This action might not be possible to undo. Are you sure you want to continue?
Udny Yule Source: Philosophical Transactions of the Royal Society of London. Series B, Containing Papers of a Biological Character, Vol. 213 (1925), pp. 2187 Published by: The Royal Society Stable URL: http://www.jstor.org/stable/92117 Accessed: 05/09/2010 19:04
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, noncommercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/action/showPublisher?publisherCode=rsl. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a notforprofit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.
The Royal Society is collaborating with JSTOR to digitize, preserve and extend access to Philosophical Transactions of the Royal Society of London. Series B, Containing Papers of a Biological Character.
http://www.jstor.org
[ 21 ]
II.A
Mathematical basedon the Conclusions of of Theory Evolution, Dr. J. C. Willis, F.R.S.
By G. UDNY YULE, C.B.E., M.A., F.KRS., UniversityLecturerin Statistics, and Fellow of St. John's College,Cambridge.
(Received June 11, 1923.Read February 7, 1924.)
CONTENTS, Page Introduction . ..... . ..................... 21 Section. I.The increase in the number of species within the genus with time, and the frequency distribution of sizes of genera all of the same age 33 .... ............ II.The limiting form of the frequency distribution for the sizes of derived genera when time is infinite .... ............... 37 III.The frequency distribution for sizes of genera (derived and primordial) when time is finite 41 IV.The fitting to data of the expression obtained in Section III ............. 51 V.The frequency distribution of ages and the mean age, etc., for genera of a given size . . 64 VI.An attempt to estimate the order of magnitude of the doublingperiod for species in the case of the flowering plants, and the present rate of occurrence of specific mutations 75 . APPENDIX.Tables AE. ................................. 85 INTRODUCTION.
The following work is founded on that conception of evolution, the most recent and and precise formulation of which is due to Dr. J. C. WILLIS,* represents an attempt to the quantitative consequences of the conception. develop By his statistical studies of distributionDr. WILLISwas led to two conclusions:(1) Species occupying large areas are, on the whole,older than those occupying small areas, provided that allied forms are compared. It is to be noted that by the " area occupied" by a species is meant, not the total contour drawn round its outermost stationsthe " areal range" of the species as it might be termed. (2) The larger genera, i.e., the genera with most species, are, on the whole, those occupying the larger areas. "Age and Area,"the second as the principleof " Size and Space." From these statistical inductions the deduction follows, time being the ruling factor in both cases : (3) The larger genera are, on the whole,the older.
'Age and Area,' Cambridge University Press, 1922, and numerous papers therein cited published during the past fifteen years. VOL. CCXIII.B. 403. [PublishedMay 14, 1924.] E *
acreage actually covered by individuals of the species in question, but the area of the
The first is the conclusion which Dr. WILLISbriefly summarises as the principle of
22
. MIR GT. DNY YULE ON A MATHEMATICAL THEOlY OF EVOLUTION,
The size of the genus, that is to say, is not an absolute rnmeasure its age but is an of index to it, very much as the total numltber children born to a imarriageis an index to of the duration of that mlarriage. On the Darwinian view that species are continually dying outthat a species rises, flourishes and dies, superseded by the more advantageous forma species occupying a very small area may be young, but it is equally likely or more likely to be old (a dying species). On Darwin's own view that the whole body of individuals in a speciesbecomealtered together,* the young, species must be found occupying a large area at once, and the species occupying a small area could only be a " dying " species. On the Darwinian view therefore either there need be no relation between Age and Area, or there would be a negative relation, species occupying small areas being on the whole the oldest. Similarly,on the Darwinianview a genus of a few, or of only one, species may be either young or olda dying genusand there need be no necessaryrelation between Age and Size. That species occupying very small areas, and the species of monotypic genera are mainly " relic " forms, is, I gather, the predominantDarwinian view. Dr. WILLIS's conclusions are inconsistent with that view. We are accordingly led directly to the mutational view of evolution that has been held by more than one writer both before Darwin and after : the view that specificdifferences arise, not cumulatively by the natural selectionof slight favourablevariations,but at once per saltum as " mutations." On this view a new form must necessarilyoccupy a small area, and the required correlationbetween age and area follows at once. The new form may possibly differso largely from the parent stock as to be classed not merely in a new species, but even in a new genus or a new family. If the new form is badly unsuited to its immediate surroundings,it will be killed out at once; natural selection will strangle the species at birth. But if the species hold out long enough to spread over any considerablerange it is highly unlikely to be extinguished by natural selection, for it has a chance to find the niches that suit it best. The operation of natural selection, on Dr. WILLIS'S view, is not denied: what is denied is the origin of species by natural selection,and what is affirmedis that natural selection operates mainly on the very young species before it has time to spread. The species that is killed out at birth we may regard as a nonviable mutation, and nonviable mutations may be to all intents and purposesdisregarded: they are not likely to be seen in life or to be found in the geological record. When we speak in future therefore of a " specific mutation " or of a " generic mutation " it may be understood that a viable mutation is implied. If attention is
confined to viable mutations, we have on the present view little or no further concern with the extinction of species by natural selection, or indeed with any extinction of species, so long as conditions are constant. It seems doubtful, at the least, whether
we have any reason to predicate death as normal for a species in the same sense that it may be normal for one of the higher plants or the higher animals.
" More Letters of Charles * Letter to G. Bentham, November Darwin," vol. I, 379, 25, 1869:
by agencies which act less continuously than spasmodically and may fairly be described as cataclysmic.BASED ON THE CONCLUSIONS OF DR. not a deluge alone. we may consider "free" or undisturbed evolution as proceeding without any species being killed out or " dying " out.000 years will appear to be absolutely sudden. a change say from temperate to glacial or from glacial to temperate accomplished in 10 or 20. with some implication. Such "disasters " as I have in mind are certainly extensive in their range. Only in so far as extermination was not complete there would be selective action." i. WILLIS.land under the sea. 23 Hlowthen. the sinking of. This is the order of time estimated to have elapsed since the culnination of the last glacial epoch.. it is naturally asked. an " accident " may happen that brings about a (viable) " specific mutation. the throwing of a new form which is regarded as a new species.R. sense.e. Now a "cataclysm" in the sense explained would kill out the whole or a great part of the organic life existing in the region over which it swept. C. the number of species continually increasingwithout a break as mutations occur.S. F. for some only of all the species in the world would lie within its range. But " cataclysm" has now come to carry the meaning of any overwhelming disaster. and selective of already existing species. and if changes of climate are shown by a curve to such a scale. or other great change of climate. but a species within the same genus as the parent. but the term "cataclysm " seems to me quite applicable. the assumptions made for the mathematical work may be stated as follows:(1) Within any species. derivative. On such a scale onehundredth of an inch represents 10. but it would not act selectivelyif the cataclysm was overwhelming and the extermination complete: the species exterminated would be killed out not because of any inherent defects but simply because they bad the illluck to stand in the path of the cataclysm. the age of the flowering plants is of the order of 100 million years (cf. The chance of this occurrencein any assigned interval of time (an hour. If we then put aside for the moment " cataclysmic " destruction of species. but on the geologicaltimescale they may well be regardedas sudden. I think. They are not " sudden" on a scale of time on which a year is regardedas a long period. . of the disaster being sudden and extensive. in the main. But even so. such as desiccation. the selection would be only interspecific.g. the onset of a glacial epoch. in any interval of time. e. In the case of land sinking below sealevel it applies in the literal.000 years. I am not for a moment denying uniformitarianism or demanding catastrophism. do you account for the number of extinct species in the geological record ? The answer is by their having been killed out. J. as it seems to me. It would necessarily act different ally. With this brief introduction. Suppose this length of time to be shown on a scale 100 inches long. Section VI). or a year or a century) is taken as the same for all species within the group consideredand as constant for all time. the species surviving being on the whole the fitter to survive in the new circumstances.. So far as I can judge from the evidence.
owing to the increase in the number of species. but was deliberate. a new form so differentfrom the parent that it will be placed in a new genus. 11must towards greater and greater numbers of species and fewer individuals in each. for example. but only to the total numberof individuals. This assumption may or may not be correct. The possibility is ignored that A. the limited space available. (a) The assumptions that the chances of specific (or generic) mutation are identical for all forms within the group considered and constant for all time are unlikely to be in accordancewith the facts. B.* Possibly the fact that ignorationof the number of individualsleads to results in close accordancewith the data may only indicate this tendency first to increase and then to decreasein the size of the species. G. (c) The possible effect of size of genus (numberof species in the genus) on the chance of a generic mutation is also ignored. I see no practicallimit to the numberof species. but have to be made to simplify the work.e. These statements call for some comment. The chance of this occurrencein any assigned interval of time is similarly taken as the same for all genera within the group consideredand as constant for all time. be The area availablebeing limited. But over the very long periods which have to be consideredthere must be a countervailingtendency to ultimate decrease in the number of individuals. (2) Within any genus.and there must be dependenceif the " accident" that brings about a mutation is of the nature of a breakdownin the mechanismwhich forms the germcells. as the species spreads. that is the " size of the species. D are the existing genera and one of them throws a generic mutation. (b) The chance of a generic mutation or a specific mutation occurringis not taken as dependent on the number of individuals in the genus or species. an " accident " may happen that brings about the throwing of a (viable) " generic mutation.'the point was made that indefinite increase in the number of species was impossible owing to.. may throw a generic mutation which will be classed under the already existing genus B. C. though there very probablyis such dependence. as it seems to me." i. and the chance of occurrenceof a transferfrom one main position of stability to anoth'eris regarded as independent of the number of minor positions of stability (species) which may have been taken up within the main position (genus). and that is all that can be said. But in any case. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. 1922. the number of individuals must tend to increase. Again this is an assumption based on complete ignorance: it seems the simplest assumptionthat one can make. At first.24 MR. and if my conclusions at * During the discussion on Darwinismn the HIullmeeting of the British Association. . It cannot be said with certainty whether it is so dependent or no till we know the nature of a mutation. the tendency. We are completely ignorant as to the nature of specific mutations."'would be of the age of the species. The generic characters are regarded as representing a main position of stability. (d) If A. I was comrpelledto ignore the numiber individuals as I could feel of no confidence in any assumption as to the function which number of individuals. in any interval of time. it is assumed that this will representa new genus E. and apart from the great additional complexity it would introduce.
each belonging to a differentgenus but similarin so far that we can take the chance of a specificmutation occurring as the same for all (cf.S. above. how many of them have become ditypic. The whole scheme has therefore. the assumptions are probablycorrect.. starting at zero time from one primordialspecies. Specific mutations may occur much more frequently within one group than another. In Section I the question is considered.e. after any the given time has elapsed? It is not enoughto say that after one doublingperiod average size of a genus will be 2 speciess: we want to known how many of the genera have remained monotypic. at time 3 there will be 8 species. at time 4 there will be 16 species and so on. In so far as we find agreement. of course. It is only by a full development of the consequencesthat an effective test of the assumptions can be made.BASED ON THE CONCLUSIONS OF DR. or the more nearly we find agreement. what is the average law of increase of the number of species within the genus. 3 species may be expected to form a descending geometricseries (i. the assumptionsare probably incorrect or incomplete. and simply exhibit their consequences. J. The conclusions result. and the doubling period will then be much shorter in the first group than in the second. in the two groups. been simplified to the greatest possible extent to enable us to get a first notion. F. doubling in successive equal intervals of time. p. on an average. the number of species (that is the size of the genus) increasesin geometric progression if at zero time there is 1 species and at time 1 there are 2 species. Within any single genus. but if in each case we take the doublingperiodas the unit of time matters are reduced to the same relative scale and will proceed pari passu. and so on. The answer obtained is that. and to compare them as far as possible with the facts. In most of the following work. In so far as the deductions do not agree with the known facts. WILLIS. consequently. a series with a common ratio less than unity). what is the " frequencydistribution. the increasein the number of species will. which all started as monotypic genera from primordial species at zero time. at time 2 there will be 4 species. 25 as to their extreme rarity are correctI do not think we are at all likely in any near future to obtain direct information as to their nature. on an average."as the statistician terms it. the doublingperiodfor species within the genus is taken as the unit of time and it must be rememberedthat in terms of years it varies from group to group. I think justifiably. 5 species. This law is important: for it suggests the " doublingperiod species within the genus " for as a natural unit of time in investigations of the present kind. But the above law is only an average. proceed with complete irregularity.R. I will now endeavour to summarise the conclusions reached in general terms which I hope may be comprehensibleto the nonmathematical biologist. C.the occurrenceof specific mutations being assumed a matter of chance. 2. of course. the number of species. of the sizes of these N genera. how many have become genera of 3. from the assumptions. 23). The further question therefore arises. 4. But if we suppose some large numberN of primordialspecies. in quantitative terms. After one doubling . The answer obtained to this question is that the proportionalfrequenciesof genera of 1. to start together at zero time. of the possible consequencesof the conceptions. we can law then ask what is the average of increase.
pp. generic mutations are ignored. the table concluding with the limiting form of distribution when an infinite time has elapsed. and so on. p. and the form of the frequency distribution for size of genus is shown at times (expressedin terms of doublingperiodsfor species within the genus) 1. 4. UDNY YULE ON A MIATTHEMATICAL THEORY OF EVOLUTION. Evolution is assumed to start as before at zero time. to proceed to the much more general problem. as the figures are rounded off to the nearest unit they are necessarily subject to abrupt variation in the tail of the distribution. 3. 56. the numbersgraduallytailing away as the size of genusis increased.000. which are all precisely of the same age. common ratio being 3. Clearlywe are not in a position to compare this deduction with the facts. the common and so on.and any observed distribution will include both the derived and the primordialgenera : what will be the form of the distributionof size of genus in the resulting aggregate? This question is considered in two successive sections of the paper. therefore. let alone a group of similar genera.000 of all genera have 17 species or more. three or more doublingperiodshave elapsed.000 of all genera have 9 species or more.tritypic generaless frequent still. 5. V. and there is a considerable proportion of the larger genera such as we seem more usually to find.at its value in an actual case (the Chrysomelidbeetles. and only some 3 per 1. 1. attention being fixed on the primordial genera alone.26 MR. Besides the time. To render the distributions comparable. for we cannot pick out any group of genera. namely 1 925. and Appendix. In the above. the ratio being : after two doublingperiodsthe series should be a. and so on. Thus after one the doublingperiod percentageof monotypes is as high as 57. In Section II the limiting form is found when the time elapsed from the beginningof evolution is taken as infinite. The mathematical results here are complicated: I would refer the nonmathemnatical reader in the first place to Table II. 43. Table A. 85). 2. Those who are familiar with tables given in the forms of distributionfound in practice.ditypic generaless frequent. i. . or will refer to the typical]. and then in Section III the practical problemis solved for finite time. namely the ratio p of the chance of a specific mutation occurringin a given small interval of time to the chance of a generic mutation occurringin the same small interval of time: this has been taken. Tables IV.the number of genera at each stage has been taken as 1. G. it is necessary to fix one other quantity in order to determine the distribution. After two doublingperiodsthe percentage of monotypes has fallen to 42 and some 5 per 1. . will see that the distributions of Table II are at least of the right form so far as the eye can judge : monotypic genea are the most frequent. and the fully characteristicform (at least for such tables as those given in the Appendix) is not reached until two. But during the long lapse of tilme generic mutations as well as specific mutations will be thfrown. and 6 28 (the time found for a certain case below).1. interval the frequencies should be proportionalto . We have. After three doublingperiodsthe percentage of mnootypes has fallen further to 37. p. 54. the Appendix. if a is the first tern of the series the common ratio must be (la) to make the sum of the series unity. The length of this " tail "grows very rapidlywith time. where a series of calculated distributions is given.
If we form a chart in which the number of genera of a given size is plotted vertically and the size of the genus horizontally. sometimes even up to genera of 100 species or more. The numbers of monotypic genera observed and calculated must agree within a decimal point or so owing to the method of fitting : but I think the readerwho studies 'rablesV to VIII will admit that the agreement between observation and calculation is throughout extraordinarily close. as well as to the adjacent explanatory ' text. So far as the graphic test goes. and No in Table IV (p. Here I would only direct attention to the rather large number of primordial genera found in each case (Table IV. In Section IV the test of agreementbetween theory and fact on this point is completed by fitting a calculated distribution to the actual distribution in four cases. and the ratio p of the chance of a specific mutation to the chanceof a generic mutation: we also require the initial number No of primordialgenera. we require to determine from it the two constants that determine its form. but as time is increased the graph soon takes on the form noted for actual data. 4 to 9. now. two others will be found on pp. It is in fact better than one has any right to expect. when time is very shortno more than one or two doublingperiodsthe graphdropsaway rapidly. and one primordial genus would only yield n genera after the given lapse of time. It will be seen that at first. at first nearly straight and then falling away rather abruptly. usually up to genera of 30 species or so. * If there are N genera in the data.S. could hardly be better. that is scales on which numbers that bear equalratios to each other (like 1. 1. To fit a given distribution. the resulting points in any actual case run rather irregularlybut fairly closely round a straight line. line 6). 27 The mlathematicalform of the distribution may be better illustrated graphically.' If. 4. C. p. 54): to the comparatively limited range of values of . pp. 2 and 3 (pp. the time T (in doublingperiods)elapsed since the commencement. 5658) and the values of r. the results of this test.R.R.(4 26 to 628. which determinesthe total number of genera existing. not to ordinary scales but to logarithmic or ratio scales. I admit very considerabledifficulties of interpretationand would refer to the discussion on pp. 4749). ibid.188 to 1 925. similar charts are drawn for the calculated distributions of Table II they will be found to run as in figs.BASED ON THE. 54). accordingly. after which the points fall rather abruptly away from the line.the theory gives very well indeed precisely the form of distribution required. F. 2412 of Age and Area. ibid. J. to which the readershouldrefer. line 5): and to the comparatively limited range also of the values of p (1. 45. 16) stand at equal distances apart. 8. 2. For the longest time considered. 46). on the one point on which direct comparisoncan be made with the facts. Subject to the admitted difficulties of interpretation. . CONCLUSIO(NS OF1 D. we must have No N/n.* The readerwill find the results in TablesV to VIII (pp. line 7. Three specimen charts so drawnare given in figs. 5862. 628 doublingperiods. the point at which the abrupt fall begins lies outside the chart on the right. WtILLIS. The first two constants are found from the proportionof monotypes and the mean size of genus in the data: No is then given by the number of species (or genera)in the data. p.
of course. giving an increase of 235 units of age for an increase of 9 units in the size of the genus. but the mode and mean are thrown further and further towards the right. The next point considered (Section' V) is the frequency distribution of ages for genera of a given size given. species. that the increase in the mnean. 66. The reader should note this approximate result and also how very largely the successive distributions of fig. though the average age is. 64. or an increment of 0 92 unit of age. etc. in this case 0 68 units as shown in Table IX. if we increase the number of species in the genus from s to s" we must multiply the mean age Thus the mean age of genera of 100 species by a (or a value very near it). 1 to 10 species after an infinite time has elapsed from the beginof ning of evolution: p is taken as I 5. 10. and the unit of the scale of time is the doublings period for species within the genus as usual. The nonmathematicalreader will do best to turn first to fig. since 100 is 102. 10 species and of 056 unit of age. species or some other series in geometric progression.. etc. The means are tabulated in colunmn 2 of Table IX for genera of 1 to 10 species. For monotypic genera the curve is a simple logarithmic curve.. (6 40) is not 10 times but barely twice the mean age of genera of 10 species (322). For the tritypic genera or genera of three species the form changes slightly again. The curverises abruptly ' from zero to a maximum at a certain " modal (most frequent) age. however. 8. we require to know how many of them are likely to be of each successive age. of genera of 10 species 3. 1120. G. 128. 16. But a further increase of 10 units in the size of the genus only gives a mean age of 414 units. 67. It will be noticed. p. something greater. falling way steadily from age zero: the most probable age of a monotypic genus is zero. Apart from a further complication introduced when time is limited and not infinite. and so on.28 MR. p. is 0 87 of a unit. and then falls away to a long tail. and thence by intervals of 10 to genera of 100 species. (1) we should group the genera. say. and the next increment of 10 units in size of genus only gives an increrent For the larger genera of. (2) we must have as many genera as possible in each group in order to average out their varying ages.e. UDNY YULE ON A MATHErMAI OF TICAL THErEORY EVOLTTTION. For larger genera the form of the distribution remains the same.n is the distribution for genera of 10 species as compared with the distribution for genera of 9 species. the curve not rising from zero abruptly but starting tangentially to the base. 10 overlap each other. and these bring out the point very clearly. 4. over. in the given case. showing the limiting formsof these distributionsfor genera. For the ditypic generathe form of the distribution is quite different. they suggest that.22 units. 2130. not in groups of 110. and even so (3) we must expect considerable fluctuations of sampling as the ages . which is at first fairly rapid. if we want to compare say areas occupied by genera of different sizes. The maximum frequency is reached at a mode of 114 units and then falls away again.. N existing genera all of the same size. The mean age of monotypic genera. increasingwith the size of the genus. 32. but in gToups round 2. the mean age varies nearly as the logarithm of the size : i. that is to say. gets slower and slower: the distribution for geniera of 4 species is evidently pushed more to the right as compared with the distribution for genera of 3 species tha.
Turningfirst to Table X (p. 11.the genera really form two distinct groups: (a) the derived genera. The proportion of primordialgenera is shown by a square to the right of each distribution. but rather more scattered. then in columns 5 and 6 the proportions of derived and primordial genera respectively (or if we prefer to read it so. Table X and fig. 10. and finally in column 7 the mean age of the derived and primordial genera together. and the greater therefore is the probability that it may be a primordialgenus. 29 of genera even of one and the same size differ largely inter se. Fig.9961 X 0. using the constants found for the Chrysomelid beetles (TableV. is drawn. 71. and a proportion 0 0039 of primordial genera all of age 6. columns 4 to 7 give the essential figures: first.BASED ON THE CONCLUSIONS OF DR. matters are not quite so simple. the greater is its probable age.R.94) + (0. 56). and as we have seen the four illustrations taken show times varying only from 4 *26 to 6 28 doublingperiods. say r. (b) the primordial genera. When only a limited time has elapsed since the commencementof evolution. found from the figures of the three preceding columns. fig. for which the frequency distributions of ages are of the forms shown by fig.S.which have arisen by genericmutationsbecomes indefinitelygTeatas compared with the number of the primordialgenera. increasesrapidly with the size of the genus. F . so to speak. simply the limiting forms of the age distributions of' derived genera. a lower value of p would give more contracted distributions with lower means and modes. The proportion of primordial genera.which are. p.terms which apply even if there is only one primordialgenus).the most probable distribution is a proportion 09961 of derived genera of mean age 0 94. for p 1 5. After an infinite lapse of time the numberof derivedgeneragenera. for genera of 10 species 00842. giving a mean age for the entire aggregate (0. Conversely. J. so that the area of the curve and the square taken together is equal to unity VOL. the area on the same scale being given by the figure of column 6. but truncated at the assigned time. since p is larger. and truncated at the limiting age 628: the area is also reduced from unity to the value given by the figure in column 5 of Table X. all of age T. but the general forms would remain the same. For the monotypic genera it is only 0 0039. for genera of 60 species 0 5045. 2. A larger value of p would give rather more scattered distributions with higher means and modes. It is roughly an even chanceaccordinglythat a genus of 60 speciesis primordial. therefore. 70). 60 and 100 species. lost in the mass. The age distributions are. the chance that a genus is derived or primordial.B. 11 illustrate the case. 3. COXII. p. Table X. The larger a genus.0039 X 6 28) 096.28. 10. shows the frequencydistributionsof age for genera of 1. WILLIS. Thus for the monotypic genera. since the age of a derived genus obviously cannot exceed 7. or the chance that a genus may be primordial. 10. as stated. F. The distributions for the derived genera are of the general forms shown in fig. 10. for genera of 100 species 0 6549. the mean age of the derived genera (column 4). When time is limited. C. and the odds are nearly 2 to 1 (65 to 35) that a genus of 100 species is primordial. The chart.
78. The reader should refer to the adjacent text. 72. In the entire aggregate generaand species evolving from one or more primordialgenera this law does not. It will be rememberedthat in Section I it was found that rence the law of increase for species within the genus was geometric. giving to the nearest unit the total number of species in such an aggregate at successive units of time when p .1*5. but the ratio rapidly falls and tends to approach 2 more and more closely. (2) the number of existing species. therefore. an attempt is made to estimate the order of magnitude in of the doublingperiod the case of the floweringplants. so long as the chanceof a species survivingfrorn zero time to the time of observationis kept constant. a subject on which we have little knowledge. but only on the estimated number of species killedout. It was judged sufficienttherefore to assume. but the divergenceis only marked during the earlierstages of the evolution. section 1. has placed at roundly 160. If we fix the number of extinct species but alter the timeincidence of destruction. amongst all species of floweringplants on the whole surface of the globe (Table XV. for example. the former I have taken as The latter Dr. On this basis. the doublingperiodworks out at roundly some 6 million years. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION.000. On this assumption an estimate of the doublingperiodis not difficult if we can fix: (1) the approximate age of the flowering plants. for if species are destroyed from time to time by cataclysms or otherwise the rate of production during the undisturbed intervals must be increased in order to give the required final number of 160. 81). In the concludingsection. he will see that after some 5 or 6 doublingperiodsthe law approximatesfairly closely to the geometric law of increase with a common ratio of 2. 74. on the given assumptions. on the estimate of either the doublingperiodor the present rate of production of specific mutations. 75. p. On this point the interesting conclusionis reached that variations in the timeincidence have no effect. arises how far variations in the timeincidenceof destruction. 13 and 14. p. and the present rate of occurof specific mutations. pp. Destruction of species will clearly tend to lower our estimate of the free doublingperiod and raise our estimate of the present rate of production of specific mutations. 73. The total number of species in the aggregate more than doubles during each successive doublingperiod for species within the genus. that within the aggregate as within the genus the law might be taken as of this simple form. 12. Tables XI. assuming first.000. so that this " doublingperiod could be of taken as the usual unit of our scale. and the present rate of productionof (viable) specific mutations at 1 in some 50 or 60 years. for the rough estimation required in this section. If the reader will refer to the figures in the last column of Table XIV. pp. The important question. if evolution has proceededfreely and without any destruction of species. XII and XIII. the numbers tending to " double in successive equal intervals of time. that the has Willingout been effected almost continuously throughout the lapse of time and then . WILLIS roundly 100 million years.30 MR. may also affect our estimate. give similar information for the three other examples and are illustrated similarly by figs. VI. hold good. G.
The tentative conclusion is confirmed by sections 6 and 7 of the table. By increasingthe destruction of species (section 5) to such an extent that over 700. The second shows a chance that I have termed for brevity in the heading the chance of a species surviving for 10 million years : it is. the estimated length of the doublingperiod been lowered has to a little over one millionyears. it is more convenient to give p. it seems probable. C. 31 that it has been effectedby a series of highly destructivecataclysms occurringat intervals.S. we will arrive at different alternative estimates for the doublingperiodand the rate of production of specific mutations at the present time. p3 the chance of surviving 30 million years. F. J. We know that the chance of a species surviving over the whole 100 million years must be practically zeroa vanishingly small quantity.BASED ON THE CONCLUSIONS OF DR. for they give too high a value for the chances of survival.000 must have been killedout. In each section of the table two cases are taken. the severity of the destruction is raised from each section of the table to the next. We also know. the tenthroot of the chance of a species surviving for the whole period of 100 million years. based on a table of Mrs. 83 below) showing the approximate percentages of species which have become extinct in certain deposits of various ages 2 . on the other hand. If. In sections 2 to 5. I need hardly say that no importanceis to be attached to the precise number of cataclysms taken and the (very large) proportionof the thenexisting species killedout by each: the scheme is simply so arranged as to illustrate the effect of very different timeincidencesof destruction. But there is little to guide us as to the whereabouts of the truth on the scale of results from section 2 to section 5 of the table. we fix the chance of a species surviving from zero time to the time of observation.and the fourth the present rate of production of specific mutations. except one broadconsideration. in the first of which the destruction of species is cataclysmic. and the estimated rate of productionof specificmutations raised to 1 in some 10 or 12 years: it makes no practical differencewhether the killingout has been continuous or cataclysmic. and so on: since p'0 in several of the examples is vanishingly small.R.p2 is the chance of surviving20 million years underthe same conditions. giving too low a value. which we have assumed is the time elapsed since the genesis of the floweringplants. a series of comparisons is given on the first method. in the second continuous and of such severity as to give the same total of species killedout. while section 5. the first gives the numberof species killedout. from such cases as Ginkgo. Table XV. WILLIS. Of the four columns on the right of Table XV. The third column gives the doublingperiod. which would require if there had been 100 equally destructive cataclysms that each should have killedout no less than onethird of the thenexisting species. CLEMENT REID'S (reproducedon p. If p is the chance of surviving 10 millionyears. overestimates the amount of extinction. in fact. It therefore seems probable that sections 2 and 3 underestimate the amount of extinction.that a species may survive for enormouslylong periods of geological time. an alteration in the timeincidence of destruction will vary the estimated number of extinct species alone.
or thereabouts. or at all dogmatic. The readershould referto the more detailed discussion at the end of Section VI of the paper. of the nature of averages. in particular. WILLIS The paper is founded on his work. Admitting all the difficulties of interpretingsome results (Section IV)and as I have said. The present rate of production of (viable) specific mutations. WILLIS'S appearto me to explain a great range of facts.'or more likely still to the mathematically trained biologist. in the Pliocene: they suggest a figure somewhere between 1 7 and 32 million years for the doublingperiod. can be based on the fact that we have no knowledge or experience of such phenomena. to be natural and reasonablein themselves. To quote from p. amongst all floweringplants on the whole surface of the globe. as it seems to me. and quantitative deductions drawn. My work on the point entirely confirmsthe conclusion of Dr.or a simpledesireto avoid the wearisomereiterationof qualifying conclusions do phrases. almost certainly over 1 million and less than 6 millions. 212. and between 1 in 15 years and 1 in 29 years for the present rate of productionof (viable) specificmutations. to the cooperationof biologists with more competent mathematicians. To Dr. but would hardly have been carried through without the encouragement and stimulus of personal intercourse. It is clear that specific mutations must be events of the very greatest rarity.32 MR." The results are. I may be optimistic. I have. I have no desire at all to be dogmatic: but Dr. say. Consideringthe roughness of the basis the final conclusion is extraordinarilydefinite. " If the age of the floweringplants is 100 million years. but it seems to me that the future holds the possibility of great developments. and leaves no doubt as to the order of magnitude of the required figures. is almost certainly less than 1 in 10 years and more than one in 60 years: it probaby lies between 1 in 15 and 1 in 30 years.* If any conclusions stated in this Introduction or in the body of the paper seem to be too confident. WILLIS. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. * ' Age and Area. of course. since the floweringplants are a very heterogeneous group. G. 84. the doublingperiodfor species is probably of the order of some 2 or 3 million years: it is. . himself my acknowledgmentsare difficult adequately to express. they are considerable these may yield to further work. and no argument. to thank him for placing at my disposal the data given in Tables A to E of the Appendix and utilised as illustrations. and to present for the first time a conception of evolution at once so simple and so definitethat it can be expressedeven if only roughly owing to the gaps in our knowledge in quantitative terms. I hope the reader will attribute the appearance to inadvertenceof wording.' p.
2.e..BASED ON THE CONCLUSIONS OF DR. Number of species in the genus. THE INCREASE IN THE NUMBER OF SPECIES WITHIN THE GENUS WITH TIME.... Then.. Of the same Np generathat already possessed 2 species at the end of the first interval. . . 5 .. Let the chance of a species " throwing " a specific mutation... The proportion in which bothof the two will throw a new species is p2 giving Np3 genera of 4 species... . 3 species after 0... 2p2q genera of 3 species.... F....... 3 intervals of time..  q + q2) q +q2) + (1 + q 6 q (1 + 3q) + q3) .. the second term in the expansion of (q+p)2.. . . .. 2 . Of the Nq monotypics a proportion q will again fail to throw in the second interval and we will have. 2... . or a total of Npq (1+q) ditypic genera (cf. the scheme below). Nq2monotypics. 1 . at the end of the interval... 33 1. the genera being initially monotypic.. 2p4q (3 + 2q) 4pq 8 . (1 4 . a proportion q2 will fail to throw from either of the species they already possess. putting aside generic mutations altogether for the present.. . at the end of the interval we will have Nq genera (q....... .1p) which remain monotypic and Np genera of two species. C.. Proportionalfrequenciesof numbersof genera with 1. ... Intervals of Time. ..S...... .. givingN.. q3 t (1 +q). .. 7 .... 6 .. Of the Np generaalready possessing2 species at the end of the first interval. and suppose the interval so small that p2 may be ignored comparedwith p... 1. ...   p . 2p2q2 (1 + . .. . giving Npq2 genera of 2 species at the end of the second interval. a new species of the same genus.. .. contributingNpq to the genera of 2 species. ... J. The new species as well as the old can now throw specific mutations and the matter becomes more complex. i.. pq pq(2 ..... Of the same Nq a proportionp will throw.. .. WILLIS.. .R. if we start with N prime species of different genera. in some small assigned interval of time be p... 2pq  q p q p3 . AND THE FREQUENCY DISTRIBUTION OF SIZES OF GENERA ALL OF THE SAME AGE. 0 1 2 3 1 ... .. the proportionin which one or other will throw a new species in the second interval is 2pq.
may be verifiedby checkingthat they The formulke... after n intervals we have pq.. . At pn = st and we have the usual approximation (1q =(1) st/n)" . . that is. We may write p s. p summarisedin the scheme above.... so that the time n . . At = t is finite...p3q(1+q+6q2+q3) q2 q q2Xp . + ") q+? .2p5q(1+3q) 4p6q With 7 species : From the 4's: p3x 43q With 8 species : From the 4's : p3 p4 . omitting the common factor N.2p2q2 (1 +q+q2) . UDNY YULE ON A MATHEMATICAL THEORYOF EVOLUTION...3 .. . eA .....34 MR. we will have generaWith 1 species With 2 species From the l's From the 2's With 3 species From the 2's From the 3's With 4 species From the 2's : pq (1+q) Xp2 From the 3's 22q X 3pq2 From the 4's: 3 X q4 With 5 species : From the 3's 2p2q 32q X From the 4's p3X4pq3 With 6 species .. G. taking the timeinterval At as indefinitely small but the number of such intervals n as large. We must now proceed to the limit.... ..pq2(1+q +2) pq (I q)X q2 pq (1+q) X 2pq 2p2qX q3 } ... The work may be continued on the same simple lines. Thus at the end of the third interval.. .. if f is f..e"s.... total to unity in each column. .... (1) ql (1qT") As regards the second term...... .. e"t: Omitting the common factor N.. the first term is qnor in the limit the proportionalfrequency of monotypic genera at time t.2p4q2 (3+2q) Fromthe 3's From the 4's 22q Xp3 p3X 6p2q2 } ...nl(l q+q + q2..
2e2't + C e3t. . putting p = s.. after time t we will find the numbers that have 1... f2 being the proportionalfrequency of genera with 2 species at time t. Or proceedingto the limit. We have et) est)2 est)1 .BASED ON THE CONCLUSIONS OF DR. so that finally f2 = est (let) as before. d (e3sf) = 2s (e2 f. dt dt (e2tf2) f2 s ../f2 +df dt qf.. and so on. .f2. . . . 35 or in the limit. or in the limit....t + Ce2st... we have f3 ?+If3= 2pq..fi f2 est. F... 2.. Evaluatingf4 in the same way we find . (2) f2 est (1est) But this expression may be derived by a process which is more convenient for the higher terms. ...f2 s ... 2p 2s. Sincef3 must be zero when t is zero..S.. and therefore C must be 1. J. est).e2t). . . f est (l et)2 .. est = e. WILLIS.es. But when t is zerof2 must be zero. n+lf2 = Pfi + q2.  et . If f2.. . 3. C...R. if N prime genera start together at zero time when they are all monotypic. dt. The general form 6f the law is now obvious.f2...... . f f est (l (5) = est(1  J That is to say.. Considering in the same way. when 1q3 = 3p and pq = p + 3sf = 2s (es . 1 q2 2s. C must be + 1. (3) (4) f4 f = est f2 = est(l est (est)3 . That is ?+1/f2 2i = plf. species given by a .. . and thereforefinally .1f2 are the proportional frequencies of ditypic genera after n and n + 1 intervals respectively. . + (q2 1)9.
egt .. At time t therefore we will actually observe. start simultaneously at zero time. so that the mlean is 1/(1 .. but.36 MR. and it is this frequency distribution which is now required.r).... If N genera..r r2 r3r+ where S is the sum to infinity. not merely the frequency distribution of the sizes of the primordial genera as given by (5).3r2 . the mean number within a group of genera all starting at the same time increases as an exponential function of the time... . consider the general geometric series S 1 t. . The series (5) gives the frequency distribution of sizes of genera for genera all of the same age t. 3.. quite irregularly as chance mutations occur. what is the form of the frequency distribution for genera of all ages ? If we start at zero time with a group of monotypic genera. As regards the increase in the number of genera. given by tl s.the frequency distribution of the primordial and the derived genera together.. 2.. ... . We will first consider the limiting form of the distribution when the time t is infinite. Thereforethe is mean of (5).r)2 1 + 2r ..eg( 2 egt)l I  .. both generic and specific mutations will be thrown as time goes on. ..e ...(1  J where g is a constant proportional to the chance of a generic mutation (or mutation from genus to genus) occurring in a given time. genera will be given by N multiplied by the successive terms of the series = egt(l ... of course. . while the number of species within anrysingle genus increases.. say MM. As regards the mean.. the whole process will proceed on precisely the same lines.. and the primordial genera'become practically negligible in numbers as compared with the derived genera. (7) f..eSt s being a constant proportional to the chance of a specific mutation occurringin a given time.. The sum of the series (5) to infinity is unity as it must be. We have dS/dr = (1 .. which is the first moment about zero. . at time t the numbers of families with 1. the mean increasing as an exponential function of the time. . geometric series of which the common ratio is 1.. .et) f3 . tM. The mean number of genera in a family at time t in these circumstances will be given by (8) .. = e.. (6) That is to say.. We have next to proceed to the question. G. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. belonging to distinct families.
. J.e(Y+2s)}dx J g [(g + s)l . .. CCXmII. writing x for t throughout.s/. i Proceeding similarly for the further terms we find for the limiting form of the frequency distribution of sizes of genera of all ages after infinite time the series 1f  p I 1 p + 2p + I + I3 f 1  . The first term will give  e+ . and the latter may legitimately be ignored during the present stage of the work.R. . .. C.. say T. p/(l + 2p). as stated at the end of Section I. The proportionaged x at time T is therefore gey~dx...s)1] p +1 )p)I(1 + 2p)1 . FORMOF THE FREQUENCY II.2p 1 + 3p 1 np J The series has been wvritten the form most convenient for calculation : the fractions in 1/(1 +. Note that these are the derived genera only.J3. The second term will give g o. . are of any assigned age x. We have now got to take the series (5). . . . (12) 2p (n1)p 1 + p 1 .S. . dx multiply each term by ge"x and integrate from zero to infinity.2( 1. .. The numbercoming into existence during the interval ? fdt round time t is thereforeNge9tdt. . (.4 1I ?e(+s). and the number of age x at time T is Ng e(TX) dx. are first run out on the calculating machine and fl.1. . .F. (9) where . ... VOL. f2.OF BASED ON THE CONCLUSIONS DR. From (8) the total number of genera at time t is Negt. are then obtained by successive multiplication. but.(10) P .1 3p 2p 1P 2p ? ? ? ? ? . 37 We first require to know how many out of the totality of genera existing at any given time.. etc. . since weare going to take time as infinite the number of derived genera will be infinitely great as comparedwith the number of primordialgenera.p).. TIE LIMITING DISTRIBUTION FORTHE SIZES OF DERIVED GENERA WHEN TIME IS INFINITE. As we can only suppose that specific mutations are more frequent than generic mutations p must be greater than unity. WILLIS. .f f...ad =g(g(+ )g s) (1 . term by term. ignoring the prime genera with which we started.
2! 1 2I 1 +3 . the mean is infinite. +f n) (n J) p.( 38 MR..pnf. The frequency distribution given by the terms of (12) is therefore one of those paradoxical distributions in which.. F. y y (y+l)+y + 3 1...fn )_ (1+P)f2 (1 + p)fi (1  P(flf2) 1 + 4 Hence if Snis the sum of the first n terms p) S =..1 p (f . (13) This result (which with much of the following I owe to the kindness of Mr. .... Sn .1) f. ) .. . This is.. (y 2 n 14) giving the first moment of (12) about zero.. 2 + p. the series is therefore a hypergeometricalseries in which y Considerthe series 2.2.. . P.. 4)n 1) ( 2) y (y + .. for on our assumptions the mean size of a genus after infinite time must itself be infinite. White. G. as it should be. John's College) is exceedingly useful for checking calculations: it also follows from it at once that S ... etc.... But IL{In\ UY2 which is less than unity..c is unity.. l+ ' +y where y 2+p1. though the median.. as it should be. In this series nl +=1 Un As y > 2 the ratio exceeds unity. of course. only approachingit as n increases. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. Apart from the common factor.1) p(fi (n 2) f) 2 p( ..3. are finite.2 y (Y . 1. or 1 t p S. We have _ (n.1) p 1 +{np or (1 + p)f (1 + p)_  (n . so the series (14) is divergent. The series may be written in the form 1+ P Y 1 y2 + 2p {1 ( + 2) (1+3) p) 1.( . The series can be summedto n terms.22 . Fellow of St.pnf.
4712 1. F. (1 .4554 1.A logf.1*925. p.5008 1.4965 1. a distribution with a maximum frequency for monotypic genera.5139 1.4870 1'4895 1.4764 1.4989 1.. 1t07f7.5146 1*5152 1. . 39' Since the limiting distribution (12) is the foundation of all subsequent work.4841 1.n +...4935 15 16 17 18 19 20 30 40 50 60 70 80 90 100 7 6 6 5 5 4 2 1 1 3329 6584 0810 5820 1475 7663 5906 6786 1982 9095 7202 5883 4922 4196 1.a ) log n . 4916 1. 54).5118 15131 1.. Alogff r. The greater p.P) (1 .1 log 2n 1 a+ n (1 + a/n)n) 1a/n)n( a) ( alog r( (n a) log .2 .2 log 2w (n where a is assumed smnall comparedwith n : whence in the limit (n 1)! p"1 (1 + p1) r_ (n) log 2 f r(l + p ) n A. We have (1 .1 925. WILLIS.l/A log n.5099 1. (15) . log n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 341 135 77 51 37 28 22 18 15 13 11 10 9 8 8803 6948 1107 1856 0945 4489 7002 6515 6744 4103 6414 2285 0789 1288 13331 13939 1. TABLEI.p1) ( + P1) p (n + p1) (n1 r(n t p)' p If n be large. A l10og.) log n . the frequency tailing off at first rapidly and then more slowly as the successive multipliers (n . the smaller is the frequencyof monotypes and the more slowly do the frequencies fall away...S.4244 1.4644 1.BASED ON THE CONCLUSIONS OF DR.. Alog ' n . . for p . it may be as well to examine its form in some detail. the value suggested by the Chrysomelidbeetles (cf. as required by all the data for size of genus that have yet been tabulated.R. we may write by Stirling's theorem log r (n) (n .Values of 107fn in the pseries (12) and of .n +.. J..4807 1.5068 1.4999 1.1) p/(l + np) approachunity. C.4430 1. even when time is limited and not infinite.5156 Considerthe approximateform of the tail of the distribution.2p) .4978 1.np) (ni)! + p1) .4951 1. It evidently gives. below. Table I shows the values of the successive terms (giving only every tenth term after f20) for p =. (2 ..
or generafrom genera. considerationsof a very rough kind suggested that the limiting form of the distribution law. We have dt = sxdx t = log x1/8. When work on the frequency distributions of sizes of genera was first begun. and for no statistical data which followed the law would such a divergence from. the approximation rapidly becoming exceedingly close. Similarly it seemed reasonable to assume that the number of species x in a genus of age t would be given by X Czet. that the number of species in a genus can be taken as a continuous variable. and the approach of the slope to the limiting value 1 + p. Further trial showed that the law did in fact hold . and that the above can be taken as absolute functional relations. The size of a genus is then absolutely determined by its age. G. UI)NY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. linearity appear significant. Now suppose that chance can be ignored. the resulting points lie nearly on a straight line: how nearly is shown by the dotted line in figs. in which logf. and we can find the numberof genera of each size by eliminatingt from the first relation. The generationof species for infinite time should approachthis logarithmiclinear from species. is plotted to log n for the data of Table I. The method of approach was obviously exceedingly crude.Alogf/ A log n given in Table I bring out better than the charts the gradual increase in the slope of the curve. This formula holds fairly even for moderate values of n. The dotted line is actually concave towards the base.seemedclosely parallel to the generation of offspring in a given stock in which mortality might be ignored: hence it seemed reasonable to assume that the proportion y of genera of age t at any given epoch (apart from the primordialgenera which might be ignored if time were infinite) would follow the law y c e.or 15195. The values of . as shown by the following comparison with the true values of fn as given by (12) in Table I: Valueoff as givenby n 10 20 50 100 (12) 0013 410 0004 766 0O001 198 0000 4196 (15) 0013 932 0004 860 0001 208 0000 4212 It follows that. if we plot logfl to log n. but it suggested logarithmic plotting of the data. but the curvature is so small that it is hardly appreciableto the eye.40 MR. dt. 49. whence y ( x(Y+S)I'dx p1) (o X dx.
. up to genera of 30 species or so.S (g .. after which the points began to drop away more or less rapidly from the line.e (gs). f2. therefore.(+s)T(e.+)T 2s)1 e2(+ 2()r 2) eg =5^45)"e(/+S?)T~(( 4s (g + 2s)I e'(+2'T + 3s (g t. If time is finite we can no longer. pp. however.R. i.3s) 1 + e( +3s)T 1 ( ~~~~~~~~~~~i r .both for animals and plants.rg (g + 2s .e(g+s) T C2 = cs (g . 'ro obtain the requiredfrequency distribution there will. At time T.s)1 e(g+)r)T 2s (g s t(g + s)1 e. out of unit total. as at the beginning of Section II. .S. (b) We must subtract from each term of (12) the value given by making the upper limit of the integral T instead of infinity. J. . 2 and 3 (below.e )1 (~2S) T [ eg (g is)~ e(g+s)T 2 (g . ignore the number of prime genera as compared with the number of derived genera. The additive corrections to fi. the entire correctionsare : C. we will have geyx dx derived genera of age x.s) . THE FREQUENCY DISTRIBUTION FOR SIZES OF GENERA (DERIVED AND PRIMORDIAL) WHEN TIME IS FINITE. f3. the observed frequencies of the larger genera were smaller than those suggested by the law : figs. be two corrections to make to the limiting distribution (12). III.BASED ON THE CONCLUSIONS OF DR. C. As will be seen from the following.(g+s)T e. The approximation was. 41 very approximately. 46) bring out the point very well. are therefore . 45. (a) We must add to each term of (12) the correspondingterm of (5) (writing T for t) multiplied by e~T. sufficiently close to encourage further work. .(g+) Bringing the two together. the fundamental mistake was the notion that time might be regarded for practical purposesas infinite: it must in fact be regardedas quite short. F.e. WILLIS. together with egT of age Tthe term which vanishes if T is infinite.(+s)'l(  esT) eT)2 and the subtractive corrections g(g  s)1 e (+s)T g (g s)1 +.28s)1 (f+2s)T g (g + 3s)~ e.
. . are calculated from (19). from (17) : the values of the limiting frequencies fi.ST . It will be noticed that these may be written in the form C1 C2 c3 C. values of c. c3.ns . are at first positive. Tf3 f? f3 Jt C3 Since sufficiently extensive tables of the binomial coefficients were not. and finally the frequenciesat time X are given by . G. thence the Given p and . C/1 CI1 2 3 __ s \ ' _ ' J(17) . The corrections c1. C2. etc. but decrease steadily in value and sooner or later become negative: such a change of sign must.3p)l '! .= (?'[/X =r. . X being the time in the new units C'1= c/ (1 +_ p)l eapl(l+p)T 2p(1. and twelve terms usually give more than sufficient precision.1)1. . .3p (1 .42 MR.I/ g ..Ai . c'.. so far as I could find. c'2. the values of c'1. .c C'l . c. . may be written. .. .+2p) eapl(l+2p) e~apl(+3p)T . . . where one new term c' is introduced at each stage. C'2. As the unit of time we will take the average time X in which the number of species within the genus doubles. As before.... . of course. For numerical purposes the terms c'. . .. 2 0 6931472. we write p s/g. tables were calculated giving all the coefficients of (1 + 1)" up to n .a (1 + np)/p. must be put in a different form. af2. available. cf3..r where Then l. (18) g_.c. Section IV). c'.3c'3 c'4 j and so on. c'3.2c'  . and '1. . writing e. are found from (12). . . c'. The correctingserieson the right of (17) convergewith fair rapidity in the illustrative examples that have been tried (cf.C13  K . 3c'2 j. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. and the numerical coefficients are the coefficientsof (1 . ?.31.og. . and the coefficients up to the twelfth for values of n from 32 to 101.
54). C... p.. TABLE II.f. When C = 2. 571 Time r in doublingperiods.6 28. time found for the Chrysomethe lidca.. correcting the terms are still positive at the limit to which calculation was carried.R. . The table shows in a very interesting way the rapid growth in the tail of the distribution towards large genera. 12 6 3 1 11 . and Appendix.. These features speak for themselves... which are necessarily positive at the beginning of the series. The table shows the distribution after 1. When T. 2..... WILLIS.. c. fo02.Showing. is the first correction to become negative and the negative maximum is reached with c. .. at successive epochs. when = I1become negative with c4. 4 and 5 doublingperiodsfor species within the genus .)must be zero. C = 6'28. .1*925. C7 reaching the negative maximum. the calculated numbers of genera with 1..S. 2. ... . When X = 4... .. c. Number of species in the genus. When =3.) . 56. teg s1 1 . To illustrate the change in the form of the distribution with time. viz. 3. the value found for the Chrysomelid beetles (Table V. It may be added that the correctingterms. for X = o is also given. 3. Table II has been calculated. approximatesto . the limiting distribution being the values found for the Chrysomelidce. J. 4 3 12 10 13 11 11 10 10 9 9 8 13 14 (Tablecontinuedoverleaf..28 units (doublingperiodsfor species within the genus) and p is 1*925 (p . . 43 occur since XZ(c. . showing the successive distributions when ? = 1.8 38 27 20 15 6 40 33 27 22 19 16 13 35 29 24 21 18 16 14 31 26 21 18 16 14 13 29 24 19 16 14 12 11 28 23 19 16 13 12 10 7 8 9 10 11 12 13 14 . and the quick approximationof the first part of the distribution towards the limiting form for infinite time... and 6. F.... 2. .. .. then for 6 28 doublingperiods. . After reaching a negative maximum the corrections again diminish until c.. . . c37 is the negative first to become negative. species out of 1. . and finally the limiting form of the distribution after an infinite lapse of time. 5.. below.. p. 227 104 50 24 192 115 77 53 343 159 97 68 514 145 85 59 4 139 80 54 40 137 78 52 8 136 77 51 37 1 2 3 4 5 6 7 8 9 10 11 12 . BASED ON THE CONCLUSIONS OF DR. 3. 2 422 3 370 45 352 345 6 28 oo 342 2 3 4 5 .000 genera in all at each stage when p = 1*925. and when = 5 the first correction to become negative lies somewherebetween c60and c80. . is the first to become and the negative maximum lies (not far) beyond c31.. as for the Chrysomelidce.. Table A).. . 4.
2 2 9 3 .000 116 1. .... 1! _.. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION.. . 19 20 21 . 1... 6I28 1 T 2 3 3 8 7 6. . TABLEI(continued). . Total _ 2 2 2 2 2 2 2 2 2 '1 3 3 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0J 31 32 33 34 35 36 37 2 2 2 1)1 1 38 39 40 41... ....... 1...G.. Beforethe present theory had been developed the form of the graphs observed seemed rather puzzling. r in doublingperiods. ... but after a certain point there was always a more or less rapid falling away of the data from the line. ..... .. . .. . The initial part of the graph... ... Figs. . form is very interesting. ... 2 and 3 are given to illustrate the point.. ... .... appearedextraordinarilynearly linear. .... . ....000 85 1.. ..5 16 17 18 19 20 21 22 . . 50 Over the last frequency given above. 8 7 7 6 6 5 5 5 4c 4 s 7 7 6 6 5 5 4 4 4 _. Nu ?Timbe Nulber of species in the genus. . . __ .000 000 1 The gradual approximation of the double logarithmic graph of the frequency distribution towards its limiting. . . . .. 49 ... ... 15 . . . nearly linear. . .. . ... . 46 . ..33 3 3 3 . 1. . .. . I 1 1 1 47 . 41 42 43 44 45 45. 42 43 44 2 2 2 2 2 1 1 1 1 1 45 46 47 48 49 50 2 1. ..000 6 11 1... 2 2 2 . . . .. .. say up to genera of 30 species or more... 10 9 5 _ 9 9 8 7 7 a ...44 MIR. .. 16 .....000 1 1. 23 24 25 5 5 4 4: 4 4 4 4 4 4 3 3 26 27 28 . . .. 2 3 .. .. 2 2 2 2 2 2 1 1 1... . .. 3 3 3 3 26 27 28 29 30 29 30 31 33 34 35 36 37 38 39 40 _I I 3 0 3 3 3 3 3 3 3 3 3 2 .i 5 5 4 9 8 7 7 6 6 5 5 4 4 4 6 6 6 5 23 24 25 ... 17 18 2 2 1. . 1 .
. J..~  ^ .B. [ F+ o o.^  *. .. logarithmic chart for the frequency distribution of sizes of genera in the Cerambycinr.. r^ *e . Table A. cr I 4 \\^ ' .. CZ .^.! 'a~~~~~~~~~~~~~~'2 0 Fig.E. .Double .. Table B. 2. COXIII. *i log(number of species) number of species . 1 100 1 ' V \ . Data in Appendix.10 100 2I L 2 0 rV : .Double logarithmic chart for the frequency distribution of sizes of genera in the Chrysomelid : logarithmof the numberof genera plotted on the vertical to logarithmof the numberof species on the horizontal.BASED ON THE CONCLUSIONS OF DR..^X~ .S. 1. ~10 E O.. ^. number of species 10 100 45 2 co 16 (U 0 (U tDn 4 E 0 23 CZ1 0 0 '2 4 '6 8 1. VOL.O .0 12 14 16 18 20 Fig. C. ' .2' 4 6 log (number of species) H 8 1'0 12 14 16 I1 18 2' 0 Data in Appendix. WILLIS. F.
. 1215.46 AMR. Fig.qguminosa: 1. ssi I _ D  0 Fi 2 '4 . . are shown in figs. bat terminsating wit. 3443. Cerarnbycinca: the same way up to 53. drawn from a larger number of significantfiguresthan are given in Table II. 3. The following were the actual groupings.L. '6 1. that they gave rise in one's mind to all kinds of speculationse. etc. In fig. used in the three charts shown. up to genera of 100 species. however.e. the graph for each value of T being given by the full line and the liimitinggraphfor X = so by the brokenline. that beyond this point there is a heavy deficiency of the larger genera as compared with the numbers that would be law. IUDNY YULE ON A MATHEMATICAL THEORYOF EVOLUTION. graduated from 4 to 23 by the two groups 413. thence in groups 911. Table E. 2 for the Ce. 5173. at least in the tail of the distribution. 3. 4 to 9. calculation shows. 1619. for the Legurninosce. number i ___ ____l_o of species 10 100 OM . 4453. 2. * It is of course always necessary to group or graduate the data for such charts. i. G.rambycince (beetles) (data in Table given by the logarithmiclinear B of the Appendix) is suggestive of linearity only up to genera of some 20 or 30 species: the points given by the frequenciesof the largergeneralie well below the line. The graphs. Here the data Fig. ) 0 X t i .g. Chrysomelidp : ungrouped to 8. K. Data in Appendix. 3443.h 5483. 2023. 2427.. it appeared so unlikely that the natural form of the graph could be iearly linear over the first part of the range and then rapidly curved.Double logarithmic chart for the frequency distribution of sizes of genera ir the Leguminosce. 74103.8 10 12 14 loq (number of species) 16 1'8 20 o0 Fig.the falling away occurs at nuch the same point and is very abrupt. the possibility of the natural distributions being truncated by the last glacial epoch! But when the logarithmic graphs are drawn for the distributions of Table II it will be seen that such forms are precisely those to be expected. 2833. 1 is for tle Chrysornelid are suggestive of nothing but a straight line right up to the limit of the chart. 4453." CD .* The forms of such graphs seemed so odd..1423 thence grouped 2433.5. 3 ungrouped. beetles (data in Table A of the Appendix)..
47 ?o2 2. J. p = 1 925. form as the time is increased (cf. Table II). Figs. F. log (number of species) 20 Fig.Double logarithmic graph of the frequency distribution of size of genus for r = 1. WILLIS.1 925.S. 5. full line. 0) E 0 C CP _p 0 lo9 ( number of species) S0 20 Fig. 4. 5 to 9 show the gradualapproximationof the form of the graph towards the limiting.Double logarithmic graph (full line) of the frequency distribution of size of genus for r = 2. the brokenline showing the limiting form at r = oc for comparison. H 2 .BASED ON THE CONCLUSIONS OF DR. nearly linear.R. C. o .
UDNY YULE ON A MATHEMATICAL THEOY OF EVOLUTION. 7.0 (f1ll Fig. line) of the frequency distribution of size of genns forr = 3 2 (U cg V 0 U" 4o 0 E 0 Z2 U". p == 1 925. 0J 0 'I%/~~~~~~~~~ \ 20 species) log (number of the frequency distribution of size of genus for Fig.Double logarithmie graph (full line) r . of .4.Double logarithmic graph p ==1925. G.0 2. 6. 0 log (number of species) 1.48 lMR.
8.BASED ON THE CONCLUSIONS OF DR. 49 (U S2 (U) El C 0 0 0 1. loq(numberof species) .9.Double logarithmic graph (full line) of the frequency distribution of size of genus for r = 5. (a E c"? sF4 . F.Double logarithmic graph (full line) of the frequency distribution of size of genus for r = 6 28. 9. co U. p = 1925.S.0 log (number of species) 20 Fig.9" Q 628 o Fiq. J.  Fig. C.R. p == 925. WILLIS.
It will be seen that at first. I and 9).. the first paragraph of Section II). 9 could well suggest anything but a linear law. The values of p estimated in this way and given at the meeting of thle Linnean Society on February 2nd. is The mean number of species per genus.. the straight " portion of the graph covers the whole chart and the point at which falling away becomes conspicuous lies outside the picture on the right. were in error from this cause.. ... dividing by the number of genera Noeg"' . the deduction of the mean size of genis from the distribution would be complex. ..N . As time increases the last feature becomes more and more conspicuous.... and Since we have not obtained the expression(20) for the frequency distributionin any simple forn. When T increases to 6 28.. WILLIS myself..NT =.. ) .. . The total number of genera at time T is Noegtwhere No is the number of primordial genera... (21) which checks by putting T = 0.... 2) or the Leguminosc (fig... No(p ) . . DNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. 7 (r . 3).g) 8{s e("..... dependingon the numberof genera.NogJ eg(.e dx .. the number in a genus aged x being e"'. the graph is more or less suggestive of an arc of a parabola. the divergence from linearity with 1000 or 2000 genera available in all would never appear significant. increasesnot at a steady (percentage) rate but with decreasingrapidity..4) is very suggestive of the sort of graph given by the CerambycinB(fig..... The total number of species at time T is the number in genera of all ages. (25) These expressionsincrease continuouslywith the time and are infinite for .. It should be noted.. (23) i{pe7e . s. . when . . (24) . Within the limits shown (genera of 1 to 100 species) no actual data following the law of fig. the graph being almost straight for some portion of its length and then falling away very rapidly: fig. but the curve lies rather close to its tangent for some distance from the start. but it may be directly deduced from quite simple considerations.50 MR G... 22) In terms of the notation when the doublingperiodfor species within the genus is taken as unit the equations may be written: yN XT .. " as for the Chrysormelidc (figs.(.(p  )1pe(1p } . Hence... 1922. however..g) .. M == (s ... for very short values of the time. ..No. that the line fitted to the data within the " given range would have a smaller slope than the limiting line.m: but it is to be noted that the totalnumberof species sN. . . The secondterm." and hence if p were on estimated therefronm the assumption that time could be regardedas infinite too high a value would be assigned to that constant.Ne 0o N{s (s  g) e g (s g) . ..N being the requirednumber(cf.. when papers were read by Dr..
and the calculation of Mand f\ from the values of p and . On the mathematical foundation of theoretical statistics.t As an example of the detailed work the Cerambycince may be taken. by his (beetles).BASED ON THE CONCLUSIONS OF DR. If the assumptionthat genericmutations give fresh starting points for specific mutations is correct. " t R. vol.that mutations of a higher order still. 222. the approximate values of p and Tdetermined from Table III by interpolation as shown below. But I have not developed the matter further.. 309368.* But I have not seen my way to any simple and directsolution. Lines 10 to 12 give the results of applying the X2test of " goodness of fit " for the respective groupingsof the data shown in Tables V to VIII: the Ptables (Tablesfor Statisticicans and Biometricians)are entered with n' as 2 less than the number of groups. The method finally adopted of was(1) to fix on the proportion of monotypes and the mean numnber species to the genus as the characteristicsto be used for determining p and . the values of M and fi given by the data. such as would give rise to new families. 87 (1922)." 'Phil. WILLIS shown in full. the Snakes and the Lizards respectively. A. The method of moments would not. IV. Roy. As examples of and fitting I have taken four tables kindly given me by Dr. Stat. must emphasise the result. since two constants have been fitted to the taken data. pp. On the interpretation of X2from contingency tables and the calculation of P." ' Jour. Trans. and for all values of T from 1 to 10 by units. Table IV shows the numbers of genera and species in each group. giving fi and M for all values of p from ?0 to 3 0 by tenths of a unit. (2) to draw up a table giving the proportionof monotypes (f.S.. THE FITTING TO DATA OF THEE EXPRESSION OBTAINEDIN SECTION III. Given an actual frequency distribution for size of genus. * . p. the effect shown by equation (24) or (25) must follow. the percentage increase falls in every following interval and rapidly asymptotes towards 100. They refer to the Chrysomelidce the CerambycGince (beetles). therefore. A. R. and (3) in any given case to determine p and T from this table by inverse interpolation.) and the mean number of species to the genus (M) for all values of p and T that seemed likely to occur. and No the number of primordial genera. M is 5 *584 and " Cf. Lines 8 and 9 are simply a check on the work and the precision of interpolation in Table III: such interpolation is not accurate.' vol. permission. WILLIS. Table III is the fundamental table. J. with distributions of the present form. F. It will of course be noted that we have not consideredany group of higher order than the genus. FISHEln. C.R. 85. FISHER.be a good method to use even if it led to a simple solution. 51 becomes of less and less importance as the proportion of genera to species decreases. and it would seem. in the Appendix to this paper. the problem is to determine from it the values of p and r. Soc.determinedby interpolation is a desirable check.'A. while the total number of species more than doubles in the first unit of time. for the very long "tail " of the distribution implies high probable errors in the moments.
875 4000 40.860 40.12.6 fl M 17 fi 2818 38. Tim 1 62.897 4883 42.178 36..469 19667 37.457 8470 10119 43.781 3079 48..773 47.488 34.1l.228 1 820 43.293 8 690 36.434 6739 37.125 51.497 28080 34.977 1 802 45.713 8985 M M M 6309 12247 43.333 2477 50. TABLEIII.726 25205 44..523 19910 35.689 5 074 6 50. or approximately by any values of p and T subject to the equation .479 18121 41..141 5876 37.930 40.1 1 and = p 1 2.. or approximately by any values of p and r subject to the equation .591 5560 39.001 28 238 38.003 5 852 47.668 18 304 40. p + 0003059.638 6054 45.608 3289 46.953 2762 44. 1e. Using they must be taken between p simple interpolation only.464 25. (6) Solving (a) and (b) we find ? = 4 980. ? = 1203519 = Similarly M = 5584 is given by (p..059 40. from 1 0 to 3 0 (decimal point omitted before fr).007 9249 34.5 fl 58. (a) .000 7 238 7931 47.. .2 fi M CI 9 7 2 3 50... P. =5'520408) and by (p .500 1 '693 61.Values of the proportion of monotypes (fr) and of the mean size of genus (M) for values of the time X from 1 to 10 (the unit being the doubling period for species within the genus) and of the ratio p (of the chance of a specific mutation to the chance of a generic mutation).462 45.484 39426 1 55 180 .748 18191 34.122 12032 35.000 8 9 10 o 0 ce 1*0 fi M 53..343 2634 47.899 1 780 58.040 30262 35. p = 1 *188224).716 P^ 37.619 10656 14049 60. for a value of very near 5..318 3840 41.620 8211 47.479 0 1 .619 9.667 22.001 6545 50.395 11 971 14949 41.807 40. we find that the observed value of fi is given by (c = 6.820 6921 41.481 43.671 11500 14 566 0 0O 1.455 i 45.060 1752 59.065 16 479 38.52 MR.. fi is 045801..489 9.462 34212 1'3 fi 3 44.584 M M9 NI 1*812 57. G.455 43.750 2560 48.154 2916 42. T 2001886 .4f1 *3 c) 1 767 2701 41.379 2960 4420 35.981 41.601 45.944 45.444 49.170 5229 45.018 40.413 1792 57..761 1735 50.905 3669 43.718 34.533 11018 37. T=4'908312).224 4542 34. p = 1 188..002 22000 38.999 37.813 9790 I Q 00 0 1. r 1 1715 1. p = 1 *185165) and by (r = 5. p+ 0163373.084 45. Looking over Table III to find a pair of values near these.481 7.514 7.575 6465 35.049 4466 47. we find that ..012 5159 47.038 40 730 47 738 (a) 0 57.682 8115 37. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION.195 3.882 4153 2386 50.046 22393 35.624 7098 50..566 .485 14 784 37.186 40.660 3485 4 5 50.793 4524 43.750 0 *a a 47.000 t 50.245 6.549 5691 43.681 41..001 0 2 869 4289 38.006 7 524 10000 13119 17 049 38.597 14027 38.814 13037 34.559 4149 39 353 43..
Time. J.852 41.699 29. WILLIS.464 33.261 7248 10819 15953 23335 33948 49206 71144 32. almost with precision(0 *457979 against 045801) but give by equation (25) a mean 5 593 instead of 5584.271 33115 34643 25.831 4962 33.306 7710 11777 17795 26698 39'872 59'364 88204 30.583 28.458 5'137 30. we have two alternative values fot the initial number of genera.916 1 877 54.576 69707 106000 29.BASED . VOL.027 54404 25. a small and hardly material difference.122 34.9 ft M 30Jf1 M 54.332 28.000 41.018 25. 1 2 3 4 5 6 7 8 9 10 20 fl M 2'] f i M ' V ? 2 2 f9i M 93 fl O. reproducef. CCXI.607 29.904 1828 56.258 31.544 31.412 31.373 28.786 30.783 19543 28.064 80089 124233 85259 133436 25.801 M fi M fi 1854 3*134 38.658 25.702 33.738 14619 26.372 3189 37. C. it will be seen from Table IV.293 23500 26.403 3072 39.684 57211 25.422 29.1.265 32.327 30.835 5218 31.416 2o 55.369 30.S. 2765 =56 03 or roundly 560.250 30.764 1881 37. the number corresponding to the estimate based on the generabeing given in Table IV.482 30.179 5435 29.Lo:B: I .418 27.840 3104 39.046 59 976 2.811 27. however.129 12667 28.389 32.041 27. 53 TABLE III(continued).791 27. the numbers of genera and of species at time T are found by equations (23) and (24) to be 182765 and 102 2133 respectively: hence the initial number of genera is either or 1024/18.695 31.783 8326 13087 20381 31 556 48675 74901 115078 28.404 32.309 31.044 8130 3 26 f 55.272 31. The value not being quite precise.379 33)350 33.485 29.316 36.079 1869 3214 3237 5294 5366 8512 8691 13492 13882 26.122 51557 26.652 29965 28.280 4657 35.344 14257 25.R.312 30.277 32.659 21196 21987 25.335 1842 56. F.ON THE CONCLUSIONS OF DR.750 36137 25.010 3037 40.116 37 598 26.006 27.638 1860 32.660 0o ? M M 55.339 33.828 3162 5052 7 925 12230 18681 28345 42826 64522 97030 29.602 45763 28.984 27.253 7485 11307 16885 25'027 36909 54'251 79561 31.648 90399 142 663 25.576 27.032 1873 28. 56.944 37. Putting No .932 29.918 22 755 25.438 mIIM 1864 2*7 /fi 28 fi 38.687 5500 27.667 3.252 55.375 33.428 8 861 26.860 9024 These values.084 1849 55.812 3280 30.664 32.007 95500 151890 26.607 1836 56.439 29.404 4765 34.865 27.167 3 259 36.468 31. 2 .335 7000 10314 15000 21627 31000 44255 63000 33.311 32.356 27. 5718/1022133 = 55 94 The interpolation is less precise in the case of the Chrysomelidce.608 30.590 4866 33. where the two alternatives are 6535 and 6605.331 26.322 27.
.. r ..... . .... ... . ........ ... 16 0.... 083 0051 . r... Number of genera 2. .. . . ......035 038 830 727 878 522 The series for the Cerambycincwas calculated up to f6. .. 293 1.. . . . ..... . . ........ 0.....f ....... Mean fromp.. 9. .. . c/.... 0160 0.... . Proportionof monotypes.. . . As already mentioned..... .. are calculated from equations (19). . .. P .. rTABLEIV........ Next.. .... Meanspecies per genus..... ..... ... ... .. Co0' .... .. ..... and checked from equation (13) by summation at intervals.. . . 3..188 560 5593 0 04580 1376 3. ........ ... ... ... .107' . .. .. ... .. 627 9...... 0. . . 4. .... No . ... .... ... ... .... ....... 8.. ... c . .. ... .. x 0455 c' c6 cy . .. .Giving particulars respecting four frequency distributions for size of genus.... .  . .. . . . .. ......... .. . . . c7' .. T ....941 104. .... .... 101 x 0... the tables of binomial coefficients were only calculated up to the twelfth in the final part of my table. .... . .. . 11.. .... Cerambyeine. 7340 .* .. .. 6. . . . .74 14 039 121 11 i18 096 From the given value of p.. . Number of species .. 457 . Chrysomelida. . . the values of c.. n' 12... cnl' . rThedistributions are given in a condensed form in the following Tables V to VIII and in full in the Appendix........161 1016 x 0514 X 0 164 10"1 X 0523 9285 6658 5844 6397 . Proportion of monotypes from p.. so it was no use going beyond c2': as many correcting terms as this may not always be wanted. . ...... .. . . but it saves time to calculate all the twelve at once in case they may be needed.. ..x 107 0386 105 X 0135 3398 6267 ...... * ../.. ... G...5...... 2 ......... UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION.... . 10. 103 x 0.. . 1. .. 3' C4 . ... 259 1. . .... . 5.. .475 5034 04471 4260 1253 278 5 038 0 4466 1377 Lizards.149 1010 X 0... .. ... .580 6100 04054 4281 1496 360 6130 0 4049 350 1.. ... .024 5. 0... . 108 x 0..94 03429 628 1 925 654 1577 0 3428 11 21 Snakes. ....718 5'584 04580 4....... p 7. c/1 c. . The first few values in the present case are :1 2 3 4 5 ...997 1.*  ... ..485 1013 X 0503 14 X 0... the values of the successive terms of the pseries (12) are calculated as there described.. .54 MR. . ...... . ..156 2737 0638 1061 8885 3085 .. C12' e..980 1..... M . ..
. terms were required up to c. the products being entered in the vertical columns headed 1. With this digression on the method of fitting we may now revert to Table IV and consider first the particulars respecting goodness of fit in lines 1012. It will be seen that the values of P are highest for the Lizardsand the Chrysomelidce. F. and the decimal point and zeros thereafter are omitted for brevity as they were in working..038 160. The signs of the products are given at the heads of the columns. Finally.960 19.1 +865 +829 +794 1762 t728 83. multiplying this by the observed number of genera we have the calculated number of genera of each size.878 35.317 26. 3. I 2 . 10. As two different methods of extrapolation gave nearly the same answer it is hoped the result is fairly close to the truth..BASED ON THE CONCLUSIONS OF DR. in the next column. The distribution for the Chrysomelidce was the most laborious to calculate as the series is very extensive: terms were calculated direct up to fio2.522 25.0 37 2 274 211 Each componentcorrectingterm c' is taken in turn and multiplied by the corresponding successive binomial coefficients. 20. 35.592 86 6 54. 4. In calculating the pseries it is desirable to retain seven significant figures so as to avoid accumulating errors. for c2' they run 1. for ca' they are 1. the sign of which is always positive at the commencement. correction.' in the correctionsto f57 onwards. In the case of grouped distributions like Tables V to VIII the grouped frequency has been calculated by adding the values of Tf. by 1 2 39  +902 457. beginning withf2. 3.. .707 36. I have found it convenient to arrange the work of equations (17) in the following form: the decimal places have been reduced to six for the illustration. for c 1.. 1024. but an estimate was also made by extrapolation of the portion of the remaining frequency lying betweenf0io and f. so as to obtain an answer correct in the decimal place.and added therefore to the corresponding term of the pseries gives the required value of T.R. formulae the giving in both these cases a most excellent fit to the data. 55 Retaining seven decimal places in the work.' the binomial coefficients are unity throughout. 1 2 _ 3 ?\ 4 f + + 941 941  + + +941  _ Total correction. beginningwithf and so on.830 M. pseries. 4. 3.727 51.024.50 inclusive. 10.f.* so as to be fairly confident of correctness in the sixth place of the result. For the Cerambycince and the Snakes the fits are not quite so good.ultiplied Total. .979 161. The summation of these terms then gives the total correction. etc. C.863 84. For c. WILLIS. 6. 2. . but still well within the limits of fluctuations of * In most of the work at this stage six decimal places only were retained. . 457. 2. 15.733 4690 165 6 3 4 5 6 7 941 941 941 941 941 77 116 155 193 232 1 4 8 14 20 . not seven. . . J. beginning with f3.S.721 20.592 52.
..4 51 to 75 .... the fit is worst for genera of 3 species.56 E ... 15*9 11.. Regrouping Table VI by the runs of sign of the differences between observation and expectation only reduces P from 039 to 032... 12 to 15 to 21 to 31 to 41 to 14 20 30 40 50 . x2 is 9 59. of which there are only 38 against an expectation of 49.. .. . and P 021.... 730..Chrysomelidce: observed and calculated numbers of genera of each size.. 101 to 150 151 upwards Total .. 56. ... 17 13 7 9 627 185 111 123* 11. of which there are 40 against an expectatiol of only 274: referenceto the detailed data in the Appendix will show that there is here a marked " hump " in the data which could not be covered by any smooth curve. 76100. even in the case of the Snakes one would expect to get a worse fit.. .. Number of genera.. . For the Cerambycince most marked divergence is in the group of genera with the 1520 species... . .. .. merely owing to the chances of sampling. ...... . .. sampling. .. ... 215 90 38 35 21 16 15 14 28 20 30 32 13 14 Calculated.. n' is 8. using groups 12. 6 . .2 27*0 186 249 25*0 . ... 7 .... 5175. . there are 10 groups. 214.. .. 76 to 100 . 5 .. . .3* 6270 * The frequency of genera of 101 species and upwardswas subdivided by extrapolation.... 8 . Number Number ofI of species in genus.... 9 to 11 . ..6 489 32. 3 ..... If we regroupthe frequenciesof TableV accordingto the runs of sign of the differences from expectation. ... ... UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION... 4150. For the Chrysomelidce TABLE V.... .. .. ..... 101 upwards.. 4. 4 . ..once in some five or six trials. hIR G.. The signs of the divergences from expectation are fairly well scattered except over the range 7 to 30 where (for the given grouping) all the observed frequenciesare slightly in excess of expectation. .... Reference to Tables V to VIII fully confirmsthe impression given by the values of P.6 238 183 14... 2 . 3140. and genera of 101150 species..7 12. ..9 85. 3. of which there are only 7 against an expectation of over 12: of all genera with more than 100 species there are only 16 against an expectation of 236... Observed. so that the distribution stands the severe test very well. .
6 . . . . ..5 1.. .. It is true that we have passed from a family and a subfamily amongst the beetles to the membersof an entire order.BASED ON THE CONCLUSIONS OF I)DR.... 9 to .. 51 to 65 . 8 ... WILLIS..024 2 1 . 7 ..... .. ..1 .......... 2 .. .but the comparativepoornessof the fit can .R... 66 upwards Total TABLE VII. ... 7 . 5 ..12 8*3 65 1 ..... 12 to 14 . 3 . 12 to 14 . C.0 7 0 9.Snakes: observed and calculated numbers of genera of each size.. 4 ... 6 . 21 to 30 .024 genus... . 469 152 82 61 33 36 18 17 36 23 40 21 15 8 4 9 1.2 8........... ... 131 35 28 17 16 9 8' Calculated. 35 upwards Total .  Number of species in genus.... 9 to 11 . 31 to 40 ...... Calculated... and the data (Appendix. ... . 15 to 20 .. Observed. .. ... 2 . 1309 47 2 252 160 1..Cerambycinc observed and calculated nuinbers of genera of each size.... Number of species in Number of genera.... ..... . . . 5 .8 9 2 6 2 2930 For the Snakes the fit is clearly less satisfactory. 57 TABLE : VI.. . 4690 165 6 866 54'0 372 274: 211 168 35*1 224 27 4 24'1 130 8.. Number of genera. Table C) are irregular.. J.. .S. 8 .... ... 3 4 . 8 13 3 7 14 4 293 I I 52 11... 41 to 50 ...1 7. 21 to 34 . 15 to 20 . .. F... Observed. .. .. .
But apparently the formulaarrivedat can do far more than this.. . . .. .. and all the varied changes that have diversifiedthe organic history of our planet.but such a deficiency can hardly be held to be proved by the present tables.. more closely indeed than we have any right to expect. Why this is so seems to me to be a point which requires some discussion. One might well have expected the personal factor in classification. the calculated figures being extraordinarilyclose to those observed......the lack of precisionin interpolationhas partly contributed to the result. 259 2590 hardly be due to this cause. ' Number of species in genus. for it will be seen that the mean of the fitted distribution(Table IV. for the Lizards give the best fit of the four tables..... . .. Observed. In the case of the Chrysomelidc. Calculated...... .. to have left so many irregularitiesin the distribution that any formula could at most have given a very rough analogy with the general run of the data...moreover. ..9 399 222 145 10. . lines 3 and 8) is slightly too low. . So far as the tests go I think it must be admitted that the formula given is capable of representingthe facts with considerableprecision. . the unit of time being ... and the experience too limited to suggest a general rule. . Desire of the systematist to breakup a genus whichhe regarded as unwieldy might well tend to cause a deficiencyof very largegenera.7 9. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. the practically cataclysmic destruction of species at numerous epochs in geological time. TABLE VIII. ... G. Having shown.. 104...... . It may be noticed that in each table the numbersof the largest genera. I return to it briefly below. we can revert to Table IV. . . 8 . .. are in defect as compared with expectation..... but the deficiency is very small in one case at least (Table VI)...8 10. 6 4 13 4 10 11 6 64 52 114 7.. . that the formula is capable of closely describing the facts.. . . 9 to 11 12 to 14 15 to 20 21 to 34 .. ...5 80 1 2 3 4 5 6 . .58 MR. . It will be seen in the first place that the values found for range only from 4 26 to 6 28. 105 44 23 14 12 7 7 ..... however.. and the actual values of p and T.Lizards observed and calculated numbers of genera of each size. Number of genera.those in the terminal group.8 77 35 upwards Total .
but short: it is indeed obvious. a genus of 20 species may be reduced to a genus of 5 species. ." e. J. the postcataclysmic distribution will or will not also be closely fitted by the same formula. as we have found. In every case the number of genera at zero time required to fit the data is very substantial: roundly. It is desirable to carry out such a test on assumed data to see whether. a number of primordial genera No in excess of unity: and further he will have to remember that the time T will be measured. killing off a large number of species. in highly simplified terms. we must expect matters to be far more complex even than this. Considerthe effect on a distribution of the advent of a " cataclysm. if the precataclysmic distribution be of the form above derived. but from an initial number of N0 monotypic genera instead of a single one. of course. for the Lizards. therefore. On the passing of such a cataclysm of the utmost possible severity and the restorationof conditions favourableto life. For any species the chance of survival is p. with a maximum frequency for the monotypics. to calculate the postcataclysmic distribution. evolution will start again de novo. In the limit. be regarded as not merely finite. but I think it may have another significance and one which will render the interpretation of the figures a matter of difficulty. F. of a glacial epoch. When. What does this mean ? Such a result would presumably be shown in any case where the origin of the group was polyphyletic. 2. n species surviving out of a genus of n species will then be given by the successiveterms of (q + py). its equivalent in years or in geological time will vary from group to group and can only be determinedif the results can effectively and without fallacy be collated with the geological record. of and the existing number for the Chrysomelidca the Snakes. . if the cataclysm be very severe. Given the precataclysmicdistributionand the value of p. and I am inclined to suspect that it may be fairly closely of the same mathematical form. most genera will be killed out entirely and no genera will be left with more than a single species. 1. but have not yet carried . and. not from the origin of the group.. though a lengthy piece of work. It may be as well to emphasisethat our unit of time being a relative unit. but from the passing of the cataclysm. somewhat as follows. the biologicalstatistician examinesthe frequency distributionfor sizes of genus in the group evolved from the survivorsof the cataclysm. But line 7 of Table IV suggests that great caution will have to be used in interpreting doublingperiodsin terms of geological time. Every genus will be reduced in size.g. what is the effect on the constants.S 59 the doublingperiodfor species within the genus in each severalcase. Time must. nearly 15 per cent.long ages after. In fact. .R. of destruction q (p+q= 1). C WILLIS.that when the mean size of genus is only some 5 to 15 species time cannot well be anything but short. a genus of 10 species to a genus of 2. if it is closely fitted. 5 per cent. some 10 per cent. The action of a " cataclysm" of less than limiting severity might be represented. and so on. Somepossibly manygenera will be killed off outright.BASED ON THE CONCLUSIONS OF DR. The chance of 0. of the existing and number for the Cerambycince. he will therefore find. once the point is considered. It is evident that the distribution will remain of the same general form to the eye. it will then be possible.
750 1. . .. The postcataclysmicdistributionwould then be compound as regards time.. .. I have.... ..167 1. Using the method of interpolationdescribedabove. we would find for the compound distribution the constants . 2. The fitted distributionis.... . it will be seen. taking the total as 10. . =. It would be further desirable to find the form of the postcataclysmic distribution after a subsequent period of evolution... 3 . .. These are problems still unsolved.335 . The first 10 terms of the respective series work out as follows. .60 MR. 203 149 110 229 189 159 137 253 19 36 154 123 ! 225 84 153 131 .. .162 7641 2 Corupounid.. .. Suppose that at the time of observation the constants for the two portions are:p=21 p 2. carriedout two brief empiricaltests to find the effect of compolunding the distribution...... 775 540 385 510 371 286 642 455 335 525 375 285 7 .. 2 4..the fitted to frequenciesbeing badly in defect over the range fromnf2 f inclusive..... however. . 6 . a poor fit to the actual distribution. . ..4444 803 4 5. Over this range ..2J '~ 8 and that there is the same number of genera in each of the two groups. 9 . 3. have p 2 2 We will then 41667 0)*41667 0 33350 037508 M 3000 p"2 31000 17'000 Compound of the values of fi and M for the compound distribution being the mxeans the values for the components..... .000 : p == T1. one portion starting de novo. if the cataclysm were of limiting severity. ...= 7203 p = 1.335 1.626 963 8 3.917 1. one portion remaining untouched.751 1... TTDNY YITLE ON A MATHEMATICAL THEORY OF EVOLUTION. out the test. 278 .. 8 ..668 T tends accordingly to be nearer the higher value of the time and p is reduced below its true value. . A cataclysm mightor probably would in generalcover part only of the area occupied by a given organicgroup. 1 668 7 203 3..... G.. 10 .....
I find:Sum of f2 to f8 inclusive.. CCXITIIB.. what is the effect of a distribution being compound as regards p ? In all the precedingwork we have assumed p to be the same for all genera in the group.. 8 9 .. ...3 _~ ~ X= 2 1603 ~ 8 Compound... 8..... Snakes . an assumption which is on'the face of it very unlikely to be in accordance with the facts..... For the four Tables V to VIII..7 .. 3 ... Cerambycinea...335 1.... Observed. VOL.8 043481 0. . To carry out a test on this point I took the two distributions p 13 C p . F. ..470.. .. but the excess is relatively small..335 764 510 371 286 229 189 159 137 " 3.8 and assumed them to be compoundedin equal proportions...348 1.the fitted frequenciesbeing in excess. 10 . C. I find approximately T 8293 Here p takes a value between those of the components. 4 . ..33350 12247 31000 T ..R..465 809 525 373 282222 180 150 127 The misfit over the first part of the range is now reversed.......giving :M f.......841a deficiency of 14 per cent.. . Chrysomelid&e ...8 c .842 1. p= 1.......453 798 517 367 277 218 177 148 125 3.20 . .... 61 observation would give a total frequency 4...... 7 .. Lizards . K 121 1'10 106.S..... . while the fitted distribution would give only 3.. . . The first ten terms of the respective series run as shown in the table below.. p 13 p =20 .. 5 .. ...8 Conm. . .......1 603 0384155 216235 Fitting the compound as before...ound JL p . 2 ..842 1. .....BASED ON THE CONCLUSIONS OF DR... .. 229 399 2361 408 7 1119 6 ...... WILLIS. J. .. 6 . ... 4.... But the question also arises...570 833 524 364 269 208 166 136 114 3.. while X is thrown up above its true value..... . ... Calculated.293 8 1 ...
expectation exceeding observation for genera of 2 and 3 species and being continuouslyin defect of the graduateddata from genera of 4 to genera of 34 species. WILLIShad compiled an estimated distribution for the aggregate of all the flowering plants. (26) ) Writing for brevity 3/ap =a a/aT . . The four tables given as illustrations in this section were the first. this discussion is at least sufficientto indicate. A good deal of the preliminary work with purely graphic charts was done with data for the flowering plants. since mere inspection of Table III showed that alterations of f. This became evident at an early stage of the work. is a fair fit of the formulae data to which have been submitted to a preliminarygraduation. The floweringplants have not been completely catalogued and the numbers of species in the larger genera are estimates rounded off to the nearest 5 or 10 (or even for the very large genera the nearest 50 or 100). iT . We might reasonably. . .171species. There is no clear evidence of either type of compounding. These data give p . . the fitted frequenciesbeing in excess for the first two tables and in defect for the last two. But these constants do not give a good fit. and the goodness of fit remains rather puzzling. and at the time of writing the only..571 genera and 160. but these are not satisfactory for a precisetest of theory. G. We determine? and p virtually by solving the equations f= (p.T: ) Hence O(p o p (O }.. giving 38*605per cent. which was as far as I carried the work. of course. To all the other difficulties of interpretation is to be added the fact that errors of sampling in and p are very high. . = 6 484. that the interpretationof the values of p and T arrived at in any given case or series of cases is not a simple matter nor likely to be quite straightforward. expect compoundednessas regards both elements.594. In these circumstancesall that can be expected. Brief and inadequate as it is. . 12. at the best. . UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. No = 750 approximately (750 to 756). The conclusionis fully confirmedby the following investigation. The group is so heterogeneousthat I do not think a good fit was to be expected : but the type of misfit is not very clearly analogous with either type of misfit shown in the two illustrations above for T compound and p compoundrespectively.alba = . The differencesare all small. tables which I had completely fitted: they are in no way tables selected to show exceptionally good fit. I hope.b1  a^ a1ylaT D M26X. Dr.62 MR. and M well within the limits of fluctuations of sampling would produce relatively large changes in r and p.T) M = (P. . of monotypes and mean species per genus 12741. .
and r.+M (1 p)l} (1 + p) pl (pl)pl . and it seems best to conduct the calculation by working out the values of the partial differentialcoefficientsfrom (28). now.f. b2.. The following statement shows the results of the calculation for the four illustrations used above. If.I)I (1 + aOpi) . f3. tional frequencies. Whence. is . dM) _ dp k(b1dM .. which would reduce the standard error..f (7maf)J Where a. f f8/(lf) Af+ . . and the standarderrorof p of the order of 0 1 to 0 2. a2. () I have failed to get any effective simplification by inserting the values of a1.1) (1 f) . K 2 . 63 and solving (26) for dp and dr we have d = kl1 (a2 dfi .R. C2= a2 =.p1 (l + p)I (1 t aOp1) +fip1 + p1 (p . f f3a/(lfi).Mp{1 + p)i + iap1} 1)i . (27) bfay2 2b1b2 r... f2.a. so that the new distribution will be . .I) (1 A) Hence the regressionof errorsin M on errorsinf. C. very desirable.BASED ON THE CONCLUSIONS OF DR. (28) (28) + (pl)1}1a and the only undeterminedquantity in (27) is r. there is an error 8 in f the compensating errorsin the remaining frequencies will most probably be distributed over those frequenciesin proportionto their respective magnitudes. J.( fi ) where C is the standard deviation of size of genus: that is rr ~~ (1 ).. . f2 f2a/(1 fi). checling from Table III to see that they are approximately correct.(Ml) a a N1 (Ml)f .(M . For the time especially the determination is wholly lacking in precision: the high standard error renders a better method of fitting. { (p  b = f b. I find a.b2 df). F.1) (1fi)(M . Iff1.so that fi + f2+ +f + ? ? ? +f .. squaringand summing. It will be seen that in each case the standard error of T is of the order of half a unit. af are the standard errorsof M and fi and rnf is the correlationbetween errors of sampling in these two constants.f in (27). are the propor we have fi+2f2+ 3f3+ .1 a = . = f . and inserting the numerical values in (27). b. WILLIS./2 (b~:?2 J2 (a12 2.a`If2 2 2ala2  rmf Gnf)l . +nf = M.S. the mean of which reduces to Mor (M.
= 05430 . UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. THE FREQUENCYDISTRIBUTION OF AGES. for chance enters largely into the matter. in fact.f = . 37.. f.41762 a2 b = +125775 ap = 0164 a = 92945 r.. Lizards p IV = fi= 040541 M m 61004 0 . FOR GENERA OF A GIVEN SIZE. with assigned values of p and .015291 a . to determine what proportion of genera of n species.00052924 =0576 a. are likely to possess any assigned age.260 . and so on. = 0342903 Chrysomelidae N= 627 } p = 1925 T=. .0 3637 af =001557 . Thus for monotypic genera the distribution of ages is given by (gs)x e ym Yo Y . p.498 f 00458008 M = 55840 a = 0 3621 a =115871 .. genera of 3 species older than genera of 2 species.1 bl 018975 4. Cerambycinae N= 1024 p =1188 =.02335 C=.. On an average. On the assumptions we have made the size of a genus is not an absolute measure of its age.00055153 bl a 0563 a b2= r..1253 p r fi = 0 447099 {0 =002905 M= 50341 .020614 ai b =T rmf a  61426 00012378 b2 = +11960 0408 a. but it is an index to age.8655 a 13*9294 259 I 1496 4'281 ! =003051 . so that y (g +s) eg+s)'). It is therefore of importance to determine the frequency distribution of ages for genera of a given size.u e where yo must be assigned the value that will make the total unity. = 0077 Snakes N293 . already been given under the integrals of equations (9) and (11) at the beginning of Section II.b. ETC.. i.6'28 M = 159442 m a= 001896 a1=011550 J bl 00010771 0418 a a2= b2= 18464 116068 + 56706 0167 r.0 3023 3 8879 + 1 8652 == 0211 V.f = = 462341 .e. genera of 2 species are older than monotypes.64 MR. since the main effect of limiting time is simply to truncate the frequency distribution of the ages of the derived genera at age T. The required expressions for genera of 1 and 2 species have. AND THE MEAN AGE. G. 0 3903 a. within a homogeneous group. We will first take the limiting case when time is infinite.
I + np a l+ p + ..^ in terms of the doublingperiodfor species within the genus. time being infinite. and so on. 1 496 found for TablesV to VIII). when n = 1. 1 188. 10 species... F. as alreadypointed out.S.ea(. a simple exponential.from a frequencyzero at age 0. (1 . ((log 2)1 pH (+( 1 eP 1  1 _3_I_ log (l + 1 p) log (1 + )} . and the modal age naturally increases with the mean age. The mean age is 0 87 of a doublingperiod(Tale IX). The mean age is 1 41.+p'. (33) From (31) and (33) it will be seen that for a given size of genus the mean age and the standard deviation of ages (measuredin terms of our relative unit of time) are both the larger. 1. modes and standard deviations with size of genus. and the standard deviation is 0. and Fis the age measured wheref. and consequently the most probable age of monotypes is zero. the relative frequencies of ages 0. For mono typic generathe curve is. e" as before is 2. Generally. * * (31) .for genera of n species. 3. fig.R. 10 shows the age distributions for genera of 1. (32) the modal frequency being M / 1P) n \lnp/ n +f npo For the standard deviation of ages of genera of n species.(g+s) (1 . When n is 2 or more. C. .the most probable age being zero. . the greater p .. is given by y gf' e.esX)~l l (0) is the nth term of the pseries(12). Genera of 2 species show quite a different form of distribution.. is (time infinite) f. the modal age is (common logarithms) .. . with finite slope. The mean age of genera of n species.e)~1 . . and 0 036. considerablyin excess of the mode. P) .. p2 (12 (1  1 2p)2 ( + . 10 has been drawn and the data of Table IX calculated with p = *5 (nearlythe mean of the four values ! 925. the proportiony of genera of each age x. It rises abruptly. 0364. 1 253. I find 2 . Fig. 2 and 3 units are 1155. 65 Similarly for genera of 2 species y (g + s) (g + 2s) sl e(g+s)x (1 eX). . WILLIS. in terms of the same relative unit of time. . rises to a maximumat age 0 68 (TableIX).87 also. ..2p' As regards the most frequent or modal age. 2.BASED ON THE CONCLUSIONS OF DR. J. = P1 J. (30) reduces to a simple exponential. 2 0 2 a2 .. As an illustration of the general character of the limiting frequency distributions of age and of the variation in the limit values of means. 0*115. and then tails away slowly..
UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. Fig. 10. G. I.66 MR. p ==1 5 (cf. 1 10 species when .A 5 .Limiting forms of the frequency distributions of age for genera of 1.. time is infinite. 3. 1 l 0 2 l Age of derived qenera 3 l 4 l 5 l 6 7 6 Species [ in Genus L I i 1 2 I I I I 4 . Table IX). . I I  I 7 8 1 9 O0 0 2 3 4 5 6   I' 7 . 2."I  t 6 .
. and thence calculating the mean age for genera of 20 species by (34). . .) ]} .. . . . .. which is carriedup to genera of 100 species...R. . Using the true mean by (31) at 10 species. . but slightly les. ... we have to a high degree of approximation?)Z2 pa~l X (1 + np)1  ni+l p"l (1 +pz)1 dz .40).... WILLIS.. (n2+) p log [1 (n an expression which was used for calculating the means for genera of 40 species and over in Table IX.. . the error is only 3 in the fourth decimal place. 1 2 3 2 1 5.14 149 177 200 220 238 254 268 363 420 461 493 . ... . . . Unit of time.... but mean and mode increase with the size of the genus. .8 1 20 121 122 123 1'26 128 1 28 129 1 29 129 625 640 592 129 129 129 For genera of 3 species the distribution is tangential to the base at the start. 141 180 4 5 6 7 ... . (34) (log 2). at first fairly rapidly and them more slowly.14 470 510 542 0 068 1. . ... . 567 589 608 519 5*41 560 5. ..S..BASED ON THE CONCLUSIONS OF DR. .. the mean age of genera of 100 species (6....... .. .. the frequency is a maximum for the modal age 1 14. .. It follows then as n becomes large the mean age varies nearly as log n.. .. . 087 . is very nearly double the age of genera of 10 species (3 22). F. and the mean is 1 80. . J. .... .Limit values (time infinite) of mean and modal ages and standard deviations TABLE for genera of 1 to 100 species: pspecies within the genus.. 67 IX.. .77 1'16 1'1.. . C. say over 20.. the doublingperiodfor 3 4: Standard deviation of ages. . . (modal) age.. From genera of 3 species onwards the general form of the distribution remains the same.. for example. . The nature of the variation is well shown by Table IX. . 087 1 02 1 09 114 Meanage............ . . .. When the number of species in the genus n is moderately large. 1 Number of species in the genus. .. 40 50 60 70 80 90 100 293 308 322 4. . 211 236 258 277 8 9 10 20 30 . ... ....{log [1 . .. ..
. When time is not infinite but limited.. the frequency distribution of which is of the form (30) but truncated at age [ = . . arithmetically) from the distribution. it is 1 *286 for genera of 50 species and no more than 1 *294 for genera of 100 species. . As we saw at the beginning of Section III. G. is useful:2) p2( n] np)2 p(n2. the genera of any given size fall into two distinct groups: (1) the primordialgenera.p TMd q . All the above are limiting results.. which is very rapidly done on the machine and sufficiently precise except for the monotypic genera (see below).68 MR.[1 pj. .. . . ..by integration. all of age . obtained by substituting integration for summation. . . with an errorin the sixth place of decimals only. T7M. . .. The procedureadopted to calculate the mean ages of genera of each size for the four illustrations of the last section was accordinglyas follows. is the mlean age of the derived genera of n species. . (37) and if rMd. q be the proportionof primordialgenera of n species to all genera of n species q= . and as in any case it is desirable to have the frequency distributions.e.. . . . lHTere again the similar approximation. . (35) It can be safely used. the proportion of primordialgenera of nl species to all genera of everysize is C(yrs)T s(I  el) or in terms of the relative notation ea(+p)r (1 ea')L If. . . .. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. (38) Equation (30) leads to very complex expressions for TMd. . (2) the derived genera. .elpr (1 et) .. and proceedingshave continued ever since without break or disturbance. . . (36) where . . for genera of 30 species onwards. mean.. To carry the figures to another place of decimals beyond those given in the table. it seemed to me simpler to calculate the average directly (i. . then. All that can be said by way of comment at the present stage is that they do not look unreasonable. . .. . . The rate of increase of the standard deviation also falls off rapidly as n increases. true only when the time elapsed from the genesis of the primordial species is infinite. The proportionof derived genera is then given by p q . It may be particularly emphasised that the size of the genus is by no means a precise measure of its age.(Pn 2 [1 i]  I 1 +  ) ji . age of all the genera of n species M. . .. .jfn is the proportionalfrequency of genera of n species at time ? (equation 20).. . .
e. the final value of i is 6 14. up to genera of 100 species. . 0875. For the monotypicgenerathe mean age of the derivedgenerafoundby the above process is not very accurate and integration gives a simple result. (3) The mean age of the derived genera M. and the value for the centre of the final group in the tail. so that all the frequency distributions of age. The largera genus.. . (2) The age distributions ea(l+ "'P (1 ea\)n1 were calculated for the same genera.. 037.and for a genus of 100 species the odds are nearly 2 to 1 that it is primordial. 8. it will be seen that as the size of genus is increasedthe proportion of primordial genera also increases. and as the primordial genera are the oldest of all they will be mostlikely to be caught in the net by picking out the larger genera. 00 to 6 28. using values of . 1 25. given for the sake of comparison in column 2. 2 75. 2.B.BASED ON THE CONCLUSIONS OF DR. 225. 0 625. 9. F. or the odds are about 11 to 1 against a genus of 10 species being primordial. CCXII. for infinite time. (1) The proportionof primordialgenera q was found by (36) for genera of 1. at first with increasing and then with decreasingrapidity. and so on. E. Turningfirst to columns 5 and 6.p = effi(l+plvg )'a [ p)1 1)X]. VOL. 1 75. The modal ages of the derived genera are given in column 3: it will be seen that even for genera of 100 species the mode falls below the limit 6'28. WILLIS. i.S. are primordial. The mean ages of the derived genera are given in column 4. L . the older is it likely to be.TP which gives the factor necessary for the reduction under (5). 0125. the centre of the interval 6. show a true modal age for the derived genera.d. 5.was then calculated from these distributions on the machine. etc. the odds are about 250 to 1 against a monotypic genus being primordial. 3. being carried the up to genera of 100 speciesa limit of size exceeded by 16 only out of the 627 genera. where r 628.R. 10. Of the genera of 10 species only 8*4 per cent. The area from 0 to T of the curve y X^1(1 1. 4. (roundly)are primordial. and are necessarily less than the limiting values of the means. . (4) The mean age of all genera of n species was obtained by (38). and thence p by (37). For a genus of 60 species it is about an even chance whether the genus is derived or primordial. for the Chrysomelidce. averages. and the mean is M](1 + pl)Tl a+e pT(l"1)[1  e (ap1)T] Table X summarisesthe results for the Chrysomelidce. 6..g. 20. species. J. (5) In order to plot diagramsof the frequency distributions. Of the monotypic genera only 0 4 per cent.the ordinates calculated under (2) were multiplied by the factor necessary to give a total area p. 7.
..98 231 2 58 282 1 2 3 4 5 6 . . TDNY YULE ON A MATHEMATICAI. .4070 0. .5930 0. of course.21 2. ..69 5. . . .33 4.. . .96 1 55 1.... THEORY OF EVOLU...8064 07071 0..0039 0 0097 0'0168 0... .. uent age.Chrysomtelidte..... 658* 6.6'28.. ... the effect of limiting the time is slightly to increase the average :ge as compared with the limiting value in columnl2: the differenceincreases up to genera of 20 species or thereabouts andl then falls rather abruptly. . .28 559 5.38 4.21 1 57 1186 2...G. and pri^ 1. . i Modalor Limit fp freemost . .0248 0.3451 0 6549 600 * Beyond the limiting value of the age.54 5 *73 5 90 2.. . Derived i Primordial genera..9570 0.47 4 75 95 510 5 22 5. and for larger genera the mearn age at time T steadily falls more and more below its limiting value..89 5 95 3. . . .13 3.5526 0.02 0o0630 00735 0. 3.9370 0'0528 3.51 4.07 626 643* .. For a genus of TABLE X.. 11 shows as illustrations the age distributionsof genera of 1.33 5*54 5.... etc..85 6..88 5. 10..2929 0 3769 0..9158 0. .. ... 0 094 150 1 90 2. . 6 28. Fig.31 5 39 0'9472 0.. somewhere near 45 species the two averages are equal.74 506 532 5....46 0. .e Meanage . g .1 d genera. 60 and 100 species. mean . .. .of derived ..25 3.6265 3. 5 and 6 the mean age of derived and primordial genera together was then calculated by (38).. . ..65 2.26 405 4. and truncated ats t . From columns 4..TION... .3735 0. 100 ... i of derived .48 500 5. 10. . . Table showing mean ages.80 5.95 1 52 193 2 25 2251 2.. .0336 0'0430 0..49 2.. .....36 3.1. The proportionof the primordialgenera is shown by a square just to the right of the . . . .. ..31 2. asymptoting..10 1 7 . The distributions of the derived genera are of the same general form as the distributionsof fig..1936 0...0842 0.. .76 4. the values obtained being given in column 7.. 1 .. but slightly altered owing to the differentvalue of p. 1 p = 1925. for genera of each size: unit of time... ... 0. It will be seen that..6231 05534 0 4955 0447 0..84 8 9 10 20 30 40 50 60 70 80 90 .. for the small genera.4466 0 5045 0. 2 3 I 4 5 6 * 7 Number of species in genus. 2'93 2.9265 0. .n age.45 2'66 0 9961 0 9903 0 9832 0. 2. = 628.. .. Proportionof .. 310 432 2. .. genera.. .05 5. the doublingperiodfor species within the genus in the Chrysomelida. . to T.73 O 0 73 1.. ^ ! p?..79 3.... . ...70) MtR.9752 09664 0. .. . . the area of the truncated curve is also made equal to p and not unity. that is 6 28 in the present case.20 3.99 3.
IF . C. frequencyof primordialgenera (cf. 2 and 3 species the truncation is scarcely perceptible and the square is very small. Fig. For genera of 10 species the truncation becomes appreciable and the square is larger.S. the derived distribution is cut off.the squlares the right showingthe proportional Table X). 11.R.BASED ON THE CONCLUSIONS OF DR. WILLIS. 60 and 100 species.Chr'ysomelidZ Frequencydistributionsof age for genera of 1.r "L' ". thle mlodalage lies just to the left of the limit and the square has nearly twice the area of the small portion of the curve that is left. and finally for genera of 100 species most of Age of derived genera 2 0 1 ^ine l Sr in Genus I.. the area of this square being q. F. 2. The on curvesgive the age distributionsof the derivedgenera. The odds are nearly 2 to 1 that a genus of 100 species or more is one of the primordialgenera L2 . so that the square and the curve make up unit area when taken together. 10. For genera of 1. J 60 I1 100 0 I 1 I 2 I 3 I 4 I 5 I 6 . For genera of 60 species the area of the square is little more than that of the curve. 1 4 3 5 .1n I v1. J. 6 I *Primordial i genera i age 6 28 2 I I t 3 Il 10 . 71 limit.
. The tables and charts will repay some study but do not seem to call for special comment. .Cerambycina.. . the discussion at the end of Section IV).....not be of interest : the age distribuLtion such genera ought to be markedly compound...14 232 249 09580 0. .....27 1.. . deage derived and pria pr mordial... . genera.. . .. As a matter is of fact the number of genera with 100 species or more in the Chrysomelidcw only 16 (Table V) and we have to class some 65 genera as primordial (Table IV)... . 13 and 14 showing the age distributionsfor a few TABLEXI......being in part primordial and in part derived... . 078 129 167 196 0 063 106 1 39 0....yoince.. .. .. = 498... Chartssuch as those in figs... Proportionof iean ageof r_ age of of derived Primordial Derived genera. . etc. .... . ....67 1. the genera.. 12. 277 226 263 09157 00843 283 9 10 20 30 40 50 60 .. 1 2 3 4 .. .. or iModal ost frequientage. . . 11 to 14 suggest the question whether an examination of of the larger genera in some group might. 4 98. (or genera reckoned as such...72 G... ....09 2. the Snakes and the Lizards. for genera of each size.. . Tables..9701 00038 0. 292 3205 396 452 491 523* 548* 242 256 350 407 447 479 505* 275 355 3.... It may once more be emphasised that the unit of time is differentin each case if measuredin years: it is the doublingperiod for species within the genus in thegiven group. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. 221 242 261. :'MR. ..0104 00192 0299 0.91 0...... . . . .85 4 04 417 425 09004: 08846 07290 06012 05046 04310 03764 0j 0996 01154 0. .. . ..... ..... p = 1188...79 130 170 200 5 6 7 . Table showing mean ages. ..2710 03988 04954 05690 06236 1 297 311 394 430 451 463 471 . .. selected sizes of genus in the same way.89 2. .XI. . 1. 077 1. Beyond the limiting value of the age.9448 09306 0*0420 00552 00694 2*26 247 266 8 . 1 2 3 4 5 6 7 Number of species of genmes..9808 0. Is it beyond the bounds of possibility that one familiar with the group might be able t effect at least a tentative separationof the two ? . ' genuu~~~~~~~' genera..63 1.9962 09896 0. cf. iimit m mean age. figs. unit of time. .. XII and XIII show data arrangedin the same form for the Cerambycitnc.. the doublingperiodfor species within the genus in the Ceramb.
0
Species I in Genus i
Aqe of derived genera 4 t 3 2 I I 1 I
Primordial genera
age 498
0
Age of derived genera
I 2 3
4
1
Species Iin I Genus I
I
1
i
1
i
2
2
3
i
_
3
m*
10
I
30 ... . 
I
I
It I
I
40 4 4
5 o 1 2 3 4
0
I
2
3
Fig. 12.Ce ambycinze. Frequency distributions of age for genera of 1, 2, 3, 10, 30 and 60 species (cf. Table XI).
Fig. 13.Snakes. Frequency distribution generaof 1, 2, 3, 10, 20 and 40 species (of. T
74 TABLE
MR. G. UDNY YULE ON A MATHEMAATICAL THEORY OF EVOLUTION,
XII.Snakes. Table showing mean ages, etc., for genera of each size : unit of time, the doublingperiod species within the genus in the Snakes. 1 426, for
p = 1253. 1 1.
Number of.
2
3
Modal or
4
Mean
5
of
6
7
ean age
Llnilt sLimeit species ixn specisinmean age. genus.
imostfrequent age.
g Li ageProportion m it of derivedri r genera. Derived Primordial genera. genera.
078 126 162 1*88 209 226 240 253 2 64 273 329 3 54 369 0.9889 0.9709 0.9483 09227 08953 08670 08384 0(8100 0 7823 0'7552 05403 04086 03253 001ll 00291 0.0517 00773 01047 01330 01616 01900 02177 02448 04597 0'5914 06747
of derived der and prir m Tmordial.
082 135 175 206 2'31 252 270 286 299 311 373 397 407
...
2 3 4: 5 6 7 8 9 10 20 30 40
... ... ... ... ... ... ... ... . ... ...
... ... ... 2.00 ... .... ... ... ... ... ... ... ...
... ...
...
0.80 132 170 225 246 264 281 296 309
00U
... ...
.
... ... ...
...
... ..
456* 4.96*
0 0 64 108 142 169 192 212 229 245 259 353 410 450*
i

* Beyond the limiting value of the age, 4.26.
rABLE XIII.Lizards. p  1496.
1 Number of species in genus.
Table showing mean ages, etc., for genera of each size: unit of time, the doublingperiodfor species within the genus in the Lizards. T = 4 281,
2 3 I4 ean age of derived genera. 5 6 7
t nmean age.
L L
t
Modalor most frequent age.
Proportion. Proortio of of Derived genera.
0 9825
I Meanage n age of der and priPrimordial mnra genera.
00175 0.89
1
...
...
2
3
...
... ...
...
..
086 1 41 180
0
083
0 68
1 14
1 33
169 01
0 9564
9256
00436
00744
1 46
188
4 5 6 7 8
... ... .. ...
... .. ....
... ... ...
...
211 2 36 258 277 293
1I.48 1 76 200 2 20 238
1 95 2 16 233 247 259
0.8923 0 8580 08238 07904 07581
01077 01420 01762 02096 02419
i
2'20 2 46 267 285 300
9
10
...
...
...
....
.. ..
308
322
253
268
270
2 79
07270
0 6976
0.2730
03024
313
324
20
...
4 14
363
333
04796
05204
382
30
40
...
...
...
...
4 53*
5 1.0*
I
4 20
461*
357
371
03566
02814
06434
07186
403
4 12
* Beyond the limiting value of the age, 4' 28
BASED ON THE CONCLUSIONS OF DR. J. C. WILIIS, F.R,S.
75
0
Aqe of derived qenera , zi 4
I I I
Species nu Genus
1
Primordial a ge 4,28
ra qgene
3
U
10
.
I
20
I
I
o
I
e
3
4
Fig. 14.Lizards.
Frequency distributions of age for genera of 1, 2, 3, 10, 20 and 40 species (of. Table XIII).
VI. AN ATTEMPT TO ESTIMATE THE ORDER OF MAGNITUDE OF THE DOUBLINGPERIOD FOR SPECIES IN THE CASE OF THE FLOWERING PLANTS, AND THE PRESENT RATE OF OCCURRENCE OF SPECIFIC MUTATIONS.
Objection has been raised to the assumption of "specific mutations " on the ground that no such phenomena have been observed. " Though undeniable as possibilities "" we have to consider what warrant for such guesses " (as Coleus elongatusbeing the immediate parent of C. barbatus,and so forth) " can be drawn from the observed facts of variation. The answer is quite clear that up to the present scarcely anything comparable has been observed." To such an argument the reply seems to me to be that we are not likely to observe the actual occurrenceof a viable specific mutationit is
Dr. BATESON wrote in a review of 'Age and Area' ('Nature,' January 13th, 1923)
is based on the helium ratio and must be regardedas a lower limit for the time that has elapsed since that epoch. (.. . In the above illustration Dr. for the Lower Cretaceous..* In the absence of a definite figure Carboniferous. WILLIS has taken the mutations as occurringuniformly throughout the period of evolution. far too rare an eventunless or until we discover how to stimulate such mutations artificially. 1913). is taken by Dr. (42) Ignoring for the moment the killing out of species.(log 2) (a log e)" .ake the number of species y as given by . (40) Further.. p. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION.000. and for dyldt a lower limit.A e where t is the time in years. if x be the doubling period e' = 2 or X .000 species of flowering plants in 8.. WILLIS as roundly 160. 62) we may place its value at 3. . Let Y be the known number of species at the present time T. of that time. This is in fact the answer already given by Dr.... (41) For the present rate of occurrence of specific mutations (viable specific mutations) we have (di) =a (T log e) ' Y (log Y log A).." Let us look at the matter more closely. Y... WILLIS.log A) . Lord RAYLEIGH'S figure of 30 million years since the Eocene.but according to a law which gradually approximates to geometric progression(equations 21 and 24.000. let us see what values (41) and (42) give us for X and for dyldt at the present time: the value so obtained for Xwill be an upper limit. equation 24) is p/(p1): as p is apparently about 1. we should get the whole of the existing 160.5 (p.100 million years may probably be taken as sufficiently near the truth for a calculation in which we are really only concerned with the order of magni tude of the result: the very roundness of the figure will remind us that no precision is implied. at 30. But if we suppose one mutation in 50 years to survive. ... not in arithmetic progression.. * ARTHURHOLMES.' ha. so that the total number of species would increasein arithmetic progression. But we concluded that the total number of species increases. For the time that has elapsed since the origin of the flowering plants I propose to use the round figure of 100.76 MRE.. The constant A (cf.... Let us suppose that for the present very rough calculations it will sufficeto t. Then a = (T log e)~ (log Y . which is only 26 per cent. as mentioned above.T log 2 (log Y.000.000 years. p... (39) y.000. Where the helium ratio gives 146 million years for the age of the the lead ratio gives 340 millions.000 years.... 212): "Lord RAYLEIGH: estimated the period since the Eocene alone.log A)' . which covers but a portion of that occupiedin the evolution of the higher plants. used by Dr. WILLIs (' Age and Area... 50).000 years. 'The Age of the Earth' (Harpers....
M . J. for present purposes. If. either more or less continuously during the whole lapse of time or more or less cataclysmically at intervals. makes a relatively small change in these results. the doublingperiodis lowered to 6 1 million years and the rate of occurrenceof viable specific mutations raised to 1 in some 55 years. F.64 X106 5. When species are being killedout. The only effect of the more complex law of equation (24). The order of magnitude of the result is of mutations lowered to not affected at all.108 X 0301 106 ?. WILLIS. or 8 viable mutations which occurred (geologically speaking) at about the same time.000 lies between the 17th and 18th powers of 2: so that if there had been simple geometric increase in the number of species (correspondingin strictness to p = o) the doublingperiod would have been between 100/17 and 100/18 or 59 and 55 million years and nearer to the former than the latter (58 millions. If the floweringplants had a polyphyletic origin. on the other hand. 4.S.t/T 108 X 0. A doublingperiodof 6 4 million years would give between 15 and 16 doublingperiods in the 100 million years taken as having elapsedsince the genesis of the floweringplants.6. it may be noted. CCXIII. to reduce the number of doublingperiodsnecessary to give 160. the doublingperiod raisedto 6 8 million years and the rate of occurrence is to 1 in some 61 years.204 .477 and for the present rate of occurrence of specific mutations (dy\ _ 16 x 104 (5. as shown by column 3 of Table XIV. and the present occurrenceof viable specific mutations at the rate of 1 in some 57 or 58 yearsamongst all species of flowering plants on the whole surface of the globe.BASED ON THE CONCLUSIONS OF DR. as stated above). A rough estimate of the length of the doublingperiodmight have been given offhand by anyone who kept in front of him a table of the powers of 2 (as in colunln 2 of Table XIV)a useful thing to do when considering. we have for the doublingperiod ). no less than 64 million years. the killingout of species is ignored. C. On this table 160.starting from 2. and further to increase the estimated to length of the doublingperiod possibly as much as 8 or 9 million years.B. 13 or 14.R. however.questions of this kind. p is lowered 1 2. it suffices to VOL. If p = 2 instead of 1 *5.000 species from 17 or 18 to 1. on these assumptions.204 . I propose. the only consequence would be furtherto reduce the number of doublingperiods necessaryto give the required total of 160. with p taken as 1 5 is.0. to take it that. 77 With these data. Raising p to infinity only lowers X to 58 millions and raises dy/dt to 1 in some 52 years. it at once becomes doubtful how far equation (24) applies. In all this argument.5 or 16 and proportionately increase the length of the doublingperiod.000 species from 15 or 16 to 12. therefore.4343 1/57 1/57 5 The doublingperiod (either for species within the genus or for all species at the present time) is then. Any alteration in p.0477) _0174 0 0174 \d.
1a r = 1 .1.576 1 3 7 16 35 76 160 333 687 1.072 262. and wrrite in (39) as unity. (43) The total number of deaths from the beginningto the time of observationis 1) = qr + qpr2 J. . .048 4. . p2r2.. . by equation (24): p =l.256 1.  1) (pr I) 1 )(Y qr (pr I) .1.qp2  ??? q p' qr (p'r. pvr where p is the proportionof survivors and r gives the free rate of increase. . Time r Powers in doublingperiods.768 65. TABLEXIV.408 2. . I will first suppose the killingout to be practically continuous. . UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. .384 32.5.055 778.860 3. As we have just seen.though as stated in the Introduction I do not think this at all closely representsthe facts. .. .144 524.763 47.776 23. .536 131.93. 78 MR. 2 3 0 1 2 3 4 5 6 7 8 9 10 11 I2 13 14 15 16 17 18 19 20 1 2 4 8 16 32 64: 128 256 512 1. .aO in the limit when the timeinterval 0 is made very small we may write y  e(8) t.024 2.N.559.? r' .288 1.869 5.240 1. pr3. .357 388.25. . 1 of 2. Let the numbersat successive small intervals of time be 1.862 96. the only effect of such an assumption in the above case is to reducethe estimated doublingperiodfrom 6 4 to 5 8 million years. .085 A assume the law of free increase as simply logarithmic from the beginning.096 8. pr. If p.048.821 1.192 16. G.
Let us now turn to a scheme of cataclysmic killing.1) T log 2 (k log Y).=a (k dt/T ) k1 Y Y log Y (T log e).. a comparisonbetween the consequencesof such a scheme and the scheme of continuous killing will show whether extreme differences in the timeincidence of destruction on species have or have not any important effect on the estimated doublingperiod..... . (45) would give us an approximate value for k. C. we take k as 2.. (44) (45) 1). is given by X log 2 (a log e)~ = (k . and species are killed out at about the rate of 1 in 52 years. (48) Rate of killingoutof species :(dy 2\dt T _ a kc y =(k . . supposing that a series of practically instantaneous cataclysms occurs...BASED ON THE CONCLUSIONS OF DR. ..o) or 2 9 million years.... Y. the total number of species that has been killedout is equal to the number at present existing. k = 2. which would give an upper limit for k..R.. WILLIS..... (47) Finally (dY. at all events.'Y If.. if the number of species killedout in a small interval of time is half the number of new species coming into existence in that time...... J. it seems to me that such a scheme is a slightly closer representationof the facts. each sweeping off a certain proportion of the then existing species.. Y being large comparedwith unity and a and 8 very small. But unfortunately we are not at present in a position to give even a lower limit for this figure. . that is.. . merely as a numerical illustration. However crude.. for example..the period in which the number of species would double apart from the killingout. F. 79 That is approximately.)1 Y log Y (T log e)1 (b) ) Were we in a position to make even a rough estimate of the number of species of flowering plants that has been killedout. D= Let Then finally D =(k (a a)= a/a8 k... If. a=klog Y [(k) (46) and the free doublingperiod... M 2 . But this consists of two parts :Rate of occurrence new species:of (y aY= k (kYlog) Y (Tlog e)(a) .. We have now T loge]. new species come into existence at about the rate of 1 in 26 years.S. x takes half the previous limiting value (p =.. So much for the scheme of continuous killingout.
e. Hence Y = pnena.. . (53) Compare equations (47) and (51)... by log (Y/eT5).. Hence.. and Y Yn log p) (T log e)l. (49) (50) is where log p is. . The free doublingperiod X T log 2 (logY .... In (47) T log 2 is divided by log y"/k1). .. ..only p2e2"' surviving after the second cataclysm.. . very nearly D .n log p). In (51) the divisor of T log 2 is log (Y/pn). 1 ea pe e pe2a 2 p2e2a4 p2e3a _ p3e3a4 ea6.. 3 During the first interval the number of species vwillincrease to of which only pe"^ survive after the first cataclysm. Supposethen that each cataclysm kills off the same proportionq of the existing species. (ddj = (log The number of species killedout is D = qea (pne" .. . .g.. But the number of species killedout. Y pFeV. of course.. The rate is the same so long as PT is the same. . G. But eT in the first case and pf in the second is the chance. say pT..1) (peaqp1 yl/n (Y 1) (yl/n (51) (52) 1)1) or as Y is large.. . whether the killingout is continuous or cataclysmic.80 MR. = p. ... During the intervals between cataclysms it will be assumed as before that increasein the numberof species may be taken as approximately logarithmic. .. essentially negative.. and so on. so long as pr is the same it is of no consequence whether the killingout is continuous or discontinuous. a proportionp surviving: that the cataclysms recur at regular intervals 4: that there aren suchintervalsin the time T : and that observation is made just afterthe nth cataclysm. .. .= (log Y n log p) (T log e).e.. is not the same in the two cases: (45) and (53) do ..that is.(49) Whence a. 0 1 2 Number of species.. During the second interval these will increase to pe2%. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. the last glacial epoch.. A comparisonof equations (48a) and (52) shows that a similar statement holds good for the present rate of production of specific mutations.... of a speciessurvivingfromthe origin of the flowering plants to the time of observation. X is unaltered by the changed incidence of destruction.. it must be noted.. .. Then the changes in the number of species will take place as follows:Time in intervals.qp1 Y(n+l) (Y"l l) .... . pneaT..
... ...... the values arrived at for the doublingperiod and for the present rate of productionof species if there is no killingout at all.. killingout of existing species ..400 472.BASED ON TIlE CONCLUSIONS OF DR.. 4..of existing species ... for the sake of comparison with the remainder of the table.. .. (b) .3805 03805 34 3 4 27 21 17 13 1...R. 100 million years : present existing 1 in years species... . Continuous killing with same value of pr : (b) k = 1416 . 1 5..of existing species . . . . mutations....0050 00562 00562 0..500 129. .. Table XV gives a conspectus of the results for various assumed values of the killingout.. .. yyears... In the first line of section 2... ..500 177. In the first section of the table are given.... . killingout ? of existing species .......600 177..of production of specific Time elapsed since the origin of the flowering D of spe speciea s period million killedout. . I. 6. and of the present rate of productionof specificmutations on various assumptions.400 375...396 . (a) No killingout p = 1. .. J...... . . (a) Discontinuous killing : 50 cataclysms each .1?226 . killingout 0.500 375... stated below.S..901 ... (b) Continuous killing with same total killed out: k1426 .. 160.. .. F. . p 2.... (a) Discontinuous killing: 10 cataclysms each (b) Continuous killing with same total killedkillingout .8 61 575 55 52 In all the following increaseis taken as simply logarithmic (p = oo) 2.300 384....092 of existing species (b) Continuous killing with same value of pT: k ==2241 . 114. WILLIS.Estimates of the doublingperiod for species in the flowering plants. (a) Discontinuouskilling : 100 cataclysmseach ..000 04238 04444 02645 01317 00600 00173 0. (a) Discontinuouskilling: 20 cataclysms each ..... .000.7 17 32 32 30 33 25 19 16 12 10 15 15 29 29 3. killingout I of existing species .. (a) Discontinuouskilling : 100 cataclysms each . 7.400 708.. Nil Nil Nil Nil Unity Unity Unity Unity 68 64 6..0 (c) o . 5. . .. ... (45) being the limiting value of (53) when n is made indefinitely great....... . 81 not lead to the same result for the same value of pr... . (b) Continuous killing with same total killed. . .....2 p =l5 .. 1 1. 114.... discontinuous killing is assumned to the extent of 10 equidistant cataclysims each of TABLE XV. .. . C. .. 'killingout ... surviving i plants.. (a) Discontinuouskilling: 100 cataclysmseach .. .600 143. (b) Continuous killing with same total killedout: k.400 708. Present rate Case assumed: i e Total number hange of Doubling.600 06667 43 39 out: k =2.. out: k =1. . p (d) .
and 5 of the table similar comparisons are made for successively increased severity of the destruction of species. not PTthe same. so as to estimate whereabouts on the killingout onethird of the then existing species. as already stated. no less than 80. we know that a species may survive through very long periods of geological time. Equation (45) then gives k. is not possible. the free doublingperiodwould be lowered to 1 3 million years. the deaths are spread over the whole curve of increase. But another and more hopeful line is afforded by the second columnthe chance of a species surviving.at present at all events.000 of the 114. the lowest figure showing 10 per cent. with 100 cataclysms each would be over 700. the total number of species killedout table the truth in fact lies. for no species has so survived. and the chance of a species surviving for 10 million years is reduced to 04238. and possibly too small in section 5. In section 5. We know that the chance of a species surviving from the origin of the flowering plants to the present time must be infinitesimally small.600: the free doublingperiod would be reduced to 43 million years. It must be rememberedthat under 2a we assume observation just after the last cataclysm: our 160. 144) and reproduced below.g. and pT must naturally In sections 3. and of such severity as to make. That line of approach. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. But unfortunately. but the total numberof species killedout the same. hence in the second column of the table I have given for the sake of readiercomparisonp.000. On the other hand. there seems no basis at present for estimating even roughly the total number of extinct species.O?' the chance or of a speciessurviving 10 million years. Hence we might reasonably conjecturethat Tp is too large in sections 2 and 3 of the table. and (47) and (48a) give the doublingperiodand the present rate of productionof specific mutations.. G. in the present case. of survivals from the base of the is founded on a single local deposit only aind obviously Pliocene to the present timeit too much weight must not be attached to it. This would give the total numberof species extinct as 114. 4. Since 10 million years. e. It will be seen that on these assumptions Xis further reduced to 34 million years.82 MR. and the present rate of productionof specific mutations would be raised to 1 in some 12 years. The chance pT of a species surviving for the whole 100 million years would in several of the cases taken be vanishingly small. Taking.can we use it to suggest a value for pl ? .000 species would then be the survivors of 240. When the killingout is continuous.000 which existed immediately prior to the cataclysm. In the second line of section 2 are given the correspondingfigures when the killingout is continuous. only cover a single cataclysm. the present rate of production of specific mutations is raised to 1 in some 30 years. be smaller to give the same total of species extinct. which sweeps out of existence onethird of the then existent species. CLEMENT REID in Chapter XIV of 'Age and Area (p. The conjecture seems to be confirmed by utilising some valuable data given by Mrs.600 species extinct having been killedout in the final cataclysm. and the present rate of production of specific mutations would be raised to 1 in some 39 years. this is 2/3 or 0 6667.
... we can readily do so. REID'stable..875 p1 = 0 75 If then we assume killingout to have been effected by 100 cataclysms..125 . and this is much more than half as distant in time. Teglian Castle Eden ...... . Upper Pliocene . and this gives the values of the doublingperiodand the present rate of production of species shown in section 6 of the table. Percentage of extinct species (approximate).. . suggest that the more recent figures for the percentages of extinct species are too low (cf.. If 0 1 is the chance of survival from. the chance of a species surviving the cataclysm must be taken as 3/4. the zero with which the table begins)or possiblythe percentagesfor the olderdeposits too high. The only figure altered if we keep pTthe same but assume killingout to have been practically continuous is the number of extinct species.1.. though they run consistently from the top to the bottom of her table.. instead of 2/3 as in section 5.o = 00562 0.. Deposit. 83 AmericanAssociationof of Percentiages Extinct Species belongingto the ChineseNorth Plants in the West EuropeanPliocene at SuccessivePeriods. Then we have:PT 0o0. the age of the base of the Pliocene must be somewhere between 6 and 15 million years.. Suppose we call it 8 millionsprobably rather a low figure. Top of Pliocene .000 odd to 384. ... But the percentage of survivals from the Middle Pliocene is as high as 56. Taking these as approximately central figures for the Pliocene and the Miocene respectively. 63 millions for the Miocene.01 log pT =... base of the Pliocene.8 01 whence log T = . . . 0 35 44 70 90 If we can fix the approximate time to which it relates. PontdeGail .. . .... . which is lowered from 472. Reuverian ..R.135 while 0 1 log pT .BASED ON THE CONCLUSIONS OF DR... F..000 odd.. Lower Pliocene Base of Pliocene . this percentageof survivals gives the results shown in section 7 of the tableroundly nearly double the figuresshown by section 6.... Age of Deposit. Lord RAYLEIGH'S figure for the age of the Pliocene is 2 *5 million years as determined by the helium ratio... Middle Pliocene ... C. . If we call the age of the MiddlePliocene 6 millionyears. I have purposely taken the age rather low in the first case and possibly . WILLIS.2 75 p... the the chance of survival from an epoch only half as distant in time should under uniform conditions be in the neighbourhoodof v/0 1 or 0. J.... But the data in Mrs..S.32. figures which would correspond to about 6 and 15 millions respectively on the leadratio scale. Cromerian . .
If the age of the flowering plants is 100 million years. are of the nature of averages and may well differ considerably for different families and genera. The assumption of a polyphyletic origin for the flowering plants would not very greatly affect these figures. the value of PT for the first case (section 6 of Table XV). WILLIS'S conclusion in this respect. almost certainly over 1 nmillion and less than 6 millions. is almost certainly less than 1 in 10 years and more than 1 in 60 years. approximately7 X 105. that the order of miagnitude is wholly different.000 for a species survivingright through from the origin of the floweringplants to the present time. for this secondcase (section 7). amongst all flowering plants on the whole surface of the globe.84 MR. as it gives a probabilityof about 1 in 14. My calculations fully It should perhaps be added that of course the figures obtained. can be bbased on the fact that we have no experience of such occurrences. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION. is perhaps rather too low. be such exceedingly rare events that no valid argument. therefore. In any case the figures are qtite definite as to order of magnituLde. rather high in the second. say. On the other hand. or thereabouts. roundly 3X1013. seems rather too high. confirm Dr. On the general reasoning used above the value of p. But it does not seem probable . since we know from such instances as Ginkgothat species may survive through very long periods of geological time. it probably lies between 1 in 15 and 1 in 30 years. G. for a group so heterogeneous as the aggregate of the flowering plants. as it seems to me. Specific mutations must. The present rate of production of viable specific mutations. so as to give limiting results. the doublingperiodfor species is probably of tile order of some 2 or 3 million years: it is.
Chrysomelidce: Numbers of genera with 1. 2. (Compiled from Cat. COXIII. C.S. 1 1 I 1 16 1 1 1 1 Total 627 VOL. . 1 1 I 1 3 2 1. 1874. 74 Genera..B. t. 32 33 34 35 Genera. 1876. species. WILLIS. Gemmingerand Harold. J. F. I 1 I 1 1 Species. 3. 76 77 79 36 37 38 39 40 41 43 44 45 6 7 8 9 10 11 12 13 14 15 16 17 8 9 5 3 1 1 2 2 I 1 4 1 1 2 4 1 1 1 1 1 I 3 1 I I 83 84 87 18 19 20 21 22 23 24 25 26 6 8 6 6 3 4 27 28 29 30 3 4 4 5 4 2 3 1 3 3 3 46 49 50 52 53 56 58 59 62 63 65 66 67 69 71 72 73 89 92 93 110 114 115 128 132 133 146 163 196 217 227 264 327 399 417 681 2 1 1 I t 1 I 1 II I 1 1 1 1. 85 APPENDIX. J. .) Species. N ..! BASED ON THE CONCLUSIONS OF DR. XII. C. TABLE A. 1 2 3 4 5 Genera.R. XI. WILLIS t. and by Dr. Coleopt. 215 90 38 35 21 16 15 14 5 15 Species.
WILLISfrom Boulenger.. 1893. Genera.) Species.024 TABLE C. 1 2 3 4 5 6 7 8 9 ______ Genera. Junk and Schenkling. 3.. UDNY YULE ON A MATHEMATICAL THEORY OF EVOLUTION.Snakes: Numbers of genera with 1. J. WILLIS. 131 35 28 17 16 9 8 8 9 4 3 2 2 3 4 5 1 1 1 1 1 1 1 1 1 Total 293 ________________ __________________ . (Compiledby Dr. C. 2. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Genera. 10 13 15 17 18 21 22 23 26 t_____ _1__ ___ Genera. Mus. . Species. species.) I Species. G. Part 39. . 2 5 1 3 3 3 I Species. species. 3. Species..  I Species.86 MR. TABLE B. 1 1 1 1 I I i 469 152 82 61 33 36 18 17 14 11 11 4 10 9 8 i I I iI i i i 11 7 11 6 5 3 21 22 23 24 25 26 27 28 30 31 32 34 35 36 37 39 40 42 43 44 I 1 1 I 1 1 2 3 1 3 2 Ij 1 1 .Cerambycinv : Numbers of genera with 1. 46 47 49 50 52 53 57 59 66 67 69 89 95 104 107 120 125 Total Genera. (Compiled by Dr. 27 31 33 40 45 74 971 Genera. Cat. 1 1 1 2 2 1 2 1 1 1 1024 1. . of the Snakes in the Brit. from Coleopt. . 2. Cat. . 1912.
) Species. Genera. 1/0 53 55 60 65 70 75 80 90 100 110 120 150 160 170 175 220 290 300 350 400 500 1.S. WILLIS. Genera. J. Genera. (Compiled by Dr.9 J i 1 1 617 .R.Lizards: Numbers of genera with 1. F.4 4. 2.9 4. 2. 1885. 1 2 3 4 5 6 7 8 9 10 11 105 44 23 14 13 14 15 31 32 37 16 17 18 41 44 66 106 159 12 7 6 4 5 5 19 21 2 3 2 1 I. 2 1 3 1 Species. Graduated or averaged figures.2 3 13 14 15 16 17 18 19 20 21 22 23 24 25 27 30 4 2 I 1 3 3 2 I 15 1 8 i i 1 1 1 1 I 1 2 33 1 6 1 1..)  : Numbers of genera with (approximately) 1. Genera.. 1 1 I 1 1 1 259 22 24 25 27 3 1 12 1 3 2 Total 259 TABLEE. species. Graduated or averaged figures.Leguminosc with graduated or averaged figures. (Compiled by Dr.8 II I I I I 35 40 43 44 45 50 5 6 1 4 J l. Cat. Species.600 Total I 1 10 11 12 1 3 2 6 1 2 2 4 86 76 68 6 5. C. 1 Genera. 87 TABLED. . 3. . 1 2 3 4 5 6 7 8 9 245 66 36 24 28 30 7 13 27 3 20 1 2 18 34 I 36 345 254 19. WILLIS from Dictionary of the Flowering Plants.6 41 38 36 34 3. Species. Mus. 245 66 Species. . of the Lizardsin the Brit.8 16 133 113 9. WILLIS from Boulenger.BASED ON THE CONCLUSIONS OF DR. 3.. species.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue reading from where you left off, or restart the preview.