]\gt ; E. W. RNES , , Felon of Comr nunicated by ssor A. R. , Sc. D. , .
S1 .
Introduction . . . .
tcment of tion .
PART S2 .
Definition of the ) . . . .
S3 .
Its Symmctry . . .
27 S4 .
terms of its .
S5 .
The formula ) S6 .
The of ) . . . . . . . . .
S7 .
The 1)ouble . . . . .
76 S8 .
Pueduction for the of meters S9 .
Expression S10 .
Expression of Tcrms of 277 S11 .
The of ) ) S12 .
The Value of . . .
S13 .
The Multiplication Formul S14 .
The Transformation .
15 .
The } whose Cocfficicnts 816 .
Derived nsions .
$ 17 .
tion of Note on the ve 1opnlcnt of the II.\mdash ; il'h Dvuble its S18 .
Introduction .
) .
Dcfinition of the .
The Diffcrcncc , , of tho ) S21 .
, , , S22 .
Properties of VOL. CXCVI .
E. W. BAIINES ON THE THEORy OF THE Page .
The Diffcrcncc elations for . . . . . . . . . . . . .
293 -cxprcssion for ) . . . . . . . . . . . . .
298 .
of the Double Function when the is cqual to one of the meters .
299 .
The Conncction with the Function ) . . . . . . . . . . . . .
300 27 .
tion betwcon and the Double Modul Forms .
302 .
The of the ttcr when the 's arc . . . . .
303 S29 .
ansion of its Dcrivnves . . . . .
305 .
Lxprcssio of Prodnct of ) . . . . .
305 S31 .
of in Tcrms of Double amma Iunctions . . . . .
308 S32 .
, , , , , , , , . . . . .
309 S33 .
, , , , , , , , , , . . . . .
310 of . . . . .
311 S ' 3,5 .
, , Dilfcrcnco for the ctio . . . . . . . . .
312 of Non-existcnco of the . . . . . . . . . .
313 III.\mdash ; C'on ) with S37 .
oduction . . . . . . . . .
314 finition of when , , and arc posilive .
Asymptotic tnsion in this } . . . . .
314 .
Definition of by the Gcneral Contour ] . . . . .
318 Deduction of tho symptotic sion in the Casc . . . . .
320 , 321 $ .
tion of Contour Integral Exprcssions for og its tives . . .
322-325 The Stirling Form .S cssion of the Contour Intcgrals Linc . . . . .
327 xprcssions as Contonr for the Modular Forms . . . .
329 S48 .
The of ) . . . . .
330 S49 .
Asymptotic Expnnsion of \ldquo ; ) .
Introduction of the Symbolic tion . . . . . .
332 S xtcnsion of Theorcm to two Parameters . . . . . . . . .
333 Asymptotic Limit for 335 S D2 .
, , , , 337 { o .
Alternativo Method of ining the Contour Intr for . . . . .
337 .
The of wheIl S is a Negative Integer . . . . . . . . . .
338 S55 .
, , , , for Positive Integrftl Valnes of.s .
Tabul tion of liesults . . .
339 S C. of the Simplc . . . . .
340 .
pplication to the Double iemann Function . . . . .
34 S C8 .
Idcntity with the Prcvious Definition . . . . . . . . .
344 S59 .
Proof of the esuIt of S53 for of . . . . . . . . .
345 of MLLIflIN'S Form when .
Application deriving the Fundamental Difference tions .
Notes . . . . . . . . . . . . .
346 PART Cl. . . . . . . . . . . . . . . . . . . .
349 NlhIA UNCTlON .
tion of in tcrms of of . . .
S63 .
Exprcssion of ) in tcrms of the ) . . . . .
The Case .
The esults of ined f the Contonr of S66 .
The FuIlctions 867 .
Introduction .
of ) . . . . . . . . . . . .
S68 .
tion for , ] . . . . . . . . . .
S C9 .
The of the First ccond 1 . . .
S70 .
tion of tho tiort of tho ltcrIl ve Proofs of the . . . . . . . . .
S72 .
The logues of . . . . . . . . . .
tion of . . . . . . . . .
JC6 S74 .
xpression of the kclin C of . . .
) ) S .
The csult of S73 in the of . . . . . .
S76 .
The Valnc of . . . . .
} S77 .
The tion of Conlour of 73 . . . .
370 PART .
The .
of finity .
The Asymptotic nsion when the . . . . .
} S80 .
Form of the Asymptotic usion of when is Posive with ) to the . . .
) ) S81 .
, , , , , , , , when is not with to S82 .
The Negative . . . . .
} S83 .
The Asymptotic xpnnsio of . . . . .
S8 .
The Asymptotic of . . . . .
Mcthod of Proof . . . . . . . . . .
S86 .
The Conlour Integral . . . . . . . . . .
} Sl .
The S1 .
INTRODUCTION .
THE present .
con tinues my researches in the theory of Previously to a certain extent I obtained known results by new methods : none IIIR .
W. BARNES ON THE OF THB the tions however have , I believe , been undertaken or suggested by other ticians .
In the first ) published in the connection I attempted to give a homogeneous of the gamma ction , considered from the point of view of function .
I introduced a parameter , and showed that the was subordinate to that of a function satisfying the difference equation any plex quantity .
That theory led naturally the the , satisfying the diffel.ence equation , and substantially { whose propel.ties could be differentiating the simple function with respect to the I next an extended function satisfying the two functional relations ; , and educible to the function when .
Several points in that paper suggested the formation of a symmetrical double gamma function , in which should be replaced by the quotient of two parameters and .
In the present investigation such a function is defined , and its theory developed in , I hope , complete detail .
The function is the natural extension to two parameters of the simple gamma function It is necessary for a complete exposition of the theory to consider the properties of lvl]at I propose to call double Bernoullian numbers functions : functions which the natural xtension to two parameters of the simple Bernoullian functions , considered in Part II .
of the earliest of the series .
Such a theory is developed in Part I. of the present paper .
In Part II .
I consider the elementary of the double gamma function .
It is shown that certain symmetrical modular constants arise as finite terms of asymptotic expansions in a manner exactly analogous to the origin of EULER'S constant Such con siderations lead naturally to III .
, in which are deduced from a contour integral , which is a double isatio of RIEMANN'S function , certain noteworthy asymptotic approximations , of which the most important is an extension * BA , " " The Theory of the Gamma Function ' Messenger of Mathematics , ' vol. 29 , pp. 64-128 .
The Ttleory of the G Function ' Qnarterly Journal of Mathematics , ' vol. 31 , pp. 264-314 .
BARNES , " " Genesis of the Double Gamma Function 'Proceedings of the London Mathematical Society , ' vol. 31 , pp. 358-381 .
AhfhIA NCTION .
of STIRLING'S .
By the aid of this it possible to the of the double functiou and the double .
constants its contour similar to those in III .
of ' of the Function .
' In Part IV .
I the multiplication and ) of double gamma functions well as certain curious forInulae , which espond to RAABE'S theorem for the slmple function , and elementary of ) general theorem connecting successive dents ( higher ) In Part V. the totic e of the double fnnction is obtained , and it is shown that the function can ot arise as solution of a differential equation whose coefficients are more simple nscendents .
There exist similar functions of any number of ) meters , these lscendents I propose to call multiple gamma functions .
I resel.ve the expression ) properties for tion elsewhere .
I ] lave worked out the theol .
for double gamma functions indelJendently inusmuch as , the complex being two dimensional , there are mally points in which ] analogy owIl : also , since many proofs the theory are , in their simplest fornl , inductive and , to be orous , require a knowledge of the theorem for the two ) .
Not only so , but in the case of the double mlna functions it is possible to give easily algebraical theory ( such as that worked out in II which is nor simple if one derived all formulae from the fundamental consideration of certain contour integrals .
I append a statement of the notation adopted in this ] place in the present series of tions where such notation is used for the first time .
ivati \ldquo ; \ldquo ; \ldquo ; DOUBLE GAMMA FUNCTION .
of Bernoullian Functions .
S2 .
In the " " Theory of the Gamma Function Part II .
, we the } ) Bernoullian function that solution of the differencc equation : where is a positive integer , which is such that it is an .
And it was proved that such a solution does exist .
In exactly the same manner it be proved that the equation has an algebraic solution , which is a ) rational integral polynomial of .
The difference veen any two solutions will be a simply periodic function of period , and will therefore be a if the solutions are both algebraic polynomials .
There thus exists a unique algebraical polynomial of degree , which is a solution of- the difference equation .
' with the This solution we call double Bnoullian function of with , and we denote it by .
By symmetry with this simple Bernoullian function would be denoted by .
We shall often omit the parameters and , when there is no doubt as to their existence , and write the function simply .
S 3 .
We now proceed to show that the double Bernonlliall function of of order is also the unique polynomial yhich is the solution of the difference equation with the condition For since : we have at once .
And therefore if we put we shall have Now and algebraical polynomials of , and therefore is also such a function .
And , therefore , since it is simply periodic of period , it must be a constant .
Thus we have constant . . . . .
( 2 .
) , integrating the relation ( 1 ) with respect to between and , we have since Integrating the relation ( 2 ) in the same manner between and obtain for the value of the constant And thus is th unique algebraic solution of the equation with the condition From the sylmetrical nature of the equations which give , we see that this function itself must be symmetrical in and S4 .
we assume . . .
the calculation of the hest coefficients may be readily effected .
For we . . .
Hence if we substitute the assumed expansion for and equate corresponding powel .
S of , we find DOUBLE GAMMA TION .
and so on .
On tions successively we readily , and , since Thus . . .
Further terms cnn be calculated if necessary .
It will bc seen , owever , that form what we propose to call double Bern ulliall numbers , whose ) ) without the llecessity of their tion .
Corollary .
We note that And hence It will be found that these occurrence in the present investigation .
Note also S 5 .
We will now prove that , We have , ' .
VOL. CXCVI .
\mdash ; A. MR. W. ON THE THEORy OF THE therefore Thus if we shall ( ' ' Gamma Function S15 ) .
imilarly eforc since .
is an.algebraical polynomial in , it is a constant .
making and remelnbering t , we obtain , which is the required result .
C. We now in a position to prove that and the same time the important relation .
Since we see from the fundamental difference equation that Jlence , if we put , in ] S 5 , we see that when , and ow the relation , We thus see that is independent of , since this is the only time in the tion just obtained which depends on * Putting then , we have when and : , which is one of the relations required .
And also another of the given relations .
The second integral formula of course be written down by symmetry .
We notice that in the notation formerly introduced Gamma Functiou S15 ) we have .
and therefore that each of these expressions when is odd , when is even .
Thus we see that when is odd .
F. W. BARNES ON OF TH : .
We now oduce double Bernoullian numbers analogous to the simple numbers introduced in the theory of the gamma function .
In th theory the simple ] Bernoullian number was defined by the relation the ?
] double number is givcn ) We by the theorem of S6we put , erefore S8 .
At point we conveniently note the reduction which takes place in the double functions and numbers when the parameters are equal to one anothel If we put we have as the single difference equation of the nth double function the relation , and the fnltction is defined as the lgebraical solution of this equation with the condition Put now , and we IIence , the other definition conditions bein satisfied , we see that hat is , the double Bernoullian function when the parameters are equal reduces simple Bernoullian functions and numbers .
It will be seen later that it is for this reason that it was possible to obtain all the expansions in the theory of the function in terms of simple Bernoullian functions .
Note that the above relation may be written .
GAMMA FUNCTION .
On differentiation we , so that , and therefore .
S 9 .
We now see at once that and so complete the expansion of S 4 .
For by theorem we have since the higher differentials vanish .
From the few terms found in S 4 we see that when ?
and , we , and thus we have the expansion in question .
S 10 .
It may now be shown that or , as we may write it , Remembering the value of , we thus prove that , when is even , , when is odd .
Substitute for in the fulldamental difference equation and we find , therefore , since , we have .
If therefore we put we see that is an algebraic solution of the difference equation , therefore can only differ by a constant from .
Determine this constant by making and we have the relation .
When is odd the last two terms cancel each other , and when is even vanishes .
Hence ] .
the fundamental diHerence equations we see at once that , and therefore we have the relation stated .
S 11 .
We may now show that , when is eyen , , a sinqple expression for the even double Bernoullian numbers which corresponds in some degree to the fact that the even simple Bernoullian numbers vanish .
On differentiating with regard to the result of the previous paragraph we find From the fundamental difference equations we have and , when is But our former relation gives us , when is even , : Hence , when is even , , which is equivalent to the relation required .
S 12 .
We now proceed to show that S We have Hence , integrating with respect to But Gamma Function S 19 ) so that , if , this function is an algebraic solution of the difference equation But this difference equation is evidently satisfied by ' Hence as these two solutions both vanish when , we have .
As a corollary we haVe on differentiation .
S 13 .
The multiplication theo1y of double Bernoullian functions may be conveniently expressed by the formula From the fundamental difference relation we have Hence satisfies the difference equation ' and is the only algebraical solution such that JIence , which is the relation required .
As a corollary we see that the th double Bernoullian number is homogeneous and of reeb ( in the For in S9 we have seen that in the expansion of the part of the coefficient of which inyolyes the 's is .
S14 .
The transformation of the ameters of the double Bernoullian function is given by the relation , as we to Let , then obtain at once DOUBLE GAMMA FUNCTION .
( '(Gamlna Function , \ldquo ; S 18 ) .
Thus is an algebraic solution of the rence equation satisfied .
The two solutions can then only difler by a constant , and thus Thus is an algebraic solution of the rence equation satisfied .
The two solutions can then only difler by a constant , and thus Thus is an algebraic solution of the rence equation satisfied .
The two solutions can then only difler by a constant , and thus by where is independent of To determine this constant , let us integrate between and .
We find and therefore by S6 , by S 12 .
( ' ' Gamma Function S 18 ) .
And therefore so that .
On substituting this value we have the theorem enunciated .
As a corollary we have on making , S 15 .
We now proceed to prove the expansion of fundamental importance in the theory of double Bernoullian functions:\mdash ; . . .
In this expansion the double Bernouilian functions have and for parameters , and the expan sion is valid provided is less than the smaller of the two quantities For within a circle whose radius is less than this quantity , the function has no poles , and hence it is expansible in a Taylor 's series of powers of Thus we may assume , for all finite values of . . .
where it is obvious that the functions of which enter as coefficients are au algebraical polynomials .
Change now into , and subtract the expansion so obtained from the one just written .
We find . . . . . . . . .
But in the " " Theory of the Gamma Function S20 , we obtained the expansion which may be written . . . . . .
Equating coefficients in these two expansions , we obtain and it is obvious that a similar set of equations hold in which and are interchanged .
Hence is a constant whose value , from the first term of the expansion , is by S4 .
Again , for these two expressions can only differ by a constaRt which S4 and the actual expansion is at once seen to vanish .
Differentiate the expansion thus obtained with respect to , and we find : . . .
and hence we have finally ( S6 ) . . . . . .
which is the expansion required . .
This expansion may be used to define the double Bernoullian numbers , and all their properties may be deduced from it .
A procedure analogous to the one here suggested will be the one employed in the general theory of multiple Bernoullian functions .
S16 .
Several expansions of constant occurrence may be deduced from the one just obtained .
In the first place , note that we may write the expansion in the form * . . . . . .
Put now , and we have by definition of the double Bernoullian numbers , . . . . . .
We thus have ( S 11 ) . . . . .
And the last series may be written Hence we find as the expansion from which the odd double Bernoullian numbers may be derived .
Finally if we integrate the fundamental expansion of S15 with respect to between and we tain . . . . . . . . .
or , as we may it , . . .
\mdash ; -- . . .
as the expansion from which the double Bernoullian functions themselves are at once obtained .
All these expansions are valid within the circle whose radius is the smaller of the two quantities and S 17 .
Hitherto we have considered the double Bernoullian function as defined by one of two diflerence equations , each of which involves the simple Bernoullian function .
We now to } ) rove that , to a linear function of is the only rational integral algebraic function of satisfying the difference equation In the first ) lace it is at o1lce evident that does satisfy this equation .
the difterence of any two solutions is a solution of Putting , we have Hence , if is ] ) diflerence of two algebraic solutions of the original equation , ( will be function , and therefore a constant .
And we sball constant , so that if is to be an algebraic polynomial , it must be of the form where and are constant with respect to DOUBLE GAMMA FUNCTION .
Thus the difference of any two rational integral algebraic solutions of is of the form .
Whence the theorem in question .
[ Dr. E. W. HOBSON has kindly pointed out to me that the analysis of the preceding paragraphs would be much simplified by starting from the direct de{inition of the double Bernoullian function in S17 .
We should thus define the double Bernoullian function by the expansion . . . . . .
On differentiating with respect to , we get the expression of S 15 .
From the relation we obtain the fundamental diflerence relations for the double Bernoullian function .
The result of S 10 follows from the identity and that of S14 from Inasmuch as theoretically the properties of an algebraical polynomial should not be derived from consideration of the coefficients of an series , the original investigation has been retained .
I had already proposed to myself to work out the theo1y of multiple Benoullian functions by a method closely allied to that suggested by Dr. HOBSON.\mdash ; Note added , 1900 .
] PART The Double Gamma Fnnction and its Elementary Properties .
S18 .
In the elementary consideration of the simple gamma function it was found to be necessary to rely on two algebraical limit theorems:\mdash ; ( 1 ) EULER'S theorem ( 2 ) STIRLING'S theorem n In analogous treatment of the double gamma function we may expect that similar lilnit theorems will be required .
This , fact , is the case ; but for our present purpose it is sufficient to take particular cases of the asymptotic expansion for .
To make use of this approximation we need only remember that Theory of the Gamma Function S39 ) if and be any finite complex quantities , and a positive integer , .
We suppose that such values of the logarithms are chosen that additive terms involving do not enter .
In other words , we shall say that the logarithms have their absolute values , the just written being merely a convenient way of the identity On differentiating this identity with respect to we have with the notation of S 2 of the " " Theory of the Gamma Function S 19 .
The double gamma function of with parameters and .
we write .
there is no doubt as to their presence the parameters are omitted .
From this function we form the subsidiary system .
and so on .
As a definition we assume This double series is , by EISENSTEIN'S Theorem , *absolutely convergent , provided the ratio of to is not real and ative .
This limitation on the parameters holds tbroughout the whole theory of double gamma functions .
It corresponds to the limitatio ] ] in ] WEIERSTRASS ' theory of elliptic functions that must not be a real quantity . .
FORSYTH , " " Theory of ) tions , \ldquo ; S56 .
DOUBLE GAMMA FUNCTION .
We shall show that by successive egrations we may determine as a function symmetrical in and such that where ) ( ' ' Theory of the Gamma Function S 31 ) , and and are integers ( unity or zero ) , to be determined in accordance with the detailed theory which we proceed to give .
And the function so determined will be unique , provided S 20 .
We readily see that the function \mdash ; -1 .
' where satisfies the two difference equations , where here , as always , we suppress the parameters of the functions and when these parameters are supposed to exist in perfectly general form .
For we have at once from the definition-series .
Gamma Function S2 ) Next , we may show that the function satisfies the two difference equations , whatever be the value of the constant For the series for is absolutely convergent so long as is regarded as one term , and we may subtract two absolutely convergent series by a term-by-term process .
Hence we have immediately and therefore Theory of the Gamma Function S 2 ) .
Similarly , .
S21 .
It may ow be shown that the function satisfies the two difference relations for certain values of the numbers and , provided the principal values of the logarithms bein taken .
We may write in the form and now we obtain at once Hence we may take provided DOUBLE or , utilising S , provided .
' But ( ' ' Gamma Function Part IV .
) we know that when is vely large , and 1lot real ) negative , terms w vanish when becomes infinite .
In every case the principal value of the logal .
whose amplitude lies between and ithm is to ) , i. e. , that Now in all cases except when lies in the region nled I lines to the and ( shaded in the figure ) .
When does within this region , readily see if I , the imaginary part of , is positive , and if I , the imaginary of , is negative .
Thus terms which vanish when is infinite , where kunleVOL. .
ss lie within the region between axes to \mdash ; 1 and 1IR , .
W. BARNES ON THE THEORy OF THE where upper lower taken according is ) ositive egative , )when does lie this rion .
, we have the derived expansion* which nish when becomes infinite , value the logarithm again taken , and being determined as before .
Substituting in the expression , we have , unless does d lie in the region ) unded by lines origin to is understood , of course , principal values of the logarithnls are to taken .
] does When , in the figures yithiu the region of exception , does not lowel ' sign being taken according as is positive or negativ the , we see at once that it is impossible that should should not lie within region bounded by the lines the origin to - itnd Take now is to say , let ?
be such that we have , unless does lie in exception , according is positive or , when this exceptional circumstance takes place .
to , we in expression .
The one in question , however , may bo dily eished by the methods employed for ) .
AMMA FUNCTION .
Then But this expression is symmetrical in and and we must therefore have the analogous elatio does where , unless lie ithin the bounded by does not from the origin to ) , in which , the upper lower being taken as I is positive or ative .
Provided therefore we have , lyitl the ussiglled values of and the two difference relations The function ) ) to call the first double It will nently be ) ressed i terms of the functiolt into the theory of the functioJlq G ) enesis .
It will be seen later that ebra of the double function would have been slightly simplified had modified value been taken for function , and the analogue to be 1 did not observe this until the theory been etely developed , ctnd the matter is scarcely of sufficient importance to delnand laboul which a would entail .
Corollary.\mdash ; Notice that it ) proved incidentally that is infinite , when is infinite , like S 22 .
As the numbers ellter constantly into the analysis , it is to consider their propel.ties .
Suppose that the functions base ) , which are real when is real and positive , and which are by cross-cuts along the axes of and -.respectively .
MR. W. ON THE THEORy OF Then it is readily seen that lo By insl)ection of we see that and both vanish if the difference of the of is less than , these nmplitudes being measured between sitively or negatively the positive half of the real axis .
In particular when the of and are both positive , and both vanish .
Not but in all cases either or must vanish .
, if GIlce of the amplitudes of and is greater than and c , annot vanish .
In fact , in this case we have the important relation upper or lower being taken according as is ative or positive .
This result is eometrically ; in the figure , for instance , two cases are indicated in yhich is negatiye .
ondi to the unaccented value of ' to the ted value of in cases ?
) No such simple expl.ession can be given for , a number which is of constant Jlce i the tbeory .
the values for and previously given , we see that when the include the of within an angle less than two right angles , the values of are given by the table GAMMA FUNCTION .
and therefore , the upper or lower signs taken as and have the same or opposite signs .
S23 .
It may now ) shown that , if have any arbitrary value , the function .
where , will satisfy the two difference relations ; \ldquo ; where and are the numbers previously specified , and the principal values of the logarithms being taken .
Observe that with the notation previously introduced ( ' ' Theory of the Function and 31 ) we may write these difference relations in the -d- The proof , to which we ow proceed , is exactly analogous to the one just iven .
We have \ldquo ; , where and are understood , as always , to meau and Substitute now the value of obtained in S21 , and we find by the employment of the identity of S 18 .
Their principal values must throughout be assigned to the logarithms .
But , as has been kQeen in S21 , from the formula obtained in the " " Theory of the Gamma Function , we have when is large and not in the vicinity of the axis of terms which vanish when is infinite , where , unless lies within the region between the axes of and , in which case , the upper or lower sig being taken as is positive or tive .
The principal value of is to ) taken , and the prescription to be given to is left indeterminate : it is obvious that we only get additive telms involving , which vanish in the sequel .
Inaslnuch as when ?
is large , none of the points , , and lie in the vicinity of the negative direction of the axis of , we may substitute the values given by asymptotic expansion in ) expression for We shall On reduction we now see that : : S We must consider the three possible cases which l both vanisb .
' ( 1 ) Firstly , when ) does lie ithi the boun ded by the axes to \mdash ; 1 ) In this case the uppel lower sig being taken is positive or negative .
And we have the upper or lower sigll being taken as is positive or negative .
But in this case , the slgns being chosen in the If then we take .
W. BARNFS ON THI THEORy OF THE we shall ( 1 ) .
( 2 ) Secondly , when lie within the region bounded by axes to and In this case We shall have then the same relation provided we take the or lower sign b taken according as is positive or negative .
( 3 ) The case , when lie within the region bounded by axes , is easily seen to be impossible .
In all othel cases we shall have ] relation ( 1 ) , provided ( 2 ) .
pose now that we had investigated similarly the quotient we should have obtained the diiference equation , where has the value , let us say , given by equation ( 2 ) , except in two cases .
( 1 ) When lie within the region bounded by the axes to \mdash ; 1 and , in which case the upper or lower sign being taken according as is positive or negative .
DOUBLE GAMMA FUNCTION .
( 2 ) When lie within the region bounded by the axes to and , in which case But the cases ( 1 ) and ( 2 ) precisely the same , as an inspection of the shows at once , except that is positive when is negative , and ) versed .
And , similarly , the cases ( 2 ) and ( 1 ) are the same , with a similar change .
Hence the values which must be assigned to , in order that the equations may co-exist , are precisely the same .
We shall have then these two , provided where stands for \mdash ; ( the principal values of the logarithms being taken ) , except in two cases , ( 1 ) When does lie within the region bounded by the axes to and , in which case ( 2 ) When does lie within the regio1l bounded by the axes to and , in which case V0L .
XCVI .
MR. .
W. BARNES ON THE THEORy OF THE Since or always vanishes , we see , on combining these results , that in all cases But , the principal values of the logarithms being always taken .
Hence As a corollary , we see that , when is very large , is infinite like .
S 24 .
We now determine the constant in the expression by the condition assigned in S 19 , that This at once brives us It is evident that , with the conditions of S 19 , one and only one function can be constructed , this is the double gamma function , which is such that DOUBLF GAMhIA FUNCTION .
where are two constants , which we call the iirst and second double gamma modular functions of the parameters and These constants are given by the relations , where the logarithms are such that their principal values ) be taken .
And the theory is the natural extension of that of the simple gamma function , which is such that where the constant is given by the relation LtjL and the pl.incipal value of the rithm must again be taken .
S 25 .
We may now see at once that . .
For we have seen that where is either zero or unity according to the deterlninabion of S22 .
But and as we see immediately from the product expression for .
'300 MR. .
W. BARNES ON THE THEORY OF THE Hence , making , we have which is one of the relations required .
We thus see that is independent of , and not only so , but its value is substantially a quantity that appeared several times in the theory of the simple function .
Thus we saw cor .
" " Gamma Function that ; that ( S8 ) the most general form of Stirling 's theorem was seen to be We now see an additional reason why it was proposed to write , and to call and the two simple gamma modular forms , the latter being sometimes called the simple Stirling modular form , We shall see that there exist three double gamma modular forms , and of exactly analogous nature .
S26 .
We proceed now to connect the function with Alexeiewsky 's function , some of whose propel.ties were investigated in " " The Genesis of the Double Gamma Functions In the first place , we take , and then we have where and wherein , And it is evident that the coefficient of in the last exponential may be written .
The general solution of the difference equation is , : where is a function of simply periodic of period Hence .
But has been seen to be an expression of the form , where , and are independent of .
We thus have ; so that and , where is some integer .
Henceandsince . . .
( 1 ) , we have at once Now satisfies the where and the principal value of is to be taken .
( The same remark , of course , applies to every many-valued expression of this nature which occurs in the course of the investigation .
) Employin the relation ( 1 ) in conjunction with this equation and the equation ' we obtain which reduces to But we have seen ( S 22 ) that for , unless the difference of the amplitudes of and is greater than in which case , according as \mdash ; I is positive or negative .
We thus find , and incidentally we obtain a yaluable verification of our results .
And now finally , the relation between the two forms of double gamma functions .
S 27 .
From the relation just found we may at once express the gamma modular constants and of the former theory in terms of For we have : where , , and also " " Genesis of the Double Gamma Funotions S10 .
for we may suppose that the terms are represented by the corners of the small squares into which the positive quadrant is divided\mdash ; in the new grouping we take together all terms on a line equally inclined to the two axes .
Thus . . .
So that when is very large terms which vanish when becomes infinite similar terms .
Hence which is the first relation , In the second place we have , when and are equal , and by the same method of reasoning as before . . .
' . . . . . .
ecting terms which vanish when is infinite , and thus .
BARNES ON THE THEORy OF THE Again , it no finite zeros , for its zeros would be those of for , 1 , . . .
And its poles are given by Thus uo zeros or poles in the finite part of the plane .
now into , and we have Now the product last itten nmst be convergent , all other terms of the identity finite finite values of , iuld this product is evidently of the form , where and functions of and only .
Hence we must have Now we have proved that .
Thus if we put ) ' we shall have and similarly we shall have Hence is a doubly periodic function of the third kind , with no finite zeros or poles .
must for in HERMITE'S expression .
of such a function the functions are each associated with a finite non -congruent zero .
determine we , and obtain Differentiating logarithmically , the identity FollsyTH , ' of Gamma Functions S2 .
It is an interesting verification to actually transform the one formula into the other , making use of the relations established between and .
On account of the symmetry of the present functions , the formula corresponding to that given in S 8 of the ' Genesis of the Double Gamma Functions\ldquo ; may written The product formulae just obtained correspond to the expression of the function as an infinite product of circular functions .
Such a circumstance , of course , at once prompts us to try and find a formula corresponding to the expression of the function as a ) of exponential functions .
But it is readily seen that such is an impossibility .
We cannot express the double gamma function as a sum of an finite series of simple gamma functions of varying arguments .
It is this fact , combined with the absence any quasi-addition theorem for the double gamma functions , which precludes the possibility of any collection of formlae rivalling in number and elegance those of the doubly periodic functions .
:$08 MR. .
W. BARNES ON THE THEORY OF S31 .
We proceed now to express WETERSTRASS ' elliptic functions in terms of double fUnctions .
In ' notation of elliptic functions we have ' where and and are the periods of .
Now by definition Representing the various terms by the corners of the parallelograms of the figure , we readily see that Hence , using the natural summation to express the left-hand side of this relation , we have and therefore , on integration , where is constant with respect to Evidently Now where the values of the logarithms are to be taken .
the upper or lower sign being taken according as I is ) ositive or negative , where and are determined by the relations of S 22 that Take the expression for given in S32 , we find by use of the formulae @ , where , unless does does lie within the region bounded by axes from the origin to and in which case , according as is positive or negative , and and ?
are obtained by changing the signs of ( i ) both .
and ( ii ) respectively in this formula .
Thus the infinities in double summation being equal in absolute magnitude , that value of being takeu which is ( the same side of the axis ( cv And , similarly , where where has its usual meaning , and is obtained from just as is from tities r But we have seen iu S22 that , if the differeIlce of mplitudes of and be than the ' lower sign being taken according as is positive or negative .
Similarly it be proved that if the difference of amplitude of is , according as I is ) And . . . . .
and is while , finally , . . . .
I or ; and is I is or In a other cases the differences between corresponding But the difference of amplitude of one , and of only one , pair of the can be greater than Hence must equal And it is to see , by taking the particulal cases which can arise , that the nrust be taken is positive or negative .
have then finally and therefore the being taken according is ] ) ositive t negative .
merely indicate the relation of the which we found to the known relations : By S33 ; DOUBLE FUNCTION .
31.3 and by the previous paragraph , the expression on the -hand side of this relation will evidently reduce to the form where S is some constant whose value may be readily seen to be given by This value simplifies on detailed consideration .
When according as is positive or negative .
When , there are four possibilities ; \mdash ; ( a ) When lies within the axes to \mdash ; 1 and , , , , , , , , and , , , , , , , , and and S , , , , , , , , and In cases ( a ) and S \mdash ; , the upper or lower sign being taken according as is positive or negative ; and in cases and the upper and lower signs are interchanged .
We thus see that in all cases so that we have the required equation .
Similarly we find .
The verification of these results affords substantial proof of the general correctness of the signs which are involved in the work .
S 36 .
It is interesting finally to notice that just as the gamma functions do exist when is real and negative , so the elliptic functions demand that shall not be real .
The condition that must not be real and negative arose explicitly at several stages , and might have been predicted For , when is real and negative , it is obvious that excluded .
will have a zero value at least once .
And thus the function will be infinite independently of ; that is to say , it ceases to exist .
V0L .
cxcvr .
:$14 MR. F. W. BARNES ON THE THEORy OF THE For all other values of and the function exists .
But the product , and consequently , will not exist when either or is.real and negative ; that is , when is real .
The criterion for the existence of multiple gamma functions ( -ple where is greater than 2 ) is more intricate , and , as we know , -ply periodic functions do not exist.ae PART III .
the Double .
The Double Riema } S 37 .
In the theory of the simple gamma function it was shown that the intervention of a definite integral , coupled with the theory of asymptotic approximations , it was possible to obtain contour and line to express EULER'S constant , and the logarithm of the simple gamma function and its derivatives .
We now proceed to show that it is possible to extend the method thus previously employed so as to obtain expressions as contour and line integrals for the ganlma modular constants and , and the logarithm of the double function and its derivatiyes .
It will be noticed that when the real parts of and are positive , the numbers and which intervened in Part II .
vanish , and there is consequently a noteworthy simplification of the formulae obtained .
This simplification extends also to the definite integral expressions , and consequently we shall first investigate the theory in this case , proceeding subsequently to contour integrals of greater complexity .
Finally we make use of an extension of MELLIN'S method of defining the simple function by a series instead of a contour integral , and we show that there is complete agreement between the formulae obtained in the different ways .
S 38 .
When the real parts of and are positive , and when in addition the real part of is positive , we define the double Riemann function for all values , real or complex of , by the integral wherein being real when is negative and being rendered uniform by a cut along the positive direction of the real axis .
The integral is to be taken along a contour enclosing the origin ( but no other pole of the * The for fnnctions which substantially multiple gamma functions have been ussed 1 CRANI , ' ' vol. 29 .
DOUBLE GAMMA FUNCTION .
subject of integration ) , and the positive half of the real axis ; and extending from as in the figure .
Under the limitations specified the integral is , in genel.al , finite .
Moreovel , by a theorem previously we have under such limitations ' the latter expression having its principal value .
And therefore all the integrals being taken along the fundamental contour .
When is a large positive integer , we proceed to throw the two rals last written into the form of asymptotic series .
For this purpose consider the expansion obtained in S which may be Wl.itten . . . . . .
We showed that this expansion is valid provided was less than the smaller of the two quantities Outside this circle the series diverges .
But within the region bounded by lines goin to infinity from the poles of ' Theory of the Gamma Function $ 22 , 33 and 34 .
the series will , in the language of M. BOREL , summable , that is to say , within this region ( which is conveniently bounded by straight lines from the points which pass to infinity through the remaining poles ) it is possible from the values of terms of the series at any point to obtain , by the employment of intermediate functions , a magnitude , independent of these particular intermediate functions , which is the value of the function at the point .
If then on any of such a series we perform the operation which is expressed by we shall expect to obtain a quantity which is the n-th term of a possibly , and even probably , divergent which in turn is , by suitable operations , summable to the value which results from the performance of the fundamental operation on the function of which the original series is the expression .
Such considerations being understood to underlie the operations , we have Now it may be readily deduced from the results obtained in Part I. , that BOREL , ' ville , Ser. , vol. 2 , pp. 103 de Ecole Normals Superieure , ' 3 Ser. , vol. pp , 1 et se 3 ] 7 In an exactly similar manner we find , since that ; If now we group together all the results which have been obtained , we find the asymptotic quality , true for all values of , and for values of , and , whose real parts are positive , S. proceeding bowers overgency i is aseries of powers of , whose line of con .
( A ) It may readily be seen , as for the case of asingle parameter , that the serles This series is summable by an almost evident modification of the application of BOREL'S ideas , which was employed in the " " Theory of the Gamma Function S38 .
It is thus the asymptotic equivalent of the sum ; and it satisfies POINCARE'S* criterion for asymptotic equality that the difference between this sum and the first terms of the series has its absolute value less than a quantity of order S39 .
The function ) has been defined , and the asymptotic equality ( A ) has been deduced only for the case in which the real parts of , and are all positive .
It is natural to try and use the equality to define the function for all values of , and In the first place , it is evident that when and are both positive , the equality ( A ) holds for all values of , for the various terms of the sum and the equivalent asymptotic series are continuous for all but an enumerable number of values of , and .
Hence in this case may be defined as the term independent of in the equality .
So also when is a real positive or negative the equality will hold for all values of , and But when is not an integer , the various terms involving in their index are multiform functions , and to ensure uniformity we have to assign definite cross-cuts to the logarithms which arise in the equivalent exponentials .
When , as under the limitations for which the equality ( A ) has been established , these cross-cuts are formed by a line outside the smaller angle between the axes of and , the expan- sion is perfectly valid ; but when the common cross-cut lies within this angle , terms arise similar to those which occurred in Part II .
of this paper , which are multiples of and involve And , therefore , if we attempted in this case ( see the second figure ) to define absolute term asymptotic equality such ( A ) S 38 , where for complex values of the principal value of each term is taken , we should ultimately 'Acta Mathematica , ' vol. 8 , pp. 295-344 ; Ilique .
lest , ' vol. 2 , pp. 12-14 .
DOUBLE GAMMA FUNCTION .
{ 321 unless the axes of and embrace the negative half of the real axis .
In is latter case , since the imaginary part of the initial value of lies between and , as also does that of , we are giving a different prescription to the many-valued function which occurs in the subject of integration .
We therefore have clz where has the values which have been assigned to it .
Now , the logarithm having its principal value ( ' ' Theory of the Gamma Function p. 107 ) .
Hence where , unless and embrace the axis of \mdash ; 1 , in which case , the upper or lower sign being taken as is positive or negative .
Finally then where the latter function , where has a cross-cut along the axis of - , and is real when is real and positive .
In other words , has its principal value with respect to the axis of S41 .
If now we apply to the integral the procedure of S , we shall , for all values of , and , such that is ) ositive with respect to the , obtain the asymptotic equality VOL. CXCVI.\mdash ; A. MR. E. W. BARNES ON THE HEORY OF THE ' wherein all the functions with as index have their principal values with respect to the axis .
It proves convenient to consider these functions as having their principal value with respect to the axis of .
In order that this may be the caHe , we must multiply the integral by where , as in S 39 , , ullless and includes the axis of \mdash ; 1 , in which case , as is negative or positive .
Remembering the definition of given at the end of S 39 , we see that we obtain for our fundamental asymptotic equality an expression which in form is identical with ( A ) S 38 , but in which the many-valued functions with as index have their principal values with respect to the 'axis of .
It is evident that the equality will hold for all values of will serve define for all values of , and S 42 .
We proceed now to take such particular cases of the general asymptotic equality which has just been obtained as lead to expressions for the logarithm of the double gamma function and its derivatiyes .
Suppose that is a positive integer greater than 2 ; then , making 1 infinite in the general asymptotic equality , we see that , where .
This relation is true for all values of and ; it is the first of a series connecting the double zeta and double gamma functions .
Let us next put for , where is a small real quantity and is , as before , positive integer greater than 2 .
Then , provided is positive with respect to and the integral being taken along the -contour , and being the integer defined at the end of S 39 .
For take the integral along the contour embracing the axis , and suppose the contour to expand so that it embraces also the axis of Then since is positive with respect to and , and since the between the axes of and ) is less than , the value of the integral will be unaltered , for its value along the part of the great circle at infinity between the axes of and is zero .
Suppose now that the contour is taken to lie on the infinite-sheeted Neumann sphere , whose sheets intersect in the cross-cut from to , on which the subject of integration of the integral is uniform .
We may , without the value of the integral , deform the cross-cut so as to take up a position along the axis of instead of along the axis , provided that in doing so we do not give a new specifica- tion to the logarithm .
The latter phenomenon will occur when and embrace the axis of , in which case we take the first contour in a sheet in which can assume real values , while the second is taken in in which for real negative values of is equal to After deformation of the cross-cut we may compress the contour so that it embraces the axis of .
It is easy to see by this repetition of the argument previously employed in the " " Theory of the Gamma Function that we have , * \amp ; 4 where has the value previously given .
S45 .
The assumption made in S 43 for the values of the constants and will therefore lead to the relation Theory of the Gamma Function S37 ) .
E. W. BARNES ON THEORY OF THE Now , by considering the various cases which can arise , it may be readily seen that These constants all yanish unless and , embrace the axis of .
When this take place , that lies above , and below , the real axis .
Then when ( lies above below the real axis .
when ( both lie below the real axis .
when ( and both lie above the real axis .
when ( lies below and above the real axis .
We get similar sets of values when the imaginary parts of and have opposite signs to those just assumed .
In all cases and , therefore , with the values assumed for and Similarly we should find But these are identically the fundamental formulae for the double gamma function found in S 23 .
The values assumed for and are therefore correct .
We have , then , the two important formulae and DOUBLE GAMMA FUNCTION .
which express as contour integrals the double gamma function , and the third double gamma modular form .
The first relation only holds when is positive with respect to the .
The second is valid for all values of and )ject of course to the dominant condition that is not real and negative .
It is wolth noticing that the first formulae may also be written S46 .
Subject to the condition that the real parts of and are positive , we may now express our contour-integrals as line-integrals .
Consider the By hypothesis the logarithm has a cross-cut along the axis of , the initial value of its imaginary part lying between and .
Hence if the contour of the integral be reduced to a straight line from to , where is a point on the axis of very near the origin , a circle of small radius round the origin , and a straight line from back again to , we shall have The logarithm in this second integral , which results from the small circular contour su1Tounding the origin , has its principal value .
The integral itself is evidently equal to ( S 15 ) terms involving positive powers of ] \mdash ; terms which vanish with Thus ) terms which vanish with Mli .
1 W. BARNLS ON OF TH Now , as has been seen in " " Theol.y of the Gamma Function S28 cor .
, when is ositive , terms which vanish with And cvidcntly under consideration is thus equal to terms which vanish with If we make coincide with the origin , the integral last written remains finite and we have .
equality may equally be written under the assigned conditions that is positive with respect to the , and that the real l of is positive .
We thus expl'ess the arithm of the double gamma function as line-integral .
In to a line integra ] , we notice that we have therefore DOUBLE GAMMA FUNCTION .
Therefore on making , we have by S45 , On differentiating the fol.lnulae which express the of the double gamma function as line and contour integrals , we obtain .
Similarly , again differentiating , And , if be greater than 2 , Notice that , when we have the more narrow restrictions , the real parts of , and are all positiye , the constants , and all vanish , and there is a substantial simplification in the formulae .
S 47 .
We may now deduce expressions as line and contour integrals for the first and second double gamma modular forms and .
We have seen ( S 22 ) that and , therefore , on making so that , by the last paragraph , VOL. CXCVI .
The additive term will of course vanish when the real parts of and are positive .
Similarly , we have so that the first double gamma modular form is expressed as a line integral by the formula It will be noted that for the modular forms , we have , making vanish , obtained line integrals which are in general finite , although in our fundamental formula the restriction was made that the real part of should be positive .
This restriction was necessary to ensure that the contour integral should be finite at .
It is clear from the mode of generation of the line integrals , that the process which has been carried out is perfectly valid , since by the introduction of the terms , \mdash ; allowance has been made for the manner in which the contour integral tends to an infinite value as tends to zero .
S 48 .
At the beginning of S 43 we entered on the investigation which has just been given by integrating with respect to under the sign of contour integration , and in this we deduced the contour for from that for .
WG now ) roceed to show how the contour integral for may be obtained without the employment of this process .
For this purpose we take the fundamental asymptotic equality of S 38 , valid for all values of and , the many-valued functions with as index having their principal values with respect to the axis .
OVER DOUBLE GAMMA FUNCTION .
where , if is positive with respect to the may be expressed by the integral .
( S39 .
) Make now , where is very small ; expand the various terms of the asymptotic equality in powers of those higher than the first , and we obtain , if the real part of is positive , log pn This equality will hold for all values of if the integrnl be eplaced by , the logarithms having their principal value with respect to .
Equate now the absolute terms in this asynnptotic equality , and we find , if is positive with respect to the But we have seen ( S 17 ) that MR. .
W. BARNES ON THE THEORy OF THE Hence , provided is positive with respect to the and for all values of and .
S49 .
Secondly , equate terms involving the first power of We obtain the asymptotic equalit ) valid for all values of and , provided the logarithms have their principal values with respect to the axis .
In order that the labour of writing down cumbrous formulae like the one just obtained may be diminished as much as possible , we propose to introduce a symbolic notation gested by CAYLEY'S notation of matrices .
* If be any function of , we shall represent symbolically by the suffix 2 denoting that we are dealing with two parameters .
the difference equation for double Bernoullian numbers is written Similarly ] denotes the function *CAYLEY , ' Collected , vol. 2 .
The corresponding theory for multiple fumctions will be developed by ] ) a symbolic 1totation ab iniiio .
DOUBLE GAMMA FUNCTION .
[ The analogy with the matrix notation would be more complete if were replaced by ) and convention here is , And now our asymptotic equality be Wl.itten ( pcv ) ] for , as may be readily proved , When the variable is positive with respect to the note the the absolute term in this asyml ) totic equality is equal to may be written , which is the expression which has been proved equal to ) by the process of differentiatio under of contour integration .
S 50 .
But if we take the expression for which has been obtained in ) and write it we may obtain this expression independently .
DOUBLE GAMMA FUNCTION .
is so evident as to determine the nomenclature .
Note in the second place that the ) asymptotic expansion ( A ) S 38 may , with the symbolic notation subsequently introduced , be written S 51 .
If in this equality , truc for all values of and , we make , we obtain , since which is equivalent to the result rentiating the ) totic expansion of S 49 .
If now we make , we obtain We now see that which is a quantity independent of and ( 1 .
E. W. BARNES ON THEORy OF THE But when , the expression on the left-hand side may be written ince the values of the logarithms with respect to the axis in all cases to taken , we see from S 23 that this expression is equal to denoting by a capital letter the logarithm which has its principal value with respect to the axis of \mdash ; 1 , we have , And therefore We notice that this formula agrees with our previous results .
For by S 47 and by S42 this last expression It ticing that incidentally has been obtained as alimit in the 1nore general form [ 0VElt ] DOUBLE FUNCTION .
where has its principal value with respect to the axis S52 .
Let us next put iu the final equality of S 50 .
We shall obtain , in the limit when is finite , F. , the logarithms having their principal values with espect to the axis .
Thus ) Now the left-hand side of this relation is independent of and we see , by S22 , on , that .
This formula again agrees with results which cau be deduced from the integlal formulae .
Incidentally has been expressed as the lole general limit , where has its principal value with respect to the axis of .
S53 .
We can now finish the investigation indicated at the beginning of S50 , and obtain the expression for without integrating under the ) of integration .
We have where ) VOL. CXCVI.\mdash ; A. X the results of SS , we find , wheu is positive with espect to the he vintJs their ) values with } ) to .
NUW the higher terms in expansion all vanish .
Hence which is the ) ression which has been previously obtained in S45 .
S54 .
We next proceed to consider the values of the double Riemann function , , when is a negative integer .
By the definition of S39 , usitive with respect to the Now in S15 it has been proved that . . . . . .
the latter integral being taken along a small circle surrounding the origin .
[ Since is integel , the two line integrals in the standard reduction of S46 destroy one another .
] FUNCTION .
'339 Hence if , when is.q positive inteo.erb greater than or equal to zero , Thus , when is a negative integer , But by S12 corollary , , and therefore finally , when is a negative integer , .
When is not positive with respect to the 's this formula continues to hold , as is immediately evident by the theory of the function , to which we shall shortly proceed .
S 55 .
We proceed ext to find the values of for positive integral ( including zero ) values of We have seen , in S48 , that when is positive with respect to the , so that .
Differentiate with regard to , and we find .
Now by S 39 , when is a small quantity , so that , neglecting powers of ) the first , But we immediately deduce from S 53 that .
W. ON OF .
Again differentiating , we have .
Also , if be small , we have and Thus ' so that , , we have .
Tabulating our results , we see that , when ; ; ; ; These formulae hold for all values of the yariable , though we have only established them for the case when is positive with respect to the .
They evidently with the results established for the case in " " The Theory of the G Function S 34 .
S 56 .
In a note appended to the " " Theory of the Gamma Function it was stated that a theory of the simple Riemann function had been developed by MELLIN .
* It * MELLIN , ' Acta Societatis ' vol. 24 , No. 10 , 1899 .
IA FUNCTION .
.341 .
was , howeyer , published after my paper had been sent to press , and I was therefore ignorant of his elegant results .
Expressed in the notation which I adopted , his method is as follows .
It is evident from the expression of the function as an asymptotic limit , its importance lies in the fact that it is a solution of the difference equation ' when has any value , real or complex .
[ From this result we see at once that we should expect that , when is a positive integer , the simple function should be substantially a derivative of the gamnla function , and when is a negative integer , Bernoullian function .
] Now when , the simplest solution of our difference equation is evidently This solution becomes when } , but MELLTN has succeeded in a modified solution by the ingenious modification of process .
We construct the function when , by writing in place in the simple Bernonllian function . . . . .
, taking the sum of the first terms , being of course a positive Thus ) .
And now is defined by the relation We readily see that this function formally satisfies our fundaJnental difference equation , and we may at once prove that the series does define a tion e over the whole plane .
For when , . . .
, it is evident that constant , and therefore When is large and has any value , the left-hand side of this relation may be expanded in the form . . .
, where the 's are integral polynomials of of degree indicated by their suffices .
As they vanish for values of , we must have Thus is convergent with and is therefore convergent , ' provided , or which is the condition with which we started .
Tbere is one point which does not arise in the work of MELLIN , who takes the case cv .
It is that throughout we must work with many-valued functions with as index , which have their principal values with respect to the axis of .
For in expanding , where is large , we have tacitly assumed that , which , unless co is real , is not the case when is large and nearly real and negative , so that and lie on opposite sides of the axis of \mdash ; 1 , if this axis is the axis of the logarithms .
S 57 .
It is now possible to construct the double Riemann function by extending the previous analysis .
The function so constructed might be made fundamental in the theory of double gamma functions and double Bernoullian numbers , these functions arising for particular values of the variable .
We will indicate the development of the theory from this point of view , for breyity establishing only the results , or those which , as in SS 54 , 55 , have been established only over part of the plane .
The double Riemanm function is the simplest solution of the differeItce equation and having any complex values such that is not real and negative .
The determination of will appear in the course of the investigation .
In the first place , it is evident that when , a solution is given by the series which will then , by EISENSTEIN'S theorem , be convergent .
DOUBLE GAMMA FUNCTION .
When being some positive integer , we form a modified solution by the introduction of the function , fonmed follows .
We take the th double Bernoullian function , , write in place of , nd theu take the sum of the first terms .
We thus have and then is given by - .
, au expression which tness we shall sometimes write .
It is at once evident the function so defined formally satisfies the fundamental diflerence equation , we may readily prove that the series is in general convergent .
For when , . . .
, , obviously where and are constants .
And therefore yauishes when , . . . .
When is large this expression admits of expansion in the form . . .
where , , . . .
integral polynomials in of deglee indicated by their suffixes , provided that the rithms which interveue in functions with index such that , when is com with .
If the axes of and include the axis of \mdash ; 1 , this will not be the case for terms of the double series , for which the numbers and , in the are large , unless the logarithms have their .
value with respect to some line between the axes of and .
We take this line to be : vanishes values of , we see that its expansion is large must be . . .
The series is therefore convergent with It is then convergent when or } being finite .
We have then obtained a solution of the difference equation lvhele has its principal value with respect to the axis , which is valid all values of and S 58 .
The identity of the function just defined with that previously employed is easily seen .
From the mode of formation of it is evident that when is positive with respect to the , we have sum of the first terms in the expansion in powers of of itnd therefore when is large we have the symptotic expansion a which may t ) to be rue for all values of by a term-by-term ulsio of the series for ) Now the expression for given in S57 , we see on taking the first values of and the qn first values of , that AMMA FUNCTION .
S45 Putting , and employin the asymptotic expansion just obtained for { obtain the same i } uality as that by which it has previously been defined in S 59 .
As example of the ) in which we should proceed in theory based on the double gamma function defined in , we will prove the relation established in for the case when is positive with respect to the In the first place , when and is small , we see that higher powers of Therefore taking as the limit , when is infinite , of the sulll obtained by taking the first terms of the double series the logarithms having their principal values with respect to the axis Putting , we find And now , in the limit where is infinite , we find VOL. CXCVI . .
2 .
E. W. BARNES ON THEORY OF where the complete form of the result established in S 55 for the case where is positive with respect to the We may establish the other results of that paragraph in a similar manner .
S 60 .
will briefly consider the reduction of the second form of the double function to the double gamma function in the case when If we put , where is very small , we obtain , , . . . . . . . .
higher powers of And therefore , by the second definition of the double function . . . . .
J ) . . . . . .
j ) where Th and [ 0VEB ] DOUBLE GAMMA FUNCTION .
in the limit when is made infinite .
Substituting the value in the various terms , .
the logarithms , and re-arranging the result in powers of , we find , with the symbolic notation of 49 , in the limit when , is made infinite .
Since the left-hand side of this equality is by the definition of fiuite unless , is at one of the points , 1 , 2 , . . .
, , 1 , 2 , . . .
, we see that we have thus been led in a purely algebraical manner to the tion of the dominant terms of the fundanlental expansion of If we make , and remember the definition of .
:\mdash ; we arrive at the dominallt of the extension of STIRLING'S theorem to two parameters .
lf we utilise this result in conjunction with the one just obtained , we find and by the definition of the double gamma function the expression last written reduces to 2 .
Thns We utilise this lnula to ] the fundamental difference equations for the double function .
By the definition of the double function of S 57 Therefore , in the limit when ?
is infinite , On reduction we obtain latter ) ressio is , by the expansion obtained in the " " Theory of the Gamma Function S30 , equal to [ The term } which al.ises is absorbed by the identities which change into } .
The prescription of the absolute logal.ithms has been throughout left indeterminate .
] We thus have one of the formulae of S ) Sufficient indication has perhaps now been given of the alternative development of the theory of the double function .
DOUBLE GAMMA Note .
expansions of this part were ) ined in my original cast of this theory , to which reference ] already been made in the note which follows the " " Theory of the Gamma Function by the assumption that thevu would ] merely powers and coupled with inductive processes .
Such method , though long , is , could the fundamental assumption be justified , probably the most elementary way of obtaining these results .
[ Additional Notc added , 1900.\mdash ; Dr. HOBSON pointed out , if we admit the validity of the application of the calculus of operators to parametel in the subject of integration contour , the from onwards may be developed in a very elegant manner .
We take the formula the known theorem , ' .
and deduce PART IV .
ltiplicatio , Integral } the Double Garnma Function .
S61 .
After the developments of Part IIT .
, we now return to the ) theory of the double gamma function .
} the and tion theories , two distinct courses open to .
We eithel proceed entirely algebraically , making use of been established , leduce the required esults without the intervention of , or may directly utilise these lattel ' in question .
formel ' course is , on grounds , } ) referable we ) ) deduce by processes .
But it is ) the fatal )jectiolt ( to very lengthy algebra .
We ) the two llethods , side by .deduce the multiplication formulae , be obselwed that the ) more elegant and more speedy .
For the sake brevity , results ) In the first sketch of ) } results were tained in this theory obtained solely by this course .
Inasmuch as the function citn be expressed in terms of the function , the multiplication theory can be deduced from the theory of tion .
As the work of obtaining the new expression is in every case almost equal to that of obtaining the results ab initio , we adopt the latter course .
Multiplication Theor ?
S 62 .
We have the definition ( S 19 ) and therefore ' , the pal.ameters bein understood to ) and not explicitly Wl.itten .
with respect to and we obtain , lvhele is constant with respect to Now Substitute this relation in the identity ( 1 ) , take the same number of telms involving on both sides of the equality , and remember that \mdash ; is always to be regarded as a single entity .
We find that , in the limit when is infinite , we have where Now we have seen that ( S22 ) , in the limit when is infinite , DOUBLE UNCTION .
the principal values of the logarithms being taken .
Therefore so that we have again with respect to and we find , where is constant with respect we have , in the limit wheJl 7 is infinite , Hence we find , on equating the irresoluble terms involving in the same way on both sides of the identity ( ii .
) , that , in the limit when is infinite , ' and , therefore , when is infinite , Now in S 23 we have seen that { Jog } , and , therefore , after some reduction , we see that .352 MR. .
W. ON OF We thus have ' Integrate with respect to determine the consta , nt by making and remember that : we obtain the formula the principal value of the logarithm taken . .
By means of the extension of STIRLING'S theorem to two parameters ( S50 ) , it is possible to express the form } which has thus arisen in terms of the double function For , in the limit when 21 is infinite , , under the condition , we obtain the result of the previous paragraph Make now , and we find where denotes that in the product the term for simulis to be excluded .
DOUBLE GAMMA FUNCTION .
351 ' } We have then , in the limit when is infinite , \ldquo ; Utilise now the extension of theorem the formulae and .
We find that F. , Fe where the logarithms have their principal values with respect to the axis of .
The labour of reducing such an expression is evidently very great .
It is diminished by observing that the result must be independent of , so that we may neglect all terms which involve this letter ; but even then it ; only aftel several steps that prove that the right-hand side is equal to VOL. CXCVI . .
2 MR. .
W. ON OF We th H see that .
In the case when the preceding result has an especial importance .
Tt is coYtvenient to , in with this hese f evidently respontl t the functions ' in ' theory of elliptic functionH .
argument , we take two similar equations , so that nleters .
being omitted .
And no the result of the preceding ) aragraph , , We the double fnlction of and .
in terms of the product of double functions whose al.gument is a half quasi-period .
We ] ) reviously seen in the theory of the simple gamma function that , the formla just obtained is the natural extension of this result .
Fron ] the results of SS 62 and 63 we see that we may express the multiplicafornnla for in the form DOUBLE GAbIMA We now proceed to this result from expression as a contour integral .
We have seen ( S45 ) , when is positive with espect to the and , therefore , being a positive integer , But . . .
I and ( S 14 ) Therefore when is positive with respect to the ?
Since is a positive integer , the defined reference the neters is the same as that defined rence to the } ) for the lines representing these two sets of rameters coincident .
Tf th ) into ?
\ldquo ; the last becomes one is the arithmetic value of log being taken .
.356 MR. I W. BARNES ON THEORY OF THE And therefore ( ma ) ' which is the result required .
result of course only been proved by means of the contour integral , under the assumption that is positive with respect to the .
To establish it in general we should appeal to the principle of continuity .
G6 .
Before the multiplication theory , we deduce )ressions for the values of , 2 , 3 , . . .
We recall 2 that within circle of ciently small radius sulrounding the , we . . .
Again , the nnultiplication ) of the preceding paragraph , we have , by a . .
Combine these two theorems , and equate coetiicients of various powers of in the resulting identity .
We find , when ?
is reater than 2 , There exists no such theory for the present functions\mdash ; at , no theory having the simplicity and elegance which is characteristic of the functions , and the reason is obvious\mdash ; the change of into makes no difference of form in such a series as ' but it makes a change of comparatively great complexity in such series as The fol.luer series is the basis of those occurring the theory of elliptic functions , .
the latter of those occurring in double gamma functions .
We shall then limit our consideration to transformations which result from the change of and into and respectively , and being positive By definition we have where .
Hence ) , it being understood that the parameters when not expressed are always lIld On integrating successively three times with respect to , we shall find where is an algebraical polynomial in of order 2 .
As has been stated , it is possible to obtain by purely algebraical processes , use being made of limit theorems p- .
reviously established .
We may , however , obtain its value as follows:\mdash ; In the which has been obtained , cha1lge into and subtract the first esult from the one so formed .
We find by S 23 , , where has the asigned in S 22 , where , unless does does not lie between \mdash ; 1 and - , in which case as I is positive or negatiye Now Theory of the Gamma Function S 7 ) ' by S 18 of the same papel .
We therefore have .
Similarly , where ?
has the tssiglled in S 21 , and ) from it in that must in the ition be substituted for Now we have seen that ( S 22 ) \mdash ; , the upper or lower sign being taken as tive o positive ; since and are integers , the same is true of Thus ( say ) , and vow satisfies the two diffel.ence relations ; fhIA FUNCTION .
and , , since is dratic o in .
have If we determine the constant by making , we have finally .
From the values of ? ?
, it is readily seen that , unless the axes to and inclnde the , in case if if i , s and .
The constant which into the tion f of the paragraph can be expressed in terms of ) this pose we consider the contour hich resents the double function .
Since is a positive And therefore if the in tegral , its ) the logarithm which occurs in the subject of integration , be as in IIT .
, I for the bisector of the allgle between the of ) is the same line the bisector of the angle between the axes .
We have by S45 , when is positive with efil ) the MR. .
W. BARNES ON THE THEORy OF THE Now by S , unless the axes of and include the axis of \mdash ; 1 , in which case as is positive or negative .
Therefore , unless the axes of and include the axis of \mdash ; 1 , in which case , as is positive or negative .
, by S 14 , We therefole have But from the values which have just been given , it is clear that We thus have , when is positive with respect to the This result agrees with that of S67 , and on comparison of the absolute term we see that DOUBLE GAMMA FUNCTION .
This may be regarded as the transformation formula for the double Stirling function .
Notice that when , and therefore the preceding formula becomes .
Comparing this with the result obtained in S 63 , we find This result may also be obtained by the transformation of the line integral which expresses .
S 69 .
We have still to consider the transformation of the first and second double gamma modular iorms and .
With this object we write symbolically and we write , where ; so that , is a form analogous to the modular forms introduced into the theory of elliptic functions by ABEL .
By S 29 we know that within a circle of sufficiently small ladius there exists the expansion . . .
Take now the formula of ' and expand in powers of .
CXCVI.\mdash ; A. 3 @o eInUIJOf p S$Im I UO pug sqns p J@oI u.lq S q Iqnop 9 pooJd S a uo1$ H$ sassaJdxa ?
Bu1ure@ S .364 MR. .
W. BARNES ON THE THEORY OF THE .
There are two alternative methods of obtaining this formula , or rather of obtaining the fundamental relation ( 1 ) .
Firstly , we may directly transform the series which expresses .
We have , from the definition , Put , a change which is equivalent to grouping together terms corresponding to points on the cross lines of the figure .
There we have Now we have seen in S 28 that We therefore have , which is equivalent to the former relation .
Secondly , we may make use of the contour-integral expression in the following manner .
We have ( ' ' Theory of the Gamma Function S 30 ) , when is positive with respect to DOUBLE FUNCTION .
Therefore , under the same limitation , Now , since ) ' it is evident that is homogeneous of degree\mdash ; 2 in and Therefore , by EULER'S theorem , , so that .
On integrating twice with respect to , we obtain , where is a linear function of Changing into , and subtracting , we see that satisfies the difference equation characteristic of , so that it can only differ from this function by a constant , which will vanish , as we see by making We therefore obtain the relation required .
S 72 .
We proceed now to the analogues of RAABE'S formula .
This formula be written Theory of the Gamma Function .68 ) We will evaluate and The method which will be employed is the same as that by lvhich formula itself was originally ined ; it was , in fact , first invented the proof of the present theorem .
Let Then , and , therefore , on integration , S 73 .
We proceed to evaluate the constant term by an application of the multiplication theory the case We have and therefore , by S65 , Put , and successively in the formula which we have just obtained for and substitute in the formula just written .
We obtain .
By means of the formula DOUBLE GAMMA FUNCTION .
we see that and this expression in turn , by the multiplication formulae when ? ?
for the simple and double gamma functio1oe and the simple Bernoullian function , is equal to .
Now , on making , we obtain from the multiplication formula Therefore the expression which we have just found reduces to And therefore MR. E. W. BARNES ON THE THEORy OF THE since the coefficient of is equal to We have , then , finally , .
We thus see that substantially the double Stirling function of and .
is expressed by , or , by symmetry , by .
We have , in fact , the relation where A is the Glaisher-Kinkelin onstant Theory of the G Function S3 ) ; and therefore for , by S22 , S74 .
We may readily prove these results by the relation which exists between and .
[ The latter is a convenient way of writing For , when each of the parameters is equal to , we have Theory of the G Function S 29 ) and therefore Now from the multiplication theorem for the double gamma functions , when we have Hence .
a formula which is equivalent to the result of S 73 .
S76 .
By combining the results of SS 72 and 73 , we may now write down the value of ; 1 For we have seen in S 72 that this integral is equal to which expression in turn is equal to .
ThusVOL .
cxev Since , by S 64 , we see that this formula may equally be written This and the corresponding formula , ed by the interchange of and and , are the analogues for double gamma functions of RAABE'S formula for simple gamma functions . .
In the particular case when is positive with respect to the , it is possible to obtain more simply the value of by means of the contour integrals investigated in Part III .
We give this method of proof as it leads incidentally to au expression as a contour integral for ) ) .
We find , on integrating the expression for , giyen in S45 , and the right-hand side , by an application of the formulae of SS 6 and 44 , becomes Now , reducing the contour to a small circle round the origin , we see that and therefore DOUBLE FUNCTION .
Since ' we see that , to establish the formula of , it is necessary to show that This may be readily done as follows:\mdash ; We have Integrate with respect to between and , and we find for it may be readily seen that [ This vanishing contour integral occurs when RAABE'S formula is proved by means of the expression of the simple gamma function as a contour integral .
] But as we have deduced in S 75 , from theorem , The contour integral has , therefore , the required value .
We here conclude our investigations of the algebra of the double gamma functions , It is evident that the formulae admit of still further development ; they lead , for instance , to many curious relations between the rals obtained in Part III .
Such considerations are , however , foreign to our immediate purpose .
The development of the integral formulae and the theories of multiplication and transformation in the case of the double Riemann function is interesting in that we thus combine many of the formulae which have been obtained separately for Bslnoulliall and gamma functions , and the algebra by which such developments are obtained by the extension of MELLIN'S definition of the simple function is in many ways attractiva Owing , however , to the length of this paper , we do not propose to consider it in this place .
DOUBLE GAMMA FUNCTION .
exists , and then it will be shown that , the possibility of such an expansion being established , its actual form is readily obtained by a process of difference-integration .
Subsequently we shall verify the results by an alternative proof by means of contour integration , this proof being the natural extension to double gamma functions of the one employed for functions of a single parameter Theory of the Gamma Function Part IV It should be noticed that the expansion under consideration differs materially from the expansions obtained in Part III .
In those expansions the limits of the number of terms of the series and products , the quantities and ormed the infinite basis terms ; but in the present case that basis is the variable itself .
We first write down the asymptotic expansion for .
We have obtained Theory of the G Function S 15 ) the asymptotic expansion this expansion being valid for all values of and such that is large , and the principal value of being taken .
There is , in the language previously employed , a barrier-line along the axis of \mdash ; 1 .
Now in S 70 it has been seen that and Therefore where now there is a barrier-line along the axis But by S 5 so that we have finally a barrier-line along the axis When , we see that we have asymptotic expansion S80 .
We will now prove that , provided be positive with respect to the , there exists an asymptotic expansion for of the form where denotes a quadratic function of In the first place it is evident that we have where is some finite quantity , and and are singly or together large positive integers .
Now , from the fundamental difference equations of the double gamma function , it is at once seen that .
A term has been neglected which is an integral multiple of , and which is therefore absorbed by a suitable specification of the logarithms involved .
The above formula may be rewritten In the first place , if we put in place of and in place of in the procedure DOUBLE GAMMA FUNCTION .
of the paragraphs leading up to S 49 , we evidently obtain for an expansion in powers of and ' every term of which involves a ebraically , and of which the non-ultimately-vanishing terms are typified by In the second place , consider .
We have seen , in S 30 , that the product being absolutely convergent .
The typical term may be written . .
Therefore ( ' ' Genesis of the Double Gamma Function SS4 and 5 ) admits an asymptotic expansion of the form each term of which involves algebraically .
Combining these results we see that admits an asymptotic expansion powers of and ) , each term of which involves algebraically , and of which the terms which do not ultimately vanish are ] ) ified by But is a function of Tt must then be capable of an asymptotic expansion in powers of \mdash ; , each term of which involves algebraically , and of which the terms which do not ultiInately are typified by And now by mere re-arrangement we may include with the term each time that the latter occurs , and we obtain for , where ) an asymptotic expansion of the form S 81 .
We can readily extend the previous proof to the case where lies between the axes and , so that it is given by By writing the fundamental difference equation in the form we readily see that But by the theorems just quoted in the previous paragraph the three expressions in the square brackets severally admit of asymptotic expansions in powers of and , whose terms which do not ultimately vanish are typified by and whose coefficients all involve algebraically .
Thus , by a repetition of the previous argument , ) admits when , an asymptotic expansion of the form In an exactly similar manner we may show that will admit of an asymptotic expansion of the same lorm when is large , lying between the axes and DOUBLE GAMMA FUNCTION .
S82 .
But when we come to the case of the negative drant given by it is interesting to notice that the ) proof breaks down .
As before , by the use of the ence equation , ) obtain relation The several expressions bracketed on the right-hand side of this identity tWlnit of asymptotic verse of and , of which the terlns volve c algebraically ; and therefore the whole of the right-hand side admits of nn of this fornn .
But there the terms and when we seek to ) with forced back 011 ] function ) .
Thus as the possibility is negative with regard to the , our results are , as we should S 78 , entirely negative .
The region between the axes and ) is ) the asymptotic expansion of the double gamma function .
) this region closes up into the barrier-line which occurs for the functions . .
We can find syInptotic of for values of which are such that does not lie in the ) ) any complex quantity of finite modulus .
For such values of and we have the expansion um cxcv1 .
W. BARNES ON THE THEORy OF THE where f and are algebraical polynomials of degree indicated by their suffixes , is , so long as is finite , likewise algebraical function .
by the mental difference equation .
Again ( ' ' Theory of the Gamma Function S 41 ) , we have the asymptotic expansion where .
is that natural logarithm of which has its principal value with respect to the axis It is thus equal to the latter ithm having its principal value with respect to the axis .
We have then , if its principal value with respect to the axis of Tf we corresponding powers of on both sides of this result , we find nnd similar tions alnong the 's and We shall get , in like manner , another set of relations in which and are .
Relnembering that the , and 's are all algebraical polynomials which vanish with , we thus prove that By this process , which may } ) ) riately be called a process of finite integration , obtain the ) DOUBLE FUNCTION .
having its principal value with respect to the axis S 84 .
We may now obtain the asymptotic expansion for , under the limitations assigned at the commencement of the preceding paragraph .
For this purpose integrate the relation just obtained with respect to between the limits and Then , by the formulae of SS 12 and 76 , we find Substitute now the asmtotic expansions for and of which the former has been quoted in the preceding paragraph , and the latter obtained in S 79 .
Then we find ?
Remember that ; then we obtain by an easy reduction , the asymptotic expansion W. BARNES ON THE which is the complete asymptotic expansion for when is large , and does lic the barrier region negative with respect to the axes and If we colnbine this result with that obtained in S 83 we find the more general ' ' the assigned limitations .
expansion is written in the precise , in order that the analogy corresponding formula in the theory of multiple gamma functions may be more ayed .
S 85 .
We might now conclude this investigation .
Since , however , this would appear to be the first time in analysis in which an asymptotic expansion with a barrier obtained , it seems better to give an alternative proof which shall not need the diflicult argument of SS 80-82 .
This proof is the direct extension of that revionsly given for the case of the simple gamma function .
* We therefore proceed as iefly as possible .
In the investigation of S 57 it was shown that when is finite and where positive integer , the series for is absolutely convergent .
Suppose now that , where , and suppose further that does not lie within the region bounded by axes to and , and that is positive with respect to the Then , since it evident that , if be any positive integer , the absolute value of each term of the ssion tellds to as to infinity .
For , by the restrictions on and , where therefore ( Theory of the G$mmn Function Part IV .
I regret to say that the Lemma of S40 is faulty ; the theorcm is cvidently only true when is real .
A slight modification will , however , establish the truth of the under the condilions enunciated .
DOUBLE GAMMA FUNCTION .
where is finite , be .
Hence the absolute value of tends to zero as tends to infinity ; and this theorem is true if is ) lacedb where are positive integers .
Hence the absolute value of X ( where is the function introduced for brevity in S 57 ) tends to zero tends to infinity .
But and therefore .
Now the double series on the right-hand side is absolutely convergent for } finite values of , and the absolute value of each term tends to as tends to infinity .
Therefore remains finite as tends to infinity , being finite and not greater than '2 .
When is greater than 2 , we have and therefore the expansion on the left-hand side is finite however large be , provided .
Consider now the integral The subject of integration is a uniform function of , wherein is to have its principal value with respect to the axis of , is to lie within the region bounded by axes to and is to be positive with respect to the DOUBL GAMMA FUNCTION .
Let now the perpendicular contour be distorted into a contour which encloses the points 2 , 1 , , .
, and which after the point goes off to infinity perpendicularly to the real axis .
So far as the value of the integral is concerned this contour will differ from the second contour only by two strips at infinity of length less than parallel to the real axis : and by the previous paragraph the integral these strips will vanish .
By CAUCHY'S theorem the value of the integral along this third contour will be equal to minus the sum of its residues at the points 2 , 1 , , . . .
, together with the integral along the perpendicular line cutting the axis of between the points and .
Now , when or 2 , the residue of the integral is equal to the coefficient of in by S , and is therefore equal to where the logarithm has its principal value with respect to the axes When , the residue is the of in by S and is therefore equal to When ] lesidue is We therefore have \mdash ; . . .
W. ON THE OF THE where the quantity]og its principal value with respect to the axis ( is fundamental taken along a perpendicular cutting the axis of between and .
It is evident that the is large is of ordor of magnitude less than We iore have tl ] ) totic expansion , when is large , residue :fter n terms of the have been taken is of the same order nitude its th final term taken .
expansion } ; evidently the same as that previously obtained .
The limitation that must ) positive with respect to the 's may evidently be removed by the fundamental difference relations for the double gamma function and totic C lsion .
We are finally left with the essential that not lie within the region bounded by the axis to ?
-transcendental . .
We finally the theorem that the double gamma function cannot arise whose coefficients are not generated from the nctio itself difying sligbtly the nomenclature introduced by MOORE , we is transcendentally-transcendental function .
The proof is a slight given the fnnctio1t ( S 30 ) , which in turn was similar V. of the " " of the Function In it be roved exactly before that if the theorem is true for it is true for ) .
We shall therefore cufine ourselves to the considera{ion the ctio .
By fl equations , we have where , nicnce , ) .
' vol. 48 , pp. 49 .
MOORE uses tho term only to describc functions which tion coc GAMMA Suppose that satisfies the differential equation so transformed that it is rational and integral in and its derivatives .
Let the terms of class be symbolically , , .
; in terms of class being , . . . . .
, , and the functions , being holomorphic .
If satisfies the differential equation , will satisfy the equation in which is written for and the equation in which is written for Make the first substitution , divide the equations by and respectively , and subtract one from the other .
We find terms of lower class But consists solely of terms of lower class than Hence either the equation which has been obtained vanishes identically , or we can reduce the equation for .
to one in which there are fewer terms of class The equation cannot vanish identically unless the coefficients of the various terms of class all vanish , which necessitates that the ratios ' . . .
' are doubly periodic functions of of periods and The equation for can thus be always reduced to one of the form . . .
terms of lower class where all the coefficients are holomorphic functions , and , in addition , the functions are doubly periodic of periods and VOL. .38 .
BARNES ON THE THEORy OF THE Divide the equation by and subtract it from the equation which results from changing into .
We obtain . . . . . . .
terms of lower class This equation will not vanish identically unless the functions . . .
all satisfy relations of the form therefore by symmetry relations of the form The quantity which gives the number of terms on the right-hand side these two relations will not vanish unless the original equation can be reduced to the form . . . . . .
terms of lower class . . . . . . . .
( 1 ) , where the coefficients are holomorphic functions of and the 's are doubly periodic functions of periods and Either then the original equation can be reduced to this form , or at least one of the composed of an additive number of solutions of difference ' . .
equations of the form But the most general solution of such a pair of equations is DOUBLE GAMMA FUNCTION .
where and are doubly periodic functions of of periods and .
And therefore one of the ratios must be afunction generated from the function The original equation therefore either contains the double gamma function implicitly among its coefficients , or it is reducible to the form ( 1 ) .
Continue our former procedure , and we see that either at least one of the ratios is composed of an additive number of equations of the type Q. and is therefore generated from the double gamma function , or the ol'iginal equation is reducible to one in which the atios of terms of the highest classes are doubly periodic functions of of periods and The successive repetitions of the argument are now evident .
Ultimately we reduce the equation to one in which either all the coefficients are doubly periodic functions ( which is absurd ) , or to one in which the last term is generated from the double gamma function .
Thus the proposition is established .
The double gamma function cannot satisfy a diflerential equation in which the coefficients are finite combinations of , e.g. , ( 1 ) Rational or irrational algebraic functions of ( 2 ) Simply or doubly periodic functions , ( 3 ) Simple gamma functions , ( 4 ) functions , ( 5 ) Theta functions , or , in fact , of any functions which are not substantially educible to or compounded of the double gamma function itself

