]\gt ; VIII .
Femoir on Inteqral nnctions .
By .
W. of Trinity College , Cambridge .
by Professor A. R. ) .
D. , .D .
, F.R. Beceived July November 21 , 1901 .
INDEX .
Section Page 1 .
Knon types of functions . . . . .
414 2 .
Exponentinl nlCtion . . . . . . . .
414 3 .
Gamma functions .
, . . . . . .
415 4 .
Elliptic functions . . . . . .
415 5 .
BESSEL'S function . . . . . .
416 6-8 .
Questions connected with the snbiect of the paper . . .
416 HISTORY OF THE SUBJECT .
9 .
Development of the theory by , LAGUERPE , POINCAP .
E , HADAMARD , . . . . . .
417 10 .
Scope of the development made by the present memoir . . . .
CLASSIFICATION or fCTIONS .
11 .
Definition of integral function .
Its formal expression .
The standard reduced simple integral function with a single simple sequence of non-repeated zeros . . . . .
419 12 .
Definitions of genre , order and convergence-exponent . . . . . .
420 13 .
Functions of infinite genre . . . . . . . . . . .
420 14 .
Functions of algebraic and transcendental seqnence of zeros . . . . .
420 15 .
metions of zero order . . . . . .
421 16 .
Density of the zcros of a function .
Character of the zero-lines .
, .
421 17 .
Product of simple functions with symmetrical sequences . . . .
422 18 .
Simple repeated integral nctions . . . . .
422 19 .
Their order . . . . . . . . .
423 20 .
Definition of multiple integral functions .
Their classification . . .
423 21 .
Other types of integral functions\mdash ; ring functions . . . . . . . .
424 22 .
Impossibility of quite general laws . . . . . . . .
425 ( 319 .
) PART \mdash ; THE THEORY 0F DIVERGENT SERIES .
Section Page 23 .
arithmetic theory of asymptotic series . . . . . . .
426 24 .
The functional theory of divergent series with finite radius of convergence . . .
427 25 .
Domain of existence of the integral ' sum\ldquo ; . . . . .
429 .
The serics and . . . . .
429 27 .
The allied problem .
References . . . . . . . . . .
430 28 .
Series which are truly asymptotic .
Character of the problem . . . . .
430 29 .
Asymptotic series of the first order and their summation .
9 . . . . .
431 30 .
The " " sum\ldquo ; and series are arithmetically dependent . . . . .
432 31 .
Such an asymptotic series can be differentiated . . . . .
433 32 .
Series of higher orders . . . . . . . . . .
434 33 .
Their summation by the repeated exponential process . . . . . .
435 34 .
The result has arithmetic dependence . . . . . . . . .
436 35 .
Fx mple colmected with the extended Riem function . . . . .
437 36 .
Some properties of an extended hypergeometric function . . . . . .
438 37 .
Its tion to the summation of divergent series of finite order . . . . .
440 38 .
xtension of the theory to series of any order . . . . .
441 39 .
L'xample of the previous theory . . . . .
441 40 .
Contrast between the two summation processes . . . . .
442 41 .
Application of the theory to the MACLAURIN sum formula . . . . .
443 42 .
, , , , , , series . . . . .
445 43 .
The re-arrangement of asymptotic series . . . . .
447 44 .
The theory of the asymptotic expansion of an integral function near infinity 448 45 .
Such expansions may differ in form in various regions round infinity . . . . .
451 46 .
Conclusion . . . . .
451 \mdash ; THE ASYMPTOTIC EXPANSION OF SIMPLE INTEGRAL UNCTIONS .
47 .
Order of progress .
48 .
The MACLAURIN constants and integral limits 49 .
Their expression , when , in terms of fimction 50 .
Example of the function 51 .
The remainder expressed as a definite integral 52 .
Asymptotic expansion of the function 53 .
The nature of the expansion 54 .
Asym expansion of the general function of order less than unity 55 .
Consideration of the integral which gives rise to the dominant term 56 .
Case and . . . . . . .
57 .
Functions to which S54 is applicable 58 .
Asymptotic expansion of a function with an algebraic sequence of zeros .
59 .
The dominant term of the expansion of 47 .
Order of progress .
48 .
The MACLAURIN constants and integral limits 49 .
Their expression , when , in terms of fimction 50 .
Example of the function 51 .
The remainder expressed as a definite integral 52 .
Asymptotic expansion of the function 53 .
The nature of the expansion 54 .
Asym expansion of the general function of order less than unity 55 .
Consideration of the integral which gives rise to the dominant term 56 .
Case and . . . . . . .
57 .
Functions to which S54 is applicable 58 .
ymptotic expansion of a function with an algebraic sequence of zeros .
59 .
The dominant term of the expansion of MR. E. W. BARNES ON TNTEGRAL FUNCTIONS .
Section Page 60 .
Application to functions of zero order . . .
466 61 .
The function . . . . . .
466 62 .
Deduction of function from that in S61 . . . . .
467 63 .
Note on the indefinite integral of S55 . . . . .
468 64 .
The further theory of functions of zero order . . . . . . . .
468 65 .
metions of finite non-integral order greater than unity .
The metion Q . . .
66 .
Asymptotic expansion of the general fimction of non-integral order greater than unity .
471 67 .
Analogy with the former expansions . . . . .
472 68 .
Application to metions with algebraic sequence of zeros . . . . . .
472 69 .
Case when the coefficients become infinite . . . . . .
474 70 .
Functions of finite integral order not less unity .
The function . . . . .
47 71 .
Deductions of the previous expansion as a limit of former results . . . . .
477 72 .
Construction of an extension of the gammn function . . . . .
478 73 .
Asymptotic expansion of a corresponding fimction with algebraic sequence of zeros .
479 74 . . . .
, 480 PART ] PTOTIC 75 .
Introduction . . . . .
480 76 .
Asymptotic expansion of the general simple repeated integral function of finite order less than unity . . .
, . . . . . .
481 77 .
The asymptotic expansion of , where . . . . .
483 78 .
Deductions from the previous expansions . . . . .
484 79 .
Asymptotic expansion of the general simple repeated integral function of finite nonintegral order greater than unity . . . . .
485 80 .
Application . . . . .
485 81 .
Simple repeated functions of finite integral order . . . . .
486 82 .
Verification of the asymptotic expansion of the -function . . . . .
488 83 .
Asymptotic expansion of a repeated function with transcendental index . . . . . .
84 .
Case of integral order .
Conclusion . . . .
490 V.\mdash ; APPLICATIONS THE PREy1OUS ASYJMPTOTIC EXPANSIONS .
85 .
The number of 7 , eros of a function within circle of given large radius .
The } of the gamma fimction . . . . . . . . . .
491 86 .
Case of the function . . . . .
492 87 .
The formula , when is not integral . . . . .
493 88 .
The case when is an integer . . .
89 .
Repeated functions 90 .
Integral metions formed by the sum or product of simple functions .
494 91 .
BOREL : S extension of PICARD'S Theorem .
MR. W. ON GRAL FUNCTIONS .
Section Page 92 .
Resemblance between an integral function and its derivative .
The extension of ROLLE'S Theorem . . . . . . . . . . .
495 93 .
is of the same order as and an expression for its zeros can be theoretically obtained .
Verification when . . . . .
496 94 .
The general theorem . . . . .
497 95 .
The nature of asymptotic expansions which involve a parameter . . . . .
497 96 .
Proof of S94 in the case of function of finite non-integral order greater than unity with algebraic zeros . . . . . . . . . .
498 97 .
If be an integral function of finite order with real zeros , can at most have only zeros . . . . .
499 98 .
The the zeros of lie adong any line through the origin . . . . .
500 99 .
Possibility of further applications . . . . .
PART I. Introdu ction .
S 1 .
Since the fundamental discoveries of , much progress has been made with regard to uniform transcendental functions ; but the advances of modern mathematics appear to have included no attempt formally to classify and investigate the properties of natural groups of such functions .
Consider , for instance , the case of transcendental integral functions which admit one possible essential singularity at infinity .
They form the most simple class of uniform functions of a single variable , and yet of them we know , broadly speaking , the nature of but four types:\mdash ; ( 1 ) The exponential function , with which are associated circular and ( rectangular ) hyperbolic functions ; ( 2 ) The gamma functions ; ( 3 ) The elliptic functions and functions derived therefrom , such as the theta functions and APPELL'S generalisation of the Eulerian functions ; ( 4 ) Certain functions which arise in physical problems ( such as ) whose properties have been extensively investigated for physical purposes .
There are , of course , isolated examples of other types of functions ; yet , broadly speaking , except for algebraic polynomials , the four types just mentioned comprise extent of our knowledge .
S2 .
Take now an example of the first type of function .
We may write , and hence we have ; so that when is very large , the approximate value of is long as .
W. ON INTEGRAL FUNCTIONS .
That is to say , for } ) ) in the of which are not at a finite distance from the zeros of , this function admits what we may call the asymptotic expansion , and similal propel .
tie is true of all functions of the first class .
S S. Consider next the secolld class of.functions .
We have as the simplest example Since the time of TIRLlNG it has beel l known that , when is a large positive integel terms ] leglected ivolving exponentials of lower order than those retained .
1889 proved that this asymptotic expansion is valid for all values of in the region of , except those which are at a finite distance from the zeros of .
By a different method it is ) ossible to establish both STIELTJES ' esult and the analogous theol.em that the double galnlna functiont in which ) dmits the ) This expansion was shown to be valid for all points in the plane of the complex variable near infinity , which are not a finite distance ) the zeros ) the function .
A similar theorem is true for multiple gamma functions .
S4 .
As regards the elliptic functions and the integral functions associated with them which constitute the third type , there are no in the urht ) infinity which are not at a finite distance from the zeros of the function and no asymptotic approximations are known to exist .
* Liouville ' , vol. 5 , pp. 425-444 .
See a paper by the , 'Phil .
, vol. 196 , pp. 265-387 .
S 5 .
The best known example of the fourth type is BESSEL'S function It is evident that is a uniform function .
The investigations of POTSSON , *STOKES , LIPSCHITZ , and JORDAN , S have finally led to rigorous demonstration by the latter that asymptotically , when is real , , when is positive , and , when is negative .
The complexity of this esult is reduced by the transformation - or which gives , an integral function of And now we have for the asymptotic value of the unique expressio1l which is valid for all values of between and This at once that , qua function of , has no imaginary roots which are not at a finite distance from the negative part of the real axis .
In point of fact , these roots are known to be real and negative when Hence the asymptotic expansion for is valid for all points in the neighbourhood of except those which are at a finite distance from the zeros of the function .
S 6 .
The question now forces itself upon us:\mdash ; " " Do functions oj a single variable of for points but those )in the vicinity of the zeros of ) functions * POISSON , 'Journal de l'Ecole Polyt , 1823 , pp. 349 et seq. STOKES , 'Camb .
Phil. Trans vol. 9 , 1856 , pp. 166 SCHlTZ , 'Crelle , ' vol. 56 , pp. 189 et seq. S JORDAN , 'Cours d'Analyse , ' 1896 , vol. 3 , pp. 254-274 .
When is negative and between and in absolute value , there may be a finite number of imaginary roots of , but these are not associated with the essential singularity . .
MACD0NALD , Proc Lot ( .
Math. Soc vol. ) ) ) .
W. ON The present memoir is devoted to answer of this stion ; the question } closely connected with other ) of S7 .
Soon after WEIERSTRASS , in 1876 , publislled his great theorem relating to the formation of uniform functions with assigned zeros , remarked the fundamental nature of the number of terms in the function which is necessary to form the " " prime factor The number was by him termed the " " genre\ldquo ; of the function ; and the at ' ' Is the genre of a function equal to the genre of its derivative ?
\ldquo ; ' ' Is the sum of two functions of the or different a function of enre equal to the common genre or equal to the respectively ?
\ldquo ; S 8 .
Again , by ROLLE'S Theorem it is that the roots of cl1ly algebraic equation , 40 , separate , and are by those of Is this true when i ) ( x ) is an integral function ?
Closely connected with this enquiry is the further If the oots of are all real , are those of real , in the case when is any integral function Again , it is evident that the more quickly the zeros of an integral function the more quickly will the TAYLOR'S series for the function converge .
Can any nection be discovered between itude of the fficients of 's series and the ] ) sion for the of the function it rep ] ?
In other words , if we ars giyen term of series for an integral function , can we cietermine the nature of its zeros All these questions fundamentally depend on the asymptotic approximation for the function .
The of the latter ) to ssifcy the of the ttcgral of S 9 .
As already remarked , WEIERSTRASS founded the theory of transcendental integral functions by constructing functions with any assigned zeros .
invented the term " " genre\ldquo ; to denote the number of terms in the associated in the prime-factor\mdash ; and for functions of genre and 1 proved that the real roots of the transcendental integral function are separated by those of He also proved , as HERMITS had previously proved for that if the roots of ' are real , those of are real , provided is of " " genre\ldquo ; This question is not formally considered in the present memoir , the expansions which obtained , although they will give closer inequalities than any hitherto published , must be still further developed before inequality can be replaced by that asymptotic equality which alone bc a complete solution of the problem .
, " " Theory der eindeutigen analyt .
Funct ' Gesamm .
LAGUERRE , 'Compt .
Rend vol. 94 , pp. 160-163 , ; vol. 95 , pp. 828-831 ; v-ol .
98 , pp. 79-81 .
S HERMITO , ' Crellc , ' vol. 90 , p. 336 .
V0L .
His principal proposition is , 'If , as tends to , a very great value of can be found for which the limit of tends uniformly to the value zero , then is of genre ' Shortly afterwards , POINCABf*gave further criteria for the genre of a function , and made the important step of pointing out that the near connection between the genre of the function and its behayiour near infinity lead to an approximate determination of the magnitude of the general term of the TAYLOR'S series for the function .
After a succession of notes by CESARO , proved that the derivative of a function is of the same genre as the function itself ) , and HERMIT , S the subject remained in abeyance until HADAMARD in a memoir crowned by the French Academy , gave a valuable extension of POINCARfi'S results .
The latter had proved that in the TAYLOB'S series for an integral function of genre , the coefficient of multiplied by the root of !
tends to , as indefinitely increases .
HADAMARD proved that , if the coefficient of is less than , the function is , in general , of genre less than .
He also showed that when the coefficient is of order , where is not an integer , the function represented by the series is of genre , designating by the integer immediately superior to Finally , BOREL , continuin researches , introduced a more precise notion than that of genre ( S 12 ) , and attacked the difficult problem of functions of infinite order whose convergence is very slow .
[ added 20th , 1902 .
] In his recent text-book , " " Legons sir les Fonctions BoBEL has given a valuable precis of our present knowledge of integral functions .
And a paper by MELLIN has recently come to my notice , which should be carefully read by all interested in the subjects with which the present memoir deals .
S 10 .
The present contribution to this interesting theory differs from previous investigations in that it is shown to be possible to substitute actual asymptotic equalities for the inequalities which have been previously obtained .
* POINCARE , ' Bull .
des Sciences Math vol. 15 , pp. 136-144 .
CESARO , ' Count .
Rend vol. 99 , pp. 26 , 27 .
VIVANTI , ' Battaglini , ' vol. 22 , pp. 243-261 , and 378-380 ; vol. 23 , pp. 96-122 ; vol. 26 , pp. 303-314 .
S HERMIT , 'Battaglini , ' vol. 22 , pp. 191-200 .
HADAMARD , 'Liouville ' ( 4 ) , vol. 9 , pp. 171-215 .
BOREL , ' Acta Mathematica , ' vol. 20 , pp. 357-396 .
Paris , , 1900 .
MELLIN , ' Acta Societatis Fennicae , ' 1900 , vol. 29 , No. 4 [ MELLIN has obtained results of this nature .
] .
E. W. BARNES ON INTEGRAL FUNCTIONS .
The memoir deals almost exclusively with simple integral functions of finite or zero order ( vide the definitions of the succeeding paragraphs .
) I reserve the consideration of functions of infinite order , and also the results which I have obtained in connection with functions of double or multiple sequence .
The latter form a self-contained theory , which is a natural extension of the investigations of the present memoir .
The consideration of the asymptotic expansion of integral functions defined by a TAYLOR'S series is also postponed , although certain noteworthy extensions of HADAMARD'S results can be at once deduced from the present theory .
My thanks are due to Professor FORSYTH for the kind way in which he has supplied me with references and criticism .
The fication of tegral Functions .
S 11 .
An integral function we define to be a uniform transcendental function with no poles , and a single essential singularity at infinity .
[ Sometimes it is convenient to include algebraic polynomials .
] An integral function is thus the same as a holomorphic function , to use the translation of CAUCHY'S name ; it is the equivalent of the French " " fonction entiere and the German " " Funktion Every meromorphic function can be expressed as the quotient of two integral functions .
The most simple integral function can be written in the form where is an integral function of , where the zero depends ] upon and certain definite constants , and where the law of dependence of upon is the same for all zeros .
Such a function we call a simple integral function with a single sequence of non-repeated zeros .
The law of dependence may be broken for a finite number of arbitrary zeros in the finite part of the plane .
The existence of such zeros is equivalent to the multiplication of the transcendental function by an arbitrary polynomial coupled possibly with an exponential function of the type , where is another algebraic polynomial .
Such terms do not substantially alter the character of the function .
Functions of type e , where is an integral function , belong to a class apart .
The integral function which we consider we shall suppose to be deprived of such extraneous factor .
* The present memoir was largely written during the summer of the year 1898 .
In consequence : and in spite of rigorous revision , results may sometimes appear to be tacitly claimed as new which have since been published in papers to which reference is made in connection with other investigations of the memoir .
ndard reduced simple integral function with a single simple sequence of -repeated zeros is thus We shall call this briefly a } ?
integral fnnction .
S12 .
The quantity is the smallest integer such that the series is absolutely convergent .
When the convergency of the series be assured by taking for ) some nuluI ) independent of , the function to be of finite gcme* .
In this case , if is a real positive quantity such that converges and diverges , vever small the real positive be , the function is said to be of order , and is called the the series ' , , .
It is sufficient the function depends uniquely upon ?
; if we put , the quantity is not necessarily a uniform function : it may a definite value of some multiform function of S13 .
When there is no finite quantity which will make the series converge , the function is said to be of infinite genre and infinite order .
The convergency of the series , as WEIERSTRASS first showed , always be obtained by taking .
A theorem due to CAUCHY proves this at once , since It is equally sufficient to take , for then ; and the latter series is convergent , since increases indefinitely with But a smaller number still is a sufficient value for , namely , the greatest integer The great difficulty in the theory of asymptotic approximations for functions of infinite order consists in finding the minimum value of .
I do not intend to consider such functions in the present memoir .
Functions of the type , where is holomorphic , are of course integral functions of infinite order .
S 14 .
It is evident that if ( does not increase more quickly than some ( possibly fractional ) power of , however small , the associated integral function will be of -k LAGUERRE , 'Comptes Rendus , ' vol. 94 ; '(Euvres , ' vol. 1 , pp. 167 et , ' tica , ' vol. ) , p. 360 .
chen Etionelt , ' p. 35 ; BOREL , ' Fonctions Entie ' p. 18 .
, ' ' vol. 20 , p. .
E. W. BARNES ON INTEGRAL FUNCTIONS .
order .
On the other hand , if increases faster than any algebraic power of , however large , provided it be not actually infinite , function is of zero order .
In other words , functions whose order is " " finite both vays to use DE MORGAN'S phrase , have zeros which are to a first approximation algebraic .
The zeros of the function are said to be actually when they ars given by when is of course positive and rational , the 's are constants , and , . .
are .
in ascending order of magnitude .
It is evident that we can form a scale of integral functions ; thus , in between functions with the zeros and , where will come functions with zeros like and so Such functions we call integral functions of finite order with a single simple transcendental sequence of zeros ; or , in brief , functions of transcendental sequence .
Thus is a function of transcendental sequence of order and genre zero .
S15 .
Functions of zero order , which must always be of transcendental sequence , can be classified in the same way .
The most simple is Then we consider functions whose zeros are obtained by multiplying by an algebraic function of .
The next step is obviously to introduce intermediate functions by means of logarithmic terms , and so on .
Then we introduce functions formed from sets of zeros of still more emphatic vergence , such as The range is obviously limitless .
S 16 .
It is worth noticing that the density of the zeros along the ( possibly curved ) line on which they lie , decreases with the increase of the the function .
The zeros of the higher functions of zero order have therefore a density which becomes less as we go higher .
The conception of the density of a function is perhaps the most easy way of intuitively classifying it .
The analogy of the DL and scalos of convergcnce is almost too obvious to need mention .
The investigation of the character at infinity of the zero-lines of simple integral functions belongs to the theory of functions of real variables .
I do not propose to undertake it here .
It is , however , evident that such lines cannot curl round infinity when they belong to functions of finite non-zero order with algebraic zeros : they approach this poin in a lin which becomes ultimately straight .
S 17 .
A function with a finite number of simple sequences of zeros can evidently be built up of a number of functions , each with a single sequence of zeros .
The function will thus have a finite number of lines of zeros tendin to infinity .
When the zeros of a function of order are all of the same character and form lines symmetrically ranged round the origin , the function will be equal to a function of of order Thus a function of order with the sequences where is given by the product \mdash ; which , considered as a function of , is of order S18 .
A function , each of whose zeros is repeated a definite number of times , ( say ) , is substantially the powe.r of a function with the same sequences of nonrepeated zeros .
When the zero of a function of simple sequence is repeated a number of times dependent upon , we call the function in brief a simple repeated function .
We can obviously have repeated functions with any number of sequences of zeros .
We may , as before , limit our consideration to a function with a single sequence of zeros ; such a one may be written ' The quantity must , in order that the repetition of the zero may not be meaningless , be an integral number depending upon ; but , if we take the principal values of the ensuing expressions , it is evident that we may get a generalised repeated function by regarding as a general function of The quantity must be so chosen that is convergent .
$ This statement does not deny that they can curl a finite number of times in the finite part of the plane .
MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
We must then , in general , have L ; that is to say , must increase indefinitely with .
We can no longer assign as a value for , which is always sufficient to ensure convergence , as was the case with simple non-repeated functions .
S 19 .
It is eyident that we may regard the value of for which , being a small real positive quantity , is convergent and is divergent an as the order of the repeated function .
When is an integer , the order is equal to the genre : in other cases the genre is the integer next greater than If the order is not infinite , and the sequence of zeros to a first approximation algebraic , , must be algebraic also .
Suppose that and then , we shall have for the determination of , or Repeated functions of infinite order will not be considered in the present memoir .
S20 .
Hitherto we have limited ourselves to integral functions which possess a finite number of simple sequences of zeros .
But we have not thus exhausted the category of integral functions .
Instead of the typical zero being a definite function of the single number necessary to define its position in the series to which it belongs , it may be a function of two or more numbers and belong to a doubly or multiply infinite sequence .
In such case we say that the function is a double or multiple integral function .
Thus the Weierstrassian function is a double , integral function , and another function of the same category is the double-gamma function to which reference has been made in S3 .
The multiple integral functions always have ultimately a lacunary in the region near infinity .
In the WEIERSTRASS ' function , this lacuna1y space covers the whole region near infinity ; for the double-gamma function this space lies between the directions of the axes of and By a well-known theorem due to JACOBI , functions of treble or higher sequence with periodic zeros cannot exist .
This theorem may be extended , and we may prove that there must , in functions whose sequence is greater than double , be such relations * The zeros will , of course , only crowd together indefinitely on the equivalent NEUMANN sphere .
The possibility , or , of summable divergent expansions is the reason for the nomenclature .
'Ges .
Werke , ' vol. 2 , pp. 27-32 . .
the parameters that the region near infinity is not ultimately a lacunary space .
The parameters are , of course , the constants which enter into the expression of the general zero in the form .
Such functions have been scarcely considered in analysis .
The ?
gamma function is the most simple example which it is possible to give .
The theory requires development , since from quotients of multiple integral functions can be built up the general solution of a linear difference equation .
It is to be noticed that , by the coalescence of the parameters , multiple integral functions give rise to functions of lower sequence with repeated zeros .
Thus the function* arises from the double gamma function when the parameters and each become equal to unity .
The separation of multiple functions into functions with repeated and non-repeated zeros and their classification would be carried out on parallel lines to the process adopted for simple functions .
As , however , detailed developments of the asymptotic expansions of such functions are not investigated in the present memoir , I do not intend to consider such functions further .
It has been already observed that by the substitution of ( integral ) for , we derive from any simple integral function a function with times as many sequences of zeros .
The ostitution of for will transform a simple function into one of double sequence .
[ An example of this is given subsequently ( S 62 ) , where LAMBERT'S function is derived from one of simple sequence .
] By transformations of greater complexity we may evidently construct functions oflimitless range .
S21 .
We are still far from exhausting the category of integral functions .
For instance , we may have functions , that is to say , functions whose zeros are situated on concentric circles : the number of zeros on the circle upon ? ?
We can readily see that such a function is given by.the product where is a function of which is equal to an integer for all values of , and where , if , inyersely , and For , the product will converge with See a paper by the author , 'Quart .
Journ. Math vol. 31 , pp. 264 seq. .
W. BARNES ON GRAL 1UNCTIONS .
the first terms for which being omitted .
Thus it converges with The modulus of the term inside the bracket is less than Now has for its greatest a iinitc quantity ( say ) .
The product then converges if converges , which is ensured by the condition assigned at the outset .
The function whose existence has thus been established has zeros on a circle of radius .
If , since the assigned condition makes the order of greater than that of , the zeros will ultimately be separated by arcs of infinite length .
S 22 .
A little ingenuity will enable us to construct other functions of types innumerable , among them what BOREL has called functions " " a croissance guliere * The survey gradually forces upon us the conclusion that we cannot expect to find any general law as to the behaviour of all integral functions near their essential singularity which is not a disguised truism .
MM .
HADAMARIr and BOREL have given relating to the increase of all integral functions .
It seems to me that such laws must be limited to particular classes of functions , and that such delimitation cannot be stated too explicitly .
Consequently in this memoir I have taken the most simple functions and have endeavoured to study in detail their behaviour near the essential singularity , for I believe that by such means the progress made will be sure , if slow .
BOREL , [ ' Fonctions Entie III gives an example of such a function in the form of TAYLOR'S series .
Such a term I should apply to BoIiEL'S law " " the maximum value of a function is equal to the inverse of its minimum value on an infinite number of circles at infinity For this law is immediate consequence of the possibility of asymptotic expansions ( see Part II .
of this memoir ) .
OSGOOD ( ' Bulletin of the American Math. Soc Nov. , 1898 , note , p. 80 ) states that the analysis used to prove HADAMARD'S most general law requires revision .
And it is to be noted that DAlfARD ( ' Liouville , ' 4 ser. , t. 9 , p. 173 ) assumes that is continuous , increasing , and such that constantly increases ultimately .
VOL. CXCIX.\mdash ; A. PART II .
of Divcrgent .
The development of the theory of divergent series is an interesting instance of the progress of mathematical thought .
The beginning was purely arithmetic : to find some to the value of !
, where is a very large integer .
In the result it appeared that the value of log !
could be more and more nearly calculated by adding on successive terms of a series proceeding by powers of .
The error is of the order of magnitude of the term of the series next after the one at which we stop .
And , most important fact of all , the series is divergent .
If !
be replaced by , a similar result can be obtained , which holds for all real positive values of Finally , there comes the enquiry as to what meaning , if any , can be attached to the equality in the case in which is any complex quantity .
Other approximations undergo the same process of development , so that it becomes necessary to try and construct a formal theory .
What we may call the arithmetic theory has been given by POINCABE , for the case in which all the quantities involved are real : restriction which the author subsequently assumes to be unnecessary .
For the more extended case , when is any quantity , we may say that the divergent series .
of which the sum of the first terms in , will , when is very large , be an asymptotic expansion for a function if the expression tends to zero , as tends to infinity .
Thus , if be sufficiently large , where is very small .
The error committed in taking for the function the first terms of the series has a modulus which is infinitely smaller than the modulus of the error obtained by taking only the first terms , for is in general finite , and is very small .
In view of subsequent results , it proves necessary to define the equality of the function and divergent series for values of which lie along some definite line tending to infinity .
We do not then assume that the expansion is possible all round the point It will be sufficient to recapitulate the results which POINCARE obtains .
We may multiply two asymptotic series together by the same rules as we should apply to absolutely convergent series .
Stirling , ' Methodus Differentialis ' ( 1730 ) .
'Acta Mathematica , ' 8 , pp. 295-344 ; 'Mecanique Ce'leste , ' vol. 2 , pp. 12-14 .
MR. E. W. BARNES ON INTEGRAL In particular , we may raise an asymptotic series to any finite power , and it will then represent the corresponding power of the function represented by the original series .
The term-by-term integral of an asymptotic series is equal to the integral of the function which it epresents : in brief , we may integrate asymptotic series .
In general , we may not differentiate an asymptotic equality .
[ Nevertheless , we may differentiate most of the expansions which arise naturally in analysis , and are not constructed artificially .
] Similarly , if an asymptotic equality inyolves an arbitrary parameter , we may no in general ( but we may fairly safely in practice ) differentiate with respect to that parameter .
Such are the main propositions of the arithmetic theory of asymptotic expansions .
The difficulties inherent in the theory are obvious when we attempt its application .
We have , in all cases , to investigate a superior limit to the remainder of the series after the first terms have been taken ; and , to do this , we must have command , even for the simple cases , of analytical processes of great complexity and power .
S 24 .
We proceed then to consider these series from the function-theoretic point of view .
That is to say , on the one hand , we attempt to a to a divergent series which shall harmonise with the development of WEIEBSTBASS ' theory , and on the other , we entel more deeply into the nature of the essential singularity of the of which the divergent series is the expansion .
Suppose first that we have a series .
of finite radius of convergency , so that by CAUCHy'S rule , When is greater than , the series is divergent and our fundamental conception of a series as a command to add in order successive terms leads to no result .
And yet , if the function which the series represents be not which has the circle of radius as a line of essential singularity , the function exists outside this circle , and admits an analytic continuation .
Thus the function exists even when the series is divergent .
Can we not then regard the sel.ies when divergent as a command to perform certain operations which shall yield the analytic continuation of the function ?
We can do so , and in an infinite number of ways .
The most simple is , perhaps , given by an extension of a process developed by BOREL .
* Let the plane of the variable be dissected by some line going from to to the right of the axis of " " The'orie des series divergentes sommables ' Liouville , ' 5 ser. , t. 2 , pp. 103 et .
" " bIemoire sir les series divergentes Ann. de cole Normale Superieure , 3 , t. 16 , pp. 1 et .
E. W. BARNES ON INTEGRAL FUNCTIONS .
This line of section wiIl render uniform , and we shall take that value which is real when is real and negative .
Then it is that where the contour of the integral embraces the line of section as in the figure .
From the original series . . . . .
and an auxiliary function in any arbitrary quantity\mdash ; construct the function This function will be an integral function , for now the integral integral is equal to , or That is to say , when , the intet represents the same function as the original series .
For values of , the integral , provided it has a meaning , represents the analytic continuation of the series .
And if , when the series is divergent , we it as a command to perform the processes which lead to the integral ( ) we shall obtain a conception of such a divergent series which is in harmony with WrIERSTRASS ' theory of functions .
* See a by the author , ' senger of ' p. 105 . .
E. W. ON FUNCTIONS .
S25 .
We now enquire whether the domain of existence of the integral is coextensive with the domain of existence of the analytic function defined by the original series .
Just as the series ceases to define the function by becoming divergent , so the integral may cease to be an adequate expression by becomin infinite Consider the series The " " sum\ldquo ; of this series , when divergent , is represented by the integral ( xz ) , in which and the integral is taken round some contour embracing an axis iu the positive lf of the -plane .
Make now tend to unity .
Then ( xz ) becomes and the integral becomes , taken along some line in the positive half of the -plane .
now that ' where and are both in absolute value not than .
Since the axis of the integral lies in the positive half of the , where is a positive quantity as ] as we please .
The amplitude of is , and that the integral may bo finite quantiby must be such that is negative .
Therefore ) or These conditions can always be by values of within the } , if does lie between or at the limits of the range bounded by thus see that the ] \mdash ; is represented by the series within a circle of adius unity ; and } ) the for all of except those which lie on that part of the real axis between the points 1 and S 26 .
Similarly the series , or its integral when it is represent provided does not lie on that part of the real axis between 1 .
And the same is true of and its equivalent series , when is not necessarily an integer .
These statements form easy examples which the reader can at once work out for himself It is interesting to notice that the lines the singularities to infinity intervene to give uniformity to the -uniform functions to which series may " " sum.\ldquo ; Thus the divergent series .
represents the non-uniform function , which becomes uniform when a cross-cut is made along the real axis from 1 to .
Suppose now that we have any function with singularities lying outside a circle of radius , within which the function is represented by the convergent series . . . . . .
We may join the singularities by straight lines to infinity , each line being the continuation of the direction from the origin to its initial point .
Then within the simply connected area thus formed we may replace the function by a set of integrals of the type Which we can therefore regard as the sum\ldquo ; of the divergent series within the region in question , whenever this set of integrals has a meaning .
Although in general this will not be the case , we can nevertheless , if the function represented by the series has only a finite number of poles outside its circle of convergence and within a circle of finite radius , greater than the radius of convergence , split up the given series into a sum of others each of which , except the last , will be divergent , but capable of bein represented by an integral of the foregoing type , while the last series is convergent within this circle of radius .
The circumstances under which the whole series can be represented by a definite integral over the region of its existence I hope to discuss elsewhere .
The problem is bound up with the determination of the number and nature of the singularities of a TAYLOR'S series and is , therefore , connected naturally with the researches of DARBOUX , HADAMARD , BOREL , , FABRY , S LE Roy , LINDELOF , and LEAU .
aeae S28 .
So far we have been concerned with the summation of divergent series of ascending powers of which are convergent for sufficiently small values of .
We will now asymptotic series as those which are divergent , however small may be , and we proceed to consider their summation .
At the outset we can see that the problem is essentially different from the one * DARBOUX , ' Liouville ' ( 1878 ) , 3 ser. , t. 4 , pp. 5-56 , 377-416 .
HADAMARD , 'Liouville ' ( 1892 ) , 4 se'r .
, t. 8 , pp. 101-186 .
BOREL , ' Comptes Rendus , ' October 5 and December 14 , 1896 ; December 12 , 1898 'Acta Mathcmatica , ' 21 ; 'Liouville ' ( 1896 ) , 5 ser. , t. 2 .
S FABRY , 'Ann .
de .
Nor .
' ( 1896 ) , 3 , t. 13 , pp. 367-399 ; 'Acta Mathematica ' ( 1899 ) , t. 22 , pp. 65-87 ; 'Liouville ' ( 1898 ) , 5 ser. , t. 4 , pp. 317-358 .
LE , ' Comptes Rendus , ' October 21 , 1898 , and February 20 , 1899 .
LINDELOF , 'Acta Societatis Fennicae , ' 1898 .
LEAU ' ( 1899 ) , 5 , t. 5 , pp. 365-425 .
MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
just considered .
stead of the process of summation leading to the same result , whatever the nature of the integral process chosen , we can obtain an infinite number of results , each associated function leading to a different function , of which the series may be regarded as the asymptotic expansion .
For when the divergent ser.ies is convergent for sufficiently small values of , it defines a function oyer that area of convergence , and any summation process can only lead to the analytic continuation of a definite branch of that function .
But a true asymptotic series has no area of convergence , and any meaning which we care to attach to it will harmonise with WEIERSTRASS ' theory of functions .
The essential nature of the difference between the two kin of series may be brought out in another way .
A series convergent for sufficiently small values of represents a function regular in the neighbourhood of the origin .
But any function which a true asymptotic series can represent will have the origin as an essential singularity .
And , therefore , not only can many functions with an essential singularity at the origin have the same asymptotic expansion , but also same function may have different asymptotic expansions in different areas having the origin as apex .
It is almost impossible to imagine a vagary which an essential singularity will not possess , and this fact we calmot , throughout the whole of the investigation , too carefully bear in mind .
Inasmuch as any means of regarding an asymptotic series leads to a result peculiar to that means , we must choose our process with care so as to obtain the most simple result , and , if possible , so as to that our co1lception of such series agrees with the arithmetic point of view by which historically they were generated .
S29 .
Suppose , in the first place , that we have given the asymptotic series . . . . . .
, in which , by CAUCHY'S rule , .
And suppose further Then the associated function , in which will be an integral functio1l .
' It is a natural extension , then , of our previous ideas to regard the asymptotic series as the expansion of the integral and , conversely , to regard the integral as the " " sum\ldquo ; of the asymptotic series .
ON FUNCTIONS .
The poin will be an essential singularity of the function of the integral is the formal expression .
For certain values of near the integral can probably take values which differ infinitely for the smallest change in the value of This will happen when lies on a line of zeros or poles crowding to .
Along such lines , or quite possibly within areas of the same nature , the asymptotic series will cease to represent the function .
Further , there must always be such lines or areas of non-representation , for the only functions to the essential ities of which poles or zeros or other singularities do not ore of the types , , which cannot admit of :symptotic ) ansion .
We have then the fundamental result that the integral cannot represent the ' ' sum\ldquo ; of the series round .
There will be certain lines or areas with as extremities or vertices along which the asymptotic series cannot be " " summed\ldquo ; by any process which we llay employ : these lines or areas will differ with the different processes , but will never be absent altogether .
There are , of course , asymptotic series of the prescribed type which can never be ' ' summed\ldquo ; by any process which we may employ .
Such a one is , in which .
But such series will never arise naturally in analysis , and we do not , therefore , need to trouble about them .
S30 .
We have now to consider whether , when a series of the prescribed type is " " summed\ldquo ; by means of the process indicated , the function which results admits the series as an arithmetically asymptotic expansion according to POINCARE'S definition .
Denote by the integral which is the result of summing the series . .
in The associated for the series is and for the sum of the first terms of the series , is Hence where .
W. BARNES ON INTEGRAL Now x ) is an absolutely rgent series , and tends to zero as tends to infinity .
Moreover and are functions which , while / has any finite value , have the same character near , for they only differ by the polynomial Therefore the integral analytic function of whenever ( xz ) does so ; and the two functions will have the same character near Therefore , within those areas for which the second integral represents the sum\ldquo ; .
the given asymptotic series , the first integral is finite , and tends to zero as tends to zero .
Thus the nptotic equality satisfies ' arithmetic definition .
The reader must note very carefully that this theorem does not apply to divergent series which have a finite radius of convergence .
It is necessary that should tend to zero .
No computer , for instance , could make . .
tend to S 31 .
Suppose now that we differentiate the series . . . . . .
in which We shall obtain the series lf we " " sum\ldquo ; this ries b exponential process tlJe name which is convenient to giye to the process employed in the preceding paragraphs ) we obtain the integral , in which We thus see that , since this sel.ies is an function , .
Therefore the " " sum\ldquo ; of the .
is VOL. CXCIX .
\mdash ; A. MR E. W. BARNES ON INTEGRAL FUNCTIONS .
Now so long as the integral ( xz ) is within the ions surrounding , for which the original series can be summed , the differential coefficient of the function which it represents is the function represented by the integral ( ) for we do not ress the rules which govern differentiation under the sign of integration .
Therefore , within the for which an asymptotic equality is valid , such equality may be differentiated .
Similarly such equality may be integrated .
And the process of differentiation or integration may be repeated any number of times .
S S2 .
We have hitherto limited ourselves to the consideration of asymptotic series of the type . . . .
in and The first condition is necessary that the series may have zero radius of convergency , that is to say , that it may be asymptotic .
The second condition was requisite in order to ensure the applicability of the exponential process .
It is convenient to call an asymptotic series for an asymptotic series of the first order ; one for which this limit is greater than zero , but a series of the second order , and so on .
We have given in the preceding paragraphs the theory of summation of series of the first order .
But suppose that we wish to sum one of the most simple asymptotic series , that for , namely , where is HERMITE'S Bernoullian function .
By CAUCHY'S theorem , re-discovered by HADAMARD , we know that ; * JORDAN , d'Analyse , ' 2nd edition , vol. 2 , pp. 151-157 .
MR. E. W. BARNES .
ON INTEGRAL FUNCTIONS .
for the expansion is only valid within a circle of radius We see then that the asympt , otic series is such that , if we denote the coefficient of by , and The series is thus of the second order .
And the associated function formed as in the preceding paragraphs will be which is not an integral function .
Our analytical machinery therefore breaks down , and we must attempt to extend it .
Just as the problem admitted of an infinite number of solutions , so we may now proceed in an infinite number of ways to give an analytical meaning to asymptotic series of the second or orders , Of these two would appear to be most natural .
We may either use some more powerful associated function than we used in the exponential process , or we may repeat the exponential process until we arriye at a finite analytical function .
S 33 .
Let us consider in the first place the second of these rnatives .
If we have the asymptotic series , . . .
, we have agreed to say that this series is the expression of the analytic function ( xz ) , whenever this integral has a meaning , that is , whenever ( xz ) is an integral function , and the is not infinite .
Now ( say ) , and , if the series is not absolutely convergent over the whole plane , we shall be consistent with our former generalised point of view , if we regard it as determining an analytic function ( ) whenever this integral has a meaning .
W. ON AL Now \ldquo ; ( say ) If the series had a finite radius of conyergence , or zero radius of the first order , ttJe function will be an integral function , and by the process just sketched , a definite meaning has been assigned to and the original series .
When , however , has zero radius of convergency of the second or higher order , will not be an function , but we must regard the series which it denotes as determining an analytic function whenever this has a meaning , that is , as preliminary condition , whenever is convergent over the whole plane .
The procedure may be repeated indefinitely .
If have started with an asymptotic series which does not ultimately give rise to a function whose finite radius of convergency is a line of sssntial singularity , we shall ultimately get an analytic function of which the original series is the asymptotic expansion in the vicinity of its essential singularity S 34 .
The extension which we have just indicated is in harmony with the general theory , but we have still to determine the important point as to whether the asymptotic equality of series and functions satisfies arithmetic definition .
Take for simplicity the series of the second order for which the associated series . .
has finite radius of conyergency and represents the function .
The given series gives rise to the function ( ) Lct W. ON INTLGRAL IONS .
then ( ) where Now ( xz ) is an analytic function of of the same character as ( xz ) : hence the natures of the two functions and near substantially same .
therefore , tends uniformly to a finite limit as tends to zero in any direction , also tends uniformly to a finite limit as tends to zero in the same direction .
That is to say , tends to zero as to zero , so that the divergent series is arithmetically asymptotic for the function .
It is evident that a repetition of the same argument will prove the arithmetic nature of the asymptotic dependence of a series of any order and the function to which it giyes rise by successive applications of the exponential process .
But one case of exception must be noticed .
At each step the equivalence of the asymptotic series and the derived function fails along certain lines or within certain areas radiating from .
And , since the effect of such failure is cumulative , it may happen that before the process is finished the equivalence has failed over the whole area around .
Either the series is hopeless\mdash ; an artificial monstrosity that cannot arise in practice\mdash ; or we some other process by means of which it can be interpreted .
S 35 .
As an exa mple of the process just sketched , consider the asymptotic expansion which , function of , is an asymptotic series of the second order and wherein and are any complex or real parameters .
Applying the integral process associated with the exponential function to the series , we obtain the integral where and we have , for convenience , taken the auxiliary function to be so that is orbed in Now is a series of finite radius of convergency , and the analytic function which we take the series to represent is the function of which the series is the expansion within the circle of convergency .
The series will therefore denote , with our present conceptions , the function The series with which we commenced may therefore be regarded as giving rise to the analytic function - on making the substitution This function admits when is small , the arithmetically asymptotic expansion from which we started .
When is alarge real positive integer , the series and integral become fundamental in the asymptotic definition of the extended Riemann function .
But there can be obtained by other processes an indefinite number of analytic functions , each of which has an essential singularity at , near which point it admits the given series as an arithmetically asymptotic expansion .
We proceed to indicate one alternative process by which such an analytic function can be obtained at a single step .
S 36 .
For this purpose we use certain results of the theory of the connection between linear difference and differential equations .
Consider the function It is evidently a transcendental integral function which is a solution of the differential equation wherein the operator If be any solution of this equation , form the function where the contour of the integral and the prescription for are exactly those employed in the definition of the integral for previously employed ( S24 ) .
On integrating by parts , we have MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
a relation which may also be written We thus see that and also Therefore , since is a solution of the equation we have .
The general solution of this difference-equation is , where is a simply periodic function of of period unity .
We have then established the identity . .
When , the expression on the left-hand side , and therefore that on the righthand side must inyolye . .
only .
Thus , when involves only .
It must therefore be a function of , . . .
Not only so , but it cannot involve these quantities at all ; for when will be a function of , . . .
, and yet it is independent of ; and so on when , . . .
Thus is a function of , simply periodic of period unity , independent of , . . .
and Let ; then becomes , and the integral becomes .
MR. E. W. BARNES ON INTEGRAL ]TIONS .
Thus , and we have finally for all values of , . .
and the identity .
This identity is the direct generalisation of the identity , and we may therefore expect to be able to use it to extend our former process of " " summing\ldquo ; asymptotic series .
S37 .
We may , in fact , show at once that we can sunl any series of convergency zero , in , where is any finite quantity .
For this purpose we put ; , and we have Then , with our former notation , we take the auxiliary function , where And now is defined by the integral in which We take , and then will be an integral function .
; and fore , In connection with the proof of this formula , the reader may with advantage refer to MELLIN , ' Acta Mathematica , ' 8 , pp. 37-80 ; 9 , pp. 137-166 ; 15 , pp. 317-384 .
' Acta Societatis Fennicae , ' t. 20 , pp. POINCARL " " American Journal vol. 7 , pp. 203-258 .
PINCHERLE , ' Accad .
dei Lincci , ' ser. , t. 4 , pp. 694-700 .
POCHAMMER , ' Matbematische Annalen , ' Bd .
38 , pp. Bd .
41 , pp. 197-218 .
' Crelle , ' Bd .
71 , pp. 316-352 .
ORR , ' ridge Phil. Trans vol. 17 , pp. 182-199 . .
W. ON INTEGRAL FUNCTIONS .
We can thus sum any natural series of convergency zero whose coefficient is of the same order as a finite power of S 38 .
But we call go further than this : we can construct iuvel.se functions which will enable us to sum any series gency zero .
For we have seen that Suppose , now , that we the function This function will be of ' which is absolutely convergent for values of But , if we operate by our integral on function , we have ; and the function is infinite if be positive .
If , now , we take is so cbosen to make the siu series , for all finite values of , it is obvious that will itself converge for all values of , and so be an integral function .
We may now take for the associated function .
, where ; and by suitable choice of the coefficients we may make vanish to an order as great as we please .
We can then sum the series , where is infinite with to an order as high as we please .
In other words , we have invented the analytical machinery necessary to sum any ( natural ) asymptotic series .
S39 .
As an example , suppose that we wish to sum a series , where is infinite like , where * This theorem corrects a mistake in my paper , ' Theory of the Gamma Function , ' p. 11 VOL. CXCIX .
\mdash ; A. 3 { .
ON INTEGRAL FUNCTIONS .
With our previous notation we take and so that is a transcendental integral function .
Then we We take the associated function where c ; the function .
in Then the sum of the series will be represented S 40 .
We have now , by means of the generalised exponential functions , given the machinery by which we may expect to be able to " " sum\ldquo ; a natural asymptotic series of any order .
It may be provedjust as the fundamental exponential process that the series and the function derived from it have asymptotic equality of the arithmetic type .
Moreover , if we regard the series as having a finite radius of convergency , on which one or more singularities lie , which has shrunk indefinitely , we , as it were , nlagnify it again by means of the function so as to obtain the associated function whose radius of convergency is infinite .
The alternative process consists in successive magnifications by means of the function These two processes will in general lead to different results : in each case we shall obtain functions with as an essential singularity ; both functions will have the * When we take , we have ; and , when is large , the series is , by a theorem due to STOKES , infinite like to the first approximation .
We thus any series for which is of order , by taking , greater than MiX .
E. W. BARNES ON INTEGRAL FUNCTIONS .
same arithmetically asymptotic expansion .
But the expansion in all probability will not be valid in the two cases along the same lines or within the same areas tending to .
Moreover , such a result is not surprising .
The original series , except from the poin of view of the computer , had no meallin ; it did not define an ytic function over any area of the plane of the complex variable and therefore could not uniquely represent such a function .
We have , however , now two processes ( out of an infinite number ) by which we may conceive the series to define an analytic function , and the functions thus defined each satisfy all that the computer can demand .
S41 .
It will , perhaps , elucidate the which has been developed if we give actual ples of its application .
We will first investigate the Maclaurin sum formula , which gives an asymptotic value when is large , under certain restrictions as to the nature of the function .
In the first place it is evident that such restrictions must exist : the function must either be uniform or be limited to a definite branch of a multiform function ; and , as takes increasing integral values , must increase uniformly .
We will assume , therefore , that is an integral function , which may lee represented by a TAYLOR'S series , Then , if the integral be taken along a contour embracing an axis in the half of the plane , we shall have , by the usual expression for the gamma function , Suppose now that the series has a finite radius of convergence Then will be the expansion of a function convergent outside a circle of radius We can always make the bulb of the contour along which the fundamental integral is taken expand so as to entirely include this circle of radius and the subsequent integral will then be finite .
Let now , so that is a definite finite quantity depending on the coefficients in the expansion of .
Then The second integral may be written in the form .
if we postulate that we are reversing the process by which we " " sum " " an asymptotic series .
The integral is equal to the asymptotic expansion .
When is odd : the integral is therefore equal to the asymptotic expansion .
We have finally asymptotic equality* .
This equality is valid when is a uniform integral function of such that if it be expanded in the form , the series has a finite or infinite radius of convergency .
We must therefore have !
equal to a finite or zero quaJltity , so that must increase as fast as or faster than .
The function must therefol.e be a function whose " " order is greater than or equal to unity .
In the particular case when the represents an function , we may conyeniently express in terms of the Riemann functions of negative integral equent paper I shadl show that it !
better to wlicc this formla in the form in to its to more genernl extensions .
this case the ) of the contour which expresses may taken so s:nall as not to include the ) ] , and we therefore This series is evidently convergent is an function whoss order is greater than or equal to 2 , a condition which is to the convergency for values of of the series It is evident the Maclaurin sum will hold good in ) cases in which is not a uniform function .
If it like or , or either of these functions multiplied by integral function of order greater than unity , the Maclaurin formula will be yalid if we suitably specify the of the function considered .
Instead of attempting to tabulate such , it is perhaps better that we should go back to the genesis of the when they actually arise .
Applications of the formulae which will be made subsequently in this emoir will usually be to cases in which has very nple values ; and ffi general formulae bs tacitly supposed subordinate to what we may call the Maclaurin restrictions .
S 42 .
As a second example of the theory of asymptotic series we propose to try and find the function of which series . .
is the asymptotic expansion near the essential singularity We know that , if be a positive integer , where the line integral is taken along any straight line from the to infinity which lies in the half of the plane to the right hand side of the imaginary axis .
Therefore the given expansion asymI ) totically represents the function where is the ction which is the series } , and the is taken the ] The given series is thel.efore the asymptotic expansion of the fUnction Suppose first that the real part of is positive ; then , putting we have the being taken along a line along which is positive , that is , along which is negative .
[ These two conditions only differ when we consider a line lly parallel to the imaginary axis and therefore itially excluded .
] \mdash ; , we have , the integral being taken along a line which , ( y ) is positive ; and therefore ( 1 ) , the integral being taken along a line still in the positive half of the plane .
Thus , where we take to be very small .
Hence , if be constant , * terms which vanish with Finally , on makin It will be noticed that the integral ( 1 ) obtained for has a pole along the line of integration so that it has an infinite number of values , all differing by , which are implicitly involved in the logarithmic term .
We see then that , when the real part of is positive , the given series is the asymptotic expansion of the function Take next the case when the 1eal part of is negative .
As in the first case , the ries in the asymptotic expansion , the integral being taken along a line for which is negative .
Thus it is the expansion of along a line for which is negative along a line for which is positive .
* See the author 's paper , 'Messenger of Mathematics , ' vol. 29 , pp. 98 and 99 .
MR E. W. BARNES ON INTEGRAL FUNCTIONS .
On pursuing the same course as before we find that the given series is the asymptotic expansion of the function But there is the important difference that now the integral which has led to result has no pole along the line of integration .
And , instead of being allowed to take any one of an infinite number of values , has such a value that is real when is real and tive , and has for complex values of real part is negative an amplitude which lies tween We see then that the process mployed has led , when is positive to an infinite number of functions , all of which have the same asymptotic expansion ; and , when is negative , to but one such function .
Evidently when we seek an asymptotic expansion for the function ae we may say that we get , when is positive , . .
for terms like are negligible compared with the least term of the asymptotic series ; but when is negative , we get in which successive terms are of decreasing order of magnitude .
The zeros of the function near the essential singularity , are ultimately along the imaginary axis .
We thus have an illustration of two important propositions:\mdash ; ( 1 .
) A uniform integral function may admit of asymptotic expansions of different form in different areas with their vertices at its essential singularity .
( 2 .
) These portions of the plane are separated by lines of zeros of the function .
S43 .
Inasmuch as in Parts III .
and IV .
of this paper we proceed to actually obtain asymptotic expansions satisfying these laws for all the most simple types of I was asked to investigate this function by Mr. G. W. WALKER , Fellow of inity College , who desired to compute it in certain physical .
Originally I obtained the expansion by considering the differential equation , in a benring great resemblance to )loyed by HORN , ' Crslle , ' vol. 120 , pp. 17 and 18 .
integral functions , we now proceed to sketch the process which will be adopted , and , in the of ' outline , to at once the validity of that process and the laws which its results .
Supl ) , in the first place , that we have the absolutely convergent expansion .
in which tho coefficients are functions of a variable , asymptotically given for large values of by expansions of the type are numbel .
S arranged in of magnitude and tending to as a limit , the first )bers of the sel.ies being possibly negative .
Suppose thab substitute these asymptotic values of the coefficients and the expression for in powers of We in , when asymptotic equality . . . . . .
sion will be asymptotic : the computer would use it to calculate values of and is large .
The series which enter as coefficients will be , in all probabi]ity , divergent ; but , as we looking at the whole matter from the of view of the computer , we are liberty to " " sum\ldquo ; them by the methods which have been developed the prese1lt of this If , the the plications which we subsequently make of this theory , these sel.ies have a finite radius of convergence , we can " " sum\ldquo ; them each to a definite , possibly non-uniform , analytic function ; and we shall have an expansion , which will satisfy POINCARE'S definition of arithmetically asymptotic dependence .
We shall thus have obtained unique asymptotic ) ansion for the function F The case in which the series of the type . . . . . .
have / adius of convel.gence does not .
In such a case we should be able to obtain an number of functions , of which these series are the asymptotic expansions , and should have the absurdity that the asymptotic expansion of in ascending powers of is not unique .
S44 .
A function cannot , as has already been stated , be represented by the same asymptotic expansion for all values of in the neighbourhood of , unless the function is an integral function of , and the sel.ies absolutely convergent .
MR. E. W. ON INTEGRAL FUNCTIONS .
44 !
) For unless and the series .
have such a \ldquo ; it is impossible that \mdash ; . . .
should , as approaches infinity , always tend uniformly zero , whatever } ) the argument of But , as we proceed to show , a uniform function of finit .
' genre with an essential singularity at infinity can , in general , be represented by one or more asymptoic expansions valid for all points near infinity except those in the immediate vicinity of the zeros and poles of the function .
Two different asymptotic expansions cannot exist within the same region , and the regions are separated by the lines or areas of zeros or poles of the function .
theorem is true whether the function be a quotient of repeated or non repea.ed integral functions , with zeros of simple or multiple sequence .
We need only consider the case of integral functions\mdash ; the general theorem follow , since every function of the type just mentioned can be represented as a quotient of two integral functions of finite genre .
The zeros of the function must proceed according to fixed laws , and therefore , in diagram of the region neal infinity , they will mass themselves infinitely close together as we approach infinity itself They will therefore form certain lines ( not necessarily straight ) or areas of ultimate singularity .
If the areas entirely surround will be no asy1nptotic expansion possible .
We thus assume that there exists an area such as AB , non-shaded in the figure , within which , if the radius A is sufficiently small , there are no zeros of Suppose first , that the zeros of the function form a single simple sequence , are non-repeated ; then it may be written , where is the " " genre\ldquo ; ( independent of ) , and is a holomorphic function .
Suppose that lies between circles of radii and where is large , then those terms of the product which may be written , in the proof of WEIERSTRASS ' fundamental theorem , where is a function represented by a series of positive powers of For those V0L .
CXCIX .
terlns which correspond to the first zeros , we can expand in the form We where , when is bsolutely convergent double series of positive and negative powers of , together with logarithmic terms .
Now , unless be in the immediate vicinity of the zeros of the function , this expression , considered the point of view of divergent summable series , will be valid for all values of .
For , when , the expression of 1 \mdash ; the form still exists as a divergent mmable series outside the circle of radius for all points except those near Therefore the form of exists continuously as increases , provided we do not cross the line of , come within the immediate vicinity of , the zeros of the function , And thus , if we treat the series entering into the expression cf as series which are summable though divergent , the expansion will be of Now the expansion may be written where is a function of which depends also on .
Expand } asymptotically in a series of successive differentials of by the Maclaurin fol.lnula , and rearrange the series .
We shall get ( A ) a certain series of positive and negative powers of , each multiplying terms like ; and ( B ) expansion consisting of a finite number of positive and an infinite number of negative powers of , each associated with a constant arising from a corresponding Maclaurin expansion .
The other terms depend upon and vanish identically ; the coefficient of each Bernouillian number is zero .
When we apply the processes of divergent summation which have been previously developed , the series which the of terms will reduce to a definite W. ON FUNCTIONS .
40- ] possibly ( say ) .
The ining ( B ) ) ) the summable divergent series [ There will only be a finite number of ) ositive P , of , of the function is finite .
] We have then In other words admits the sylnptotic expansion valid for points but those in the vicinity of the zeros of . .
The process just sketched will become ] nucll more clear when it is to various particular cases ns in the followin .
The proof may , by verbal alterations , be extended so as to include functions of simple sequence with relJeated zeros .
A function with a finite number of simple sequences of zeros can be expressed as a product of functions , each with a single simple sequence .
The logarithm of each of these functions will admit an asymptotic expansion , and the sum of such expansions will be the asymptotic expansion for the logarithm of the function .
But terms of the category may be of different weight in different regions , separated by bands of zeros , and thus the expansions may differ in such regions , as has previously been seen in the case of the integral function S 46 .
The genel.al theorem which has just been given may be p ) for integral or meromorphic functions with multiple .
We refrain proof , as the consideration of such functions is omitted from the subsequent development of this paper .
Neither do I make any attempt to considel functions of infinite order , 01 expansions near isolated essential singularities of uniform functions .
The difficulties which arise all subordinate to the main necessity of limiting the ) of under considel.ation ; it seems whethel it is possible to give any concerning integl.al functions and behavioul near infinity , which will apply to every function which can be constructed .
For exceptional classes must always be infinite in number compared with those which can be mally defined .
PART III .
The Asymptotic of Integral Functions .
.S47 .
We now proceed to consider in detail simple integral functions .
After the discussion given in Part I. , we may confine ourselves to functions with a single sequence of zeros .
We shall find that such functions divide themselves naturally into three groups:\mdash ; ( 1 ) Fullctions whose order is less than unity , ( 2 ) Functions of non-integral order greater than unity , ( 3 ) Functions of integral order greater than unity .
In connection with each group of functions with algebraic sequence of zeros we first consider a standard type with which all functions of the group may be compared .
These standard functions are here , where is and not integral , and is an integer such that , where is an integer the logarithms of each of these functions we obtain in turn the complete asymptotic expansion near .
We then show how all functions of ths same order with algebraic sequence of zeros yield by the same method similar asymptotic expansions .
And we indicate how it is possible to apply the same methods to wide classes of simple functions with a transcendental sequence of zeros .
48 .
The constants which enter into the analysis arise from the Maclaurin sum formula ( S41 ) , which may for our present purpose be written , 1yeing ) any integer , positive or llegative .
What have called the claurin r ) the function are always MR. .
W. BARNES ON INTEGRAL FUNCTIONS .
45.3 supposed to apply .
We shall cadl the bIaclaurin integral-limit for .
We shall also put , and call the constant for .
When , we have the formula .
, where is the aurin integral-limit for .
We put , and call the absolute constant for .
When is a positive integer and ] , it is evident that and S 49 .
In the .
case when being real or con plex , the Maclaurin constants are cases of fullction .
For , for all values of When , we have We have also the special values , when is an evell positive integer , , when is an eve negative integer , , when is a negative odd integer equal to .
We write , when is any quantity real or conJl ) , unless in which case we put .
Order them S 50 .
Before we proceed to consider the general theory of the asymptotic expansion of functions typified by , where will consider the ction , which is known to be equal to We , and , if we obtain on .
the arithms . . . . . .
If now we employ the arithmetic asymptotic approximations giyen by the Maclaurin sum formula for ' and , we obtain , in the limit when is infinite , where we have re-arl.anged the terms of our double series in accordance with S 43 .
Now by the theory of summable divergent series and Hence have , when is , the approximation , asymptotic with egard to MR E. W. BARN ON TEGRAL FUNCTIONS .
4.55 or Suppose now that is large , and that where [ This assumes that is a large , a restriction which , as will be seen later , can easily be removed .
Then , when is large , we have asymptotically The sum of the FOURIER'S series inside the square bracket is , when equal to Therefore , when is large , we have asymptotically S 51 .
In the preceding ation we have assumed that the sum formula expresses the values , when is large , of the functions and ( positive or negative ) .
Accurately we have of course ; , when is and when is positive .
Hence , in the limit when is infinite , This mula is accurate and holds }whatever positive integral value may MR. W. ON INTEGRAL -FUNCTIONS .
nfortunately we may not say that the sum of an infinite series of integrals is the whose subject of integration is the function to which the series of sub.jects of integration can be " " summed The two integrals last written down can then only be evaluated by reducing them to an exceeding complicated extension of the known as DrRICHLET'S integrals .
analysis is utterly intractable .
If we make , and expand the subjects of integration in powers of , then we can say that the last two terms will not contribute terms whose order of nitude when is large is comparable with that of any positive or negative power of And , as we know , the sum of these two terms is equal to .
The formula log is thus asymptotic exactly as the Maclaurin series for !
and , from which it is derived , are asymptotic .
That is , for large values of , the expression for all yalues of tends to zero as tends to infinity .
There is , in fact , POINCARE'S arithmetic asymptotic dependence .
The preceding example will serve to show the nature of the asymptotic expansions which we can now proceed to obtain .
S52 .
We consider first the function , where We have . . . . . .
Therefore , if we substitute the approximations for !
, , and , given the Maclaurin sum , we shall obtain the expansion , arithmetically with egard to MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
or , This expansion is arithmetically asymptotic with regard to , and the coefficients of various powers of are ultimately to be summable divergent series .
Let now be large , and such that lies between .
Then the modulus of is a quantity which is very nearly equal to unity .
We proceed to ' ' sum\ldquo ; the , and Write , then the first series becomes Thus and So that If now we " " sum\ldquo ; the last series we ; and therefore , where A is independent of When Hence VOL. CXCIX\mdash ; A. 3 , by usual theory of ] series , 1 ) ovided Therefore so that ] Hence The second ) is at once seen to be equal toAnd , since the term by term rential of a summable divergent series is equal to the of its SUln , the third series vanishes identically for all positive integl.al values of Thus , when lies between and , or possibly is equal to the latter quantity , we have the asymptotic expansion , while ; or , Thus , when has any large value , , we have the arithmetically expansion S53 .
The approximation represents an ithmetic not a functional equality .
It does not vary with the argument of , and it exists eyerywhere in the neighbourhood of infinity except at points on or near the line of zeros of the function .
Not only so , but at points on the line of zeros of which are not in the immediate vicinity of one of its zeros , both the function and the asymptotic series have arithmetic continuity , and tljerefore the equality will hold at such points .
These results accord with the general theory developed in Part II .
MR. E. ON INTEGRAL FUNCTIONS .
The series for \mdash ; can be " " summed\ldquo ; by the methods of The function .
written tends to zero near its essential singularity , and the same will be true of the function which we get by any process of tion .
But , in general , the function derived from will not be equal to the ) from which the series has been obtained .
Since , the series is equal ) the integral being taken round the fundamental contour of S24 .
The series is thus equal to ' where The series for is divergent and of order .
The integral is interesting in that , in place of , we have used as our auxiliary of summation . .
We now pass on to consider the most general simple integral function with a single sequence of non-repeated zeros , whose order is any number ( zero included ) less than unity .
The function may be written , where is absolutely convergent .
The zero , , is a definite function of and any finite number of given constant quantities .
Suppose that if , then inversely .
Let and suppose that / is a large integer such that Then ) , so that if we expand the logal.ithms in convergent series we shall get . .
Now , by the Maclaurin sum , if be positive , where is a constant quantity , depending on and the form of , which we have proposed to call the Maclaurin integral limit for . . .
If be negative , we have where , there being no term independent of on the right-hand side .
And \mdash ; . . . . .
Hence , in the limit when , Now , when the niting values for of the summable series are taken , Hence , asymptotically , MR. .
W. ON FUNCTIONS .
Now , where is a given form , and such that we can integrate expressions like positive negative , integral ) we need only carry out this process and ( sum\ldquo ; the ensuing series of positive and negative powers of to obtain the dominant term of the asymptotic expansion of .
If , however , is not thus formally given , we have to face the difficulty that the lower limits of the definite integrals are different quantities .
The lower limits , however , corresponding to negative values of , are such as to give rise to .
If , then , we consider only indefinite integrals of the type , and take care that in any transformation of these we do not introduce arbitrary additive constants , we may take the symptotic expansion in the form \mdash ; S 55 .
It is the integral which gives rise to the dominant term of the asymptotic expansion of This integral is evidently equal to Suppose now that and take , the logarithm , when having a cross-cut along the negative half of the real axis , so that , where is arithmetic .
Then Now the of the dominant term I does not depend on the quantity which vanishes when is sufficiently large .
We have then an inteo r of the type first considered by DIRICHLET.:ff .
' Crelle , ' vol. 4 , pp. 157 , et The theory previously developed in Part II .
tells us that this integral must , function of , be independent of ; in other words , when , that But this is precisely result : we thus have a valuable .
verification of our theory .
Finally , then , the dominant term of the asymptotic expansion of is the function Since we may evidently change the sign ol without altering the value of we have Now is the function inverse to .
If , then , we suppose that and are determined from the relation , principal values of inverse expressions being taken , and and being functions of and , we shall have finally , as the simplest form in which we may write .
There is no doubt that it is possible to construct functions for which the preceding analysis will not hold good.$ It would appear , however , to be ] icable to most of the types of functions which would ordinarily arise , and a more accurate investigation will need the exquisite finesse of certain developments of the theory of functions of a real variable .
Note that , for the case in which , we have established that S 56 .
The dominant term of the asymptotic expansion of takes a very simple form for the case in which can , when is large , be nded in descending powers of in the form , where We have the asymptotic expansion *We have assumed , for nstance , that we can apply the urin sum formula ) , and , therefore , the conditions of are satisfied .
MR. E. W. BARNES ON INTEGRAL NCTIONS .
Since is a very large quantity we may utilise the expansion of .
The integral last written is , thel'efore , equal to If , then , we introduce the constants , , we shall obtain the asymptotic expansion Such values of the many-valued functions introduced to be taken as would be indicated by the analysis .
S57 .
It is evident that the investigation of S 54 applies to all simple integral functions whose primary factors need no } ) onential to ensure convergency .
Thus it includes all simple functions of order where is real positive and with ' algebraic zeros .
It includes all simple functions with non-algebraic zeros of the type given by , where and are both real positive and is positive or ; where denotes . .
} , these being repetitions of the logarithm , and where , are decreasing quantities tending to as limit .
MR. .
W. BARNES ON INTEGRAL FUNCTIONS .
But , because of the validity of the Maclaurin sum formula , it includes functions with very rapid convergence\mdash ; such as those for which function of of lower order than S58 .
We can now extend the result which was obtained in S 52 , and find an asymptotic ) proximation for a simple integral function with an algebraic sequence of zeros , that is to say , of a function of which the zero , , admits , when is , an expansion of the form .
, where is greater unity and the quantities , . .
are real in ascending order of magnitude .
We take so that , by reversion of series , ( say ) .
Since , when is positive , we may expand ectly by the binomial theorem , we have , when is large , . .
, where . .
is a definite finite quantity vanishing with the quantities which can be expressed in terms of a series of Riemann functions .
Again , when is positive , .
And ; nd we shall put MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
If now we substitute in the general mula we shall obtain And when we sum the Fourier series which result from the last integral , we find \mdash ; . .
S 59 .
This expansion is valid for all values of any which lie between and It is arithmetically asymptotic in the same way as the expansion from which it is derived .
We see from the results just obtained that the asymptotic approximation for log , where .
exceeds that for by a quantity whose first term is . .
when , and by a quantity whose first term is - , when When , the difference of the two asymptotic approximations commences with the indeterminate form \mdash ; , which arises from the integTal VOL. But this integral is equal to Thus , where , the difference of the two asymptotic approximations commences with the term , a result which may be obtained without much difficulty by elementary algebra .
Note that is the expression obtained when we reject the infinite part of the function for , when expanded in powers of Note also that the constant . .
which , when is the first term of the asymptotic expansion of the logarithm of the ratio of our two products is equal By means of the formula ' where is a real quantity lying between , we may prove that this series is absolutely convergent when Application to Functions of Zero : S60 .
Hitherto no example has been given of a function of zero order , although the general investigation of S36 applies equally to functions of this nature .
In such cases it becomes necessary to introduce Maclaurin constants of a complexity which seems , except in special cases , beyond the reach of present analytical ocesses .
They can no longer , as for functions of finite order , be expressed in terms of RIEMANN'S function nor , I believe , in terms of any functions which have so far been introduced into analysis .
An example will now be given of a very rapidly conyerging integral function .
It obviously would serve as the point of a series of interesting researches dealing with the classification of simple integral functions of zero 61 .
We propose to obtain the asymptotic expansion of the function the notation of the general theory we have now Therefore ; By the Maclaurin sum formula , if be positive , \mdash ; where is the Maclaurin constant corresponding to , which may be determined as iu S84 .
If we put , we have .
E. W. BARNES ON TEGRAL FUNCTIO S. If then we out the general process , we shall obtain the asymptotic expansion .
As before , we have to ( sum\ldquo ; the final divergent series .
We take to be a large quantity such that is very nearly equal to unity , and then we consider the Fourier series But \mdash ; .
And Therefore we have the asymptotic expansion or finally* \mdash ; S62 .
It should be noticed that if , in the function whose asymptotic expansion has thus been obtained , we substitute for , we shall btain the function This is an integral function whose zeros are of the form , 1 , . . .
It is substantially what I propose to call LAMBERT'S function .
The function has properties which are a sort of mean between those of the elliptic and double gamma functions .
We can express LAMBERT'S function as a product of two double gamma functions .
It is closely colmected with the well-known LAMBERT'S series , and in terms of it we can express in a very elegant form the coefficients of capacity of two spheres .
*The dominant terms of this result are equivalent to those given by MELLIN , 'Acta bSocietatis .
Fennicae , ' t. 24 , p. 50 .
S63 .
The reader will notice that in the preceding analysis we have used the methods and not the result of the general formula .
reason is that with an exponential subject of integration we are unable to ensure that we do not introduce arbitrary additive constants when the indefinite integrals are transformed as formerly .
For in this case and ; and we have to consider a series of integrals of which the first is ) We are tempted to say that this integral is equal to whereas we only avoid introducing an additive constant by saying that S64 .
The integral function just considered is the most simple function of zero order .
In carrying out the algebraical analysis of a theory of such functions , it would be necessary to consider the types \amp ; .
The asymptotic expansions for these successive functions are of successively lower orders of greatness\mdash ; they are never , however , of so low an order as , where is finite .
This agrees with the known theorem that an algebraical polynomial is the only uniforn function of such an order .
Unfortunately , unless we introduce new analytical functions defined by definite integrals , we cannot investigate formally asymptotic approximations for such types ; and until the properties of such new functions are investigated , we but express one unknown form in terin of another .
Simple Integral Functions of Finite Non-integral Ordcr Unity .
S65 .
In the investigations to which we now proceed of simple integral functions of finite non-integral order greater than unity , the theoretical considerations which have been given in detail for functions of order less than unity will for the most part be suppressed , and for brevity only the bare analysis will be written down .
We consider first the standard function ' where and is an integer such that Let , and suppose that is very large .
W. BAliNES ON UNCTIONS .
Take an integer such that Then so that ; and hence . .
Now , when is a very large integer , .
And , when is positive and greater than We use these Maclaurin approximations and rearrange the double series which results as the arithmetically asymptotic approximation for .
We obtain , in the limit , when the limits of the summable divergent series are taken for infinite , the asymptotic expansion MR. W. INTEGPAL FUNCTI \mdash ; \mdash ; I We suppose now that is a quantity which , when is very large , is ultimately equal to unity .
Then we may " " sum as before ( S 52 ) , the Fourier series , which are the various coefficients in the preceding expansion .
We have , provided is not integral , and provided .
And \mdash ; Also , exactly as before , the coefficient of in the asymptotic approximation for vanishes identically .
Therefore we have , provided , the asymptotic equality Thus , provided , we have finally This expansion is exactly analogous to the one previously obtained for and is to be regarded in the same way .
It must be borne in mind that has been assumed to be the arithmetic root of .
Had any other root been taken\mdash ; say the arithmetic root multiplied by , where is an integer , we should have obtained the asymptotic expansion valid , when , i. e. , when The expansion is thus valid everywhere except along the new lin of zeros .
MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
S66 .
We proceed now to investigate the asymptotic expansion for where is such a function of , let us say , that converges , and diverges , however small may be , being real , finite , positive , non-integral , and greater than unity , while is an integer such that Suppose that the result of reversing the equality is to give .
Let be a very large integer , such that As previously , we have . .
Substitute now the arithmetically asymptotic approximations giyen by the Maclaurin sum formula , and we have In this expansion is infinite , and there is no corresponding Maclaurin constant if , and only if , Use definite integrals and transform by integrating by parts in the same way and under the same restrictions as in S 54 , and we get The final integral gives rise to the dominant term of the asymptotic expansion .
As , it be written If we denote the value of this integral by , and if we put we have the final asynlptotic equality S 67 .
We notice the exact analogy between this expansion and the one previously obtained in the case when the order of the function is less than unity .
The only difference arises from the Maclaurin constants .
In the former case , all the constants corresponding to negative values of were zero ; in the present case , the first of them are formed from asymptotic expansions like .
, and give rise consequently to finite constants ; while only the remaining ones , formed from expansions like .
are such that We notice also the great elegance with which WEIERSTRASS ' exponential factor enters to ensure the finiteness of the expressions obtained in the course of the analysis .
Could we conceive an attempt to investigate , for functions of order greater than unity , the theory which we carried out for functions of order less than unity in the first paragraphs of this part of the present paper , we should at the outset be forced to invent again WEIERSTRASS ' great theorem .
to Fvnctions with Algebraic \amp ; quel ?
ce of Zeros .
S 68 .
We will now evaluate the first few terms of the asymptotic expansion for , where , and the 's are positive real quantities arranged in ascending order of magnitude .
Let , then on reversion of series we find When is positive , MR. ON GPAL FCTIONS .
47.3 . .
- . . . .
; and , where can be expressed in terms of Riemann functions , or the equivalent Maclaurin constants by formula As formerly , we put , so that arises as the Maclaurin constant corresponding to the asymptotic expansion .
Proceeding exactly as for the case when the order of the function is equal to unity we see that the asymptotic expansion oflog is \mdash ; . . .
, this expansion valid when V0L .
XCIX .
Thus , when , the first term of the expansion of the ratio And , when , the first term is , which is readily seen to be equal to ] S69 .
The expansion which we obtained is valid for all points except those near the line of zeros of the function and for all finite values of the quantities and such that has a finite value .
It must be carefully noticed that when any term becomes infinite through the occurrence of sines of integral multiples of , we must reyert to the genesis of that term to find the true form of the expansion .
when , the first term of the ratio just considered is that which arises from that is , where the double accent denotes that the terms corresponding to and are to be omitted from the summation .
Put now , then , with the argument previously used , we have to sum the series Now , when is not integral , we have seen that , provided Let us put , where is a positive integer and is very small .
Then we have , retaining only first powers of so that MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
47.5 The term which we seek is then the value of This is , of course , the term independent of in the expansion of in ascending powers of In exactly the same manner , if ( say ) is an integer , the corresponding term of our asymptotic expansion must undergo the same process of eyaluation and will give rise to a logarithmic term .
If one of the , say , is equal to obtain in the asymptotic expansion a corresponding logarithmic term Integral Functions of Finit Order .
S70 .
We proceed to considel the standard function , where is an integer Let , where is very large , and let be a large integer such that Then , employing the same process and argument as before , , when , and in accordance with the definition of S49 we put .
If , then , we suitably modify the analysis formerly employed we shall obtain , when I$ W. ON INTEGRAL S. the limits of the summable divergent series as tends to infinity are taken , the ithmetically asymptotic expansion where the double accent denotes that in the corresponding summation the terms for which and are to be omitted .
As , the of vanishes identically .
The is equal to It is then only necessal.y for us to consider the series If we put , we may write this series in the form ] that a summable divergent series may be differentiated , we find or Therefore where A is a constant of integration .
Now when , MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
and , by putting in the FOURIER'S series , which we considered in the preceding paragraph , we see that this is equal to Therefore so that evert now to the asymptotic expansion for .
We find on substitution that \mdash ; And thus , when is an integer , very large , and As formerly , this expansion is , in form , independent of the argument of S71 .
We may easily deduce this theorem independently as the limit of our former results .
Take the asymptotic equality where lies between and Put now ; then is the limit , when vanishes , of It therefore possesses all asymptotic expansion which is the limit , when vanishes , of rememberin that , and that terms which vanish when We thus obtain the same asymptotic expansion as in the previous paragraph .
Note that we have obtained our expansion by making increase up to the nearest integer .
If , on the contrary , we make decrease down to the nearest integer , there is breach of continuity in the introduction of an additional exponential factor .
Thus we have and therefore we have the asymptotic expansion Now , unless ; and therefore we have asymptotically the same expansion as before .
This paragraph is instructive in that it shows how the asymptotic expansion calls for another exponential factor in each term of WEIERSTRASS ' product as the order passes through an integral value .
S72 .
If now it is desired to construct a function which is the natural extension among simple integral functions of the ordinary gamma function , we take .
MR. W. BARNES ON INTFGRAL FUNCTIONS .
And now the asymptotic expansion of when is very large and is given by When , this formula is exactly the asymptotic expansion of for complex values of , which , as stated in S3 , was first obtained by STIELTJES .
For , when is an even positive integer , When is an odd positive integer , let us say , And Thus , when becomes , and the series becomes which accords with the usual result .
S .
It is obvious that we can now at once write down the asymptotic expansion for , where and is an integer , from the corresponding expansion for the function in which and is not integral .
The , of course , are assumed to be positive and in ascending order of magnitude .
The result is .
provided . .
be not integral .
Thus , not being integral , the first term of the asymptotic nsio of the quotient is We note that . .
When is an integer , we see on evaluating the limit which arises , that the dominant term of the asymptotic expansion is still the one just written down .
For , in this case , the only other term which might be considered first in the asymptotic expansion of the quotient is which , since is positive , is of lower order than S74 .
It is now evident that , if we are given any simple function of finite integral order , we can find its asymptotic expansion .
The analysis just given solves completely the case of algebraical zeros .
When the zeros are not algebraic we may , and , in fact , we shall have to introduce new analytical functions defined as indefinite integrals ; but there will be no essential difference in the theory .
It should be noticed that just as we have to take the principal values of the algebraically many-valued expressions which occur in the asymptotic approximation for functions of non-integral order , so we must assign principal values to the logarithms which occur when the functions are of integral order .
PART IV .
The Asymptotic Expansion of Repeated Integral Functions .
S75 .
As has been stated in the general classification of Part I. , an integral function , which is such that its zero is repeated a number of times dependent upon , is called a repeated function .
If the number of sequences of zeros be not infinite , the function is called a simple repeated function ; and it is obyious that such a function may be built up of functions , each of which possesses a single sequence of zeros .
We shall limit ourselves to the consideration of such functions .
The order of simple repeated functions with a single sequence of zeros has been previously defined .
Taking this definition , we consider , in turn , in the ensuing paragraphs , functions ( 1 ) of finite ( non-zero or zero ) order less than unity , ( 2 ) of finite non-integral order greater than unity , ( 3 ) of finite integral order greater than or equal to unity .
And , finally , an example is given of the asymptotic expansion of a repeated function with a transcendental Inasmuch as the principles which underlie the analysis are exactly the same as those which have been previously discussed , we shall give but a bare outline of the methods by which the results are obtained .
J8 ] Simple ctions of Order less than Unit , t S76 .
The most ' nnction of this type may be written the principal value of each term is taken when is any function of ; and where is a function of which increases without as increases , and which is such that is absolutely convergent .
The function is of finite less than unity : and , when ) is an integer , its zero is repeated times .
We take .
When we suppose that inversely .
Suppose that , then if we take to be a large integer such that ) , we have We carry out our analysis a ) which depends exactly upon the rgument employed in the corresponding case for -repeated functions .
We have at once , in the limit when Now , if be positive , where is a constant ) ending on and on the forms of and .
We call the Maclaurin i limit for and .
If be negative , the previouf , expansion will hold , but in this case , and the constant term vanishes .
Again we have .
an . .
VOL. CXCIX.\mdash ; A. 3 W. ON FUNCTION We shall find it convenient ) put , so that .
And now , if the limiting values when of the summable divergent series be taken , \mdash ; The last term vanishes as for the corresponding case of non-repeated functions .
After reduction , we have \eth ; f last If then we put , so that and may be called the zero and Maclaurin constants for and , we shall have the asymptotic approximation .
BARNES ON FUNCTIONS .
S 77 .
Considel now , as all of the genel.al formula just obtained , asymptotic expansion of ositive quantities such that is , and , therefore , snch that ] .
With our former notation The constant rises f the asymptotic equality and , therefore , . .
The constant arises from the asymptotic equality .
We may dily show that ) .
For , as has been stated , for all values If , then , we put , and expand each in powel .
S of , we may equate coefficients of similar powers in the If we equate coefficients of the first power , we find * Compare the process cal.ried out in $ 27 and 30 of the ' Theory of the Function or , chano s ' into Thus .
We have , therefore , The last integral is equal to Thus we have the asymptotic expansion We note that the first of this product vanishes when is even integer .
78 .
It is now possible to write down the expansion of where is ] of the , in which For such function is ltlerel , y , the product of the ower of power ) , and so on .
MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
We note that the constants which enter will be expressible in of the coefficients of and values of tIJe Inann function .
We might now inyestigate the asymptotic nsio of a repeated function of order less than unity with ebraic sequence of of the . .
the quantities . .
are , positive , and in ascending order of magnitude .
The analysis is , , such obvious extension of the correspolldigesult of Part III .
that it may be once supplied by the reader .
Repeated Simple Fv , nctions of -integral Or .
We next consider the asymptotic lnsio of function where , and is such that 2 is convergent , and 2 divergent , when is a small real positive quantity .
The analysis is an obvious modification of that in S , lo i ' , .
As example , we lllay ider the ) is not integrral , order of the function is We have , .
And , as in S 77 , Thus the asymptotic expansion of may be written Note that Hence , when we get the asymptotic expansion ' which with the of S 65 .
Simple Repeated Functions of Intenrol Order .
S81 .
It is obvious from the ations of SS70-73 that the asymptotic expansion obtained , where and is such , is convergent , and 2 divergent , will hold in the limit when vided that in any terms which become we reject the infinite part and keel ) only corresponding finite expression found by applying the usual methods of the calculus of limits to the subsidiary Fourier and Maclaurin series .
Consider , for exa1 , the function , where The asymptotic expansion obtained previously was ' Theory of the Gamma Function S27 .
Ibid. , S30 .
Now , when integer , becomes infinite .
sm We have then to substitute a corresponding finite expression del.ived by proceeding to the limit in the infinite terms of the subsidiary Fouriel series .
is not , the series and equivalent value re given by When is integral , this series , ( luitting the term in the second summation , is equal to the finite part of when ; that is to say , it is qual to We thus replace by since is a positive integer , the only term of the form , . .
\mdash ; 1 , 1 , 2 , . .
which becomes infinite is that for which This term is , ] lich arises from the Maclaurin series We have already taken account of substitution of for ] ) ; we need , therefore , only replace by We have nptotic expansion in which and both tegers , and is the " " genre\ldquo ; of the function .
It is inte ] esting to notice that the constants which enter into the asynlptotic xpansion this very , eneral function all values of the Riemann function .
S 82 .
When , the function is to an exponential factor an important function which I have proposed to call the -ple function .
These functions are derived fionl the nnultiple Gamma functions the coalescence of the parameters .
The theory of the simple function has been developed in the second of a series of papers on Gamma functions .
In that development I took and obtained the asymptotic expansion where A is the Glaisher-Kinkelin constant .
Putti n , the asymptotic expansion which we have obtained for the same nction in the present paragraph is Now , when is odd , and when ' ' Quarterly Journnl of Mathematics , vol. 31 , pp. 264 et Ibid. , S15 .
Ibd .
, S23 .
MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
, since A is given by the identity , terms which vanish when in is infinite , we have Thus the asymptotic expansion of the present paragraph may be written ' We thus obtain a yaluable verification of our results .
Simple Integral Functions of Indcx .
S 83 .
It is obvious that such a function as is of infinite order when is algebraic .
If , however , is of the same order as , the order of the function is finite , and can therefore be expanded in the neighbourhood of infinity by our methods .
We shall take , as an example of repeated functions of transcendental index , the product This function is of order , greater than or equal to unity .
Suppose first that is not an integer , so that is the integer next greater than without former notation - ' And where we take the limit when of the summable divergent series .
VOL. CXCIX .
; Thus Therefore Again , is given by ( 1 ) , , when Thus Also - is given by putting in this same expansion .
Thus Again , by substituting for , expanding in power of , and equating coefficients of the first power of in the asymptotic identity ( 1 ) , we readily find If , then , is not an integer , we have the asymptotic expansion S84 .
Suppose next that is an integer\mdash ; so that .
The analysis will , of course , be slightly more complicated .
The constant will be given by or , so that MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
And we shall have the double dash denoting that the terms for which are to be omitted .
We therefore have Therefore We thus have the asymptotic expansion We have now given examples of the asymptotic expansions of repeated simple functions with transcendental index in the cases when the order is or is not integral .
And it is evident that such examples might be multiplied indefinitely .
In the more complex cases the difficulties of the analysis will , no doubt , be very great ; but such difficulties in no way invalidate the theory which has been developed .
PART V. Applications of the Previous Asyrnptotic nsions .
S85 .
We proceed now to consider some applications of the previous theorems to such questions concerning integral functions as have been raised in the Introduction to the present paper .
In the first place , a knowledge of the asymptotic expansion of a function serves to determine the number of roots which it possesses inside a circle of given large radius .
Let us consider the simple example of the Gamma function , for which we have the asymptotic equality in which the terms neglected on the right-hand side are of lower exponential order than those retained .
By CAUCHY'S theorem the number of roots within a circle of given large radius 7 ' is determined by the integral being taken round the circle in question .
Now we may , to terms which vanish exponentially with substitute for its ' value given by the asymptotic expansion .
And this expansion is valid for all values of for which .
It is also valid right up to the two limits of provided the circle on which lies passes between two consecutive zeros of If , now , , we have , to terms which vanish exponentially with Now Therefore , to terms which are ultinuately exponentially small , Of course we know independently that the nulnber of roots is the greatest integer less than .
And the entrance of the terln might have been predicted priori , for when the circle of radius passes through a zero of we jump from \mdash ; to as we integrate round a small circle enclosing this zero .
S 86 .
It is interesting to notice that the analysis verifies itself in the same way for MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
the function , which , by a change of the indep endent variable , reduces to LAMBERT'S function .
For this function we have obtained the asymptotic expansion Therefore The number of roots within a circle of radius is , therefore , to terms which exponentially small when is large Since the function has no zero at the origin , we should have predicted the occurrence of the term \mdash ; S 87 .
We may now prove that , if the term of the of an integral function is algebraic and such that the zero is of non-integral order ( where is neither zero .
infinite , but greater or less than unity ) , then the number of roots of the function within a circle of large radius is to a first approximation is the maximum value of the modulus of the ction on the circle in question .
There are two cases to be considered according as is greater or less than unity .
We take the former , the argument will hold in detail for the latter by changing into Let be the function in question ; then , under the conditions enunciated , is equal to terms of lower order .
Hence , the number of roots is , to a first approximation , given by We thus complete and prove BOREL'S intuition .
S 88 .
When is an integer , the preceding theorem ceases to be valid .
But we now proye that the nunlber of oots to a first approximation is For , the same conditions still being supposed to hold good , has to a first approximation been proved to be equal to lower terms .
And therefore lower terms .
Hence , to a first approximation , which establishes the theorem in question .
S89 .
In the two preceding paragraphs we have assumed that we were dealing with non-repeated functions .
From the analysis of Part IV .
it is , however , evident that the theorems hold toto for repeated functions , the order being that which has been assigned to such functions .
We cannot , of course , attempt to prove the theorems for functions of multiple sequence until we have investigated the corresponding asymptotic expansions .
S 90 .
We may next write down a number of theorems relating to two or more integral functions .
It is obvious that the sum of two or more integral functions of simple sequence is an integral function of order equal to the largest order of the component integrafunctions .
We may replace additive signs by those of subtraction if the two component functions of largest order are not identically equal .
The large zeros of the compound expression are to the first of approximation equal to the large corresponding zeros of the component function of largest order .
The product of two or more integral functions of simple sequence is an integral function of order equal to the largest order of the component integral functions .
The number of zeros of the equation where the 's are integral functions of simple sequence and the 's algebraic polynomials , within a circle of large radius is ultimately to a first approximation equal to the number of zeros within that circle of the function of largest order .
S91 .
The expansions which have been obtained may be utilised to give a proof of BOREL'S extension of a theorem due to PICARD .
* * PICARD , 'Annales de l'Ecole Normale Supe'rieure , ' 2 ser. , t. 9 ( 1880 ) .
BOREL , ' Acta blathematica , ' t. 20 , } ) MR. W. BARNES ON INTEGRAL FUNCTIONS .
The identity in which the 's are integral functions of simple sequence of any finite or zero order not greater than some number , and the functions are polynomials of order greater than or transcendental integral functions , necessarily involves Since the identity holds for all points in the plane of the complex variable we may consider it in the neighbourhood of uppose first that the 's are functions of simple sequence of non-integral finite order .
If be the order of , we shall have near the identity where we have neglected in each term terms lower exponential order than those retained .
The identity will hold for all values of such that is not within a finite distance of the zeros of the The functions by hypothesis cannot be equal to one another .
As tends to infinity , one of them must become infinite to an order which exceeds the order to which all the others become infinite by a quantity of order greater than The corresponding term ( say ) is then infinite to an order greater than the order of any other term of the identity Since cannot vanish , we must then have The same argument may now be applied to the identity , and it may be pro successively that all the functions vanish .
And thus the theorem will be proved .
When any of the quantities are integral , a suitable modification of the formulae in accordance with S73 shows that the theorem is still true .
When the 's are repeated functions , a corresponding modification again establishes the theorem .
When the functions 's reduce to constants , so that , the theorem is still true , the functions being unequal .
S92 .
We pass now to the consideration of the resemblance between an integral function of simple sequence and its derivative .
And I would remark that , in the same manner as ROLLE'S theorem is proved , it may be established that the real zeros of such a function with real coefficients are separated by zeros of its derivative .
[ It cannot , however , be proved that the derived function has not other real or ( necessarily even ) number of other imaginary zeros .
] This theorem I shall not prove , as it is not connected with the main developments of the present paper .
We proceed , however , to show that such developments complete and to some extent verify this extension of ROLLE'S theorem , and that incidentally they furnish many criteria as to the nature of the derivative of a given integral function .
S 93 .
Let us consider , as an elementary example , the function of genre zero and order , where We have the asymptotic equality Remember , now , that it has been proved in Part II .
that we may differentiate an asymptotic equality of this type , and we obtain together with terms whose ratio to the terms retained tends to zero as tends to infinity .
From this expansion we see that ( 1 ) is of the same order as , ( 2 ) The zeros of are such that , when is large , we have with the usua notation lower terms .
Not only so , but theoretically , by finding successive terms in the expansion for , we ought to be able to determine the form of its zero as nearly as we please .
Practical difficulties will , of course , arise when we come , in the asymptotic expansion , to a term which arises from a transcendental term in the zero of .
Note that the formula for may be readily verified when For and , therefore , asymptotically .
MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
The large zeros of are approximately those of and the latter are such that , with the usual notation , We notice that the general form of given above shows that the large zeros of separate and are separated by those of , a fact which agrees with the extension of ROLLE'S theorem . .
From the example it is now evident that we are in a position to prove that , for all the types of integral functions of which asymptotic expansions have been obtained in this memoir , the order of the function is equal to the order of its derivative .
And not only so , but we are theoretically in a position to determine as nearly as we please a formula for the ( large ) of the derivative .
It would be tedious to consider in turn all cases which can arise : we will take one or two as typical of the rest .
As an immediate corollary of the eceding example it may be seen that the derivative of a simple non-repeated function of order less than unity with algebraic zeros of the type .
is a similar function of equal order , whose zeros are typified by S 95 .
As a suggestion of the possibility of extending the expansions of Parts III .
and .
let us next write down the first few terms of the asymptotic expansion of , where is any quantity of finite modulus , and The expansion will be ( S68 ) , and may be transformed into . . .
\mdash ; . . . . . .
VOL. CXCIX . .
3 . .
By the employment of extended Riemann functions of parameter , it is impossible to give a form of this expansion which shall include all powers of , analogous to the expansion of log , which involves BERNOULLIAN functions of as coefficients .
For brevity we content ourselves with the preceding first approximation .
S 96 .
By differentiating the expansion for , given in S 68 , we have at once , as is evident by the preceding paragraph , Thus the asymptotic expansion for is given by terms involving positive fractional ) powers of Thus is a function of the type where higher powers of Thus the differential of an integral function of order , where is not integral , is itself an integral function of order whose zero , when is large , will MR. E. W. BARNES ON INTEGRAL FUNCTIONS .
differ from the corresponding zero of the original function by the term \mdash ; together with terms involving lower powers of In an exactly similar manner it may be proved that the function admits an asymptotic expansion of which the dominant term is so that is ofintegral The term in the exponential just written down shows that we shall come , sooner or later , to a transcendental term in the expansion of the zero of .
Similarly the theorem may be established for the general simple non-repeated function of finite integral order .
As regards the application of the same methods to simple repeated functions it is only necessary to notice that corresponding to a zero times repeated of the original function there will be a zero times repeated of its derivative .
S97 .
We have now to consider whether the derivative of an integral function , all of whose roots are real , can have zeros other than the real zeros which by the extension of ROLLE'S theorem separate the roots of the original function .
For this purpose let us consider the difference between the number of roots of and of within a circle of very large radius This number will be , where the integral is taken round the circle in question .
Now by examining the various cases which can arise , it may at once be seen that the asymptotic expansion of is given by terms which yanish when .
Therefore to a first approximation we have If then the function is of genre , its derivative can at most have only zeros besides those demanded by the extension of ROLLE'S theorem .
And therefore when is odd , can at most have only imaginary roots .
* In particular when is of genre or ] , can have no imaginary roots .
From this theorem coupled with the expansion given in S5 and the equality BOREL , ' Fonctions Entieres , ' p. 44 .
LAGUERRE , ' ( Euvres , ' t. 1 , pp. 167 we see that all the zeros of are real if the zeros of are real\mdash ; a theorem due to MACDONA If the function be by an algebraic polynomial with real coefficients whose degree is , the derivative of the product can at most have only imaginary roots .
S 98 .
So far we have only considered integral functions whose roots are all real and negative .
If , however , we have an , function all of whose roots lie along a line other than the negative half of the real axis , a change of the independent variable will at once reduce it to an integral function all of whose roots are real and negative .
If then an integral function of genre have all its roots but lying in a sequence along a straight line through the origin , its derivative will at most have roots which do not lie along this line .
S 99 .
We now conclude for the present the applications of the expansions which have been obtained .
There are many questions which are still to be discussed\mdash ; for instance :\mdash ; ( 1 ) Functions of infinite order ; ( 2 ) Functions of multiple sequence ; Asymptotic expansions deducible from linear differential equations ; ( 4 ) The rate of increase of the coefficients of the TAYLOR'S series expansion of an integral function ; and so on .
Investigations in connection with each of these questions have been tentatively undertaken\mdash ; notably by BOREL , HORN , HADAMABD and POINCABE .
And I find it possible to extend , by methods of this memoir , many of the results which have hitherto been obtained .
But such investigations I leave for future publication .
* MACDONALD , " " Zeros of the Bessel Functions ' Proc. Lond. Math. Soc vol. 29 , p. 575 .

