environmental resistance')<br />

<strong>and</strong> the attributes <strong>of</strong> the subject (population growth), so the latter remain unknown.<br />

The logistic equation was derived through metaphoric inferences rather than through<br />

comparisons between the attributes <strong>of</strong> the subject <strong>and</strong> those <strong>of</strong> a model in which<br />

factors involved are known.<br />

Thus, we can see that differences between (a) a deductive model (deduced only<br />

by reasoning) <strong>and</strong> (b) a descriptive equation like that <strong>of</strong> the logistic law, lie in<br />

differences between (a) a comparison <strong>of</strong> components in the subject with equivalent<br />

parts <strong>of</strong> the model <strong>and</strong> (b) a metaphoric juxtaposition <strong>of</strong> the observed trend <strong>of</strong> the<br />

subject with that <strong>of</strong> some known concepts.<br />

While the importance <strong>of</strong> metaphor, as SCHON (1967) emphasized, is appreciated,<br />

it should be borne in mind that metaphor alone does not necessarily lead to explanations<br />

<strong>and</strong> underst<strong>and</strong>ings. Quoting one <strong>of</strong> SCHON'S examples, the original concept <strong>of</strong><br />

'foot', restricted to an animal's foot, can be shifted to a much broader concept including<br />

'the foot <strong>of</strong> a mountain'. Although this example certainly shows the importance<br />

<strong>of</strong> metaphor in, say, the evolution <strong>of</strong> languages, such juxtaposition does not immediately<br />

imply the underst<strong>and</strong>ing <strong>of</strong> the structure <strong>of</strong> the foot <strong>of</strong> mountains. In other words,<br />

metaphor, playing its important role in one situation, or in a certain part <strong>of</strong> the process<br />

in the <strong>for</strong>mation <strong>of</strong> ideas, can be too vague to be useful in another. In the<br />

example <strong>of</strong> the logistic law, metaphor led to the <strong>for</strong>mulation <strong>of</strong> the equation that can<br />

<strong>of</strong>ten describe observed relationships satisfactorily, but such success <strong>of</strong>ten depends<br />

on how the observed relationships are described deterministically.<br />

Normally, in the field <strong>of</strong> population ecology, a deterministic description <strong>of</strong> phenomena<br />

is <strong>of</strong>ten so difficult that a descriptive, empirical equation can be adopted only<br />

casually. Such casual equations <strong>of</strong>ten involve some coefficients whose nature is not<br />

known. The equation is then hard to rationalize as there can be some other <strong>for</strong>ms<br />

<strong>of</strong> equations which fit the same observation equally well. Also, the coefficients must<br />

be estimated from a limited set <strong>of</strong> observed data (our observations are, at any rate,<br />

limited), <strong>and</strong> the more limited the number <strong>of</strong> observations, the less generalized the<br />

estimate will be. Further, the more coefficients that are involved <strong>and</strong> that need to be<br />

estimated, the more flexible the equation becomes since the degrees <strong>of</strong> freedom <strong>for</strong><br />

fitting increase. The above statement simply suggests that a good fit does not imply<br />

that the equation concerned explains the mechanism.<br />

Conversely, if an equation, derived from metaphoric inference, did not fit observed<br />

relationships, it would have to be rejected. The rejection, however, involves a risk<br />

<strong>of</strong> rejecting a correct assemblage <strong>of</strong> right components. This is because the disagreement<br />

could be due to some other components or conditions which were missed, <strong>and</strong><br />

not due to inappropriate metaphor; if this is so, the equation need not be rejected<br />

but only improved by further search <strong>for</strong> these overlooked factors. The difficulty is,<br />

however, that there is no systematic way to know whether the disagreement is due<br />

to the inadequate assemblage <strong>of</strong> factors or to inappropriate metaphor.


Hence, the fitting <strong>of</strong> an empirical equation to observed relationships in certain<br />

subjects, <strong>and</strong> I imply that animal <strong>predation</strong> is one such subject, has a limited value<br />

theoretically. In the following, another method, i.e. analogies by attributes, will be<br />

explored.<br />

Normally, reasoning starts from a set <strong>of</strong> tentative propositions. This set <strong>of</strong><br />

propositions is one kind <strong>of</strong> hypothesis. Because it is only tentatively assumed, it does<br />

not necessarilly <strong>and</strong> immediately postulate mechanisms underlying the subject.<br />

Often, an early, tentative hypothesis is a mere collection <strong>of</strong> all the factors that<br />

can be conceived, whereas what one can observe is the integrated complex <strong>of</strong> factors<br />

interacting with each other. It is, however, difficult in many cases to extract each<br />

component <strong>of</strong> the subject to compare with an assumed one purely by the observational<br />

method. It is possible, instead, to integrate the assumed components on a theoretical<br />

basis so that the assumption-system can be compared with the observed whole. The<br />

difficulty is that a mere list <strong>of</strong> components will not necessarily provide the method<br />

<strong>of</strong> integration. By some means, we have to assume the structure as well. It is at<br />

this stage that analogies can play a role, <strong>and</strong> it is the structure thus derived from<br />

analogies (or some known examples) that I call a 'model' here. When the model to<br />

adopt is determined, a method <strong>of</strong> calculating the model's attributes will follow. An<br />

analytic (i. e. mathematical) method can be used <strong>for</strong> the calculation, or the method<br />

commonly called 'Monte Carlo simulation' may be useful. Here, mathematics is used<br />

not as a convenient means <strong>of</strong> description, but as a means <strong>of</strong> inference.<br />

Now, we recognize two early stages <strong>of</strong> inferences; the collection <strong>of</strong> components,<br />

<strong>and</strong> the arrangement <strong>and</strong> integration <strong>of</strong> them by a tentative model. The tentative<br />

model may be called a hypothesis, but it should be borne in mind that it is only<br />

tentative <strong>and</strong> not more than a convenient assumption. Such tentative hypotheses do<br />

not enable us to postulate the mechanism <strong>of</strong> the subject. The tentative hypothesis,<br />

however, is now compared with observation <strong>and</strong> will in general need refinement, as<br />

it <strong>of</strong>ten does not agree with the facts with a desirable degree <strong>of</strong> precision. A refinement<br />

will be made through alteration <strong>of</strong> the arrangement, adding some more components<br />

which have previously been missed, etc. As the stage <strong>of</strong> refinement advances,<br />

the hypothesis would enable one to postulate more confidently. Finally, as the degree<br />

<strong>of</strong> agreement with the facts increases, the postulational hypothesis would eventually<br />

emerge as a theory or even a principle.<br />

There are three important points in the gradual process <strong>of</strong> inferences mentioned<br />

above; they will be discussed more in detail below. First is the <strong>for</strong>mulation <strong>of</strong> a<br />

tentative hypothesis; second, the evaluation <strong>of</strong> agreement ~nd disagreement between<br />

the theoretical <strong>and</strong> the observed; third, the fact-observation relationship. The third<br />

one is a question <strong>of</strong> whether an observation can be accepted as fact.<br />

For the following discussion, some symbols will be used as defined below:<br />

0 : the result <strong>of</strong> observation,<br />

Ko :the set <strong>of</strong> all major components involved (not particularly known) in the


observed system,<br />

So : the structure <strong>of</strong> the observed system,<br />

KA : the set <strong>of</strong> all assumed components in the model system,<br />

S~ : the structure <strong>of</strong> the model system,<br />

E : theoretical expectation deduced from KA <strong>and</strong> S.x.<br />

When E <strong>and</strong> 0 are compared, we will get either an agreement or a disagreement,<br />

i.e. E=O or Er respectively, to which various conditions (causes) contribute as<br />

below:<br />

Conditions<br />

C1. K~ <strong>and</strong> S~ are involved in Ka <strong>and</strong> So respectively (so that both K~4 <strong>and</strong> S~<br />

are, at least, not false).<br />

cu. if KA <strong>and</strong> S~ are both sufficient, then E=O.<br />

c~2. if either KA or S:~ is inadequate, then E~ O.<br />

c~a. if O is false or inadequate under c~I, then E:~O.<br />

Cz. Ko does not involve the whole <strong>of</strong> K~, <strong>and</strong>/or So does not involve the whole<br />

<strong>of</strong> S.~ (so that KA <strong>and</strong>/or S.n are/is, at least partly, false),<br />

c21. if false parts <strong>of</strong> Ka <strong>and</strong> Sn, or false parts <strong>of</strong> O <strong>and</strong> K~, (or S:,z), are<br />

adjusted so that they cancel out each other, then E-O.<br />

c~2. if not c21, then Er<br />

Now, one can claim that his hypothesis is right only when c, under C~ holds.<br />

However, the fact that an agreement (E~O) exists is not sufficient to establish the<br />

hypothesis, since E=O also occurs when c~1 under C2 is involved. There<strong>for</strong>e, if a<br />

comparison between E <strong>and</strong> O is the only available method, we have to be contented<br />

with an assessment <strong>of</strong> the relative credibility <strong>of</strong> these causes. The assessment can<br />

be done much the same way as <strong>for</strong> the calculation <strong>of</strong> the LAPLACIAN probability (see<br />

BURNSIDE 1928 ; POL~CA 1955).<br />

Let Pr {E=O} be the probability <strong>of</strong> event (E=O) taking place. As it takes place<br />

either when cu or when C~l is involved (the probability <strong>of</strong> which will be written as<br />

Pr {(E=O) ]c~} <strong>and</strong> Pr {(E=O) Icy} respectively), we get<br />

Pr{E=O} =Pr{(E=O) i c~} +Pr{(E=O) !c2~.<br />

Also, as C~l is dependent on C~,<br />

Pr{(E=O) [c,} =Pr~c,}Pr{C~}<br />

<strong>and</strong> similarly,<br />

Pr { (E= O) I ce~} =Pr {c~x} Pr {C~}.<br />

From these <strong>for</strong>mulae, the following conclusions are drawn. If Ka is comprised<br />

<strong>of</strong> only those components which are either axiomatic, a priori (known to be true<br />

without appeal to the particular facts <strong>of</strong> evidence), or can be deduced from concepts<br />

already known to be true, Ka must be involved in Ko. In other words, Pr {C~} is<br />

high but Pr {C2~ is low. There<strong>for</strong>e. if an agreement (E=O) was observed under<br />

these circumstances, Pr{(E-O) [cH[ is high as compared with Pr {(E=O) lc2~} ; i. e.<br />

the credibility<br />

<strong>of</strong> reasoning that the agreement is due to a right hypothesis is com-


paratively high. However, the more axiomatic K~ is, the lower Pr {cn} will be, <strong>and</strong><br />

so the less likely is event (E=O) to occur.<br />

For the above reason, a simple, deductive model <strong>of</strong>ten fails to agree with obser-<br />

vation. But such failures in deductive <strong>models</strong> are more likely to be caused either<br />

by c~2 or c~3 than by c2z. If so, there is no reason to reject the hypothesis ; it only<br />

needs further elaboration. The only case in which at least a part <strong>of</strong> the components<br />

or the structure should be rejected, is C~. Here, a careful observation <strong>of</strong> the disagree-<br />

ment is <strong>of</strong> paramount importance.<br />

There are, broadly speaking, two possible kinds <strong>of</strong> alterations when a disagree-<br />

ment is observed. A method frequently seen in the literature is to adjust the structure<br />

<strong>of</strong> the model or to add some more components to obtain E=O. Here, Pr{E=O}<br />

certainly increases, but at the same time there is a risk <strong>of</strong> getting a high Pr{C2},<br />

<strong>and</strong> hence Prl(E=O)[c21}. The risk is greater if the added components are those<br />

whose trend is not fully understood. The estimation <strong>of</strong> coefficients involved in the<br />

empirical equation could amount to this kind <strong>of</strong> adjustment, as the coefficients <strong>of</strong>ten<br />

have to be estimated by comparing E with O. The recent <strong>predation</strong> <strong>models</strong> in fact<br />

involve such a risk, as will be shown later. The worst thing is to obtain E=O by<br />

adjustment when the first disagreement was in fact caused by c22 ; it only increases<br />

Prl(E=O)]c21} <strong>and</strong> has no meaning at all.<br />

The second type <strong>of</strong> improvement is to look <strong>for</strong> more <strong>of</strong> the axiomatic components,<br />

or <strong>of</strong> those which are known to be true <strong>for</strong> any reason, without making a particular<br />

ef<strong>for</strong>t to obtain E=O. This keeps Pr {C2} to a low level, <strong>and</strong> there<strong>for</strong>e the improve-<br />

ment, if any, increases, though only gradually, Pr {c1~}.<br />

Although the second method will provide a steady approach to the goal, a question<br />

arises whether a collection <strong>of</strong> axiomatic assumptions can eventually produce a suffi-<br />

cient model. WALKER (1963) argued that "it is a common misconception that new<br />

<strong>models</strong> are constructed by strict logical deduction from observed facts <strong>and</strong> from<br />

previous <strong>models</strong>". Certainly, nothing new will come from mere accumulations <strong>of</strong><br />

known concepts. However, a model is not a mere collection <strong>of</strong> already known com-<br />

ponents but involves a positive recombination <strong>of</strong> them which is applied to a new<br />

situation. And the role <strong>of</strong> the model is to produce a useful recombination by analogy.<br />

The efficiency <strong>of</strong> finding a useful model depends on the efficiency in selecting<br />

axiomatic components <strong>and</strong> recombining them. A model is there<strong>for</strong>e required to have<br />

room <strong>for</strong> accommodating added components <strong>and</strong> recombining them. This calls <strong>for</strong><br />

a general <strong>and</strong> idealized model to start with: too specific a model has to be rejected<br />

upon finding a disagreement because <strong>of</strong> its limited capacity <strong>for</strong> modification, or it<br />

could involve a high value <strong>of</strong> Pr{(E=O)Ic2~}, particularly when some coefficients<br />

involved have to be estimated rather than determined by independent <strong>and</strong> direct<br />

observations <strong>of</strong> what these coefficients' represent.<br />

The role <strong>of</strong> idealization is again seen in the history <strong>of</strong> the physical sciences,<br />

which should be understood in the context <strong>of</strong> fact-observation relationships <strong>and</strong> <strong>of</strong>


the notions 'realistic' <strong>and</strong> 'unrealistic'. In the ARISTOTELIAN doctrine, certain natural<br />

phenomena as observed were taken <strong>for</strong> granted as axioms. Thus, a cart pulled by a horse<br />

(a constant <strong>for</strong>ce) moves at a constant speed, but comes to a stop (a natural state<br />

<strong>of</strong> rest) when the <strong>for</strong>ce is removed. Inorganic chemical processes were explained by<br />

analogy with physiological processes, such as seeds becoming ripe, which were accepted<br />

as natural, axiomatic, <strong>and</strong> were not questioned.<br />

A significant change in the way that natural order was regarded came at the<br />

time <strong>of</strong> the Renaissance when BURIDAN (OPPENHEIMER 1956), <strong>and</strong> later GALILEI,<br />

made the earliest announcement <strong>of</strong> the principle <strong>of</strong> physical inertia. In 1612, GALILEI<br />

wrote to a pupil <strong>of</strong> his:<br />

"For I seem to have observed that physical bodies have physical inclination to<br />

some motion ...... through an intrinsic property ...... And there<strong>for</strong>e, all external impediments<br />

removed,. ..... it will maintain itself in that state in which it has once been<br />

placed" (translation by DRAKE 1957).<br />

The recognition <strong>of</strong> the "physical inclination through an intrinsic property" is<br />

important, in the context <strong>of</strong> the present discussion, as GALILEI could not have been<br />

able to observe a ship floating on a perfectly calm, smooth, resistanceless water <strong>and</strong>,<br />

once pushed, moving at a constant speed without the faintest sign <strong>of</strong> slowing down.<br />

The discovery, or recognition, <strong>of</strong> inertia must there<strong>for</strong>e have been made with only<br />

an idealized situation in mind, a situation which to other natural philosophers <strong>of</strong> the<br />

period must have been 'unrealistic'. A similar example is found in the history <strong>of</strong><br />

chemistry, when the existence <strong>of</strong> chemically pure substances was recognized only<br />

under idealized, artificial, <strong>and</strong> there<strong>for</strong>e unnatural conditions (TouLMIN 1961).<br />

These examples illustrate the point that a fact as observed in a natural state is<br />

not ultimate, <strong>for</strong> it is only the visible part <strong>of</strong> the whole. Inferences by <strong>models</strong> can<br />

only help one to generalize an observation, <strong>and</strong>, as POINCAR~ (1952) pointed out,<br />

"without generalization, prediction is impossible". It is perhaps particularly true with<br />

ecological studies that generalization is possible, not in a thing in itself which we<br />

observe under natural conditions, but in an idealized situation. Here, analogies by<br />

<strong>models</strong> play an important role.<br />

3. A BACKGROUND THEORY OF THE STRUCTURE OF PREDATION AND PARASITISM<br />

In the first place, it will be made clear that what I mean by 'background' in this<br />

section involves only those components <strong>and</strong> conditions which, under each idealized<br />

assumption, are known a priori; that is, they are known to be involved in the idealized<br />

process <strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> without any need <strong>of</strong> confirmation by observation.<br />

The need <strong>for</strong> such theories is undeniable since, as already pointed out in w 2,<br />

they are the starting point <strong>for</strong> gradual inferences. It should be borne in mind that a<br />

direct comparison <strong>of</strong> this background theory with any observation might result in<br />

disagreement; but such disagreement, unlike that caused by condition c~2, will not<br />

invalidate the theory.


10<br />

The structure <strong>of</strong> <strong>predation</strong> is considered first. Suppose that there are x prey <strong>and</strong><br />

y predator individuals per unit area, <strong>and</strong> that each individual predator consumes, on<br />

the average, f(x) prey individuals per unit time. For the moment, the analytic <strong>for</strong>m<br />

<strong>of</strong> f(x) is not specified, but it is an assumed, increasing function <strong>of</strong> x.<br />

It is also<br />

assumed <strong>for</strong> the moment that the prey <strong>and</strong> predator numbers are fixed at X <strong>and</strong> Y<br />

respectively throughout one observation period t, i.e. that during t the prey population<br />

is replenished as fast as it is reduced by <strong>predation</strong>, <strong>and</strong> no increase or decrease<br />

occurs among the predators. Under these assumptions, the following will hold :<br />

n-~f(X) Yt (3.1)<br />

where n is the total number <strong>of</strong> prey killed by predators per unit area during t.<br />

capital letters <strong>for</strong> x <strong>and</strong> y indicate that these values are fixed during an observation<br />

period t.) At this stage, neither the effect <strong>of</strong> changes in the predator's psycho-physio-<br />

logical state nor the effect <strong>of</strong> social interaction is considered.<br />

(The<br />

Equation (3.1) is shown graphically in Fig. 1 where hypothetical values <strong>of</strong> n are<br />

plotted against X; note that Y <strong>and</strong> t are both fixed <strong>for</strong> all X's. The evaluation <strong>of</strong> n is<br />

nl when X is X~ <strong>and</strong> n2 when X is Xz. Of course, the measurements <strong>of</strong> n~ <strong>and</strong> n~<br />

must be made in two separate observations to meet the condition under which eq.<br />

(3.1) holds. It should also be noticed that eq. (3.1) does not provide any means <strong>of</strong><br />

evaluating the effect <strong>of</strong> <strong>predation</strong> upon the prey density because the latter is fixed in<br />

each observation period.<br />

rl<br />

n~<br />

N<br />

y<br />

X2<br />

Fig. 1. A hypothetical example <strong>of</strong> curves <strong>for</strong> eq. (3. 1) The prey density,<br />

fixed at X during time-interval t, is plotted on the horizontal axis,<br />

<strong>and</strong> the total number (n) preyed upon <strong>for</strong> t, when the predator density<br />

y is fixed at Y, is plotted on the vertical axis.<br />

Now, suppose a new situation in which the prey population is not replenished so<br />

that the prey density is gradually depleted while the predators are hunting in one<br />

observation period, i.e.t.<br />

Xl<br />

As the prey density decreases during the period, the number<br />

that the predators can kill per unit time per unit area must also decrease. This<br />

>•


11<br />

situation is easily seen from Fig. 1. Suppose XI, is the initial prey density. At this<br />

moment, the prey are killed at the rate <strong>of</strong> nJYt. But if the prey density is depleted<br />

to X2, the rate <strong>of</strong> <strong>predation</strong> is decreased to n~/Yt. Hence, the overall rate <strong>of</strong> <strong>predation</strong><br />

must be something between nl/Yt <strong>and</strong> n2/Yt. To evaluate the overall rate <strong>of</strong> <strong>predation</strong>,<br />

we must use calculus.<br />

As mentioned be<strong>for</strong>e, eq. (3.1) holds only when the prey density does not change<br />

during t. In our new situation in which the density x decreases as t increases, eq.<br />

(3. 1) holds only <strong>for</strong> such a short period that a reduction in x at this moment can<br />

practically be ignored. Let us denote this short period by `it <strong>and</strong> an accordingly small<br />

fraction <strong>of</strong> number killed by `in. Substituting x, `it, <strong>and</strong> `in <strong>for</strong> X, t, <strong>and</strong> n respectively<br />

in eq. (3.1), we have<br />

`in =f(x) Yztt, or `in/dt =f(x) Y (3.2)<br />

<strong>and</strong> <strong>for</strong> `it->0, we have<br />

dn/dt :f(x) Y (3.3).<br />

Clearly, the derivative dn/dt is the rate <strong>of</strong> capturing prey, <strong>and</strong> so it is a positive<br />

function <strong>of</strong> x. If, however, the rate <strong>of</strong> depletion in prey density, i.e. dx/dt, is considered,<br />

it is a negative function <strong>of</strong> x, but its absolute value must be equal to dn/dt,<br />

because the prey population is reduced according to the number consumed. So, we<br />

have<br />

dx/dt = -f(x) Y (3.4).<br />

Let xo be the initial prey density (when t=0) which is reduced to x over a period<br />

<strong>of</strong> time, i.e. t, <strong>and</strong> integrating eq. (3.4), we have<br />

dx<br />

fx (3.5).<br />

=- xo fCxi-<br />

Now, I shall explain in more detail the reason why the differential equation <strong>and</strong><br />

the integration (i. e. eqs. (3.4) <strong>and</strong> (3.5) respectively) are used as a means <strong>of</strong> deduction,<br />

because this means <strong>of</strong> deduction should be understood thoroughly so that my criticism<br />

<strong>of</strong> various <strong>models</strong> in later sections will be followed readily.<br />

Suppose that the initial prey density was xo when t=0, <strong>and</strong> that it took `ito to<br />

reduce the prey density by ,Ix. Assuming that `ito <strong>and</strong> so `ix were sufficiently small,<br />

<strong>and</strong> substituting xo, -`ix, <strong>and</strong> `it0 <strong>for</strong> x, `in, <strong>and</strong> `it in eq. (3.2), we have<br />

Ydto = - `ix/f (xo) .<br />

At this moment, the prey density is reduced to Xo-`ix. Suppose, <strong>for</strong> further reduc-<br />

tion in the prey density by as much as `ix, it took `itl. Then <strong>for</strong> the same reason as<br />

above, we get<br />

Y`it~ = - `ix/f (xo- dX) .<br />

In general, at the i th interval, it takes `its to reduce the density by another `ix. As<br />

the prey density has been reduced to xo-i`ix by this time, the evaluation <strong>of</strong> `it~ is<br />

given by


12<br />

Y Jr, = -Jx/f(xo-iJx).<br />

Thus we have the following summation :<br />

YJto = - Jx/f(xo)<br />

YJt, Jx/f(xo - ,fx)<br />

Y~at~ = - Jx/f (xo - 2Jx)<br />

+) YJt,=-Jx/f(xo-iJx)<br />

i<br />

i<br />

Y ZJt, = - ~ {Jx/f(xo- iJx) }.<br />

i=0 i=0<br />

Now, let t be the total time taken to the i th interval. Then t is the summation <strong>of</strong><br />

all Jt's to the i th interval, i.e.<br />

so that<br />

i<br />

t=~dt,<br />

i=0<br />

i<br />

Y ~,ft, = Yt.<br />

i-O<br />

Similarly, let x be the prey density at the i th interval, which is the difference between<br />

the initial density xo <strong>and</strong> the total number <strong>of</strong> prey taken per unit area, i.e. iJx. So,<br />

X=Xo-iJx.<br />

As x is a continuous variable, we can make Jx infinitesimally small, which is now<br />

written as dx. Also, under these circumstances, the summation sign ~ is replaced by<br />

the integral sign ~. Further, it is clear that x varies from xo to x when i varies from<br />

0 to i. Thus<br />

i fx dx<br />

lim ~ {Jx/f(xo- lax) } : f(x)"<br />

Jx~O i~O<br />

xo<br />

Hence, yt=_<br />

f; ~ dx f(X)' <strong>and</strong> we have eq. (3.5).<br />

For further discussion, the integral in the right-h<strong>and</strong> side <strong>of</strong> eq. (3.5) must be<br />

evaluated. As the <strong>for</strong>m <strong>of</strong> f(x) has not been specified, a few <strong>for</strong>ms will be assumed<br />

below <strong>for</strong> convenience.<br />

Let us assume first that f(x) is a linear function <strong>of</strong> x ; that is, the prey are killed<br />

in proportion to their density. Then,<br />

f(x) =ax (3.6)<br />

where a is any positive constant. As will be seen later, eq. (3.6) is the basis <strong>of</strong> the<br />

classical <strong>models</strong> by LOTKA (!925), VOLTERRA (1926), <strong>and</strong> NICHOLSON <strong>and</strong> BAILEY<br />

(1935), but I shall not discuss its ecological meaning as this is not needed at the<br />

moment. Substituting the right-h<strong>and</strong> side <strong>of</strong> eq. (3.6) <strong>for</strong> f(x) in eq. (3.5), we have<br />

which yields<br />

a Yt = - (x<br />

d xo<br />

dx<br />

X<br />

a Yt = - In x (3. 7).<br />

Xo


As the prey density is reduced from Xo to x during time t, the difference (xo-x) is<br />

the number <strong>of</strong> prey individuals killed per unit area during t. So, removing the 'ln'<br />

sign <strong>and</strong> rearranging, eq. (3. ,7) will be solved with respect to xo-x as below,<br />

Xo - x = Xo (1 - e- ~')<br />

or, setting z equal to Xo--X,<br />

z =Xo (1 - e -art) (3.8).<br />

This is in fact the familiar NICHOLSON-BAILEY 'Competition equation' (see w<br />

Now we have three variables in eq.<br />

13<br />

(3.8), z being the dependent variable <strong>and</strong><br />

x0 <strong>and</strong> Yt independent ones. In this particular example, the predator density Y <strong>and</strong><br />

the time t (<strong>for</strong> which the prey population is exposed to <strong>predation</strong>) are mutually com-<br />

plementary. That is to say, the effect <strong>of</strong> <strong>predation</strong> upon prey density exerted by twice<br />

as many predators <strong>for</strong> half the time, is exactly the same as the effect by half as many<br />

predators <strong>for</strong> twice the time since<br />

(2 Y) (t/2) = (Y/2) (2 t).<br />

This holds only because neither social interference (or social facilitation) among pred-<br />

ators nor changes in physiological state are considered: they have been ignored,<br />

<strong>for</strong> the time being, <strong>for</strong> simplicity.<br />

Under the above circumstances, eq.<br />

(3. 8) represents a surface in a three-dimen-<br />

sional coordinate system, i.e. the z-, x0-, <strong>and</strong> Yt-axes, in which z is the only<br />

dependent variable, <strong>and</strong> x0 <strong>and</strong> Yt are mutually independent.<br />

This does not mean<br />

that Yt is ecologically independent <strong>of</strong> x0, particularly in a closed system in which the<br />

predator density in one generation is determined by the prey density in the preceding<br />

generation. But, within a generation, Yt <strong>and</strong> xo are mathematically independent <strong>of</strong><br />

each other, in the sense that we can think <strong>of</strong> any Yt-value <strong>for</strong> a given x0-value.<br />

Figure 2a shows a surface generated by eq. (3.8), the surface being determined<br />

primarily <strong>for</strong> a given value <strong>of</strong> the constant a.<br />

In this figure, any cross-section <strong>of</strong> the surface parallel to the Z-Xo plane is linear,<br />

which suggests that <strong>for</strong> any fixed value <strong>of</strong> YL the number <strong>of</strong> prey killed per unit<br />

area increases linearly with the initial prey density. However, the cross-section parallel<br />

to the z-Yt plane is exponential, suggesting that <strong>for</strong> any fixed value <strong>of</strong> x0, the share<br />

<strong>of</strong> food <strong>for</strong> each predator decreases progressively as the predator density increases,<br />

or it becomes progressively harder <strong>for</strong> each individual to find its food as the time<br />

spent hunting increases. This is in fact the 'law <strong>of</strong> diminishing returns' when the<br />

predators put more ef<strong>for</strong>t (or predator-hours, i.e. Yt) into hunting.<br />

If both sides <strong>of</strong> eq. (3. 8) are divided by x0, then<br />

Z/Xo: (1-e -~') (3. 9).<br />

The right-h<strong>and</strong> side <strong>of</strong> eq.<br />

(3. 9) does not involve x0, <strong>and</strong> there<strong>for</strong>e z/xo is uninflu-<br />

enced by changes in x0. Graphically, the surface on the Z/Xo-Xo Yt coordinate system<br />

is perfectly parallel to the xo-Yt plane so that the cross-sections parallel to the z/xo<br />

-Yt plane maintain a constant shape along the x0-axis (Fig. 2b). Under these<br />

circumstances, we do not need a three-dimensional coordinate system but a simple


14<br />

Z-Xo PLANE ~ ...'~<br />

0 Xll ~<br />

Fig. 2a. An example <strong>of</strong> surfaces generated by eq. (3.8). x0 is the initial<br />

prey density, Yt the hunting ef<strong>for</strong>t (i.e. predator-hours), <strong>and</strong> z<br />

the reduction <strong>of</strong> the prey density at the end <strong>of</strong> the interval t.<br />

A<br />

zl~-x, PLANE<br />

,*'"'i<br />

Fig. 2b. Same as Fig. 2a, but the proportion <strong>of</strong> the prey density reduced<br />

from the initial density, i.e. Z/Xo, is plotted on the vertical axis, cf.<br />

eq. (3. 9).<br />

two-dimensional one, i.e. a Z/Xo-Yt system. This is in fact the method <strong>of</strong> presenta-<br />

tion originally used by N[CHOLSON (1933) who called the curve the 'competition curve'.<br />

This simple method <strong>of</strong> presentation is possible, however, only under the particular


15<br />

assumption that f(x) is a linear function <strong>of</strong> x. If f(x) is not a linear function, gener-<br />

ally speaking, the x0-axis is still required since the ratio z/xo again changes as x0<br />

changes, This will be shown in the following example.<br />

Observations by various authors have shown that the function f(x) is not normally<br />

linear, <strong>and</strong> there is a good reason to believe that it should not be so (see w 4c).<br />

fact, f(x) is more like the curve shown in Fig. 1, which increases as x increases but<br />

gradually approaches a plateau. This type <strong>of</strong> curve can be generated by various<br />

equations.<br />

For convenience, however, we shall assume the following function used<br />

extensively in IVLEV'S (1955) <strong>study</strong> on fish <strong>predation</strong> (see w 4d):<br />

f(x) -~ b (1 - e- a,~) (3.10)<br />

where b <strong>and</strong> a are any positive constants.<br />

Although the biological meaning <strong>of</strong> this<br />

equation is as open to criticism as the NICHOLSON-BAILEu one, this is not important<br />

at the present stage <strong>of</strong> the argument.<br />

Substituting the right-h<strong>and</strong> side <strong>of</strong> eq. (3. 10) <strong>for</strong> f(x) in eq. (3.5), we have<br />

In<br />

which yields<br />

Yt=- fx dx (3.11),<br />

xo 1 -- e -~<br />

Z--<br />

a 1 ln{(l_e_axo)e_~br~+e_~,, } (3.12)<br />

where z = x0- x.<br />

Equation (3. 12) generates a surface on the z-xo-Yt coordinate system (Fig. 3a)<br />

which has a more complex shape than that generated by eq. (3.8) or Fig. 2a.<br />

Although the cross-sections parallel to the z-Yt plane are very similar to those in<br />

Fig. 2a, as they also represent the law <strong>of</strong> diminishing returns, those parallel to z-xo<br />

Z<br />

9 -,.. \ ",, :~<br />

9 ., \ -,. :<br />

0 x~<br />

Fig 3a.<br />

Same as Fig. 2a, but the surface is generated by eq. (3. 12).


16<br />

z/x, T x o PLANE<br />

/// ,' !"--r , j,<br />

Fig. 3b.<br />

0 • ~<br />

Same as Fig. aa, but proportion z/xo is plotted on the vertical axis.<br />

plane are also curvilinear <strong>and</strong> similar to the curve generated by eq. (3. 10), Clearly,<br />

we cannot present eq. (3. 12) in a two-dimensional coordinate system showing the<br />

relationship between Z/Xo <strong>and</strong> Yt, since the relationship changes as Xo changes (Fig.<br />

3b).<br />

These two examples show that, though no ecological reality is attached to them<br />

at the moment, the number <strong>of</strong> prey killed by predators per unit area, i.e. z, is expressed<br />

as a function <strong>of</strong> two independent variables (x0 <strong>and</strong> Yt). So we can write this rela-<br />

tionship in a general <strong>for</strong>m using a functional symbol F as<br />

z=F (Xo, Yt) (3.13).<br />

Equation (3. 13) will be called an 'overall hunting equation' i<strong>and</strong> the function F<br />

an 'overall hunting function' as opposed to eq. (3.1), or (3.4), which is called an<br />

'instantaneous hunting equation' <strong>and</strong> the function f an 'instantaneous hunting function'.<br />

(The instantaneous hunting function may be a function <strong>of</strong> x, y, <strong>and</strong> t as a general<br />

case ; see later. )<br />

The essential difference between the overall <strong>and</strong> the instantaneous<br />

equations is that the <strong>for</strong>mer involves the effect <strong>of</strong> diminishing returns whereas the<br />

latter holds only at an instant <strong>and</strong> so does not involve this effect. If one intends to<br />

build a model to <strong>study</strong> a predator-prey interacting system, what is needed, from a<br />

theoretical point <strong>of</strong> view, is the overall hunting equation, since this is the equation<br />

which provides the estimates <strong>of</strong> the number <strong>of</strong> prey killed <strong>and</strong> <strong>of</strong> the final density<br />

<strong>of</strong> prey at the end <strong>of</strong> a hunting period. The <strong>for</strong>mer estimate gives a basis <strong>for</strong> calculat-<br />

ing the number <strong>of</strong> predators' progeny <strong>and</strong> the latter the number <strong>of</strong> prey's progeny.<br />

Equation (3. 13), however, does not take into consideration a number <strong>of</strong> other<br />

factors which are likely to be involved in an actual <strong>predation</strong> process, e.g. the effect


17<br />

<strong>of</strong> social interactions among predators <strong>and</strong> the effect <strong>of</strong> hunger. One way to incorpo-<br />

rate these factors <strong>and</strong>~ their influence on the <strong>for</strong>m <strong>of</strong> an overall hunting equation will<br />

be shown in the following paragraphs.<br />

Social interactions among predators may be classified into two major categories,<br />

social interference <strong>and</strong> facilitation. These cause a reduction or increase, respectively,<br />

in the instantaneous hunting efficiency <strong>of</strong> each predator as compared to what it would<br />

potentially exhibit if these factors were not operating (<strong>for</strong> a detailed discussion, see<br />

w 4i). Let S be the factor by which f(x) is reduced or increased. Then an effective<br />

instantaneous hunting function will be Sf(x). Also the effect <strong>of</strong> social interaction<br />

must vary as the densities <strong>of</strong> both predators <strong>and</strong> prey vary. For instance, too many<br />

predators hunting too few prey would experience more intense interference, than other-<br />

wise, among the predators. There<strong>for</strong>e, S must at least be a function <strong>of</strong> both Y <strong>and</strong> x,<br />

which will be written as S (Y, x). Incorporating the factor S (Y, x) into eq. (3.4),<br />

we have<br />

<strong>and</strong> so<br />

dx/dt= -S(Y, x)f(x) Y (3. 14),<br />

Yt = _ ~,jx dx (3.15).<br />

xo S (Y, x)f(x)<br />

However, the intensity <strong>of</strong> social interaction might change with time, in which case, S<br />

may also be a function <strong>of</strong> t. The complex function Sf in eq. (3.14) is a generalized<br />

instantaneous hunting function <strong>and</strong> can be written as f(x, Y), <strong>and</strong> <strong>for</strong> further gener-<br />

alization as above it may be written as f(x, Y, t). But I shall avoid such complica-<br />

tions at the moment.<br />

The integral on the right-h<strong>and</strong> side <strong>of</strong> eq. (3.15) generally involves Y but not<br />

t. This suggests that if z(=xo-x) is evaluated in eq. (3. 15), it would be a function<br />

<strong>of</strong> xo, Y, <strong>and</strong> t, rather than one <strong>of</strong> x0 <strong>and</strong> Yr. Here, Y <strong>and</strong> t no longer <strong>for</strong>m a single<br />

complex variable. Thus the overall hunting equation becomes<br />

z =F(x0, Y, t) (3.16).<br />

Equation (3. 16) has three independent variables, <strong>and</strong> so it can be presented only<br />

in a four-dimensional coordinate system, or more practically in a series <strong>of</strong> three-dimen-<br />

sional coordinate systems;if, <strong>for</strong> instance, z, x0, <strong>and</strong> Y <strong>for</strong>med the three axes <strong>of</strong> a<br />

graph, separate graphs would be needed <strong>for</strong> each l. This means that if any social<br />

interaction is involved, different results should be expected between observations with<br />

different values <strong>of</strong> t.<br />

The effect <strong>of</strong> hunger can be incorporated in much the same way as is that <strong>of</strong><br />

social interactions. Suppose f(x) is an instantaneous hunting function <strong>of</strong> an individual<br />

predator when it can potentially exert its maximum output in hunting. If the predator<br />

is partially satiated, the maximum per<strong>for</strong>mance will only be partially realized.<br />

partial realization will be expressed by a factor H, which is an index <strong>of</strong> the hunger<br />

level <strong>and</strong> is naturally defined between 0 <strong>and</strong> 1 ; H may also be less than unity when<br />

the animal has been so starved that it cannot exert its full potential ef<strong>for</strong>t. Under<br />

This


18<br />

these circumstances, the effective instantaneous hunting function is Hf(x) instead <strong>of</strong><br />

f(x), so that we have, from eq. (3.4),<br />

dx/dt =--Hf(x) Y (3.17).<br />

Naturally, H is dependent on the net food intake into the stomach <strong>and</strong> the speed<br />

<strong>of</strong> digestion. No doubt, the net food intake depends on the density <strong>of</strong> food, the density<br />

<strong>of</strong> predators, <strong>and</strong> the time spent in hunting; 3nd the speed <strong>of</strong> digestion is also a<br />

function <strong>of</strong> time, at least. There<strong>for</strong>e, an argument similar to that in social interaction<br />

applies here too. One essential difference between the effect <strong>of</strong> social interaction <strong>and</strong><br />

hunger is that the latter involves the effect <strong>of</strong> initial state ; i. e. the factor H is<br />

influenced by the level <strong>of</strong> satiation or hunger just be<strong>for</strong>e the start <strong>of</strong> the observation.<br />

So if this initial state is denoted by the symbol I0, we can write the factor H as<br />

H(x, Y, t]I0), <strong>and</strong> so the instantaneous hunting equation will be <strong>of</strong> the <strong>for</strong>m<br />

dx/dt=-H(x, Y, t! Io)f(x) Y (3.18).<br />

Both functions S <strong>and</strong> H in the above examples are indices <strong>of</strong> the partial realization<br />

<strong>of</strong> the potential per<strong>for</strong>mance that an individual predator could exert if the influence<br />

<strong>of</strong> social interaction or hunger did not exist. Of course, this index method <strong>of</strong> building<br />

a model may not toke account <strong>of</strong> the actual <strong>and</strong> detailed processes <strong>of</strong> such psycho-<br />

logical <strong>and</strong> physiological states, although these states must actually have influences on<br />

particular components <strong>of</strong> the hunting activity ; e.g. the threshold at which searching<br />

or catching action is triggered must be reflected in, say, the effective speed <strong>of</strong> search-<br />

ing or the distance at which a predator reacts to a prey. Nevertheless, the index<br />

method has the advantage <strong>of</strong> illustrating some basic properties that a model must<br />

have, without going into too minute <strong>and</strong> unnecessary details <strong>of</strong> the structure, <strong>and</strong><br />

provides a criterion <strong>for</strong> evaluating some <strong>of</strong> the <strong>models</strong> reviewed in later sections.<br />

For instance, it shows that all the components that one wants to incorporate into a<br />

model have to be considered in the <strong>for</strong>m <strong>of</strong> an instantaneous hunting equation from<br />

which the overall equation will be derived. To incorporate new components directly<br />

into the overall function that had been derived be<strong>for</strong>e these components were dis-<br />

covered is not valid, unless the new components are known to have no influence on the<br />

effect <strong>of</strong> diminishing returns.<br />

treatment will be reviewed later.<br />

Some examples <strong>of</strong> <strong>models</strong> containing such erroneous<br />

A model <strong>for</strong> <strong>parasitism</strong> has a different structure than that <strong>for</strong> <strong>predation</strong>, <strong>and</strong> a<br />

brief account <strong>of</strong> it will be given below.<br />

In <strong>predation</strong>, prey individuals normally disappear from the hunting area one after<br />

another as they ~re preyed upon, <strong>and</strong> so these "already eaten" prey are no longer<br />

available to the predators. This process is described by a differential equation, e.g.<br />

eq. (3. 4), which is the basis <strong>of</strong> a <strong>predation</strong> model. In <strong>parasitism</strong>, however, host<br />

individuals do not necessarily disappear <strong>and</strong> are still available to parasites during the<br />

course <strong>of</strong> hunting. Under these circumstances, the approach based on a differential<br />

equation loses its logical basis. Also, the availability <strong>of</strong> already parasitized hosts has<br />

different influences on those parasites that do not discriminate between parasitized


19<br />

<strong>and</strong> unparasitized hosts <strong>and</strong> on those that do.<br />

A typical, idealized parasite <strong>of</strong> the indiscriminate type can be defined as one<br />

which parasitizes fresh host individuals <strong>and</strong> already parasitized ones with equal prob-<br />

ability. In the following, <strong>for</strong> simplicity, it is assumed ideally that a parasite individual<br />

lays only one egg at a time.<br />

Suppose the host density is X<br />

(the capital letter indicates, as be<strong>for</strong>e, that the<br />

density is not subject to change during the course <strong>of</strong> attack) <strong>and</strong> n eggs are laid by<br />

Y parasites per unit area <strong>for</strong> time interval t. Then eq.<br />

(3.1) holds here too. As the<br />

parasites do not recognize already parasitized hosts, some hosts receive more than<br />

one parasite egg. Then our task is to find the total number <strong>of</strong> hosts receiving at<br />

least one egg, since those hosts receiving at least one egg are assumed to be killed<br />

eventually.<br />

Let Pr{i} be the probability <strong>of</strong> one host individual receiving i eggs. Then XPr{i}<br />

is the number <strong>of</strong> hosts per unit area, each <strong>of</strong> which receives i parasite eggs. There-<br />

<strong>for</strong>e, the total number <strong>of</strong> hosts parasitized, i.e. z, will be<br />

Clearly, since<br />

n<br />

z =X~ Pr{i} (3. 19).<br />

i=l<br />

Pr {i} = 1- Pr {0},<br />

i~l<br />

the right-h<strong>and</strong> side <strong>of</strong> the above equation is substituted <strong>for</strong> that in eq.<br />

we have<br />

z = X(1 - Pr ~0} ) (3.20).<br />

(3. 19), <strong>and</strong><br />

Normally, the frequency distribution <strong>of</strong> a probability is determined by its mean <strong>and</strong><br />

variance about the mean. Since n eggs are laid in X hosts per unit area, the mean<br />

number <strong>of</strong> eggs laid in each host is n/X, <strong>and</strong> so, if the variance V is known, we can<br />

write<br />

Pr{O~ =r V) (3.21)<br />

where ff is a functional symbol. Since n is given by eq. (3. 1), we have<br />

Pr{0} =r<br />

Yt/X, V).<br />

Substituting the right-h<strong>and</strong> side <strong>of</strong> the above equation <strong>for</strong> Pr{O} in eq. (3. 20), we get<br />

z=X[1-O(f(Y) Yt/X, V)] (3. 22).<br />

Equation (3. 22) is an overall hunting equation <strong>for</strong> an indiscriminate parasite comparable<br />

to eq. (3. 13) <strong>for</strong> predators. If social interaction is involved among the parasites<br />

concerned, the same argument as in <strong>predation</strong> applies here too ; the function f is then<br />

S(Y, X)f (X), or in general f(X, Y, t).<br />

Generally, eqs. (3. 13) <strong>and</strong> (3. 22) differ from each other, even if they have the same<br />

f(X), Y, <strong>and</strong> t. Only under a few special circumstances will these two turn out to<br />

be <strong>of</strong> the same <strong>for</strong>m. For instance, if the parasites are assumed to distribute their<br />

eggs at r<strong>and</strong>om over the host individuals, <strong>and</strong> if the number <strong>of</strong> hosts is sufficiently<br />

large so that the probability <strong>of</strong> a given host individual being found by each parasite


20<br />

individual is sufficiently small, the number <strong>of</strong> hosts receiving no egg, i.e. Pr {0}, will<br />

be the first term (or the 0 term) <strong>of</strong> a POISSON series, i.e.<br />

Pr{O} =e -~/x<br />

So if we assume f(X)=aX, we have from eq. (3. 1),<br />

n =aXYt<br />

so that<br />

Pr {0} = e -~r~<br />

<strong>and</strong> substituting the right-h<strong>and</strong> side <strong>of</strong> the above equation <strong>for</strong> Pr{O} in eq. (3.20),<br />

we get<br />

z=X(1-e -art) (3.23).<br />

Since X is equivalent to x0 in the case <strong>of</strong> <strong>predation</strong>, the above equation is identical<br />

in <strong>for</strong>m to eq. (3. 8).<br />

If, however, we assume that eq. (3.10) holds instead <strong>of</strong> eq. (3. 6) <strong>for</strong> f(x), other<br />

things being equal, we have <strong>for</strong> <strong>parasitism</strong><br />

z =X(1-e -b(1-e-ax) Yt/X) (3.24),<br />

which is not the same as eq. (3.12). Obviously, a predator does not find 'already<br />

eaten' prey individuals nor spend any time eating such imaginary prey, <strong>and</strong> this<br />

makes the difference. In the first example <strong>for</strong> <strong>parasitism</strong>, no account is taken <strong>of</strong> the<br />

time that the parasite has to spend laying eggs, so that it becomes the same as in a<br />

<strong>predation</strong> model in which the time spent eating prey is not considered. Also, as will<br />

be discussed in detail later, we cannot assume without contradiction that the instantaneous<br />

hunting function is the same <strong>for</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>. This implies that<br />

<strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> <strong>models</strong> cannot logically be considered to have the same <strong>for</strong>m.<br />

As far as I know, this point has been entirely overlooked in population theories.<br />

If the parasite concerned has the ability to detect a host already carrying one or<br />

more eggs, then one assumption set <strong>for</strong>th in the above indiscriminate <strong>parasitism</strong><br />

model breaks down. That is, discriminate parasites would not spend the same amount<br />

<strong>of</strong> time on already parasitized hosts as on fresh hosts, since in the <strong>for</strong>mer case only<br />

the time spent in examination would be involved whereas, in the latter, the time<br />

spent laying eggs is also involved. Even their paths <strong>of</strong> search may be influenced if<br />

they can detect an already parasitized host from some distance by scent <strong>and</strong> do not<br />

approach <strong>for</strong> a close examination. The situation is then halfway between <strong>predation</strong><br />

<strong>and</strong> indiscriminate <strong>parasitism</strong>.<br />

Bearing these background theories in mind, we can now take a close look at the<br />

existing <strong>models</strong>.<br />

4. THE EXISTING MODELS<br />

The <strong>models</strong> to be studied here are those by LOTKA (1925)-VoLTERRA (1926),<br />

NICHOLSON-BAILEY (1935), HOLLING (1959b), IVLEV (1955)-GAosE (1934), ROYAMA<br />

(1966), WATT (1959), THOMPSON (1924)-SToY (1932), HASSELL-VARLnY (1969), <strong>and</strong>


21<br />

HOLLING (1966). In order to maintain consistency throughout this <strong>study</strong>, an ef<strong>for</strong>t<br />

will be made, as far as possible, to use the same symbols denoting the same factors,<br />

parameters, etc. For example, x st<strong>and</strong>s <strong>for</strong> the density <strong>of</strong> a prey (host) species as<br />

against y <strong>for</strong> the predator (parasite) density, <strong>and</strong> t <strong>for</strong> a time-interval during which<br />

the prey (host) species are exposed to <strong>predation</strong> (<strong>parasitism</strong>). Symbols used extensively<br />

are listed <strong>and</strong> defined in Appendix 4. The consistency <strong>of</strong> using the same symbols<br />

<strong>for</strong> the same meaning in different <strong>models</strong> makes it difficult to use those <strong>of</strong> the<br />

original authors.<br />

Each subsection begins with the presentation <strong>of</strong> the model concerned, more or<br />

less in the manner that the original author presented it, so that the way he reasoned<br />

can be studied easily.<br />

a). The LOTKA-VOLTERRA model<br />

LOTKA (1925) <strong>and</strong> VOLTERRA (1926) independently proposed equations which<br />

are essentially the same. Both authors' methods are largely analytical (i. e. by mathematical<br />

analysis), though considering to some extent analogies from kinetics. VOLTE-<br />

RRA was thinking <strong>of</strong> <strong>predation</strong> whereas it was explicitly stated by LOTKA that his<br />

equations were <strong>for</strong> <strong>parasitism</strong>.<br />

Their first assumption is the geometric increase <strong>of</strong> a population; in the case <strong>of</strong><br />

the prey population, its instantaneous rate <strong>of</strong> increase per individual, i.e. (dx/dt)/x, is<br />

assumed to be constant in the absence <strong>of</strong> predators. Thus we have dx/dt-rx where<br />

r is a coefficient <strong>of</strong> increase (or <strong>of</strong> net birth -= birth minus death due to factors other<br />

than <strong>predation</strong>). Similarly, <strong>for</strong> the predator population, we have dy/dt=-r'y where<br />

-r' is a coefficient <strong>of</strong> decrease in the absence <strong>of</strong> the prey population, as predators<br />

will die if no food is available. However, if the two populations are put together,<br />

the prey population will now diminish as much as it is preyed upon. That is to say,<br />

in the presence <strong>of</strong> predators, the coefficient <strong>of</strong> increase must be equal to the difference<br />

between the net birth in the absence <strong>of</strong> predators <strong>and</strong> the death due to <strong>predation</strong>.<br />

It is assumed secondly that the number preyed upon is proportional to the number<br />

<strong>of</strong> encounters between prey <strong>and</strong> predator individuals, <strong>and</strong> so the rate <strong>of</strong> loss due to<br />

<strong>predation</strong> is equal to the rate at which an individual prey is encountered by predators,<br />

i.e. ax where a is a proportionality factor <strong>of</strong> encounters. Then r, under these circumstances,<br />

should be replaced by the expression (r-ay). Similarly, the predator population<br />

can now increase because food is available, <strong>and</strong> its rate <strong>of</strong> increase per predator<br />

must be equal to the difference between the death rate <strong>and</strong> the birth rate due to the<br />

intake <strong>of</strong> food. So, under the assumption that the birth rate is proportional to the<br />

amount <strong>of</strong> food eaten, which in the above assumption is proportional to the number<br />

<strong>of</strong> encounters with prey, the coefficient <strong>of</strong> the net increase in the predator population<br />

is equal to the expression (-r'+a'x), where a' is a positive constant. Thus we have,<br />

dx/dt = (r- ay) x<br />

=rx-ayx<br />

(4a. la)


22<br />

dy/dt: (-r' +a'x)y<br />

= -r'y+a'xy (4a. lb).<br />

Both LOTKA <strong>and</strong> VOLTERRA, assuming that all the coeffients involved were constant,<br />

solved the above two equations simultaneously, from which emerged the familiar<br />

'LOTKA-VoLTERRA oscillation' in a predator-prey interacting system. Both LOTKA <strong>and</strong><br />

VOLTERRA were aware that the assumption that the coefficients a <strong>and</strong> a' were cons-<br />

tant was too simple, but VOLTERRA justified his assumption by stating that the<br />

frequency <strong>of</strong> encounters between the prey <strong>and</strong> the predators<br />

proportion to the densities.<br />

must be in linear<br />

For LOTKA, however, the justification <strong>of</strong> the constant<br />

coefficients seemed to be purely <strong>for</strong> operational convenience, that is, to solve the<br />

simultaneous equations. LOTKA carefully stated that factor a can, in a broad assump-<br />

tion, be exp<strong>and</strong>ed as power series in x <strong>and</strong> y, i.e.<br />

a:oz+~x +ry + ............<br />

<strong>and</strong> a=oz can be an approximation if ~, r, etc. are sufficiently small <strong>for</strong> values <strong>of</strong><br />

both x <strong>and</strong> y not too large.<br />

Some unreasonable aspects can be pointed out in the LOTKA-VoLTERRA equations<br />

from a theoretical point <strong>of</strong> view.<br />

First, LOTKA stated that the model is primarily<br />

<strong>for</strong> <strong>parasitism</strong>, although he did not exclude <strong>predation</strong> explicitly. As I have pointed<br />

out in w 3, however, the instantaneous hunting equation <strong>for</strong> <strong>parasitism</strong> would not take<br />

the <strong>for</strong>m <strong>of</strong> a differential equation as in eq. (4a. la). There<strong>for</strong>e, LOTKA was mistaken<br />

in this respect. Secondly, because VOLTERRA was thinking <strong>of</strong> a <strong>predation</strong> process,<br />

the way he reasoned to get eq. (4a. lb) is not acceptable. First, it is obviously<br />

incorrect to assume that predators die <strong>of</strong> starvation at the same rate when prey is<br />

available as when no prey is available. In other words, the presence <strong>of</strong> the prey<br />

population causes not only the rise <strong>of</strong> predator population by reproduction but also a<br />

decrease in the death rate, because the predators are not as starved as when no prey<br />

was given. This suggests that the first term in the right-h<strong>and</strong> side <strong>of</strong> eq. (4a. lb)<br />

must also involve x, the prey density. It is acceptable, however, to assume that, in<br />

eq.<br />

(4a. la), the coefficient <strong>of</strong> increase <strong>for</strong> the prey population is equal to the<br />

difference between r, the rate <strong>of</strong> net increase in the absence <strong>of</strong> enemies, <strong>and</strong> ay, the<br />

rate <strong>of</strong> death due to <strong>predation</strong> when the predator population is added to the system<br />

concerned. This is because it can be assumed that r is not influenced directly by<br />

the presence <strong>of</strong> predators; its influence, if any, operating only through changes in<br />

the prey numbers due to <strong>predation</strong>.<br />

At this stage, let us rewrite eqs. (4a. la) <strong>and</strong> (4a. lb) in general <strong>for</strong>ms <strong>for</strong> further<br />

discussion, i. e.<br />

dx/dt=g~(x) -f(x, y) (4a. 2a)<br />

dy/dt=g2(x, y) (4a. 2b)<br />

where gl, ge, <strong>and</strong> f are functional symbols. Note that the linear term <strong>for</strong> x in the<br />

second equation is now excluded <strong>for</strong> the reason given above.<br />

The second function, i.e.<br />

f(x, y), on the right-h<strong>and</strong> side <strong>of</strong> eq. (4a. 2a) is <strong>of</strong>


course an instantaneous hunting function which has been referred to in w 3 as f(x) Y.<br />

The expression f(x, y) is a general one, <strong>and</strong> f(x) Y more specific. Whichever expression<br />

is convenient will be used in this paper.<br />

The above <strong>for</strong>m <strong>of</strong> presentation was in fact used by GAUSE (1934) in his explana-<br />

tion <strong>of</strong> VOLTERRA'S theory, though GAUSE did not explicitly explain why the linear<br />

term <strong>for</strong> x in the second equation was excluded.<br />

The following two points were raised by GAUSE (1934). First, the assumption<br />

<strong>of</strong> a geometric increase in the prey population in the absence <strong>of</strong> predators is not<br />

correct, since the population growth in any animal species, in the absence <strong>of</strong> natural<br />

enemies, would normally follow the logistic law, <strong>and</strong> so a population would not grow<br />

indefinitely. That is to say, function g~(x) in eq.<br />

23<br />

(4a. 2a) would not be a linear<br />

function <strong>of</strong> x but should be an instantaneous <strong>for</strong>m <strong>of</strong> the logistic law, <strong>and</strong> this sugges-<br />

tion sounds reasonable. GAUSE'S second point is that in the predator population the<br />

rate <strong>of</strong> increase per predator at different densities <strong>of</strong> prey would not be a linear<br />

function <strong>of</strong> the prey density either ; i.e. (dy/dt)/y is not a linear function <strong>of</strong> x.<br />

This conclusion <strong>of</strong> GAUSE'S was based on an observation by SMIRNOV <strong>and</strong> WLADIMI-<br />

gOW (see GAUSE 1934, p. 139), which showed that the rate <strong>of</strong> increase <strong>of</strong> a parasite<br />

population, Morrnoniella vitripennis, in relation to the density <strong>of</strong> its host, Phorrnia<br />

groenlaudica, was not linear, <strong>and</strong> an exponential function <strong>of</strong> x was more appropriate<br />

<strong>for</strong> gs(x, y). This suggestion, however, is not immediately acceptable <strong>for</strong> reasons<br />

discussed in detail in w 4d.<br />

My last point is concerned with the interpretation <strong>of</strong> t. If g~ is a non-zero positive<br />

function <strong>for</strong> all x's~0, (x <strong>of</strong> course is never negative), it means that progeny are<br />

produced in the prey population, <strong>and</strong> at the same time these progeny are susceptible<br />

to <strong>predation</strong> during t. Similarly, if g2 takes at times a positive value, it means that<br />

the predator population must also produce their progeny which attack the prey during<br />

t. Hence, there is no clear distinction between generations ; generations are continuous<br />

as in protozoa.<br />

<strong>and</strong> (4a. lb)<br />

Under these circumstances, the solution <strong>of</strong> simultaneous eqs. (4a. la)<br />

generates the predator-prey oscillations that were actually shown by<br />

both LOTKA <strong>and</strong> VOLTERRA.<br />

However, generations can be discrete, as in many insect species, in which case<br />

the progeny <strong>of</strong> prey produced during the time vulnerable to <strong>predation</strong> in the present<br />

generation may be attacked only in the following generation. Also, the progeny <strong>of</strong><br />

predators produced in the present generation may not attack the prey in this generation.<br />

The populations in the present generation are then only subject to decrease during t<br />

(within the generation), in which case both functions g~ <strong>and</strong> g2 will never become<br />

positive. Under these circumstances, a solution <strong>of</strong> the two simultaneous equations<br />

gives changes in numbers in both populations <strong>of</strong> the present generation, only during<br />

the period <strong>of</strong> <strong>predation</strong> within the generation (see w 4b). Hence, in this case, separate<br />

equations are required to compute the number <strong>of</strong> progeny to be produced to <strong>for</strong>m<br />

the next generation by the survivors <strong>of</strong> each population in the previous generation (s).


24<br />

This problem will not be discussed any further in this paper.<br />

The solution <strong>of</strong> simultaneous eqs. (4a. la) <strong>and</strong> (4a. lb) under the assumption <strong>of</strong> discrete<br />

generations was not considered by the original authors. The solution, as I will<br />

show in the next section, is in fact possible <strong>and</strong> is related to the NICHOLSON-BAILEY<br />

model.<br />

b).<br />

The NICHOLSON-BAILEY model<br />

This model is known as the 'Competition model'. It is, primarily, constructed <strong>for</strong><br />

the purpose <strong>of</strong> demonstrating NICHOLSON'S hypothesis that animal populations are in<br />

the state <strong>of</strong> balance fluctuating around a steady density <strong>of</strong> each species concerned,<br />

<strong>and</strong> that this steady density (or steady state) is brought about by competition among<br />

the members <strong>of</strong> the parasite species (NICHOLSON 1933). NICHOLSON with the collabo-<br />

ration <strong>of</strong> a mathematician, BAILEY (NICHOLSON <strong>and</strong> BAILEY 1935), intended to con-<br />

struct a model on the assumption <strong>of</strong> a very simple, idealized situation, concerning a<br />

theoretical relationship in densities between host <strong>and</strong> parasite species. By altering<br />

conditions in this simple model, they drew numerous conclusions about the mode <strong>of</strong><br />

existence <strong>of</strong> steady states.<br />

Whether or not NICHOLSON'S basic philosophy that animal populations are in the<br />

state <strong>of</strong> balance is a useful one, is not <strong>of</strong> concern here.<br />

It is more important to<br />

determine whether the basic premises in the NICHOLSON-BAILEY theory can produce<br />

a reasonable model <strong>for</strong> <strong>parasitism</strong>, so that a comparison between the model <strong>and</strong><br />

observation can provide any useful direction.<br />

original authors.<br />

The following is the reasoning by the<br />

Let x0 be the number <strong>of</strong> objects (hosts) originally present in a unit area, <strong>and</strong> let<br />

x be the number left undiscovered after an area s has been traversed (by all parasites<br />

concerned). Then the number <strong>of</strong> previously undiscovered objects discovered in a<br />

fraction <strong>of</strong> area traversed, i.e. ds, is xds. This must be equal to the decrease, -dx,<br />

<strong>of</strong> the number <strong>of</strong> undiscovered objects per unit area, i.e.<br />

-dx--xds (4b. 1),<br />

<strong>and</strong> since x=xo when s=0, integrating eq. (4b. 1) <strong>for</strong> the range (0, s), <strong>and</strong> hence<br />

(x0, x), we obtain<br />

X ~Xoe-*<br />

from which we have<br />

z/xo: 1- e-' (4b. 2)<br />

where z=xo-x. Factor s is called by the authors the 'area traversed', which is the<br />

area that is searched effectively by all parasites involved <strong>and</strong> includes possible overlaps.<br />

In passing, the average area traversed by each individual parasite is called the 'area <strong>of</strong><br />

discovery'. As against the area traversed, the net total area searched by all parasites<br />

concerned is called the 'area covered', which excludes areas already searched. Thus,<br />

the right-h<strong>and</strong> side <strong>of</strong> eq.<br />

(4b. 2) shows that the proportion <strong>of</strong> the 'area covered'<br />

increases only asymptotically as the 'area traversed' increases, <strong>and</strong> that there<strong>for</strong>e the


number <strong>of</strong> hosts attacked in terms <strong>of</strong> a proportion <strong>of</strong> the initial number present per<br />

unit area, i.e. Z/Xo, increases only asymptotically. Hence, the equation shows a simple<br />

example <strong>of</strong> the law <strong>of</strong> diminishing returns.<br />

25<br />

As it is important to underst<strong>and</strong> the<br />

geometric meaning <strong>of</strong> the above equation in order to see if the assumptions involved<br />

are reasonable, an illustration will be given.<br />

Be<strong>for</strong>e doing so, however, it should be pointed out that NICHOLSON <strong>and</strong> BAILEY<br />

failed to recognize the distinction between the <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> processes.<br />

For the reason already given in w 3, the differential equation as in eq. (4b. 1) is a<br />

starting point <strong>of</strong> deduction in the <strong>predation</strong> process, whereas NICHOLSON <strong>and</strong> BAILEY<br />

were aiming at constructing a <strong>parasitism</strong> model. Since I am examining the reasoning<br />

<strong>of</strong> NICHOLSON <strong>and</strong> BAILEY, their differential equation as a means <strong>of</strong> deduction has to<br />

be taken seriously. Since their reasoning is based on this differential equation, it is<br />

unreasonable to use the word 'parasite', <strong>and</strong> hence, <strong>for</strong> the remaining part <strong>of</strong> this<br />

section, I shall use the word 'predator' instead.<br />

Although the NICHOLSON-BAILEY<br />

equation can be regarded as one <strong>for</strong> <strong>parasitism</strong> because, as pointed out in w 3, an<br />

equation <strong>for</strong> <strong>predation</strong> can take the same <strong>for</strong>m as one <strong>for</strong> <strong>parasitism</strong> under a particular<br />

assumption, the maintenance <strong>of</strong> consistency between terminology <strong>and</strong> reasoning is<br />

more important here. The case in which the NICHOLSON-BAILEY equation is considered<br />

to be a <strong>parasitism</strong> model will be discussed in w 4g.<br />

Suppose a number <strong>of</strong> prey individuals are scattered at r<strong>and</strong>om over a plane where<br />

one predator searches with an average speed V, completely independently <strong>of</strong> the<br />

distribution <strong>of</strong> the prey individuals, from point A to B (see Fig. 4). The path <strong>of</strong> the<br />

predator between A <strong>and</strong> B is assumed to be rectilinear, <strong>and</strong> all the prey individuals<br />

in the plane remain stationary. (It can be shown that an irregular path may be as-<br />

sumed without influencing the conclusion, or that there is no need to assume a stationary<br />

distribution <strong>of</strong> prey individuals.) As in Fig. 4a, each prey individual has an area around<br />

it within, <strong>and</strong> only within, which the predator can recognize the prey. To simplify<br />

[" 9<br />

9 .<br />

(a)<br />

(b)<br />

Fig. 4. A geometric interpretation <strong>of</strong> the NICHOLSON-BAILEY (1935) model.<br />

For explanation see text.


26<br />

the situation again, though it is not quite necessary, the area around each prey is<br />

assumed to be a circle <strong>of</strong> radius R. Then, as the predator moves from A to B, it<br />

sees those prey individuals with hatched circles (Fig. 4a) ;or if, alternatively, the<br />

predator, rather than the prey, is given a circle <strong>of</strong> radius R as in Fig. 4b, then those<br />

prey within the hatched belt along the predator's path will be recognized.<br />

To calculate the number <strong>of</strong> prey found by the predator along its path <strong>of</strong> search,<br />

Fig. 4b will be used. First, if the predator can see a prey anywhere in the circle,<br />

the size <strong>of</strong> the effective area in which prey are found between A <strong>and</strong> B must be the<br />

size <strong>of</strong> the hatched area plus the circle at A. If, however, the distance between A<br />

<strong>and</strong> B is sufficiently large as compared with radius R, the area covered with the<br />

circle around A can be neglected as compared with the size <strong>of</strong> the hatched area.<br />

Second, <strong>and</strong> alternatively, if the path between A <strong>and</strong> B is considered to be a given<br />

fraction <strong>of</strong> a path <strong>of</strong> search, point A is the last point reached in the preceding section<br />

<strong>of</strong> search, <strong>and</strong> so the circle at A is the area already searched. Thus, it is sufficient<br />

to know the size <strong>of</strong> the hatched area in order to calculate the number <strong>of</strong> prey found<br />

between A <strong>and</strong> B. The size <strong>of</strong> the hatched area is clearly the product <strong>of</strong> the width<br />

2R <strong>and</strong> the length Vt, so that the number <strong>of</strong> prey found in the area is 2RVXt, where<br />

X is the density <strong>of</strong> prey fixed during t in each observation.<br />

If there are Y predator individuals searching at the same time, their paths being<br />

entirely independent <strong>of</strong> each other, the total number <strong>of</strong> prey found by these Y predators<br />

<strong>for</strong> time t, i. e. n, will be<br />

n-:2RVXYt (4b. 3).<br />

Equation (4b. 3) is clearly equivalent to eq. (3.1). That is, expression 2RVX is the<br />

instantaneous function f(X) in eq. (3.1), i.e.<br />

f(x) :2 RVX.<br />

Consequently, if R <strong>and</strong> V are assumed to be independent <strong>of</strong> X, i.e. changes or variation<br />

in both R <strong>and</strong> V are independent <strong>of</strong> X, we can replace the complex factor 2RV<br />

by a single factor, say, a, which can conveniently be treated as a constant. Thus we<br />

have in this model<br />

f(X) -aX (4b. 4)<br />

or<br />

n =aXYt. (4b. 5).<br />

Clearly, the complex factor aYt(=-2RVYt) is the area traversed by all the predators<br />

<strong>for</strong> time t, <strong>and</strong> is there<strong>for</strong>e equal to the NICHOLSON-BAILEY factor s. Also, if t is the<br />

whole length <strong>of</strong> time that each predator spends hunting in the generation concerned,<br />

the expression at is the whole area effectively searched by each individual predator<br />

hunting <strong>for</strong> the generation. So, this factor at is the 'area <strong>of</strong> discovery' <strong>and</strong> is assumed<br />

in NICHOLSON-BAILEY'S argument to be constant <strong>for</strong> a given species. (It should be<br />

mentioned here that NICHOLSON-BAILEY did not find any reason to separate <strong>parasitism</strong><br />

from <strong>predation</strong>, <strong>and</strong> so the above factor was in fact called the area <strong>of</strong> discovery <strong>of</strong><br />

a parasite species <strong>for</strong> its life time which usually ends at the end <strong>of</strong> a generation.)


Now, if the prey density in the present model is subject to decrease because <strong>of</strong><br />

<strong>predation</strong>, X should be replaced by variable x. Then eq. (4b. 4) becomes .f(x)=ax,<br />

which is identical to eq. (3.6) in every respect. Thus in conclusion we have eq. (3.8)<br />

which is identical to eq. (4b. 2). The above discussion will be summarized below.<br />

If the predator's paths are independent <strong>of</strong> each other as well as <strong>of</strong> the distribution<br />

<strong>of</strong> prey individuals, <strong>and</strong> also if the paths are deflected every now <strong>and</strong> then, predators<br />

will sooner or later cross those paths already traversed by themselves or by others,<br />

where the probability <strong>of</strong> finding still-undiscovered prey will be effectively nil, provided<br />

that all the prey discovered are eaten. (If a proportion <strong>of</strong> prey in the area traversed<br />

is not discovered, the predator's effectiveness is reduced by lessening the effective<br />

area <strong>of</strong> discovery by that proportion. If, however, this proportion is independent <strong>of</strong><br />

prey density, it does not influence the end conclusion.) Now, the paths intersect each<br />

other more frequently as either the time spent hunting by each predator or the<br />

number <strong>of</strong> predators hunting increases, <strong>and</strong> so the efficiency <strong>of</strong> finding prey drops<br />

progressively.<br />

This is the geometric meaning <strong>of</strong> 'competition' in NICHOLSON'S concept <strong>and</strong> is, as<br />

already mentioned, synonymous with the 'law <strong>of</strong> diminishing returns'. The effect <strong>of</strong><br />

diminishing returns still exists even when only an individual predator is hunting.<br />

The effect can still be called 'competition' since the predator is competing with itself,<br />

so to speak. In this respect, the NICHOLSONIAN competition should be distinguished<br />

from competition caused by social interference.<br />

As already mentioned, the NICHOLSON-BAILEY model assumes animals with discrete<br />

generations. Let us introduce this condition into the LOTKA-VOLTERRA eqs. (4a. la)<br />

<strong>and</strong> (4a. lb). As generations are discrete, no birth will take place during t in both<br />

populations; the coefficient <strong>of</strong> increase <strong>for</strong> the prey species will never exceed zero,<br />

i.e. r


28<br />

since<br />

z =Xo (1 - e-ay~<br />

lira (1 - e-,'t)/r': t.<br />

r~-~O<br />

Clearly, y0 corresponds to my previous notation Y, <strong>and</strong> so we obtain the NICHOLSON-<br />

BAILEY eq. (3. 8).<br />

Now it is very clear that the NICHOLSON-BAILEY model is only a special case <strong>of</strong><br />

the new solution <strong>of</strong> the LOTKA-VOLTERRA model, i.e. eq. (4b. 6), in which r, r', <strong>and</strong><br />

a' are all zero. The above conclusion is contradictory to a statement by NICHOLSON<br />

<strong>and</strong> BAILEY (1935, second paragraph, p. 551) :<br />

"... , we have not been able to derive our theory from LOTKA'S fundamental<br />

equations. Competition does not appear explicitly in any <strong>of</strong> his equations, <strong>and</strong> few,<br />

if any, indicate the existence <strong>of</strong> this factor."<br />

It should be mentoned that NICHOLSON <strong>and</strong> BAILEY appeared to refer to 'LOTKA'S<br />

fundamental equations' as those in chapter VI <strong>of</strong> LOTKA'S book, but that those which<br />

are relevant to the NICHOLSON-BAILEY treatise, i.e. eqs.<br />

(4a. la) <strong>and</strong> (4a. lb) in the<br />

present paper, appear in chapter VIII. However, LOTKA called the equations in chapter<br />

VIII a 'special case' <strong>and</strong> those in chapter VI, a 'general case'. Since a general case<br />

involves a special case, the NICHOLSON-BAILEY criticism quoted above must be meant<br />

to apply also to eqs. (4a. la) <strong>and</strong> (4a. lb), <strong>and</strong> such a criticism cannot be taken seriously.<br />

Contrary to the NICHOLSON-BAILEY view, the LOTKA-VOLTERRA equations are<br />

<strong>comparative</strong>ly more general <strong>and</strong> detailed than the NICHOLSON-BAILEY one. Obviously,<br />

the only necessary condition which makes the LOTKA-VOLTERRA equations match the<br />

condition <strong>of</strong> discrete generations is that a'-0.<br />

And r <strong>and</strong> r' are, unlike the simpler<br />

assumption by NICHOLSON-BAILEY, not generally zero. That is to say, the whole <strong>of</strong><br />

the NICHOLSON-BAILEY model is covered by the LOTKA-VOLTERRA one, <strong>and</strong> so we<br />

do not need the <strong>for</strong>mer. However, some specific assumptions tentatively adopted by<br />

LOTKA are not satisfactory from an ecologist's point <strong>of</strong> view. What is needed is the<br />

generalized LOTKA-VOLTERRA eqs. (4a. 2a) <strong>and</strong> (4a. 2b), which have both necessary <strong>and</strong><br />

sufficient conditions <strong>for</strong> computation <strong>of</strong> the final densities <strong>of</strong> both populations, if<br />

appropriate functions <strong>for</strong> f, gl, <strong>and</strong> gz are found. As animals with discrete generations<br />

are assumed here, we need separate equations to evaluate the initial densities in the<br />

following generation. If, however, generations are not discrete, eqs. (4a. 2a) <strong>and</strong> (4a. 2b)<br />

are sufficiently comprehensive.<br />

Although TINBERGEN <strong>and</strong> KLOMP (1960) introduced into the NICHOLSON-BAILEY<br />

model the effect <strong>of</strong> mortality in both populations (the authors considered <strong>parasitism</strong><br />

rather than <strong>predation</strong> as they did not think that a distinction was needed), it was<br />

assumed that their mortality factors acted only after the period <strong>of</strong> attack had ended,<br />

but not during the attack. This was perhaps because the arithmetic method used by<br />

TINBERGEN <strong>and</strong> KLOMP was not quite capable <strong>of</strong> incorporating the effect <strong>of</strong> mortality<br />

during the attack period. An analysis <strong>of</strong> the process in which death takes place during


29<br />

the attack period is now possible with the aid <strong>of</strong> the generalized LOTKA-VOLTERRA<br />

eqs. (4a. 2a) <strong>and</strong> (4a. 2b).<br />

The final point <strong>of</strong> investigation in this section will be concerned with the real<br />

meaning <strong>of</strong> the 'area <strong>of</strong> discovery' in the NICHOLSON-BAILEY terminology, i.e. at, or<br />

2RVt in my notation.<br />

In the dynamics <strong>of</strong> gas molecules, the movement <strong>of</strong> a molecule can be regarded<br />

as ideally haphazard <strong>and</strong> independent <strong>of</strong> other molecules be<strong>for</strong>e it collides with another.<br />

Then the number <strong>of</strong> collisions that will occur during time interval At will be<br />

2RVxAt in which R is the effective radius, V the average speed <strong>of</strong> each molecule,<br />

<strong>and</strong> x the population density <strong>of</strong> the molecules. Clearly, the NICHOLSON-BAILEY model,<br />

as well as the second term in LOTKA-VOLTERRA eq. (4a. la), is an analogy to the<br />

'law <strong>of</strong> mass-action' in physical chemistry. (This was perhaps the reason by which<br />

VOLTERRA justified his assumption <strong>of</strong> the linear relationship between the frequency<br />

<strong>of</strong> encounters <strong>and</strong> the densities <strong>of</strong> prey <strong>and</strong> predator species.) The competition equation<br />

as expressed in eq. (3.8) is in fact identical to what is called in chemistry<br />

the 'velocity equation <strong>for</strong> a unimolecular reaction'.<br />

It is not difficult to visualize what the effective radius <strong>of</strong> a gas molecule is,<br />

since it is the radius in which an effective contact with another molecule is made so<br />

that a reaction takes place. However, what is the effective radius <strong>for</strong> a predator ?<br />

NICHOLSON <strong>and</strong> BAILEY assumed that this was the radius within which a predator<br />

could recognize the prey. However, <strong>for</strong> the competition equation to be an exact analogy<br />

to the velocity equation, as the <strong>for</strong>m <strong>of</strong> the NICHOLSON-BAILEY equation implies, the<br />

path <strong>of</strong> a predator (a molecule) has to be completely independent <strong>of</strong> the position <strong>of</strong><br />

the prey (other molecules) immediately be<strong>for</strong>e the collision. In other words, the<br />

predator's recognition <strong>of</strong> a prey has to be made, strictly speaking, by bodily contact.<br />

If, however, recognition was made well away from the prey, the predator would have<br />

to approach it, <strong>and</strong> this immediately means a digression from a free path. The problem<br />

now is, how much deviation from the competition equation would be expected by<br />

the digression from a free path. This degression can be serious under certain circumstances<br />

as will be discussed in w 4e.<br />

Also, if it is assumed that recognition occurs at some distance from the prey,<br />

the predator may see more than one prey individual at a time. Unless the predator<br />

can catch the prey in a sweeping action, each prey must be h<strong>and</strong>led individually.<br />

Under these circumstances, recognition will not result in immediate capture. In other<br />

words, the number <strong>of</strong> prey that a predator can capture would not increase as fast as<br />

the number <strong>of</strong> recognitions increases with increasing density <strong>of</strong> prey population.<br />

This problem will be discussed in ~ 4d.<br />

The last point, which is more important than any other, is the effect <strong>of</strong> the time<br />

spent catching, killing, digesting, etc., <strong>for</strong> each victim. In the analogy <strong>of</strong> unimolecular<br />

reaction, there is no need to think about the time involved in actions taking place<br />

after a collision is made since each molecule ceases to capture more. In the <strong>predation</strong>


30<br />

<strong>models</strong>, each predator continues to catch a number <strong>of</strong> prey during the period concerned.<br />

While it is h<strong>and</strong>ling the victim, the predator temporarily stops searching, <strong>and</strong> this<br />

h<strong>and</strong>ling time should be separated from effective searching time. No doubt, as the<br />

number <strong>of</strong> captures increases within a given period, the smaller is the proportion <strong>of</strong><br />

the period available <strong>for</strong> searching, <strong>and</strong> hence the number <strong>of</strong> captures increases only<br />

asymptotically as the prey density increases (<strong>for</strong> details see w 4c).<br />

It is clear that because <strong>of</strong> their neglect <strong>of</strong> some unavoidable physical properties<br />

the NICHOLSON-BAILEY concept <strong>of</strong> the 'area <strong>of</strong> discovery' is quite unsatisfactory as a<br />

basis <strong>of</strong> calculating the number <strong>of</strong> captures. The term may be used as one denoting<br />

the efficiency <strong>of</strong> a predator's catching activity now <strong>for</strong> a historical reason, but if so it<br />

should be borne in mind that the term loses its geometric implication <strong>and</strong> becomes<br />

only metaphorical.<br />

The usage <strong>of</strong> the term 'area <strong>of</strong> discovery', widely seen in the literature <strong>of</strong> popu-<br />

lation dynamics, is in fact <strong>of</strong> this metaphorical nature, because it is not an 'area'<br />

measurable in physical dimensions. It is a value to be calculated from eq. (3. 7),<br />

i.e. in the NICHOLSON-BAILEY model <strong>for</strong> <strong>predation</strong>,<br />

at= 1 in Xo (4b. 7)<br />

X0 -- z<br />

(note that X=Xo-Z), <strong>and</strong> <strong>for</strong> a special case <strong>of</strong> <strong>parasitism</strong>, i.e. in eq. (3. 23),<br />

at =+In X (4b. 8).<br />

X-z<br />

If we accept this metaphorical implication <strong>of</strong> the term, <strong>for</strong> historical reasons,<br />

<strong>and</strong> redefine it as a calculated value in the right-h<strong>and</strong> side <strong>of</strong> eqs. (4b. 7) or (4b. 8), the<br />

concept <strong>of</strong> 'area <strong>of</strong> discovery' becomes one way <strong>of</strong> expressing the hunting efficiency<br />

(though not necessarily a useful one: see the appendix to w 4i), completely emanci-<br />

pated from its original implication as a geometric measure <strong>of</strong> the effective area <strong>of</strong><br />

recognition (or catching) as in Fig. 4. Naturally, the value under the logarithmic<br />

sign in eqs. (4b. 7) <strong>and</strong> (4b. 8) involves various factors which are <strong>of</strong> no geometric<br />

significance, e.g. the time spent in activities other than searching, mortality in both<br />

hunting <strong>and</strong> hunted species, <strong>and</strong> social interactions among the hunters.<br />

In order to make a clear distinction between the two concepts : (1) 'the area <strong>of</strong><br />

discovery' as a measure <strong>of</strong> hunting efficiency, defined by the right-h<strong>and</strong> side <strong>of</strong> eqs.<br />

(4b. 7) <strong>and</strong> (4b. 8), <strong>and</strong> (2) 'the effective area <strong>of</strong> recognition' as truly a geometric measure<br />

in <strong>models</strong> appearing in this paper, I shall use, throughout this paper, a symbol /~ <strong>for</strong><br />

the <strong>for</strong>mer measure as distinguished from a plain symbol a <strong>for</strong> the latter. Thus, the<br />

/~ <strong>for</strong> <strong>predation</strong> is :<br />

<strong>and</strong> the /~ <strong>for</strong> <strong>parasitism</strong> is :<br />

i~=1 In Xo (4b. 9),<br />

X0 --Z<br />

i~ =~ln X (4b. 10).<br />

. [ X-z


31<br />

In this scheme, the NICHOLSON-BAILEY model is a particular case in which ii=-at<br />

(constant). However, as will be shown in later sections (also summarized in the<br />

appendix to w 4i), the value /~ cannot normally be constant but a function (<strong>and</strong> <strong>of</strong>ten<br />

a decreasing function) <strong>of</strong> both predator <strong>and</strong> prey (or parasite <strong>and</strong> host) densities.<br />

It should be recalled again here that the NICHOLSON-BAILEY model, although in<br />

their paper used extensively as a model <strong>of</strong> para~<br />

is essentially a <strong>predation</strong> model,<br />

a special case <strong>of</strong> the LOTKA-VOLTERRA model. For the reason given in w 3, it turns<br />

out to be <strong>of</strong> the same <strong>for</strong>m as a parasite model only because the instantaneous hunting<br />

function is assumed to be linear. This assumption is no longer reasonable, however.<br />

c). HOLLING'S disc equation<br />

HOLLING (1959b) per<strong>for</strong>med the following simulation experiment.<br />

A number <strong>of</strong><br />

cardboard discs were placed on a table in a casual way. A blind-folded subject tapped<br />

the table with a finger to discover discs, again in a casual manner. When a disc was<br />

found, it was picked up <strong>and</strong> carried to a corner <strong>of</strong> the table, then a new series <strong>of</strong><br />

taps was per<strong>for</strong>med to find another disc, <strong>and</strong> so on. After a series <strong>of</strong> experiments at<br />

various disc densities, the number <strong>of</strong> discs h<strong>and</strong>led in a given time was plotted against<br />

the density <strong>of</strong> the discs. The curve thus obtained looked like that in Fig. 1, i.e. the<br />

number <strong>of</strong> discs taken increased as the disc density increased but gradually leveled<br />

<strong>of</strong>f.<br />

HOLLING'S explanation <strong>for</strong> this trend was as follows. It was assumed that the<br />

probability <strong>of</strong> finding a disc with a tap was proportional to the density <strong>of</strong> the discs.<br />

Thus, letting a be the proportionality factor, which HOLLING called the 'instantaneous<br />

rate <strong>of</strong> discovery', <strong>and</strong> t, be the time spent tapping, the number <strong>of</strong> discs touched<br />

(i.e. n) <strong>for</strong> t, would be<br />

n=aXt, (4c. 1)<br />

where X is the density <strong>of</strong> the discs. This evaluation is reasonable provided that the<br />

disc density is kept constant during each experiment ; thus capital letter X is used<br />

above. HOLLING, however, did not explicitly mention that X was kept constant to<br />

meet this condition.<br />

Now if an average time h was spent h<strong>and</strong>ling a disc each time one was found,<br />

the total time spent h<strong>and</strong>ling n discs must be hn. Thus the total hunting time t (i. e.<br />

total time tapping+total time h<strong>and</strong>ling) must be<br />

t=t,+hn (4c. 2).<br />

Eliminating t~ from eqs. (4c. 1) <strong>and</strong> (4c. 2) we have<br />

n =aXt/(1 +ahX) (4c. 3).<br />

This is HOLLING'S disc equation <strong>and</strong> describes his experiments very well. HOLLING<br />

showed that this equation also fitted observed relationships in experiments with<br />

various living predator <strong>and</strong> parasite species very well. Equation (4c. 3) clearly shows<br />

that, because <strong>of</strong> the involvement <strong>of</strong> factor h, the number <strong>of</strong> prey captured by a<br />

predator <strong>for</strong> a given time is limited, <strong>and</strong> under no circumstances can exceed l/h, since


32<br />

lim n/t=l/h.<br />

X~co<br />

(It should be mentioned in passing, however, that there is a logical jump from the<br />

assumption <strong>of</strong> the disc model to the above conclusion (see w 4d).)<br />

Although this model is excellent to demonstrate the effect <strong>of</strong> the h<strong>and</strong>ling time<br />

upon the number <strong>of</strong> captures, the application <strong>of</strong> this model to various observations<br />

with actual animals, by HOLLING himself <strong>and</strong> by others, is open to criticism. Let us<br />

examine the disc experiment more critically below.<br />

First, the assumption that the probability <strong>of</strong> finding a disc is proportional to the<br />

density <strong>of</strong> discs is justified only when the discs do not overlap <strong>and</strong> only <strong>for</strong> as long<br />

as the disc density is kept constant. Clearly, the mathematical probability <strong>of</strong> touching<br />

a disc at a tap must, under these circumstances, be equal to the proportion <strong>of</strong> the<br />

total area covered with the discs to the area <strong>of</strong> the table, provided that every part<br />

<strong>of</strong> the table has an equal probability <strong>of</strong> being tapped. Hence, letting R, S, <strong>and</strong> X be<br />

the radius <strong>of</strong> each disc, the size <strong>of</strong> the table, <strong>and</strong> the density <strong>of</strong> discs (fixed during<br />

t in each set <strong>of</strong> experiment), respectively, the probability <strong>of</strong> discovery, P, at each tap<br />

will be P=1rR~SX/S-TrR2X. As the frequency with which discs are touched (i. e. n)<br />

<strong>for</strong> tapping time to must be the product <strong>of</strong> P, t~, <strong>and</strong> the frequency <strong>of</strong> tapping per<br />

unit time (i. e. k), we have<br />

Comparing eq.<br />

n =Pkts<br />

= ~R~Xkts (4c. 4).<br />

(4c. 4) with eq. (4c. 1), we find that HOLLING'S factor a, his instanta-<br />

neous rate <strong>of</strong> discovery, is in fact ~R2k. As zc, R, <strong>and</strong> k can all be assumed to be<br />

constant, factor a can also be constant.<br />

Now, if eq. (4c. 1) is compared with eq. (4b. 5), it is clear that the <strong>for</strong>mer is a<br />

special case <strong>of</strong> the latter; that is, t in eq.<br />

(4b. 5) is replaced by t~, <strong>and</strong> Y is set equal<br />

to 1. Hence, HOLLING'S disc equation also applies to the NICHOLSON-BAILEY geometric<br />

model as in Fig. 4. This means that HOLLING'S model seems reasonable if, <strong>and</strong><br />

perhaps only if, the predator is either a sweep-feeder or tap-feeder (such as plankton-<br />

feeders or shore birds probing their beaks into the s<strong>and</strong>), but in either case discovery<br />

<strong>of</strong> prey should be made by bodily contact. This implies that the size <strong>of</strong> the disc is<br />

either the size <strong>of</strong> the body <strong>of</strong> the prey in the case <strong>of</strong> a tap-feeder, or the ambit <strong>of</strong><br />

the catching apparatus in the case <strong>of</strong> a sweep-feeder. Although both NICHOLSON-BAILEY<br />

<strong>and</strong> HOLLING assumed that the factor s or a involves the distance at which the<br />

predator perceives a prey (HOLLING 1961), the distance should be restricted, <strong>for</strong> the<br />

reason raised above, only to a very limited area around the prey or the predator. If<br />

this limitation is removed, the size <strong>of</strong> the disc should, in the case <strong>of</strong> the tap-feeder,<br />

set the upper limit <strong>of</strong> the disc density, because the probability <strong>of</strong> discovery is zcR2X,<br />

which should not exceed unity. This is somewhat artificial, <strong>and</strong> the problem will be<br />

discussed in w 4d <strong>and</strong> e.<br />

In HOLLING'S disc eq. (4c. 3), the number <strong>of</strong> predators searching does not explicitly


33<br />

appear. This is perhaps because HOLLING used only one finger <strong>and</strong> also because n is<br />

the number <strong>of</strong> discs removed per table. If, however, there are two fingers tapping<br />

independently, the frequency <strong>of</strong> tapping, k in eq. (4c. 4), will be doubled provided<br />

that there is no interference. In general, if there are Y fingers tapping per unit<br />

area <strong>of</strong> the table we have, in place <strong>of</strong> eq. (4c. 1),<br />

n =aXYG (4c. 5)<br />

in which n is the number <strong>of</strong> discs removed per unit area rather than per table.<br />

This generalization would not influence eq. (4c. 2), <strong>and</strong> so, eliminating t~, we get<br />

n = aXYt/(1 + ahX) (4c. 6).<br />

As the disc density has been fixed in the above model situation, eq. (4c. 6) is obviouslY<br />

an instantaneous equation comparable to eq. (3.1), in which<br />

f(X) =aX/ (l +ahX) (4c. 7).<br />

Hence the overall hunting equation <strong>for</strong> <strong>predation</strong>, i.e. eq. (3.13), will be evaluated<br />

by integrating<br />

dx/dt = -ax Y/ (1 + ahx) (4c. 8).<br />

Thus we have<br />

z =x0 (1 - e -~(rt-7~)) (4c. 9).<br />

Equation (4c. 9) is an overall <strong>for</strong>m <strong>of</strong> HOLLINC'S disc equation, taking account <strong>of</strong> the<br />

effect <strong>of</strong> diminishing returns, <strong>and</strong> is thus comparable with the LOTKA-VOLTERRA <strong>and</strong><br />

the NICHOLSON-BAILEY equations. It is at once clear that the equation is a generalized<br />

NICHOLSON-BAILEY model, or that the latter is a special case <strong>of</strong> the <strong>for</strong>mer in which<br />

factor h-0 (cf. eq. (3.8)). Equation (4c. 9) represents a surface in a Z-Xo-Yt coordinate<br />

system, the shape <strong>of</strong> which is very much like that in Fig. 3a. A cross-section<br />

parallel to the z-Yt plane shows the effect <strong>of</strong> diminishing returns similar to the<br />

cross-section in the NmHOLSON-BAILEY competition surface (Fig. 2). The shape <strong>of</strong><br />

a cross-section parallel to the Z-Xo plane in HOLLING'S surface is, however, curvilinear,<br />

unlike the NICHOLSON-BAILEY one which is rectilinear (see Fig. 2a). (It is difficult to<br />

make the variable z in eq. (4c. 9) perfectly dependent : nevertheless the surface can<br />

be drawn by assuming that Xo is the dependent variable <strong>and</strong> z <strong>and</strong> Yt independent<br />

ones.)<br />

If the disc model is applied to <strong>parasitism</strong> <strong>of</strong> the indiscriminate type, the righth<strong>and</strong><br />

side <strong>of</strong> eq. (4c. 7) should be substituted <strong>for</strong> f(x) in eq. (3. 22). Hence, if the<br />

fingers tap, <strong>for</strong> example, entirely at r<strong>and</strong>om, the function r will be the zero-term <strong>of</strong><br />

a POISSON series, <strong>and</strong> so we have<br />

z=X(1-e -a~/(l+ahz~) (4c. 10).<br />

This also generates a surface similar to that in Fig. 3a.<br />

My intention in generalizing HOLLING'S disc equation is in fact to point out three<br />

mistakes commonly seen in the scattered publications in which the original disc equation<br />

was applied directly to observed data (see e.g. HOLLING 1959). First, the<br />

density <strong>of</strong> the predators is not always 1. If eq. (4c. 3) is fitted to observed data in<br />

which Yr the estimates <strong>of</strong> the factor a <strong>and</strong> h in the equation inevitably involve


34<br />

the effect <strong>of</strong> Y. In order to show the point, let us assume that a <strong>and</strong> h in eq. (4c. 6)<br />

are the true estimates <strong>of</strong> the factors originally defined, whereas those in eq. (4c. 3)<br />

are superficial <strong>and</strong> are written a' <strong>and</strong> h' respectively. Then the fit <strong>of</strong> eq. (4c. 3) to<br />

data, which should be correctly fitted by eq. (4c. 6), means that<br />

a'Xt/ (1 + a'h'X) = aXYt/ (1 + ahX)<br />

<strong>and</strong> transposing we get<br />

h'=h/Y+ (a'-aY)/aa'XY<br />

or<br />

a'=aY/ (l +aX(h- h'Y) ).<br />

Thus, estimates a' <strong>and</strong> h' not only change as the predator density changes but also<br />

they are a decreasing function <strong>of</strong> the prey density, unless a'=aY <strong>and</strong> h'=h/Y. Fur-<br />

ther, they are not independent <strong>of</strong> the units in which the densities <strong>of</strong> both populations<br />

are measured. Hence, such estimates have no universal value.<br />

Secondly, the depletion <strong>of</strong> prey or, in the case <strong>of</strong> <strong>parasitism</strong>, super<strong>parasitism</strong>,<br />

<strong>of</strong>ten occurred in the observations. If eq. (4c. 3) is applied to such data, it is the<br />

same as fitting the instantaneous equation to one <strong>of</strong> the cross-sections parallel to the<br />

Z-Xo plane <strong>of</strong> the surface generated by eq. (4c. 9) in the case <strong>of</strong> <strong>predation</strong>, or eq.<br />

(4c. 10) in the case <strong>of</strong> <strong>parasitism</strong>. Obviously, the curve <strong>of</strong> a cross-section parallel to<br />

the Z-Xo plane, as in Fig. 3a, changes as Y or t changes, <strong>and</strong> there<strong>for</strong>e the estimates<br />

<strong>for</strong> factors a <strong>and</strong> h should change accordingly. Again such estimates have no univer-<br />

sal value.<br />

The third point is concerned with a precaution that should be taken when prey<br />

density is changed by changing the size <strong>of</strong> the experimental universe, a method <strong>of</strong>ten<br />

adopted to obtain an extremely high prey density with relatively few individuals (e. g.<br />

MORRIS 1963; HAYNES <strong>and</strong> SISOJEVIC 1966).<br />

If this is done, however, it should<br />

be noticed that the predator density changes too. To assess z in this method, we<br />

measure a cross-section <strong>of</strong> the surface which goes diagonally across the xo-Yt plane.<br />

As far as I know, all the published literature in which HOLLING'S disc equation<br />

is applied, involves at least one <strong>of</strong> the above three misapplications. Yet, the goodness<br />

<strong>of</strong> fit is, in many cases, remarkably high. This is because the equation involves two<br />

factors that are normally estimated from the observed relationships between z <strong>and</strong><br />

x0. That is to say, the nature <strong>of</strong> the estimation ensures that the resultant curve fits<br />

the observation well. Hence, a good fit does not constitute verification <strong>of</strong> the theory.<br />

In my opinion, the significance <strong>of</strong> HOLLING'S simulation experiment with discs<br />

<strong>and</strong> tapping is to show the importance <strong>of</strong> the factor h, <strong>and</strong> this simple model is<br />

sufficient to show that "the area <strong>of</strong> discovery", i.e. /i, cannot be independent <strong>of</strong> prey<br />

or host density, as clearly deduced from eqs. (4c. 9) or (4c. 10). However, an application<br />

<strong>of</strong> the original model as in eq. (4c. 3) to a situation in which the effect <strong>of</strong> diminishing<br />

returns is obviously involved, is incorrect.


35<br />

d).<br />

The IVLEV-GAusE equation<br />

IVLEV (1955), in his <strong>study</strong> <strong>of</strong> the feeding ecology <strong>of</strong> fish, proposed an equation<br />

<strong>and</strong> used it rather extensively <strong>for</strong> the analysis <strong>of</strong> <strong>predation</strong> processes. One <strong>of</strong> IVLEV'S<br />

fundamental ideas which led him to the <strong>study</strong> <strong>of</strong> feeding ecology appears to have<br />

emerged from his dissatisfaction with one <strong>of</strong> VOLTERRA'S assumptions that the number<br />

<strong>of</strong> prey taken by predators is a linear function <strong>of</strong> the prey density.<br />

his book (pp. 20-21, English edition, 1961):<br />

IVLEV wrote in<br />

"... , according to VOLTERRA'S position, in the case <strong>of</strong> an unlimited increase in<br />

the concentration <strong>of</strong> the food material, there must also be an unlimited increase in<br />

the amount <strong>of</strong> the food taken. This 'unlimited' increase is a biological absurdity,<br />

since each individual is only capable <strong>of</strong> consuming a strictly limited quantity <strong>of</strong><br />

food in each unit time."<br />

This criticism led IVLEV to propose a new hunting equation in the following quota-<br />

tion:<br />

"... the actual ration <strong>of</strong> food eaten by the predator over a certain period <strong>of</strong> time<br />

will, under favorable feeding conditions, tend to approach a certain definite size,<br />

above which it cannot under any circumstances increase <strong>and</strong> which also corresponds<br />

to the physiological condition <strong>of</strong> full saturation. Hence the mathematical interpret-<br />

ation <strong>of</strong> the given law takes a <strong>for</strong>m which has been used fairly widely in quanti-<br />

tative biology <strong>and</strong> physical chemistry. If the amount <strong>of</strong> the maximal ration is taken<br />

as b, then the relation between the size <strong>of</strong> the actual ration u <strong>and</strong> the density <strong>of</strong><br />

the prey population v must be proportional to the difference between the actual <strong>and</strong><br />

maximal rations <strong>and</strong> can be expressed by<br />

du/dv =a (b- u), (4d. 1)<br />

where a represents the coefficient <strong>of</strong> proportionality. Integrating this equation, we<br />

get<br />

u =b (1 - e -~') ." (4d. 2)<br />

(The symbols are mine).<br />

Although IVLEV'S criticism <strong>of</strong> VOLTERRA is a correct one, there appear to be some<br />

ambiguities <strong>and</strong> confusions in the second statement quoted above. It is not clear<br />

what IVLEV was aiming at in these equations, because <strong>of</strong> inadequate definition <strong>of</strong> the<br />

symbols in the equations. As far as I know, it is not explicitly mentioned anywhere<br />

in his book, whether eq. (4d. 2) represents an overall hunting equation or an instan-<br />

taneous relationship.<br />

IVLEV'S treatise covers a wide range <strong>of</strong> problems in feeding ecology, which<br />

includes problems in natural population <strong>and</strong> competition between species feeding on<br />

the same food resources. Needless to say, these problems cannot be solved unless<br />

the depletion <strong>of</strong> the resources is considered ; the notion <strong>of</strong> 'competition' in the VOL-<br />

TERRA-GAusE (as well as in the NICHOLSON-BA1LEY) line <strong>of</strong> thought would not have<br />

emerged without this fundamental phenomenon, the depletion <strong>of</strong> some essential requi-


36<br />

site, e.g. space <strong>and</strong> food. It should have there<strong>for</strong>e<br />

been <strong>of</strong> fundamental importance<br />

<strong>for</strong> IVLEV to make his position absolutely clear; whether he was aiming at an overall<br />

relationship, which includes the effect <strong>of</strong> depletion, or an instantaneous relationship,<br />

in which the effect is not considered.<br />

Having studied IVLEV's inference critically, I reached the conclusion that eq.<br />

(4d. 2) is not an overall hunting equation. This conclusion would seriously influence<br />

the value <strong>of</strong> IVLEV'S treatise, <strong>and</strong> there<strong>for</strong>e I should present the pro<strong>of</strong> <strong>of</strong> this conclu-<br />

sion in order to eliminate the possibility that my remarks are too critical.<br />

In order to prove my point, let us assume firstly that eq. (4d. 2) is an overall<br />

hunting equation equivalent to z=F(xo, Y, t) in my system presented in w 3. Thus,<br />

under this hypothesis, IVLEV'S symbol u is equivalent to my z, <strong>and</strong> obviously v cannot<br />

be x. it follows that v=xo, the initial prey density. Then eqs. (4d. 1) <strong>and</strong> (4d. 2)<br />

become respectively<br />

<strong>and</strong><br />

dz/dxo = a (b- z)<br />

z ~ b (1 - e- aXo )<br />

(4d. 1')<br />

(4d. 2').<br />

Now, according to IVLEV, Z approaches, over a certain period <strong>of</strong> time, the maximal<br />

ration b. In the meantime, b must be a positive value, <strong>and</strong> it has to be independent<br />

<strong>of</strong> x0, since otherwise eq. (4d. 2') will not be yielded by eq. (4d. 1'). However, since<br />

x0 is any given initial density <strong>of</strong> the prey, it can take any positive value. Thus, we<br />

can choose a value <strong>of</strong> Xo smaller than b. Under these circumstances, z can exceed<br />

the value <strong>of</strong> x0, which suggests that the predators can eat more than supplied, <strong>and</strong><br />

this is <strong>of</strong> course absurd.<br />

It is clear that IVLEV's symbol u cannot be z, <strong>and</strong> there<strong>for</strong>e we have to give up<br />

the hypothesis that eq. (4d. 2) represents an overall hunting equation. The second<br />

alternative is there<strong>for</strong>e that u is the number <strong>of</strong> prey taken by a given number <strong>of</strong><br />

predators per unit area <strong>for</strong> a given period <strong>of</strong> time when the prey density is kept<br />

constant; the equation represents an instantaneous relationship. Under this hypothesis,<br />

u is equivalent to n, <strong>and</strong> it follows that v is X, a fixed prey density. Hence, from<br />

eqs. (4d. 1) <strong>and</strong> (4d. 2) we have<br />

<strong>and</strong><br />

dn/dX:a (b- n) (4d. 1")<br />

n = b (1 - e -~x) (4d. 2").<br />

Equation (4d. 2") has no contradiction as an instantaneous<br />

hunting equation, i.e.<br />

n =f(X)Yt, if the factor Yt in eq. (4d. 2") is either unity or involved in the factor<br />

b. The latter situation is a general one, <strong>and</strong> so we can write:<br />

b~bYt.<br />

So that we have<br />

n =b(1-e -~x) Yt (4d. 3).<br />

IVLEV, when fitting his eq. (4d. 2) to observations with some fish species (figure<br />

1 <strong>of</strong> his book), stated that five young fish were introduced into a container <strong>and</strong><br />

allowed to feed <strong>for</strong> 1.5 to 2 hours. No mention was made, however, <strong>of</strong> either how


37<br />

big the container Was or if the prey density was kept constant; though, judging from<br />

his statement (chapter 2, p. 18 <strong>of</strong> his book) that "Food consumption was studied...<br />

by estimation <strong>of</strong> the food left over out <strong>of</strong> the quantity given .... ", the prey density<br />

was apparently not kept constant. Obviously, in the light <strong>of</strong> my analysis leading to<br />

eq. (4d. 3), IVLEV's variable u must have been influenced by the size <strong>of</strong> container <strong>and</strong><br />

consequently the density <strong>of</strong> fish: more precisely, the size <strong>of</strong> container influence-~ the<br />

density <strong>of</strong> fish, <strong>and</strong> consequently the coefficient b, if the density <strong>of</strong> food species was<br />

kept constant to meet the condition required <strong>for</strong> describing the instantaneous relationship.<br />

If the food species diminished gradually during the course <strong>of</strong> <strong>predation</strong>, the<br />

estimates <strong>of</strong> both coefficients a <strong>and</strong> b obtained by fitting eq. (4d. 3), even though it<br />

takes factors Y <strong>and</strong> t into account, would have been different between observations<br />

with different predator density, simply because it amounts to fitting an instantaneous<br />

equation to one particular cross-section <strong>of</strong> the hunting surface. Hence, these coefficients<br />

a <strong>and</strong> b estimated in IVLEV'S experiments are specific to these experiments <strong>and</strong><br />

have no universal meaning. This is the same criticism I raised in regard to HOLLINO'S<br />

model. In order to eliminate the awkwardness pointed out above, it is necessary to<br />

deduce an overall hunting equation. But be<strong>for</strong>e doing so, I shall examine in more<br />

detail the reason why IVLEV'S instantaneous equation takes such specific <strong>for</strong>m: although<br />

eq. (4d. 3) has no apparent contradiction as an instantaneous hunting equation, the<br />

justification <strong>for</strong> starting our inference from the differential eq. (4d. 1) is yet to be<br />

rationalized.<br />

IVLEV obtained his idea concerning eq. (4d. 1) from three existing equations<br />

developed in physical chemistry <strong>and</strong> physiology. The first was an equation <strong>for</strong> unimolecular<br />

reaction. The velocity equation <strong>for</strong> a simple unimolecular reaction takes<br />

the <strong>for</strong>m<br />

dx/dt = -ax<br />

where x is the density <strong>of</strong> molecules at time t, <strong>and</strong> a the reaction coefficient. So,<br />

letting x0 be the initial density <strong>of</strong> the molecules, we have<br />

z =x0 (1 - e -~)<br />

where z--xo-x. As already seen, this is the NICHOLSON-BAILEY equation in which<br />

Y=I. Of course, there is a resemblance between the velocity equation <strong>and</strong> IVLEV's<br />

equation in their mathematical <strong>for</strong>m, but the meanings are entirely different since<br />

the derivative dx/dt in the velocity equation is the rate <strong>of</strong> change in time, whereas<br />

dn/dX in IVLEV'S is the rate <strong>of</strong> change with density. These two attributes are,<br />

needless to say, totally irrelevent to each other. There<strong>for</strong>e, the quotation <strong>of</strong> the<br />

velocity equation by IVLEV is absolutely irrelevant in his context.<br />

IVLEV also quoted the "WEBER-FECHNER law" (but without citing the literature<br />

source). As far as I know, this is a law in neuro-physiology representing the relationship<br />

between the strength <strong>of</strong> a stimulus <strong>and</strong> the reaction <strong>of</strong> a nerve. However,<br />

there appears to be no possible resemblance between this law <strong>and</strong> IVLEV'S equation<br />

in their mathematical <strong>for</strong>ms.


38<br />

The last one is MITSCHERLICH'S <strong>for</strong>mula concerning the relationship between<br />

plant growth <strong>and</strong> nutrient supply. MITSCHERLICH (cited by RUSSELL 1961) assumed<br />

that a plant or crop should produce a certain maximum yield if all conditions were<br />

ideal, but ins<strong>of</strong>ar as any essential factor is deficient there is a corresponding short-<br />

age in the yield. Further, it is assumed that the increase <strong>of</strong> crop produced by unit<br />

increment <strong>of</strong> the lacking factor is proportional to the decrement from the maximum,<br />

thus<br />

dn/dx=a(b-n)<br />

where n is the yield obtained when the amount <strong>of</strong> the factor present is X, b the<br />

maximum yield obtainable if the factor was present in excess, <strong>and</strong> a is constant.<br />

While it may be underst<strong>and</strong>able that IVLEV quoted MITSCHEgLmH'S equation,<br />

since prey density can, in a way, be compared to the nutrient supply in plants, the<br />

comparison is more like a metaphorical juxtaposition <strong>and</strong> does not really aid his<br />

argument in <strong>predation</strong>. IVLEV'S failure to recognize the distinction between the overall<br />

<strong>and</strong> instantaneous relationships, <strong>and</strong> also to incorporate the factors Y <strong>and</strong> t, implies<br />

that such comparisons did not give a sufficient insight into the structure <strong>of</strong> <strong>predation</strong>.<br />

Also, as already mentioned, there will be a number <strong>of</strong> mathematical equations which<br />

can equally well describe the trend <strong>of</strong> the hunting curve that IVLEV obtained <strong>for</strong><br />

fish <strong>predation</strong> (figure I <strong>of</strong> IVLEV'S book).<br />

For instance, HOLLING's disc equation<br />

shows a very similar trend, yet the meaning <strong>of</strong> the equation is quite different from<br />

MITSCHERLICH'S <strong>for</strong>mula. As I have shown in w 3, the <strong>for</strong>mulation <strong>of</strong> a differential<br />

equation in <strong>predation</strong> <strong>models</strong> has a definite significance as a means <strong>of</strong> inference. That<br />

is, it is <strong>of</strong>ten an instantaneous relationship, between the components concerned, that<br />

is self-explanatory <strong>and</strong> so can be <strong>for</strong>mulated intuitively <strong>and</strong> correctly. IVLEV's equa-<br />

tion, however, appears not to have been developed theoretically, but has been borrow-<br />

ed by metaphor from other fields in which the development <strong>of</strong> a differential equation<br />

has no relevant significance to <strong>predation</strong>.<br />

It is, however, true that IVLEV'S equation has a remarkable descriptive power.<br />

It might there<strong>for</strong>e be that some essential meaning or some fundamental structure <strong>of</strong><br />

<strong>predation</strong> processes is reflected in the equation. To find this out, the following sim-<br />

ulation experiments will be considered.<br />

Suppose a set-up similar to the disc experiments by HOLLING. The only difference<br />

is that the distribution <strong>of</strong> the discs is ideally at r<strong>and</strong>om, <strong>and</strong> hence the discs may<br />

overlap each other. To make calculation easier, however, the model is slightly modified,<br />

though the principle <strong>of</strong> the model process remains unaffected. It will be assumed that<br />

the prey individuals are particles <strong>and</strong> a predator has a ring <strong>of</strong> radius R around itself.<br />

Then, the ring, representing the recognition zone <strong>of</strong> the predator, is tossed over the<br />

hunting area (a table), again at r<strong>and</strong>om; it should be strictly at r<strong>and</strong>om so that every<br />

part <strong>of</strong> the table has an equal chance <strong>of</strong> receiving a toss <strong>of</strong> the ring.<br />

Under these circumstances, the total number <strong>of</strong> particles (written as n~) falling<br />

within the ring after a number <strong>of</strong> tosses will be in proportion to the density <strong>of</strong> the


39<br />

particles.<br />

As the average number <strong>of</strong> particles within the ring is 7rR2X <strong>and</strong> the<br />

number <strong>of</strong> tosses in time ts is kt,, where k is the frequency <strong>of</strong> tosses per unit time,<br />

the total number caught by the ring will be<br />

n~ =~R~kXt~.<br />

Although the right-h<strong>and</strong> side <strong>of</strong> the above <strong>for</strong>mula is the same as in eq. (4c. 4)<br />

describing HOLLING'S experiment, notation n, in the left-h<strong>and</strong> side is not n as in eq.<br />

(4c. 4). It was assumed in HOLLING'S model that all the discs tapped were taken,<br />

<strong>and</strong> it was a justifiable assumption since only one disc was discovered at a time as<br />

there was no overlap <strong>of</strong> the discs. In the present model, however, more than two<br />

particles may occur within the ring at one toss. Thus, if the predator is allowed to<br />

catch only one prey individual within its recognition zone at a time, the total number<br />

captured (i. e. n) will be the number <strong>of</strong> successful tosses rather than ns, the total<br />

number discovered.<br />

In the present model, the distribution <strong>of</strong> the prey individuals <strong>and</strong> the tosses <strong>of</strong><br />

the ring are ideally at r<strong>and</strong>om, <strong>and</strong> so, provided that R is not too large as compared<br />

with the table, the frequency <strong>of</strong> tosses in which a given number <strong>of</strong> particles falls<br />

must follow a POISSON series with its mean equal to the average number <strong>of</strong> particles<br />

in the ring at one toss, i.e. rcR~X. Hence, as the frequency with which no particle<br />

occurs within the ring in the period t, will be kt,e -~R*X the total frequency <strong>of</strong><br />

successful tosses in t, will be<br />

n =kt, (l-e- ~R~X ).<br />

In the above model, only one ring was assumed to be tossed at the frequency <strong>of</strong> k<br />

per unit time per table.<br />

If there are Y rings per unit area tossed independently,<br />

each at the average frequency <strong>of</strong> k per unit time, the total frequency is kYts, <strong>and</strong><br />

so the above equation becomes<br />

n=kYt~(1-e -'~R~X ) (4d. 4).<br />

If kYt~ in the above equation is set equal to b, <strong>and</strong> rcR 2 to a, we have IVLEV'S eq.<br />

(4d. 2"). Thus, we find that IVLEV'S equation, in terms <strong>of</strong> the toss-a-ring simulation,<br />

is a generalized version <strong>of</strong> HOLLING'S experiment, since eq. (4d. 4) is a generalization<br />

<strong>of</strong> eq. (4c. 5). That is, recognition zones can overlap, <strong>and</strong> so there is no restriction<br />

to the range <strong>of</strong> variation in the prey density X (remembering that HOLLING'S eq.<br />

(4c. 5) holds in the disc experiment only <strong>for</strong> X


40<br />

finding a disc. So in HOLLING's disc experiment, the value n/Yt should never exceed<br />

k/(l+hk). It should be noticed, however, that eq. (4c. 6) describing HOLLING'S disc<br />

situation also holds <strong>for</strong> sweep-feeding <strong>predation</strong> as in the NICHOLSON-BAILEY model.<br />

In this case, there is no limitation to the prey density, <strong>and</strong> so the maximum ration<br />

is set at n/Yt~l/h.<br />

The effect <strong>of</strong> the h<strong>and</strong>ling time, h, can easily be incorporated into the toss-a-ring<br />

equation by eliminating t~ from eqs.<br />

kYt(l_e -~R2X)<br />

n=l + hk(l_e- ~R2x )<br />

(4c. 2) <strong>and</strong> (4d. 4). Thus we have<br />

(4d. 5).<br />

In the above equation, the maximum value <strong>of</strong> n/Yt is k/(l+hk) which is reached<br />

asymptotically as X becomes infinity. This elaboration, however, may be <strong>of</strong> doubtful<br />

value, <strong>for</strong> the following reasons. First, the purpose <strong>of</strong> the toss-a-ring model is only<br />

to explain why IVLEV'S equation takes that <strong>for</strong>m, <strong>and</strong> also to find where the missing<br />

variables, the predator density <strong>and</strong> the time factor, should appropriately be placed.<br />

Secondly, if the model is to reflect a certain type <strong>of</strong> <strong>predation</strong> process, it may be<br />

compared only to that <strong>of</strong> some shore birds probing with their beaks to find food,<br />

but it is not a common method <strong>of</strong> hunting <strong>for</strong> the predators <strong>of</strong> interest. There<strong>for</strong>e,<br />

it would be wise to leave IVLEV's equation in the <strong>for</strong>m, i.e. eq. (4d. 3) :<br />

n = b (1 - e -~x) Yt,<br />

as purely descriptive without attaching any serious meaning to it. Equation (4d. 3)<br />

has an excellent power <strong>of</strong> describing instantaneous hunting curves <strong>of</strong> the shape <strong>of</strong><br />

the one in Fig. 1, which has been widely observed with various predators.<br />

As long as it is remembered that eq. (4d. 3) is an empirical <strong>for</strong>m <strong>of</strong> an instant-<br />

aneous hunting equation, it may be used safely, according to the situation to which<br />

the equation is applied, either as in the present <strong>for</strong>m or as the basis <strong>of</strong> deducing a<br />

theoretical trend in the higher order <strong>of</strong> synthesis. For example, I have used IVLEV'S<br />

equation <strong>for</strong> my model <strong>of</strong> the clutch size variation in birds (RoYAMA 1969) <strong>and</strong> the<br />

instantaneous <strong>for</strong>m <strong>of</strong> HOLLING'S equation <strong>for</strong> explaining the hunting behaviour <strong>of</strong><br />

the great tit (Parus major L.) (ROYAMA 1970). In both cases, I could neglect depletion<br />

<strong>of</strong> the prey density caused by <strong>predation</strong> by the birds. If, however, eq. (4d. 3) is to<br />

be applied to a situation in which the prey density is being depleted, further synthesis<br />

is needed to deduce an overall hunting equation. This has in fact been done in w 3,<br />

as in eq. (3. 12) derived from eq. (3.10). If eq. (4d. 3) is used to describe the<br />

instantaneous relationship in <strong>parasitism</strong> <strong>of</strong> the indiscriminate type, eq.<br />

than eq.<br />

(3.12) should be used.<br />

(3. 24) rather<br />

A brief mention will be made <strong>of</strong> an equation that GAUSE (1934) had proposed<br />

be<strong>for</strong>e IVLEV. GAUSE, in his <strong>study</strong> <strong>of</strong> protozoan predators, found that the rate <strong>of</strong><br />

increase per individual in the predator population was not a linear function <strong>of</strong> prey<br />

density, <strong>and</strong> he proposed an exponential function to replace the equation, i.e. eq.<br />

(4. lb), proposed by LOTKA <strong>and</strong> VOLTERRA <strong>for</strong> that relationship. Thus


41<br />

(dy/dt)/y = C (1- e-~) (4d. 6)<br />

where C <strong>and</strong> ,t are positive constants. I shall examine in the following: (1) whether<br />

GAUSE'S proposal <strong>of</strong> eq.<br />

(4d. 6) takes the same <strong>for</strong>m as that <strong>of</strong> IVLEV'S eq.<br />

(4d. 6) is acceptable, <strong>and</strong> (2) why the right-h<strong>and</strong> side <strong>of</strong> eq.<br />

(4d. 2').<br />

Although GAUSE was aiming at the <strong>for</strong>mulation <strong>of</strong> the relationships between a<br />

protozoan predator <strong>and</strong> its prey, his experimental justification <strong>of</strong> eq. (4d. 6) came<br />

from an observation by ~MIRNOV <strong>and</strong> WLADIMIROW (cited by GAUSE 1934, pp. 139-140)<br />

<strong>of</strong> a parasite, Mormoniella vitripennis, attacking its host, Phormia groenl<strong>and</strong>ica. It<br />

should be remembered that the relationship expressed in the <strong>for</strong>m <strong>of</strong> a differential<br />

equation as in eq. (4d. 6) will not be appropriate in the case <strong>of</strong> an entomophagous<br />

parasite in which generations are discrete.<br />

This is because, while the expression<br />

C(1-e -~*) st<strong>and</strong>s <strong>for</strong> the number <strong>of</strong> progeny (per unit area) produced per parasite<br />

during t in the present generation to <strong>for</strong>m the next generation, these progeny will<br />

not reproduce in the present generation. There<strong>for</strong>e, it is incorrect to equate the (dy/<br />

dt)/y to C(1-e-~*). Be<strong>for</strong>e making further comments on eq. (4d. 6), however, I shall<br />

investigate the second point, i.e. why were changes in the density <strong>of</strong> the parasites'<br />

progeny, in relation to changes in host density as observed by SMIRNOV <strong>and</strong> WLADI-<br />

MIROW, described by the <strong>for</strong>mula C(1-e -~) which is <strong>of</strong> the same <strong>for</strong>m as IVLEV'S ?<br />

Let Y' be the density <strong>of</strong> progeny in the parasite population produced to <strong>for</strong>m the<br />

next generation, <strong>and</strong> z the density <strong>of</strong> hosts attacked in the present generation. Let<br />

us assume hypothetically that Y' is more or less proportional to z, i.e. Y'=c'z where<br />

c' is a proportionality constant. In the meantime, it has been shown that although it<br />

is an instantaneous equation, eq.<br />

(4d. 3) can nevertheless describe one cross-section,<br />

parallel to the z-X plane, <strong>of</strong> an overall hunting surface <strong>for</strong> <strong>parasitism</strong>, e. g. eq.<br />

(3.22). In other words, although eq. (4d. 3) should correctly be used to estimate the<br />

value <strong>of</strong> n, the equation can, because <strong>of</strong> its considerable flexibility in fitting, also<br />

describe the z-X relationship, provided that the coefficients a <strong>and</strong> b are appropriately<br />

chosen <strong>for</strong> a given value <strong>of</strong> Yt, i.e. <strong>for</strong> a given cross-section <strong>of</strong> the hunting surface.<br />

Under these circumstances, one may obtain, though only superficially,<br />

Y'=c'b(1-e -ax) Yt,<br />

or by transposing Yt to the left-h<strong>and</strong> side,<br />

Y'/Yt =c'b (1 -e -ax) (4d. 7).<br />

In the SMIRNOWWLADIMIROW observation, as presented in figure 40 <strong>of</strong> GAUSE'S<br />

(1934) book, the factor Yt appears to be fixed at the same value <strong>for</strong> different values<br />

<strong>of</strong> X throughout the observation, <strong>and</strong> this consistency satisfies the condition under<br />

which eq. (4d. 7) has been derived.<br />

Clearly, the expressions Y'/Yt <strong>and</strong> X in eq.<br />

(4d. 7) are equivalent to (dy/dt)/y <strong>and</strong> x respectively in eq. (4d. 6), <strong>and</strong> the constants<br />

c'b <strong>and</strong> a in eq. (4d. 6) can be written as C <strong>and</strong> 2 respectively as in eq. (4d. 6).<br />

And this is why GAUSE's equation can be deduced from IVLEV'S.<br />

The reasoning which led to eq. (4d. 7) explains why GAUSE adopted eq. (4d. 6),<br />

<strong>and</strong> also why GAUSE <strong>and</strong> IvLgv proposed the same function, though GAUSE was


42<br />

thinking <strong>of</strong> a function equivalent to g2 in eq. (4a. 2b) while IVLEV was thinking <strong>of</strong><br />

one equivalent to the function f in eq.<br />

(4a. 2a). This reasoning suggests, however,<br />

that GAUSE'S propgsal <strong>of</strong> his equation, the justification <strong>of</strong> which was based erroneously<br />

on an observation <strong>of</strong> parasites with discrete generations, is not acceptable, <strong>for</strong> two<br />

reasons, if applied to predators with continuous generations. First, as pointed out by<br />

NICHOLSON <strong>and</strong> BAILEY (1935, p. 552, first paragraph), the number <strong>of</strong> progeny in a<br />

predator population, unlike that in a parasite population, will not in general be propor-<br />

tional to the number <strong>of</strong> prey eaten by the parental predators.<br />

be avoided if it could be verified that the exponential function as in eq.<br />

This objection might<br />

(4d. 6) still<br />

holds <strong>for</strong> protozoan predators in which the value <strong>and</strong> the ecological significance <strong>of</strong><br />

the coefficient ~ are different from those <strong>of</strong> the coefficient a in eq. (4d. 7), <strong>and</strong> there-<br />

<strong>for</strong>e that the similarity between the two equations is only coincidental.<br />

My second objection, however, seems unavoidable. Clearly, dy/dt is a linear func-<br />

tion <strong>of</strong> y in eq. (4d. 6), suggesting that the predator population exhibits a geometric<br />

increase <strong>for</strong> any given value <strong>of</strong> x. This contradicts GAUSE'S own suggestion <strong>of</strong> a<br />

logistic law with respect to the natural increase <strong>of</strong> the prey population in the absence<br />

<strong>of</strong> predators; in particular with respect to the function gl (x) in the generalized LOTKA-<br />

VOLTERRA eq. (4a. 2a). GAUSE'S inconsistency in this respect, i.e. adopting the logistic<br />

law <strong>for</strong> gl <strong>and</strong> neglecting it <strong>for</strong> gz, appears to be attributed to the misconception in<br />

his statement (GAuSE 1934, p. 53, last paragraph): "In the general <strong>for</strong>m the rate <strong>of</strong><br />

increase in the number <strong>of</strong> individuals <strong>of</strong> the predatory species resulting from the<br />

devouring <strong>of</strong> the prey dN2/dt Eequivalent to dy/dt in my notation] can be represented<br />

by means <strong>of</strong> a certain geometrical increase which is realized in proportion to the<br />

unutilized opportunity <strong>of</strong> growth.<br />

This unutilized opportunity is a function <strong>of</strong> the<br />

number <strong>of</strong> prey at a given moment." It appears that GAUSE overlooked in the last<br />

sentence above that the 'unutilized opportunity <strong>of</strong> growth' in the predator population<br />

is a function not only <strong>of</strong> the density <strong>of</strong> prey but also <strong>of</strong> the density <strong>of</strong> the predators<br />

at the same time.<br />

To summarize, in section IV <strong>of</strong> chapter III in his book, GAUSE (1934) tried to<br />

improve the LOTKA-VOLTERRA <strong>for</strong>mulation <strong>of</strong> the prey-predator system. His sugges-<br />

tions, however, were reasonable only with respect to the function gl (x) in eq. (4a. 2a)<br />

<strong>and</strong> not with g2(x, y) in eq. (4a. 2b). As to the function f(x, y) in eq. (4a. 2a),<br />

there was no suggestion by GAUSE, <strong>and</strong> the function f was left uncriticized as a<br />

linear function <strong>of</strong> x, which is unreasonable. For these reasons, GAUSE'S mathematical<br />

investigations into the prey-predator interaction system are unsatisfactory.<br />

e). ROYAMA'S model <strong>of</strong> r<strong>and</strong>om searching <strong>and</strong> probability <strong>of</strong> r<strong>and</strong>om encounters<br />

In the previous sections, it has been shown that the instantaneous hunting func-<br />

tion, i. e. f, is, unlike the simple assumption by LOTKA <strong>and</strong> VOLTERRA <strong>and</strong> by<br />

NICHOLSON <strong>and</strong> BAILEY, not a linear function <strong>of</strong> prey or host density. This finding<br />

influences the notion <strong>of</strong> 'the area <strong>of</strong> discovery' that was originally used by NICHOLSON


<strong>and</strong> BAILEY but redefined in w<br />

43<br />

as /~=(1/Y) In{xo/(Xo-Z)} <strong>for</strong> <strong>predation</strong>, <strong>and</strong> /i=<br />

(l/Y) ln{X/(X-z)} <strong>for</strong> <strong>parasitism</strong>. If the instantaneous function f(x)is a non-linear<br />

function, z is also a non-linear function <strong>of</strong> x0 or X, <strong>and</strong> hence /~ cannot be constant.<br />

Although the <strong>models</strong> proposed by HOLLING <strong>and</strong> IVLEV, reviewed in w 4c <strong>and</strong> d, are<br />

adequate to show that /~ will not be constant theoretically, these <strong>models</strong> are not quite<br />

sufficient to show how /~ can logically vary. One <strong>of</strong> the key points <strong>for</strong> this analysis<br />

seems to lie in underst<strong>and</strong>ing the coefficient a (not /~).<br />

The coefficient a appeared consistently in the instantaneous hunting function <strong>of</strong><br />

<strong>models</strong> reviewed in preceding sections, with the same geometric meaning, namely the<br />

size <strong>of</strong> an area immediately around either a prey or a predator individual within<br />

which the predator would take action to catch the prey. In all the <strong>models</strong>, it was<br />

assumed that the coefficient a was constant, or independent <strong>of</strong> variables x, y, <strong>and</strong> t.<br />

Some objections arose among ecologists (see below) against the validity <strong>of</strong> the assump-<br />

tion that the coefficient was constant,<br />

yet. The main purpose <strong>of</strong> the present<br />

nature <strong>of</strong> this coefficient.<br />

The justification <strong>of</strong> the assumption<br />

<strong>and</strong> the debate has not quite been settled<br />

section is to give a much clearer idea <strong>of</strong> the<br />

that a is constant is partly concerned with<br />

the validity <strong>of</strong> the assumption <strong>of</strong> r<strong>and</strong>om searching--whether or not such an assumption<br />

is reasonable in <strong>predation</strong> or <strong>parasitism</strong> theories. The idea <strong>of</strong> r<strong>and</strong>om searching <strong>and</strong><br />

r<strong>and</strong>om encounter is not new. It has been used very widely among theorists, <strong>and</strong><br />

while the concept <strong>of</strong> r<strong>and</strong>omness is clearly defined in the field <strong>of</strong> mathematics,<br />

confusion has resulted from loose application <strong>of</strong> the concept to an ecological situation.<br />

First, one must separate the concepts <strong>of</strong> r<strong>and</strong>om searching <strong>and</strong> r<strong>and</strong>om encounters,<br />

as the one does not necessarily follow the other. Perhaps the most satisfactory defini-<br />

tion <strong>of</strong> the term 'r<strong>and</strong>om searching' is that the path <strong>of</strong> each individual predator is<br />

affected neither by the location <strong>of</strong> prey individuals in the hunting area nor by the<br />

paths <strong>of</strong> other members <strong>of</strong> the predator population. 'R<strong>and</strong>om encounter', as opposed<br />

to 'r<strong>and</strong>om searching', can be defined as meaning that every prey individual in the<br />

area concerned has an equal probability <strong>of</strong> being encountered by predators searching<br />

<strong>for</strong> a limited time interval. Now, the simple theory <strong>of</strong> the kinetics <strong>of</strong> gas molecules,<br />

used by LOTKA, VOLTERRA, <strong>and</strong> NICHOLSON <strong>and</strong> BAILEY, assumes r<strong>and</strong>om encounters<br />

between molecules, <strong>and</strong> so the theory can be a model <strong>for</strong> r<strong>and</strong>om searching to a<br />

limited extent. This is because, as pointed out earlier, the analogy can be reasonably<br />

applied to hunting behaviour only when an encounter is made by bodily contact;<br />

otherwise, r<strong>and</strong>om encounters are approximately guaranteed only if either (1) the<br />

number <strong>of</strong> predators searching independently is high, (2) the time <strong>for</strong> searching is<br />

unlimited, or (3) the density <strong>of</strong> prey is relatively low.<br />

This can be illustrated by assuming that a few molecules are marked as predators,<br />

<strong>and</strong> all other molecules, unmarked, are prey. Then the movement <strong>of</strong> these few marked<br />

molecules <strong>for</strong> a certain limited time interval must be limited to a small fraction <strong>of</strong><br />

the total area concerned, so that the unmarked molecules located in the immediate


44<br />

vicinity <strong>of</strong> any one <strong>of</strong> the marked molecules must have higher chances <strong>of</strong> being<br />

encountered by the marked ones than those located remotely. However, if the number<br />

<strong>of</strong> marked molecules is high, all the unmarked molecules have an equal chance <strong>of</strong><br />

being encountered by any <strong>of</strong> the marked one~.<br />

LAING (1937) <strong>and</strong> ULLYETT (1947) were aware <strong>of</strong> a part <strong>of</strong> the above point <strong>and</strong><br />

stated, in their criticism <strong>of</strong> the NICHOLSON-BAILEY theory based on r<strong>and</strong>om searching,<br />

that a predator or a parasite can perceive the location <strong>of</strong> prey or host at a certain<br />

distance away from it ; the search there<strong>for</strong>e can only be at r<strong>and</strong>om (a free path or<br />

undirected path would be better words <strong>for</strong> r<strong>and</strong>om searching : my comment) as long<br />

as the predator or parasite remains outside this zone <strong>of</strong> perception;as soon as it<br />

enters this zone, its actions cease to be r<strong>and</strong>om <strong>and</strong> become, instead, directed towards<br />

the prey or host. THOMPSON (1939) thus wrote that a theory that equates animal<br />

action <strong>and</strong> r<strong>and</strong>om action covers at least only a fraction <strong>of</strong> the field, <strong>and</strong> that any<br />

theory based on the assumption that search is r<strong>and</strong>om cannot be accepted as a valid<br />

general theory.<br />

Although these criticisms sound reasonable, no attempt has been made, as far<br />

as I am aware, to find what <strong>and</strong> how much bias might be involved in a theory based<br />

on r<strong>and</strong>om searching as against the more likely situation raised by the above authors.<br />

I investigated this problem to some extent in a previous paper (RoYAMA 1966), but<br />

because it was written in a language not normally accessible to English-speaking<br />

readers, <strong>and</strong> also because I have since found some mathematical errors in it, all the<br />

points will be revised here.<br />

The investigation will be made with a situation in which a predator recognizes a<br />

prey from some distance. For the moment, it is assumed that only one predator is<br />

searching <strong>and</strong> that the prey density is kept constant. Suppose a number <strong>of</strong> particles<br />

<strong>of</strong> kind P (prey) are scattered at r<strong>and</strong>om over a sufficiently large plane (a two-dimensional<br />

space), each particle being given a circle <strong>of</strong> radius R as the recognition radius.<br />

One particle <strong>of</strong> another kind Q (a predator) can recognize P only within the area<br />

covered by the circles, so that Q's path is undirected when Q is outside the recognition<br />

area, but it is directed towards the nearest P within the recognition area. For<br />

simplicity, it is supposed that all P's remain stationary, but that Q moves around in<br />

the manner described. Also the radius R is the same <strong>for</strong> each P (although this assumption<br />

is not quite essential). At the moment, the time spent h<strong>and</strong>ling P's after<br />

they are caught by Q is not considered.<br />

Now, the starting point <strong>of</strong> Q will be determined on the plane at r<strong>and</strong>om. Then<br />

one or other <strong>of</strong> the following two cases wilt occur (see Fig. 5). Case 1: the starting<br />

point happens to be outside any <strong>of</strong> the circles <strong>of</strong> P's. Then Q determines the direction<br />

<strong>of</strong> its movement at r<strong>and</strong>om <strong>and</strong> moves along a (straight) free path until it contacts<br />

a P's circle; then goes to the centre <strong>of</strong> the circle to capture the P. Case 2: the starting<br />

point happens to be within one <strong>of</strong> the circles. Then Q immediately goes to the<br />

nearest centre, where the P is located, along the shortest path. A situation is first


45<br />

Fig. 5. ROYAMA'S first geometric model in which a predator (Q) recognizes<br />

its prey (P's, black dots) within a circle around each P, <strong>and</strong> searching<br />

by Q is discontinued each time a P is captured. A starting ponit (cross)<br />

is determined at r<strong>and</strong>om <strong>for</strong> each new search. For further explanation<br />

see the text.<br />

considered in which Q discontinues searching each time it captures a P, <strong>and</strong> so the<br />

next starting point is determined again at r<strong>and</strong>om over the plane.<br />

This is perhaps<br />

comparable to a bird collecting food <strong>for</strong> its young, <strong>and</strong> each time a food item is found<br />

it is taken to the nest, <strong>and</strong> the next hunting starts independently <strong>of</strong> the previous<br />

search.<br />

Suppose n <strong>of</strong> P's were caught by Q during the total time spent in searching, ts,<br />

(note that the number caught can be smaller than the number seen).<br />

Let n~ be the<br />

number <strong>of</strong> occurrences <strong>of</strong> case 1, L1 the average distance that Q travelled between<br />

the starting point outside any circle <strong>and</strong> the periphery <strong>of</strong> the first circle that Q hap-<br />

pened to encounter, <strong>and</strong> G the average speed <strong>of</strong> movement while Q was on an undi-<br />

rected path. Similarly, let n2 be the number <strong>of</strong> occurrences <strong>of</strong> case 2, L2 the average<br />

distance between the starting point inside the circles <strong>and</strong> the nearest centre <strong>of</strong> the<br />

circles, <strong>and</strong> Ve the average speed <strong>of</strong> movement along a directed path. Then we have<br />

the following <strong>for</strong>mula<br />

ts = ( L1/ VI +R/V2) nl q- ( L2/ V2) ne (4e. 1).<br />

Now, let Pr{nff <strong>and</strong> Pr{ne} be the probability that cases 1 <strong>and</strong> 2 occur respectively,<br />

then<br />

but since nl+n2=n,<br />

m=nPr{n,}<br />

n2 = nPr {n2},<br />

n2 =n (1-Pr {nd),<br />

<strong>and</strong> so eq. (4e. 1) becomes


46<br />

t,=[ (LJV~ + R/V2) Pr {nl} + L2(1-Pr{n~} )/V2]n (4e. 2).<br />

There<strong>for</strong>e, if L~, L~, <strong>and</strong> Pr{nl} are evaluated as functions <strong>of</strong> the density <strong>of</strong> pts (i. e.<br />

X), n can be determined as a function <strong>of</strong> X. Details <strong>of</strong> the evaluation <strong>of</strong> L1 <strong>and</strong><br />

L2 will be given in Appendices 1 <strong>and</strong> 2 respectively, <strong>and</strong> the end results alone will<br />

be shown below:<br />

L1 = 1/2RX (4e, 3)<br />

<strong>and</strong><br />

L2= {$(V'2rcX R)/1/X-Re ,R X }/(1--e ,,R X ) (4e. 4)<br />

where $(1/2~XR)=~2~f~Re<br />

-02/2d0"<br />

Now if we assume a sufficiently large area, the probability <strong>of</strong> Q's starting point<br />

happening to be outside any <strong>of</strong> the circles <strong>of</strong> P's must be the 0-term <strong>of</strong> a PomsoN<br />

series with its mean zcReX. Thus<br />

-~rR2X<br />

Pr {Ytl} =e (4e. 5).<br />

Eliminating L~, L2, <strong>and</strong> Pr{n~} from eqs. (4e. 2) to (4e. 5) inclusive <strong>and</strong> solving with<br />

respect to n, we have<br />

- ~r R~X<br />

n=[2RV~ V2/{V2 e +2R V~/X ~(v~2rcXR)} ]Xt8 (4e. 6).<br />

Now, if there is Y number <strong>of</strong> Q's per unit area, the total number <strong>of</strong> P's caught will<br />

be obtained simply by multiplying the right-h<strong>and</strong> side <strong>of</strong> eq. (4e. 6) by II, since the<br />

equation is an instantaneous one. So, if the expression in the outer brackets on the<br />

right-h<strong>and</strong> side <strong>of</strong> eq. (4e. 6) is denoted by a function a(X), where a is a functional<br />

symbol, we have<br />

n =a(X) XYt, (4e. 7).<br />

Equation (4e. 7) is clearly comparable to eq. (4b. 5) in the NICHOLSON-BAILEY model<br />

(note that in the NICHOLSON-BAII.EY model there is no distinction between t <strong>and</strong> ts)<br />

<strong>and</strong> is also comparable to eq. (4c. 5) in HOLLING'S model. That is to say, the coeffi-<br />

cient a <strong>for</strong> the two <strong>models</strong> is in fact comparable to the function a(X) in my model.<br />

Now, it is clear that the coefficient a can no longer be considered to be independent<br />

<strong>of</strong> X if a predator recognizes a prey from a distance <strong>and</strong> has to approach to seize it.<br />

The new coefficient a(X) is normally a decreasing function <strong>of</strong> prey density. (It should<br />

be noticed that the coefficient a in the NICHOLSON-BAILEY model is equal to 5 per<br />

unit time, but neither the a in HOLLINa'S disc model nor a(X) in eq. (4c. 7) is /i<br />

per unit time.) Let us examine the nature <strong>of</strong> a(X) more closely.<br />

If the density <strong>of</strong> P's (i. e. X) is very low, then<br />

lim a(X)=2RV1.<br />

X+O<br />

So, if R <strong>and</strong> 171 are independent <strong>of</strong> X, eqs. (4e. 6) or (4e. 7) converges with the NICHOLSON-<br />

BAILEY equation (cf. eq. (4b. 3)), or with HOLLISG'S be<strong>for</strong>e the introduction <strong>of</strong> the<br />

factor h. This is reasonable because, if the prey density is low, a large part <strong>of</strong> the<br />

predator's movement should be undirected <strong>and</strong> there<strong>for</strong>e independent <strong>of</strong> the location<br />

<strong>of</strong> prey individuals, in which case the analogy <strong>of</strong> unimolecular reaction is approxi-


47<br />

mately true. Consequently, HOLLING'S disc equation should be a good approximation<br />

<strong>for</strong> low prey densities.<br />

On the other h<strong>and</strong>, if X is very large, then<br />

lira a(X)=lim 2V~/v'X.<br />

X~o~ X~oo<br />

Thus, when X increases, a (X) decreases in inverse proportion to the square root <strong>of</strong><br />

X. This is because the probability <strong>of</strong> any one P being found in the recognition<br />

radius <strong>of</strong> a O increases as X increases, <strong>and</strong> so in its extreme situation, the movement<br />

<strong>of</strong> the Q is largely governed by the location <strong>of</strong> P's. Under these circumstances, the<br />

movement <strong>of</strong> the Q can no longer be independent <strong>of</strong> the location <strong>of</strong> P's. More<br />

precisely, if the Q is most attracted to the closest one <strong>of</strong> the P's found in the recog-<br />

nition area, then O's path consists largely <strong>of</strong> the distance to the closest P. As already<br />

demonstrated by MORISITA (1054) <strong>and</strong> CLARK <strong>and</strong> EVANS (1954), the distance between<br />

an arbitrarily selected point <strong>and</strong> the closest one <strong>of</strong> a number <strong>of</strong> r<strong>and</strong>omly distributed<br />

points in a two-dimensional plane is inversely proportional to the square-root <strong>of</strong> the<br />

density <strong>of</strong> the r<strong>and</strong>om points. The same conclusion applies when the recognition<br />

radius R is large relative to X. This suggests that HOLLING'S disc equation would<br />

show its bias, even if X is low, when R is very large. Conversely, if R is <strong>comparative</strong>ly<br />

small, the disc equation is again a good approximation even if X is high. That is to<br />

say, the equation holds whenever the R is reduced to an immediate area around the<br />

prey so that a capture is made practically by bodily contact, <strong>and</strong> this confirms my<br />

previous conclusion.<br />

It should be pointed out, however, that the inverse proportionality <strong>of</strong> a(X) to<br />

the square-root <strong>of</strong> X holds only when the hunting area is a two-dimensional plane.<br />

If the hunting area is one-dimensional, such as thin branches <strong>of</strong> a tree, a (X) is in-<br />

versely proportional to X;<strong>and</strong> if it is a three-dimensional space, a(X) is inversely<br />

proportional to the cubic-root <strong>of</strong> X (<strong>for</strong> the pro<strong>of</strong>, see CLARK 1956).<br />

The above investigation suggests that if HOLLING'S disc eq. (4c. 6) is applied to<br />

my present model, the coefficient a thus estimated will show a decreasing trend as X<br />

increases. MORRIS (1063) applied the disc equation to the instantaneous hunting curve<br />

observed with Podisus maculiventris (Hemiptera) eating larvae <strong>of</strong> Hyphantria cunea<br />

(Lepidoptera). (MORRIS kept the density <strong>of</strong> the prey constant during the observation<br />

to meet the condition <strong>for</strong> an instantaneous hunting curve.) It was found that, when<br />

the factor h was assumed to be constant, the factor a decreased as the prey density<br />

was increased. MORRIS thought that this was due to satiation by the predator. While<br />

this interpretation is reasonable, it is equally tenable that the trend observed was due<br />

to a decrease in the factor a as in my a(X). Hence, it is not known, unless the<br />

experiment is designed accordingly, to what extent the trend is attributable to satiation<br />

<strong>and</strong> to what extent to the geometric property <strong>of</strong> the hunting behaviour.<br />

MILLER (1960) pointed out that both HOLLING'S <strong>and</strong> WATT'S equations (to be<br />

reviewed in the next subsection) tended to deviate from the observed trend in various


48<br />

animals when the prey densities were increased. My factor a(X) suggests that in<br />

the case <strong>of</strong> <strong>parasitism</strong>, in which satiation might not be important, a deviation from<br />

the observed would become greater as host density became higher.<br />

Now, the h<strong>and</strong>ling-time factor h can be incorporated into my model by eliminating<br />

t~ from eqs. (4c. 2) <strong>and</strong> (4e. 7), i.e.<br />

n = {a (X) XYt}/{1 + a (X) hX} (4e. 8).<br />

Some interesting characteristics are observed in curves generated by eq. (4e. 8).<br />

The shape <strong>of</strong> the curves will change with changes in the ratio between the speed <strong>of</strong><br />

the predators' undirected movement <strong>and</strong> directed movement (see Fig. 6). As pointed<br />

out already, eqs. (4e. 6) or (4e. 7) has two asymptotes, the one <strong>for</strong> X-~0 where n approac<br />

2RVI Xt,, a straight line passing through the origin <strong>of</strong> the n-X coordinates; <strong>and</strong> the<br />

other <strong>for</strong> X~oo where n-~2V~v'Xt,. There<strong>for</strong>e: (1) If the speed <strong>of</strong> undirected movement<br />

is not lower than that <strong>of</strong> directed movement, i.e. V~ > V~, the curve generated<br />

by eq. (4e. 8) will be a monotonically increasing one with its tangent ever-decreasing;<br />

this is very much like the curve generated by HOLLING'S or IVLEV'S equations. (2) If<br />

Va is considerably smaller than V2, the curve may be a gentle sigmoid. (3) If V1<br />

<strong>and</strong> V~ have a certain ratio, the curve may be approximately linear <strong>for</strong> a wide range<br />

<strong>of</strong> the prey density, in which case NICHOLSON-BAILEY'S model, rather than HOLLING'S<br />

or IVLEV's, could be a better approximation (the Iinearity must sooner or later break<br />

down as X increases, however). All <strong>of</strong> these trends have been observed with living<br />

predators <strong>and</strong> parasites (see e.g. HOLLING 1959a). Equation (4e. 8) would not,<br />

however, generate very strongly sigmoid curves. A probable mechanism which causes<br />

such a strong sigmoid <strong>for</strong>m has been discussed elsewhere (RoYAMA 1970).<br />

R= 2<br />

v2-- 200<br />

h = 0.0'1<br />

(~) V~ = 600<br />

{2) = 50<br />

{3) = 1 0<br />

Fig. 6. Three examples <strong>of</strong> curves generated by eq. (4e. 8).<br />

:~X<br />

Now, from eqs. (4e. 8) <strong>and</strong> (3. 4), we have<br />

dx/dt = -a (x) xY/ {1 +a (x) hx} ,<br />

<strong>and</strong> so the overall hunting equation <strong>for</strong> <strong>predation</strong> is obtained by integrating the above<br />

differential equation, i.e.<br />

Yt = (1/2R V1) [ln (nR 2x) - nR 2x + (nR 2x) 2/2.21 - (nR ~x) 3/3.3 ! + ....<br />

]xo


49<br />

-- -- rr R~x Xo<br />

+ (l/V2) [2fO(V'2rrxR)l/x + (e )/rrR] x +h(xo-x) (4e. 9)<br />

(the expression [f (0)] 0~ reads f(Oo) -f(O)).<br />

Although I am unable to solve eq. (4e. 9) with respect to z:xo-x, the hunting surface<br />

on the Z-Xo-Yt coordinate system can be calculated numerically with a computer.<br />

Incidentally, the numerical values <strong>for</strong> the normal (or GAUSSIAN) integral ~ in eq.<br />

(4e. 4) can easily be obtained from a table <strong>of</strong> the normal (or GAUSSIAN) probability<br />

function.<br />

Surfaces generated by eq. (4e. 9) will almost certainly fit a wide variety <strong>of</strong> hunting<br />

surfaces actually exhibited by various predators, since the equation has four (rather<br />

than two as in HOLLING'S <strong>and</strong> IVLEV'S ones) independent coefficients, R, V,, Vz, <strong>and</strong><br />

h, to be estimated from the observed surfaces. This suggests that a close fit does<br />

not prove anything, unless the coefficients are measured independently <strong>and</strong> directly<br />

in separate observations, which at the moment is technically difficult. However, the<br />

model was designed primarily to show what degree <strong>of</strong> deviation <strong>of</strong> a(X) from the<br />

constant a in simpler <strong>models</strong> could be expected. No attempt was there<strong>for</strong>e made to<br />

fit the equation to any observed data.<br />

Another inquiry by a similar model will be made, secondly, into a situation in<br />

which a predator or a parasite continues to hunt without leaving the area each time<br />

a victim is captured, so that the starting point <strong>of</strong> a new path <strong>of</strong> search is the place<br />

where the previous victim is captured. This situation is too complex to h<strong>and</strong>le with<br />

an analytic (mathematical) method, <strong>and</strong> so a Monte Carlo simulation has been used.<br />

Two hundred points, representing prey, were plotted at r<strong>and</strong>om on a sheet <strong>of</strong><br />

graph paper (30• each point being given a circle <strong>of</strong> 2cm in radius. The<br />

predator's starting point was determined also at r<strong>and</strong>om. The direction <strong>of</strong> walk was<br />

then determined again by chance with predetermined probability <strong>of</strong> occurrence; five<br />

angles measured from a reference line, which was the previous path except <strong>for</strong> the<br />

starting point, i.e. 0, • <strong>and</strong> • were given an equal chance <strong>of</strong> occurrence (a<br />

backward movement with reference to the immediately previous path was excluded).<br />

When the predator had moved up to the periphery <strong>of</strong> a circle, it went to the centre,<br />

<strong>and</strong> the point was removed permanently from the area. If a second point was within<br />

the 2-cm recognition radius from that point, the predator went straight to the second<br />

one to catch it. If another point was not within 2cm, the predator determined its<br />

direction in the manner described above, <strong>and</strong> carried on. A part <strong>of</strong> this simulation<br />

experiment is shown in Fig. 7. The number <strong>of</strong> points removed is plotted on the<br />

horizontal axis against the distance traversed by the predator on the vertical axis<br />

(Fig. 8).<br />

If we assume an equal speed <strong>of</strong> movement <strong>for</strong> the undirected <strong>and</strong> the directed<br />

paths (i. e.<br />

VI= Vz), the distance traversed, L, per unit area is the product <strong>of</strong> the<br />

speed <strong>of</strong> movement <strong>and</strong> the predator-hours. Thus, if VI~V2=I, then L=Yt,. Also<br />

the effect <strong>of</strong> h<strong>and</strong>ling time is not essential in the present discussion <strong>and</strong> so is not


50<br />

Fig. 7. I{OYAMA'S second geometric model in which predator Q continues to search<br />

<strong>for</strong> prey without shifting the starting point after each capture. Other features<br />

are the same as in Fig. 5.<br />

(3<br />

LI.I<br />

0<br />

~E<br />

hl<br />

r,."<br />

U")<br />

I.I.I<br />

200<br />

~.~ 100<br />

I'--<br />

n"<br />

R= 2CN<br />

~._--<br />

I.L<br />

0<br />

6<br />

Z<br />

I I f I I I<br />

I00 200 300 400 500 600<br />

DISTANCE TRAVERSED (CM)<br />

Fig. 8. An observed relationship (solid line with black dots) between the<br />

number <strong>of</strong> prey taken (vertical axis) <strong>and</strong> the distance traversed by a<br />

predator (horizontal axis) in a Monte Carlo simulation <strong>of</strong> the <strong>predation</strong><br />

model <strong>of</strong> continuous search (cf. Fig. 7). The broken curve is generated<br />

by eq. (4e. 9) <strong>for</strong> R=2, 12"1=1, <strong>and</strong> h=0 as in the present Monte Carlo<br />

simulation. For details see text.<br />

considered; thus L= Yr. So the horizontal axis in Fig. 8 is equivalent to the Yt-axis<br />

in Fig. 2a, <strong>and</strong> the vertical axis is <strong>of</strong> course the z-axis. Thus the graph in Fig. 8 is<br />

equivalent to a cross-section in which x0=200/(30x39) parallel to the z-Yt plane in<br />

Fig. 2a, i.e. it is a competition curve in the NICHOLSON-BAILEY sense.<br />

The curve with solid circles is not monotonic but wavy, <strong>and</strong> there is a clear ten-


5T<br />

dency towards a number <strong>of</strong> miniature waves. The explanation <strong>of</strong> this trend is rather<br />

simple. As the direction <strong>and</strong> the length <strong>of</strong> each path are determined by chance, the<br />

predator's track is a kind <strong>of</strong> MA~KOV'S chain, although the probability distributions<br />

<strong>of</strong> both the direction <strong>of</strong> directed paths <strong>and</strong> the length <strong>of</strong> all paths are dependent<br />

stochastically on the location <strong>of</strong> the prey which were encountered9 This is a complex<br />

(or a generalized) 'r<strong>and</strong>om walk'9 Thus the predator's path <strong>of</strong> search was <strong>of</strong>ten<br />

deflected in an irregular manner <strong>and</strong>, because <strong>of</strong> the nature <strong>of</strong> a 'r<strong>and</strong>om walk', the<br />

predator tended to stay in a restricted area <strong>for</strong> some time. Consequently, the prey<br />

density in that vicinity was gradually depleted, <strong>and</strong> this caused a temporary drop in<br />

the predator's hunting efficiency. However, as the density <strong>of</strong> prey in the vicinity was<br />

lowered, the predator's undirected paths increased in length <strong>and</strong> eventually led to a<br />

pIace where the prey had not been exploited. Then the hunting efficiency increased<br />

temporarily be<strong>for</strong>e decreasing again. Thus the hunting curve was a kind <strong>of</strong> composite<br />

competition curve <strong>and</strong> became wavy. Figure 8 also includes a curve (broken line)<br />

calculated from eq. (4e. 9) <strong>for</strong> the same values <strong>of</strong> R <strong>and</strong> V (h=0, <strong>of</strong> course)9 The<br />

observed curve in this simulation model is always lower than the calculated one,<br />

but this is because factor h is not considered (see p. 54).<br />

Now, the same principle should also apply to an indiscriminate parasite9 The<br />

same set-up was used again except that none <strong>of</strong> the points (hosts) was removed from<br />

the area <strong>and</strong> parasitized hosts were left exposed to super<strong>parasitism</strong>. It was assumed<br />

that one egg was laid at each encounter. The result <strong>of</strong> the first experiment is shown<br />

•,,• 100<br />

ILl<br />

N<br />

I-.-<br />

8C<br />

/-<br />

2 OBS. .o'"<br />

/ "/jl~ -o<br />

c>- ..... ~ THEOR. /-'/<br />

O'/<br />

oJ"<br />

o<br />

n<br />

k.-<br />

if)<br />

O<br />

212<br />

tL<br />

O<br />

o<br />

Z<br />

9<br />

4s<br />

2c<br />

/<br />

~<br />

..ff7<br />

I I L i i i<br />

0 50 73 93 I10 13C, ~5~<br />

No. OF EGGS LAID (.n)<br />

Fig. 9. An observed relationship (solid line with black dots) between<br />

the total number <strong>of</strong> hosts parasitized (z) <strong>and</strong> the total number <strong>of</strong><br />

eggs laid (n) in the~first serie~<strong>of</strong> Monte Carlo simulation <strong>of</strong> the<br />

<strong>parasitism</strong> model <strong>of</strong> continuous search. The broken line with open<br />

circles is a theoretical relationship expected from the binomial distribution9


52<br />

in Fig. 9, in which the total number <strong>of</strong> hosts parasitized, z, is plotted on the vertical<br />

axis against the total number <strong>of</strong> eggs laid, n, on the horizontal axis. It should be<br />

noticed, however, that this method <strong>of</strong> presentation is comparable to that in Fig. 8<br />

<strong>for</strong> the <strong>predation</strong> model, showing the effect <strong>of</strong> diminishing return3. Here again, a<br />

wavy tendency is discernible, which shows that the occurrence <strong>of</strong> super<strong>parasitism</strong> is<br />

periodically increased in accordance with changes in the number <strong>of</strong> eggs laid.<br />

One important difference between the <strong>predation</strong> model <strong>and</strong> the present <strong>parasitism</strong><br />

one is that the number <strong>of</strong> hosts freshly found hardly increased after 130 eggs had<br />

been laid in this example. This is because the parasite could not get out <strong>of</strong> the area<br />

already searched because no host individuals were removed from the area <strong>and</strong> the<br />

parasite repeatedly re-parasitized them. Also, it is noticeable that the observed number<br />

<strong>of</strong> hosts attacked was always lower than that expected in a r<strong>and</strong>om encounter, i.e.<br />

a binomial series (SToY's <strong>for</strong>mula; see w 4g), because <strong>of</strong> excessive super<strong>parasitism</strong>.<br />

Although the indiscriminate parasite in this model is unable to avoid superparasit-<br />

ism, to continue searching in the area already searched is obviously inefficient. If,<br />

however, the parasite periodically moved to start a new search elsewhere, it would<br />

raise its hunting efficiency, <strong>and</strong> so would be favoured by natural selection. If the<br />

parasite is unable to know whether the area where it starts a new search has already<br />

been searched by itself or by other parasites, the shift <strong>of</strong> hunting area can be only<br />

at r<strong>and</strong>om. Of course, a r<strong>and</strong>om shift <strong>of</strong> hunting area certainly involves the risk <strong>of</strong><br />

hitting an area which has already been searched by itself or by other parasites; but<br />

the parasite at least avoids the disadvantage <strong>of</strong> staying in an area which has just<br />

been searched by itself.<br />

In the second series <strong>of</strong> 'experiments', the parasite left the area after each five<br />

eggs had been laid <strong>and</strong> travelled <strong>for</strong> a certain distance in a r<strong>and</strong>omly determined<br />

direction (Fig. 10 a). Again, a wavy trend is seen, but the observed number <strong>of</strong> hosts<br />

v<br />

: ~- OBS,<br />

80<br />

....... THEOR.<br />

/o".-'" /<br />

__N<br />

nr<br />

ff~ 4O<br />

l.--<br />

ffl<br />

0<br />

212<br />

~ 20<br />

o<br />

6<br />

Z<br />

o 40 ~ 1oo 12o ~o<br />

No. OF EGGS LAiD (n)<br />

Fig. 10a. A result from the second series <strong>of</strong> Monte Carlo simulations<br />

<strong>of</strong> <strong>parasitism</strong> similar to the first series, except that a parasite, after<br />

every five eggs have been laid, moves 7. 5 unit lengths in a r<strong>and</strong>omly<br />

determined direction.


.... -o<br />

~ 300<br />

O<br />

1.1.1<br />

N<br />

I.--<br />

9 0BS.<br />

o- ..... ~ THEOR.<br />

20O<br />

nr"<br />

n<br />

U~<br />

U')<br />

0<br />

"r 100<br />

b-<br />

o<br />

Z<br />

/<br />

I<br />

I<br />

I I i i i i i J<br />

SO0 1000<br />

No. OF EGGS LAID (n)<br />

Fig. 10 b. A relationship between z <strong>and</strong> n actually observed with a parasite<br />

species, Encarsia <strong>for</strong>mosa G^~AN, laying eggs on its host species,<br />

Trialeurodes vaporariorum (W~sTW.), in BURNETT'S experiment. The<br />

solid line with black dots is the observed relationship <strong>and</strong> the broken<br />

line with open circles is the one expected from the binomial distribution<br />

(adapted from BURI~ET1" 1958; table IV).<br />

attacked is much closer to that expected by THOMPSON'S <strong>for</strong>mula <strong>for</strong> a r<strong>and</strong>om distri-<br />

bution <strong>of</strong> eggs. Thus the efficiency was on the whole raised in the second experi-<br />

ment as compared with the first; it should be noticed that <strong>for</strong> the lower numbers <strong>of</strong><br />

eggs laid, less super<strong>parasitism</strong> occurred than expected in a r<strong>and</strong>om distribution.<br />

Only two published sets <strong>of</strong> data are available to compare with my simulation<br />

model; in other published data, changes in the distribution pattern were not observed<br />

in accordance with the number <strong>of</strong> eggs laid. The first is BURNETT'S (1958) <strong>study</strong> <strong>of</strong><br />

the distribution <strong>of</strong> the eggs laid by Encarsia <strong>for</strong>mosa GAHAN on Trialeurodes vapora-<br />

riorum (WESTW.), in which periodic deviations from a r<strong>and</strong>om distribution in accord-<br />

ance with the number <strong>of</strong> parasites searching is clearly shown.<br />

For <strong>comparative</strong><br />

purposes, the number <strong>of</strong> hosts attacked is plotted against the number <strong>of</strong> eggs laid<br />

by all parasites (Fig. 10b, adapted from BURNETT'S table IV). A striking similarity,<br />

in the way that the observed curve deviates from the expected, is immediately apparent<br />

in Fig. 10a<strong>and</strong> b.<br />

BURNETT (Op. cit.) also showed a periodic deviation from a r<strong>and</strong>om distribution<br />

in accordance with changes in host density, rather than parasite density. Although I<br />

did not attempt to make any simulation experiment to compare with this experiment<br />

by BURNETT, the result may be deduced from Fig. 10a <strong>and</strong> b. Obviously, the curves<br />

shown in Fig. 10a <strong>and</strong> b are cross-sections <strong>of</strong> the hunting surface (as in Fig. 3a)<br />

parallel to the z-Yt plane, whereas the figure <strong>for</strong> BURNETT'S second experiment is<br />

a cross-section parallel to the z-xo plane. Now, the periodicity <strong>of</strong> waves (or wave<br />

length) is likely to be subject to change according to the initial prey density x0.<br />

That is, it is likely that the wave length becomes longer as x0 increases, since the


54<br />

effect <strong>of</strong> local depletion <strong>of</strong> prey upon the hunting efficiency will be less at higher<br />

values <strong>of</strong> x0. If a cross-section parallel to the z-xo<br />

circumstances, it should again be wavy.<br />

plane is observed under these<br />

A similar tendency to periodic deviation from a r<strong>and</strong>om distribution in accordance<br />

with changes in the host density was shown by SIMMONDS (1943, see also citation by<br />

WILLIAMS 1964). SIMMONDS, <strong>and</strong><br />

odic deviation in relation to the<br />

SIMMONDS' experiments, however,<br />

WILLIAMS as well, presented <strong>and</strong> analysed the peri-<br />

parasite/host ratio, rather than to the density. In<br />

parasite density was kept constant <strong>and</strong> host density<br />

changed <strong>for</strong> ratios between 1/200 <strong>and</strong> 1/25, but <strong>for</strong> those between 2/25 <strong>and</strong> 10/25 the<br />

parasite density was changed <strong>and</strong> the host density kept constant.<br />

There<strong>for</strong>e, the<br />

<strong>for</strong>mer corresponds to a cross-section parallel to the z-xo plane in the hunting surface,<br />

but the latter is a cross-section parallel to the z-Yt<br />

parable even if the ratio declines continually.<br />

plane, <strong>and</strong> so they are not com-<br />

Keeping this point in mind, one can<br />

compare my simulation model <strong>and</strong> SIMMONDS' observation, <strong>and</strong> find again a close simi-<br />

larity between them. It should be borne in mind that from the st<strong>and</strong>point <strong>of</strong> my<br />

simulation model, it is not the parasite/host ratio which is essential: it is the geo-<br />

metric properties <strong>of</strong> parasites' searching activity, changing as the densities change,<br />

that results in the pattern described above.<br />

Now, a question is posed as to whether the wavy trend in the hunting surface is<br />

inherent in <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>. As already shown, a hunting curve comes closer<br />

to the one expected in the r<strong>and</strong>om encounters between prey <strong>and</strong> predators, <strong>and</strong><br />

between hosts <strong>and</strong> parasites, when the predators, or parasites, shift their hunting area<br />

frequently.<br />

Of course, an approach to the curve expected in r<strong>and</strong>om encounters<br />

suggests a rise in the average hunting efficiency. This is, however, just an apparent<br />

relationship, because the time needed to h<strong>and</strong>le a victim has been excluded in the<br />

above <strong>models</strong>. If, in reality, a hunter shifts its hunting area much too <strong>of</strong>ten, the time<br />

involved in travelling from place to place will also increase. Consequently, the advan-<br />

tage <strong>of</strong> shifting will eventually be cancelled by the disadvantage.<br />

Hence, there must<br />

be an optimal frequency <strong>of</strong> shifts <strong>and</strong> an optimal distance (average) <strong>of</strong> travel that<br />

result in the highest hunting efficiency. There<strong>for</strong>e, a perfectly smooth hunting surface<br />

would not at any rate be expected.<br />

If, however, the prey or host individuals also<br />

move around independently <strong>of</strong> each other <strong>and</strong> <strong>of</strong> the predators or parasites, then<br />

r<strong>and</strong>om encounters might again be expected.<br />

It should be mentioned that the wavy pattern <strong>of</strong> a hunting curve (i. e. z plotted<br />

against n) is caused entirely by uninterrupted searching in the first simulation experi-<br />

ment, i.e. the predator's path is continuously increased with time. If the experi-<br />

ment was designed, however, so that the paths <strong>of</strong> some predators consisted <strong>of</strong> a<br />

number <strong>of</strong> short ones, as in the second simulation experiment, the wavy trend will<br />

be less pronounced. If the predator's path was completely discontinuous, that is, if n<br />

was varied entirely by Y, <strong>and</strong> t was extremely short, the waves would eventually<br />

disappear. However, as long as the path <strong>of</strong> each predator is allowed to be continuous


55<br />

<strong>for</strong> a sufficiently long period <strong>of</strong> time, as in experiments on insect parasites cited in<br />

this section, the effect <strong>of</strong> the uninterrupted searching will not be eliminated entirely,<br />

<strong>and</strong> hence the wavy trend will result.<br />

Finally, it should be mentioned that my simulation <strong>models</strong> are not a postulational<br />

hypothesis but suggest some points to be borne in mind in order to make observa-<br />

tions systematic. A problem is left unconsidered as to what would be expected when<br />

the distribution <strong>of</strong> prey or hosts is non-r<strong>and</strong>om. However, the assumption <strong>of</strong> a non-<br />

r<strong>and</strong>om distribution <strong>of</strong> the prey or hosts further requires the consideration <strong>of</strong> non-<br />

r<strong>and</strong>om searching by predators or parasites. As countless non-r<strong>and</strong>om distributions<br />

or non-r<strong>and</strong>om searching patterns are conceivable from a theoretical point <strong>of</strong> view,<br />

it is impractical to try every combination <strong>of</strong> them.<br />

The one way to tackle the<br />

problem is to assume that natural selection will favour those predators or parasites<br />

which adjust their searching pattern in such a way that the highest efficiency is<br />

achieved. Then our task in model building is to find out the best searching pattern<br />

<strong>and</strong> to compare it with observations. My theoretical <strong>study</strong> <strong>of</strong> this problem will be<br />

published elsewhere, however.<br />

f). WATT'S equation<br />

WATT (1959) proposed an equation which describes the relationship between the<br />

attacking <strong>and</strong> attacked species <strong>and</strong> which has been admired as one <strong>of</strong> the most precise<br />

<strong>and</strong> complete equations proposed to date (HoLLING 1966). The equation was fitted<br />

to some observed data <strong>for</strong> various parasites <strong>and</strong> predators in order to demonstrate<br />

its high descriptive power.<br />

WATT'S equation, however, is most difficult to comprehend, firstly because the<br />

definition <strong>of</strong> some notations is not clear, secondly because some assumptions are<br />

incomprehensible, <strong>and</strong> thirdly because there seem to be some errors in mathematical<br />

operation.<br />

In order to show the above points, WATT'S own presentation (WATT 1959, p. 133)<br />

will be quoted first <strong>and</strong> will later be compared with my general <strong>for</strong>mulae shown in<br />

w 3. The following quotation is verbatim except <strong>for</strong> equation numbers, the omission<br />

<strong>of</strong> one unnecessary equation, <strong>and</strong> changes in three symbols, i.e. g->r, a->a, <strong>and</strong> bo B.<br />

"Definition <strong>of</strong> Symbols:<br />

Na the number attacked<br />

No the initial number <strong>of</strong> hosts or prey vulnerable to attack<br />

P<br />

A<br />

K<br />

the number <strong>of</strong> parasites or predators actually searching<br />

coefficient <strong>of</strong> attack, the Na per P (an instantaneous rate)<br />

the maximum number <strong>of</strong> attacks that can be made per P during<br />

the period the No are vulnerable.<br />

"Since we seek an integral equation <strong>of</strong> <strong>for</strong>m Na :f(No, P), this could be obtained<br />

from a partial differential equation <strong>of</strong> type<br />

ON.~/ONo=r~(No, P) (4f. 1)


56<br />

or one such as<br />

ON~/OP=rz(No, P) (4f. 2).<br />

"The reason <strong>for</strong> starting with (4f. 1) rather than (4f. 2) was that the structure<br />

(4f. 1) should have was more intuitively obvious. In general, there will probably be<br />

more than one road to a complex integral equation, <strong>and</strong> a large part <strong>of</strong> the problem<br />

<strong>of</strong> obtaining it has been solved when we have ascertained the easiest route.<br />

"Equation (4f. 3) states that all parasites or predators can generate a total <strong>of</strong> PK<br />

attacks, <strong>and</strong> ON.#ONo diminishes gradually as NA approaches this maximum.<br />

ON~/ONo = PA (PK- N.4) (4f. 3)<br />

"However, the larger A is, the greater dA/dP will be [For this statement to be<br />

consistent with eq. (4f. 4) below, it must be ]dA/dPL rather than dA/dP: TANAKA<br />

in lit.I, because it will be more difficult <strong>for</strong> parasites or predators to find unattacked<br />

host or prey. At the same time, dA/dP [This must be [dA/dP[ again] must decrease<br />

in inverse ratio to P, because the greater P is, the greater inter-attack competition<br />

will be. This competition might take any <strong>for</strong>m from active interference to superpara-<br />

sitism. The above ideas are expressed in the equation<br />

dA/dP= - [~A/P (4f. 4)<br />

or<br />

A =trP-~ (4f. 5)<br />

where ~ <strong>and</strong> /~ are positive constants. Substituting (4f. 5) in (4f. 3) we get<br />

ON.~/ONo = PerP-~ (PK- N.4) (4f. 6)<br />

<strong>and</strong> integrating,<br />

NA=PK(1-e aNoP ) (4f. 7)."<br />

This is the first part <strong>of</strong> WATT'S modelling process, <strong>and</strong> some statements need to<br />

be reinterpreted <strong>for</strong> the following reason.<br />

First, the definition <strong>of</strong> Na is insufficient, since it does not specify whether the<br />

density <strong>of</strong> the attacked species is (1) subject to reduction as it is attacked, or (2)<br />

kept constant. If it is (1), Na is equivalent to my notation z, but if (2), N~ must be<br />

equivalent to n.<br />

Although WATT was not specific in this respect, it seems certain<br />

that he was aiming at the evaluation <strong>of</strong> z rather than n, since he stated, in the<br />

sentence following eq. (4f. 3), "because it wilt be more difficult <strong>for</strong> parasites or pred-<br />

ators to find unattacked hosts or prey", <strong>and</strong> also because his equation was fitted to<br />

data in which the reduction in the attacked individuals was unmistakable.<br />

Let us there<strong>for</strong>e assume (1), from which it automatically follows that No <strong>and</strong> P<br />

are equivalent to x0 <strong>and</strong> Y, respectively, in my notation.<br />

Also it is clear that the<br />

model is concerned with animals with discrete generations, because no account is<br />

taken <strong>of</strong> any changes in the density <strong>of</strong> the attacking species during the period the<br />

attacked species is vulnerable. Thus, WATT'S eq. (4f. 7) must be what I called an<br />

overall hunting equation, i.e. z=F(xo, Y, t), in which t is not considered as a vari-<br />

able. In fact, the factor t in WATT'S equation is concealed in factor K <strong>and</strong>, hence,<br />

is treated as constant, as will be shown later. Thus, NA=f(No, P) must be equivalent


57<br />

to z=F (xo, Y, t). Also one must remember that z should not under any circum-<br />

stances exceed x0 as no animal can eat more than is supplied.<br />

This is synonymous<br />

with saying that NA should not exceed No no matter how large P is. However, the<br />

calculation <strong>of</strong> N,,=lim f(No, P) as in eq. (4f. 7) shows the following results:<br />

p~<br />

when 1) 0~2, NA-~c~,<br />

2) ~=2, Na-~KNo, <strong>and</strong><br />

3) ~2, NA-~O,<br />

The first case is <strong>of</strong> course contradictory, <strong>and</strong> there<strong>for</strong>e the coefficicnt t~ should not<br />

be smaller than 2. However, WATT did not give any such restriction to /~. Now, if<br />

we assume that no social interference between individuals <strong>of</strong> the attacking species is<br />

involved, then theoretically all the attacked individuals must eventually be wiped out<br />

if P increases indefinitely, i.e. NA must tend to No under these circumstances. The<br />

only possible instance in which NA can tend to No is the second case in which /~=2.<br />

However, since NA-~ozKNo there, ezK ought to be unity to be consistent with the<br />

above condition, i.e. NA-)No. In the meantime, ez <strong>and</strong> K are defined by WATT as<br />

constants.<br />

Then no matter what situation is considered (i. e. whether or not the<br />

system under consideration involves social interaction), the relationship ezK=I should<br />

hold. So that eq. (4f. 5) becomes A = 1/KP~ which should hold <strong>for</strong> all real <strong>and</strong> positive<br />

values <strong>of</strong> P. It follows that the smaller P is, the larger will A be without any limit.<br />

On the other h<strong>and</strong>, the coefficient A is defined by WATT as "the NA per P (an<br />

instantaneous rate)". So, it has to be concluded that N• per P-~c~ when P becomes<br />

infinitesimally small. This <strong>of</strong> course contradicts the assumption that each individual<br />

attacker has a capacity limited by K.<br />

In the third case, where /~2,<br />

NA-~0, suggesting that if social interference is<br />

involved, no attack could be made when the attacking species is crowded. Although<br />

this suggestion sounds reasonable, it is only superficial, because it does not remove<br />

the contradiction with respect to A <strong>and</strong> K as pointed out above.<br />

In subsequent<br />

paragraphs, it will be shown that WATT made serious mistakes in mathematics which<br />

reflect a certain misconception in his model-building approach, <strong>and</strong> that what is re-<br />

presented by the factor ~ is an artifact.<br />

Clearly, the partial derivatives ~NJONo <strong>and</strong> ON.~/OP are equivalent to Oz/Oxo <strong>and</strong><br />

Oz/OY, respectively, in my notations. The first is the tangent <strong>of</strong> cross-sections <strong>of</strong> the<br />

hunting surface f(No, P), equivalent to F(x0, Y) <strong>of</strong> my notation, parallel to the N•<br />

-No (or Z-Xo) plane; <strong>and</strong> the second is the tangent <strong>of</strong> cross-sections parallel to the<br />

N.~-P (or z-Y) plane. Hence, the functions rl <strong>and</strong> r2 should satisfy the following<br />

three conditions.<br />

No <strong>and</strong> P respectively, we have<br />

(1), a mathematical condition : integrating rl <strong>and</strong> r2 with respect to<br />

<strong>and</strong><br />

f lvo rldN~=NA+c~<br />

f p r2dP=N.4+c2


58<br />

where ci <strong>and</strong> ce are integral constants. Hence, the following relationship should hold:<br />

f ~f,dN~-cl= f<br />

r~dP-c~.<br />

That is, the functions rl <strong>and</strong> r2 are not independent <strong>of</strong> each other.<br />

(2), an ecological condition that the function rl should reflect: since rl is a partial<br />

derivative ON~/ONo, it should reflect the limited capacity <strong>of</strong> each predator according<br />

to its maximum number <strong>of</strong> attacks, i.e.K.<br />

(3), an ecological condition that the function r2 should reflect: since r2 is a partial<br />

derivative ONA/~P, it should reflect the effect <strong>of</strong> both types <strong>of</strong> competition, i.e. the<br />

effect <strong>of</strong> diminishing returns <strong>and</strong> the effect <strong>of</strong> social interference.<br />

In the derivation <strong>of</strong> eq. (4f. 7), WATT evaluated the factor A arbitrarily, neg-<br />

lecting the first condition <strong>and</strong> part <strong>of</strong> the third condition, i.e. the effect <strong>of</strong> diminishing<br />

returns.<br />

Although it seems as though WATT was considering the effect <strong>of</strong> compe-<br />

tition, i.e. condition (3), his consideration in terms <strong>of</strong> the factor A was concerned<br />

only with the effect <strong>of</strong> social interference. This is obvious because eq. (4f. 4) does<br />

not involve No, whereas the effect <strong>of</strong> diminishing returns should be a function <strong>of</strong><br />

No. The neglect <strong>of</strong> the effect <strong>of</strong> diminishing returns in WATT'8 mathematics, <strong>and</strong><br />

hence the failure to determine the function r2 in relation to r~, resulted in the contra-<br />

diction when eq. (4f. 7) was regarded as an overall hunting equation.<br />

My alternative interpretation <strong>of</strong> eq. (4f. 7) is there<strong>for</strong>e that it is an instantaneous<br />

hunting equation equivalent to eq. (3. 1); i.e. Na is not z, but n. It automatically<br />

follows that No is X, <strong>and</strong> this means that the density <strong>of</strong> the attacked species is kept<br />

constant during the attack period. Then P is Y. Now, K was defined by WATT as<br />

'the maximum number <strong>of</strong> attacks made per P during an attack period' (though the<br />

expression 'per P' is not clear, this is perhaps 'per individual predator'). Then K<br />

corresponds to the expression bt used in my toss-a-ring experiment, i.e. b was the<br />

frequency <strong>of</strong> tosses per unit time so that the maximum possible number <strong>of</strong> attacks<br />

<strong>for</strong> time t per individual ring was bt (see w 4d).<br />

mine as above, except A, eq. (4f. 3) is rewritten as<br />

On/OX= A Y(Ybt - n)<br />

So, changing WATT'S notations to<br />

<strong>and</strong> eliminating A from the above equation, using the relationship A--aY-a (cf. eq.<br />

(4f. 5) in which P~ Y, <strong>and</strong> a-=a), we have<br />

n=b(1-e -aYI-pX) Yt (4f. 8).<br />

If we set ~=1, then eq. (4f. 8) becomes<br />

n=b(1-e -~) Yt,<br />

<strong>and</strong> this equation is identical to eq. (4d. 3), i.e. IVLEV'S equation with the addition<br />

<strong>of</strong> the attack period t <strong>and</strong> the density, Y, <strong>of</strong> the attacking species.<br />

While WATT stated that the structure <strong>of</strong> eq. (4f. 3) was intuitively obvious, it<br />

appears that intuition was a poor guide in this case, <strong>and</strong> the structure has become<br />

intelligible in the light <strong>of</strong> the toss-a-ring experiment.<br />

That is, eq. (4f. 7), which is<br />

equivalent to eq. (4f. 8), represents a generalized instantaneous equation <strong>of</strong> the toss-


59<br />

a-ring model, m winch the area <strong>of</strong> the ring (t. e. a=a) diminishes, or increases, by<br />

the factor Y:-P, as the number <strong>of</strong> rings per unit area, i.e. Y, increases. There<strong>for</strong>e<br />

the equation is considered to imitate the effect <strong>of</strong> social interaction incorporated into<br />

the effective area <strong>of</strong> each attacking individual; the effective area is now aY a-~ rather<br />

than simply a.<br />

Clearly, (1) if 0


60<br />

diminishing returns (e. g. super<strong>parasitism</strong>). However, eq. (4f. 7)could behave as<br />

though the effect <strong>of</strong> diminishing returns was taken into account. This is because <strong>for</strong> a<br />

certain range <strong>of</strong> B (i. e. in the vicinity <strong>of</strong> 2) the cross-section <strong>of</strong> the hunting surface<br />

parallel to the NA-P plane (i. e. the z-Y plane in my notation) resembles, though<br />

only superficially, the effect <strong>of</strong> diminishing returns. There<strong>for</strong>e, if the equation is fitted<br />

to data in which the density <strong>of</strong> available attacked individuals is reduced, the estimate<br />

<strong>of</strong> 0 will inevitably be close to 2.<br />

Such estimates <strong>of</strong> 0 are artifacts which, while<br />

they avoid the logical contradiction that is inevitably encountered when N• applies<br />

to a diminishing population, reflect no logical interpretation with respect to social<br />

interactions. These clearly illustrate that fitting a curve is not in itself a verification<br />

<strong>of</strong> the model concerned. It is only necessary to add that case c2~ under C2 in w 2<br />

actually happened.<br />

In the second half <strong>of</strong> WATT's modelling process, it was assumed that the rela-<br />

tionship between the number <strong>of</strong> attacking individuals, P~, in one generation <strong>and</strong> that<br />

<strong>of</strong> the next generation, P~, could be derived from the following differential equation<br />

dP~/dNo~,: -cP~<br />

<strong>and</strong> this, according to WATT, yields<br />

-cNop<br />

P. =P2,e<br />

where No. is the initial density <strong>of</strong> the attacked species in the first generation, <strong>and</strong> c<br />

is a proportionality factor. The equation is totally incomprehensible, <strong>and</strong> I do not<br />

quite underst<strong>and</strong> why the attacking species manages to rear as many <strong>of</strong>fspring as<br />

the parents when food is practically unobtainable, i.e.<br />

when No, approaches zero.<br />

Because all the variables involved in the above equation should be able to take any<br />

positive real values, it should hold when No, tends to zero, in which case P~ tends to<br />

P..<br />

When he was working on gravitation, ISAAC NEWTON invented calculus, by which<br />

he managed to reduce the scattered distribution <strong>of</strong> the gravitational <strong>for</strong>ce on a solids<br />

sizable body into a sizeless point called the centre <strong>of</strong> gravity. The principle is to<br />

start from a simpler <strong>and</strong> self-explanatory situation <strong>and</strong> to draw a conclusion that is<br />

not intuitively obvious. As already mentioned in w 3, the process <strong>of</strong> <strong>predation</strong> is <strong>of</strong>ten<br />

self-explanatory only when an instantaneous situation is assumed, <strong>and</strong> this is why we<br />

<strong>of</strong>ten start from a differential equation.<br />

It may well happen, too, that an empirical<br />

equation is first found to describe an observed trend, then a differential equation is<br />

derived from this empirical one to see if it indicates something immediately intel-<br />

ligible. The differential equations used by WATT fit into neither <strong>of</strong> the two cases. Even<br />

though WATT claimed that eq. (4f. 3) was obvious, it turned out to be due to his<br />

illusion. The same criticism applies to IVLEV (see w 4d).<br />

If a differential equation<br />

does not suggest anything obvious, why do we have to start from it ? Clearly, the<br />

differential equations used by WATT <strong>and</strong> IVLEV have nothing to do with inferences;<br />

the same applies to the array <strong>of</strong> differential equations listed by WATT (1961, figure


61<br />

7; or 1968, figure ii. 2). It is a misuse <strong>of</strong> mathematical language as a <strong>for</strong>mal system<br />

<strong>of</strong> inferences, <strong>and</strong> its consequence is now obvious.<br />

Finally, model oscillations in the densities <strong>of</strong> an attacking <strong>and</strong> attacked species<br />

calculated by MILLER (1960) <strong>and</strong> IT5 (1963) using WATT'S equation must be men-<br />

tioned.<br />

Both authors found stable oscillations rather than <strong>of</strong> a relaxation type as<br />

expected in the NICHOLSON-BAILEV model. While the difficulty <strong>of</strong> the latter model<br />

was thus removed, this was done at the expense <strong>of</strong> logical meaning. First, because<br />

WATT'S equation is an instantaneous one, there is no way to estimate the final density<br />

<strong>of</strong> the attacked population. There<strong>for</strong>e in these examples the final density had to be<br />

calculated by subtracting the number eaten from the initial density, i.e. No-NA<br />

(=X-n in my notation). This was <strong>of</strong> course logically incorrect, but a numerical<br />

value was obtained anyway. In fact, as pointed out in the early part <strong>of</strong> w 3, n is<br />

always larger than z, <strong>and</strong> so the calculated value <strong>of</strong> X-n is smaller than xo-z. In<br />

other words, the estimates <strong>of</strong> the final density by MILLER <strong>and</strong> by IT3 must have<br />

been lower than they should actually be.<br />

Secondly, however, the factor /~ in the<br />

vicinity <strong>of</strong> 2 sets the upper limit <strong>of</strong> the number attacked when the density <strong>of</strong> the<br />

attacking species increases, <strong>and</strong> in conjunction with factor K <strong>and</strong> a in the range <strong>of</strong><br />

not greater than ozK=l, the upper limit <strong>of</strong> the number attacked could be set much<br />

lower than it could in the NICnOLSON-BAILEY model. In other words, with ~-~2 <strong>and</strong><br />

aK_


62<br />

hunting t, <strong>and</strong> z the total number <strong>of</strong> hosts parasitized per unit area by the end <strong>of</strong><br />

that period. If encounters between parasites <strong>and</strong> hosts are at r<strong>and</strong>om (<strong>for</strong> the defini-<br />

tion <strong>of</strong> r<strong>and</strong>om encounter, see w 4e), a further increment <strong>of</strong> the number <strong>of</strong> eggs laid,<br />

i.e. An, will be distributed with equal probability among the unparasitized <strong>and</strong> the<br />

already parasitized hosts (this statement is not strictly true; see later). Then, letting<br />

Az be an increment in the number <strong>of</strong> hosts parasitized, the ratio Az/Jn must be<br />

proportional to the proportion <strong>of</strong> the hosts unparasitized, i.e.<br />

,~z/,~n = ( X- z) IX<br />

<strong>and</strong> thus <strong>for</strong> An-~O, the following differential equation is obtained,<br />

dz/dn = (X-z)/X (4g. 1)<br />

<strong>and</strong> integrating, we get<br />

z = X(1 - e -~/x) (4g. 2).<br />

This is THOMPSON'S equation. Note that the meaning <strong>of</strong> the differential equation in<br />

eq. (4g. 1) is entirely different from that in eq. (3. 5) which represents a <strong>predation</strong><br />

process.<br />

There<strong>for</strong>e the derivation <strong>of</strong> a parasite model by means <strong>of</strong> a differential<br />

equation as above may be justified, but this is not inconsistent with my earlier state-<br />

ment in w 3.<br />

If eq. (4g. 2) is compared to eq. (3. 20), it will be found that the probability <strong>of</strong><br />

a host individual receiving no parasite egg, i.e. Pr{0}, is equal to the expression<br />

e -'/z which is the 0-term <strong>of</strong> a PoIssON series. So it becomes clear that THOMPSON'S<br />

statement (in italics above) is not quite rigorous but requires an additional condi-<br />

tion: the statement is correct if the number <strong>of</strong> hosts is sufficiently large so that the<br />

probability <strong>of</strong> a given host individual being found by each parasite individual is<br />

sufficiently small.<br />

In a laboratory experiment, however, the animals are <strong>of</strong>ten confined to a small<br />

cage, <strong>and</strong>, unless the density <strong>of</strong> hosts is sufficiently large, the condition which could<br />

ensure the PoIssoN distribution is not satisfied. So, let us look at the problem more<br />

closely from the probabilistic point <strong>of</strong> view. As already mentioned, the precise ex-<br />

pression <strong>of</strong> 'r<strong>and</strong>om encounters' is that each host individual in the area concerned has<br />

an equal probability <strong>of</strong> being parasitized. Let this probability be p <strong>and</strong> the probability<br />

that a given host individual does not receive any egg be q, i.e. p+q=l.<br />

Then, the<br />

frequency distribution <strong>of</strong> nM eggs over the hosts in area M will be given by the<br />

following bionomial series,<br />

(p+q)'~ =q'~ +nMq~t-lp/1 !+ nM(nM- 1) q'~-2pz/2 ! + ...... +p~M<br />

where q~ is the proportion <strong>of</strong> hosts unparasitized, so that 1-q "~ is the proportion<br />

<strong>of</strong> the hosts parasitized.<br />

If the density <strong>of</strong> hosts is X, the probability <strong>of</strong> each host<br />

being parasitized (i. e. p) is equally 1/MX, <strong>and</strong> as q=l-p, we have<br />

q~ = (1 - 1/MX) "~.<br />

Also, the total number <strong>of</strong> hosts parasitized per area M (i. e. zM) is the proportion<br />

<strong>of</strong> hosts parasitized (i. e. 1-q "~) multiplied by the total number <strong>of</strong> hosts in area M<br />

(i. e. MX), thus


<strong>and</strong> so<br />

zM=MX {1- (1-1/MX) n~}<br />

z =X {1 - (1-1/MX) ~'~} (4g. 3).<br />

This is SwoY's model <strong>for</strong> r<strong>and</strong>om <strong>and</strong> indiscriminate <strong>parasitism</strong> (appendix to SALT<br />

1932), <strong>and</strong> holds <strong>for</strong> all values <strong>of</strong> p; that is, STOY'S equation is more general than<br />

THOMPSON'S.<br />

In order to compare THOMPSON'S model with STOY'S, let us calculate the deriva-<br />

tive dz/dn in eq. (4g. 3) which is differentiable <strong>for</strong> all real values <strong>of</strong> MX larger than<br />

1, <strong>and</strong> thus<br />

dZ (MX ln MX ) X-z<br />

dn MX- 1 X (4g. 4).<br />

If eqs. (4g. 4) <strong>and</strong> (4g. 1) are compared, it is at on ceclear that THOMPSON's differential<br />

equation is a special case <strong>of</strong> eq. (4g. 4) in which the expression in the brackets tends<br />

to 1. This occurs only when MX tends to infinity, i.e.<br />

lim MX ln{MX/(MX-1)} =1<br />

MX~oo<br />

<strong>and</strong> MX~c~ means that p=I/MX->O. This is <strong>of</strong> course the well-known relationship<br />

between the POISSON <strong>and</strong> bionomial distributions; i.e. the PolssoN distribution is a<br />

special case <strong>of</strong> the binomial distribution in which p is sufficiently small.<br />

The above analysis illustrates that although THOMPSON'S reasoning, as it appeared<br />

in his differential eq. (4g. 1) under the assumption <strong>of</strong> r<strong>and</strong>om encounters, was appar-<br />

ently reasonable, it was in fact not sufficiently precise because it is not obvious that<br />

the equation requires the condition p~0. Here again a differential equation was used<br />

rather uncritically; it should have been noticed that calculus was not quite an appro-<br />

priate method <strong>of</strong> reasoning in a <strong>parasitism</strong> model.<br />

The above argument, however, excludes the possibility that THOMPSON'S eq.<br />

(4g. 1) might be more appropriate than STOY'S if, paradoxically, r<strong>and</strong>om encounters<br />

are not assumed. That is to say, there is a possibility, though not demonstrated here,<br />

that some non-r<strong>and</strong>om encounters might again satisfy the condition expressed in eq.<br />

(4g. 1). One such example is given in my simulation model <strong>for</strong> <strong>parasitism</strong> in w 4e, in<br />

which the observed frequency was fairly close to that expected in a PoIssoN series,<br />

while the underlying mechanism is clearly <strong>of</strong> a non-PoIssoN type. TORII (1956) has<br />

already pointed out that an agreement between the observed <strong>and</strong> expected frequencies<br />

alone would not imply that the same mechanism is involved, as it is possible that<br />

different mechanisms could yield an almost identical frequency distribution.<br />

suggests that an agreement between the observed <strong>and</strong> the expected in THOMPSON'S<br />

equation could be entirely irrelevant to the test <strong>of</strong> the hypothesis <strong>of</strong> r<strong>and</strong>om encoun-<br />

ters. Conversely, it may be suggested that the assumption <strong>of</strong> r<strong>and</strong>om encounters is<br />

not very important.<br />

63<br />

This<br />

In passing, we may note that TORII (1956) also showed that the index <strong>of</strong> disper-<br />

sion, i.e. the variance-mean ratio, which is unity in the PolssoN distribution, could<br />

statistically be less than unity if the binomial series was involved. This is because,


64<br />

in the binomial series, the variance is npq as against the mean which is np, <strong>and</strong> so<br />

the variance-mean ratio is npq/np=q (where n is the number <strong>of</strong> eggs laid, p the<br />

probability <strong>of</strong> a given host receiving one egg, <strong>and</strong> q=l-p).<br />

small, the variance-mean ratio, which is q=l-p,<br />

If p is not sufficiently<br />

must be smaller than unity. This<br />

refutes the common belief that if the ratio is less than unity the parasites concerned<br />

are <strong>of</strong> the discriminate type. The interpretation <strong>of</strong> frequency distribution <strong>of</strong> parasite<br />

progeny on hosts should there<strong>for</strong>e be made with the utmost caution.<br />

Another relevant point here is the use <strong>of</strong> the negative binomial series developed<br />

by BLISS <strong>and</strong> FISHER (1953). The expectation <strong>of</strong> the 0-term (the proportion <strong>of</strong> hosts<br />

receiving no parasite eggs) in the negative binomial series is given by<br />

Pr{0} = (1 +m/k) --k (4g. 5)<br />

where m is the mean number <strong>of</strong> eggs per host, which in my notation is n/X. Sub-<br />

stituting n/X <strong>for</strong> m in eq. (4g. 5) we have<br />

Pr {0} = (1+ n/kX) -k (4g. 6).<br />

Here k is a positive constant, <strong>and</strong> if k-~oo the negative binomial distribution coincides<br />

with the PoISsON distribution, <strong>and</strong> if k-~0, it becomes the logarithmic series (BLISS<br />

<strong>and</strong> FISHER 1953). When we choose an appropriate value <strong>of</strong> k, the negative binomial<br />

series describes various types <strong>of</strong> non-r<strong>and</strong>om distribution (excluding an even distribu-<br />

tion). Substituting the right-h<strong>and</strong> side <strong>of</strong> eq. (4g. 6) <strong>for</strong> Pr{O} in eq. (3.20) we have<br />

z =X {1 - (1 +n/kX) - k} (4g. 7).<br />

GRIFFITHS <strong>and</strong> HOLLING (1969) proposed the use <strong>of</strong> eq. (4g. 7) <strong>for</strong> <strong>parasitism</strong> in<br />

which the parasite egg distribution is not r<strong>and</strong>om (or, more precisely, the variance-<br />

mean ratio is larger than unity).<br />

The negative binomial series, however, does not<br />

distinguish the type <strong>of</strong> underlying mechanisms involved, <strong>and</strong> so it is purely a descrip-<br />

tive <strong>for</strong>mula. Although the negative binomial distribution is identical with what is<br />

known as the POLYA-EGGENBERGER distribution (IT(5 1963) which has a specific model<br />

structure, I found it difficult to relate this model structure to the process <strong>of</strong> para-<br />

sitism. In passing, although GRIFFITHS <strong>and</strong> HOLLING (lOt, cit.) suggested the use <strong>of</strong><br />

eq. (4g. 7) also <strong>for</strong> <strong>predation</strong>, it is not legitimate to do so.<br />

As already discussed in w 3, <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> <strong>models</strong> will not differ from<br />

each other in the <strong>for</strong>m <strong>of</strong> the instantaneous hunting function, i.e. eq. (3. 1) holds<br />

<strong>for</strong> both cases as long as the prey or host density can be considered to be constant.<br />

In <strong>parasitism</strong>, the overall equation in general <strong>for</strong>m is eq. (3.22), in which f(X) can<br />

be anything as long as it is not influenced by the pattern <strong>of</strong> encounters between<br />

parasites <strong>and</strong> hosts. There<strong>for</strong>e, THOMPSON did not have to assume a particular <strong>for</strong>m<br />

<strong>for</strong> f(X) but only needed to say that n was the number <strong>of</strong> eggs laid.<br />

It does not<br />

matter whether n is an observed value or an assumed function <strong>of</strong> X, as long as one<br />

can justify the assumption <strong>of</strong> the POISSON or similar distribution. Consequently, if<br />

f(X) Yt in eq. (3. 1) is used as the theoretical expectation <strong>of</strong> n, the following simul-<br />

taneous equations will hold,<br />

,~n =f(X) Y, Jt


Az/An = (X-z)/X<br />

from which we get<br />

dz/dt = (X- z)f(X) Y/X<br />

<strong>and</strong> integrating we have<br />

z =X(1- e -f(I) r~/I) (4g. 8).<br />

Clearly, HOLLING'S introduction <strong>of</strong> the factor h, or IVLEV'S equation justified in a<br />

toss-a-ring model, does not influence the assumption <strong>of</strong> Po~ssoN-type encounters, <strong>and</strong><br />

so eqs. (4c. 10) or (3.24) as specific <strong>for</strong>ms <strong>of</strong> eq. (4g. 8) are obtained <strong>for</strong> these two<br />

cases respectively (n in STOY'S equation can <strong>for</strong> the same reason be replaced by<br />

f (x) Yt).<br />

In the NICHOLSON-BAILEY <strong>predation</strong> model, however, it is crucial to assume a<br />

particular type <strong>of</strong> f(X), because the evaluation <strong>of</strong> the overall hunting equation is<br />

influenced by f(X), even if encounters are made at r<strong>and</strong>om. As already shown, if<br />

f(X) is a linear function <strong>of</strong> X, the overall equation is coincidentally <strong>of</strong> the same<br />

<strong>for</strong>m as THOMPSON'S, but if f(X) is <strong>of</strong> HOLLING'S type or IVLEV'S, the overall equation<br />

is eq. (4c. 9) or (3. 12), which are quite different from eqs. (4c. 10) <strong>and</strong> (3. 24) respectively.<br />

THOMPSON (1939) argued against NICHOLSON-BAILEY (1935) <strong>and</strong> stated that, while<br />

the NICHOLSON-BAILEY assumption <strong>of</strong> r<strong>and</strong>om searching was not justifiable, the fact<br />

that THOMPSON himself arrived at the same equation "merely illustrates the well-<br />

known fact that identical quantitative relationship may be developed from biologically<br />

different postulates, since these postulates are not, in their ontological significance,<br />

incorporated in the <strong>for</strong>mula". Now it is clear that THOMPSON was mistaken in that<br />

he was comparing incomparables, i.e. <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>, <strong>and</strong> that the resem-<br />

blance does not signify anything. The ontological significance <strong>for</strong> the two postulates<br />

becomes obvious under general circumstances in which f(X) is not a linear function<br />

<strong>of</strong> X.<br />

WATT (1959), in his review <strong>of</strong> various <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> <strong>models</strong>, made<br />

similarly erroneous comments that the NICHOLSON-BAILEY <strong>and</strong> THOMPSON equations<br />

are identical, <strong>and</strong> furthermore, that THOMPSON'S equation should have a constant<br />

factor in front <strong>of</strong> the exponent, to express the efficiency <strong>of</strong> different parasites. The<br />

suggestion is nonsensical because the exponent n/X (in my notation) is just a straight-<br />

<strong>for</strong>ward "mean number <strong>of</strong> parasite eggs per host" laid by all the parasite individ-<br />

uals <strong>for</strong> the entire observation period, <strong>and</strong> the mean number is a mean number no<br />

matter how efficient are the parasites concerned.<br />

plied by a constant factor signify ?<br />

65<br />

What does a mean number multi-<br />

A correct interpretation is as follows. (1) If n<br />

is an observed value, it should be observed under st<strong>and</strong>ard conditions in which the<br />

time <strong>of</strong> observation <strong>and</strong> the densities <strong>of</strong> both host <strong>and</strong> parasite populations, i.e. t, X,<br />

<strong>and</strong> Y respectively, are fixed (a st<strong>and</strong>ard may be determined conveniently); then<br />

differences between values <strong>of</strong> n <strong>for</strong> different parasite species reflect differences in<br />

efficiency between the species. (2) If n is an expected value, i.e. a theoretical expec-<br />

:tation when t, X, <strong>and</strong> Y are known, it should be replaced by f(X) Yt as in eq.


66<br />

(4g. 8). Then f(X) <strong>for</strong> a st<strong>and</strong>ard X is the efficiency <strong>of</strong> the species concerned.<br />

further discussion, see the appendix to w 4i).<br />

It should be pointed out here that, on the whole, the review <strong>of</strong> <strong>models</strong> by WATT<br />

(1959) is invalid, firstly because his mathematics is <strong>of</strong>ten wrong, <strong>and</strong> secondly because<br />

he was confused between<br />

(For<br />

instantaneous <strong>and</strong> overall functions, between <strong>parasitism</strong><br />

<strong>and</strong> <strong>predation</strong>, <strong>and</strong> between the Z-Xo <strong>and</strong> z-Y relationships. It should also be noticed<br />

that the criticism against the assumption f(x)=ax invalidating the NICHOLSON-BAILEY<br />

<strong>predation</strong> equation does not invalidate THOMPSON'S <strong>parasitism</strong> equation, since the latter<br />

does not assume f(x)=ax.<br />

h).<br />

The HASSELL-VARLEY model <strong>of</strong> social interference in parasites<br />

Although this model is called by the authors (HASSELL <strong>and</strong> VARLEY 1969) 'a<br />

new model' based on the NICHOLSON-BAILEY competition equation (see w it is in<br />

fact a special case <strong>of</strong> the generalized THOMPSON'S model <strong>for</strong> indiscriminate parasites,<br />

eq. (4g. 8), in which the instantaneous hunting function is a modified NICHOLSON-<br />

BAILEY linear function. As already pointed out, THOMPSON'S equation <strong>for</strong> <strong>parasitism</strong><br />

takes the same <strong>for</strong>m as the NICHOLSON-BAILEY 'competition equation' <strong>for</strong> <strong>predation</strong><br />

if the instantaneous hunting function f(X) is assumed to be a linear function <strong>of</strong> X,<br />

i.e. f(X) =aX, in which the coefficient a is the 'effective area <strong>of</strong> recognition per<br />

unit time'. Under these circumstances, the value <strong>of</strong> /~, as defined by eq. (4b. 8), becomes<br />

at, a constant.<br />

It has been shown in w 4e, however, that the /~ cannot be constant, but is at least<br />

a function <strong>of</strong> X, host density. HASSELL <strong>and</strong> VARLEY, however, found that in some<br />

published data the value <strong>of</strong> ~ was not independent <strong>of</strong> Y, parasite density. These data<br />

are shown graphically in Fig. 11; this is a reproduction <strong>of</strong> figure 1 in HASSELL <strong>and</strong><br />

VARLEY (1969) with a slightly different arrangement. These data show that the<br />

value <strong>of</strong> In ii tends to decrease as the value <strong>of</strong> In Y increases. The interpretation <strong>of</strong><br />

these relationships by HASSELL <strong>and</strong> VARLEY is that the parasites interfered with each<br />

other more strongly as their density increased, <strong>and</strong> hence the reduction in "the area<br />

<strong>of</strong> discovery", i.e. /f. The authors stated that "the striking feature <strong>of</strong> the relationships<br />

in (Fig. 11) is that they are linear over several orders <strong>of</strong> magnitude", <strong>and</strong> that "the<br />

data <strong>for</strong> Chelonus texanus CRESS. [curve (c)] cover a narrow range <strong>of</strong> parasite densities<br />

but seem to imply a curvilinear relationship".<br />

Thus, it was concluded that the relationships were described by the following<br />

<strong>for</strong>mula<br />

In ii=ln Q-m In Y (4h. 1)<br />

or<br />

= O Y-~ (4h. 1'),<br />

in which the factor Q is called "the quest constant" <strong>and</strong> m "the mutual interference<br />

constant".<br />

If the above relationships are incorporated into the NICHOLSON-BAILEY model


67<br />

0.200<br />

Iol<br />

0,100<br />

0.050<br />

0,020<br />

0,0]0<br />

0,005<br />

:O<br />

(~)0,002' , , , , T ~ I<br />

1.0<br />

W.<br />

0,05( 9<br />

I I I I I i m'f I<br />

10.0<br />

I n I I I I I II I I I<br />

100.0<br />

o.0~<br />

(el<br />

0,OLC<br />

0,00 ~ .<br />

0.002<br />

o.ool i i i i i i I i i i i i i i i I t i i i t i i i i I t<br />

1.o 1o.o IOO.O<br />

No. OF PARASITES PER FT 2<br />

Fig. 11. Observed relationships between the values <strong>of</strong> fi <strong>and</strong> parasite densities,<br />

in natural logarithmic scale on both axes. (a) Dahlbominus fuscipennis (ZETT.)<br />

attacking Neodiprion sertifer (GEoFr.) between 17. 5 <strong>and</strong> 24.0~ (BuRNETT<br />

1956; table I); (a') the same as (a) but below 17.5~ (b) Encarsia <strong>for</strong>mosa<br />

GAHAN attacking Trialeurodes vaporariorum (WEsrw.) (BURNETT 1958; table<br />

IV) ; (c) Chelonus texanus CRmss. attacking Ephestia kiihniella ZmLL. (ULLYEXT<br />

1949a; table II); (d) Cryptus inornatus PRAtt attacking Loxostege stricticalis<br />

L. (ULLYETT 1949b; table III); (e) Nemeritis canescens (GRAY.) attacking<br />

Anagasta kiihniella (ZsLL) (HuFFAK~m <strong>and</strong> KENN~rr 1969; figure 1). All figures,<br />

except (e), are calculated from original numerical data, <strong>and</strong> the densities <strong>of</strong><br />

parasites are expressed consistently as the number per square foot. However,<br />

the constant host density per square foot differs considerably between observations<br />

(i.e. 4/ft 2 in (a) <strong>and</strong> (a'), 2074 in (b), 3600 in (c), <strong>and</strong> 354 in (d)),<br />

<strong>and</strong> there<strong>for</strong>e these curves are not comparable with each other.


68<br />

(more correctly, into THOMPSON's equation, i.e. eq. (4g. 8), since the authors were<br />

dealing with <strong>parasitism</strong> rather than <strong>predation</strong>), the following relationship will be<br />

obtained:<br />

_Qyl-~.<br />

Z = X (1- e ) (4h. 2).<br />

The main theme <strong>of</strong> the HASSELL-VARLEY paper is to show that host-parasite<br />

oscillations, which are unstable under the NICHOLSON-BAILEY assumption <strong>of</strong> a constant<br />

~, can be stabilized if the value <strong>of</strong> m in eq. (4h. 1') is sufficiently large, so that the<br />

value <strong>of</strong> ~ becomes sufficiently small as Y increases. While the end conclusion by<br />

HASSELL <strong>and</strong> VARLEY, that the effect <strong>of</strong> social interference among parasites plays an<br />

important role in stabilizing host-parasite oscillations, might still be correct, their<br />

usage <strong>of</strong> eq. (4h. 2) as a basis <strong>of</strong> reasoning is not logically sound.<br />

First, if eq. (4h. 1) holds <strong>for</strong> any given constant value <strong>of</strong> Q <strong>and</strong> <strong>for</strong> all positive<br />

values <strong>of</strong> Y, then <strong>for</strong> m~0, the value <strong>of</strong> ~ increases without limit as Y increases.<br />

This is exactly the same misconception that is involved in WATT'S equation reviewed<br />

in ~4f; it has been shown that eq. (4f. 8) is a generalized instantaneous hunting equ-<br />

ation <strong>for</strong> the toss-a-ring model in which the effective area a diminishes by the factor<br />

YI-,~ as Y increases. If we use the same analogy, that the effective area a diminishes<br />

by the factor yl-.~ as Y increases, in the THOMPSONIAN model <strong>of</strong> the NICHOLSON-<br />

BAILEY type, rather than the toss-a-ring model, we have<br />

z :X(1-e -av'-~rt) (4h. 3).<br />

Equation (4h. 3) is perfectly equivalent to eq. (4h. 2) because, if the coefficients involv-<br />

ed in these two equations are estimated from the same set <strong>of</strong> data, the value <strong>of</strong> Q<br />

is the same as that <strong>of</strong> at <strong>and</strong> the value <strong>of</strong> m is the same as that <strong>of</strong> ~-1. As already<br />

pointed out in w 4f, however, the assumption that the effective area, d, changes by the<br />

factor yl-, as Y changes, is not acceptable.<br />

And if one criticizes NICHOLSON <strong>and</strong><br />

BAILEY in that the assumption <strong>of</strong> a constant a is biologically absurd, one may on the<br />

same logical basis criticize the postulation <strong>of</strong> an unlimited increase in 6 as equally<br />

absurd. It might have been that HASSELL <strong>and</strong> VARLEY, as well as WATT, thought<br />

that such an assumption was an approximation <strong>and</strong> could be used <strong>for</strong> practical purposes<br />

in the context <strong>of</strong> their argument. But, then, they should have referred to an objective<br />

criterion to set limits within which such an approximation could be tolerated <strong>for</strong> their<br />

further speculation on host-parasite oscillations based on that relationship. If, on the<br />

contrary, we start from an axiomatic view that the // will not exceed a certain finite<br />

value as Y decreases, a conclusion will be deduced that the relationship between the<br />

values <strong>of</strong> In gi <strong>and</strong> In Y has to be curvilinear. And from this point <strong>of</strong> view, it is<br />

obvious that every observed relationship in Fig. 11 is in fact curved to some degree;<br />

it varies from the most pronounced trend in (c) to the least pronounced one in (e).<br />

Secondly, it is clear, upon comparison with eq. (4h. 3), that eq. (4h. 2) ignores<br />

the fact that the d is also a function <strong>of</strong> X, the host density; the fact established in<br />

the preceding sections <strong>of</strong> this paper. Since HASSELL <strong>and</strong> VARLEY were fully aware


9<br />

<strong>of</strong> this fact, it is curious that they ignored it.<br />

One possible justification may be<br />

related to two statements. The first one (the last paragraph <strong>of</strong> the introductory part<br />

<strong>of</strong> their paper) stated:<br />

"These oscillations can be stabilized [the NICHOLSON-BAILEY equations generate<br />

host-parasite oscillations with ever-increasing<br />

amplitude] by reducing the area<br />

<strong>of</strong> discovery as parasite density increases, but changes in area <strong>of</strong> discovery in<br />

relation to host density do not promote stability".<br />

The second statement (the second <strong>and</strong> third paragraphs on p. 1135 <strong>of</strong> their paper)<br />

is summarized below:<br />

The NICHOLSON-BAILEY equation did not exactly fit the observed relationship<br />

between the winter moth, Operophtera brumata (L.), <strong>and</strong> its parasite Cratiech-<br />

neumon culex (MuELLER) in two aspects: (1) the calculated peak <strong>of</strong> the parasite's<br />

density lagged two generations after the peak <strong>of</strong> the host's density whereas the<br />

observed lag was only one generation; (2) while in the NICHOLSON-BAILEy model<br />

more than two parasite species could not coexist [because <strong>of</strong> the competitive<br />

exclusion <strong>of</strong> one species by another], there were several parasite species coexisting<br />

in the field. But, when eq. (4h. 2) was used instead <strong>of</strong> the NICHOLSON-BAILEY,<br />

it was found that both <strong>of</strong> the above difficulties disappeared.<br />

With respect to the first statement, it is certainly agreeable that reduction <strong>of</strong> "the<br />

area <strong>of</strong> discovery", i. e. d, in relation to increase in host density will not promote<br />

stability. It should be noticed, however, that such changes in the d tend to accelerate<br />

instability (see TINBERGEN <strong>and</strong> KLOMP 1960). This implies that the effect <strong>of</strong> changes<br />

in the d in relation to host density, which must be involved in actual host-parasite<br />

interaction systems, has to be counteracted by other factors, or conditions, more<br />

strongly than in a hypothetical situation in which changes in host density have no<br />

influence on the value <strong>of</strong> d. Since the HASSELL-VARLEY equation assumes that the d<br />

is independent <strong>of</strong> host density, this bias has to be cancelled out by another bias, <strong>and</strong><br />

this latter bias is in fact involved in the assumption expressed by eq. (4h. 1). Hence,<br />

the fact that the observed relationship between O. brumata <strong>and</strong> C. culex agreed with<br />

the theoretical relationship expressed in eq. (4h. 2) suggests that this model is another<br />

example <strong>of</strong> c21 under C2 in w 2.<br />

It was pointed out in w<br />

that HOLLING'S disc equation involved some bias, be-<br />

cause it assumed that the discovery <strong>of</strong> a prey was regarded as the capture <strong>of</strong> it.<br />

Nevertheless, the model enhanced the importance <strong>of</strong> the factor h, the h<strong>and</strong>ling time.<br />

By the same token, although these <strong>models</strong> involve certain contradictions, the HASSELL-<br />

VARLEY model, as well as WATT'S model, implies strongly the significance <strong>of</strong> social<br />

interference among parasites as one important regulatory mechanism in host-parasite<br />

interaction systems. As the effect <strong>of</strong> social interference seems very important, I shall<br />

investigate it more in detail in the following subsection.


7O<br />

i). A geometric model <strong>for</strong> social interaction among parasites (this <strong>study</strong>)<br />

In w 3, I introduced a function S, by which the effect <strong>of</strong> social interaction among<br />

attacking species upon the instantaneous hunting efficiency is indicated.<br />

Thus, if<br />

social interaction is involved, the instantaneous hunting equation <strong>for</strong> <strong>predation</strong> is given<br />

by eq. (3. 14) rather than eq. (3. 4). As already explained, however, the instantaneous<br />

equation <strong>for</strong> <strong>parasitism</strong> does not take the <strong>for</strong>m <strong>of</strong> a differential equation as in eq.<br />

(3. 14), but, <strong>for</strong> indiscriminate parasites, it is expressed in terms <strong>of</strong> the number <strong>of</strong><br />

eggs laid per unit area, i.e. n, as in eq. (3. 1). Thus, the equation <strong>for</strong> indiscriminate<br />

<strong>parasitism</strong>, equivalent to eq. (3. 14) <strong>for</strong> <strong>predation</strong>, will be written as:<br />

n=S(Y, X)f(X) Yt (4i. 1),<br />

<strong>and</strong> from eq. (3.22), we have an overall hunting equation <strong>for</strong> indiscriminate parasites<br />

as below:<br />

z=X{1-r (Y, X)f(X) Yt/X, V)} (4i. 2).<br />

In order to <strong>study</strong> some fundamental characteristics <strong>of</strong> the function S, I shall again<br />

use a geometric model similar to those used previously.<br />

Suppose, firstly, that a given parasite individual encounters, within an area 8<br />

around itself, other parasite individuals in the course <strong>of</strong> hunting.<br />

these other parasites encountered within the area 8 is 0, 1, 2 .......<br />

If the number <strong>of</strong><br />

or i, the instan-<br />

taneous hunting efficiency <strong>of</strong> the given parasite, i.e. f(X), is changed by factors 2o,<br />

21, ~2 ...... or 2~ respectively. It is conceivable, as a more general case, that 2 is<br />

influenced not only by the number <strong>of</strong> parasites in the 8, but also by the number <strong>of</strong><br />

hosts. This is because, as already pointed out in w 3, the effect <strong>of</strong>, say, interference<br />

might be strengthened or weakened if a lesser or greater number <strong>of</strong> hosts, respec-<br />

tively, is available within the 8. Thus, it is more appropriate to indicate the number<br />

<strong>of</strong> hosts too. Thus, 1~ is the index <strong>of</strong> the degree <strong>of</strong> social interaction when there<br />

are i parasites <strong>and</strong> j hosts within the area 8; it should be noted that both i <strong>and</strong> j<br />

take discrete values, 0, 1, 2 ..... , independently <strong>of</strong> each other.<br />

Secondly, let m (j) be the probability <strong>of</strong> finding j hosts within the 6. Then the<br />

average partial realization <strong>of</strong> the potential efficiency <strong>for</strong> a fixed value <strong>of</strong> i will be<br />

oo<br />

2~:~o(j). Similarly, let o(i) be the probability <strong>of</strong> finding i parasites within the 3.<br />

j-0<br />

Then the overall degree <strong>of</strong> changes in the instantaneous hunting efficiency <strong>for</strong> all j's<br />

<strong>and</strong> i's, i.e. S(Y, X), will be<br />

S(Y, X)= ~{ ~ 2~:~o(j)} o(i) (4i. 3).<br />

i=o j=0<br />

Now, ~o is the probability-distribution function <strong>of</strong> j (<strong>and</strong> can be determined when both<br />

the average number <strong>of</strong> hosts within the area 8 <strong>and</strong> its variance are known).<br />

So the<br />

oo<br />

value <strong>of</strong> ~2~jco(j) can be determined <strong>for</strong> a given value <strong>of</strong> X <strong>and</strong> <strong>for</strong> each value <strong>of</strong> i.<br />

j-0<br />

There<strong>for</strong>e, if the value <strong>of</strong> X is fixed in the following argument, the expression<br />

co<br />

~,~o(j) can be indicated simply by 2,(X). Then, eq. (4i. 3) will be written as<br />

j=o<br />

co<br />

S(Y, X) = ~ ,~,(X) o(i) (4i.4).<br />

i~0


71<br />

Now, if interference is involved among parasites, the value <strong>of</strong> ~l~ must decrease as i<br />

increases while j is fixed, i.e. 2~j~2~+lj, <strong>and</strong> facilitation is indicated conversely by<br />

,~ 2~+I(X), this indicates that the effect <strong>of</strong> interference<br />

outweighs that <strong>of</strong> facilitation, <strong>and</strong> vice versa. If all 2's are equally unity, this indicates<br />

that there is no social interaction, since from eq. (4i. 3),<br />

c~ co<br />

S(Y, X)=32,~o(j) ~o(i)=1.<br />

j=0 i=0<br />

In order to make further investigations <strong>of</strong> the nature <strong>of</strong> the function S <strong>and</strong> its<br />

influence on eq.<br />

(4i. 2), it may be more convenient to assume a certain concrete<br />

<strong>for</strong>m <strong>of</strong> the function 0. For this purpose, let us assume that 0 is a PomsoN distribu-<br />

tion function, i.e.<br />

p (i) : e -~r' (8 Y') '/i ! (4i. 5),<br />

where 6Y' is the mean number <strong>of</strong> parasites within the area 6 around a given parasite<br />

individual (excluding the given individual), <strong>and</strong><br />

6Y' = 8Y/(1-e -~r) -1 (4i. 6).<br />

(For the derivation <strong>of</strong> eq. (4i. 6), see Appendix 3.)<br />

If we adopt at this stage the<br />

THOMPSONIAN model, i.e. eq. (4g. 8), as a concrete <strong>for</strong>m <strong>for</strong> eq. (4i. 2):<br />

z :X(1 -e- {f(X) Yt/X} e -~Y' X (2~ (X) (5II')*/i !} ) (4i. 7).<br />

The evaluation <strong>of</strong> the // in eq. (4i. 7) is, from eq. (4b. 8),<br />

d= {f (X)t/X} e -at' X {,is(X) (6Y')'/i !} (4i. 8).<br />

Equation (4i. 8) is compared to eq. (4h. 1'), <strong>and</strong> if we take the logarithm <strong>of</strong> both<br />

sides <strong>of</strong> eq. (4i. 8), i.e.<br />

In ii =In {f (X) t/X} + ln[e-Sr'X {~, (X) (6 Y') '/i !} ] (4i. 9),<br />

<strong>and</strong> this equation is directly comparable with eq. (4h. 1) or with the curves in Fig.<br />

11.<br />

The following are comparisons between eqs.<br />

(4h. 1) <strong>and</strong> (4i. 9), or between eqs.<br />

(4h. 1') <strong>and</strong> (4i. 8). First, while the value <strong>of</strong> In Q in eq. (4h. 1) is constant, the equi-<br />

valent term (i. e. the first term <strong>of</strong> the right-h<strong>and</strong> side) in eq. (4i. 9) is a function <strong>of</strong><br />

X; this term in eq. (4i. 9) can be treated as constant when the value <strong>of</strong> X is fixed,<br />

since the term is independent <strong>of</strong> Y.<br />

Secondly, while the second term <strong>of</strong> the right-<br />

h<strong>and</strong> side <strong>of</strong> eq. (4h. 1) is a linear function <strong>of</strong> In Y, <strong>and</strong> independent <strong>of</strong> X, the equi-<br />

valent term in eq. (4i. 9) is not a linear function <strong>of</strong> In Y, <strong>and</strong> at the same time it is<br />

generally a function <strong>of</strong> X too; the term becomes independent <strong>of</strong> X only when 2,5 is<br />

independent <strong>of</strong> X <strong>for</strong> a given value <strong>of</strong> i. Thirdly, while the value <strong>of</strong> // in eq. (4h. 1)<br />

will increase without limit as Y decreases, the d in eq. (4i. 9) will converge to a<br />

finite value <strong>for</strong> a given fixed value <strong>of</strong> X, i.e.<br />

lim ?i =20 (X) f (X) t/X (4i. 10).<br />

Y~0<br />

Now, I shall examine the shape <strong>of</strong> curves that are generated by eq. (4i. 9), <strong>and</strong><br />

compare them with the observed data in Fig. 11. For the purpose <strong>of</strong> maintaining<br />

the generality <strong>of</strong> this model, the examination will be made analytically (i. e. mathemati-<br />

cally), <strong>and</strong> some concrete examples will be shown later.<br />

In order to find conditions


72<br />

Under which the value <strong>of</strong> In ~, <strong>for</strong> a given fixed value <strong>of</strong> X, is increasing, decreasing,<br />

or remaining constant, the first order partial derivative Oln ii/Oln Y will be calculated<br />

below:<br />

Oln ii .. dY' ~-Y,~+I (X) (~Y')~ 1 ) (4i. 11)<br />

Oln Y= o" ~ d Y ----;~Y'\~<br />

L z2,(x) ~ ~i )<br />

in which the derivative d Y'/dY is, from eq. (4i. 6):<br />

d Y'/d Y= { (1 - e -~r) -/~ Ye -~r } / (1 - e -st) 2>0.<br />

From the above evaluation <strong>of</strong> the partial derivative, the following conclusions will be<br />

drawn:<br />

(1). When Y->O, the partial derivative converges to zero, so that the curve is parallel<br />

to the In Y axis at the level <strong>of</strong><br />

In ii=ln{f (X)t/X} +ln 20(X)<br />

(see eq. (4i. 10)).<br />

~ oo<br />

(2). When Y is sufficiently small, so that ~ 2i+1 (X) (~Y') ~/i [ <strong>and</strong> 2E 2i (X) (~Y') ~/i !<br />

i=1 i~l<br />

are negligible as compared with ,h(X)<strong>and</strong> ;o(X) respectively, then<br />

Oln ii/Oln Y~--~Y(dY'/dY) {2~(X)/2o(X)-1}.<br />

There<strong>for</strong>e: (a) if 2~ (X) >20 (X) , i.e. social facilitation, the curve is increasing<br />

as Y increases, but (b) if At (X) ~ At(X) (SY')'/i [,<br />

i=0 i=0<br />

the partial derivative in eq. (4i. 11) is positive, <strong>and</strong> so the curve is increasing,<br />

but (b) if the effect <strong>of</strong> interference outweighs that <strong>of</strong> facilitation, the curve is<br />

decreasing.<br />

(4). When Y becomes sufficiently large, both lower <strong>and</strong> higher terms in the series<br />

{2,(X)(~Y')~/i!} will become negligible as compared with mid-terms, i.e. <strong>for</strong><br />

certain numbers k <strong>and</strong> k', we have<br />

co<br />

k t<br />

2~(X) (~Y')~/i ! ~ ~ At(X) (~Y')'/i !<br />

i=O<br />

i=k<br />

<strong>and</strong> the same applies to the series {A,+~(X)(~Y')*/i!}. Now it is unlikely that<br />

the degree <strong>of</strong> social facilitation increases indefinitely as i increases; the effect<br />

<strong>of</strong> interference must sooner or later become apparent. Hence, beyond a certain<br />

number <strong>for</strong> i, e. g. k, the inequality A, (X) >Ak+~ (X) will always hold. Under<br />

these circumstances, the partial derivative becomes always negative, <strong>and</strong> hence


73<br />

the curve must be decreasing <strong>for</strong> large values <strong>of</strong> Y. Although the pro<strong>of</strong> is<br />

curtailed here (because it can easily be confirmed by calculating the second<br />

order derivative), it should be mentioned that whether the rate <strong>of</strong> decrease is<br />

accelerated or decelerated depends on the rate <strong>of</strong> decrease in 2~(X) with increasing<br />

i; the curve is decreasing with an increasing rate if the value <strong>of</strong> 2<br />

decreases <strong>comparative</strong>ly fast as i increases, but the rate <strong>of</strong> decrease in the curve<br />

may become lower if the value <strong>of</strong> 2 decreases only slowly with increasing i.<br />

Some examples <strong>of</strong> curves generated by eq. (4i. 9) are shown in Fig. 12. These hypothetical<br />

curves cannot be compared directly with the observed curves <strong>and</strong> scattergram<br />

in Fig. 11, because the values <strong>of</strong> f(X), t, <strong>and</strong> ~ are not known in these observations.<br />

tOO<br />

0,50<br />

Xo(x} = 1.0<br />

e<br />

)h(x) = 0,5 I<br />

(i~<br />

1l<br />

0,10<br />

(1} o= 0<br />

(2) = 0,25<br />

0,0. ~<br />

13) = 0.50<br />

[z, ) = 0,75<br />

X<br />

:0<br />

l,f,-<br />

0<br />

=,<br />

~ 1.0C<br />

i ' ' ' I I<br />

0,1<br />

I I I L I I I i| I I I I I I I II I I<br />

1.0 10,0<br />

o~5o<br />

t<br />

{1)<br />

(2)<br />

0 , 2 0 , 5 , B,0,2,,,5 \ \,~,<br />

1.00 1,50 0,80 0.50 0.25 0,21 0.18 014 011 0.09 0.08 0.07<br />

1oo ,.,2o loo .... o.o oJo 0,50 o: 030 21; 0.27 ::: 0.25 ::; 0,23<br />

0,IC<br />

(1}<br />

..... 0:, . . . . . . . . ,'0 . . . . . . . . ,~0 ' '<br />

MEAN No. OF PARASITES PER EFFECTIVE AREA OF INTERACTION<br />

Fig. 12a. Hypothetical relationships between the values <strong>of</strong> ii/{f(X)t/X} <strong>and</strong> 6Y,<br />

(mean number <strong>of</strong> parasites per effective area) calculated from eqs. (4i. 6) <strong>and</strong><br />

(4i. 8), plotted in the natural logarithmic scale on both axes. The values <strong>of</strong><br />

2i (X) shown in the figure decreases as i increases, indicating that social<br />

interference only is considered here.<br />

Fig. 12b. The same as in Fig. 12a, but the value <strong>of</strong> 2i(X) increases from i=0<br />

to 1, indicating social facilitation, <strong>and</strong> then decreases towards higher values <strong>of</strong> i.


74<br />

However, the similarity in their shapes can be compared, if desired, by parallel<br />

translation <strong>of</strong> the relative position <strong>of</strong> the coordinate systems between the observed<br />

<strong>and</strong> hypothetical relationships, since the curves are drawn on the ln-ln scale. Then,<br />

the shapes <strong>of</strong> curves (a) <strong>and</strong> (a') in Fig. lla are comparable to a certain part <strong>of</strong><br />

curve (1) in Fig. 12b. The scattergram (e) in Fig. llb resembles curve (3) in Fig.<br />

12a, <strong>and</strong> so on. A strict comparison will not be attempted here <strong>for</strong> the above reason,<br />

however.<br />

It should be mentioned finally that fitting a straight line to these observed relationships<br />

may be justified only <strong>for</strong> the purpose <strong>of</strong> showing the declining tendency <strong>of</strong><br />

the value <strong>of</strong> ~ with increasing parasite density. In other words, the only conclusion<br />

that one can draw from such linear regression analysis is restricted to the suggestion<br />

that social interference is involved among parasites. However, there is no justified<br />

basis <strong>for</strong> adopting the hypothesis that the relationship is linear. Also, the assumption<br />

<strong>of</strong> the 'quest constant' by HASSELL <strong>and</strong> VARLEY (1969) (see w 4h) is justified only in<br />

the linear regression analysis <strong>of</strong> those data in which host density is known to be<br />

constant: the assumption is, however, hardly justified <strong>for</strong> speculating about the stability<br />

<strong>of</strong> host-parasite oscillations in which host density is changing all the time. The<br />

possibility <strong>of</strong> stable oscillations induced by social interference among parasites is yet<br />

to be demonstrated on a more reasonable basis; until it is, the suggestion by HASSELL<br />

<strong>and</strong> VARLEY is only a possibility.<br />

Appendix to w 4i. Is the concept <strong>of</strong> 'area <strong>of</strong> discovery' useful in studies<br />

<strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> ?<br />

It has been shown in this paper that the concept <strong>of</strong> 'area <strong>of</strong> discovery', originally<br />

introduced by NICHOLSON (1933), cannot be used as a geometric attribute <strong>of</strong> the<br />

hunting process, since this simple, but highly hypothetical, concept involves a contradiction<br />

from the energetics point <strong>of</strong> view. But the concept has been shifted, as one<br />

way <strong>of</strong> expressing the hunting efficiency <strong>of</strong> predators or parasites, <strong>and</strong> has been widely<br />

used in the literature <strong>of</strong> population dynamics. The shifted concept is now defined as<br />

d in eq. (4b. 7) <strong>for</strong> <strong>predation</strong> <strong>and</strong> in eq. (4b. 8) <strong>for</strong> <strong>parasitism</strong>. The definition, however,<br />

is not a straight<strong>for</strong>ward expression <strong>of</strong> hunting efficiency, as it is the logarithm<br />

<strong>of</strong> the reciprocal value <strong>of</strong> the survival rate, <strong>for</strong> a specified value <strong>of</strong> the initial density<br />

<strong>of</strong> the hunted species per hunter.<br />

My question here is whether this concept <strong>of</strong> 'area <strong>of</strong> discovery' is altogether<br />

useful in the <strong>study</strong> <strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>. Of course, the concept has played a<br />

significant role in its original context as a species specific constant under a given<br />

condition. But, once the original meaning <strong>of</strong> this index as a species specific constant<br />

is lost, what does the shifted concept signify ? Is there any particular advantage in<br />

using this index in shifted, <strong>and</strong> more general, situations ? In order to answer these<br />

questions, the index d will be evaluated in various <strong>models</strong> reviewed in this paper<br />

<strong>and</strong> will be compared with the instantaneous hunting function on which each model


75<br />

is based. As the evaluation <strong>of</strong> the//is in general different between <strong>models</strong> <strong>for</strong> preda-<br />

tion <strong>and</strong> <strong>parasitism</strong>, I shall use a symbol //, <strong>for</strong> <strong>predation</strong> <strong>and</strong> i/2 <strong>for</strong> <strong>parasitism</strong>, so<br />

that:<br />

//1 : 1 x0<br />

In<br />

_r Xo--z<br />

ii2=l_ln<br />

X<br />

X-z"<br />

Also, the expression f(x, Y) will be used as a general <strong>for</strong>m <strong>of</strong> the instantaneous<br />

hunting function; x should be replaced by X <strong>for</strong> <strong>parasitism</strong>.<br />

1. The LOTKA-VOLTERRA model<br />

The definition <strong>of</strong> //1 does not fit here, since the model takes into account changes<br />

in the densities <strong>of</strong> both predator <strong>and</strong> prey populations during the hunting period, t.<br />

In other words, first, the value <strong>of</strong> Y, defined as a fixed predator density during t,<br />

does not exist, <strong>and</strong> secondly the value <strong>of</strong> z is influenced by mortality in the prey<br />

population due to factors other than <strong>predation</strong>.<br />

Under these general circumstances,<br />

the redefined concept <strong>of</strong> 'area <strong>of</strong> discovery' just does not exist. If, however, it is as-<br />

sumed as a specific case that mortality in the prey population does not occur except<br />

by <strong>predation</strong>, <strong>and</strong> that predator density is fixed during t, the model converges to the<br />

NICHOLSON-BAILEY model.<br />

2. The NICHOLSON-BAILEY model<br />

f(x, Y)=ax<br />

//1 =at.<br />

3. HOLLING'S disc model<br />

f(x, Y)=ax/(l+ahx)<br />

//1 :at-ahz/Y<br />

//2-at/(l+ahX), under the THOMPSONIAN assumption.<br />

4. IVLEV'S model<br />

f(x, Y) = b (1 - e -~')<br />

//1 = - (1/Y) In [-1 + (1/axo) In { (1 - e- axo) e- abYt + e- aXo } ]<br />

5. WATT's model<br />

Same as IVLEV'S, but<br />

a~ayl-~.<br />

iiz-bt(1-e-~x)/x, under the THOMPSONIAN assumption.<br />

6. ROYAMA'S model in w 4e<br />

f(x, Y) =a (x) x Yt/{1+ a (x) hx}<br />

where a (x) is defined in p. 46, w 4e.<br />

7. THOMPSON'S model<br />

~i2=n/XY,<br />

//1: may be evaluated from eq. (4e. 9), but since its analytical solution<br />

with respect to z is difficult, the evaluation will not be attempted<br />

here.<br />

62 =a (X) t~ {1 +a (X) hX}, under the THOMPSONIAN assumption.


<strong>and</strong> if n is evaluated as f(X, Y)Yt,<br />

ii2=f(X, Y)t/X.<br />

So, if f(X, Y)=aX as in the NICHOLSON-BA[LEY model,<br />

gi 2 - at.<br />

8. The HASSELL-VARLEY model<br />

ii2_O y -,~<br />

or, using my system <strong>of</strong> notations,<br />

5z =aY1-Bt.<br />

(Note that the THOMPSONIAN assumption is inherent to this model. )<br />

9. ROYAMA'S model in w<br />

52-e -~y' ~. ~2~(X) (~Y')~/i!} f(X)t/X, under the THOMPSONIAN assumption.<br />

i-0<br />

(For symbols, see eq. (4i. 8).)<br />

A comparison between the evaluation <strong>of</strong> the 52 <strong>and</strong> the function f (as in the<br />

second equation in 7 above) clearly shows that the 'area <strong>of</strong> discovery' <strong>of</strong> parasites<br />

<strong>of</strong> the indiscriminate type is directly related to the instantaneous hunting efficiency<br />

under the THOMPSONIAN assumption, i.e.<br />

lows a POISSON series.<br />

that the distribution <strong>of</strong> parasite eggs fol-<br />

In other words, the shifted concept <strong>of</strong> 'area <strong>of</strong> discovery',<br />

under the THOMPSONIAN assumption, still maintains its significance as an index <strong>of</strong><br />

the hunting efficiency <strong>of</strong> the parasites concerned.<br />

Such significance, however, is re-<br />

stricted only to the situation in which the THOMPSONIAN assumption holds. If a gener-<br />

alization is made to cover those indiscriminate parasites which do not distribute<br />

eggs after the POISSON fashion, those which discriminate between parasitized <strong>and</strong><br />

unparasitized hosts, or predators, the 'area <strong>of</strong> discovery' is not directly related to the<br />

hunting efficiency (compare, <strong>for</strong> instance, the 5Vs with corresponding f's in the above<br />

list). Furthermore, if the 'area <strong>of</strong> discovery' is calculated from data in which mortality<br />

in either hunting or hunted, or both, species occurs during the hunting period, as is<br />

the case with the LOTKA-VOLTERRA model, the index cannot be calculated.<br />

suggests that the calculation <strong>of</strong> the index from the data obtained in the field is theo-<br />

retically difficult, since mortality among the hunting species certainly occurs; calcu-<br />

lating the index by using the average density might be attempted, but then it has to<br />

be remembered that the index could not be linearly related to the efficiency.<br />

Thus, the concept <strong>of</strong> 'area <strong>of</strong> discovery' loses its significance on general ground,<br />

<strong>and</strong> there is no particular<br />

This<br />

advantage in using it. What is essential is to find the<br />

method <strong>of</strong> determining the instantaneous hunting function directly. This problem is,<br />

however, beyond the scope <strong>of</strong> this paper.<br />

j). HOLLING'S hunger model<br />

In w 3, I showed one method <strong>of</strong> incorporating the hunger component. The hunger<br />

level there was defined by function H which expressed a partial realization <strong>of</strong> the<br />

potential maximum per<strong>for</strong>mance that each predator can exert in hunting at given<br />

prey <strong>and</strong> predator densities. HOLLING (1966), in his <strong>study</strong> <strong>of</strong> the <strong>predation</strong> behaviour


77<br />

<strong>of</strong> Hierodula crassa GIGLIO-TOs., approached the problem from a different direction.<br />

A female mantid, H. crassa, had been deprived <strong>of</strong> food <strong>for</strong> various lengths <strong>of</strong> time<br />

be<strong>for</strong>e flies (as prey) were <strong>of</strong>fered, <strong>and</strong> then the weight <strong>of</strong> flies eaten by the mantid<br />

was measured <strong>for</strong> each length-class <strong>of</strong> deprivation time. It was found that the weight<br />

<strong>of</strong> flies eaten increased as the deprivation time increased (<strong>and</strong> hence the mantid was<br />

hungrier), gradually leading to a plateau.<br />

The effect <strong>of</strong> hunger revealed itself not only in the mantid's increased dem<strong>and</strong><br />

<strong>for</strong> food to the level <strong>of</strong> satiation, but also in other components, e.g. the size <strong>of</strong> the<br />

area <strong>of</strong> reaction to the prey, speed <strong>of</strong> reaction, capture success, time spent in pursuing<br />

<strong>and</strong> in eating prey, <strong>and</strong> in the digestive pause. The influence <strong>of</strong> the deprivation time<br />

on each <strong>of</strong> these components was expressed by separate descriptive equations which<br />

were then synthesized to describe the relationship between the number <strong>of</strong> prey killed,<br />

the density <strong>of</strong> prey, <strong>and</strong> the time involved; the relationship thus obtained was illustrated<br />

by HOLLING (1966) in his figure 29.<br />

It is not my intention here to review critically every detail <strong>of</strong> HOLLING'S mathematical<br />

treatment, as the <strong>study</strong> <strong>of</strong> the effect <strong>of</strong> hunger is still in its infancy, <strong>and</strong> also<br />

because I have not had sufficient experience with the problem myself. There are,<br />

however, a few things to be pointed out which HOLLINC seems to have missed.<br />

First, in this <strong>study</strong> again HOLLING did not recognize the effect <strong>of</strong> diminishing<br />

returns. It is not certain whether, in the observation that appeared in his figure 29,<br />

the prey density was kept constant during each set <strong>of</strong> observations. If so, the figure<br />

represents an instantaneous hunting surface equivalent to eq. (3. 18) in which dx/dt<br />

is written simply as n. If, however, the prey density was depleted during the course<br />

<strong>of</strong> observation, the figure represents a particular one <strong>of</strong> the overall hunting surfaces<br />

which is specific only to the mantid density used in this particular experiment. In<br />

this case, the density <strong>of</strong> predators should have been stated (the number <strong>of</strong> mantids<br />

used might have been just one, but as the fly density was expressed per square centimetre,<br />

the mantid density could not be unity). Also, the theoretical curve fitted to<br />

the data in the same figure is in fact an instantaneous rate, <strong>and</strong> so if the density was<br />

depleted, the comparison is not justifiable.<br />

Secondly, if our aim is to obtain an overall hunting equation, which is no doubt<br />

needed in population dynamics, an appropriate instantaneous hunting equation is<br />

required, the reason <strong>for</strong> this being explicit in earlier sections <strong>of</strong> the present paper.<br />

To obtain an instantaneous hunting equation, from which the final synthesis is made,<br />

the experimental analysis <strong>of</strong> the elemental components should have been designed<br />

accordingly. However, the observed relationship, <strong>for</strong> instance, between the amount <strong>of</strong><br />

prey eaten <strong>and</strong> the deprivation time in HOLLING'S original paper (1966, figures 4 <strong>and</strong><br />

5) is not appropriately tailored <strong>for</strong> the above purpose. This is because the time<br />

involved in consuming a given amount <strong>of</strong> prey was not explicitly considered by the<br />

author. Suppose the amount eaten up to the state <strong>of</strong> satiation (see HOLLING'S definition,<br />

1966 p. 16) was W~ <strong>and</strong> W2, when the deprivation time was 7", <strong>and</strong> 2"2, <strong>and</strong> tl


78<br />

<strong>and</strong> t2 hours were required to consume WI <strong>and</strong> Wz, respectively, after the flies were<br />

<strong>of</strong>fered. Then the rates <strong>of</strong> consumption W~/t~ <strong>and</strong> W2/t~ could be considered as<br />

instantaneous rates if tl <strong>and</strong> t2 were not too large. It might be technically difficult to<br />

keep tPs sufficiently small, <strong>for</strong> otherwise W's could not be measured. If t's are long,<br />

then digestion may take place during those hours, <strong>and</strong> this must influence the value<br />

<strong>of</strong> W. Then what is required is the measurement <strong>of</strong> the relationship between W <strong>and</strong><br />

t <strong>for</strong> various Tts, from which the instantaneous rate, d W/dt, may be obtained. And,<br />

<strong>of</strong> course, it is d W/dt which should be incorporated in the synthesized instantaneous<br />

hunting equation.<br />

Although HOLLING'S approach, which he called an 'experimental component analy-<br />

sis', is no doubt important, some technical difficulties are expected, namely how to<br />

design experiments to meet theoretically required conditions; the example cited above<br />

clearly illustrates these difficulties. This is why I proposed a simpler approach in w 3,<br />

which can tentatively be used <strong>for</strong> calculating a predator-prey interaction without going<br />

through the details <strong>of</strong> physiological studies <strong>of</strong> hunger.<br />

In passing, HULLING tOO used differential equations, which could yield curves<br />

resembling observed ones, without attaching any significance to the equations as the<br />

means <strong>of</strong> inference. I must again suggest avoiding this unjustifiable operation.<br />

5. DISCUSSION AND CONCLUSIONS<br />

In this section, I shall deal with problems that are more methodological than<br />

technical. Be<strong>for</strong>e doing so, however, what was dealt with in w 4 will be summarized<br />

in the following diagram (Fig. 13).<br />

It is a flow diagram <strong>of</strong> reasoning leading to<br />

each model reviewed, <strong>and</strong> shows the scope that is covered by that model. The dia-<br />

gram is based on my own <strong>study</strong> <strong>and</strong> not necessarily identical to what the authors<br />

claimed in their original papers, as their verbal statements were <strong>of</strong>ten wrong.<br />

The reasoning starts from (A), a generalized instantaneous hunting equation.<br />

This generalization is obvious from eq. (3. 18), in which H(x, Y, t)f(x) can be written<br />

as f(X, Y, t) if x is fixed at X. From (A) there are two main streams, dealing with<br />

<strong>parasitism</strong> <strong>and</strong> <strong>predation</strong>.<br />

The <strong>predation</strong> flow is further divided into subflows 1 <strong>and</strong> 2.<br />

Subflow 2 goes<br />

directly to the determination <strong>of</strong> specific <strong>for</strong>ms <strong>of</strong> (A) to evaluate n, the number <strong>of</strong><br />

prey taken per unit area in time interval t when the prey density is kept constant.<br />

All the <strong>models</strong> after 1955 (except mine) belong to this flow. Since (A) is an instan-<br />

taneous equation, it cannot be used <strong>for</strong> comparison with observation except <strong>for</strong> cases<br />

in which reduction in the prey density can be neglected. Also (A) does not give any<br />

means <strong>of</strong> estimating the final density <strong>of</strong> the prey or host population at the end <strong>of</strong><br />

each generation.<br />

Hence, these equations in the category <strong>of</strong> (A) cannot be used, as<br />

they st<strong>and</strong>, <strong>for</strong> the <strong>study</strong> <strong>of</strong> prey-predator or host-parasite interaction systems.<br />

Subflow 1, however, incorporates the effects upon the number <strong>of</strong> prey taken per<br />

unit area (i. e. z) <strong>of</strong> (a) diminishing returns, (b) changes in the number in the prey


79<br />

population caused by factors other than <strong>predation</strong>, <strong>and</strong><br />

(c) changes in the numbers<br />

in the predator population. All the classical <strong>models</strong> be<strong>for</strong>e 1935, <strong>and</strong> also mine, come<br />

into this flow. In the diagram, (a), (b), <strong>and</strong> (c) are assumed to be independent <strong>of</strong><br />

each other, <strong>and</strong> under this assumption the simultaneous equations (B) are a priori.<br />

However, if there is any interaction between these effects, the equations are not a<br />

priori <strong>and</strong> so need experimental confirmation. Also, the use <strong>of</strong> calculus can be justi-<br />

fied only under the assumption that the pattern <strong>of</strong> the~ispatial distribution <strong>and</strong> move-<br />

ments <strong>of</strong> the animals concerned remain unchanged throughout the time-interval t.<br />

This is perhaps only approximately so, or may be even a very poor approximation as<br />

my first simulation model <strong>for</strong> <strong>predation</strong> in w<br />

clearly shows, <strong>and</strong> it requires experi-<br />

mental verification. Until then calculus is only a tentative method.<br />

In the step under the heading 'special cases' in the diagram, the flows diverge<br />

according to the more specific assumptions adopted, <strong>and</strong> within the scope <strong>of</strong> each<br />

hypothetical situation (or assumption adopted), all <strong>of</strong> them are legitimate in that<br />

there is no logical contradiction at this stage. This step is followed by the specification<br />

<strong>of</strong> the functions involved, under the heading 'specific equations'.<br />

The specific <strong>for</strong>ms <strong>of</strong> the function f(X, Y, t) are one <strong>of</strong> the major concerns in<br />

this <strong>study</strong>. The meaning <strong>of</strong> each <strong>for</strong>m was interpreted in the light <strong>of</strong> the geometric<br />

properties <strong>of</strong> hunting behaviour under the assumption <strong>of</strong> r<strong>and</strong>om distribution in the<br />

prey population. The assumption <strong>of</strong> r<strong>and</strong>om distribution is legitimate in a theoretical<br />

<strong>study</strong> like this, as the first step towards more general, irregular distribution patterns<br />

in future studies (the problem will be discussed elsewhere). Some other assumptions<br />

appearing in certain specific <strong>for</strong>ms in WATT (1959), NICHOLSON-BAILEY (1935), LOTKA-<br />

VOLTERRA (1925-1926), <strong>and</strong> GAUSE (1934) are not legitimate: in the WATT equation,<br />

as well as in HASSELL <strong>and</strong> VARLEY, rl-,~ as a measure <strong>of</strong> the degree <strong>of</strong> social inter-<br />

action is contradictory to the premise that a predator has a limited capacity to attack<br />

its prey; in the NICHOLSON-BAILEY equation, f(x) as a linear function <strong>of</strong> x is a priori<br />

impossible; in the LOTKA-VOLTERRA equations, the same criticism as in the NICHOLSON-<br />

BAILEY applies, <strong>and</strong> also the assumption <strong>of</strong> a constant r' is theoretically incorrect;<br />

finally, in GAUSE'S equation, his suggestion concerning g~(x)--f(x)y (GAusE 1934,<br />

<strong>for</strong>mula (25), p. 57) is not comprehensible.<br />

The evaluation <strong>of</strong> z (the reduction <strong>of</strong> the prey density during time-interval t in<br />

a system with discrete generations) is possible only through the reasoning <strong>of</strong> subflow<br />

1. Such an evaluation was made in the original literature only by NICHOLSON <strong>and</strong><br />

BAILEY. All <strong>of</strong> the three recent <strong>models</strong> (i.e. IVLEV, HOLLING, <strong>and</strong> WATT) were<br />

concerned only with the evaluation <strong>of</strong> n, <strong>and</strong> LOTKA <strong>and</strong> VOLTERRA gave only one<br />

special solution <strong>for</strong> a system with continuous generations. There<strong>for</strong>e, tho~e evaluations<br />

in the diagram were made in the present <strong>study</strong> (w 4). It should also be mentioned<br />

here that the evaluation <strong>of</strong> z made in this <strong>study</strong>, except <strong>for</strong> the LOTKA-VOLTERRA<br />

equations, assumed that both functions gl <strong>and</strong> g2 were zero as in the NICHOLSON-<br />

BAILEY equation, but this is only possible in an idealized, experimental set-up. The


80 84<br />

assumption, however, plays a legitimate role in the process <strong>of</strong> inferences as discussed<br />

in w 2. If the assumption does not hold, the functions gl <strong>and</strong> gz have to be determined<br />

experimentally, as there seems no method presently available to deduce specific<br />

<strong>for</strong>ms <strong>for</strong> these functions by analogy. But this problem is not relevant to the present<br />

<strong>study</strong> <strong>of</strong> hunting behaviour.<br />

GAUSE'S model is inadequate, because the logistic law was taken into account<br />

only in gl <strong>and</strong> not in g2. Also, f is not specified. It is pointed out here, however,<br />

that GAUSE'S equation is applied specifically to a system with continuous generations,<br />

since the introduction <strong>of</strong> a logistic function as a specific <strong>for</strong>m <strong>for</strong> g~ positively excludes<br />

the case <strong>of</strong> discrete generations.<br />

It is clearly shown in this <strong>comparative</strong> <strong>study</strong> that it is LOTKA <strong>and</strong> VOLTERRA,<br />

the pioneers in the theoretical <strong>study</strong> <strong>of</strong> the prey-predator interaction system, whose<br />

thought <strong>and</strong> insight covered the widest scope, <strong>and</strong> who laid the foundation <strong>for</strong> the<br />

<strong>for</strong>malization <strong>of</strong> the system, although their specific <strong>for</strong>ms were unsatisfactory. All<br />

other later <strong>models</strong> covered only a fraction <strong>of</strong> the basic structure <strong>of</strong> the system. Yet,<br />

surprisingly, none <strong>of</strong> the later authors appeared to be aware <strong>of</strong> this fact. Thus,<br />

NICHOLSON <strong>and</strong> BAILEY proposed their model as an alternative to the LOTKA-VOLTERRA<br />

one without noticing that the scope <strong>of</strong> their model had already been potentially covered<br />

by the earlier model. In other words, while LOTKA <strong>and</strong> VOLTERRA gave one solution<br />

to a system with continuous generations, the NICHOLSON-BAILEY competition equation<br />

is nothing more than another solution <strong>of</strong> the LOTKA-VOLTERRA equations in a system<br />

with discrete generations. The attempts by HOLLING, IVLEV, <strong>and</strong> WATT were concerned<br />

only with the improvement <strong>of</strong> the specific <strong>for</strong>m <strong>of</strong> function f, <strong>and</strong> completely<br />

neglected the reasoning along subflow 1 in Fig. 13, <strong>and</strong> thus these authors failed to<br />

separate n from z. WATT'S proposal <strong>of</strong> his equation even involves a contradiction;<br />

the proposal can hardly be called an improvement. Yet, the claims by these recent<br />

authors that their <strong>models</strong> were more realistic than the classical ones have been accepted<br />

by many ecologists in this field. These facts are an indication <strong>of</strong> the lack <strong>of</strong><br />

rigorousness in the attitude <strong>of</strong> ecologists, <strong>and</strong> this will be considered below.<br />

First, some concepts <strong>and</strong> terminology used in this field <strong>of</strong> ecology are too loose;<br />

thus confusion occurs in communication between ecologists or even within the mind<br />

<strong>of</strong> a single person. For example, WATT (1962) stated that the classical <strong>models</strong> failed<br />

because the reasoning started from 'a priori assumption' <strong>and</strong> was purely deductive,<br />

<strong>and</strong> thus he proposed (WATT 1959) what he called a 'deductive-inductive method'.<br />

This criticism <strong>of</strong> the classical <strong>models</strong> <strong>and</strong> the proposal <strong>of</strong> the alternative are not<br />

convincing <strong>for</strong> the following reasons.<br />

First, the premise in the classical <strong>models</strong> as appeared in specific <strong>for</strong>m, e.g. f(x)<br />

=ax, is not a priori. Strictly speaking, the phrase a priori means that which is<br />

"marked by being knowable by reasoning from what is considered self-evident <strong>and</strong><br />

there<strong>for</strong>e without appeal to the particular facts <strong>of</strong> evidence" (WEBSTER'S 3 rd International<br />

Dictionary 1968), from which is derived secondarily that which is 'intuitive' or


81<br />

'without experience'. If one uses the phrase to mean just 'intuitive or without expe-<br />

rience' completely emancipated from the original meaning, one might include any as-<br />

sumption set <strong>for</strong>th without confirmation by observation. Then, the expression 'a priori<br />

assumption' is tautological, since an 'assumption' is a premise adopted be<strong>for</strong>e a thing<br />

is known. Although the premise in the classical <strong>models</strong> was set <strong>for</strong>th without support<br />

from factual evidence, it is not a priori.<br />

On the contrary, the premise is a priori<br />

impossible as it violates the second law <strong>of</strong> thermodynamics. If any reasoning starts<br />

from an assumption which is a priori impossible, the conclusion drawn deductively<br />

is bound to be contradictory upon comparison with a fact. (This is the case with c22<br />

under C2 in w<br />

Thus, it is to be expected that the specific <strong>for</strong>m <strong>of</strong> the classical<br />

<strong>models</strong> would fail to describe real events satisfactorily.<br />

entirely invalidate the classical <strong>models</strong>, as will be shown later.<br />

This, however, does not<br />

The second point is concerned with deductive <strong>and</strong> inductive methods <strong>of</strong> inference.<br />

Today everyone knows, as WALKER (1963) pointed out (see w 2), that deduction<br />

does not produce more than has been involved in the premise, <strong>and</strong> there<strong>for</strong>e this<br />

method <strong>of</strong> inference alone will not contribute to our knowledge <strong>of</strong> natural order. It<br />

is impossible to believe that mathematicians like LOTKA <strong>and</strong> VOLTERRA did not know<br />

this rule: rather, they must have had a firm reason to present their <strong>models</strong> as deduc-<br />

tive ones. We know, as FRANCIS BACON himself had long ago pointed out, that a<br />

simple induction, i.e. a mere enumeration <strong>of</strong> facts, is no better, <strong>and</strong> even 'childish',<br />

<strong>and</strong> that a new concept would be <strong>for</strong>med only by what BACON called 'gradual induc-<br />

tion', i.e. a gradual passage from concrete facts to broader <strong>and</strong> broader generaliza-<br />

tions (DucASSE 1960). The process <strong>of</strong> gradual induction, however, does not exclude<br />

a phase which is deductive, e.g. once a certain assumption is made, perhaps by<br />

induction, a conclusion can be drawn only by reasoning, <strong>and</strong> then this conclusion is<br />

compared with observation. This is very similar to what I described in w 2. My<br />

interpretation <strong>of</strong> the LOTKA-VoLTERRA method is that they were showing what such<br />

deductive phases <strong>of</strong> reasoning could be like. That is, what they have shown is a<br />

model <strong>of</strong> deductive reasoning which is conveniently separated from the entire process<br />

<strong>of</strong> inference.<br />

It was rather un<strong>for</strong>tunate that the premise in the classical <strong>models</strong> was in fact<br />

biologically absurd (IVLEV 1961) <strong>and</strong> did not appeal to ecologists. However, <strong>for</strong> a<br />

mathematician who tries to show the basic structure <strong>and</strong> the method <strong>of</strong> analysis (to<br />

<strong>for</strong>m a hypothesis), the adoption <strong>of</strong> such specific <strong>for</strong>ms might have been merely<br />

trivial <strong>and</strong> only tentative since it can be changed readily if desired; but the principle<br />

<strong>of</strong> the mathematical method remains uninfluenced. This attitude is very clear in<br />

LOTKA'S work. It is un<strong>for</strong>tunate that ecologists became too much concerned with<br />

such a casual premise <strong>and</strong> failed to see the more fundamental aspect <strong>of</strong> the idea.<br />

This point is clearly illustrated in the three recent model builders, as reviewed above,<br />

who failed to distinguish n from z. This failure to underst<strong>and</strong> the classical model<br />

is not only seen in these authors but also in others (e. g. ANDREWARTHA <strong>and</strong> BIRCH


82<br />

1954; MILNE 1957) who thought that the premise <strong>and</strong> the structure <strong>of</strong> the classical<br />

<strong>models</strong> were far too simple to be realistic.<br />

It should be pointed out, however, that<br />

those who proposed what was claimed to be more realistic, taking so many conceiv-<br />

able factors into account, have never been able to <strong>for</strong>malize the ideas that they stated<br />

only verbally, or have not even tried to do so.<br />

From such verbal statements, one<br />

cannot draw a quantitatively expressed conclusion that can be compared with ob-<br />

served quantities <strong>for</strong> testing.<br />

Now it is clear that the criticism <strong>of</strong> the classical <strong>models</strong> was due to insufficient<br />

underst<strong>and</strong>ing <strong>of</strong> the nature <strong>of</strong> inferences.<br />

As pointed out in w 4f, although WATT<br />

claimed that the assumption <strong>of</strong> the coefficient A as it appeared in eq. (4f. 5) was<br />

based on an empirical fact, it was in fact an illusion, since the assumption proved to<br />

be nothing but dogmatic <strong>and</strong> even impossible a priori. Obviously, the author did not<br />

test his hypothesis (i, e. eq.<br />

(4f. 5)) by any means <strong>and</strong> this positively violates, con-<br />

trary to what was claimed, the code <strong>of</strong> rules <strong>for</strong> inferences by induction. The same<br />

criticism applies to the HASSELL-VARLEY model in w 4h.<br />

The above discussion suggests that the stage we are in is still very primitive,<br />

with an evident lack <strong>of</strong> rigor in methodology. This, however, may well be because<br />

the nature <strong>of</strong> the objects we are <strong>study</strong>ing have influenced the development <strong>of</strong> ideas<br />

in this field. My point may be illustrated by contrast with the development <strong>of</strong> the<br />

physical sciences.<br />

In physics, some properties <strong>of</strong> certain objects were, very <strong>for</strong>tunately, describable<br />

deterministically (sensu BORN 1964--predictable without the causal relationships being<br />

known; a timeless <strong>and</strong> spaceless link between the events, e.g. a railway time-table).<br />

The arithmetic prediction <strong>of</strong> the stars' motion by the Babylonians or, more recently,<br />

KEPLER'S Law, are perhaps typical examples. As modern physicists went into the<br />

more minute details <strong>of</strong> atoms, <strong>and</strong> as the required measurements became finer <strong>and</strong><br />

finer, they eventually reached a stage where the classical method <strong>of</strong> induction was no<br />

longer applicable. A positive barrier was encountered when HEISENBERG enunciated<br />

his Uncertainty Principle in 1927; this predicts that some physical attributes <strong>of</strong> the<br />

object being measured are influenced by interaction between the object <strong>and</strong> the meas-<br />

uring system. However, be<strong>for</strong>e this stage was reached, there were enough examples<br />

<strong>of</strong> success in macrophysics, i.e. in NEWTONIAN physics, which encouraged the phys-<br />

icists to explore thoroughly the method <strong>of</strong> induction.<br />

In the field <strong>of</strong> population dynamics, however, difficulties similar to those that<br />

modern physics is currently facing have been a major problem from the beginning.<br />

Some may be only technical difficulties in obtaining accurate measurements.<br />

example, the concept <strong>of</strong> the h<strong>and</strong>ling time (h), originally suggested by HOLLING<br />

(1956), was found to be highly idealized in my <strong>study</strong> <strong>of</strong> the great tit, Parus major<br />

L. (ROYAMA 1970). I tried to time the tit as it searched <strong>for</strong> food <strong>and</strong> as it h<strong>and</strong>led<br />

each item.<br />

The in<strong>for</strong>mation was used to calculate a theoretical value <strong>for</strong> the amount<br />

<strong>of</strong> food that the tit could collect per day using HOLLING'S disc equation (<strong>for</strong> the<br />

For


83<br />

justification <strong>of</strong> its use, see ROYAMA 1970). It was found that, <strong>for</strong> an intuitively rea-<br />

sonable magnitude <strong>for</strong> the factor a, the calculated value <strong>for</strong> h was ridiculously high.<br />

When the factor a was so adjusted as to obtain a more reasonable value <strong>for</strong> h, then<br />

such values <strong>of</strong> a were inexplicably low. My conclusion was there<strong>for</strong>e that the estima-<br />

tion <strong>of</strong> h by observation was far lower than it actually was. This is perhaps because<br />

what was recorded as searching time must have contained a high proportion that was<br />

spent upon various activities other than pure searching, e.g. watching <strong>for</strong> enemies.<br />

These activities must have occupied such short intervals that they were hardly separable<br />

by direct observation.<br />

Beside such difficulties in measuring each activity separately <strong>and</strong> accurately, there<br />

are more pr<strong>of</strong>ound ones which may not be solved technically. The first is the time<br />

factor. In order to take a sufficiently reliable measurement, say, <strong>of</strong> the fluctuation in<br />

numbers <strong>of</strong> an animal species from year to year, the life span <strong>of</strong> a single ecologist<br />

may not be sufficiently long: perhaps he can <strong>study</strong> only some twenty generations <strong>of</strong><br />

a univoltine species.<br />

From a mere accumulation <strong>of</strong> sampling data, he can draw<br />

conclusions by guessing, not by induction. The second difficulty lies in differences<br />

between natural <strong>and</strong> experimental situations. The development <strong>of</strong> <strong>comparative</strong> ethology<br />

in the past decade shows that the behaviour <strong>of</strong> animals has so evolved that its biologi-<br />

cal goal is attained by responding appropriately to a chain <strong>of</strong> stimuli provided in<br />

the animals' natural environment (see e.g. TINBERGEN 1951; <strong>for</strong> more recent develop-<br />

ment see HINDE 1966). We cannot be certain on the one h<strong>and</strong>, however, if some <strong>of</strong><br />

the necessary stimuli are lacking in an experimental set-up in which the animals<br />

concerned may not behave in an intelligible manner. But, on the other h<strong>and</strong>, we may<br />

not be able to know, without experimental studies, what stimuli are involved in the<br />

animals' normal environment where evolution has taken place. Then we will not be<br />

certain whether we can establish a fact from which to follow the <strong>for</strong>mula <strong>of</strong> induction<br />

laid out by the classical physics, or rather if a fact at h<strong>and</strong> is a meaningful one that<br />

can be used to start the gradual process <strong>of</strong> induction.<br />

These arguments may be metaphysical problems, but are sufficient to show that<br />

the primitive stage we are in at the moment, as compared with physical sciences, is<br />

perhaps due to the above-mentioned difficulties, which prevented us consciously or<br />

subconsciously from developing the method <strong>of</strong> inference by induction, starting from<br />

an elemental stage where a deterministic prediction would not have been hard to make.<br />

For the time being, ecology will perhaps remain largely descriptive, even if we<br />

cannot expect to develop a deterministic law governing the hunting behaviour <strong>of</strong><br />

animals directly from such enumerations. A theoretical <strong>study</strong> by <strong>models</strong>, i.e. infer-<br />

ence by analogy, will not replace the tedious process <strong>of</strong> enumeration, but it will at<br />

least partially help in the interpretation <strong>of</strong> observed facts. The method will be useful,<br />

however, only provided that the <strong>models</strong> are appropriately constructed <strong>and</strong> used. The<br />

idea is perhaps the same as that suggested by OPPENHEIMER (1956) to a group <strong>of</strong><br />

psychologists as 'pluralism' in the method.


84<br />

What is required at the present stage is to make use <strong>of</strong> the <strong>for</strong>malism <strong>of</strong> mathematics<br />

to a thorough extent in the quest <strong>of</strong> certitude. Some ecologists might suggest<br />

that the complexity <strong>of</strong> natural order almost prohibits this <strong>for</strong>malism. But this amounts<br />

to giving up the quest <strong>for</strong> certitude, because, if mathematics cannot h<strong>and</strong>le the complexity,<br />

it would be even more difficult <strong>for</strong> the verbal method <strong>of</strong> inference to cope<br />

with the problem, <strong>and</strong> we do not have a better alternative, at least at the present<br />

stage <strong>of</strong> the development in population dynamics.<br />

6. SUMMARY<br />

1. This is a critical <strong>study</strong> <strong>of</strong> major existing <strong>models</strong> <strong>for</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>,<br />

<strong>and</strong> its aim is to evaluate the roles played, or those not played, by these <strong>models</strong> in<br />

helping our insight into the relationships underlying prey-predator <strong>and</strong> host-parasite<br />

interaction systems.<br />

2. The concept <strong>of</strong> a model, <strong>and</strong> the role it plays in the process <strong>of</strong> reasoning that<br />

leads eventually to underst<strong>and</strong>ing <strong>of</strong> the system, is considered first in general terms.<br />

It is pointed out that a model is a means through which a<br />

hypothesis is <strong>for</strong>med;<br />

the model is an analogy to the thing to be understood, constructed from already<br />

known concepts.<br />

3. The basic structure <strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong> are then presented in terms <strong>of</strong><br />

mathematical equations <strong>of</strong> general <strong>for</strong>ms. This is to help readers to underst<strong>and</strong> the<br />

examination <strong>of</strong> existing <strong>models</strong> which follows in the subsequent sections. Two differ-<br />

ent sets <strong>of</strong> inferences are discussed, the one leading to a model <strong>for</strong> <strong>predation</strong> <strong>and</strong><br />

the other to a model <strong>for</strong> <strong>parasitism</strong>. The role played by a differential equation as a<br />

means <strong>of</strong> deduction is explained.<br />

4. Models proposed by LOTKA, VOLTERRA, NICHOLSON <strong>and</strong> BAILEY, HOLLING,<br />

IVLEV, GAUSE, ROYAMA, WATT, THOMPSON, STOY, <strong>and</strong> HASSELL <strong>and</strong> VARLEY are<br />

critically examined, with particular attention to their logical structures rather than<br />

to their ability to generate a theoretical trend that superficially fits an observed one.<br />

The logical structures <strong>of</strong> these <strong>models</strong> are summarized in the <strong>for</strong>m <strong>of</strong> diagrams in<br />

w 5. The critical <strong>study</strong> is developed by means <strong>of</strong> analogies to various geometric<br />

<strong>models</strong>, which clearly show some misconceptions involved in each model reviewed.<br />

It also shows that the logical structure in the classical <strong>models</strong> (those proposed be<strong>for</strong>e<br />

1935) are sound, but the assumptions involved are <strong>of</strong>ten inadequate to describe an<br />

actual biological system; on the other h<strong>and</strong>, recent <strong>models</strong> (proposed after 1955) either<br />

involve contradiction in logic or are much too <strong>of</strong>ten erroneously applied to the analysis<br />

<strong>of</strong> actual <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>. It is pointed out that the mere fitting <strong>of</strong> curves<br />

to data can neither establish a particular mechanism nor provide verification <strong>of</strong> the<br />

model concerned.<br />

5. Finally, the types <strong>of</strong> inferences, which can or cannot help us in underst<strong>and</strong>ing<br />

the processes <strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>, are considered by comparison to similar<br />

problems in the development <strong>of</strong> physical sciences. In ecology, inference by induction,


85<br />

which has been <strong>and</strong> Still is to a large extent the major scientific method, has inherently<br />

many <strong>of</strong> the same problems as modern physics, such as HEISENBERG'S Uncertainty<br />

Principle. Perhaps certain types <strong>of</strong> difficulty that are encountered by the observational<br />

method can be overcome, if not entirely, with the aid <strong>of</strong> inferences by analogy, <strong>and</strong><br />

it is in inferences where the model plays its role.<br />

ACKNOWLEDGEMENTS: I am grateful to the following people <strong>for</strong> their very constructive comments<br />

<strong>and</strong> suggestions which led to much greater clarity in my presentation: Drs. R.F. MORRIS,<br />

G.L. BASKERVILLE <strong>and</strong> M.M. NEILSON (Canadian Forestry Service, Fredericton), M.E. SOLOMON<br />

(University <strong>of</strong> Bristol), <strong>and</strong> Y. IT6 (National Institute <strong>of</strong> Agricultural Sciences, Tokyo). Pr<strong>of</strong>essor<br />

S. TANAKA'S (Ocean Research Institute, University <strong>of</strong> Tokyo) comments on my mathematical treatments<br />

helped to minimize erroneous arguments.<br />

REPERENCES<br />

ANDREWARTItA , H.G. <strong>and</strong> L.C. BIRcH (1954) The distribution <strong>and</strong> abundance<br />

Chicago Press, Chicago.<br />

<strong>of</strong> animals. Univ.<br />

BLiss, C.I. <strong>and</strong> R. A. FISHER (1953) Fitting the negative binomial distribution to biological data<br />

<strong>and</strong> note on the efficient fitting <strong>of</strong> the negative binomial. Biometrics 9 : 179-200.<br />

BORN, M. (1964) Natural philosophy <strong>of</strong> cause <strong>and</strong> chance. Dover, New York.<br />

BURNETT, T. (1956) Effects <strong>of</strong> natural temperatures on oviposition <strong>of</strong> various numbers <strong>of</strong> an insect<br />

parasite (Hymenoptera, Chalcidoidae, Tenthredinidae). Ann. Ent. Soc. Amer. 49:55-59.<br />

BURNETT, T. (1958) Effect <strong>of</strong> host distribution on the reproduction <strong>of</strong> Encarsia <strong>for</strong>mosa GAHAN<br />

(Hymenoptera : Chalcidoidae). Can. Ent. 90 : 179-191.<br />

BURNSIDE, W. (1928) Theory <strong>of</strong> probability (1959 ed.). Dover, New York.<br />

CHAeMA~, R.N. (1931) Animal ecology. McGraw-Hill, New York & London.<br />

CLAaK, P.J. (1956) Grouping in spatial distributions. Science 123 : 373-374.<br />

CLARK, P.J. <strong>and</strong> F. C. EVANS (1954) Distance to nearest neighbor as a measure <strong>of</strong> spatial relation-<br />

ship in populations. Ecology 35 : 445-453.<br />

DRAKE, S. (1957) Discoveries <strong>and</strong> opinions <strong>of</strong> GALILEO. Doubleday, New York.<br />

DUCASSE, C.J. (1960) FRANCIS BACON'S philosophy <strong>of</strong> science. Chapter 3 in "Theories <strong>of</strong> scientific<br />

method : Renaissance through the nineteenth century". Univ. Washington Press, Seattle.<br />

ELTON, C. (1935) Review <strong>of</strong> A. J. LOTKA (1934). ~ Anita. Ecol. 4 : 148-149.<br />

GAMOW, G. (1962) Gravity. Doubleday, New York.<br />

GAuss, G.F. (1934) The struggle <strong>for</strong> existence. Williams & Wilkins, Baltimore.<br />

GRIFFITnS, K.J. <strong>and</strong> C.S. HoLLInG (1969) A competition submodel <strong>for</strong> parasites <strong>and</strong> predators.<br />

Can. Ent. 101 : 785-818.<br />

HASSELL, M.P. <strong>and</strong> G. C. VARLEY (1969) New inductive population model <strong>for</strong> insect parasites <strong>and</strong><br />

its bearing on biological control. Nature 223 : 1133-1137.<br />

HAYNES, O.L. <strong>and</strong> P. SIsojEvtc (1966) Predatory behavior <strong>of</strong> Philodromus rufus WALCKENAER (Ara-<br />

neae : Thomisidae). Can. Ent. 98 : 113-133.<br />

HINDS, R.A. (1966) Animal Behavior. McGraw-Hill, New York.<br />

HOLLIN6, C.S. (1959a) The components <strong>of</strong> <strong>predation</strong> as revealed by a <strong>study</strong> <strong>of</strong> small mammal<br />

<strong>predation</strong> <strong>of</strong> the European pine sawfly. Can. Ent. 91 : 293-320.<br />

HOLLINC, C.S. (1959b) Some characteristics <strong>of</strong> simple types <strong>of</strong> <strong>predation</strong> <strong>and</strong> <strong>parasitism</strong>. Can.<br />

Ent. 91 : 385-398.


86<br />

HOLLING, C.S. (1961) Principles <strong>of</strong> insect <strong>predation</strong>. Ann. Rev. Ent. 6 : 163-182.<br />

HOLLING, C.S.<br />

(1966) The functional response <strong>of</strong> invertebrate predators to prey density. Mere.<br />

Ent. Soe. Can. 48:3-86.<br />

HUFFAKER, C.B. <strong>and</strong> C. E. KENNETT (1969) Some aspects <strong>of</strong> assessing efficiency <strong>of</strong> natural enemies.<br />

Can. Ent. 101 9 425-447.<br />

ITb, Y. (1963) DSbutsu Seitai-gaku Ny~mon (Introduction to animal ecology]. Kokin-shoin, Tokyo.<br />

IVLF.V, V.S.<br />

(1955) Experimental ecology <strong>of</strong> the feeding <strong>of</strong> fishes (English translation by D. SCOTT<br />

1961). Yale Univ. Press, New Haven.<br />

LAING, J. (1937) Host finding by insect parasites. I. Observations on the finding <strong>of</strong> hosts by<br />

Alysia m<strong>and</strong>ucator, Mormoneilla vitripennis, <strong>and</strong> Trichogramma evanescens. J. Anita. Ecol.<br />

6 : 298-317.<br />

LOTKA, A.J.<br />

MILLSR, C.A.<br />

(1925) Elements <strong>of</strong> physical biology. Williams & Wilkins, Baltimore.<br />

(1960) The interaction <strong>of</strong> the spruce budworm, Choristoneura fumiferana (CLsM.),<br />

<strong>and</strong> the parasite Glypta fumiferana (VIEs.). Can. Ent. 92 : 839-850.<br />

MILNE, A. (1957) Theories <strong>of</strong> natural control <strong>of</strong> insect populations.<br />

Quant. Biol. 22 : 253-267.<br />

MORRIS, R.F.<br />

Cold Spring Harbor Syrup.<br />

(1963) The effect <strong>of</strong> predator age <strong>and</strong> prey defence on the functional response <strong>of</strong><br />

Podisus maculiventris SAY to the density <strong>of</strong> Hyphantria cunea DRURY. Can. Ent. 95 : 1009-1020.<br />

MORISXTA, M. (1954) Estimation <strong>of</strong> population density by spacing method. Mere. Fac. Sci. Kyushu<br />

Univ. E 1 : 187-197.<br />

NICHOLSOS, A.J. (1933) The balance <strong>of</strong> animal populations. J. Anim. Ecol. 2 9 132-178.<br />

NmaoLsor~, A.J. <strong>and</strong> V. A. BAILEY (1935) The balance <strong>of</strong> animal populations. Part 1. Proc. Zool.<br />

Soc. London 1935, Part 3 : 551-598.<br />

OeeENamMsa, R. (1956) Analogy in science. Amer. Psychol. 11 : 127-135.<br />

PEARL, R. (1927) The growth <strong>of</strong> populations. Quart. Rev. Biol. 2 : 532-548.<br />

POINCARL H. (1952) Science <strong>and</strong> hypothesis (English ed.). Dover, New York.<br />

PogvA, G. (1955) Mathematics <strong>and</strong> plausible reasoning. Vol. 2. Patterns <strong>of</strong> plausible inference.<br />

Princeton Univ. Press, Princeton.<br />

ROYAMA, T. (1966) Mathematical <strong>models</strong> <strong>of</strong> predatory behaviour in population dynamics.<br />

Japanese.) Biol. Sci., Tokyo, 18 : 97-105.<br />

ROYAMA, T. (1969) A model <strong>for</strong> the global variation <strong>of</strong> clutch size in birds. Oikos 20 : 562-567.<br />

RO~rAMA, T.<br />

(1970) Factors governing the hunting behaviour <strong>and</strong> selection <strong>of</strong> food by the great<br />

tit, Parus major L. ]. Anita. Ecol. 39 : 619-668.<br />

(In<br />

RUSSELL, E.M.<br />

(1961) Soil conditions <strong>and</strong> plant growth (9th ed.) Longmans, Norwich.<br />

SCHON, D.A. (1967) Invention <strong>and</strong> the evolution <strong>of</strong> ideas. Butler & Tanner, Frome & London.<br />

SMITH, H.S. (1939) Insect populations in relations to biological control. Ecol Monogr. 9 : 311-320.<br />

SIMMONDS, F.J. (1943) Occurrence <strong>of</strong> super<strong>parasitism</strong> in Nemeritis canescens. Rev. Can. Biol.<br />

2 : 15-40.<br />

STov, R.H. (1932) Appendix to G. SALT (1932) [Super<strong>parasitism</strong> by Collyria calcitrator GRAV.]<br />

Bull. Ent. Res. 23 : 215-216.<br />

THOMPSON, W.R. (1924) La th6orie math6matique de Faction des parasites entomophages et le<br />

facteur du hassard. Ann. Fac. Sci. Marseille 2".69-89.<br />

THOMeSON, W.R. (1939) Biological control <strong>and</strong> the theories <strong>of</strong> the interaction <strong>of</strong> populations.<br />

Parasitology 31 : 299-388.


87<br />

TINBERGEN, L. <strong>and</strong> H. KLOMP (1960) The natural control <strong>of</strong> insects in pinewoods. II. Conditions<br />

<strong>for</strong> damping <strong>of</strong> Nicholson oscillations in parasite-host systems. Arch. ne~rl. 7.ooi. 13 : 344-379.<br />

TINBERGEN, N. (1951) The <strong>study</strong> <strong>of</strong> instinct. Clarendon Press, Ox<strong>for</strong>d.<br />

TORII, T. (1956) Konchfe-Shftdan no Suikei-h5 [Statistical analysis <strong>of</strong> insect populations]. Chapter 7<br />

in "Seitai-gaku Gaisetsu" [Outline <strong>of</strong> ecology,, (ed. by N. YAGI <strong>and</strong> K. NOMURA). Yoken-do,<br />

Tokyo.<br />

TOULMIN, S. (1961) Foresight <strong>and</strong> underst<strong>and</strong>ing. Harper (Torchbook ed. 1963), New York & Evanston.<br />

ULLY~TT, G.C. (1947) Mortality factors in populations <strong>of</strong> Plutella maculipennis CURTIS (Tineidae :<br />

Lep.), <strong>and</strong> their relation to the problem <strong>of</strong> control. Ent. Metal, Dep. Agr. Forest., Union <strong>of</strong><br />

South Africa 2 (part 6) : pp. 202.<br />

ULLY~TT, G.C. (1949a) Distribution <strong>of</strong> progeny by Chelonus texanus CRESS. (Hymenoptera : Braconidae).<br />

Can. lent. 81:25-44.<br />

ULLYETT, G.C. (1949b) Distribution <strong>of</strong> progeny by Cryptus inornattus PRATT (Hymenoptera : Ichneumonidae).<br />

Can. Ent. 81:285-299.<br />

VBRnURST, P.E. (1838) Notice sur la loi que la population suit dans son accroissement. Corresp.<br />

Math. Phys. 10 : 113-121.<br />

VOLT~RRA, V. (1926) Variazioni e fluttuazioni del numero d'individui in specie animali conviventi.<br />

Mere. R. Accad. Naz. dei Lincei, Ser. VI, Vol. 2 [English translation in CHAeMAr~ 1931].<br />

WALKER, M. (1963) The nature <strong>of</strong> scientific thought. Prentice-Hall, New Jersey.<br />

WATT, K. E.F. (1959) A mathematical model <strong>for</strong> the effect <strong>of</strong> densities <strong>of</strong> attacked <strong>and</strong> attacking<br />

species on the number attacked. Can. Ent. 91 : 129-144.<br />

WATT, K. E.F. (1961) Mathematical <strong>models</strong> <strong>for</strong> use in insect pest control. Can. Ent. Suppl.<br />

19 : 1-62.<br />

WATT, K.E.F. (1962) Use <strong>of</strong> mathematics in population ecology. Ann. Rev. Ent. 7:243-260.<br />

WATT, K. E.F. (1968) Ecology <strong>and</strong> resource management. McGraw-Hill, New York.<br />

WILLIAMS, C.B. (1964) Patterns in the balance <strong>of</strong> nature <strong>and</strong> related problems in quantitative<br />

ecology. Academic Press, London & New York.


88 84<br />

Appendix 1. The pro<strong>of</strong> <strong>of</strong> LI=I/2RX :<br />

Suppose particles <strong>of</strong> kind P are distributed at r<strong>and</strong>om with density X over a two-dimensional<br />

plane <strong>of</strong> sufficiently large area. Let l be the distance between Q's starting point A <strong>and</strong> the periph-<br />

ery <strong>of</strong> the circle <strong>of</strong> radius R around P that Q first encountered from outside the circle when<br />

moving in an arbitrarily determined direction. Suppose that l comprises very small sections <strong>of</strong><br />

length ,~l, <strong>and</strong> that l~ is defined by izll (i=1, 2 .....<br />

c~o). Let ~7l be the probability that at least<br />

one particle <strong>of</strong> kind P falls within, <strong>and</strong> only within, a locus (li, li+dl). Then the average distance<br />

that Q travels from outside to encounter the periphery <strong>of</strong> the first, i.e. L~, is<br />

oo<br />

Lt=lim ~2 IpT~<br />

dl~O i=l<br />

Now, ~2~ is the product <strong>of</strong> the following probabilities:<br />

(1) probability that at least one particle <strong>of</strong> kind P is found within area 2R(li+zff), which <strong>for</strong><br />

the assumption <strong>of</strong> the Po~ssos distribution is, <strong>for</strong> j=l, 2 .....<br />

~,, [ {2R(l~+dl) X} J/j !] e -2R(I,+aI)X,<br />

j-1<br />

(2) probability that all <strong>of</strong> these j particles within area 2R(li+dl) are found within locus (li, li<br />

+dl), i.e.<br />

{2RJ1/2R (l~ +all) } ~.<br />

Thus,<br />

~= ~ [ (2R (li + AI) X} J/j !]e-2R(h + ~l)X {2Rdl/2R (li + Jl) }<br />

cr<br />

(i).<br />

= ~, { (2RJIX) J/j !} e<br />

j=l<br />

2RJIX- 2RI~X<br />

=(1-e<br />

- 2RdlX - 2RliX<br />

Since, by a theorem in the theory <strong>of</strong> limit,<br />

) e<br />

lira (1 - e- 2R~IX) ~all = 2 RX,<br />

Jl~0<br />

<strong>and</strong> so, by writing l~ simply as l, we have<br />

Hence, from eq. (i),<br />

lira ~7~ = 2RXe- 2RIX dl.<br />

6l ~0<br />

L~=2RX f :Ie -2RlX dl<br />

=I/2RX.<br />

Appendix 2. The pro<strong>of</strong> <strong>of</strong> L2={O(V'~XR)/1/X-Re -TrR2X }/(1-e -~rR~X )<br />

on a two dimensional plane<br />

Suppose that the radius R consists <strong>of</strong> a number <strong>of</strong> very small sections <strong>of</strong> length dr, <strong>and</strong> that<br />

ri is defined as idr(i=l, 2 ..... oo). Let tti be the probability <strong>of</strong> at least one point <strong>of</strong> A's falling<br />

within, <strong>and</strong> only within, a locus (ri, ri+dr). Then the average distance between the centre <strong>of</strong> a<br />

circle <strong>of</strong> radius R <strong>and</strong> the nearest A within the circle, i.e. L2, is<br />

oo<br />

co<br />

Lz=limar~o { i=5-2'1 r~pd / { ~lt~}~<br />

Now, pi is the product <strong>of</strong> the following probabilities:<br />

(ii).


(1)<br />

assumption <strong>of</strong> a PoxssoN distribution,<br />

~E [ {z (r, + Jr)'Aq s~ j!] e - ~'(' '+ dr)2X ,<br />

j~l<br />

(2) probability that all <strong>of</strong> these j points fall within a locus (ri, ri+dr), i.e.<br />

[ {z (r~ +dr) 2_ r, ri~}/~r (ri +dr) 2],,.<br />

With a similar calculation as <strong>for</strong> 7] in Appendix 1, we have<br />

Thus,<br />

Similarly,<br />

probability that at least one point <strong>of</strong> A's falls within radius ri+dr; this is, under the<br />

l. - wr~.X<br />

~m ta~=2~rXe dr.<br />

d r--*O<br />

co PR - *rr2X<br />

lira i=<br />

J r--~0 '=<br />

a~lFi=|o 2r, rXe dr<br />

=l_e-~rR~X<br />

oo PR ~ -wrO-X<br />

lira ]E rq2~=l 2.'rr Xe dr<br />

zIr~O i=1 dO<br />

<strong>and</strong> integrating by parts,<br />

(iii).<br />

fR2~,r,e-,rrO~X, ar=Jo ['R-wr~<br />

e ' dr-Re -'R~'x (iv).<br />

Let O(t) be the normal probability function, i.e.<br />

1 t - t 2/2<br />

$(t) =. 7,-- f e dt,<br />

V~Tr JO<br />

<strong>and</strong> setting t equal to l/2zXr,<br />

9 (1/~R)/V'X.<br />

the integral in the right-h<strong>and</strong> side <strong>of</strong> eq. (iv) will be written as<br />

Thus, from eqs. (ii), (iii), <strong>and</strong> (iv) <strong>and</strong> using symbol ~, we have<br />

L.=[, )<br />

89<br />

Appendix 3. The pro<strong>of</strong> <strong>of</strong> eq. (4i. 6)<br />

Let Y be the density <strong>of</strong> particles distributed at r<strong>and</strong>om over an area A, each one <strong>of</strong> the<br />

particles having a circle <strong>of</strong> area 6 around it. Then the density <strong>of</strong> the particles, Y", within the<br />

total area covered with these circles, i.e. A', is<br />

Y"-=-AY/A'<br />

Suppose that a large number <strong>of</strong> points, i.e. U, with circles <strong>of</strong>, also, area ~ are placed to<br />

cover the whole area <strong>of</strong> A independently <strong>of</strong> the distribution <strong>of</strong> the particles. Then, the number<br />

U' <strong>of</strong> these points that include at least one <strong>of</strong> the particles will he, on the assumption <strong>of</strong> the<br />

Poissos distribution,<br />

U'= U(1-e -~r)<br />

(vi),<br />

<strong>and</strong> the proportion U/U ~ must be equal to the proportion A/A'. Thus from eqs. (v <strong>and</strong> vi), we<br />

find<br />

Y"=Y/(1-e-~Y).<br />

Thus a circle <strong>of</strong> area c~ around each particle contains on the average 6Y~P number <strong>of</strong> particles.<br />

However, since ~Y~ is, as defined in w 4i, the mean number <strong>of</strong> particles in each circle less<br />

the one at the centre, we have<br />

6Y'=6Y"--I<br />

=~Y/(1-e -~r) -1.


90<br />

Appendix 4.<br />

List <strong>of</strong> symbols<br />

The following symbols with the same meaning appear in more than two sections <strong>and</strong> in the<br />

flow diagram <strong>of</strong> Fig. 13. Sections indicated in the parentheses are the places where definitions<br />

are given. Symbols used in one section only, or those defined each time they appear, are not<br />

listed.<br />

Variables :<br />

x (w Prey or host density.<br />

X (//) Fixed prey or host density during t. (For t see below. )<br />

x0(H) Initial prey density when /-=-0.<br />

y (H) Predator or parasite density.<br />

Y (/t) Fixed predator or parasite density during t.<br />

t (tt) Interval <strong>of</strong> a hunting period.<br />

(It) Number <strong>of</strong> prey taken, or number <strong>of</strong> parasite eggs laid, per unit area, when the density<br />

<strong>of</strong> the hunted species is fixed during t.<br />

z (tt) Number <strong>of</strong> prey taken, or number <strong>of</strong> hosts parasitized, per unit area during t, when<br />

the density <strong>of</strong> the hunted species is not replenished.<br />

L(w 4c) Time spent in searching only.<br />

:Functional symbols :<br />

f (w 3) Instantaneous hunting function.<br />

F ( tt ) Overall hunting function.<br />

gl(w 4a) Function characterizing the instantaneous rate <strong>of</strong> increase (or decrease) <strong>of</strong> prey population<br />

in the abscense <strong>of</strong> predators.<br />

g, (M) Characterizing the instantaneous rate <strong>of</strong> increase (or decrease) <strong>of</strong> predator population<br />

in the presence <strong>of</strong> food species.<br />

S(w Characterizing the degree <strong>of</strong> social interaction (interference or facilitation) among<br />

predators or parasites by which the f changes.<br />

H( ~t ) Characterizing the partial realization <strong>of</strong> the potential per<strong>for</strong>mance in hunting in accordance<br />

with the degree <strong>of</strong> hunger or satiation.<br />

(tt) Probability <strong>of</strong> a host receiving no parasite egg.<br />

:Factors independent <strong>of</strong> the variables listed above :<br />

a (w 4b) Effective area <strong>of</strong> recognition.<br />

b (w 4d) Positive proportionality factor.<br />

c (zt) Positive proportionality factor.<br />

/~ (w 4f) Coefficient <strong>of</strong> social interaction. (See also w 4h. )<br />

r(w 4a) Coefficient <strong>of</strong> increase in prey population in the absense <strong>of</strong> predators.<br />

r t (//) Coefficient <strong>of</strong> decrease in predator population in the absense <strong>of</strong> food species.<br />

a'(#) Coefficient <strong>of</strong> increase in predator population due to feeding.<br />

h (w 4c) Time spent in h<strong>and</strong>ling an individual prey or host.<br />

M(w 4g) Size <strong>of</strong> hunting area.<br />

Other parameters :<br />

~(w 4b) 'Area <strong>of</strong> discovery' defined as in eqs. (4b. 9) <strong>and</strong> (4b. 10).<br />

k (w 4g) Factor characterizing the degree <strong>of</strong> aggregation in the negative binomial distribution.<br />

(The symbol k used in w w 4c <strong>and</strong> d st<strong>and</strong>s <strong>for</strong> the frequency <strong>of</strong> tapping fingers or that<br />

<strong>of</strong> tossing rings. )


91<br />

4. LOTKA, VOLTERRA, NICHOLSON ~ BAILEY, HOLLING, IVLEV, GAUSE, ~JJ, WATT, THOMPSON, STOY,<br />

~ , ~ ~'~:~i~~.l:l;~b~'~:~6~bt: P__ ~8, ~A.~~ ~, ~/~6~o~

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

Saved successfully!

Ooh no, something went wrong!