You are on page 1of 815

G.N.

WATSON

G.N.WATSON

PRESS

A TREATISE ON

THE THEORY OF BESSEL FUNCTIONS


G. N. Watson
The late Professor G. N. Watson wrote his monumental treatise on the theory of Bessel functions with two objects in view. The first
was the development of applications of the fundamental processes of the theory of complex variables, and the second the compilation of a collection of results of value to mathematicians and physicists who encounter Bessel functions in the course of
their researches.

The completeness of the theoretical account, combined with the wide scope of the
of practical examples and the extensive numerical tables, have resulted in a book which is indispensable to pure mathematicians, to applied mathematicians, and to physicists alike.
collection

'In Professor Watson's treatise, which is a of erudition and its often too rare accompaniment, clear exposition, we have a rigorous mathematical treatment of all types of Bessel functions, their properties,

monument

representations, asymptotic expansions, integrals containing them, allied functions, series, zeros, tabulation, together
integral

with extensive numerical tables.' L. M. Milne-Thomson in Nature

'A veritable mine of information.


pensable to all those use Bessel functions.'
S.

.indis-

who have

occasion to

Chandrasekhar in TheAstrophysical Journal

Also available as a paperback

THEORY OF
BESSEL FUNCTIONS

W.

B. F.

A TREATISE ON THE

THEORY OF
BESSEL FUNCTIONS
BY
G. N.

WATSON

SECOND EDITION

CAMBRIDGE
AT THE UNIVERSITY PRESS
1966

PUBLISHED BY THE SYNDICS OF THE CAMBRIDGE UNIVERSITY PRESS


Bentley House, 200 Euston Road, London, N.W. 1 American Branch: 32 East 57th Street, New York, N.Y. 10022 West African Office: P.M.B. 5181, Ibadan, Nigeria

First Edition Second Edition Reprinted

1922 1944 1952 1958 1962 1966

First paperback
edition

1966

~--fcU^6\\*s

First printed in Great Britain at the University Press, Cambridge

Hazell Watson

Reprinted by lithography in Great Britain by & Viney Ltd, Aylesbury, Bucks

PREFACE
THIS
book has been designed with two objects in view. The first is the development of applications of the fundamental processes of the theory of
functions of complex variables.

For this purpose Bessel functions are admirably adapted; while they offer at the same time a rather wider scope for the application of parts of the theory of functions of a real variable than
is

provided by

trigonometrical functions in the theory of Fourier series.

The second object is the compilation of a collection of results which would be of value to the increasing number of Mathematicians and Physicists who
encounter Bessel functions in the course of their researches. The existence of such a collection seems to be demanded by the greater abstruseness of properties
of Bessel functions (especially of functions of large order) which have been

required in recent years in various problems of Mathematical Physics.

While my endeavour has been to give an account of the theory of Bessel functions which a Pure Mathematician would regard as fairly complete, I have
consequently also endeavoured to include
special, which,
all

formulae, whether general or

although without theoretical interest, are likely to be required in practical applications; and such results are given, so. far as possible, in a form appropriate for these purposes. The breadth of these aims, combined
with the necessity for keeping the size of the book within bounds, has made it necessary to be as concise as is compatible with intelligibility.
is, for the most part, a development of the theory of funcexpounded in the Course of Modern Analysis by Professor Whittaker and myself, it has been convenient to regard that treatise as a standard work of reference for general theorems, rather than to refer the reader to original

Since the book

tions as

sources.

draw attention here to the function which I have regarded namelt the function which was defined by Weber and used subsequently by Schlafli, by Graf and Gubler and by Nielsen. For historical and sentimental reasons it would have been pleasing to have felt justified in using Hankel's function of the second kind; but three
It is desirable to

as the canonical function of the second kind,

considerations prevented this.

The

first is

the necessity for standardizing the

function of the second kind; and, in

my

opinion, the authority of the group

of mathematicians who use Weber's function has greater weight than the authority of the mathematicians who use any other one function of the second kind. The second is the parallelism which the use of Weber's function exhibits

between the two kinds of Bessel functions and the two kinds (cosine and

sine)

VI
of trigonometrical functions.
interpolation

PREFACE

The third is the existence of the device by which made is possible in Tables I and III at the end of Chapter XX, which seems to make the use of Weber's function inevitable in numerical work.

It has been my policy to give, in connexion with each section, references any memoirs or treatises in which the results of the section have been previously enunciated; but it is not to be inferred that proofs given in this book are necessarily those given in any of the sources cited. The bibliography at the end of the book has been made as complete as possible, though doubtless omissions will be found in it. While I do not profess to have inserted every memoir in which Bessel functions are mentioned, I have not consciously omitted any memoir containing an original contribution, however slight to the theory to

of the functions; with regard to the related topic of Riccati's equation, I have

been eclectic to the extent of inserting only those memoirs which seemed to be relevant to the general scheme.
In the case of an analytical treatise such as this, it is probably useless to hope that no mistakes, clerical or other, have remained undetected; but the
of such mistakes has been considerably diminished by the criticisms and the vigilance of my colleagues Mr C. T. Preece and Mr T. A. Lumsden, whose labours to remove errors and obscurities have been of the greatest value. To these gentlemen and to the staff of the University Press, who have given every assistance, with unfailing patience, in a work of great typographical

number

complexity, I offer

my

grateful thanks.

G. N.
August
21, 1922.

W.

PREFACE TO THE SECOND EDITION


To
incorporate in this work the discoveries of the last twenty years would
necessitate the rewriting of at least Chapters

XII

XIX;

Bessel functions, however, has

waned

since 1922,

and I

prepared to undertake such a task to the detriment In the preparation of this new edition I have therefore limited myself to the correction of minor errors and misprints and to the emendation of a few assertions (such as those about the unproven character of Bourget's hypothesis) which, though they may have been true in 1922, would have been
definitely false

my interest in am consequently not of my other activities.

had they been made in 1941. thanks are due to many friends for their kindness in informing me of errors which they had noticed; in particular, I cannot miss this opportunity of expressing my gratitude to Professor J. R. Wilton for the vigilance which

My

he must have exercised in the compilation of his

list

of corrigenda. G. N.

W.

March

31, 1941.

CONTENTS
CHAP.
I.

PAGE

BESSEL FUNCTIONS BEFORE

1826

II.

THE BESSEL COEFFICIENTS


BESSEL FUNCTIONS

H
38

III.

IV.

DIFFERENTIAL EQUATIONS

85
132

V.

MISCELLANEOUS PROPERTIES OF BESSEL FUNCTIONS


INTEGRAL REPRESENTATIONS OF BESSEL FUNCTIONS
ASYMPTOTIC EXPANSIONS OF BESSEL FUNCTIONS
BESSEL FUNCTIONS OF LARGE ORDER
.

VI. VII.

160 194

VIII.
IX.

....
.

225
271

POLYNOMIALS ASSOCIATED WITH BESSEL FUNCTIONS


FUNCTIONS ASSOCIATED WITH BESSEL FUNCTIONS
ADDITION THEOREMS
DEFINITE INTEGRALS
INFINITE INTEGRALS

X.
XI.
XII.

308
358 373

XIII.

383 450
.

XIV.

MULTIPLE INTEGRALS

XV.
XVI.

THE ZEROS OF BESSEL FUNCTIONS

477

NEUMANN
VARIABLES

SERIES

AND LOMMEL'S FUNCTIONS OF TWO


522
551
.

XVII.

KAPTEYN SERIES
OF FOURIER-BESSEL AND DINI

XVIII. SERIES

576 618
.

XIX.

SCHLOMILCH SERIES

XX.

THE TABULATION OF BESSEL FUNCTIONS


TABLES OF BESSEL FUNCTIONS
BIBLIOGRAPHY

654
665 753

INDEX OF SYMBOLS
LIST OF

AUTHORS QUOTED

....

789
791

GENERAL INDEX

796

To stand upon every

point,

and go over things at

large,
:

and to be curious

in

particulars, belongeth to the first author of the story

but to use brevity,

and avoid much labouring of the work,

is

to be granted to

him that
ii.

will

make an abridgement.
2

Maccabees

30, 31.

CHAPTER

I
1826

BESSEL FUNCTIONS BEFORE


1*1.

Riccati's differential equation.

The theory of Bessel functions is intimately connected with the theory of a certain type of differential equation of the first order, known as Riccati's equation. In fact a Bessel function is usually defined as a particular solution
of a linear differential equation of the second order (known as Bessel's equation)

which

is

derived from Riccati's equation by an elementary transformation.


earliest

The
this

appearance in Analysis of an equation of Riccati's type occurs

in a paper* on curves which was published

by John Bernoulli in 1694. In paper Bernoulli gives, as an example, an equation of this type and states that he has not solved it"f\
In various letters! to Leibniz, written between 1697 and 1704, James Bernoulli refers to the equation, which he gives in the form

dy = yydx + xxdx,
and
states,

more than

1697): "

Vellem porro ex Te

Ego

in mille
lusit."

Thus he writes (Jan. 27, dy yydx + xxdx. meam operam improbum sed Problema performas transmutavi,
once, his inability to solve
scire
it.

num

et hanc tentaveris

Five years later he succeeded in reducing the equation to a linear equation of the second order and wrote to Leibniz (Nov. 15, 1702) " Qua

petuo

occasione recordor aequationes alias memoratae

quam

separare potui indeterminatas a se


:

simpliciter differentialis

sed separavi

illas

qua nuninvicem, sicut aequatio maneret reducendo aequationem ad hanc


in

dy yydx + a?dx

differentio-differentialem||

ddy.y x& dx*."

had been made, it was a simple step to solve the last and so to obtain the solution of the equation of the first order as the quotient of two power-series.
this discovery
series,

When

equation in

* Acta Eruditorum publicata Lipsiae, 1694, pp. 435 437. i f "E8to proposita aequatio differentialis haec x dx + y*dx = a?dy quae an per separationem indeterminatarum construi possit nondum tentavi " (p. 436). t See Leibnizens gesamellte Werki, Dritte Folge (Mathematik), in. (Halle, 1855), pp. 5087. Ibid. p. 65. Bernoulli's procedure was, effectively, to take a new variable u defined by the

formula
1 du u dx

_ ~*
by
y.

in the equation
il

dyldx=x'2 + y 2 and then


,

to replace

The connexion between

this equation

and a

special form of Bessel's equation will be seen

in 4-3.

THEORY OF BESSEL FUNCTIONS


And, in
fact, this

[CHAP.

James Bernoulli within a year

form of the solution was communicated to Leibniz by (Oct. 3, 1703) in the following terms*:

"Reduco autem aequationem dy= yydx+xxdx ad fractionem cujus uterque


terminus per seriem exprimitur,
ita

X
V

3
,

X 3.4.7
7

X 3.4.7.8.11
11

XVa
3.

X19
.

4.7.8.

11

12.

15^3.

4. 7.8. 11. 12. 15. 16. 19

X*

3.4

X*

3.4.7.8

X 12 3.4.7.8.11.12

x 3.4.7.8.11.12.15.16
lti

quae

series

ratio progressionis

quidem actuali divisione in unam non tarn facile patescat, scil.


3

conflari possunt, sed in

qua

_a?
y

x1

3.3. 7

2x11

373. 3. 7. 11

13-r15

3.3.3.3.5. 7777II

+
far

Of

course, at that time,

mathematicians concentrated their energy, so

as differential equations were concerned, on obtaining solutions in finite terms,

and consequently James Bernoulli seems to have received hardly the full credit which his discovery entitled him. Thus, twenty-two years later, the paperf in which Count Riccati first referred to an equation of the type which now bears his name, was followed by a note:,: by Daniel Bernoulli in which it was
to

stated that the solution of the equation

ax dx + uudx = bdu
11

was a hitherto unsolved problem. The note ended with an announcement in an anagram of the solution " Solutio problematis ab 111. Riccato proposito characteribus occultis involuta 24a, 6b, 6c, Sd, 33e, hf, 2g, 4>h, 33i, 61, 21m,
:

26n, 16o, Sp, oq, I7r, 16s, 25t, 32m, 5#, By, +, -,

, =,

4, 2, 1."

The anagram appears never


his solution
||

to

have been solved

but Bernoulli published

of the problem about a year after the publication of the anagram.

The

solution consists of the determination of a set of values of n,

4m/(27/i 1),

where

is

any

integer, for

soluble in finite terms; the details of this

namely any one of which the equation is solution will be given in 4*1, 4*11.

The prominence given

to the

with the fact that Riccati's


*

work of Riccati by Daniel Bernoulli, combined equation was of a slightly more general type than
m.
(Halle, 1855), p. 75.
in

See Leibnizens gesamellte Werke, Dritte Folge (Mathematik),

t Acta Eruditorum, Suppl. equation was

vm.

(1724), pp.

6673.

The form

which Riccati took the

xmdq = du + uu dx:q,

where
other

= xn

% Ibid. pp. 73

75. Daniel Bernoulli mentioned that solutions had been obtained by three members of his family John, Nicholas and the younger Nicholas. The reader should observe that the substitution

u=
gives rise to
|j

bdz
z

dx

an equation which is easily soluble in series. Exercitationes quaedam matheviaticae (Venice, 1724), pp. 77 pp. 465473.

80;

Acta Eruditorum, 1725,

1*2]

BESSEL FUNCTIONS BEFORE 1826

John Bernoulli's equation* has resulted in the name of Riccati being associated
not only with the equation which he discussed without solving, but also with

still

more general type of equation.


Riccati's generalised equation to

It is now customary to give the namef any equation of the form

where P, Q,
It is
if

R are given

functions of x.

P=0,

supposed that neither P nor R is identically zero. If U=0, the equation is linear; the equation is reducible to the linear form by taking 1/y as a new variable.
last

The

equation was studied by Euler J

it is

reducible to the general

linear equation of the second order,


to Bessel's equation

and

this equation is

sometimes reducible

by an elementary transformation

(cf. 3*1, 4*3, 4'31).

Mention should be made here of two memoirs by Euler. In the first it is proved that, when a particular integral y x of Riccati's generalised equation is known, the equation is reducible to a linear equation of the first order by replacing y by y x + 1/u, and so the general solution can be effected by two quadratures. It is also shewn (ibid. p. 59) that, if two particular solutions are known, the equation can be integrated completely by a single quadrature; and this result is also to be found in the second of the two papers. A brief discussion of these theorems will be given in ChaDter iv.
||

1*2.

Daniel Bernoulli s mechanical problem.

In 1738 Daniel Bernoulli published a memoir! containing enunciations of a number of theorems on the oscillations of heavy chains. The eighth ** of
these
is

as follows

"

uniformiter gravis et perfecte

Defigura catenae uniformiter oscillantis. Sit catena AG flexilis suspensa de puncto A, eaque oscillationes
si turn

facere uniformes intelligatur: pervenerit catena in

AMF;

fueritque
ejus

longitudo catenae
valorisff ut
fit
/

= 1:

longitudo cujuscunque partis

FM x, sumatur n
'

n
*

Jl + 4ww

P_

4 9n
.

l^_
4 9 16n4
.

4.9.16. 25ns

See James Bernoulli, Opera Omnia, n. (Geneva, 1744), pp. 1054 1057 it is stated that the is the determination of a solution in finite terms, and a solution which resembles the solution by Daniel Bernoulli is given. t The term Hiccati's equation was used by D'Alembert, Hist, de I' Acad. R. de Set. de Berlin,
;

point of Riccati's problem

'

xix. (1763), [published 1770], p. 242.

X Iiutitutiones Calculi Integralis, n. (Petersburg, 1769), 831, pp. 8889. the reduction, see James Bernoulli's letter to Leibniz already quoted.

||

In connexion with

Novi Coram. Acad. Petrop. vni. (17601761), [published 1763], p. 32. 163164. If " Theoremata de osoillationibus corporum filo flexili connexorum et catenae verticaliter suspensae," Comm. Acad. Set. Imp. Petrop. vi. (1732 3), [published 1738], pp. 108 122.
Ibid. ix. (17621763), [published 1764], pp.

** Loc.

cit: p.

116.
is n.

ft The length of the simple equivalent pendulum

THEORY OF BESSEL FUNCTIONS

[CHAP.

Ponatur porro distantia extremi puncti F ah linea'verticali = 1, dico fore ab eadem linea verticali aequalem distantiam puncti ubicunque assumpti

x n

xx
i

<*

**

**

4>nn

4.9w3

4.9.16n 4

4.9.16.25w8

f ctc

He

goes on to says "Invenitur brevissimo calculo n


littera

= proxime 0691

I....

Habet autem

infinitos valores alios."

The last series is now described as a Bessel function * of order zero and argument 2 V(#/n); and the last quotation states that this function has an infinite number of zeros.
viii,

Bernoulli published f proofs of his theorems soon afterwards; in theorem he obtained the equation of motion by considering the forces acting on

the portion

FM

of length
later

x.

The equation

of motion was also obtained

by

Euler \

many years
following
is

from a consideration of the forces acting on an element

of the chain.

The

the substance of Euler's investigation

Let p be the line density of the chain (supposed uniform) and let T be the tension at height x above the lowest point of the chain in its undisturbed position. The motion being transversal, we obtain the equation bT=gpbx by resolving vertically for an element of chain of length bx. The integral of the equation is Tgpx.

The

horizontal component of the tension


;

zontal) displacement of the element

is, effectively, T{dyjdx) where y and so the equation of motion is

is

the (hori-

'*-( r 2)If

we

substitute for

T and

proceed to the limit, we find that

dt*
If

y dx\ dxj

'

is

the length of the simple equivalent pendulum for any one normal vibration,

we

write

where

A and

f are constants

and then

n (x/f)
d

is

a solution of the equation

A
If

-\ ( dx\ dx)
x/f= ii, we obtain the solution
1

+- =
f
series,

in the

form of Bernoulli's
v?

namely

u ul v= l H
*

1.4

+ u* 1.4.9 1.4.9.16

On

the Continent, the functions are usually called cylinder functions, or, occasionally, func-

tions of Fourier -Bessel, after Heine,

Journal fur Math. lxix. (1868), p. 128; see also Math. Ann.

m.

(1871), pp.

609610.

t Comm. Acad. Petrop. vn. (17345), [published 1740], pp. 162179. X Acta Acad. Petrop. v. pars 1 (Mathematics), (1781), [published 1784], pp. 157177. Euler took the weight of length e of the chain to be E, and he denned g to be the measure of the distance (not twice the distance) fallen by a particle from rest under gravity in a second. Euler's

notation has been followed in the text apart from the significance of g and the introduction of
p

and

5 (for d).

1*3]

BESSEL FUNCTIONS BEFORE

1826

D are constants.
If a is the

where

C and

Since

is finite

when #=0, C must be

zero.

whole length of the chain, y =0 when . = a, and so the equation to determine/* is a% a3 _


1

"

j. +
1

./

TT4/2 ~

4. 9/3

""'

By an extremely ingenious analysis, which will be given fully in Chapter xv, Euler proceeded to shew that the three smallest roots of the equation in a/f are 1-445795, 7*6658 and 18-63. [More accurate values are 1*4457965, 7*6178156 and 18*7217517.]
In
'the

memoir'"' immediately following this investigation Euler obtained the general


-j-

solution (in the form of series) of the equation

(u-r-\+v0, but

his statement of the

law of formation of successive coefficients is rather incomplete. The law of formation had, however, been stated in his Institutiones Calculi Integrality, n. (Petersburg, 1769), 977,
pp. 233-235.
1*3.

Euler 's mechanical problem.


vibrations of a stretched

The
1764.

membrane were
Idz
1

investigated by Euler}: in

He

arrived at the equation

ld?z_d?z
e*

dt2

~ dr* + rdr* r'cty

d*z
3
'

where z

is

the transverse displacement at time


(r,
<f>);

at the point whose polar

coordinates are

and tension of the membrane.

e is

a constant depending on the density and

To obtain a normal
where
a,

solution he wrote

z = u sin (at

+ A) sin (/9< + B),


is

A,

fi,

are constants
is

and u

a function of r; and the result of

substitution of this value of z

the differential equation


/a8

*w,l^w,
The
of Euler's memoir;

p\
at the origin
is

solution of this equation which


it is

is finite

given on

p.

256

?-P

\l
{

2(w+l)e
is
||

+ 2.4(w+l)(n
2/3

+ 3)e4 ~)

where n has been written in place of


This differential equation
of order

+ 1.
as Bessel's equation for functions
0, 1, 2,
is
. .

now known

and

may

have

any of the values

..

Save

for

coefficient of order /S

an omitted constant and argument

factor the series


arje.

now

called a Bessel

The

periods of vibration, 2irja, of a

* Acta Acad. Petrqp. v. pars 1 (Mathematics), (1781), [published 1784], pp. 178190. t See also 935, 936 (p. 187 et seq.) for the solution of an associated equation which will be

discussed in 3*52.

X Novi Comm. Acad. Petrop. x. (1764), [published 1766], pp. 243260. The reason why Euler made this change of notation is not obvious.
||

If

/3

were not an integer, the displacement would not be a one-valued function of position,
{p</>

in view of the factor sin

+ B).

6
circular

THEORY OF BESSEL FUNCTIONS

[CHAP.

membrane of radius a with a fixed boundary* are to be determined from the consideration that u vanishes when r = a.
This investigation by Euler contains the earliest appearance in Analysis of

a Bessel coefficient of general integral order.


1*4.

The researches of Lagrange, Carlini and Laplace.

Only a few year* after Euler had arrived at the general Bessel coefficient on vibrating membranes, the functions reappeared, in an astronomical problem. It was shewn by Lagrangef in 1770 that, in the elliptic motion of a planet about the sun at the focus attracting according to the law of the inverse square, the relations between the radius vector r, the mean anomaly and the eccentric anomaly E, which assume the forms
in his researches

M= Egive rise to the expansions

e sin

E,

= a(l-e cos E),


00

Z A E=M+ n=l
in

j,

sinnM,

-=l + he + 2 B H cosnM, &


2

n=\

which a and
.

are the semi-major axis and the eccentricity of the orbit,


ti

and

_9 ~ "m=o

(-)" M -n-i 6 n+m

2+"*"ro (n
!

w)'!

n
'

^ ~~
n

(_)m H + ( ^ l+2 t

n-t-m-i ro

n+2m e

mt

2'

w!(w + m)!
in

Lagrange gave these expressions


is

for

1, 2, 3.

The

object of the expansions

to obtain expressions for the eccentric

anomaly and the radius vector

terms of the time.


In modern notation these formulae are written

A n -2Jn {tu)/n, B n = -2(e[n)Jn '(m).


It

was noted by Poisson, Connaissance des


j?

Terns,
f

1836 [published 1833],

p.

6 that

dA n

n at
a

memoir by
is

Lefort, Journal de Math. xi. (1846), pp. 142

152, in which an error made by

Poisson

corrected, should also be consulted.

A remarkable investigation of the approximate value of A n when n is large and < e < 1 is due to Carlini J: though the analysis is nob rigorous (and it would be difficult to make it rigorous) it is of sufficient interest for a brief
account of
it

to

be given here.
in. (1866), pp.

* Cf. Bourget,

Ann. Sci. de VEcole norm. sup.

5595, and Chree, Quarterly


204233. [Oeuvres,

Journal, xxi. (1886), p. 298.


t Hist, de VAcad. R. des Sci. de Berlin, xxv. (1769), [published 17711, pp.
in. (1869), pp.

113138.]

(Milan, 1817).
col.

X Ricerche sulla convergenza delta serie che serva alia soluzione del problema di Keplero This work was translated into German by Jacobi, Astr. Nach. xxx. (1850),

197254 [Werke, yii. (1891), pp. 189245]. See also two papers by Scheibner dated 1856, reprinted in Math. Ann. xvn. (1880), pp. 531544, 545560.

14]
It is easy to

BESSEL FUNCTIONS BEFORE


shew that

1826
equation

An
c
2

is

a solution of the

differential

+ <^-"- 2 (l-e 2 )J B ==0.


i

Define n by the formula

A n =2nn 2

/udt e /n

and then
2

(^+2 ) + tt-
u-

(l-2 ) = 0.
large.
;

Hence when n
If u
(n a
)

is

large either

u or u2 or du/dt must be

(n a ) respectively and and du/dt to l>e (n 2a ) and on considering the highest powers of n in the various terms of the last differential equation, we find that a= 1. It is consequently assumed that u admits of an expansion in descending powers of n in the form

we should expect

where

?<

u lt u 2

...

are independent of n.
first

On
/-ero

substituting this series in the differential equation of the

order and equating to

the coefficients of the various powers of n,


o
2

we

find that
...

= (l- f 2 )/*2
fl

+ 2oi)+Mo=0,
,

where v n'=du

/dt

so that

= +-

..

2,

and therefore

judt-n^ogj^f2)vA(1-e
and, since the value of

2
)

+ l}-ilog(l- f 2 ) + ...,
is

A n shews that Jude ~ n log ^e when be taken and no constant of integration is to be added.
From
Stirling's

small, the upper sign

must

formula

it

now

follows at once that


f

n exp

B
v/(^

{nj{\- t2)J *(i-<2 )*{i+N/(i-2 )} n

'

and

this is the result obtained

by

Carlini.

much
which

further

by Meissel

(see 8*11), while

This method of approximation has been carried Cauchy* has also discussed approximate
8"42), for

formulae for

An

in the case of

comets moving in nearly parabolic orbits (see


obviously inadequate.

Carlini's

approximation

is

The

investigation of which an account has just been given is

much more
corre-

plausible than the

arguments employed by Laplace + to establish the


for

sponding approximation

Bn

The
he uses
taking

investigation given
is

by Laplace

is

quite rigorous and the

of considerable

importance when

the value of

all

the coefficients in the series to be positive

Bn

or,

method which modified by alternatively, by


is

e is a pure imaginary. But Laplace goes on to argue that an approximation established in the case of purely imaginary variables may be used sans crainte in the case of real variables. To anyone who is acquainted

supposing that
'

'

with the

modem

theory of asymptotic series, the fallacious character of such

reasoning will be evident.


* Comptes Rendus,

xxxvm.

(1854), pp.
t.

990993.
v.
[first

f Mecanique Celeste, supplement,


pp. 486489.

published 1827].

Oeuvres, v. (Paris, 1882),

THEORY OF BESSEL FUNCTIONS


The
earlier portion of Laplace's investigation is based

[CHAP.

on the principle

that, in the case of a series of positive

terms in which the terms steadily in-

crease

up

to a certain point

and then steadily decrease, the order of magnitude


frequently be obtained from a consideration of
series.

of the

sum

of the series

may

the order of magnitude of the greatest term of the

For other and more recent applications of this principle, see Stokes, Proc. Camb. Phil. 362366 [Math, and Phys. Papers, v. (1905), pp. 221225], and Hardy, Proc. London Math. Soc. (2) n. (1905), pp. 332339 Messenger, xxxiv. (1905)., pp. 97101. A statement of the principle was given by Borel, Acta Mathematica, xx. (1897), pp. 393
Soc. vi. (1889), pp.
;

394.

The
The

following exposition of the principle applied to the example considered


interest

by Laplace may not be without


series considered is

iL<" = 2
in

2
,=o

(n

+ 2m) nn + 2m ~ 2 t n + *>" 2 + 2m m\{n + m)\


The

'

which n

is

large
/u

and

t.

has a fixed positive value.

greatest term

is

that for which

mfi, where

is

the greatest integer such that


4/x (n

+ /x)

(n + 2/i

- 2) ^ (n + 2M ) n2 e2

and so

fi

is

approximately equal to

iW(l + 2 )-l} + *e2/(l+ 2


Now,
if

).

u m denotes the general term in BJ~ l \

it is

easy to verify by Stirling's theorem

that, to a first approximation, -H^<**>q&,

where
).

"

Hence
since* q
is

.BB< 1

=-2V(l + *2 )/(e 2 )~M {1 + 2q + 2q* + 2 q9 +


log ?
f
,

. .

.}

~2ttMX/{7r/(l-?)},
nearly equal to
Stirling's
1.

Now, by

theorem,
6"- 1 exp {ws/(l
7T% 2

+ g2 )|
'

{l+ x/(l+ 2 )}
)

and

so

BJ l)r

2 v'(l

nn3

+ c2 )) ^*
J

exp
{1

+ 2 )} + ^(1 + ,')}
{

WU

The

inference which Laplace

drew from

this result is that


2

jD,l

~_ V

/ 2V(1

**) \*

6"expW(l-e )}
{l

ttw*

+ V(l-e )}
2

'

This approximate formula happens to be valid when e < 1 (though the reason for this restriction is not apparent, apart from the fact that it is obviously necessary), but it is difficult to prove it without using the methods of contour

The
cf.

formula 1 + 2 2<f'

~ N/W(l-g)} maT

be inferred from general theorems on series;

Bromwieh, Theory of

Infinite Series, 51.


elliptic functions,

It is also

a consequence of Jacobi's transformation


1

formula in the theory of


see

*3
Modern Analysis,
21*51.

(0 \r)

= (-*)-* M0 -TI

);

1*5]
integration

BESSEL FUNCTIONS BEFORE


(cf. 8*31).

1826

Laplace seems to have been dubious as to the validity

of his inference because, immediately after his statement about real

and

imaginary variables, he mentioned, by way of confirmation, that he had


another proof; but the latter proof does not appear to be extant.

The researches of Fourier. In 1822 appeared the classical treatise by Fourier*,


1"5.

La TMorie analytique
It is

de la Chaleur; in this work Bessel functions of order zero occur in the discussion of the symmetrical motion of heat in a solid circular cylinder.

shewn by Fourier ( 118 120) that the temperature x from the axis of the cylinder, satisfies the equation

v,

at time

t,

at distance

dv_K_/d?v lcfoA Jt~lW\dtf + xdx)'


where K, G, D denote respectively the Thermal Conductivity, Specific Heat and Density of the material of the cylinder; and he obtained the solution
v

e~
ni

gx*
f

tf<*

x*
\ ')
'

22

2 2 .42

2 2 .42 .6 2

where g
at the

= rnCD/K and

has to be so chosen that


hv

+ K(dvldx) =
is

boundary of the cylinder, where h


(

the External Conductivity.

theorem that and no complex roots. His proof is slightly incomplete because he assumes that certain theorems which have been proved for polynomials are true of integral functions; the defect is not difficult to remedy, and a memoir by HurwitzJ
Fourier proceeded to give a proof

307

309) by

Rolle's

the equation to determine the values of

m hasf an infinity of

real roots

has the object of making Fourier's demonstration quite rigorous.


It should also be

formula ( 313) for dervate; generalisations of this formula Another formula given by Fourier, namely
22 2 2 42
.

mentioned that Fourier discovered the continued fraction the quotient of a Bessel function of order zero and its
will

be discussed in

5-6, 9*65.

2 2 42 6 2
.

1 =

f*
|

cos (a sin x) dx,


it is

had been proved some years earlier by Parseval; are now known as Bessel's and Poisson's integrals
*

a special case of what

22, 23). memoir deposited in the archives of the French Institute on Sept. 28, 1811, and crowned on Jan. 6, 1812. This memoir is to be found in the Mim. de VAcad. des Sci., iv. (1819), [published 1824], pp. 185555; v. (1820),
(

The

greater part of Fourier's researches was contained in a

[published 1826], pp. 153246. t This is a generalisation of Bernoulli's statement quoted in 1*2. t Math. Ann. xxxin. (1889), pp. 246266.
648. This paper also contains the formal M4m. des savant itrangers, i. (1805), pp. 639 statement of the theorem on Fourier constants which is sometimes called Parseval's theorem another paper by this little known writer, Mim. des savans Strangers, i. (1805), pp. 379 398, con-

tains a general solution of Laplace's equation in a form involving arbitrary functions.

10

THEORY OF BESSEL FUNCTIONS

[CHAP.

The expansion of an arbitrary function into a series of Bessel functions of order zero was also examined by Fourier ( 314 320); he gave the formula for the general coefficient in the expansion as a definite integral.

was examined much more recently by Hankel, 471494; Schlafli, Math. Ann. x. (1876), pp. 137142; Diui, Serie di Fourier, i. (Pisa, 1880), pp. 246269 Hobson, Proc. London Math. Soc. (2) vn. (1909), pp. 359388; and Young, Proc. London Math. Soc. (2) xvni. (1920), pp. 163200.
validity of Fourier's expansion

The

Math. Ann.

vm.

(1875), pp.

This expansion will be dealt with in Chapter xvni.

1/6.

The researches of Poisson.

The unsymmetrical motions of heat in a solid sphere and also in a solid cylinder were investigated by Poisson* in a lengthy memoir published in 1823.
In the problem of the sphere f, he obtained the equation

where r denotes the distance from the centre, p


integer (zero included), and

is

a constant, n

is

a positive

R is that

factor of the temperature, in a It

mode, which is a function of the radius vector. a solution of the equation is


cos (rp cos
a>)

normal was shewn by Poisson that

sin2n+1 coda

Jo

and he discussed the cases


( 3*3)

?i

= 0,

1,

2 in detail.

It will appear subsequently

that the definite integral

is

(save for a factor) a Bessel function of

order n

+ ^.
(ibid. p.

In the problem of the cylinder

340

et seq.)

the analogous integral

is

V
.'o

cos (h\ cos


is

co)

sin2n G>cifo>,

where

w= 0,

1, 2, ...

and \

the distance from the axis of the cylinder.


2'3).

The

integral is

now known

as Poisson's integral (

In the case n

0,

and
is

its

derivate was obtained by Poisson

an important approximate formula for the last integral (ibid., pp. 350 352) when the variable

a>

large; the following is the substance of his investigation:

Let J

(k)

=it

cos (i cos ) da,

'

(k)=
ir

cos

sin (k cos ) dm.

Then J

(k) is

a solution of the equation

*&ffl + 1+ i),_a (
249403. 300 etseq. The equation was also studied by Plana, Mem. della R. Accad. delle Sei. di Torino, xxv. (1821), pp. 532 534, and has since been studied by numerous writers, some of whom are mentioned in 4*3. See also Poisson, La TMorie Mathimatique de la Chaleur (Paris,
+ Ibid.
p.

* Journal de I'ficole JR. Polytechnique, xn. (cahier 19), (1823), pp.

1835), pp. 366, 369.

t See also Rohrs, Proc. London Math. Soc. not used by Poisson.

v. (1874), pp.

136

187.

The notation J

(k)

was

1*6]

BESSEL FUNCTIONS BEFORE


large, l/(4 2 )
is

1826

11

When k is

may

be neglected in comparison with unity and so

we may expect

that Jo{k)sJk

approximately of the form


that

cos k + Bsm k where

and

B are constants.
2

To determine A and B observe


cos k
.

(k)

- sin k
for

'

()

=-

{cos 2 ba> cos (2k sin 2 a>)


o

"J

+ sin 2 <a cos (2 cos

)}

ofo.

Write

ir

- to

a
.

in the latter half of the integral

and then

cos k

(k)

- sin k

/'

(k)

cos 2

Ao cos

(2k sin 2 i)

o?<u

^j
/"

I
V(2 * }

-2-)

"^'

and similarly

sin ../(*)

+ cos . J

'

()

= ^^

fl

-^Y"

nrnx^dx.

But

fo *-^oo y

lim

VW

(l-^V.^
\

2kJ sin

f J sin

^V.^K/^),

by a

well

known formula*.

[Note. It is not easy to prove rigorously that the passage to the limit is permissible the simplest procedure is to appeal to Bromwich's integral form of Tannery's theorem,

Bromwich, Theory of Infinite


It follows that
(

Series, 174.]

cos k

(k)

-sink.

'

(k)

1 = j^

1 4- t ),

sin

J
;

(k)

+ cos k J
.

'

(k)

= -jip- (1

rj

k ),

where

**-*-()

and

ri

k -*-0

as k-+-x>

and therefore

J ^ =

J(*k)

K1 +

ffc

C S
^

*+^+

sin

^'

i
It

^o'

(*)

= 77^m t - (!+*) sin *+(! + ?*) cos


t/ () is

^'3-

was then assumed by Poisson that

expressible in the form

where

A =.5=1. The

series are, however, not convergent

but asymptotic, and the validity

of this expansion was not established, until nearly forty years later,

gated by Lipschitz, Journal fur Math. lvi. (1859), pp. 189

when

it

was

investi-

196.

The

result of formally operating

on the expansion assumed by Poisson


1

for the function

(k) *J(irk)

with the operator -jt + 2

eft

+ rn

is

-*[

.rz.l.B -jA

2.

2B"- (I. 2 + 1)

A'

2.

3B'"-(2. 3 + j) A"

& p
+

&
Ji

*
+

+-

1 J
J'
15, 1914),

+ 8m *[_
* Cf.
p. 71, for

+-

Watson, Complex Integration and Cauchy's Theorem (Camb. Math. Tracts, no. a proof of these results by using contour integrals.

12
and
so,

THEORY OF BESSEL FUNCTIONS


by equating
to zero the various coefficients,

[CHAP.

we

find that

A'A---H,

A" -" A

A *, 2 g2

A'"*= A -

9 25 2.3.83
'

R
'

'

and hence the expansion of Poisson's integral


/""

is

OM(*ooe.)A,-^J
j
o

9-25 9 \ /V\*IY 1 - 1 I--2T8^ + 2V3."83l3+-" > 8 LV


/.
1
,

C08 *

+ 8l"2T8^~2.3.88ife + -; 8m

9.25

,1
*J-

But, since the series on the right are not convergent, the researches of Lipschitz and subsequent writers are a necessary preliminary to the investigation of the significance of

the latter portion of Poisson's investigation.


It should

was first given by W. was expressed thus

be mentioned that an explicit formula for the general term in the expansion B. Hamilton, Trans. B. Irish Acad. xix. (1843), p. 313; his result

"["

cos (20 sin a) da =

-^ 1

[0]

([

- *]) 2 (43) cos

(20 - $n,r
[0]

- }n),

and he described the expansion as semi-convergent; the expressions to be interpreted as 1/n and (-|)( f) ... (-+)
!

-B and [-$]" are

A result of some importance, which was obtained by Poisson in a subsequent


memoir*,
is

that the general solution of the equation

is

y = Ax%

Jo

e- hxcoSu> da>

+ Bx$

~ 2 e hxcoaa log (x sin

m) da>,

Jo

where

and

B are

constants.

It follows at

once that the general solution of the equation

dxr
is

x dx

= A\
.'o

e~ hxcoau da)

+B

e hxcoa '\og

(a?

sin2 w)

cZo>.

Jo

likely that

This result was quoted by Stokesf as a known theorem in 1850, and it is he derived his knowledge of it from the integral given in Poisson's memoir; but the fact that the integral is substantially due to Poisson has

been sometimes overlooked J.


* Journal de VEcole R. Polytechnique, xn. (cahier 19), (1823), p. 476. The corresponding general integral of an associated partial differential equation was given in an earlier memoir,
ibid, p. 227.

t Camb. Phil. Trans, ix. (1856), p. [38], [Math, and Phys. Papers, hi. (1901), J See Encyclopidie des Set. Math. n. 28 ( 53), p. 213.

p. 42].

1'7]

BESSEL FUNCTIONS BEFORE


The researches of Bessel.

1826

13

17.

bear his

The memoir* in which Bessel examined in detail the functions which now name was written in 1824, but in an earlier memoir f he had shewn
is

that the expansion of the radius vector in planetary motion

+ 2

Bn cosnM,

ne sin u)

where

Bn = Bn should

sin

u sin (nu

du

this expression for

be compared with the series given in

1'4.

In the memoir of 1824 Bessel investigated systematically the function I h k


defined by the integral J
Ikh

If = k zvrJo
I

2"

cos (hu

k sin u) ' du.


many
is

He

took h to be an integer and obtained


is

of the results which will be

given in detail in Chapter n.


function which

not adapted for defining the most worth study when h is not an integer (see 10-1) the function which is of most interest for non-integral values of h is not Ikh but the function defined by Lommel which will be studied in Chapter in.
Bessel's integral
;

that

After the time of Bessel investigations on the functions became so numerous it seems convenient at this stage to abandon the chronological account
logical order.

and to develop the theory in a systematic and

account of researches from the time of Fourier to 1858 has been compiled by Wagner, Bern MittheUungen, 1894, pp. 204266 a briefer account of the early history was given by Maggi, Atti delta It. Accad. dei Lincei, (Tra/isunti), (3) iv. (1880), pp. 259263.
historical
;

An

* Berliner Abh. 1824 [published 1826], pp. 1 52. The date of this memoir, " Untersuchung des Theils der planetarischen Storungen, welcher aus der Bewegung der Sonne entsteht," is Jan. 29, 1824.

t Berliner Abh. 181617 [published 1819], pp. 4'J 55. J This integral occurs in the expansion of the eccentric anomaly

with the notation of 1-4,

nA n

= 21\ t

a formula given by Poisson, Connaissance des Terns, 1825 [published 1822],

p.

383.

CHAPTEE
21.

II

THE BESSEL COEFFICIENTS


The
definition

of the Bessel
is

coefficients.

The

object of this chapter

the discussion of the fundamental properties

of a set of functions

known
;

as Bessel coefficients.

There are several ways of

the method which will be adopted in this work is to in a certain expansion. This procedure is due coefficients the as them define properties of the functions from his defimany derived who Schlomilch*, to

denning these functions

nition,

and proved incidentally that the functions thus defined are equal to the definite integrals by which they had previously been defined by Bessel f. It should, however, be mentioned that the converse theorem that Bessel's integrals are equal to the coefficients in the expansion, was discovered by Hansen:}:

fourteen years before the publication of Schlomilch's memoir. results had been published in 1836 by Jacobi ( 222).

Some

similar

The generating

function of the Bessel coefficients

is

It will be shewn that this function can be developed into a Laurent series, n qua function of t; the coefficient of t in the expansion is called the Bessel is denoted by the symbol Jn (z), coefficient of argument z and order n, and it

so that

*('"?)(1)

I
=-oo

t"Jn

(z).

expanded into an and for all values of t, ielt can be expanded into an absolutely converwith the exception of zero, e~ of t. When these series are multiplied powers descending gent series of convergent series, and so it may be absolutely together, their product is an is to say, we have an expansion of the arranged according to powers of t that and t, t = excepted. form (1), which is valid for all values of z

To

zt can be establish this development, observe that e^

absolutely convergent series of

ascending powers of

* ZeiUchriftfiir

namely that of

SCO80 , see Frullani,

Math, und Phys. n. (1857), pp. 137 165. For a somewhat similar expansion, Mem. Soc. Ital. (Modena), xvm. (1820), p. 503. It must be

pointed out that Schlomilch, following Hansen, denoted by Jftn what we now write as Jn {2z) but the definition given in the text is now universally adopted. Traces of Hansen's notation
are to be found elsewhere, e.g. Schlafli, Math. Ann. 22. t Berliner Abh. 1824 [published 1826], p.
theil, [Schriften

m.

(1871), p. 148.

+ Ermittelung der Absoluten Storungen in Ellipsen von beliebiger Excentricitdt und Neigung, der Sternwarte Seeburg : Gotha, 1843], p. 106. See also the French translaab$olues (Paris, 1845), p. 100, and Leipziger tion, Memoire sur la determination des perturbations
i.

Abh. n. (1855>, pp. 250251.

2*1, 2 '11]

THE BESSEL COEFFICIENTS


we
write

15

If in (1)

\jt

for

t,

we get
n -x

= 5 i-tyj_ n { Z
n
x

),

on replacing n by - w. Since the Laurent expansion of a function a comparison of this formula with (1) shews that
(2)

is

unique*,

J-n{z)
is

= (-YJn (*),

where n

any integer

a formula derived by Bessel from his definition of

Jn (z)

as an integral.
(2) it
is

From
(3)

evident that (1)


eiz(t-llt)

may be
{t

written in the form


)n t - n]

=J

{z)

+ J
w-1

+{_

Jn {2)

The Analyst, of elementary applications of these functions to problems of Mathematical Physics has been compiled by Harris, American Journal of
i.

A summary of elementary results concerning Jn (z) has been given by Hall,


8184, and an account

(1874), pp.

Math, xxxiv. (1912), pp. 391420.

The

ix. (1909), pp.

function of order unity has been encountered by Turriere, Nouv. Ann. de Math. (4) 433441, in connexion with the steepest curves on the surface z=y (5.r2 -y).

2*11.

The ascending

series

for

Jn (z).
the form of an ascending series of powers
series for

An explicit expression for Jn (z) in


of z is obtainable

by considering the

exp

(h zt)

and exp ( - kz/t), thus

exp (**(-- 1/0)=

I flT v
r =o

- **>'"'L
ml

rl

l=t>

term of the first series on the associated with the general term of the second series gives rise to a term involving tn is the term for which r = n + ; and, since n > 0, there is always one term for which r has this value. On associating these
right which,

When n is a

positive integer or zero, the only

when

terms

for all the values of

m, we see that the

coefficient of t n in the

product

is

m=o(n-rm)l

ml

We
(1)

therefore have the result

Jn(z)= S

* Z)

series in t, in which some of the were not zero. If we then multiplied the expansion by t and integrated it round a circle with centre at the origin, we should obtain a contradiction". This result was noticed by Cauchy, Comptes Rendus, xm. (1841), p. 911.

* For, if not, zero could be

expanded into a Laurent


<")

coefficients (say, in particular, that of

'

16

THEORY OF BESSEL FUNCTIONS


is

[CHAP.

II

where n

a positive integer or zero.


zn

The

first

few terms of the series are

given by the formula


(2)

Jn () =

z*
2
.

2^1 { 1 ~ 2
Jo (?)

(n

+
1

2-

* + 1) (n +

2)

'"]'

In particular
(3)

22

^ 22 42
.

22

42 6 2
.

"

To obtain the Bessel coefficients of negative order, we select the terms involving irn in the product of the series representing exp tyzt) and exp(- \z\t\ where n is still a positive integer. The term of the second series which, when
n associated with the general term of the first series gives rise to a term in t~ is the term for which ra = n + r ; and so we have

J -n{Z) ~

rZo

r\"

(n

+ r)l

'

whence we evidently obtain anew the formula

2'1 (2),

namely

J_^) = (-)
It is to

^(*)-

be observed that, in the series (1), the ratio of the (w + l)th term 2 oo for all to the rath term is - s /{ra (n + ra)}, and this tends to zero as ra that follows it values of z and n. By D'Alembert's ratio test for convergence, it so and and n, the series representing Jn (z) is convergent for all values of z
,

is

an integral function of z when n =


It will

0,

1,

2,

3,

. . .

appear later (4"73) that


is is

Jn

(z) is
;

and so

it

a transcendental function

not an algebraic function of z moreover, it is not an elementary

to say it is not expressible as a finite combination of exponential, logarithmic and algebraic functions operated ou by signs of

transcendent, that

indefinite integration.

From
ance
(cf.

(1)

we can obtain two

useful inequalities, which are of

some importis

Chapter xvi) in the discussion of series whose general term


z be real or complex,

multiple of a Bessel coefficient.

Whether

we have
ao
i

i2W

*
and
so,

n!

mZom\(n

+ l)m

'

when n ^ 0, we have
!

(4)

J.

< & exp (|!^) <


I
r l

^
l.

exp (J

|>).

854

This result was given in substance by Cauchy, Comptes Rendus, a similar but weaker inequality, namely
;

xm.

(1841), pp. 687,

lJ

.l< -^P(l*l ,

was given by Neumann, Theorie der BesseVschen Functionen

(Leipzig, 1867), p. 27.

2*12]

THE BESSEL COEFFICIENTS


considering
all

17

By
()

the terms of the series for

Jn (z)

except the

first, it is

found that

j.()-^(i+),
||

where
It should

"P <M p(il4')-l < r \n + 1/

(*

' ,'>- 1
l

any, bounded
*-plane.

be observed that the series on the right in 2*1 (1) converges uniformly in domain of the variables z and t which does not contain the origin in the
if 8,

For

A and

R are positive constants and if


d<|t|<A,
|*|

<

A.I,

the terms in the expansion of exp (\zt) exp ($z/t) do not exceed in absolute value the corresponding terms of the product exp ($RA) exp QR/8), and the uniformity of the convergence follows from the test of Weierstrass. Similar considerations apply to the series obtained

by term-by -term

differentiations of the expansion 2^*t/ n


z

(z),

whether the differentiations be

performed with respect to


2" 12.

or

or both z and

t.

The recurrence formulae.

The equations*
(1)

Jn-i (*)

Jn (z), + Jn+1 (*) = ~ z

(2)

Jn_ (z)-Jn+1 (z) = 2Jn'(z),


1

which connect three contiguous functions are useful in constructing Tables of


Bessel coefficients; they are

known

as recurrence formulae.

To prove the
namely

former, differentiate the fundamental expansion of 21,

,*'-= 5 tj (z), n
with respect to
t
;

we get
$z(l

+ l/t*)e
2

iz(t ~

m=

2 nt^Jniz),
nt^Jniz).
t

so that

^(1 + 1/e ) 5
If the expression on the left
is

Jn (z)= 2

arranged in powers of

and

coefficients of n_1
tf

are equated in the two Laurent series, which are identically equal,

it is

evident

that

\z {Jn-i (z) + Jr* which


*
its

(z))

=nJn (z),

is

the

first

of the formulae f.

Throughout the work primes are used to denote the derivate of a function with respect to

argument.
f Differentiations are permissible because ( 2*11) the resulting series are uniformly convergent. of coefficients is permissible because Laurent expansions are unique.

The equating

18

THEORY OF BESSEL FUNCTIONS


iz<t ^(t-l/t)e

[CHAP.
;

II

Again, differentiate the fundamental expansion with respect to z and then

m=

= -oo

Jn '(z),
t

so that

\(t

- Ijt) 2
=-oo

Jn (z) = 5
w=-oc

Jn

'

(z).

By equating
The
(3)

coefficients of tn

on either side of this identity we obtain formula

(2) immediately.

results of adding

and subtracting (1) and (2) are


'
x

(4)

zJn (z) + nJn (z) = zJn__ (z), zJn'(z)-nJn (z) = -zJn+1 (z).
to

These are equivalent


(5)

^{^ (*)}=*"</-!(*),
^{z-Jn (z)} = -z-"Jn+1 (z).
is trivial

(6)

In the case w=0, (1)


(7)

while the other formulae reduce to

/,'(#)

= -/(*).

(1) and (4) from which the others may be derived were discovered by Abh. 1824, [1826], pp. 31, 35. The method of proof given here is due to Schlomilch, Zeitschrift fur Math, und Phys. ir. (1857), p. 138. Schlomilch proved (1) in this manner, but he obtained (2) by direct differentiation of the series for Jn (2).

The formulae

Bessel, Berliner

A
(g)

formula which Schlomilch derived


2r

{ibid. p.

143) from (2)

is

^T-)=
\m

{-TrCm .Jn -r+n {z\


we have

where

Cm

is

a binomial

coefficient.
(6),

By

obvious inductions from (5) and

(d
/

n
{*

zTz)

Jn(z)}=2 n m Jn-m(z),

(10)

(is)
is

m ~ {2 nJn (z)] = (-r*-"^+ (*).


in is

where n

any integer and

any positive

integer.

The formula

(10)

is

due

to Bessel (ibid. p. 34).

As an example
*Ji

of the results of this section observe that

0) = 4t/a 0) - zJs (*)

= 4J"

(z)

- 8J, (z) + zJ> (z)


1

= 4 2 (-Y~ nJi,(z)^(-rzJiN+1 (z)


=i

= 4 1 (-y-*nJm (z),
since

zJ^+i (z) -*

as iV-*- 00

by

211

(4).

2-13, 2'2]

THE BESSEL COEFFICIENTS


obtained,

19

The expansion thus


(11)
is

*./,(*)

=4 I
n-l

{-)n

nJ2n (z),
functions ( 3'o7).

useful in the developments of

Neumann's theory of Bessel

2*13.

The

differential equation satisfied by

Jn {?)>

When

the formulae 2'12 (5) and (6) are written in the forms

the result of eliminating J, _,


t

(z) is

seen to be

d
dz
that
is

f-^Wni*)} = -z*- n Jn (z),

to say

and

so

we have

Bessel's differential equation*


zi

(1)

W)

+z

dJAz)

+ (z2 _ wl) Jh {z) ^


z (dldz).

The

analysis

is

simplified

by using the operator ^ defined as


are

Thus the recurrence formulae


(S

+ n) Jn (z) = zJn _
(^

{z),

(^ - n

1)

Jn -i (z) = - zJn (z),

and so
n + 1)
[z-i

(*

+ n) Jn (z)} = -zJn (z),

that

is

z-> (^

- n) (* + n) Jn (z) =-zJn (z),


C&-n*)Ju {z) = -z*Jn (z)

and the equation


reduces at once to Bessel's equation.
Corollary.

The same
(S

differential equation is obtained if

Jn +

(z) is

eliminated from the

formulae

+ n + 1) Jn +

(z)

= zJn (z), (-n)Jn (z)=-zJn ^

(z).

2*2.

Bessel's integral for the Bessel coefficients.


shall

We
(1)

now prove

that

Jn (z) =

JLit

"'
I

cos {n$

- z sin B) d6.

'o

This equation was taken by Bessel f as the definition of

Jn (z),

and he

derived the other properties of the functions from this definition.


*

Berliner Abh. 1824 [published 1826], p. 34

see also Frullani,

Mem.

Soc. ItaL.(Modena),

xvm.

(1820), p. 504.

t Ibid. pp. 22 and 35.

20

THEORY OP BESSEL FUNCTIONS


It is frequently convenient to

[CHAP.

II

modify

(1)

by bisecting the range of

in-

tegration and writing


(2)

2tt

for

in the latter part.


f"

This procedure gives

Jn (z)=IT

cos (n0

- z sin 0) dO.
first

JO
27r,

Since the integrand has period


into
1

the

equation

may

be transformed

/*2ir+a
I

(3)

J n (z) = 2~
is

cos(n0-zsm0)d0,
~x

where a

any angle.
(1),

To prove

multiply the fundamental expansion of

(1)

by t~n

and

integrate* round a contour which encircles the origin once counterclockwise.

We

thus get

2771 J

=-
all

Z7rt

so

The integrals on the right we obtain the formula

vanish except the one for which

m = n;

and

Take the contour

may

be a circle of unit radius and write t = e~ i0 so that be taken to decrease from 2tt + a to a. It is thus found that
to
t

rtor+a

(5)

Jn(*)^-/

*<**-*) <$

77" J a

a result given by Hansen f in the case a = 0.

In this equation take a


former part, replace

by

= ir, bisect the range of 0. This procedure gives

integration and, in the

Jn (z) = -L
and equation
(2),

e*(*-z n )

+ e - Une - z sin 9) Q
J

from which (1)

may be

deduced,

is

now

obvious.

Various modifications of Bessel's integral are obtainable by writing

Jn (z) =
If

f
I

cos

nd cos (z sin 0)d0-i

If... n0
I

sin

sin (z sin 0) d0.

IT

Jo

TT J

be replaced by 7r in these two integrals, the former changes sign when n is odd, the latter when n is even, the other being unaffected in each case and therefore
s Jn (&) J i ; =i

n n & sm ( z sm 0) d0
(odd).
.

(6)

T JO

sin

w0 sin (s sin 0) d0
I

Term -by- term


is

integration

is

permitted because the expansion


(a+)

contour. It

convenient to use the symbol

uniformly convergent on the to denote integration round a contour encircling


is

the point a once counterclockwise.


t Ermiitelung der absoluten Storungen (Gotha, 1843), p. 105.

"

2*21]

THE BESSEL COEFFICIENTS

21

Jn {z)K)
2

'**
I

cos

n$ cos (z sin

6)

dd
("even).

= ir

,^
I

cos w

cos (z sin &)

d0

Jo

If

be replaced by tt /n (*)

77

in the latter parts of (6)


1}
I

and

(7), it is

found that
(w odd),

8)

--

(-)*<

cos

n?7

sin (s cos

17) <fr;

(9)

/ii

(z)

=-(-)**

cos nr) cos (z cos 17) cfy

(n even).

The
that

last

two
It

results are

due substantially to Jacobi*.


1.

[Note.

was shewn by Parseval, Mem. des savans Grangers,


.

(1805), pp.

639648,

2s

+ 22742 ~

a*

2 2.42.62 +

1
,,

[*
cos (

and
will

so, in

the special case in which n = 0, (2) will be seen in 2-3 that two integral representations of
identical

J be described as ParsevaVs integral.

8m *) <**

It

Jn (z),

namely

Bessel's integral
for

and Poisson's integral become


justified.]

when w=0,

so a special

name

thL

case

is

The reader
2-12 (4)

wijl find it interesting to obtain (after Bessel)

the formulae 2-12

(1)

and

from Bessel's

integral.

2 '21.

Modifications of Parseval's integral.

formulae involving definite integrals which are closely connected with PaTseval's integral formula are worth notice. The first, namely
(1)
is

Two

Jo{J(*-&)} = \ /"**" cos (lain*)***,


The simplest method of proving
it is

due to Bessel f.

to write the expression on the

right in the form

/>
J

e v cos 0+vs sin

dfy

expand in powers of y cos d+iz sin 6 and use the formulae


J
^
(y cos 6

+ iz sin 6f +

dd = 0,

f"

(y cos 6

+ iz sin 6f dd =

2r ( w + *)r(i)
(y2

the formula then follows without

difficulty.

The other
(2)
is

definite integral,

due to Catalan J, namely

J
(1)

(2 <Jz)

=IT

/"" e<i+)cos

cos {(l

- z) sin
1

0} dd,

J
1

a special case of

obtained by substituting

-z and

+z

for 2

and y

respectively.

* Journal fUr Math. xv. (1836), pp. 1213. integrals actually given by Jacobi had limits

[G<?s.

Math. Werke,

vi. (1891), pp.

100102]
1,

the
2/*-.

and w with

factors I/*- replacing the factors

See also Anger, Neueste Schriften Comptes Rendus, xxxvm. (1854), pp. t Berliner Abh., 1824 [published Ges. in Danzig, v. (1855), p. 10, and

der Naturf. Ges. in Danzig, v. (1855), p.

and Cauchy,

910

913.
See also Anger, Neueste Schriften der Naturf.
p. 151.
(2) xli. (1876), p.

1826], p. 37.

Lommel, ZeiUchrift fUr Math, und Phys. xv_(1870),


938.

t Bulletin de Vdcad. R. de Belgique,

22
Catalan's integral

THEORY
may

OJF

BESSEL FUNCTIONS

[CHAP.

II

be established independently by using the formula


'(0+) i'(0+)

no that
./

(2i\'*)

-i_-J-.
=
J_.

2
(0
/"

-,
+

t-^-Wdt

ex

p^ + ! )^ = l
N
;

[* exp {*+*-}

<W,

by taking the contour


integration.

to be a unit circle

the result then follows by bisecting the range of

2"22.

Jacobi' s expansions in series of Bessel coefficients.

Two 'series, which


covered by Jacobi*.
t

are closely connected with Bessel's integral, were disis

ie

in

The simplest method of obtaining them the fundamental expansion 21 (3). We thus get
M=l

to write

= J 0) + 2 2 Jm 0)cos 2nd
n=\

2i

2 Jm+1 (*) sin (2 +


w=0

1)

0.

On adding and
we
find

subtracting the two results which are combined in this formula,

(1)

cos (z sin 0)

= J O) + 2 2 Jm (z) cos 2n0, n=


\

(2)

sin(*sin0)=

2 /2n+1 (*)sin(2n + l)0.

Write \ir
(3)

r)

for 6,

and we get
t?)

cos

cos

= Jo (z) + 2 2
=i

(-)./,(*) cos 2ni7,

(4

sin (s cos

t/)

2 (-)M J2n+1 0) cos (2n +


=o

1)

rj.

The results (3) and (4) were given by Jacobi, while the others were obtained later by Anger t. Jacobi's procedure was to expand cos (s cos 17) and sin (z cos q) into a series of
cosines of multiples of
integrals
t}, and use Fourier's rule to obtain the which are seen to be associated with Bessel's integrals.

coefficients in the

form of

In view of the fact that the first terms in (1) and (3) are not formed according to the same law as the other terms, it is convenient to introduce Neumann's factor* en> which is defined to be equal to 2 when n is not zero,

and

to be equal to 1
*

when n

is zero.

The employment
[Ges.

of this factor, which

Journal fUr Math. xv. (1836), p. 12.

Math.

U'erke, vi. (1891), p. 101.]

t Neueste Schriften der Xaturf. Gen. in Danzig, v. (1855), p. 2. % Neumann, Theorie der BesteVschen Functionen (Leipzig, 1867), p.

7.

2-22]
will

THE BESSEL COEFFICIENTS

23

be of frequent occurrence in the sequel, enables us to write (1) and (2) in

the compact forms:


oo

(5)

cos (^ sin 0)

= X m J2n (z) cos 2nd,


e.

=o
ao

(6)

sin (z sin 0)

= X
=o
find

e2w+1

Jm+1 (z) sin (2n + 1) 6.

If

we pub

in (5),

we

0)
If

1=2

m Jm (z).

of. Bessel coefficients when m is any positive any polynomial is thus expansible. This is a special case of an expansion theorem, due to Neumann, which will be investigated in Chapter xvi.
is

that zm

we differentiate (5) and (6) any number of times before putting 6 = 0, we obtain expressions for various polynomials as series of Bessel coefficients. shall, however, use a slightly different method subsequently ( 27) to prove

We

expansible into a series

integer.

It is then obvious that

For the present, we will merely notice before 6 is put equal to 0, there results
(8)

that, if (6)

be differentiated once

*= 2
%-ir

e 2n+1

(2n

+ l)J2n+1 (z),
(5)

while, if 6 be put equal to


(9)

after
2

two differentiations of
i

and

(6),

then

zsinz
zcosz

= 2{2*J (z)-4?J (z)+6*J


=2
jl./;(*)-3 s

(z) -...},

(10)

(z)

52

!i

(z)-

...}.

These results are due to Lommel*.


Note.

The expression exp {$z{t-

1/0} introduced in $ 2 1 is not a generating function

in the strict sense.

The generating function t

associated with

en

Jn (?)

is

**"/

(z).

If this expression be called S,

by using the recurrence formula

2-12 (2),

we have

If

we

solve this differential equation

we get
Z
(t

(11)

S=ei*-W+l(t + tyj*V-wJ e-*V-V')j


was given by Brenke,
Bull.

)di:
Soc. xvi. (1910),

result equivalent to this

American Math.

pp. 225230.
* Studien ilber die Bessel'schen

It will

Functionen (Leipzig, 1868), p. 41. be seen in Chapter xvi. that this is a form of " Lommel's function of two variables."

24
2
3.

THEORY OF BESSEL FUNCTIONS


Poissons integral for the Bessel
coefficients.

[CHAP.

II

Shortly before the appearance of Bessel's memoir on planetary perturbations, Poisson had published an important work on the Conduction of Heat*, in the
course of which he investigated integrals of the types f

Jo

^ cos (z cos 6) sin

2"-1" 1

Odd,

(' cos (z cos 6) sin*1 0d0,

Jo

where n

is

a positive integer or zero.

He

proved that these integrals are

and gave the investigation, which has already been reproduced in 1*6, to determine an approximation to the latter integral when z is large and positive, in the special case n = 0.
solutions of certain differential equations^

We

shall

now prove

that

and, in view of the importance of Poisson 's researches,

it

describe the expressions on the right as Poissons integrals for n case n = 0, Poisson's integral reduces to Parseval's integral ( 2'2).
It is easy to prove that the expressions

seems appropriate to J (z). In the

Jn (z);
I

we expand the integrand term-by-term we have


for, if
||,

in powers of z

under consideration are equal to and then integrate

fir

oo

7tJ

cos (* cos

0)8^0(20 = - t 7T m =o

V (2W)!

\m sm

fir

cos2'0sin=0<W

Jo

=
j!

~(>0r

'

2.4.6... (2n
'

+ 2m)

~ -1.3.5...(2-1) 2
and the result
is

H+m m!(n oli 2

+ m)r

obvious.

* Journal de VEcole R. Polyteckuique, xn. (cahier 19), (1823), pp. 249403. previously been t Ibid. p. 293, et seq. ; p. 340, et seq. Integrals equivalent to them had examined by Euler, hut. Calc. Int. u. (Petersburg, 1769), Ch. x. 1036, but Poisson's forms are more elegant, and his study of them is more systematic. See also 3'3.

J E.g. on p. 300, be proved that,

if " cos (rp cos ) sin 2


" 1 1

R=rn+1
then

udu,

Jo

satisfies the differential

equation

dr1
Nielsen,

r-

Handbuch der

Tfieorie der Cylinderfunktionen (Leipzig, 1904), p.

51, calls

them
is

Bessel's second integral,


||

but the above nomenclature seems preferable. The series to be integrated is obviously uniformly convergent ; the procedure adopted

due

to Poisson, ibid. pp. 314, 340.

2-3, 2*31]

THE BESSEL COEFFICIENTS

25

Poisson also observed* that


(' e^* sin8'* Odd Jo
this is evident
left

=
it

Jo

cos (z cos 0) sin 0d0\

when we

consider the arithmetic

mean
by

of the integral on the


it

and the integral derived from

by replacing

0.

We thus get A
(3) slight modification of this formula,

namely
(1

/(*)

r(n

4|^V,
(cf.

_ PY - iM>

has suggested important developments


functions.
It should also be noticed that
!

6*1) in the theory of Bessel

(4)

' cos (z cos 0) sin2"

0d0 = 2

'
\

cos {z cos 0) sin1"

0d0

Jo

Jo

=2

f*

cosCssin^cos*1

Jo

^,
was obtained

and each of these expressions gives rise to a modified form of Poisson's integral.

An interesting application of Bessel's and


by Lommelf who multiplied the formula
, / ,m cos2n0= 2 (-)

Poisson's integrals

-o

4w8 {4n8 -28 }...{4n2 -(2ro-2)2} . -sin*"^ /a v.


.

(zm)\

by cos (z cos 0) and


(5)

integrated.
(

It thus follows that

J (z)-(-Y I

4 8 f4n-2r... (W-(2iii -*Y)

Jm (s)

2'31.

Bessel's investigation of Poisson's integral.


proof, that
is

The
Bessel J,

Jn (z)

is

equal to Poisson's integral, which was given by


it is

somewhat elaborate;

substantially as follows

It is seen

on differentiation that
sin2"
-1

-^ cos
I

cos (z cos 0)

9
8

,-

sin2*1
**

" "
1

sin (z cos 0) sin2"*8

=
|

(2w - 1) sin8"-8
made

- 2n sin " +
p. 340,

cos (* cos 0),

* Poisson actually
sinSH+1 6
;

the statement

but, as

he points out on

(p. 293) concerning the integral which contains odd powers may be replaced by even powers throughout

his aualysis.

f Studien Uber die BesseVscken Functional (Leipzig, 1868), p. 30. + Berliner Abh. 1824 [published 1826], pp. 3637. Jacobi, Journal filr Math. xv. (1836), p. 13, [Get. Math. Werke, vi. (1891), p. 102], when giving his proof ( 2*32) of Poisson's integral formula, objected to the artificial character of Bessel's demonstration. w. B. jr. 2

26

THEORY OF BESSEL FUNCTIONS


1,

[CHAP.

and hence, on integration, when n >


(2ra

-1)1

cos (z cos 0) sin8"" 2

6d0 - %n
^2

cos (z cos 0) sin3"

0d0

+ 2n + i J
If

r*

cos (* cos 0) sin2"-*" 8

0d0 = 0.

now we

write

rin + ^ra) /^
the last formula shews that
z<f>

008 <* cos 0) sin2n

s * <>

(n

- 1) - 2w0 (w) +
it is

*<-6

+ 1) = 0,

so that

(w)

and

Jn (z)

satisfy the

same recurrence formula.


evident that
(z),

But, by using Bessel's integral,

=J <f>(0)
<f>

(1)

^J
l r

cos (* cos 0) sin2

0d0 = -

-J
'

isin (s cos 0)1 sin

0d0

=and
so,

sm (z cos

0) cos

Odd

(z)

(z),

by induction from the recurrence formula, we have

when w

= 0,

1, 2, 3, ....

2*32.

Jacobus investigation of Poisson's integral.


of the direct transformation of Poisson's integral into Bessel's
this

The problem
integral

was successfully attacked by Jacobi*;


dM -i sin2M -i0

method

necessitates the use

of Jacobi's transformation formula

si^ = ^
We
shall
it

1.3.5. ..(2/i-l) . sinn *>


.

where

fi

cos

0.

simple direct proof of

assume this formula for the moment, and, since no seems to have been previously published, we shall

give an account of various proofs in 2*321

2323.

fi,

we observe that the vanish when fi = 1, it


If
zn j

first
is

/i2 ) M-i with respect to evident that, by n partial integrations, we have


1 derivates of (1
,

cos (z cos 0) sin-'1

Odd

= zn

cos (zfi) (1
.

/a

)"-* dfi

= (-) n
*

J_

COS foa
i

- \ mr)

-^ d. di
}
vi. (1801),

Journal fiir Math. xv. (1830), pp. 1213. [Ges. Math. Werke,
i.

pp. 101102.] See

also Journal de Math.

(1836), pp.

195196.

2*32, 2*321]
If

THE BESSEL COEFFICIENTS


Jacobi's formula, this
1
.

27

we now use

becomes
{Zfi

1.3.5...(2n-l)f If

cos

\n-jr)

.dsinnd dsinnd

dfi

=1

3 5
.

. .

(2n

- 1)
Jo

cos (z cos

- ^nir) cos n0d0

= 1.3.5...(2n-l)wJ n (5),
by
to

Jacobi's modification of 2'2 (8)

and

(9), since

cos^cos 6
is

\nir)

is

equal

( )***
2*321.

cos (z cos 6) or (-)^ n-1)

sin(^cos 9) according as n

even or odd; and

this establishes the transformation.


Proofs of Jacobus transformation.
it

Jacobi's proof of the transformation formula used in 2-32 consisted in deriving

as a special case of a formula due to Lacroix*; but the proof which Lacroix gave of his formula is open to objection in that it involves the use of infinite series to obtain

a result of an elementary character. A proof, based on the theory of linear differential equations, was discovered by Liouville, Journal de Math. vi. (1841), pp. 69 73; this proof will be given in 2*322. Two years after Liouville, an interesting symbolic proof was published by Boole, Camb. Math. Journal, in. (1843), pp. 216224. An elementary proof by induction was given by Grunert, Archio der Math, und Phys. iv. (1844), pp. 104 109. This proof consists in shewing that, if

<p- i(W) -
/"(* dQa --2/i9B -M(H-l) 5 J

then

2 n+1 = (l- M )-

B rfM ,

and that ( -

n~
)

'

1.3.5.;. (2

1) (sin nff)jn satisfies

the same recurrence formula.

Other proofs of this character have been given by Todhunter, Differential Calculus (London 1871), Ch. xxviil, and Crawford t, Proc. Edinburgh Math. Soc. xx. (1902), 15, but all these proofs involve complicated algebra. pp. 11

depending on the use of contour integration is due to Schlafli, Ann. di Mat. (2) The contour integrals are of the type used in establishing 202. Lagrange's expansion; and in 2*323 we shall give the modification of Schlafli's proof, in which the use of contour integrals is replaced by a use of Lagrange's expansion.
v.

A proof

(1873), pp. 201

To prove

Jacobi's formula, differentiate


t
<

by Leibniz' theorem, thus

i.l!5^-l) g^ 1 -^'
"

1+ '->- ,
\

=0

a"

m ~t"2J

=
and
*

m=0

2 (-)".C'!m + i(ainitf)* + (oo8itf)"-*- 1

=sin(2wx0),
this is the transformation required
J.

183. See also a note written by edition), pp. 182 Catalan in 1868, Mim. de la Soc. R. des Sci. de Liege, (2) in. (1885), pp. 312316. t Crawford attributes the formula to Rodrigttes, possibly in consequence of an incorrect statement by Frenet, Recueil d'Exercices (Paris, 1866), p. 93, that it is given in Rodrigues' dissertation,
Diff.
i.

TraiU du Calc.

(Paris, 1810,

2nd

Corresp. sur VEcole R. Poly technique, in. (18141816), pp. X I owe this proof to Mr C. T. Preece.

361385.

28
2*322.

THEORY OF BESSEL FUNCTIONS


LiouvMds proof of Jacob?* transformation.
Liouville of Jacobi's formula is as follows
let

[CHAP.

II

The proof given by


Let y=(l fi2 )
n ~k

and

D be written for d/dp


n times ; and then

then obviously

(l-V)Z)y+(2n-l) My=0.
Differentiate this equation
(1

- fi2 )

Dtt+l

i/

but

(l"^-^-^^^-"^.
(^ +
D*~ 1 y= A
;

- (iD^+^D^-^^O

so that

Z)- y = 0.

Hence
where

sin nd +

BcoB n6,

A and B are constants since D y is obviously an odd function of 6, B is zero. To determine A compare the coefficients of 6 in the expansions of D*- 1 y and A sinnd in ascending powers of 6. The term involving 6 in D*~ l y is easily seen to be
n"1

n~ l

(-^)
so that

^" 1=( "


namely

"~ 1(2w " 1)(2w ~ 3)

-3

<9'

%J = (-)- 1 1.3.5...(2n-l),
result,

and thence we have the

d-sin2-i0

.1.3.5...(an-l)
v
'

dp*~ l
2*323.

SchlaJWs proof of Jacobi's transformation.


'

We first recall

Lagrange's expansion, which

is

that, if z=fi+hf(z), then

so that

(,)

| = *

*l{ fWjr if 0)],


<'(*)
), i.e.

subject to the usual conditions of convergence*.

Now
it

take

/(2) 3 -|(l-2*),

v/(l-*

8
),

being supposed that 0'

The

singularities of 2
,J{\

pansion of

22 )
de

when A-*-0. (2) reduces to ^(1 ja function of h are at A=>e *' ; and so, when 6 is real, the exin powers of h is convergent when both h and 2 are less than unity.
-/*2
to sin 6
| \

Now
and so

2={l- v'(l-2/iA+# )}/A,


J

Jfl

J)

(l^-*)*-(l^*)
is

/_\-i

Hence
pansion of

it

follows that

d -1 sin 2B_1 d
,

rr-r \ ; 2n_1 .(n-l)!

d/t"

. . -1

the coefficient of hn ~ l in the ex-

J(l

- 22 )
z

(62/8/*)

in

powers of

h.

But

it is

evident that

,n

^*

(l-^y-i-q-^-*)-* coefficient of A*

j?

1.3.5...(2n-l) e* - "" ...,

and a consideration of the


Jacobi's formula.

-1 in the last expression establishes the truth of

* Cf.

Modem

Analyst*, 7-32.

2-322-2*33]

THE BESSEL COEFFICIENTS


application of Jacobi s transformation.

29

233.

An

The formal expansion


/'(cos x)
Jo
in

cos

nxdx =

2
Jo t-0

(-)m am fi n+2m >(cosx)dx,

which am is the coefficient of tn+2m in the expansion of Jn (t)/J (t) in ascending powers of t, has been studied by Jacobi*. To establish it, integrate the expression on the left n times by parts it transforms ( 2*32) into
;

1.3.5...(2n-l) Sl
and,

f(n) (C S

*} 8inm * dx

>

when
1

sin2n #
f

is

replaced by a series of cosines of multiples of x, this becomes

XM ,

J.
(n,

2n

2n(n-l)

We

now

integrate

.(cosa?)cos2#, /<"> (cos x) cos

4a?,

...

by

parts,

and by

continual repetitions of this process,


of the type stated:

we

evidently arrive at a formal expansion

When

/(cos x)

is

a polynomial in cos
is

a?,

the process

obviously terminates and the transformation

certainly valid.

To determine
[

the values of the coefficients a^ in the expansion

5 (-)m am f in+sm) (cosx) dx f (cos x) cos nxdx=j JOw-0

thus obtained, write

/(cos x)
according as n
is

= (-)* cos (t cos x),

(-)*<-> sin (t cos x), 2*2 (8)

even or odd, and we deduce from

and

(9) that

Jn (t) = 2
so that am has the value stated.

(_)-,+

{(_), J%

It has been stated that the expansion is valid when /(cos x) is a polynomial in cos a;; it can, however, be established when /(cos x) is merely restricted to be an integral function of cos x, say

* 6w cos n a?

provided that lim %\ bn


Jo (t)

is less

than the smallest positive root of the equation


it

the investigation of this will not be given since

seems

to be of

no practical importance.
*

Journal filr Math. xv. (1836), pp.

2526

also Jacobi, Astr. Nach.

xxvm.

(1849), col.

[Qes. Math. Werke, vi. (1891), pp. 117118]. 94 [Ges. Math. Werke, vn. (1891), p. 174].

See

30
2*4.

THEORY OF BESSEL FUNCTIONS


The addition formula for the Bessel
coefficients possess
coefficients.

[CHAP.

II

The Bessel

an addition formula by which


z.

Jn (y + z)

expressed in terms of Bessel coefficients of y and which was first given by Neumann* and Lommelf, is
00

may be

This formula,

(1)

J(y

+ z) = 2 Jm (y)Jn-r,i(z).
m-oo

The simplest way


gives

of proving this result


1
0+)

is

from the formula 2*2 (4), which


~ vt
'

Jn {y

+ *) = 5^-.

f<

'~
po+)

M_1 * {y+*
oo

{t

dt

~
I i
00

X
oo

m -n- Jm (y)e*z 1
<t

1/t

>dt

no+)

= _i-.
analysis

Jm (y)\

p--i eft-Milt

2
wi -oo

Jm (y)Jn-m{z),
line of the

on changing the order of summation and integration in the third and this is the result to be established.
;

Numerous

generalisations of this expansion will be given in Chapter xi.

2'5.

Hansen's series of squares and products of Bessel

coefficients.

Special cases of
early as 1843.

The

Neumann's addition formula were given by HansenJ as first system of formulae is obtainable by squaring the
2*1 (1), so that

fundamental expansion
.-!/,

.j I- rJr (z))\ I-oo tm Jm (z)\.


[

r=

oo

Km-

By expressing

the product on the right as a Laurent series in t, and equating n the coefficient of tn in the result to the coefficient of t in the Laurent expansion of the expression on the
left,

we

find that

J(2s)
In particular, taking
(1)
?i

= Jr {z)Jn-*(z).
have

= 0, we
r=l

Jo (2s)

= Jo* (:) + 2S

(- )' Jr2 (*)

-
r=0

(-)' e r

Jr

(z).

* Theorie der BesseVschen Functionen (Leipzig, 1867), p. 40. f Studienilber die BesseVschen Functionen (Leipzig, 1868), pp. 26

27

see also Schlafli,

Math.

Ann. in. (1871), pp. 135137. + Ermittelung der absoluten Stiirungen (Gotha, 1843), p. 107 et seq. Hansen did not give (4), and be gave only the special case of (2) in which n = l. The more general formulae are dne to Loramel, Studien Uber die BesseVschen Functionen (Leipzig, 1868), p. 33.

For

brevity,

JH2 (z)

is

written in place of

{</

(z)) 2 .

2'4-2'6]

THE BESSEL COEFFICIENTS


the general formula

31

From
(2)

we

find that

Jn (2z) -S/f (#) Jn-r (*) + 2 (-Y Jt (*) Jn+r (z), r=0 r=l
Bessel coefficients of negative order are removed

when the

by using

2*1 (2).

Similarly, since
j

rjr (s)\\ 5 {-rv*J

ni

{z)\
1/t)}

= exp [z (t =1
>

exp [\z (-

1/t)}

it

follows that

(3)

J >(z) + 2% Jr*(z) = l,
r-0

(4)

2 (-y Jr (z) Jm_r (z) +

2X Jr (z) Jm+r (z) = 0.


r-1

Equation (4) is derived by considering the coefficient of tm in the Laurent expansion the result of considering the coefficient of tm+1 is nugatory.
;

A
(5)

very important consequence of

(3),

namely

that,

when x

is real,

\J,{*)\*1,

|^,(*)j<W2,

where r
2"6.

= 1,

2, 3, ...,

was noticed by Hansen.

Neumann's

integral for

Jn

(z).

It is evident

from 2*2 (5) that

Jn (z) = land so

I" eUn0-z amo) d0)

Jn2 (z)

1 = |-5

f*
j

f*
I

*>+*> e -(nfl+sin +,

dOdj).

To reduce this double by the equations


so that

integral to a single integral take

new

variables defined

8-<f>

= 2X

0+<f>

2yJr,

d(0,<f>)

2.

9 (x Vr )
It follows that

J 2 (*) =

*"""*

2^*11
is
y

""** 8in * cos *

d*;cty,

where the

field

of integration

the square for which


>
-

tt^x k^'7r "" < X + ^ ^ 7r


Since the integrand
is is

unaffected
yfr

if

both

% and

yfr

are increased by

tt,

or

if

increased by

gration

may

simultaneously decreased by ir, the evidently be taken to be the rectangle for which
ir is

while

field

of inte-

0<X^ 7r _ TT ^

l/r

7T.

32

THEOEY OF BESSEL FUNCTIONS


Hence
Jn
(*)
2"**- 2** sin

[CHAP,

- 2^-

f f*

* * cty dy

=If
result
(1

1 C"

if

Jo

J** (2* cos x)

<%
acute or obtuse,

we

replace

% by |tt T

0,

according as
t

is

we

obtain the

Jn*
may

W = -f
If J

Jm (2z sin 0) dd.

This formula
(2)

obviously be written in the form

Jn

(z)

= - f J*n (2z sin 0) d0,


IT

JO

which is the result actually given by Neumann*. It was derived by him by some elaborate transformations from the addition-theorem which will be given in H'2. The proof which has just been given is suggested by the proof of that addition- theorem which was published by Graf and Gublerf.

We obtain a different form of the integral with respect to x instead of with respect to yjr.
Jr? (s)
so that
i

if

we perform the integration This procedure gives


j***
<ty,

^f
r*
1 if

Jo (2* sin

yjr)

(3)

^""^"Srl
a

/o(2*sin^)cos2n^(fy

f*

(2z sin

i/r)

cos 2nyjr

dijr,

a result which SchlafliJ attributed to Neumann.

2'61.

Neumann's

series

a for J

(z).

taking the formula 26(1), expanding the Bessel coefficient on the right in powers of z and then integrating term-by-term, Neumann shewed

By

that

lf

(-)* zm+wt

sin 2n+^B,

JQ

_ ~

TO =

-) (2n

-f

2m) (^zYn+im
! '

m\{tn+m)\{(n + m)l\ %

Theorie der BesseVtchen Funetionen (Leipzig, 1867), p. 70. t Einleitung in die Theorie der BesteVtchen Funktionen, n. (Bern, 1900), pp. 81 85. of Neumann's treatise. J The formula is an immediate consequence of equation 16 on p. 69 was given, was first pub Math. Ann. in. (1871), p. 603. The memoir, in which 4his result lished in the Leipziger Berichte, mi. (1869), pp. 221256.


2*61, 2'7]

THE BESSEL COEFFICIENTS


Neumann
in the form

33

This result was written by

W
where

Jn{Z)

(n\f I

l(2n

l)

1.2.(2n

l)(2n

+ 2)

J'

1\

l = 2w + 2' 2m-

(2)

(2n+l)(2n + ~(2n + 2)(2w +


(2,.

3) 4)'

+ l)(2tt + 3)(2n + 5)
,

"(2n + 2)(2n + 4)(2n +

6)

This expansion

is

Schlafli) for the product of

a special case of a more general expansion (due to any two Bessel functions as a series of powers with
( 5*41).

comparatively simple coefficients

2*7.

Schlomilch's expansion of z m in

series

of Bessel

coefficients.

We shall now obtain the result which was foreshadowed in 2*22 conis any cerning the expansibility of zm in a series of Bessel coefficients, where = has already been given in 2*22 (7). positive integer. The result for

their expansions in powers of sins 6.

In the results 2*22(1) and (2) substitute for cos 2nd and sin(2n+ 1)0 These expansions are*

cos2n#= 2
,=o

n.(n + s -1)!
(-)'

wo\t (n-s)!(2s)!
,

(2 sin ey,

The
(

results of substitution are

cos (x sin 6)

- J, (x) + = 1 Jm+

2 ^1

J (x)
j

(-) ij&"^i (2 sin *)}


!

sin (x sin 9)

(X)

(-)

gj$| (* - )""(
series in sin 6

If

we rearrange the

series

that
(

it is

permissible to do

so),

on the right as power we have

(assuming

-/i\

irMLPvrnL? (-)"(2sin0)f 3 2n.(n + s-l)\


(~)*(2sin0) +1 5 X L -L
f

, n\ sin (z sin 6)=


" (2n

ro TT\i
Hobson,

r^-^r

l).(n +

s)!

J+i(*)

* Cf.

PZajie Trigonometry (1918), 80, 82.

34
If

THEORY OF BESSEL FUNCTIONS


we expand
find that

[CHAP.

II

the left-hand sides in powers of sin $ and equate coefficients,

we

l=Jo(s) + 2

n=l

5 J2n (4

(
I

'

"-.

Jn^7)\

Jm+1

(z >-

*' 2>

>

The first of these is the result already obtained bined into the single formula
(1)

the others

may be com-

(i .

I C + *.)( + r = w=0

. -Dl

Jm>

(m _ 1(

2|

3...)

m = 1, 2, 3, were given by Schlomilch* shewed how to obtain the general formula which was given explicitly some years later by Neumann f and LommelJ.
The
particular cases of (1) for which

He

also

permissible

The rearrangement of the double series now needs justification the rearrangement if we can establish the absolute convergence of the double series. If we make use of the inequalities
;

is

IA +1

(*)|<||^exr(iM*),
{z

^'<1,
-

<>.),

in connexion with the series for sin

sin 6) T

we
,

see that

I2sin4j*+i (2w

+ l) .(n+)!,

Nl

|2sin^p +

|i*|

fc+

,.
,

l2N

= sinh
and so the series of moduli a similar manner.
is

(|

2 sin

1)

exp (

2
|

),

convergent.

The

series for cos (z sin $)

may

be treated in

The somewhat elaborate analysis which has just been given is avoided Lommel's proof by induction, but this proof suffers from the fact that it
supposed that the form of the expansion
tion.
If,

in
is

is

known and merely needs

verifica-

following

Lommel, we assume that

i tor- n=0
* ZeitschriftfUr Math,

< + * t
>-iT
n. (1857), pp.

1 )'

jM ,(A

und Phys.

140141.

f Theorie der BesseVschen Functionen (Leipzig, 1867), p. 38. t Studien liber die BesseF?chen Functionen (Leipzig, 1868), pp. 35
in

36.

Lommel's investigation

given later in this section.

2*71]

THE BESSEL COEFFICIENTS


2'22 (8) in
.

35

[which has been proved in


/i
\,+,

the special case


!

m = 1],

we have

S (m + 2n) (m + n 1)
n=0
i
x.

n
\

r / 5 ((wi + n)! (ra + w 1)!) = \z m J + r _l m (z) + 2^ r


.

^i

* J~+ (*)

2
=0

(wi -Mi)!

,,

n\

_
Since (m

"
=0

(m+l + 2n).Q + n)
w! as

+ m)' ^i+sn (z)/n\-*-

-* oo

the rearrangement in the third

line of the analysis is permissible.

It is obvious

from this result that the

in-

duction holds for

m = 2,

3, 4,

An
Let
t

extremely elegant proof of the expansion, due to A. C. Dixon*,


t

is

as follows
2
,

be a complex variable and

let

u be denned by the equation u= 1


1

so that

when

describes a small circuit round the origin (inside the circle

= 1),

u does the same.

We then

have

wi!

/"

expi-ia^-l/O}---^*

^/

when we
n=0
calculate the

lw

-p<-^'-'/',
*

r
\

+2n)

-'

i!

-ir'

'--"
the inter-

(m +2n).(m + n-l)! . / m+2n * !


'

zh

sum
;

of the residues at the origin for the last integral

change of the order of summation and integration is permitted because the series converges uniformly on the contour and the required result is obtained.

Note.

When m

is zero,

-y-

has to be replaced by

2*71.

Schlomilch's expansions of the type

2n r JH

(z).

The formulae
(1)

1 (?nypJin {z)= 2

/^"S
i^?* -*
9
1
,

(2)

n=0
in

2 (2n+lJ*+i J2n+1 (*)- 2 m=0

which p

is

any positive integer

[zero included in (2) but not in (1)]

and 1*

is

a numeri-

cal coefficient, are evidently very closely connected with the results of 27.
* Messenger, xxxii. (1903), p. 8; a proof

The formulae
had been
pre-

viously given by Kapteyn,

on the same lines for the case Nieuw Archief voor Wiskunde, xx. (1893), p. 120.

=l

36

THEORY OF BESSEL FUNCTIONS


Zeitschrift

[CHAP.
(1857), p. 141,

II

were obtained by Schlomilch,


gave, as the value of P^ p \ m

fur Math, und Phys.

II.

and he

{)
is

w>

<* i-r-c*(-**r
2m

m ~A=o

.m!

where m Ck is a binomial coefficient and the last term of the summation is that for which k hm - 1 or \{rti 1). To prove the first formula, take the equation 2*22 (.1), differentiate
to
0,

2p times with respect


2

and then make 6 equal to


cos

zero.

It is thus found that

/ <

- ).<*.). *.

xn

/o

x*.

/x R**" w _[__

(* sin *)"! ;

j,. o =

V Z [^ Mi;

d *p

(-yfaiiPT]
(2m)
,

j,,,,

The terms

of the series for which

m>p,

when expanded

in ascending powers of

6,

contain no term in 0*, and so

it is sufficient

to evaluate

[_d6*>JLo

(2^Ti

J-o"

JL

(a*)l".iJoLM*l
P
fAz^ 2m 2m
!

(20""*

iJ-o

m=0 v * wl7

Jfc-0

= 2(-)l S
=o

jP-Zff,
and the end of the summation with respect to evident, and equation (2) is proved in a
which were stated by

since terms equidistant from the beginning


I-

are equal.

similar

The truth of equation manner from 2*22 (2).

(1) is

now

The reader
Schlomilch
4)

will easily establish the following special cases,

P J (*) + 33 Ja (i + 53 Jb (*) + = (+**), '2V2 ?)+4V4 W+eV6 ()+...=^,


x

...

2.3.473 (*) + 4.5.6./6 (*)+6.7.8.f7 (*)+...=$*3.

2*72.

Neumanns expansion of z as a seizes of squares of Bessel coefficients.


Schlomilch's expansion
it is

From

( 2'7)

of

as a series of Bessel coefficients

of even order,

easy to derive an expansion of z* as a series of squares of

Bessel coefficients,

by using Neumann's

integral given in 2*6.

Thus,

if

we take the expansion

am (2sin0)=
,

v 2

(2m
v

n-l)! + 2n).(2m '-A- +


we
find that

, /a J Wi+m (2zsm0),
.

-.

and integrate with respect to

0,

I?.? &>*- I
ir

(^ + 2)-(2m + n-l),

Jo

=o

so that
(I) (1)

(when

m > 0)
!

v (2m + 2n).(2m + r-l) a *.S2 "(2m)!*, .*!


M (**>

/-*<*>

2-72]

THE BESSEL COEFFICIENTS


An
alternative form
is

37

This result was given by Neumann*.


(2)

<^-S.L-ri^<*
when

and

this is true

m = 0,

for it

then reduces to Hansen's formula of

25

As

special cases,

we have

**=
(3)

~
tJ.
.

4=2
.

4n* (4

- 22) J* (*),
- 22 )
(4n* - 4*)
./* (*),

*6= Il-f-'-J I

4 5 b =3

4n* (4n2

If

we

differentiate (1), use 2*12 (2)


/i

and then rearrange,

it is

readily found that

tA\

uh- 1

.!(- 1)1

Z (2m + 2n-l).(2m+-2)!

ST

an expansion whose existence was indicated by Neumann.


* Leipziger Berichte, xxi. (1869), p. 226.

[Math. Ann. in. (1871), p. 585.]

CHAPTER
3*1.

III

BESSEL FUNCTIONS
The generalisation of
Bessel's differential equation.
II,

The

Bessel coefficients, which were discussed in Chapter


n,

are functions

of two variables, z and

of which z is

unrestricted but n has hitherto been


generalise these functions so as to

required to be an integer.

We

shall

now

have functions of two unrestricted (complex) variables.


This generalisation was effected by Lommel*, whose definition of a Bessel
function was effected by a generalisation of Poisson's integral
of his analysis he
;

in the course

shewed that the function, so defined, is a solution of the linear differential equation which is to be discussed in this section. Lommel's definition of the Bessel function Jv {z) of argument z and order v wasf

(*)

= rfr +ffirft)

cos {z cos 6) sin2 "

0d0

'

and the integral on the right is convergent for general complex values .of v for which R(v) exceeds |. Lommel apparently contemplated only real values of v, the extension to complex values being effected by HankelJ; functions of order less than - \ were defined by Lommel by means of an extension of the recurrence formulae of 212.

on comparing 3*3 with 1*6 that Plana and Poisson had investigated Bessel functions whose order is half of an odd integer nearly half a century before the publication of Lommel's treatise.
will observe,

The reader

We shall now replace the integer n which occurs in Bessel's differential equation by an unrestricted (real or complex) number v, and then define a Bessel function of order v to be a certain solution of this equation it is of
;

course desirable to select such a solution as reduces to

Jn (z) when

v assumes

the integral value

n.

We
(1)

shall therefore discuss solutions of the differential equation

z>*l
will

+z

^+

(z>- v>)

y = 0,
v.

which

be called Bessel's equation for functions of order

Functionen (Leipzig, 1868), p. 1. + Integrals resembling this (with v not necessarily an integer) were studied by Duhamel, Cours d Analyse, n. (Paris, 1840), pp. 118121.
%

* Studieji iiber die BesseVschen

Math. Ann.

i.

(1869), p. 469.
v,
ft.

Following Lommel, we use the symbols


being reserved for integers.

to denote unrestricted numbers, the symbols

is customary on the Continent, though it has not yet come into general use in this country. It has the obvious advantage of shewing at a glance whether a result is true for unrestricted functions or for functions of integral order only.

This distinction

3-1]

BESSEL FUNCTIONS

39

Let us now construct a solution of (1) which is valid near the oiigin; the form assumed for such a solution is a series of ascending powers of z, say

y= 2
m-0

+m cm z* ,
be determined, with the pro-

where the index o and the


viso that c is not zero.

coefficients c m are to

For brevity the


so that
< 2>

differential operator

which occurs in (1)

will

be called V,

w+s+*-*
is

It

easy to see that *

V, 2

cm z* +m

= 2

c w {(a

+ m)

-i/2 }s" +m

+ 2 cm z a+m+a

namely

The expression on the right reduces to the first term of the first series, c (a8 Vs) za if we choose the coefficients cm so that the coefficients of
,

corresponding powers of z in the two series on the right cancel.

This choice gives the system of equations


/(^{(a

+ l)*-* ca {(a + 2) -*, j+c c {(a + 3) -i/ + Cl


2
}

2 2

=0 =0 =0

(3)
c,{(a

+ ra) -v }+cM_2 =
2

If,

then, these equations are satisfied,


(4)

we have

2
m-0

cTO ^+'tt

=c

(aa -i/2 )2

a
.

From
values of

this result, it is evident that the postulated series can


if

be a solution

of (1) only
z.

=+

v; for c is not zero,

and za vanishes only

for exceptional

consider the mth. equation in the system (3) written in the form
cm (o

Now

when

w > 1.

It can

be

- v + m) (a + v + m) +
Cm^
negative integer, that
is

Cm-*

= 0,

and

so it determines cm in terms of

for all values of


is,

greater than 1

unless

a-i>ora + i/isa

integer (when

a= v)
(o

or unless 2v

is a negative a negative integer (when a v).

unless

2v

We
3*5),
*

disregard these exceptional values of v for the

moment
1,

(see 3" 11,


It

and then
When

+ m) p*
2

does not vanish

when

2, 3, ....

now

the constants a and c m have been determined by the following analysis, the series

is easily seen to be convergent and differentiable, so that the formal procedure actually produces a solution of the differential equation.

obtained by formal processes

>

40

THEORY OF BESSEL FUNCTIONS


d = c3 = c5 =
...

[CHAP.
and that

Ill

follows from the equations (3) that

= 0,

c 2Ml is ex-

pressible in terms of c
c"*

by the equation

(-r*
(a-v + 2)(a-v + 4,)...(a-v + 2m)(a+v +
of equations (3)
is

2)(a

The system
from (4) that
(5)
is

now

satisfied;

+ v + 4>)...(a + v + 2my and, if we take a = v, we see

c zv

1+ 2
-i

(-)" (**)""

(v

+ 1) {v + 2)
If

. . .

soli a formal solution of equation

(1).

we take

+ m) a= v, we
(v

obtain a second

formal solution
(6)
c

'z-"\i+ s
,-i
'

(-)m (^y

m\(- v + 1) (- v + 2)

...

(-v + m)]'

In the latter, c has been written in place of c because the procedure of obtaining (6) can evidently be carried out without reference to the existence of (5), so that the constants c and c ' are independent.
,

Any

values independent of z

may be

assigned to the constants

and

'

but, in view of the desirability of obtaining solutions reducible to v -* n, we define them by the formulae *
(')
i \ 2T(v+l)' and (6) may now be
i

Jn (z) when

co

= oirv..

2-T{-p + l)'
(-) w (i-g)~" +m

The

series (5)

written

|
,=o

(-)m (bz) v+2m

*
(

m\r(v + m +

l)'

m!r(-i/ + m + l)'

In the circumstances considered, namely when 2v is not an integer, these series of powers converge for all values of z, (z = excepted) and so term-by-term
differentiations are permissible.

The operations involved in the analysis f by which they were obtained are consequently legitimate, and so we have obtained two solutions of equation (1). The
first

of the two series defines a function called a Bessel function of


z,

order v and argument

of the
is

first kindX',

and the function


is

is

denoted by
(z).

the symbol

(z).

Since v

unrestricted (apart from the condition that, for

the present, 2v

is

not an integer), the second series

evidently J_

Accordingly, the function


(8)
It is evident

(z) is defined

by the equation

MZ) = ? m!I> + m+l)m


from
2*11 that this definition continues to hold

positive integer (zero included),

when v is a a Bessel function of integral order being

identical with a Bessel coefficient.


* For properties of the Gamma-ftmction, see Modern Analysis, ch. xn. t Which, up to the present, has been purely formal. X Functions of the second and third kinds are denned in 3*5, 3 54, 3-57,

3-6.

'

'

3*11]

BBSSBL FUNCTIONS
solution of Bessel's equation has been given

41

An interesting symbolic
form

by Cotter*

in the

[l+zv D- 1 z- 2
where Dsd/dz while

"- 1

D- 1 zv+1 ]- 1 (Azv +Bz- v


This

),

and

B are constants.

may be

derived by writing successively

[D(zD~2v)+z]zv y=0,

[zD-2p+D~ 1 z]z"y= -2vB, zD (z~ vy) + z~ 2v D~ V+V= - 1vBz- % \ x tv z- vy+D- z-^- x D- x z v+l y=A +Bz~
which gives Cotter's
3'11.
result.

Functions whose order

is

half of an odd integer.

two cases of Bessel's generalised equation were temporarily omitted from consideration, namely (i) when v is half of an odd integer, (ii) when v is an integerf It will now be shewn that case (i) may be included in the general
In
3*1,
.

theory for unrestricted values of

v.

When
where r
If
(1) ' v
is

v is half of

an odd

integer, let

S = (r + $r,
a positive integer or zero.

we

take a = r

in the analysis of 3*1,


(c 1

we find =0, 1 / ,v r! (cTO .m(m + 2r+l) + cm _ 2 = 0,


.l(2r + 2)
.

that

ci

(m>l) '
v

,,

and so
(2)
Ctm
is

( y q
2.4...(2m).(2r + 3)(2r + 5)...(2r +
27/i

+ l")
by r

which
If

the value of

c^ given by
c

3'1

when a and

v are replaced

+ \.

we

take
1

-2~r"+4r(r + f)'

we

obtain the solution


,=

m!r(r +

m + f)'
Jr+ ^(z),
so that the definition of

which

is

naturally denoted

by the symbol

31

(8) is still valid.

If,

however,

we take a = r
\cm

the equations which determine cm become

(S) (3)

foi.l(-2r)
before, c lt c3 ,
cv -i are all zero,

=0,

.m(m-l -2r) + Crn-* = 0.


Cgr+i

(m>1)
is

As
and
is

...

but the equation to determine c^+x


l

+ Czr
(-)w

0,
Car+r>

this equation is satisfied

by an arbitrary value of

when

ra

> r,

C2m+i

defined by the equation


.
27,1+1

=
(2r

-r c

2r

+i
.

+ 3) (2r + 5) ... (2m +

1)

...

(2m- 2r)

* Proc. R. Irish.

Acad. xxvn. (A), (1909), pp. 157161. f The cases combine to form the case in which 2v is an integer.

42
If

THEORY OF BESSEL FUNCTIONS

[CHAP.

Ill

Jv (z)

be defined by
2-"-*

3*1 (8)

when v r

\,

the solution

now

con-

structed is*
Co

T(\-r) /_,_! (z) + Czr+1 2*+* r (r + f) Jr+i (z).

It follows that

no modification in the definition of

Jv (z)

is

necessary
is

when

v=(r + );

the real peculiarity of the solution in this case

that the

negative root of the indicial equation gives rise to a series containing two
arbitrary constants, c

and

c^+i,

i.o.

to the general solution of the differential

equation.

3*12.

A fundamental system

of solutions of Bessel's equation.

It is well known that, if yx and ya are two solutions of a linear differential equation of the second order, and if y( and y2 denote their derivates with respect to the independent variable, then the solutions are linearly inde'

pendent

if

the Wronskian determinant^


y*

y* y*

yl
does not vanish identically; and
then, either one of the

if the Wronskian does vanish identically, two solutions vanishes identically, or else the ratio of

the two solutions


If the

is

a constant.
identically,

Wronskian does not vanish

then any solution of the

differential equation is expressible in the

form

cx

yx

+ d y2

where

cx

and

ca are

constants depending on the particular solution under consideration; the


solutions

y and y2 are then


x

said to form a fundamental system.


will

For brevity the Wronskian of yx and y%


W,{y,,y},
the former being used

be written in the forms

8K{yi,y.},

when

it is

necessary to specify the independent variable.

We now
If

proceed to evaluate

we

"multiply the equations

V v J_ v {z) = 0,
which

V v J,(z) = Q
results,

by J (z), J_ v (z) respectively and subtract the

we obtain an equation

may be

written in the form

[zm{j z j
* In
Sci. di Torino, xxvi. (1821), pp.

v {z),

j-^yi-q,
delle

connexion with series representing this solution, see Plana, Mem. della R. Aeead.

519538.

For references to theorems concerning Wronskians, see Encyr.lqpidie des


109.

( 23), p.

Sci. Math. n. lfr Proofs of the theorems quoted in the text are given by Forsyth, Treatise on

Differential Equations (1914), 72

74.

3*12]

BESSEL FUNCTIONS

43

and hence, on integration,

a)
where

m[j
G is
C,

v {z),

J-Az))=-,
z

a determinate constant.

To evaluate we have

we observe

that,

when

v is not

an

integer,

and

\z\ is

small,

"

{z)

l>TT)

{l

{z% J"' {z) =


/'_ (z);

m>5

W>
1 )}

with similar expressions for


J.

J-V (z) and


{

and hence

W^W- J-W: W-i r(^T(^)-r W r(-, +


2sini/7r
TTZ

-.
V

'

If

we compare
is

right which
(2)

this result with (1), it is evident that the expression on the 0(z) must vanish, and so*

m
is

[jv {z\ j_ v (z)}


is

= _ *!E^r.
TTZ

Since sini^r
/_ {z)

not zero (because v

not an integer), the functions

Jv (z),
(2),

form a fundamental system of solutions of equation


v is

31

(1).

When

an

integer, n,

we have seen

that,

with the definition of

21

and when v

is

made equal

to

in 3*1 (8),

we

find that

nK)
Since the
first

mZ

m\r(-n + m + Ty
the series
is

n terms of the

last series vanish,

easily reduced to

( )n Jn(z)y so that the two definitions of J-n(z) are equivalent, and the functions Jn (z), J_n (z) do not form a fundamental system of solutions of Bessel's equation for functions of order n. The determination of a fundamental
system jn this case will be investigated in

363.

Jv (z) is defined, for all values of v, by the and J (z), so defined, is always a solution of the equation Vt/ = 0. When v is not an integer, a fundamental system of solutions of this equation is formed by the functions Jy (z) and /_ (z).
To sum
up, the function

expansion of

31

(8);

generalisation of the Bessel function has been effected


Briefly,

researches on "basic numbers."

a basic number []

is

H. Jackson in his vn 1 defined as * , where pis


F.

by

the base, and the basic

Gamma

function

Tp

(u

+ l) = [u].rp (v).

rp

(v) is defined to satisfy

the recurrence formula

The

basic Bessel function is then defined

series for the Bessel function


*

by replacing the numbers which occur in the by basic numbers. It has been shewn that very many theorems
p. 104.

This result
z
-*
<x>

making

is due to Lommel, Math. Ann. iv. (1871), and using the approximate formulae which

He

derived the value of

by

will be investigated in

Chapter vn.

44

THEORY OF BESSEL FUNCTIONS

[CHAP.

Ill

concerning Bessel functions have their analogues in the theory of basic Bessel functions, but the discussion of these analogues is outside the scope of this work. Jackson's main
results are to

6572

be found in a series of papers, Proc. Edinburgh Math. Soc. xxi. (1903), pp. xxn. (1904), pp. 8085; Proc. Royal Soc. Edinburgh, xxv. (1904), pp. 273276; Trans. Royal Soc. Edinburgh, xli. (1905), pp. 128, 105118, 399-408; Proc. London Math. Soc. (2) I. (1904), pp. 361366; (2) II. (1905), pp. 192220; (2) ill. (1905), pp. 123.
;

The more obvious generalisation of the Bessel function, obtained by increasing the number of sets of factors in the denominators of the terms of the series, will be dealt with in 4 4. In connexion with this generalisation see Cailler, Mem. de la Soc. de Phys. de
#

Geneve, xxxiv. (1905), p. 354; another generalisation, in the shape of Bessel functions of two variables, has been dealt with by Whittaker, Math. Ann. lvil (1903), p. 351, and Peres,

Comptes Rendus, clxi. (1915), pp. 168

170.

3'13.

General properties of Jv (z).


series

The
closed

which defines

Jv (z) converges absolutely and uniformly* in


v.

domain

of values of z [the origin not being a point of the

any domain when

R (y) < 0], and in


For,

any bounded domain of values of

when

\v\

N and

\z\

% A, the test ratio for this series

is

-I*
m(v + m)
whenever

m (m A )
r
8

m is taken to be greater than the positive root of the equation


m'-miV-|A = 0.
z,

This choice of to being independent of v and


the test of Weierstrass.

the result stated follows from

being excepted)

an analytic function of z for all values ofz(z = possibly an analytic function of v for all values of v. An important consequence of this theorem is that term-by-term differentiations and integrations (with respect to z or v) of the series for Jv (z) are
Hence f

Jv (z)

is

and

it is

permissible.

An
(i)

inequality due to Nielsen J should be noticed here, namely

W=tvt)( 1+ >
|

where

and

+l

is

the smallest of the

{jJi]}v+2 numbers v + 1
|

6\

<exp

1'

1,

|,

1+3|,

....

This result may be proved in exactly the same way as 2'11 pared with the inequalities which will be given in 3*3.
Finally, the function
z",

(5)

it

should be com-

which

is

a factor of

Jv (z),

needs precise specifica-

* Bromwich, Theory of Infinite Series, 82. + Modern Analysis, % 5-3. % Math. Ann. lii. (1899), p. 230; Nyt Tidsskrift,
(1902), p. 494.

ix.

(1898), p. 73; see also

Math. Ann.

lv.

3-13, 3-2]
tion.

BESSEL FUNCTIONS
define
it to

45
is

We

be exp (v log z) where the phase (or argument) of z

given

its principal

value so that

ir < arg z
When
it is

ir.

necessary to "continue" the function


explicit

values of arg

z,

mention

will

Jv {z) outside this range of be made of the process to be carried out.


(z).

3*2.

The recurrence formulae for

Lommel's generalisations* of the recurrence formulae


efficients ( 2'12) are as follows:

for

the Bessel co-

(1)

^(fJ + ^W-v/rW z
J _ (z)-Jv+1 (z)=2J,'(z),
v
l

(2) (3)
(4)

zJv'(z) + vJv (z) = *J*-^)>


zJ: (z)

-vJ

{z)

= - zJv+1 (z).
2*12, for

These are of precisely the same fonn as the results of being the substitution of the unrestricted number v

the only difference


n.

the integer

To prove them, we observe

first

that

it V - A v

(-)j*h

**
7\.

_ \m ^2 v

+ai

Zo2"-

1+im

.m\r{v + m)

- z" Jv^ (z).


When we
differentiate out the product

on the

left,

we

at once obtain (3).

In like manner,

dz

i \

T rJ z-k

'WJ-rfj mt 2".i!r(v +

=1
oc

\\-A 5

(-) m z*m

l)

/_\r^2wi

2 K+2m-i

W _ 1)! T
\m+i n+i
(v

(l/

+W+

1)

/
.

+m+1 w = 2'

ml r

w + 2)

whence

(4)

is

obvious; and (2) and (1)


(4).

may be

obtained by adding and sub-

tracting (3)
*

and

when

StUdien Uber die BesseVschen Functionen (Leipzig, 1868), pp. 2, 6, 7. Formula (3) was given v is half of an odd integer by Plana, Mem. delta R. Accad. delie Sci. di Torino, xxvi. (1821),

p. 533.

THEORY OF BESS EL FUNCTIONS


[CHAP.
Ill

46

We
(5)

can

now obtain the

generalised formulae
>

(AJ [zvj
W

{z)]

= *-~ j- <*>'

(6)

(^)
obtained
all

ir-J.{*)} =(-)' z>"Jv+m (z)

by repeated

differentiations,

when

m is any positive integer.


3*1.

Lommel
integral

these results from his generalisation of Poisson's

which has been described in


(1) has

The formula

been extensively used* in the construction of Tables

of Bessel functions.

By
and
(7)

expressing

/_!

0) and

J_
x

(z) in

terms of J (z) and J' v


sin vtr
7TZ

(z)

by

(3)

(4),

we can

derive Lominel's formula f

J. (z) J^ v

/vtv + JL T-/\r/\2 = - J
(z)

(z)

v .,

(z)

from formula (2) of

312.
aud
(2) is that, if

An
(8)

interesting consequence of (1)

Q y () = J"2 (2),

then

<?v-,w-^ + i(^y;wi.
was discovered by Lommel, who derived various consequences of it, Studien Functionen (Leipzig, 1868), pp. 48 et seg. See also Neumann, Math.

this formula

iiber die Bessel'schen

Ann.

111.

(1871), p. 600.

321. Bessel functions of complex order.

and imaginary parts of the function J +M (x), where v, /* and x been discussed in some detail by Lommel J, and his results were have real, are subsequently extended by B6cher.

The

real

In particular, after denning the real functions


equation
1

VtH.{x)

and

v>ll

(x)

by the

Lommel
(1)

obtained the results

dx*

~ {K

Vi

*(a)

iS(x)}

+ {K Vth (x) iS(w)} + l*


dx
x
[Kv ^x) iS{*))
VttL

+ 2(fr)
(2)

- 0,

w+ltM (x)
.

(3)
* See, e.g.

S, +I>M (x)

= K {x) + K" (x), = 8VtH (x) + S"(x),


Vtll
.

Lommel, Milnchener Abh. xv. (18841886), pp. 644647. t Math. Ann. iv. (1871), p. 105. Some associated formulae are given
X Math. Ann. in. (1871), pp. 481486. Annals of Math. vi. (1892), pp. 137160. The reason for inserting the factor on the right
I!

in 3*63.

is

apparent from formulae which

will

be

established in 3-3.

3-21, 3-3]

BESSEL FUNCTIONS

47

with numerous other formulae of like character. These results seem to be of no great importance, and consequently we merely refer the reader to the memoirs in which they were published.

In the special case in which v = 0, Bessel's equation becomes

solutions of this equation in the form of series were given

by Boole* many

years ago.

33. Lommel's expression of Jv (z) by an integral o/Poisson's

type.

We
(1)
It

shall

now shew

that,

when R(v) > -

, then

Jw (z) = _^g.'_ JJcos (* cos 6) sm*>0d0.


was proved by
Poisson-f- that,

cluded), the expression


this

when 2v is a positive integer (zero inon the right is a solution of Bessel's equation and expression was adopted by Lommelj as the definition of Jv (z) for positive
;

values of v

+ \.
that the function, so denned,
is

Lommel subsequently proved


generalised equation

and that
(1).

it satisfies

the recurrence formulae of

a solution of Bessel's 32 and he then


; t

defined

Jv (z)

for values of v in the intervals

(-, -#),(_#, _), (J| -J),

...

by suc-

cessive applications of 3 - 2

To deduce

(1) from the, definition of


for

form the general term of the series


(-)rw (i*),+swt

Jv {z) adopted in this work, we Jv (z) in the following manner:


** r(y+j)r(m + i)

trans-

m!I> + m + l) I> + )r(*)-(2ro)f

(-y*Q*Y

I> + m + l)

provided that

R (v) > |. Now when R (v) > , the

series

l= i

(2wz)!

converges uniformly with respect to t throughout the interval (0, ), and so it 1 may be integrated term-by-term on adding to the result the term for which
;

* Phil.

Tram, of

the

Royal Soc. 1844,

p. 239.

See also a question set in the Mathematical

Tripos, 1894.
t Journal de VEcole R. Poly technique, xn. (cahier 19), (1823), pp. 300 et teq., 340 et $eq. Strictly speaking, Poiason shewed that, when 2v is an odd integer, the expression on the right

multiplied by Jz is a solution of the equation derived from Bessel's equation by the appropriate change of dependent variable.

t Studien

liber die

BetseVschen Functionen (Leipzig, 1868), pp. 1

et seq.

48

THEORY OF BESSEL FUNCTIONS


namely
/

[CHAP.

Iir

m = 0,

<"~*(1

t^dt,

which

is

convergent,

we

find

that,

when

whence the result stated follows by making the substitution t = sin2 6 and using the fact that the integrand is unaffected by writing ir 6 in place of 6-

When -&< R(v)<$,


more
elaborate.

The simplest procedure seems terms omitted and integrate by parts, thus
(2m)!

the analysis necessary to establish the last equation is a little to be to take the series with the first two

m _2

Jo

m=2>'

+i

(2m)!

./

Jov+iU
/

(2mj!

(1

'>

J*
j*

+**ui

(2m)i

(1

on integrating by parts a second time.


convergence of the
series.

The interchange

of the order of

summation and

integration in the second line of analysis is permissible on account of the uniformity of

On

adding the integrals corresponding to the terms

m0, m = 1

(which are convergent), we obtain the desired result.


It follows that,

when R{v) > \, then

J- <" "

r(5(i) />-* V ~ ** cos


(1 nffir-w ,C <'>"""'

I*

G - ' dt

Obvious transformations of this


<2 >

result, in addition to (1), are the following:

<*>

cos <*>

*
*

<3>

* <*> -

nMm fj
2

- ,r * cos (zt)

<5)

(z)

r(-

+i)r( C'

cos (z cos 9) sin" " d "'

<6 >

J " =

i>^f)Ta)J.
1

>0,,sin8

"^
namely

The formula obtained by a


(7)
is

partial integration of (5),

(*>

= ^TlVffe
^s

""^ cos ( cos ^) sin2


\.

^'

sometimes usefuiy^t

valid,

only

when R{v)>

3*31]

BESSEL FUNCTIONS
(4)

49
by Nielsen*

An expansion involving Bernoullian polynomials has been obtained from with the help of the expansion

f-*-(i-.-)i
'= in which
<f> n ()

rfl

'1

*" +,tf+1)
(n

+ 1)!

denotes the nth Bernoullian polynomial and aizt.

[Note.

Accad. delle

Integrals of the type (3) were studied before Poissou by Plana, Mem. delta R. Sci. di Torino, xxvi. (1821), pp. 519 538, and subsequently by Kummer,

Journal

Math, xn. (1834), pp. 14-4 147; Lobatto, Journal fiir Math. xvn. (1837), pp. 363371; and Duhamel, Cours d' Analyse, II. (Paris, 1840), pp. 118 121.
filr

A function, substantially equivalent to Jv


J{fjL,x)=\
J o
(1

(2),

denned by the equation


.

- v2 )* cos vx

dv,

was investigated by Lommel, Archiv der Math, und Phys. xxxvu. (1861), The converse problem of obtaining the differential equation satisfied by

pp.

349360.

zv

f*

{v-af- (v-P)"- 1 dv

was also discussed by Lommel, Archiv der Math, und Phys. XL. (1863), pp. 101126. In connexion with this integral see also Euler, Inst. Cole. Int. II. (Petersburg, 1769), 1036,

and

Petzval, Integration der linearen Differentialgleichungen (Vienna, 1851), p. 48.]

3*31.

Inequalities derived from Poissoris integral.

From
(i)

33 (6)

it

follows that, if v be real

and greater than

\,

then

!-/.<*)!

rj+ffrft-j

/' exp

'

'

sin "

ede

By

using the recurrence formulae

3-*2

(1)

and

(4),

we deduce

in a similar

manner that
(2)

|JpW

<TT

i^jl + jFi lggL


{2/(7rz)}* cos

}exp|/W|
J]

<-|<-i>,
(>-*).

(3)

i^wK^lJi + j-^y^i/wi
using the expressionf
z for

By

J^ (z)

it

may be shewn

that

(1) is valid

when

= \.

These inequalities should be compared with the less stringent inequalities obtained in 313. When v is complex, inequalities of a more complicated character can be obtained in the same manner, but they are of no great importance.
*

Math. Ann.

lix. (1904), p. 108.

The notation used

in the text is that given in

Modern

Analysis, 7*2; Nielsen uses a different notation.

t The reader should have no

difficulty in verifying this result.

formal proof of a more

general theorem will be given in 3*4.

*
50
3*32.

THEORY OF BESSEL FUNCTIONS


Gegenbauer's generalisation of Poisson's integral.
integral formula

[CHAP. HI

The
(1 >

~ - fRPp|^ />--- 9.CJ


is
;

(cos *)

rfft

in

which

Gn v (t)

the coefficient of a" in the expansion of (1

2at + o

-"
)

in

ascending powers of

R (v) > \

and n

is

the formula is valid when a, is due to Gegenbauer* any of the integers 0, 1, 2, .... When n = 0, it obviously

reduces to Poisson's integral.

In the special case in which v =


(2)

the integral assumes the fonn


z cos 9

Jn +

(z)

= (_i)

(ff/
we

Pn (cos 0) sin dd

this equation has

been the subject of detailed study by Whittakerf.


formula,
(

To prove Gegenbauer's
'

take Poisson's integral in the form


(1

'*

W - r(, + i'5*)r( )i>


t

- (=) ' + ""'

and integrate

times

by parts

the result

is

^
Now

,,()

(-2i)nr(,

+ ,,4-^ra)J_ 1
!

e
1

<**

pCn "(0,

it is

known thatj
(- 2) n T ( v + n +
$)

dn (l- t*)*+-l
whence we have

T (2i/)
(1

-)-*

(3)

J
t/

(2) (

- (-O -r(2,).7 ,!(^ p ^-r( v + i)r^)r(2 + n)J_


?

e
1

(1

I;

c w<.

and Gegenbauer's result

is

evident.
is

symbolic form of Gegenbauer's equation

this

was given by Rayleigh

in the special case v

=h
aid of the

The reader

will find it instructive to establish (3)

by induction with the

recurrence formula

* Wiener Sitzunasberichte, lxvii. See also Bauer, (2), (1873), p. 203; txx. (2), (1875), p. 15. Milnchener Sitzungsberichte, v. (1875), p. 262, and 0. A. Smith, Giornale di Mat. (2) xn. (1905), The function Cn " (t) has been extensively studied by Gegenbauer in a series of pp. 365 373.

memoirs

in the

Wiener Sitzungsberichte

some

of the

more important

results obtained by

him

are

given in Modern Analysis, 15'8. + Proc. London Math. Soc. xxxv. (1903), pp. 198-206.

See 6-17, 10

5.

t Cf. Modern Analysis, 15*8. Proc. London Math. Soc. iv. (1873), pp. 100, 263.

"

3*32, 3*33]

BESSEL FUNCTIONS
namely*
*"'

51

A formula which is a kind of converse of (4),


(5)

***
i*"* denotes
;

(7(^)) - r(^i^+V^ (P )
is

in

which

a generalised Legendre function,


is left to

due to Filon,

Phil.

Mag.

(6) vi.

(1903), p. 198

the proof of this formula

the reader.

3*33.

Gegenbauer's double integral of Poisson's type.

It has
(1 )

been shewn by Gegenbauerf that, when


v

R (v) > 0,
cos ^r)]
3-

Jv (r ) =
3

*
I
I

exp [iZ cos

tV (cos <f>cos0 + sin < sin


sin
2

""1

^ sin

0d^d0,

r = Z* + z* - 2 Zz cos < and Z, z, <f> are unrestricted (complex) variables. This result was originally obtained by Gegenbauer by applying elaborate integral transformations to certain addition formulae which will be discussed in Chapter xi. It is possible, however, to obtain the formula in a quite natural

where

manner by means of transformations of a type used


sphere J.
After noticing that,

in the

geometry of the

when

= 0,

the formula reduces to a result which

is

an obvious consequence of Poisson's integral, namely

Jv (Z) =
we proceed
unit sphere
this point

TT 1

%-r-r
and

(* eiz sin8"
as longitude

(V) Jo

f sin *"
8
.'

yjr dyjr,

to regard
;

yfr

and colatitude of a point on a

we denote
(I,

by

the direction-cosines of the vector from the centre to m, n) and the element of surface at the point by dto.

then transform Poisson's integral by making a cyclical interchange of the coordinate axes in the following manner
:

We

J* ()

e ^FJ-\ I"!*

iv *

*9

sin2 '

sin8-- 1 fd0dyfr

ttJ

{v))} m> o
[(

(W

= -^A
* It is supposed tbat

e%v sm 9 cos * cos2" _1 e sin

Ofy d0-

9*y _ Y{y + l)z v ->x ~ r("-M+i) dz*

t Wiener Sitzungsberichte, ucxiv. (2), (1877), pp. 128 129. t This method is effective in proving numerous formulae of which analytical proofs were given by Gegenbauer ; and it seems not unlikely that he discovered these formulae by the method
in question
;

cf.

12*12, 12-14.

The

device is used by Beltrami, Lombardo Reiidiconti, (2)

xiii.

(1880), p. 328, for

a father different purpose.

The symbol

which

jj i is positive.

m ^ means

that the integration extends over the surface of the hemisphere on

52

THEORY OF BESSEL FUNCTIONS


the integrand
is

[CHAP.

Ill

an integral periodic function of yjr. and so the limits of yfr may be taken to be a and a + Itt, where a is an arbitrary (complex) number. This follows from Cauchy's theorem.
integration with respect to

Now

We

thus get

(w)

(ktffY
-,

/'i"'/" a

+2w
e iv sin
e cos

/.

* cos2 " -1

sin Bdslr d$

TTl

(v)Jo J*
W
'

= -te K
irl

Wio
o

I *l
.'o

iw8i,l * C08 (* + >

cos2 ""1

sin ddylrdd.

We
so that

now

define a

by the pair of equations

r cos

= Z z cos
exp
[i

<f>,

r sin

z sin
cos
yfr

<f>,

(a)

(Aw)"
-

/*" f 2 "
1
I

L,

(Z z cos

<f>)

sin

iz sin
2

<f>

sin

t/t

sin 0]

con

'- 1

6 sin Bdyfrde.

The only

difference

between
v

this formula

and the formula


cos -f] cos 2
"- 1

Jv {m) =
is in

l, 7Tl (v) Jo

exp [iw
.'

sin

sin 0ety> d0

the form of the exponential factor

and we now retrace the steps of the

analysis with the modified form of the exponential factor.

When

the steps are

retraced the successive exponents are

i(Z z cos <f>)l iz sin


i(Z z cos
<f>)n
<f>)

(f>

m,
.

%{Zz cos
The
last expression is

cos

iz sin iz sin < cos


<f>

I,

yfr

sin

0.

iZ cos

iz (cos

<j>

cos

+ sin

<f>

sin

cos

\jr),

so that the result of retracing the steps is

(IwY
y,,

ff ex P [*^ cos

** ( C(>s

<}>

cos

+ sin

<f>

sin
2

cos

yfr)]
2"

sin

"" 1

^ sin

0d\frd0,

and consequently Gegenbauer's formula


[Note.
Sci.,

is

established.

The

device of using transformations of polar coordinates, after the

manner of
I'

this section, to evaluate definite integrals

seems to be due to Legendre, M4m. de


Sci. in. (1818), p. 126.]

Acad, des

1789, p. 372,

and Poisson, Mem. de FA cad. des

3*4.

The expression of </( M +d


shall

{z) infinite terms.

We

now deduce from

Poisson's integral the important theorem that,


in finite teivns

when v is half of an odd integer, the function Jv (z) is expressible by means of algebraic and trigonometrical functions of z.
It will
is

appear later
;

not so expressible

such a value, then ,TV (z) but of course this converse theorem is of a much more
(

474)

that,

when

v has not

recondite character than the theorem which

is

now about

to be proved.

3*4]
[Note. Solutions in
finite

BESSEL FUNCTIONS
terms of differential equations associated with
;

53

Jn+

(z)

were ob-

tained by various early writers

it

was observed by Euler, Misc. Taurinensia,

in.

(1762
;

1765), p. 76 that a solution of the equation for e i2

)t

(z) is

expressible in finite terms

while
Terns.

the equation satisfied by

2*

Jn+i

(.s)

was solved

in finite

terms by Laplace, Conn, des


;

1823 [1820], pp. 245257 and Mecanique Celeste, v. (Paris, 1825), pp. 8284 della R. Accad. delle Sci. di Torino, xxvi. (1821), pp. 533534; by Paoli, di

by Plana, Mem. Mem. di Mat. e

Fis. (Modena), xx. (1828), pp. 183188; and also by Stokes in 1850, Trans. Camh. Phil. Soc. ix. (1856), p. 187 [Math, and Phys. Papers, n. (1883), p. 356]. The investigation

which will now be given is based on the work of Lommel, Studien Fimctionen (Leipzig, 1868), pp. 51 56.]

liber die

BesseVsehen

It

is

convenient to restrict n to be a positive integer (zero included), and


3-3 (4),

then,

by

^(^ST.^d-^
n
!

dt

\Zv '_!

nly/ir

\_

r+x r=0 z
;

dtr

J_i'
t*)

+ times degree 2n, the process then terminates.


integrate by parts 2n
1

when we

since (1

n is a polynomial of

To simplify the last expression we observe that if dr (l -t2 n /dtr be cal) culated from Leibniz' theorem by writing (1 -t2 n = (1 - t)n (l + t) n the only ) term which does not vanish at the upper limit arises from differentiating n
,

times the factor (1 - t)n and therefore from differentiating the other factor r n times; so that wo need consider only the terms for which r^n.
,

Hence
and similarly

[tOfff]

=(-), 0.

-5^

[*&=!*j

_(_>-., Q, .

J^
/"_
v

It follows that

Um

* KZ)

Jir

rn

^+1 .(r-n)!(2n-r)!
2

+ (_yi+i e-i2 v V '


and hence
(1)
</ +i (s)

V+l Oan-r V L

-,
-

~\

,.-s' +1

.(r-w)!(2n-r)!

'

-J [>

*~"<n + r)l

(- t-r-n-x (n

+ r)

;-|

This result
(2)

may be

written in the form*

J^(.)-fiy[n(,+1W i )'S' -,<-<*2L_ 5 (2r)I(n-2r)i(2*)


r-c,

WL

cos^-i^r' _o
2

(-M* ++!)!
(2?is

1
'

+ l)!(n-2r-l)!(2^'+'

compact method of obtaining

this

formula

Soc. Sci. de Bruxelles, xxix. (1905), pp. 140

143.

given by de la Vallee Poussin, Ann. de la

J
54

THEORY OF BESSEL FUNCTIONS

[CHAP.

Ill

In particular we have
(3)

Jj (*)

= ()

sin

m,

(z)

(-)

-^ - cos z)
power

the former of these results

is

also obvious from the

series for J (z).

Again, from the recurrence formula we have


/* <*)

= (")n * n+i (^)" {*"* J* (*)i.


_.

and hence, from


6
iz

(1),

-n

-(n

+ r)l
*

(~i)r n .(n+r)\

,~ rl(n-r)\(2zy

rl(n-r)l(2zy

But, obviously, by induction

we can express
\zdz)
z

as a polynomial in \jz multiplied


oiz
'

by

iz
,

and
;

so

we must have

,_"o

(y-(~ + r)
r (n
!

- r)

(2*)

y.W<*

WW

V****.
*
'

for, if not,

the preceding identity would lead to a result of the form

where fa

(z)

and fa (z) are polynomials in


follows thatf
"

1/z;

and such an identity is obviously

impossible*.

Hence

it

r -o

r!(n-r)!(2^"

,=

r!

(w-r)! (2-) r

= (-r (2tt>
Consequently
(4)

*+>

{*-*

(
J

i)
(^).

j_j (*>i

= (-)(27r^./_n_
J
</_,

-J [> W _ v(te) M;a)= ^


j-* u

| 2

**.( + + r)n _____ I (-Y+.(n + ^ _ ^


r)i
e

rS(n

r)!(2

* Cf.

Hobson, Squaring the Circle (Cambridge, 1913),


the series

p. 51.

From

m-_r(j)
J,

(-r <**)
(

,to4-# ...(m-i)'

it is

obvious that

(e)

cos

z.

3*41]

BBSSEL FUNCTIONS

55

and hence
(5)

j^ w -(i)*r

00 .(,

+ j,?
*

77

yr^L!^_

v
'

r-o

(2r+l)!(>i-2r-l)!(2)2

+1
_

In particular, we have
(6)

mo- )'.* ^w-(i,y(-sf'-.)


have now expressed in
finite terras any Bessel function, whose order is by means of algebraic and trigonometrical functions.

We

half of an odd integer,

The explicit expression of a number of these functions can be written down from numerical results contained in a letter from Hermite to Gordan, Journal fur Math. lxvi.
(1873), pp.

303311.

3*41.

Notations for functions whose order

is

half of an odd integer.

Functions of the types /(n+j)(s) occur with such frequency in various branches of Mathematical Physics that various writers have found it desirable to denote them by a special functional symbol. Unfortunately no common notation has been agreed upon and none of the many existing notations can

be said to predominate over the others. Consequently, apart from the summary which will now be given, the notations in question will not be used in this work.
In his researches on vibrating spheres surrounded by a gas, Stokes, Phil. Trans, of the clviii. (1868), p. 451 {Math, and Phys. Papers, iv. (1904), p. 306], made use of

Royal Soc.
the series

"

n( + l ) 2.imr

(-!)( + !)

( + 2 ) " 2. 4. fair) 2

K "'

which

is

annihilated by the operator

d* dr2

d
dr

n(n + l)
r2

This series Stokes denoted by the symbol /(') and he wrote


>*

Sm e ~ *</ (r) + Sn
;

'

'*/.

" *'),
fa
is'

where

<S'

and

Sa

'

are zonal surface harmonics


rf2

so that

annihilated by the total

operator

dr1

+ 2 d 4-,a
r

M (" + 1 )
r2
'

dr

and by the

partial operator

'' l 2 cr*

+ -*-+., . - r- sin 6 " or r


, , .

2 d

v)
dff)

55 c6

\ {

sin 6 52}

+ m2

In this notation Stokes was followed by Rayleigh, Proc. London Math. Soc. iv. (,1873), 103, 253283, and again Proc. Royal Soc. i.xxii. (1903), pp. 4041 [Scientific v. (1912), pp. 112114], apart from the comparatively trivial change that Rayleigh would have written/,, fair) where Stokes wrote/, (/).
pp. 93 Papers,

'

56

THEORY OF BESSEL FUNCTIONS


In order to obtain a solution
finite at

[CHAP.
it

Ill

the origin, Rayleigh found

necessary to take

Sn = '

n+
)

Sn

in the course of his analysis,

and then

t>

ir

It follows

from

3-4 that

rn+1

i (ir\ Q?-' =
n [e- ir i +

/
( %

r \ n e~

ir
,

-4-

\ rcrj

and that

Jn+i {r) = -j^

fn (ir) + e-i--ifn {-ir)].

ir In order to have a simple notation for the combinations of the types e* /(tr) which write convenient to are required for solutions finite at the origin, Lamb found it

** ~ l ~ 2(2n + 3)
in his earlier papers, Proc.

+2

4.

(jSh

+ 3)

(2/i+5)~

"*'

(1884), pp.

pp.
pp.

London Math. Soc. xm. (1882), pp. 5166; 189 212; xv. 139149 xvi. (1885), pp. 2743 Phil. Trans, of the Royal Soc. clxxiv. (1883), 519549; and he was followed by Rayleigh, Proc. Royal Soc. lxxvii. A, (1906), 486499 [Scientific Papers, v. (1912), pp. 300312], and by Love* Proc. London
; ;

Math. Soc. xxx. (1899), pp. 308321. With this notation it is evident that

.=p^,=(-)-...(
!

!.

+ .).( i)'^.
Soc.

Subsequently, however,
ingly in his treatise on
pp. 11

Lamb

found

it

convenient to modify this notation, and accordalso Proc.

Hydrodynamics and

London Math.

xxxn.

(1901),

20, 120 150 he used the notation f *n W = 1.3.5... (2n + L ~ 2^+3) + 2


1

(d

l)

"" 4(2 + 3) (S + 5) ~ J

e~"

-j-\

=*n(*) -l'V'i(*)

so that

* n () =

>t+ ^ pr+i

"

, '

= vnw<A
()

to Y~

while Rayleigh, Phil. Trans, of the Royal Soc. cciii. A, (1904), pp. 87110 [Scientific Papers, Love, v. (t912) pp. 149161] found it convenient to replace the symbol /(*) by x*(*)Phil. Trans, of the Royal Soc. ccxv. A, (1915), p. 112 omitted the factor (-) and wrote

dye-**

/rf\"sin2

while yet another notation has been used by Sommerfeld, Ann. der Physik

und

Chemie, (4)

xxvm.

(1909), pp.

Chemie, (4) (1913), p. 191

665736, and two of his pupils, namely March, Ann. der Physik und xxxvu. (1912), p. 29 and Rybczyriski, Ann. der Physik und Chemie, (4) xli.
;

this notation is

^(.HW^-^
C(*)(MH'+i (*)
and
it is

(-ai)"T'
on
electric

+ (-)* tf--*(*>l
waves which was the

certainly the best adapted for the investigation

subject of their researches.


*

In this paper Love defined the function

E n (z)

as

- ) . 1

...

(2n - 1)

^
J

but, as

stated, he modified the definition in his later work.

t This is nearer the notation used


p.

82

except that Heine defined

^ n (z)

by Heine, Handbuch der Kugelfiinctionen, i. (Berlin, 1878), to be twice the expression on the right in his treatise, but

not in his memoir, Journal filr Math. lxix. (1869), pp. 128

141.

3*5]

BESSEL FUNCTIONS

57

Sommerfeld's notation is a slightly modified form of the notation used by L. Lorenz, who used vn and vn +( ) n iwn in place of >frH and fn see his memoir on reflexion and refraction
;

of light, K.
(1898),

Danske Videnskabernes Selskabs pp. 405502.]

Skrifter, (6) vi. (1890), [Oeuvres scientifiqucs,

I.

3*5.

second solution of BesseVs equation for functions of integral order.


( 3'1 2) that,

It has

been seen

whenever v

is

not an integer, a fundamental

system of solutions of Bessel's equation


the pair of functions

for functions of order v is

Jv {z)

and

./_(z).

When

is

formed by an integer (n), this is no

longer the case, on account of the relation


It
is

J_ n (z) = () n Jn (z)is

therefore necessary to obtain a solution of Bessel's equation which

linearly independent of
will give

Jn (z);

and the combination of this solution with

JH (z)

a fundamental system of solutions.


solution which will

The
the

now be

constructed was obtained by Hankel*;

full details

of the analysis involved in the construction were first published

by BOcherf.
method of constructing HankeFs solution was discovered by Forsyth method of Frobenius, Journal fiir Math, lxxvi. (1874), pp. 214235, for dealing with any linear differential equation. Forsyth's solution was contained in his lectures on differential equations delivered in Cambridge in 1894, and it
alternative
his procedure is based on the general

An

has since been published in his Theory of Differential Equations, iv. (Cambridge, 1902), 102, and in his Treatise on Differential Equations (London, 1903 and 1914), pp. 101

Chapter

vi.

note

1.

It is evident that, if v be unrestricted,

and

if

n be any integer

(positive,

negative or zero), the function

j {z)-(-yj_ v {z)
v

is

a solution of Bessel's equation for functions of order v

and

this function

vanishes

when

n.

Consequently, so long as v

n,
v

the function
v

J {z)-(-TJ_
v
is

(z )

n
;

also a solution of Bessel's equation for functions of order v

and this function

assumes an undetermined form \ when v

= n.

We
and we

shall

now evaluate
lim

^-(-y^-fr)

shall

shew that

it is

a solution of Bessel's equation for functions of

* Math. Ann. i. (1869), pp 469472. t Annals of Math. vi. (1892), pp. 85 90.

See also Niemoller, Zeitsehri/t fiir Math, und Phys.

xxv. (1880), pp.

65-71

X The essence of Hankel's investigation is the construction of an expression which satisfies the equation when v is not an integer, which assumes an undetermined form when v is equal to

the integer

n and which has a

limit

when

v-*~n.

w.

B. F.

58
order n and that

THEORY OF BESSEL FUNCTIONS


it is linearly

[CHAP.

Ill

independent of

Jn (z)

so that it

may be taken

to be the second solution required*. It is evident that

J, (*)

- (- V J- v (*) _ Jv () - Jn (M) v n v n
dv
l
;

)n
(

J- v (Z) - J-n (z) v n

du
.

as v

- n,

since both of the differential coefficients existf

Hence
^ w

n
n.

exists; it is called

a Bessel function of the second kind of order


it

To
we

distinguish
it it

from other functions which are also called functions of


Following Hankel,

the second kind


shall

denote

described as HankeVs function. by the symbol J TTn (z) so that

may be

(1)

YH (s) =
*.(*)now
to be

j
lim
v-*v

{z)-(-yj_ v ( Z )
v

and

also

(2)

dJA z ) dJA =dv

( K

Vt dJ"-,(fY
>

dv

It has

shewn that Y

(z) is

a solution of Bessel's equation.


v,

Since the two functions

J v (z)

are analytic functions of both z and

the

order of performing partial differentiations on


is

a matter of indifference .

Hence the

v (z) result of differentiating the pair of

with respect to z and v

equations

V v Jv {z) =
with respect to v
-

may be

written

d*

dJv (z)
dv

d bJv (z)
dz
dv

^ dJ v (z)
dv

_n

dz%

When we

combine the

results contained in this formula,

we

find that

dJv {z)
dv
*
K
}

dJ_ v (z)
dv

*2v{Jv

(z)-(-rJ-Az)}>
it is

The reader
3-1.

will realise that, given


is

a solution of a differential equation,

not obvious that

a limiting form of this solution

a solution of the corresponding limiting form of the equation.


z

+ See

It is

conventional to write differentiations with respect to

as total differential

coefficients while differentiations with respect to v are written as partial differential coefficient-.

Of course,

in

% The symbol function equal to


See, e.g.

many parts of the theory, variations in v are not contemplated. Yn (z), which was actually used by Hankel, is used in this work
ljir

to denote a

times Hankel's function

( 3*54).

Hobson, Function* of a Real Variable

(1921), 312, 313.

3-51]
so that

BESSEL FUNCTIONS

59

+ 2v{Jv (z)-(-yj_v (z)}.

Now make
functions of
v,

v --n.

All the expressions in the last equation are continuous

and so we have

where v

is

to

be made equal to n immediately after the differentiations with

respect to v have been performed.


(3)

We

have therefore proved that

so that

V n Y n (*) = 0, Yn (s) is a solution of BesseVs equation for functions of order n.


r* J J T-.(,)-lim 'W-<-> V + H ^_ n

It is to be noticed that

-W
'

=
whence
(4)

M ^

im

J-A*)-(-YJAz) fi + n
due
to

follows a result substantially

Lommel *,

Y_ n (*)=(-)Yn (s).

Again,

while, because

Jv (z) is a
dv

monogenic function of v at v

0,

we have

dJ. v (zy
J-0

^(-^Jv-O

dv

J"""'

and hence

it

follows that

A result equivalent to
3*51.

this

was given by Duhamelf as early as 1840.

The expansion of

Y
is

(z) in

an ascending

series.

Before considering the expansion of the general function

Y (z),

it is is

con-

venient to examine the function of order zero because the analysis

simpler

and the resulting expansion

more compact.

We

use the formula just obtained,

SUidien
1

iiber die

this result for

what

is

BesseVschen Functionen (Leipzig, 1868), p. 87. Lommel actually proved sometimes called Neumann's function of the second kind. See 3-58 (8).
n. (Paris, 1840), pp. 122

t Court

d Analyse,

124.

60

THEORY OF BESSEL FUNCTIONS


is

[CHAP. HI

and the result of term-by-term differentiation

(*)

= 2["I

^w-^'Tt
is

x ]

lo g

fa) ~ J- log r (v + m +

1) 11

=2 2
where
yfr

rra^ (log^-^(m + l)},


customary, the logarithmic derivate of the

denotes, as

Gamma-

function*.
Since

0< ^ (i + 1) < m
in.

when w=l,
is

2, 3, ...

the convergence of the series for YoG?)

may be

established by using D'Alembert's ratio-test for the series in which

^(m+1)

is

replaced by

The convergence

concerning analytic functions.

an immediate consequence of a general theorem See Modern Analysis, 5*3.


also

The
(1)

following forms of the expansion are to be noticed

Y.(*)-2Sl

'J**'
.

{log(^)-^(m + l)},
(s)

log (*)

- X

(3)

Y. () -2

fry

+ log ft*)} ./.(*)

yy * -2 JS^ yy
^

(m

+
1)J

I+ g +

1 + raj !

The reader

will observe that

iY.(*)
is

+ (log2- 7 )J.(*)
The expansion
of

a solution of Bessel's equation for functions of order zero.

this function is

Neumann,

This function was adopted as the canonical function of the second kind of order zero by Theorie der BesseVschen Functional (Leipzig, 1867), pp. 42 44; see 3 57. But the series was obtained as a solution of Bessel's equation, long before, by Euler \.

Eider's result in his

own notatiou

is

that the general solution of the equation

xx ?dy + xdxcy +gx*y Cx*

V~

X
n*

\JW X + 1.8.27'^
.

1.8.27.64^
tT 3
('

+et -

+A +a
*

1 ( \

- 3- r + JL- X2 _
nn
1
.

_L_
1.4. 9
C

4m'

g + 1.4.9.

.*

Ion8

- etc

Ix

nn

X+ 1.4n* X
Ch. xn.

9 f

+1

16

*'
integer,

Modern

Atialy-iis,

It is to be

remembered

that,

when

m is a positive

then

*U)=-7,
t Inst. Calc. Int.
11.

V'('

1 )=

+ 2+ - +-->
233235.
See also Acta Acad. Petrop.
v.

where 7 denotes Euler 's constant, 0-5772157


(1781) [published 1784], pars
1.

....

(Petersburg, 1709), 977, pp.

Mathematica, pp. 186190.

3*52]

BESSEL FUNCTIONS
He

61

where A and a are arbitrary constants. numerators in the first line


6 = 3.2-1.0, 548 = 9.100-16.22,

gave the following law to determine successive


100 = 7.22-9.6,
etc.

22=5.6-4.2,

3528 = 11.548-25. 100

w
<r

2(1
\1

+* 1+...+!) = ^, mj

m
(r

this law is evidently expressed by the formula

m+ i = (2m+l)<r-m

ii

m.1

3*52.

The expansion of

Yn (z)

in

an ascending

series

and

the definition

of

<*) We

shall

now obtain Hankel's* expansion


is

of the

more general function


3-5.]

fl

(^),

where n

any positive

integer.

[Cf.

equation (4) of

It is clear that

?^.<*>dv

m =()dv

(- )m (i*) H * w | \m\ T(i> + l)j


(

m+

w=0 m!r(i/
ra!(w

+ w+l)
l

6V2

n
/J
'

/;

OT =o

+ ra)!

6V2

rv
is

when

i/-*-w,

where

is

a positive integer. That

to say
i

a)
The
of

|^>] o"
L

J^=n

- {7+ k g(W) j;(.)- 5 (-ra^- |i -om!(n + m)! (1


m

_li ^ + n+wj

[3/_(.s)/3i/],,_ n is a little more tedious because of the pole (- v + m + 1 ) at v = w in the terms for which m = 0, 1. 2, ., n 1. We break the series for /_ (z) into two parts, thus

evaluation of

yjr

" V

'

TO

l m!

r(-i/ +

m + l)

*! r(-i/

+m+

l)'

and in the former part we replace


1
,

V(-v + m+l)
Now, when
[~
t

by

r (v in) sin (v m) it *
1

'

^
"

ra

<

n,

(^)~'" m r (y - wt) sin (v - m) tt

= [(J*)~ r+m r(i-m) {tt-i ^r m) sin (> ra)7r + cos (v m) tr tt' - (*)-+ r (n - m) cos (n - m) tt.
(i

log ($z) sin (i/ )}]_

Math. Ann.

i.

(1869), p. 471.

62

THEORY OF BESSEL FUNCTIONS


Hence
(-) n r(-TO)(|^)-" 4 * ,t

[CHAP. HI

pM*)1

v
+

,;.

J(-,+.) it- h <*') + <- + ' +i '


w =o

that

is

to say
( **>

()
L

dp

_U(1)

m!

+(
x
{

>

.t m!(n + m)!
(m+1)},

lo g(i*)-"f

when we

replace

m by n + m in the second series.


and
(2)

On combining

we have Hankel's
a

formula,

namely

w
=2
{

m=o

*!

m=0
w

w!(n + m)!
1}

{2 log ($z) - yfr (m + 1) - i/r (n + m + 1)}


1

7 + log g *)}/,(*)

-(^r
1

(n

~r,"

(h^
,

(-)'"(^)^ m =o m:(n + m)l


j
In the
first

|l

(12
1

,111
m
1

w + m)
} {

term (m =s 0) of the

last

summation, the expression in

is

11,

12
d

It is frequently convenient (following

Lommel*)

to write

(4)

%(*) =

-^

Z)

-Mz)logz,

so that
<5

>

^^-S J<.nXLi) ll082+n ^ m+1)];


v is a negative integer,

when

^ (s)

is

defined by the limit of the expression

on the right.

We
(6)

thus have

Yn(z) = 2JJz))ogz + %(z) + (-)$_


solution of
.v

7l

(z).

The complete

-v^ + ay=0 was given in the form of a series (part of which

contained a logarithmic factor) by Euler, Inst. Calc. Int. n. (Petersburg, 1769), 935, 936 solutions of this equation are
;

**

J (2aM),
x

#*

Yi (2aM).
;

Euler also gave


this equation are

(ibid.

937, 938) the complete solution of x% -r^ + ay =0

solutions of

**

J2 (4a* **),

(4a* .v*).

* Studicn iiber die Besscl'schen Fvnctionen (Leipzig, 1868), p. 77.

3-53, 3-54]
3-53.

BESSEL FUNCTIONS
definition

63

The

o/Y v

(z).

Hitherto the function of the second kind has been defined only when its order is an integer. The definition which was adopted by Hankel* for unrestricted values of v (integral values of

2v excepted)

is

(1)

(*)

= 2ne

JAz)C0S

- J

-' (z)
.

sin2i/7r

This definition fails both when v is an integer and when v is half of an odd integer, because of the vanishing of sin 2vtt. The failure is complete in
the latter case
;

but, in the former case, the function


is

is

defined by the limit

of the expression on the right and it

easy to reconcile this definition with

the definition of

3'5.

To prove

this statement, observe that

lim

(z)

ire"**

lim

-n J

(z) cos vtt

cos vrr sin vir

/_ (z)~\ n v
\

J z COSv7r (-)* lim \ "( ) v-n v+n L


v

J-AzY

= YM (z) + lim > r+n L = Y n (s),


and
so

~n

Jv (z)
J

we have proved that


limY(*) =

(2)
It is

Yn

(*).

now

of that equation,

when

(i)
:

Y (z), defined either by (1) or by the limiting form a solution of Bessel's equation for functions of order v both v has any value for which 2v is not an integer, and when (ii) v is an
evident that
is

integer

the latter result follows from equation (2) combined with 3*5 (3). function Y (z), defined in this way, is called a Bessel function of the second kind (of Hankel's type) of order v ; and the definition fails only when

The
is

+1

an integer.
The reader should be

careful to observe that, in spite of the change of form, the qua function of v, is continuous at v = n, except when z is zero; and, in fact, Jv {z) and Y(z) approach their limits J (z) and Y (z), as v-*~n, uniformly with n respect to z, except in the neighbourhood of z = 0, where n is any integer, positive or negative.

Note.

function

Y (z),

3*54.

The Weber-Schlafli function of the second kind.


definition of the function of the second kind

The
Hankel

which was given by

( 3*53) was modified slightly by Weberf and Schlafli} in order to avoid the inconveniences produced by the failure of the definition when the

order of the function


*

is

half of an odd integer.

Math. Ann.

i.

(1869), p. 472.

f Journal fur Math, lxxvi. (1873), p. 9 ; Math. Ann. vi. (1873), p. 148. These papers are dated Sept. 1872 and Oct. 1872 respectively. In a paper written a few months before these, Journal filr Math. lxxv. (1873), pp. 75105, dated May 1872, Weber had used Neumann's

function of the second kind (see 3-57, 3-58). + Ann. di Mat. (2) vn. (1875), p. 17 ; this paper

is

dated Oct. 4, 1872.

64

THEORY OF BESSEL FUNCTIONS


The
function which was adopted by
is

[CHAP.

Ill

Weber

as the canonical function of the


first

second kind

expressible in terms of functions of the

kind by the formula*

Jv (z) cos vir J., (z )


sin vtt
(or the limit of this,
Schlafli,

when

v is

an

integer).

however, inserted a factor \nr\ and he denoted his function by

the symbol K, so that, with his definition,

<*\

- J 0) COS I/7T -</_(*)


"

Subsequent writers, however, have usually omitted this factor \tr, e.g. Graf and Gubler in their treatise f, and also Nielsen, so that these writers work with Weber's function.

The symbol
and so
to
it is

is,

however, used largely in this country, especially by

Physicists, to denote a completely different type of Bessel function ( 3'7),

advisable to use a different notation.


is to

The procedure which seems


Weber s function,

produce least confusion

use the symbol

Yv {z) to denote

after the

manner

of Nielsen J,

and to adopt

this as the canonical function of

the second kind, save in rare instances


integral order saves the insertion

when the use of the number tr in

of Hankel's function of

certain formulae.

We
(1)

thus have

Yv (z) =

"

(jOcgg

'" --*--(*>

= cos^;
Yn (0
.

(2)

Yn (z) =

lim
,-*. n

J M^^JLrl-A^ = I
Sin VTT
7T

Schlafli's fuuction has been used by B6cher, Annals of Math. vi. (1892), 8590, and by McMahon, Annals of Math. vm. (1894), pp. 5761; ix. (1895), pp. 23 30. Schafheitlin and Heaviside use Weber's function with the sign changed, so that the function which we (with Nielsen) denote by Yv (z) is written as Yv (z) by Schafheitlin and (when v = n) as GM (z) by Heaviside||.

[Note.

pp.

Gray and Mathews sometiinesIT use Weber's function, and they denote symbol Yn
.

it

by the

Weber's definition was by an integral

(see 6*1)

which
i.

is

equal to this expression

the

expression (with the factor \ir inserted) was actually given by Schlafii.
+ Einleitung in die Theorie der Bexsel'xchen Funktioiien,

(Bern, 1898), p. 34 et seq.

% Nielsen, as in the case of other functions, writes the

number
it

indicating the order as an


190-1), p. 11.

index, thus

Y v (z), Handbuch

der Theorie der Cylinderfuiilctionen (Leipzig,

There
used by

are obvious objections to such a notation,

and we
pp. 31

reserve

for the obsolete function

Neumann

( 3*58).

See, e.g.

Journal fUr Math. cxiv.

(181)5),

44, and other papers;

also Die Theorie der

BesseVschen Funktioiien (Leipzig, HK)8).


I!

Proc. lloyul Sue. liv. (1893), p. 138, and Electromagnetic Theory, n. (Loudon, 1899), p. 255;

a change in sign has been


11

made from his Electrical Papers, n. (London, 1892), TreatUe on Bessel Functions (London, 1895), pp. G5 66.

p. 445.

3-55]

BESSEL FUNCTIONS
in his later work, used

65

Lommel,
function

Neumann's function

of the second kind (see 3'57), but

in his Studien iiber die BesseVschen Functionen (Leipzig, 1868), pp. 85

86,

he used the

7rFM
where

(*)

+ {,/,(+4) + log2} Jn (z),


One disadvantage
of this function is that the

Yn (z)

is

the function of Weber.


yjr

for the function much more complicated; see Julius, Archives Nterlandaises, xxviii. (1895), pp. 221 225, in this connexion.]

presence of the term

(+) makes the recurrence formulae

3'55.

Heine's definition of the function of the second kind.

The definition given by Heine* of the function of the second kind possesses some advantages from the aspect of the theory of Legendre functions it enables certain generalisations of Mehler's formula ( 5*7 1), namely
;

lim
to be expressed in a

Pn (cos $/n) = J
The

(0),

'

compact form.
to

function,

the symbol
equal to

Kn

(z), is

expressible in terms of the canonical functions,

which Heine denoted by and it is

-%irYn (z) and

-\Yn {z);

the function consequently differs only

in sign from the function originally used

by

Schlafli.

The use of Heine's function seems to have died out on the Continent many years ago the function was occasionally used by Gray and Mathews in their treatise t, and they term it On (z). In this form the function has been extensively tabulated first by AldisJ: and
Airey, and subsequently in British Association Reports, 1913, 1914 and 1916.

This revival of the use of Heine's function seems distinctly unfortunate, both on account of the existing multiplicity of functions of the second kind and also on account of the fact (which will become more apparent in Chapters vi and vn) that the relations between the
functions
cosine

JH (z)
sine

and
;

YH (z) present many points of resemblance to the relations between the


||

and

so that the adoption

of

Jn (z)

and

On (z)

as canonical functions
It

is

com-

parable to the use of cos and

%ir sins as

canonical functions.

must

also be pointed

out that the symbol

On (z) has

two writers, namely Heaviside, Proc. Royal Soc. and Dougall, Proc. Edinburgh Math. Soc. xvm.
Note.

been used in senses other than that just explained by at least liv. (1893), p. 138 (as was stated in 3*54),
(1900), p. 36.

error in sign on p. 245 of Heine's treatise has been pointed out by Morton, Nature, lxiii. (1901), p. 29 ; the error is equivalent to a change in the sign of y in formula
3*51 (3) supra. It was also stated by Morton that this error had apparently been copied by various other writers, including (as had been previously noticed by GraylT) J. J. Thomson, Recent Researches in Electricity and Magnetism (Oxford, 1893), p. 263. A further error
i. (Berlin, 1878), pp. 185 248. Treatise on Bessel Functions (London, 1895), pp. 91, 147, 242. t Proc. Royal Soc. lxvi. (1900), pp. 32 43. Phil. Mag. (6) xxn. (1911), -pp. 658663.

An

Handbuch der Kugel/unctionen,

||

From

and also

for using

the historical point of view there is something to be said for using Hankel's function, Neumann's function ; but Heine's function, being more modern than either,
its

has not even this in

favour.

1 Nature,

xli?. (1894), p. 359.

66

THEORY OF BESSEL FUNCTIONS


;

[CHAP.

Ill

noticed by Morton in Thomson's work seems to be due to a most confusing notation employed by Heine for on p. 245 of his treatise Heine uses the symbol to denote the function called - \n J'q in this work, while on p. 248 the same symbol denotes - \n (Y -iJ ).

3*56.

Recurrence formulae for

(z)

and"Y v {z).

as those which are satisfied

The recurrence formulae which are satisfied by Yv (z) are of the same form by Jv (z) they are consequently as follows
;

(1)

Y ^(z)+Yv+1 (z)^Yv (z),


v

(2)
(3) (4)

Y ^(z)-Yv+1 (z) = 2Yv'(z),


v

zYv'(z) + v Yv (z) = zYv_ zYv'(z)-vYv (z)

(z),

zYv+1 (z),
replaced throughout by the

and in these formulae the function function Y.

may be

To prove them we take


z j
if
{*"

3*2 (3)

and

(4) in the

forms

(*)}

= *" J*-i (*)

{? Jj z
vrr,

(*)}

= - *' J-v+

>

(*)

we multiply

these by cot vir and cosec

and then subtract, we have

whence

(3) follows at once.

Equation (4)

is

derived in a similar manner from

the formulae

{*- J
By

(z)}

= - jr- Jv+1 (z),


J*-

{*- J. v

(z)}

= r- /__> (zy
(1).

addition and subtraction of (3) and (4)

we

obtain (2) and

The formulae are, so far, proved on the hypothesis that v is not an integer but since Yv (z) and its derivatives are continuous functions of v, the result of proceeding to the limit when v tends to an integral value w, is simply to
replace v

by

n.
ire rri

is

Again, the effect of multiplying the four equations by equal to 7re " 1,,ri sec {y l)7r, is to replace the functions
(

sec

vir,

which

Y by the

functions

Y throughout.
In the case of functions of integral order, these formulae were given by Lommel,
Studien
iiber die Bessel'schen

Functionen (Leipzig, 1868),


for such functions directly

p. 87.

The reader

will find it

instructive to establish

them

from the

series of 3-52.

Neumann's

investigation connected with the formula (4) will be discussed iu 3*58.

3*56, 3*57]
3'57.

BESSEL FUNCTIONS

67

Neumanns function of the second kind. The function which Neumann* adopted as the
it is

canonical function of the

second kind possesses the advantage that

represented more simply by

integrals of Poisson's type than the functions of the second kind which have

been hitherto discussed

but this

is its

only merit.

We

first

define the function of order zerof, which will be called

<0)

(z).

The second solution of Bessel's equation for functions of order known to contain logarithms, Neumann assumed as a solution the

zero being

expression

J
where

(z) log z

+ w,
,

is

a function of z to be determined.

If this expression is to be annihilated

by V we must V w = -V {J (*)log*} = -2zJ '(z).

have

But, by

212 (11),

and
so,

2z

(z)

2z J,

(s)

= 8 2 {-T~'nJ2n (*)
have

since

V Jm (z) 4>n2 JM (z), we


n=l

VoW = 2 2(-)'V 4(z)/n = 2V 2(-y-iJm (z)/n;


=i

the change of the order of the operations

2 and V

is

easily justified.

Hence a

possible value for

is

n-l

2(-y->Jm (z)/n,

and therefore Neumann's function


(1)
is

(0)

(z),

defined by the equation

F< \z)

=J

(z) log z

(-)""1

^-^

a solution of Bessel's equation for functions of order zero.


Since

/-*Oas+0, (the
it is

the origin),
solutions,

evident that

and hence

(z) is

series for w being an analytic function of z near J (z) and F (z) form a fundamental system of expressible as a linear combination of J (z) and
,0)

Yw (z); a comparison of the behaviours of the three functions near the origin shews that the relation connecting them is
(2)
*

F<> (*)

= J Y, (z) +

(log 2

- 7 ) Jo (*).

Theorie der Bessel'schen Functionen (Leipzig, 1867), pp. 42


( 9-1).

44.

Neumann calls

this function

Bessel's associated function,

and he describes another function, On (z), as the function


H
(z)

of the second
is

kind

But, because.

is

not a solution of Bessel's equation, this description

un-

and it has not survived. t Neumann's function is distinguished from the Weber- Schlafii function by the position of the suffix which indicates the order.
desirable

68
3*571.
It

THEORY OF BESSEL FUNCTIONS


The integral of Poissons type for

[CHAP.

Ill

{0)

(z).

was shewn by Poisson* that

h
is

eixcoa

<

log (x sin 2

<o)

dm
;

a solution of Bessel's equation for functions of order zero and argument x and subsequently Stokes obtained an expression of the integral in the form of an ascending series (see 3*572).

The

associated integral

f*

cos (z sin 0) log


.

(4iz

cos2 0)
,0 >

dd
;

was identified by Neumann f with the function F (z) and the analysis by which he obtained this result is of sufficient interest to be given here, with some slight modifications in matters of detail.

From

2-2 (9)

we have
2
/*

(-)M /*(*)

nir J

cos (z cos 0) cos 2n0d0,


o

and

so, if

changed,

we assume that the order we deduce that


I
2,

of

summation and integration can be

n=i

=7rJ

cos (z cos 0) 7
f*
I

z
W=1

do
d0

cos (z cos 0) log (4 sin3 0)


.

from this result combined with Parseval's integral F (z), we at once obtain the formula
,0

2 2) and the definition of

(1

F<> (*)

= !/"

"cos (z cos 0) log (4* sin3 0)


.

<*0,

from which Neumann's result

is

obvious.

The change of the order of summation and integration has now to be examined, because 1n~ l cos 2n0 is non-uniformly convergent near 6=0. To overcome this difficulty we observe that, since 2 ( - ) n J.in {z)jn is convergent, it follows from Abel's theorem that
%

S(-)^.W/= lim I (-)Ji<)/n= lim * j /*% os (s C os0)?Wcos2 ^ca?. =l o^-i-o n=l M -*M5Tn-U0
* Journal de. VEcole 2?. Poly technique, hi. (oahier 19), (1823), p. 476. The solution of an associated partial differential equation had been given earlier (ibid. See also Duhamel, p. 227). Cowr d? Analyse, n. (Paris, 1840), pp. 122124, and Spitzer, Zeitsehri/t filr Math, und Phys. n. (1857), pp. 165170. t Theorie der Bessel'sr.hen Functionen (Leipzig, 1867), pp. 45 49. See also Niemoller, Zeitschriftfllr Math, und Phys. xxv. (1880), pp. 6571.

I Cf. Bromwich, Theory of Infinite Series, 51.

3-571, 3*572]

BESSEL FUNCTIONS

69

Now, since a is less than 1, 2 (on cos 2n6)/n does converge uniformly throughout the range of integration (by comparison with 2a"), and so the interchange is permissible that is to say
;

=1

2 - 2

/"*""
/

n 2 / a .a cos2n8 . A eos(cos0) d6**>-

/"i*
I

Jo

cos(cos#) 2
lh"
/

a" cos 2nd

n-1

n dO
,

1 = --

n Jo

cos

cos d) log (1

- 2a cos 26 + a2

flW.

Hence we have

n-l

(")* (*'

=_

lim
o-*.l-0

i "

*""
|

cos

(a

cos d) log (1

- 2a cos 20

a2 )

<J9.

y o

We now proceed to shew that*


lim
/

cos(*cos0){log(l-2acos20-M2 )-log(4asin 2 0)}c?0 = O.


1

a-*-l-0 J

It is evident that

- 2a cos 26 +
log (1
j

- 4a sin2 6 ( 1 - a) 2 ^ 0,
a2 )
(z

and so
Hence,
'It
if

- 2a cos 26

^ log (4a sin 2 6).


|

A
(

be the upper bound t of cos


cos 6) {log
(1

cos 6)

when

0^6^ ,
\

we have

cos
I/,o

- 2a cos 26 + a2 ) - log (4a sin2 6)} dd

**

Jo

{log (1

- 2a cos 26+ a8 ) - log (4a sin2

6)}
,

d6
.

-A
i

/**"
\

~ ( |-22
I

a"cos20

+log (l/a>- 2

log (2 sin

6)\d6
J

io

*i

"

= J*iUog(l/a), term-by-term integration being permissible since


/"*
I

a< 1.

Hence, when
I

a< 1,
log (l/a)-*-0,

cos

(z

cos 0) {log (1

- 2a cos 26 + a2 ) - log (4a sin2 0)} dd < Jtt4

as a--l

0; and
2
n=j

this is the result to be proved.

Consequently

(-)w,7*(*) n

= _

I ( ** cos ii a-*i-o ,r y o

(*

CoS

tf)

log (4a sin2 6)

dd

-1

/"*"

cos

(z

cos 0) log (4 sin 2 8) dd,


.

and the interchange is finally justified. The reader will find it interesting to deduce combined with 3*5 (5).
3*572.

this result

from Poisson's integral for Jv (z)

Stoked series for the Poisson- Neumann integral.

The

differential equation considered

by Stokes \

in 1850

was -rf

+-

-r-

- m2 y =0, where

m is a constant. This is Bessel's equation for functions of order zero and argument imz. Stokes stated (presumably with reference to Poisson) that it was known that the general
solution was
/itr
o

{C+ D log (z sin 2 0)} cosh (mz cos 6)

d&.

The value

of this limit

was assumed by Neumann,

t If 2
p. 42.]

is real,

A l\

if

not,

A < exp

(z)
j

J Trans. Camb. Phil. Soc.

ix. (1856), p. [38].

[Mathematical and Physical Papers, in. (1901),

'

70
It is easy to see that,
is

THEORY OF BESSEL FUNCTIONS

[CHAP.

Ill

with Neumann's notation, the value of the expression on the right

\n {C- D log (4m)} J (imz)+\nD F() (imz).


The expression was expanded
bn (C + D log z)
into a series

by Stokes

it is

equal to

(imz)

+ 2D 2

m 2" z2n
,_
.,

f in
I

=o v* n )

cos2" 6 log sin 6 dd,


it

J o

and, by integrating by parts, Stokes obtained a recurrence formula from which

may

be

deduced that

-j*%o82^logsin^^=^^|^log2 + i
Neumann's

ff

(I

+ ^ + ... + i)}.

3*58.

definition of

(n)

(z).

The Bessel

function of the second kind, of integral order n, was denned

by

Neumann*
(1)

in terms of

<0)

(z)

by induction from the formula

dYi
z

- nY^

(z)

= - zYW (*),
is

which is a recurrence formula of the same type as 2*12(4). It from this equation that
(2)

evident

FW(jr)-(^)(^)*FW().
satisfies

Now Y m (z)

the equation

and, if

we apply the operator f

-j- to this equation n times,


ZCLZ

and use Leibniz'

theorem,
<3>

we get

'
*"

&T
is

''

{l)

+ (2M + 2) (a)"" 7 "


+ (2n + 2)
{z

(a)"
{Z)]

7 " (J) " *


= -

and so

{zlzf

^^

{Z)]

~ nYW

{zlz)

^^

{Z)

This equation
(4)

at once reducible to

V,Fw(5)-0,

and

so

(w)

(s) is

a solution of Bessel's equation


written in the form
>

for functions of order n.

Again, (3)

may be
l

zj{- z~n~ F(n+1


*
its

(*)}

- (2w + 2) z-*-

F<n+1

>

(*)

+ *~ n F(w

>

(2:)

= 0,
when

Theorie der BetteVschen Functionen (Leipzig, 1867), p. 51.


is

The function

is

undefined

order

not an integer.
is simplified

t The analysis

by taking \z* = f, so that d _ d zdz~ df

3*58, 3*581]
so that

BESSEL FUNCTIONS

71

dF(n+I >(*)
Tz

n+ 1 + -^t

F<+-> (z)

- F

(*)

= 0,

whence we obtain another recurrence formula


(5)

d7

(Z)

2
)

+ nYm (*) = zY-


we

(z).

When we combine
formulae
(6)

(1) with (5)

at once deduce the other recurrence

F'"- 1 (z)

F< +1 (z)
>

=
=

F<"> (),

(7)

F<"--> (*)

F<+ -> (*)

dF< "'^ as

Consequently

(w

(*) satisfies

the same recurrence formulae as

and

Yn

Jn (z),

F (s)

(s).

It follows from

357

(2) that

<

(8)

F (,) - *7rFn (,) + (log 2 - y) Jn (z)


= ^Y n (z) + (\og2-y)Jn (z).

solution of the equation V H (y) = in the form of a definite integral, which reduces to the integral of 3*571 when =0, has been constructed by Spitzer, Zeitschrift fiir Math, und Phys. ill. (1858), pp. 244-246; cf. 3-583.

3*581.

Neumann's expansion of

(n)

(z).

The
it is

generalisation of the formula

357 (1) has been given by Neumann*;

(1)

7(,)-/. W{kw .-^-jE

m-1
i

|_ -^
On ro-i
,

r /\

v. where

sn

= T + B + 5+
1

111
Z

,=i

m (n + m)
Ln (z)
/_\

...+-,

= 0.
define the functions

To

establish this result,

we

first

and

Un (z)

by

the equations
(2)

(3)

so that

- U (z). We shall prove that Ln (z) and Un (z) L^ (z) = - Zn (z) + (n/z) Ln (z), (4)
F<> (*)

=Z

,__ -ig. + ^()-^(-)+j "ff+ y-/"^^


Z.

W.J
(z)

n-l

9n-m-i

*>

.(.) log

satisfy the recurrence formulae

'

U^ (z) = - Un (z) + (n/z) Un


'

(z),

and then

(1) will

be evident by induction from

358

(2).

* Theorie der Bessel'schen Funetianen (Leipzig, 1867), p. 52.

die Bessel'schen Functionen (Leipzig, 1868), pp. 82

84

See also Lommel. Studien liber

Otti,

Bern Mittheilungen, 1898, pp. 34

and Haentzschel,

Zeitschrift fiir Math,

und Phys. xxxi.

(1886), pp. 25

35

33.


72

THEORY OF BESSEL FUNCTIONS


It is evident that
d_
[

[CHAP.

Ill

L n (z)
zn

\
)

dz dz\
1 _

d (JJs)) S dz\ zn

Mz) _
z n+1 n - r"- 1

2 -->.nl _d

m.
.!

...

Jn (z)

n z} 2

(n-m).w!^Vsn - M (.. .Jm (z) J +i(z)\l


j

Jm (z)
vl

z1

-74 _,7n+1 ^ )log ^ +


_
L n+1 Q)
zn
'

io~^r-|

1+ ^^ri| ^HriJ 2

and the

first

part of (4)
n

is

proved.

To prove the second


f

part,

we have
)

d [Un(z)]_ dz\ zn
)

d [Jn (z)) dz\ zn


1

TO =i

| (-y(n + 2m) d J^fr) m(n + m) dz\ z n


(z)

= - * -=- + 7n *

(z)

=i

~ -^ [mJr*^ (n +

( )m

- (n + m) J1l+M (z)}

*7n + l(*)

and the second part of


zn

(4) is proved.

It follows from

358 (2)

that
)

Y< (z) - L n+X (z) +

U^ {z) _

d[ Y m (,) - L n (z) + Un (z)


dz\
zn
is

j'

and since the expression on the right vanishes when n = 0, it induction that it vanishes for all integral values of n. Hence

evident by

Y(z) = L n (z)-Un (z),


and the truth of equation (1)
3*582.
is

therefore established.

The power series for

Un (z)>
3*581 (3) as

The function
coefficients, coefficients,

n (z),

which was defined in

has been expressed by namely

Schlafli* as a

a series of Bessel power series with simple

To
3*57 (1);

g-<'>-. s v.i(;v)- it is

and by straightforward differentiation, the expression on the right satisfies the same recurrence formula as that of 3*581 (4) for Un (z) equation (1) is then evident by induction.
establish this result, observe that

true

when n =

by

3*51 (3)

and

that,

Note.

It will be found interesting to establish this result

by evaluating the

coefficient

of () n+2m in the expansion


*

on the right of 3*581

(3).

Math. Ann. in. (1871), pp. 146147.

3*582-3*6]

BBSSBL FUNCTIONS
will

73

The reader
(2)
(3) (4)

now

easily prove the following formulae

3fn (*)
F> (m)

= {7 - log 2} Jn (z) - Un (z), = Ln (z) + 3 (s) + {log 2 - 7} Jn (z),


= Ln(*) + %*<*).
().

i*Yn{)
The
iivbegral

3 '583.

of PoUson's type for F<)

of 3-571 for F() (z) was generalised by Lommel, Sttidien ilber die BesseVschen Functionen (Leipzig, 1868), p. 86, with a notation rather different from Neumann's; to obtain Lommel's result in Neumann's notation, we first

The Poisson-Neumann formula

observe that, by differentiation of Poisson's integral for

Jv (*),

we have

^^-Jy(z)logz=
and
so,

^^^ f^

coa(z Sme)coH^B{log(ico^0)-ylr(u+i)}dB,

from 3*582

(3),

rW (>) "

r(n+l?r(i)
8

J**(*sin 0)co*** 6

{log

**-* (*+*)-?} d$+Ln (*),


formula

and hence, since


(l)

() \/r (1)
(

- 2 log 2 = - y - 2 log 2, we have the

jw () =
it is

r(w + ^r(i) j

** cos sin 6) cos2w * log (4 cos8 6) (*

de

in

which

coefficients

to be remembered that and powers of z.

Ln (z)

is

expressible as a finite combination of Bessel

3*6.

Functions of the third kind.

In numerous developments of the theory of Bessel functions, especially


those which are based on Hankel's researches (Chapters vi and
representations
tions of

vn) on integral and asymptotic expansions of Jv (z) and Yv (z), two combinaBessel functions, namely J, (z) iY, (z), are of frequent occurrence.
also present themselves in the theory of "Bessel functions
3*7).

The combinations

of purely imaginary argument" (


It has consequently

seemed desirable to Nielsen* to regard the pair of as standard solutions of Bessel's equation, and he describes them as functions of the third kind; and, in honour of Hankel, Nielsen denotes them by the symbol H. The two functions of the third kind are defined by the equations f
functions

Jv (z)iYv (z)

(1)

H*\z) = Jv {z) + iYv (z\


these definitions,

Hf{z) = Jv {z)-iYv {z).

From
'

combined with

354

(1),

we have

t9\

(!),.,
v
is

J- v {z)-e-Jv (z)

J. v (z)-e^Mz)
by
their limits.

When

an

integer, the right-hand sides are to be replaced

Since Jp {z) and F (z) satisfy the same recurrence formulae ( 3*2, 3'56), the functions enter linearly, and since the functions of the third kind which in
* Ofversigt over det K.
bitch

Danske Videnskabemes Sehkabs Forhandlinger,

1902,. p. 125.

Hand-

der Theorie der Cylinderfunktionen (Leipzig, 1904), p. 16.

t Nielsen uses the symbols

Hf(z)

Hv
2

(z).

74

THEORY OF BESSEL FUNCTIONS

[CHAP.
v

Ill

are linear functions (with constant coefficients) of

Jv (z) and Y
:

(z), it

follows that

these same recurrence formulae are satisfied by functions of the third kind.

Hence we can at once write down the following formulae


(3)

*.>)+*&<.)-**.

*?,+*-**(.),

(6)

.f^-^w.-^w
-&-- -<
Vv ff(z)=0,
(2) >

.f^'-^w-.dr*,.,,
-Jr- --<<>.
Vv ff(*)=0,

(7)

(8)

(^-{^.(^^. )-{^_ -^.


(

(1907), pp.

Note. Bayleigh on several occasions, e.#. PAt7. J/ogr. 350359 [Scientific Papers, iv. (1904), p. 290; v.

(5) xliii. (1897), p.

(1912), pp.

266 (6) xiv. 410418], has used the


;

symbol

Dn (z) to denote the function

which Nielsen

calls

\ triH

(2)

(s).

3*61.

Relations connecting the three kinds of Bessel functions.

It is easy to obtain the following set of formulae,

which express each


will observe

function in terms of functions of the other two kinds.


that
left.

The reader

some of the formulae are simply the

definitions of the functions

on the

(1)

/(*)

H\z) + Hf(z)
g

Y_ v {z)-Y v {z) COS VTT

W^
nnw
2t

'

J-' (jr)l
(3)

e JB (z) + e-" #?> CO


'

2
,/<*) cob

= F_ (#) cos tt Yv (#) #J


1}

'

(jr)

r- J_ r <*)

<*)

sm

" #f (*)

(5)

{x)

(z)
"

= J-M-c-iJriz)
t

= Y.
_

(z)-e-"- i
sin vir

(z)
'

sin vtt

(6)

JjWm /^'^" ^^)


"

r.,(*)-e-'r,,(*)
sini/7r

^sini/7r

From
(7)

(5)

and

(6) it is obvious that


(s),

H (z) = e- J5T

if

(,) . --

iff

(*).

3-61-3-63]
3*62.

BESSEL FUNCTIONS

75
ze mvi
.

Bessel functions with argument


is

z and
is

Since Bessel's equation


the functions

unaltered if z

replaced by

z,

we must expect
by

J v ( z)

to be solutions of the equation satisfied

Jv (z).

To avoid the slight difficulty produced by supposing that the phases of both of the complex variables z and z have their principal values*, we
shall construct Bessel functions of

argument

ze mni

where

is

any

integer,

arg z has

its principal value,

and

it is

supposed that

arg (zemni ) mnr


Since

+ arg z.
it is

Jv (z)/zv

is

definable as a one-valued function,

obviously con-

venient to assume that,


defined by the

when

the phase of z
as that

is

unrestricted,

same convention

by which

z" is defined;

Jv (z) is to be and accordingly

we have the equations


(1)

Jy (zemni - emviri J
)

(z),
v (z).

(2)

J_ v (ze

miri

e- m

J-

The

functions of the second and third kinds will

now be

defined for

all

values of the argument by means of the equations 3"54 (1), 36 (1); and then the construction of the following set of formulae is an easy matter:
(3)

(zemvi )
,Hiri

= e~ m

(z)

+ 2i sin mvrr cot vir Jv (z),


vir

(4)
(5)

r_ (ze

= e~ m

F_ (z) + 2i sin mvir cosec

J
'

(z),

m"i = e- m ir w H?\ze ) v "

(z)-2e-^~^^ JJz) sini>7r


v

= sin(l-w)i/7r ff(1)
sini/7r
"

'

_ e_

yit

sinmi/Tr

Rm
"

sinv7r
7r

(6) v '

Hf (zem) - a* H
v

(2)

"

(z) v
'

+ 2e J^T
8
"

sini/7r

Jy (z) vV '
(l)

= sin(l+t)^r H m (z) + evni sinmvTr ff


sinv7r
sinv7r
case

''

"

Of these results, (3) was given by Hankel, Math. Ann. vin. (1875), p. 454, in the special when m=\ and v is an integer. Formulae equivalent to (5) and (6) were obtained by Weber, Math. Ann. xxxvu. (1890), pp. 411, 412, when m=l see 6-11. And a memoir by Graf, Zeitschrift fur Math, und Phys. xxxvm. (1893), pp. 115120, contains the general
;

formulae.

3*63.

Fundamental systems of solutions of Bessel's


been seen
(

equation.

It has

3 12) that

solutions of Bessel's equation when,

(z) and J_ (z) form a fundamental system of and only when, v is not an integer. We shall

now examine the Wronskians


*

of other pairs of solutions with a view to detercritical case

mining fundamental systems in the


For Arg ( - z)

when

v is

an integer.

= Arg z =f t,

according as I (z)

0.

76

THEORY OF BESSEL FUNCTIONS


It is clear from 3"54(1) that

[CHAP,

III

[Jv (z),

(z)}

= - cosec vir WL [Jv (z), J.

(z)}

_2
TTz'

This result

is

established on the hypothesis that v is not

an integer; but con-

siderations of continuity
(1)

shew that

{J, (*), *,(#)}

-2/(,
and

whether v be an integer or mental system of solutions.


It is easy to (2)

not.

Hence

(z)

(z)

always form a funda-

deduce that

aa {j, (*),T

(#)}-

z cos vrr

and, in particular*,
(3)

m{Jn (z),Yn
found that

(z)}

= 2Jz.

When we

express the functions of the third kind in terms of J (z) and

v (z), it is

(4)

5129

{H (z),

hT
v.

(z)}

=-2im{J

(z),

Yv

(jr)}

= - 4i/(w),

so that the functions of the third kind also form a fundamental system of
solutions for all values of

Various formulae connected with (1) and (3) have been given by Basset, Proc. London Math. Soc. xxi. (1889), p. 55 they are readily obtainable by expressing successive differ' ' ential coefficients of Jv (z) and v (z) in terms of Jv (z), Jv (z) f and v (z), v (?) by re;

peated differentiations of Bessel's equation.

Basset's results (of

which the

earlier ones

are frequently required in physical problems) are expressed in the notation used in this work by the following formulae
(5)

Jv
JJ

(z)

IV
JV

(z)

Yv
17

(z)

Jv" Jv"

(z)

--

J.

(6)

()

()-

(,)

()-

A (l -^
(^-l),
x

(7)

Jv
j;

(z)

YJ" (z)-Yv {z)JJ" (.)-! iv"

(8)

(z)

w - JV
()

(z)

jv>"

()

A (^ _ y

(9)

JJ'iz)

JV"

Yv"{z) Jv'"

(.)-- (i

-^- + --^r-)
J,

(10)

Jv
j;

(z)

YvW{z)-Yv
jy>(*)- r;

(z)JvW(z) =
./,<">(*)=

J^

(i-^3),
+i)

(ii)

(,)

(,)

A (^i -

Throughout these formulae


are multiplied by

- sin vw

and J

Yv may be replaced by J_ v if the expressions on the right Yv may be replaced by W^ ET throughout if the
2i.
and Hankel, Math. Ann. vnr.
(1875). p. 457.

expressions on the right are multiplied by


* Cf.

Lommel, Math. Ann.

iv. (1871), p.

106,

3*V]

BESSEL FUNCTIONS
associated formula, due to

77
iv. (1871), p. 106,

An
12 )

Lommel* Math. Ann.

and Hankel,

Math. Ann.
(

vm.

(1875), p. 458, is

JAz)Yv ^{z)-Jv ^{z)Yv {z)=-~.


irz

This
3'7.

is

proved in the same way as

32

(7).

Bessel functions of purely imaginary argument.


differential equation

The

which differs from Bessel's equation only in the coefficient of y, is of frequent occurrence in problems of Mathematical Physics; in such problems, it is usually desirable to present the solution in a real form, and the fundamental systems

(iz)

and J_ (iz) or

Jv (iz)

and

(iz)

are unsuited for this purpose.

However the function e~ iviri J (iz)


of the equation.
(2)

It is

is a real function of z which is a solution customary to denote it by the symbol / (z) so that
oo

Iw(s)wg
is

y+2m <**)

J: 9 m\r( v +

m + iy

regarded as a complex variable, it is usually convenient to define its phase, not with reference to the principal value of argiz, as the consideration of the function Jv (iz) would suggest, but with reference to the principal value of arg z, so that
/ (*)

When z

= e-W Jv (ze^%

(- nr < arg z < tt),


(J

/ (*)

= e J (ze~^),

w < arg z < w).

The introduction
argument "
is

of the symbol

due

to Basset f

and

it is

Iv (z) to denote "the function of imaginary now in common use. It should be menbeen used by

tioned that four years before the publication of Basset's work, Nicolas} had suggested the use of the symbol v (z), but this notation has not

other writers.

The relative positions of Pure and Applied Mathematics on the Continent as compared with this country are remarkably illustrated by the fact that, in Nielsen's standard treatise , neither the function /(*), nor the second solution v {z\ which will be defined immediately, is even mentioned, in spite of their importance in

physical applications.
it is

The function I_ v (z) 312) that


(3)
#

is

also a solution, of (1),

and
VTT

easy to r prove

Ccf.

{/.(*X

/-(*)}

=-

28in
irz

Lommel gave the corresponding formula for Neumann's function of the second kind. f Proc. Camb. Phil. Soc. vi. (1989), p. 11. [This paper was first published in 1886.] Basset in this paper, defined the function of integral order to be i+ Jn (iz), but he subsequently changed it, in his Hydrodynamics, u. (Cambridge, 1888), p. 17, to that given in the text. The more recent definition is now universally used.
+ Ann. Sci. de Vticole norm. sup.

(2) xi. (1882),

supplement,

p. 17.

Handbuch der Theorie der Cylinderfunktionen

(Leipzig, 1904).

78
It follows that,

THEORY OF BBSSEL FUNCTIONS


when
v is not
(1).

[CHAP.

Ill

an integer, the functions Iv (z) and /_(*) form

a fundamental system of solutions of equation

In the case of functions of integral order, a second solution has to be con3*54. structed by the methods of 3'5

The

function

K
is

second solution,
<>

n (z), which will be adopted throughout defined by the equation

this

work as the

*.- ['-^](cf.

An

equivalent definition

3*5) is

It

may be
The

verified,
is

by the methods of
n.

3*5,

that

Kn (z)

is

a solution of (1)

when
v,

the order v

equal to

function

(z)

has been defined, for unrestricted values of

by

Macdonald*, by the equation


(6)

K
it

(Z)

= A-7T

and, with this definition,


(7) It is easy to
(8)

may be

verified that

#(*)=
deduce from (6) that
iT,(z)

lim

(z\

= i^'ir v

(iz)

= $>me-i"n H_ v ()-

fact that it is a v (z) lies in the physical importance of the function as z<x> through zero to exponentially tends which equation (1) solution of

The

positive values.

This fundamental property of the function will be established

in 723.

The definition of n (z) is due to Basset, Proc. Camb. Phil. Soc. vi. (1889), p. 11, and the infinite integrals by his definition is equivalent to that given by equations (4) and (5) ; 61 5. Basset subsein 614, discussed be will $ which he actually defined the function

(Cambridge, 1888), quently modified his definition of the function in his Hydrodynamics, n. rdl-y(z) 37, (*)1 1 , Ai equivalent to c pp. 1819, and his final definition is
.

^Ti^

'

cV~J v -n

In order to obtain a function which

satisfies

the same recurrence formulae as 7

(z),

Gray and Mathews


omit the factor

in their work,

Treatise on Bessel Functions (London, 1895), p. 67,


is

1/2", so

that their definition

equivalent to

'

2|_

ov

Cv

Jv= n

The only simple extension


formula

of this definition to functions of unrestricted order is

by the

K* (2) = \ir cot vrr{I- v (z) - 7 ()},


Proc.

London Math.

Soc. xxx. (1899), p. 167.

3*71]
(cf.

BESSEL FUNCTIONS

79

Modern Analysis, 1771) but this function suffers from the serious disadvantage that vanishes whenever 2 is an odd integer. Consequently in this work, Macdonald's function will be used although it has the disadvantage of not satisfying the same recurrence formulae as Iv (z).
it

An inspection of formula (8) shews that it would have been advantageous if a factor %n had been omitted from the definition of v (*) but in view of the existence of extensive tables of Macdonald's function it is now inadvisable to make the change, and the presence of the

factor is not so undesirable as the presence of the corresponding factor in Schlafli's function ( 354) because linear combinations of /() and v (z) are not of common occurrence.

3'71.

Formulae connected with Iv {z) and

(z).

We

shall

now

those constructed in

give various formulae for Iv (z) and v (z) analogous to 3'23*6 for the ordinary Bessel functions. The proofs

of the formulae are left to the reader.

(1)

/_, (,)

- J+1 (,) = ^ I, (z),


+ Iv+l (z) = 2/; (*),

K.
v

(z)

- Kv+l (z) = - ^ Kv (z),


+ Kv+1 (z) = - 2KJ (z), + vKv (z) = - z Kv . (z),
x

(2) (3) (4)

/_, (z)

K^

(z)

zi; (z) + vlv (z) zlj (z)

= zlv_

(z),

zKJ (z)

- vlv (z) = zlv+1 (z), = z^Iv_ m (z),


*'+
j
'

zK: (z) - vK v (z) = - zKv+1 (z),


(-*-)
[z

^L)

{z"Iv (.)}

*Kv (z)} =

{-r*K
>

(z),

\zdzj

z"

\zdz)

z"

]~

~^~'

(7)
(8)

/,'(,)/,(*),

K '(z)
K_
v

K,{z),

I_ n

(z)

= In (z),

{z)

=K

(z).

The
(9)

following integral formulae are valid only

when R(v +

\)

>

Iv (z) =

cosh (z cos 0)
'

sm^BdO
{zt) dt

- i>?|rr (i)

(1

" t2y

cosh

<**)'

,(1

^os s i n2 ,

dQ

r^r|^)/rcosh
"

{z cos d) sin2v0d0

-t7^fc)/.

<r4oo8hw&

'

80

THEORY OF BESSEL FUNCTIONS

[CHAP.

Ill

These results are due to Basset.

n0

(10)

r (A = I n+h (,)

-j^ [* X

We also have L_L*I (-)>+- r)i


o

rl(n

_ r)|(2 ,)r
>

+^
(11)

r!( W

-r)!(2^J'

7_ (w+i)

(s)

- ^^i) L^r r!(n-r)!(2^


+ (-)e-

^ r!(n ^

r)!(2(8)rJ

(12)

/f + , (.) =

r- 2

o;T

^-^
+ 2

(13)

*,-)'.-.
iTo (*)

(14)

- - log (*) 7
.

<*)

^
-

* <,+ 1),

\ n (15)

x (-)"(- m- 1)1 - ^ s' ^.W-jS, m!(w-m


eo

<")n+1

/1 2 \n+am
'

2
*
f

,
,

v.

{log (**)

W (m + - &
1)

(n +

m +1

)],

(16)

^o (*) = -J, (*""<)

e*

008 '

{log (2* sin2 0)

+ 7}

dO,

(17) (18) (19)


(20)

= emv7ri Iv (z),

(*"*) =
{!,(*),

e JT, (,) #(*)}


7 +1 (z)

tti

^^
=
1/z.

/, (,),

1/*,

J (*)

JT^ (,) +

(z)

The

integral involved in (16) has been discussed

by Stokes

(cf.

3"572).

The integrals involved in (9) and the series in (14) were discussed by Riemann in his memoir "Zur Theorie der Nobili'schen Farbenringe," Ann. der Physik und Chemie, (2) xcv.
(1855), pp. 130

139, in the special case in which


(2).

v=0; he

also discussed the ascending

power

series for 7

The recurrence formulae have been given by Basset, Proc. Camb. Phil. Soc. vi. (1889), 19; by Macdonald, Proc. London Math. Soc. xxix. (1899), pp. 110115; and by pp. 2

Aichi, Proc. Phys. Math. Soc. of Japan, (3)

11.

(1920), pp.

819.

Functions of this type whose order is half an odd integer, as in equations (10) and (12), were used by Hertz in his Berlin Dissertation, 1880 [Oes. Werke, 1. (1895), pp. 7791]

and he added ^et another notation

to those described in 3*41.

3-8]
3'8.

BESSEL FUNCTIONS
Thomson's functions ber
(z)

81
their generalisations.

and

bei (z)

and

which occurs in certain electrical problems consists of Bessel functions whose arguments have their phases equal to \tr or f ir.

A class of functions
The

functions of order zero were

first

examined by W. Thomson *

they

may be
(1)

defined

by the equation f
ber (x)

+ i bei (x) = J

(xi Ji)

=I

(x

^/i),

where x is real, and ber and bei denote real functions. For complex arguments we adopt the definitions expressed by the formulae
(2)

ber (z)

bei (z)

-J

(zi

Vi

*)

= h (* V

*')

Hence we have
(3)

berW=1 _|i| + |g_...,

Extensions of these definitions to functions of any order of the


third kinds have been effected

first,

second and

by

Russell:}:

and Whitehead.

The functions of the second kind of order zero were defined by Russell by a pair of equations resembling (2), the function IQ being replaced by the function thus

(5)

ker

(z)

kei (z)

=K

(zy/

i).

Functions of unrestricted order v were defined by Whitehead with reference to Bessel functions of the first and third kinds, thus
(6)
(7)

ber, (z)
her, (z)

i bei, (z) =

i hei (z)

Jv (ze** *),
1

= H (ze**).
(z),

It will be observed that|| (8)

ker (z) =

\tr hei

kei (z)

\ir her (z),

in consequence of 3'7 (8).

The
(9)

following series, due to Russell, are obtainable without difficulty

ker (*)

- log (z)

ber (z) +

fr bei

(z)

* Presidential Address to the Institute of Electrical Engineers, 1889. [Math, and Phys. Papers, in. (1890), p. 492.] f In the case of functions of zero order, it is customary to omit the suffix which indicates the order.

Mag. (6) xvii. (1909), pp. 524552. Quarterly Journal, xui. (1911), pp. 316 342. Integrals equal to ker (z) and kei (z) occur in a memoir by Hertz,
t Phil.

II

Ann. der Physik und Chemie,

(3)

xxn. (1884),

p.

450

\_Ges.

Werhe,

i.

(1895), p. 289].

82
(10)

THEORY OF BESSEL FUNCTIONS


kei (z)

[CHAP.

Ill

=-

log (\z)

bei (*)

- \ir ber (*)

It has also been observed by Russell that the ber2 ()+bei 2 () have simple coefficients, thus

first

few terms of the expansion of

but this result had previously been obtained, with a different notation, by Nielsen m 5-41) the coefficient of ($z)* in the expansion on the right is l/[(m !) 2 (2i) !].
;
.

(cf.

Numerous expansions involving squares and products


referred to Russell's

of the general
is

functions have been obtained by Russell; for such formulae the reader

memoir and

also to a paper

by Savidge*.

Formulae analogous to the results of 3*61, 3"62 have been discussed by Whitehead it is sufficient to quote the following here
;

(12)
(13)

ber_ (z)
bei_ {z)

cos vtr

ber (z)
bei (z)

sin vrr

[hei (z)

(14)
(15)

= cos vir her_(,z) = cos vir hei_ v (z) = sin vir

+ sin inr herv (z) sin vtr


her (z)

bei (z)], [her (z) ber (z)],


hei (z),

+ cos vir

hei v

(z).

The reader will be able to construct the recurrence formulae which have been worked out at length by Whitehead.

The

functions of order unity have recently been examined in some detail

by B. A. Smith f.
3*9.

The

definition

of cylinder functions.

Various writers, especially SonineJ and Nielsen, have studied the general
theory of analytic functions of two variables 9$ {z) which satisfy the pair of
recurrence formulae
(i)

rM(l)+H(') =
*_.<#) -tfH.(*) =

7rW,
W(*).

(2)

which z and v are unrestricted complex variables. These recurrence formulae are satisfied by each of the three kinds of Bessel functions.
in

Functions which satisfy only one of the two formulae are also discussed by Sonine in his elaborate memoir a brief account of his researches will be given in Chapter x.
;

* Phil.

Mag.

(6)

xk.

(1910), pp.

4958.

f Proc. American Soc. of Civil Engineer*, xlvi. (1920), pp. 375

425.

t Math. Ann. xvi. (1880), pp. 180.

Handbuch

der Theorie der Cylinderfttnktitmen (Leipzig, 1904), pp.

1,

42

et seq.

3*9]

BESSEL FUNCTIONS
call

83

Following Sonine we shall

the formulae, a cylinder function.

It will

any function ^(z), which satisfies both of now be shewn that cylinder functions

are expressible in terms of Bessel functions.

When we
(3)
(4)

combine the formulae (1) and

(2),

we

find that

z^:{z) + v%\{z)

= z^^Az), mW (jr) - v<9w (z) = - z<#+1 (z),


we deduce
that
1

and

so, if

be written

for z (d/dz),

(5) (6)

+ v)<$v (z) = z'@v _ (z), <* - v) Wr (z) = - zVv+1 (z).


(*
(&

It follows that

I*) 3? (Z)

that

= (Sr - V) {ZK-L (*)} = ^(^-I/+l)^_ (^) = -*#,(*).


1

is

to say

Hence
where a and
6 are

^ (*) = a v Jv (z) +
independent of
z,

bv

Yv (z),
v.

though they may depend on

When

we

substitute in (3)

we
x

find that

ay Jy _
and
so,

(z)

+ fcX-i {z) s

<*_,

Jv _, (*) + 6_, !;_, (z),


z,

since /_! (z)/ F^_ 2 (^) is not independent of

we must have
and,

Hence ay and
(2) are satisfied.

bv

must be periodic functions of v with period unity


v, it is

conversely, if they are such functions of

easy to see that both (1) and

Hence the general


(8)

solution of (1)
(*)

and

(2) is

%
-er^v)

i () Jv (z)

cr2 ()

F, (z),

where
It

and

r(i/)

are arbitrary periodic functions of v with period unity.


is

may be
(9)

observed that an equivalent solution


S? (z)

= ., () JETr () + r4 () #,<*> (s).

difference equation, which is more general than (1), has been examined by Barnes, Messenger, xxxiv. (1905), pp. 52 71 ; in certain circumstances the solution is expressible by Bessel functions, though it usually involves hypergeometric functions.

is used by Nielsen to denote Jv (z), Yv (z), ffvW () more general functions discussed in this section. This procedure is in accordance with the principle laid down by Mittag-Leffler that it is, in general, undesirable to associate functions with the names of particular mathematicians. The name cylinder function is derived from the fact that normal solutions of Laplace's

Note.

The name

and

HW
V

cylinder function

(z)

as well as the

equation in cylindrical coordinates are

e**>Jm {kpf r
(cf.

sin

m<t>

4-8

and Modern Analysis,

18*5).

84 Some

THEORY OF BESSEL FUNCTIONS


writers* following Heine t

[CHAP.
call

Ill

who

called

Jn (z)

a Fourier- Dessel function,

Jn {z)

a Fourier function.

Although Bessel coefficients of any order were used long before the time of Bessel 1-3, 1*4), it seems desirable to associate Bessel's name with them, not only because it has become generally customary to do so, but also because of the great advance made by Bessel on the work of his predecessors in the invention of a simple and compact notation
(cf.

for the functions.

Bessel's

name was

associated with the functions by Jacobi, Journal fiir Math. xv.

(1836), p. 13 [Ges. Math. Werke, vi. (1891), p. 101].

que usus

in determinandis integralibus definitis exposuit

"Transcendentium 7 naturam variosill. Bessel in commentatione


fc

celeberrima."

A more recent controversy on the name to be applied to the functions is to be found in a series of letters in Nature, lx. (1899), pp. 101, 149, 174; lxxxi. (1909), p. 68.
* E.g. Nicolas, Ann. Sci. de f Journal fiir cylinder function.

VEcole norm. sup.


p. 128.

(2) xi. (1882),

supplement,
to be responsible for the

Math. lxix. (1868),

Heine also seems

term

CHAPTER IV
DIFFERENTIAL EQUATIONS
4*1.

Daniel Bernoulli 's solution of Riccati's equation.


solution given

The
(!)

by Bernoulli* of the equation

-^ = azn + by 2

consisted in shewing that

when

the index n has any of the values

0;

4,-4: &

J-2

12 --

is.-

while a and 6 have any constant values f, then the equation is soluble by algebraic, exponential and logarithmic functions. The values of n just given are comprised in the formula

means of

/9\ \*)

4m 2m

where

m is zero or a positive integer.


method
it is first

Bernoulli's

equation,

of solution is as follows If n be called the index of the proved that the general equation J of index n is transformable
:

into the general equation of index

N, where

(3)

N=- n + 1'
also proved that the general equation of index
v,

and

it is

is

transformable

into the general equation of index


(4)

where
n

=-

is

4.

The

Riccati equation of index zero

obviously integrable, because the

variables are separable.

Hence, by

(4),

the equation of index


is

- 4 is integrable.

Hence by (3), the equation of index tinued by using the transformations


set of soluble cases given above,

-1
(3)

integrable.

If this process be con-

and
is

and
(2).

it

(4) alternately, we arrive at the easy to see that these cases are

comprised in the general formula


*

Exercitationes

pp. 473

475.

quaedam muthematicae (Venice, 1724), pp. 7780 Acta Eruditorum, 1725, The notation used by Bernoulli has been slightly modified and in this analysis
; ;

is

not restricted to be an integer.


is

t It is assumed that neither a nor h be separable.


%

zero.

If either

were zero the variables would obviously

That

is,

the equation in which a

and

have arbitrary values.

86
4*11.

THEORY OF SESSEL FUNCTIONS


Daniel Bernoulli's transformations of Riccati's equation.

[CHAP. IV

Now
Take

that the outlines of Bernoulli's procedure have been indicated,

we

proceed to give the analysis by which the requisite transformations are effected.
4*1 (1) as the

standard equation of index n and

make the substitutions

[Note.
n.

= Z, y = + !-"*1

Yis

The substitutions
n+ 1

are possible because

not included

among
;

the values of
its

The

factor
is

presence

was not inserted by Bernoulli the effect of that the transformed equation is more simple than if it were omitted.]
in the denominator

The equation becomes


1

dY

-T7t Y*dZ

=+ T

Y*zn

'

that

is

where

N = - n/(n + 1)

and

this is the general equation of index

N.

Again in

41

(1)

make

the substitutions

The equation becomes

where

=n4
is

and

this is the general equation of index

v.

The

transformations described in 4'1 are therefore effected,


soluble in the cases stated.

and so the

equation

But

this procedure does not give the

solution in a compact form.

4*12.

The limiting form of Riccati's equation, with index

- 2.

When

Riccati's equation, the value to

41

(2),

the processes described in $4'1, 411 are continually applied to oo in which the index tends, when not soluble by is - 2. The equation with index - 2 is consequently

a finite

number

of transformations of the types hitherto under consideration.

To

solve the equation with index

- 2, namely

write y

= v/z,

and the equation becomes


z
-.-

dz

= a + v + bv2

unci this is

an equation with the variables separable.


in this limiting case, Riccati's

Hence,

equation

is still

soluble by the use

of elementary functions.

4-114'13]

DIFFERENTIAL EQUATIONS
414) y = _ _ J?

87

This solution was implicitly given by Euler,


p. 185.

Inst. Cole. Int. n. (Petersburg, 1769), 933,


r,

If

we

write

(cf.

the equation which determines

is

dz^

z*

-U

>

which

is

homogeneous, and consequently

it is

immediately soluble.

Euler does not seem to mention the limiting case of Riccati's equation explicitly, although he gave both the solution of the homogeneous linear equation and the transformation which connects any equation of Riccati's type with a linear equation.
It will

appear subsequently
that
is

Riccati's equation is soluble in finite

examined
and

that the only cases in which terms are the cases which have now been to say, those in which the index has one of the values
(

4*7 4'75)

0;

-f, -|;

-f, -;

...,

_2

also the trivial cases in

which a or b

(or both) is zero.

This converse theorem, due to Liouville, is, of course, much more recondite than Bernoulli's theorem that the equation is soluble in the specified cases.

4'13.

Euler* s solution of Riccati's equation.

practical

soluble cases

method of constructing a solution of Riccati's equation in the was devised by Euler* and this method (with some slight changes

in notation), will

now be

explained.

First transform Riccati's equation, constants as follows


:

41

(1),

by taking new
2q

variables

and

(!)

= -v/b,
is

ab

= -c\

the transformed equation


(2)

cT

+r,2

~ &zn~^ =
is

'

and the soluble cases are those in which 1/q Define a new variable w by the equation
(3)

an odd

integer.

= c^-i + I<^, w dz
dw

so that the equation in


(4)
/A \

is _

-^ + 2cz*-> -^ + (q - 1 ) cz*--w = 0.
is
00

solution in series of the last equation

w = z-m-v 2 A r z~Qr
provided that'
Ar_+l

_ (2qr + q +

1) (2qr

+qpp.

1)

Ar
*

8qc(r
vm. (17601761)

l)

Nov. Comm. Acad. Petrop.

[1763],

363; and

ix.

(17621763)

[1764], pp.

154169.


88

THEORY OF BESSEL FUNCTIONS

[CHAP. IV

and so the series terminates with the term A m z~ qm if q has either of the values l/(2m + 1); and this procedure gives the solution* examined by Bernoulli.
The general solution of Riccati's equation, which is not obvious by this method, was given explicitly by Hargreave, Quarterly Journal, vn. (1866), pp. 256258, but Hargreave's form of the solution was unnecessarily complicated; two years later Cayley, Phil. Mag. (4)
xxxvi. (1868), pp. 348351 [Collected Papers, vn. (1894), pp. 912], gave the general solution in a form which closely resembles Euler's particular solution, the chief difference between the two solutions being the reversal of the order of the terms of the series involved.

Cayley used a slightly simpler form of the equation than (2), because he took constant multiples of both variables in Riccati's equation in such a way as to reduce it to
(5)

W-^~ f c

= 0.

4'14.

Cayley s general solution of Riccati's equation.


is

We

have just seen that Riccati's equation


dz

reducible to the form

given in
this

and we shall now explain Cayley 'sf method of solving equation, which is to be regarded as a canonical form of Riccati's

413

(2);

equation.

When we make
(1)

the substitution %

r)

= d (\ogv)/dz,

the equation becomes

*?_<-,*-.,,_<);

if U^ and Uz are a fundamental system of solutions of this equation, the general solution of the canonical form of Riccati's equation is

and,

where

d and C
z.

are arbitrary constants

and primes denote

differentiations with

respect to

To express

and

in a finite form,
v

we

write

= w exp

(cz q

/q),

so that the equation satisfied

by

is

4'13
sfl

(4).

solution of this equation

in

proceeding in ascending powers of


i

is

?-l
q(q-l)

czq+

^ Vx ^ ^q(q-l)2q(2q-l)
-^*_

(g-l) (3g-l) *
,

/,>

iq

_
and we take
*

(y-l)(3g-l )(og-l) q{q-l)2q(2q-l)Sq(Sq-l)


by
this series.

to

be exp

(czq/q) multiplied

the index n of the Riccati equation is - 2, equation (4) is homogeneous, Mag. (4) xxxvi. (1868), pp. 348351 [Collected Papers, vn. (-1894), pp. 912]. the memoirs by Euler which were cited in 4*13. cf. 1-1. X This is, of course, the substitution used in 1702 by James Bernoulli

When

t Phil.

Cf. also

4*14]

DIFFERENTIAL EQUATIONS
equation (1)
is

89
so

Now

unaffected

by changing the sign of c, and

we take

and both of these


integer.

series

?(0-l)2$(2gr-l) - l)(5i- 1)_ Z + + _il_- 1JJ3 ? q(q- l)2q(2q- l)Sq(Sq-lf -J' terminate when q is the reciprocal of an odd positive
"I

?(9-l)

Since the ratio

Ci

J72 is

the exponential function exp (2oz9 /<7)


;

multiplied by an algebraic function of

U U
lt

zv, it cannot be a constant form a fundamental system of solutions of (1).

and so
write

If q were the reciprocal of an odd negative integer, equation (1) in the form

we should

*-"W.)^(./.)-a
whence
it

follows that

d
where
lt

^ and y
2

are constants, and


afljq)
1

V F = z exp( +
The
series

-?-;.

czfl

+ l)(Sq + l) CV2 ,..] l)2q(2q + l) + q(q


(q
.-.

which have now been obtained

will

be examined in

detail in 4'4

much

greater

4'42.
following solutions of Riccati's

equation,

The reader should have no difficulty in constructing the when it is soluble in finite terms.
Equation

Values of

C72

(i)

(dr,/dz)

+ t,*=l

exp (z)

(ii)
(iii)

(dr,/dz)+Ti*=z-*i 3
{dr)ldz)

(l+3z l s )exv(3zW)
/

+ rP=Z-W

(1

+ W* + ^z2 6 ) exp ( + 5z
'

1/5

Equation

Values of

Vlt V%

(i)
(ii)

(dtj/dz)

+ i =z- i
rf

exp

1 /*)
1 /5

(dT,;dz)+q*=z-M
(dT)/dz)+t)Z=z-

z(l
z (1

+ 3Z- 1'3) exp ( 3z~ /3)

(iii)

+ 5-

+^-2/*) exp ( 5z-/s)

It is to

to terminate with the see

be noticed that the series Di, U^ (or Vu V2 as the case may be) are supposed term before the first term which has a zero factor in the numerator 4-42 and Glaisher, Phil. Trans, of the Royal Soc. clxxii. (1881), p. 773.
,

w.

B. F.

90

THEOBY OF BESSEL FUNCTIONS


Among
the writers

[CHAP. IV

who have studied equation (1) are Kummer, Journal fur Math. xn. 144147, Lobatto, Journal fur Math. xvn. (1837), pp. 363371, Glaisher (in the memoir to which reference has just been made), and Suchar, Bull, de la Soc. Math, de France, xxxn. (1904), pp. 103116; for other references see 43.
(1834), pp.

diately soluble;

when q0, the equation (1) is homogeneous and immeand that the second order equation solved by James Bernoulli ( 1*1) is obtainable by taking q2 in (1), and so it is not included among the soluble cases.
will observe that

The reader

4*15.

Schlafli's canonical form

of Riccati's equation.

The form
(i)

of Riccati's equation which was examined

by

Schlafli *

was

Zt
This
is easily

ift

~ 1ra~lu

*-

reduced to the form of

4*13(2) by taking

t~a/a

as a

new

independent variable.

To

solve the equation, Schlafli wrote

at

and arrived at the equation

Iff

F(a,t)=X
1=9

w!r(o + m + l)'
is
t).

the general solution of the equation in y

=c

F(a,

t)

+ c*t-aF(- a,

The

solution of (1)

is

then

Ct^Fia + 1,

t)

+ c2 F{- a - 1,
and

t)

^F^^ + cJ-HFi-a,*)
Bessel's equation is thus

The connexion between


rendered evident
;

Riccati's equation

to exhibit the connexion

but a somewhat tedious investigation is necessary ( 4*43) between Cayley's solution and Schlafli's solution.
<f>
:

Note.

The

function
,

z,

defined as the series


1

1+- + H2'
.

1 a3 a2 e(z+l)^2.3'z(z + l)(z+2y'"'

which

is evidently expressible in terms of Schlafli's function, was used by Legendre, Elements de Oeometrie (Paris, 1802), note 4, in the course of his proof that is irrational.

Later the function was studied (with a different notation) by fragment in his Math. Papers (London, 1882), pp. 346 349.

Clifford; see

a posthumous

Ann. di Mat.

(2)

i.

(1868), p. 232.

The reader

will see that

James

Bernoulli's solution in

series ( 1*1) is to

be associated with Schlafli's solution rather than with Cayley's solution, t This notation should be compared with the notation of 4-4.

4*15, 4*16]
It is obvious that

DIFFERENTIAL EQUATIONS
Jv (*) = (\zy F (v, - \z*),

91

and it has recently been suggested* that, because the Schlafli-Clifford notation simplifies the analysis in the discussion of certain problems on the stability of vertical wires under gravity, the standard notation for Bessel functions should be abandoned in favour of a
notation resembling the notation used by Schlafli-Clifford a procedure which seems comparable to a proposal to replace the ordinary tables of trigonometrical functions by tables
:

of the functions

-o(2)!' nZo(2n + l)\'

4*16.

Miscellaneous researches on Riccati's equation.

A solution of Riccati's equation, which involves definite integrals, was given by Murphy,
Trans. Camb. Phil. Soc. in. (1830), pp; 440

443.

The equation which he considered

is

and,

if

a be written

for l/(r+2)

and

A' d (log y)jdt


1

for u, his solution

(when

ABa2 =l)

is

y =4*
where
If 1/A

h~^[(f> (A)

exp

(*i//A)

+ < (1/A)

exp

(A*V)] dh,

<6(A)=e*A- ^

e~ h ha

- 1 dh= I

Jo

, n =oa(a

,. ^2)..,(a + n) + l)(a +\,


t
-,

for

be written for A in the second part of the integral, then the last expression given reduces to wit multiplied by the residue at the origin of h~* < (A) exp (* 1/0/A), and the
( 4*15) is evident.

connexion between Murphy's solution and Schlafli's solution

was published by Challis, Quarterly Journal, vn. (1866), pp. 51 which shewed how to connect two equations of the type of 4*13 (2), namely
investigation

An

53,

in one of which l/q is


integer.

an odd positive integer, and in the other it is an odd negative This investigation is to be associated with the discovery of the two types of

solution given in 4*14.

The equation
which
is easily

t- H

f-

bz" u 2

- czm 0,
i

new

variables,
it

transformed into an equation of Riccati's type by taking - + and z*u as was investigated by Rawson, Messenger, vn. (1878), pp. 69 72. He trans-

formed

into the equation

dv -jt

a+a

y+

bz m

-a

+a 2 y -cz" =Q,

by taking bu=czaly; two such equations are called cognate Riccati equations. A somewhat similar equation was reduced to Riccati's type by Brassinne, Journal de Math. xvi. (1851), pp. 255256.

The connexions between the various types of equations which different writers have adopted as canonical forms of Riccati's equation have been set out in a paper by Greenhill, Quarterly Journal, xvi. (1879), pp. 294 298.

* Greenhill, Engineering, cvh. (1919), p. 334 see also Engineering, cix. (1920), p. 851.

Phil.

Mag.

(6)

xxxvm.

(1919), pp.

501528

92
The reader should

THEORY OF BESSEL FUNCTIONS


also consult a short paper

[CHAP. IV

by Siacci, Napoli Rendiconti, (3) vn. a monograph on Riccati's equation, which apparently contains the majority of the results of this chapter, has been produced by Feldblum, Warschau
(1901), pp. 139

143.

And

Univ. iVach. 1898, nos.

5, 7,

and 1899,

no. 4.

4*2.

The generalised Riccati

equation.
is

An
(1)

obvious generalisation of the equation discussed in 4'1

^ = P + Qy + Rf,
;

where P, Q, R are any given functions of z. This equation was investigated by Euler*. It is supposed that neither P nor R is identically zero for, if
either
It

P or R is zero,

the equation

is easily

integrable by quadratures.
Sci.

special equation of this type

was pointed out by Enestrom, Encyclop4die des namely

Hath. n.

16, 10, p. 75,

that a

nxxdx nyydx-k-xxdyxydx
was studied by Manfredius, De constructions aequationum differentialum primi gradus (Bologna, 1707), p. 167. "Sed tamen haec eadem aequatio non apparet quomodo construibilis sit, neque enlni videmus quoinodd illam integremus, nee quomodo indeterminatas ab
invicem separemus."

The equation
order,

(1) is easily

reduced to the linear equation of the second

by taking a new dependent variable u defined by the equation f

(2)

--*--"
\r\

The equation then becomes d2 u /o\


Conversely,
/a\
if in

dR) du

the general linear equation of the second order,


_ du du ^__ + lh2 _ + p,o,
2

(4)

(where
(5)

p p lt p2
,

are given functions of

z),

we
,

write

u
is

= e^' dz

the equation to determine y

(6)

^^.fly-y., y
dz
is

which

of the

same type

as (1).

p** p The complete

'

equivalence of the generalised


is

Riccati equation with the linear equation of the second order


established.

consequently

The equations
1896, pp.
*
1

of this section have been examined by Anisimov, Warschau Univ. Nach.

33.

[Jahrbuch

iiber die Fortschritte

der Math. 1896, p. 256.]

Nov. Comvi. Acad, retrop. vin. (1760 1761) [1763], p. 32 ; see also a short paper by W. W. Johnson, Ami. of Math. in. (1887), pp. 112 115. t This is the generalisation of James Bernoulli's substitution ( 1-1). See also Euler, Inst.
Calc. Int. n. (Petersburg, 1769), 831, 852, pp. 88, 104.

4*2, 4*21]

DIFFERENTIAL EQUATIONS
Eider's theorems concerning the generalised Riccati equation.

93

4*21.

been shewn by Euler* that, if a particular solution of the is known, the general solution can be obtained by two quadratures if two particular solutions are known the general solution is obtainable by a single quadrature f. And it follows from theorems discovered by Weyr and Picard that, if three particular solutions are known, the general solution can be effected without a quadrature.
It has

generalised Riccati equation


;

To prove the

first result, let

y be a particular solution of

g-P+fc + JV.
and write y = y

+ 1/v. The
dv

equation in v

is

^ + (Q+2Ry )v + R =
first

0,

of which the solution


v

is

exp {f(Q

+ 2Ry ) dz) + jR exp {/(Q + 2%,) dz)


),

.dz

= 0,

and, since

v = l/(y y

the truth of the

theorem

is

manifest.

To prove

the second, let y and yx be two particular solutions, and write

y-yi
The
result of substituting (y x
y<>-yi

w y )/(w 1)

for

in the equation is
,

*?

d}h

(w-lfdz^w-ldz
and,

w-1 dz~
is

!_ dy*_ P >Q yw- .vo + p (yxw - y ** w-l V w - 1 )


1

when we

substitute for (dyjdz) and (dy /dz) the values

and

P + Qy

+ Ry*,

P + Qy + Ry*

the last equation

reduced to

wTz
so that

=Ry- Ry^
{f(Ry

w = c exp
y
is

Ry ) dz],
x

where

c is the constant of integration.

we
let

see that

Hence, from the equation defining w, expressed as a function involving a single quadrature.
third result, let y
let c'

To prove the
.

y% be a third solution, and reduce to y2 Then y

and y, be the solutions already specified, be the value to be assigned to c to make

yjrJf.o

=l
C
'

y*-y<>
y. 2

y-yi
and
this is the integral in a

-y

'

form

free

from quadratures.

* Nov.

Comm. Acad.

Petrop. vni. (17601761) [1763], p. 32.

t Ibid. p. 59, and ix. (17621763) [1764], pp. 163164. Math. xl. (1850), p. 361.

See also Minding, Journal fur

94

THEORY OF BESSEL FUNCTIONS


It follows that the general solution is expressible in the

[CHAP. IV
form

Hence
giving

it is

evident that,
lf 2
,

if
3
,

ylt yit y8 yt be any four solutions, obtained by


,

the values

C C C G

respectively, then the cross-ratio

(yi-y*) (y*-y*) (yi-y*)(y*-y*)


is

independent ofz; for

it is

equal to

(0 -(7# )(Ci
1

-(Q"
it

In spite of the obvious character of this theorem, been noticed until some forty years ago*.

does not seem to have

Other properties of the generalised Riccati equation may be derived from


properties of the corresponding linear equation ( 4*2). Thus Raffy f has given two methods of reducing the Riccati equation to the canonical form

these correspond to the methods of reducing a linear equation to its normal form by changes of the dependent and independent variables respectively.
Various properties of the solution of Riccati's equation in which P, Q, R are rational functions have been obtained by C. J. D. Hill, Journal fiir Math. xxv. (1843), pp. 23 37 Autoune, Comptes Rendus, xcvi. (1883), pp. 1354 1356; cxxvm. (1899), pp. 410 412; and Jamet, Comptes Rendus de V Assoc. Francaise (Ajaccio), (1901), pp. 207 228; Ann. de la

Fac. des Sci. de Marseille, xn. (1902), pp.

21.

hagen,

The behaviour of the solution near singularities of P, Q, Nieuw Archie/ voor Wiskunde, (2) vi. (1905), pp. 209
The equation
of the second order

R has

been studied by Falken-

248.

whose primitive

is

of the type

C1C1

+ C2& + C3C3'

where c x c2 , c3 are constants of integration (which is an obvious generalisation of the primitivo of the Riccati equation), has been studied by Vessiot, Ann. de la Fac. des Sci. de
,

Toulouse, ix. (1895), no. 6 and by Wallenburg, Journal fiir Math. cxxt. and Comptes Rendus, cxxxvii. (1903), pp. 1033 1035.

(1900), pp.

210

217-

* Weyr, Abh. biihm. Ges. Wiss. (G) vm. (18751876), Math. Mem. i. p. 30 ; Picard, Ann. Sci. de VEcole norm. sup. (2) vi. (1877), pp. 342 343. Picard's thesis, in which the result -is contained, is devoted to the theory of surfaces and twisted curves a theory in which Riccati's

equation has various applications.


t

Nouv. Ann. de Math.

(4)

n. (1902), pp. 529545.

4*3]
4*3.

DIFFERENTIAL EQUATIONS
Various transformatiwis of Bessel's equation.

95

The equations which we


Bessel's equation

are now about to investigate are derived from by elementary transformations of the dependent and inde-

pendent

variables.
first

The
(i)

type which

we

shall consider is*

p -*,.BiE
t dz*

where

c is

an unrestricted constant.

The equation

is

of frequent occurrence

in physical investigations, and, in such problems,

is

usually an integer.

The equation has been encountered in the Theory of Conduction of Heat and the Theory of Sound by Poisson, Journal de VlZcole Polytechnique,- xn. (cahier 19), (1823), pp. 249403; Stokes, Phil. Tram, of the Royal Soc. 1868, pp. 447464 [Phil. Mag. (4) xxxvi. (1868), pp. 401421, Math, and Phys. Papers, iv. (1904), pp. 299324]; Rayleigh, Proc. London Math. Soc. iv. (1873), pp. 93103, 253283 [Scientific Papers,
i.

(1899)',

pp. 138, 139].

The

special equation in

which

p=2

occurs in the Theory of the Figure of

the Earth; see

Ellis,

Camb. Math. Journal, n. (1841), pp. 169177, 193201.

Since equation (1)


,

may be
d

written in the form


(uz-i)
,

d? (uz-i)

its

general solution

is

(2>

= z^p+i

(ciz).

unrestricted,

Consequently the equation is equivalent to Bessel's equation when p is and no advantage is to be gained by studying equations of the form (1) rather than Bessel's equation. But, when is an

integer, the solu-

tions of (1) are expressible "in finite

termsf"

(cf.

4),

and

it

is

then

frequently desirable to regard (1) as a canonical form. various types of solutions of (1) will be examined in

The

relations

between

detail in

441-44:}.

is derived from (1) by a transformation of the dependent variable which makes the indicial equation have a zero root The roots of the indicial equation of (1) are + 1 and p -p, and so we write u = vz-p we are thus led to the equation

The second type

of equation

dz*

z dz

CV

'

of which the general solution

is

W
* See Plana,

= zP + i<@pH (ciz).

Mem.

Mem. della R. Accad. delle Sci. di Torino, xxvi. (1821), pp. 519538, andPaoli di Mat. e di Fis. della Soc. Italiana delle Sci. xx. (1828), pp. 183188.
(1)

t This was known to Plana, who studied equations has just been made.

and

(5)

in the paper to

which reference

96
Equation
der Physik
396];
in the
(3),

THEORY OF BESSEL FUNCTIONS

[CHAP. IV

(2) in. (1874), pp.

which has been studied in detail by Bach, Ann. Sci. de Vjtcole norm. sup. 47 68, occurs in certain physical investigations; see L. Lorenz, Ann.

und Chemie, (2) xx. (1883), pp. 1 and Lamb, Hydrodynamics (Cambridge,
form of continued fractions
(2)

21 [Oeuvres

Scientifiques,

1906), 287

291.

I. (1898), pp. 371 Solutions of equation (3)

de VAcad. R. de Belgique,
(1876), pp.

(cf. 5*6, 9 '65) have been examined by Catalan, Bulletin xxxi. (1871), pp. 68 73. See also Le Paige, ibid. (2) xli.

10111016, 935939.
(3),

Next, we derive from

by a change of independent

variable,

an equation

normal form. becomes


in its (5)

We write z %qjq, where q = l/(2p +1),

the equation then

Jr-c^-^o,
solution
is

and

its

(6)

= (Wqyi<$im) (cip/q).
absorbed into the symbol
^?,

When

a constant factor

is

the solution

may be

taken to be

P^iKwiciplq).
Equation
(5),

which has already been encountered

Mem.
(4)

della B. Accad. delle Sci. di Torino, xxvi. (1821), pp.

Studien

xxxvi. (1868), pp. 348351 [Collected ilber die BesseVschen Functionen (Leipzig, 1868), pp. 112

been studied by Plana, 519538; Cayley, Phil. Mag. Papers, vn. (1894), pp. 912]; and Lommel,

in 4 14, has

118.

The system

of equations which has

now been

constructed has been dis-

cussed systematically by Glaisher*, whose important memoir contains an interesting account of the researches of earlier writers.

The equations have been studied from a different aspect by Haentzschel f who regarded them as degenerate forms of Lame's equations in which both of
the invariants g 3 and g3 are zero.

pp.

The following papers by Glaisher should also be consulted Phil. Mag. 433438 Messenger, vm. (1879), pp. 2023 Proc. London Math.
:

(4) xliii. (1872),

Soc. ix. (1878),

pp.

197202.
It

may

be noted that the forms of equation (1) used by various writers are as follows:

cPy

k(k + l)

d*R n(n + l) n D _J__i __ _ p2 /2= Rf


.,

(Poisson),

S-*-*^*
Equation
(5)

(Wisher).

vertical pole of variable cross-section,


is

has been encountered by Greenhill J in his researches on the stability of a under the action of gravity. When the cross-section constant, the special equation in which <?= is obtained, and the solution of it leads to

Bessel functions of order +-J.

bridge

Tram, of the Royal Soc. clxxii. (1881), pp. 759 828 and Dublin Math. Journal, ix. (1854), pp. 272 290. t Zeitschriftfiir Math, und Phy*. xxxi. (1886), pp. 25 33. % Proc. Camb. Phil. Soc. iv. (1883), pp. 6573.
* Phil.

see also a paper

by Curtis, Cam-

4-31]
4*31.

DIFFERENTIAL EQUATIONS
Lommel's transformations of Bessel's equation.
occasions; his earlier researches* were of a

97

Various types of transformations of Bessel's equation were examined by

Lommel on two

somewhat

special

type, the later f were

much more

general.

In the earlier investigation, after observing that the general solution of


/i\ (1)
.

&y i?- s + *-
2i/-l dy
,

"(2)

y- *"#,(*).
direct transformations to construct the equation
at

Lommel proceeded by
general solution

whose
result,

is gfi"~

i^v

(y^), where

a,

#,

are constants.

His

which
(3)
is

it will

be

sufficient to quote, is that the general solution of


i

z*^ + (2a-2/3v+l)z^ + {&y*z*i + a(a-2/3v)}u =


u

(4)

= z*-*<@v (yz*).

When /3=0,
and when y=0,
unless

the general solution of (3) degenerates into

tt=g-(Cl +c2 log3);


it

degenerates into

(dv is zero.

The
be

solution of (3)

was given

explicitly

by Lommel

in

numerous

special cases.

It will

sufficient to

quote the following for reference

(7)

(8)

S+ ^+

(1

-"

S + 5"*=0;
tt=0;

*- + V,W')-**.{iV*).
u=

(1

_ v) g_i

(9)

^+/3y*-=0;
2

*Wmw (y*).
**^ (**).

(10)

l^ + 0;
<^ zu =0-

-**#S (Jld),
u=**#4 (3*),

(11)

*^j ().

An

account of Stokes' researches on the solutions of equation (11) will be given in

64, 10-2.

* Studien liber die Bessel'schen Functionen (Leipzig, 1868), pp.

98

120

Math. Ann.

in.

(1871), pp.

475487.

f Math. Ann. xiv. (1879), pp. 510586.

98

THEORY OF BESSEL FUNCTIONS

[CHAP. IV

Lommel's later researches appeared at about the same time as a memoir by Pearson*, and several results are common to the two papers. Lommel's procedure was to simplify the equation f
d*{ylx(z)\

2v-ld{y/ X (z)}
+(s)

d{ir{z)Y
of which the solution
( 12 )

d+iz)

+ y (*)~
X

'

is ( 4*3)

?-*(*) {*(*)}' *,{*<*)}.


reduction the equation becomes

On
(13)

^-li^) + (2^-l)t>) + 2 ^)l^


(*"<*)
l*'(*)

+ (2 ,-i)^ + 2^x>)_aGf) Tf ^
<

i.l

y=

0.

Now

define the function

(2)

by the equation

It will

be adequate to take

(H)
If

^W-^WfeWI'^W^
it is

1
.

we eliminate x ( 2 )>
v ;

apparent that the general solution of

ck

<(*)

dz

|_4 (</>(^j

2<(*)

y
is

=o

As a
(17
IS
>

special case, if

we take

<f>

(z)

= 1,

it is

seen that the general solution of

a?

[a

*^r - ito ++<)-+ *i i+wf J y

(18)

-V{* ()/*' CO}. #, {*<*)}


x (*) = {* (^)}M_V and we find
J

Next, returning to (13), we take


solution of

that the general

is

(20)
*

y {*<*)}* W*<*)).
f

Messenger, ix. (1880), pp. 127131.

The

Inactions

(*)

and

\p (z)

are arbitrary.

4-32]

DIFFERENTIAL EQUATIONS
following are special cases of (17):

99

The
(21)

(22)

^ + (^~^)y = H + ^^ =
will find

0;

y -#,(),

0;

y-*^^>
worked

The independent researches of Pearson proceeded on very similar lines except that he started from Bessel's equation instead of from the modified
form of it. The reader out in his paper.

many

special cases of equation (17)

partial differentia] equation closely connected with (7)


o2
0Z l
v '

and

(8),

namely

du
VZ

du

Ct

has been investigated by Kepinski, Math. Ann. lxi. (1906), pp. 397 405, and MyllerLebedeff, Math. Ann. lxvi. (1909), pp. 325330. The reader may verify that Kepinski's formula

is

a solution, when/(w>) denotes an arbitrary function of

w.

The
int.

special case of the equation

when

i/=

de

I'

Acad, des

Sci.

de

Cracoirie, 1905, pp.

198

205.

was also investigated by Kepinski, Bull,

4'32.

Malmsten's differential equation.

equation,

Twenty years before Lommel published his researches on transformations of Bessel's Malmsten* investigated conditions for the integrability in finite terms of the

equation

which
(15).
is

3+;-(^>
obviously a generalisation of Bessel's equation
;

and

it is

a special case of

^ 4*31

To reduce the

equation,

Malmsten chose new

variables defined

by the formulae

where p and q are constants to be suitably chosen.

The transformed equation

is

We choose p
so that p

Plana, and therefore

and q so that we take

this

may

reduce to the equation of 4*3

(1)

considered by

= - ir fan.

2pq-q + l+qr=0,
to

(m + 2)q = 2,

The equation then reduces

cPuV
d?
*

4A

L(

+ 2) 2+

2 2 g {4*+(l-r) }-l ~]

4*

_P

Camb. and Dublin Math. Journal, v. (1850), pp. 180 182. The case in which =0 had been previously considered by Malmsten, Journal fUr Math, xxxix. (1850), pp. 108 115.

100

THEORY OF BESSEL FUNCTIONS


4*3 this is intcgrable in finite

[CHAP. IV

By

terms

if

to{4 + (l-r)}-i-*( +

l),

where n
(2)

is

an integer

so that

m + 2= + ^+(l-r)>}
+
is

The equation

also obviously integrable in the trivial cases

A =0 and

m=

-2.

4'4.

The notation of Pochhammer for

series

of hypergeometric

type.

is

A compact notation, invented by Pochhammer* and modified by Barnesf, convenient for expressing the series which are to be investigated. We shall
now and subsequently
(a)

write

= o(a+ l)(a+
will

2)

...

(a

+-

1),

(a)

1.

The notation which


p* q (au a2
,

be used
,

is,

in general,

...,

ap

Pl

p2

...,

pq

z)=
rt=0

n
J

j-^
KPi)n \Pi)n

zn

\Pq)n

In particular,

I
CO

< g >"

n=0W!(p)n
.71

=0!(p)n'
*F,(,/9;

,;*)=*

-^7^^
three series are called generalised hyper-

The

functions defined by the

first

geometric functions.
It

may be

noted here that the function

(a

z) is

a solution of the

differential equation

and,

when p

is

not an integer, an independent solution of this equation

is

^"".^(a-p +
It is evident that

l; 2 -p;

z).

Various integral representations of functions of the types \Fi

i , 0^*3

have been studied

by Pochhammer, Math. Ann.


*

xli. (1893), pp.

174178, 197218.
(1891), pp. 227, 586, 587.
Cf. 4-15.
is

Math. Ann. xxxvi.

(1890), p.

84

xxxvm.

t Proc. London Math. Soc.


of the suffixes

(2) v. (1907), p. 60.

The

modification due to Barnes

the insertion

p and

q before and after the

F to render evident the number of sets of factors.

4'4, 4*41]

DIFFERENTIAL EQUATIONS
Various solutions in
series.

101

4'41.

We

shall

now examine

various solutions of the equation

d2 u
dz*

-j

r + \) c*u = p(p - u,

z*

and obtain relations between them, which in Pochhammer's notation.


It is

will for

the most part be expressed

supposed

for

and, equally, since the equation

the present that p is not a positive integer or zero, is unaltered by replacing p by p 1, it is


f
sfi

supposed that p
It is already

is

not a negative integer.


(4-3) that the general solution*
is

known

$p+

(ciz),

and

this gives rise to the special solutions


z*+*
.

F, (p

\c*z-)

z~p

-oF^-p;

\c*z*).

The equation may be written

in the forms

dz*

dz

z*

which are suggested by the

fact that the functions e e* are solutions of the

original equation with the right-hand side suppressed.

When ^

is

written for z (d/dz), the last pair of equations become

(^

-p -

1) (^

+ p)

(ue**) 2cz% (uei*2 )

= 0.

When we solve these in series we are led to the following four expressions for u
zP+^.tF^p + l; 2p + 2; -2cz); zp+i e-^^F^p + 1; 2p + 2; 2cz)
Now, by
sT*'e**
l

(- p

-2p; -2cz);
2cz).
left

zr* era .xFx (-p\ -2p\

direct multiplication of series, the


,

two expressions on the


,

are

expansible in ascending series involving z?+1

z^ zp+3
.

....

And the expressions

on the right similarly involve powers are the same when 2p


(1)

tr*,
is

z1 ~p,

z-'P, ....

Since none of the two sets of

not an integer,
x

we must have
x

*. ,F, (p

2p + 2

- 2cz) = e~ n F

(p

2p + 2

2cz)

(2)

ef.Ai-p; -2p; -2cz) = e-". F (-p; -2p;


1
1

2cz)

These formulae are due

to

Kummerf. When
is

(1) has

been proved

for general

values of p, the truth of (2)

obvious on replacing
is

p by p 1
integer.

in (1).

We now
* It follows

have to consider the cases when 2p


from
3*1 that

an

a special investigation

is

also necessary

when p

is

half of an odd

integer.

t Journal fllr Math. xv. (1836), pp. 138141.

102

THEORY OF BESSEL FUNCTIONS


When p
has any of the values , f f
,

[CHAP. IV

as a factor have to be replaced

and there

is

..., the solutions which contain z~p by series involving logarithms (3*51, 3*52), only one solution which involves only powers of z. By the
,

previous reasoning, equation (1)

still

holds.

When p

has any of the values

0, 1, 2, ...

a comparison of the lowest powers


still

of z involved in the solutions shews that (1) that there are no relations of the form
z-p J\(\-p\ \<?z*)
x

holds; but

it is

not obvious

= z-ve*\F (-p\ -2p; -2cz) + k zP+\F


1

^z-Pe-^F.i-p; -2p;
where
lcit

2cz)

+ k2 zr+*

<t

+ ; \&z*) F {p + %; \c*z%
1

(p

k2 are constants which are not

zero.

We
and
(2)

have to give an independent investigation of which depends on direct multiplication of series.


shall consequently

(1)

Note.
vii.

tions of the series

In addition to Rummer's researches, the reader should consult the investigaby Cayley, Phil. Mag. (4) xxxvi. (1868), pp. 348351 [Collected Papers,

(1894), pp. 912] and Glaisher, Phil. Mag. (4) xliii. (1872), pp. Trans, of the Royal Soc. clxxii. (1881), pp. 759828.

433438;

Phil.

4'42.

Relations between the solutions in series.

The equation

*Vi(p+l; 2p+2;
which forms part of equation
general formula due to
(i)

-2cz)

= e-ez F
1

(p

+ l; 2p + 2;

2cz),

(1) of

441,

is

a particular case of the more

Kummer*

i^(; p;

0-Wp-;

/>;

-),,

which holds for all values of o and p subject to certain conventions (which will be stated presently) which have to be made when a and p are negative integers.

We first suppose that p is not a negative integer and then the coefficient of n in the expansion of the product of the series for e* and (p a p ) is

&

mto(n

- m)\

ml(p)m

" Wp)n .V^*' {p " ")m (1 " P ~ n)


nl(p) n

"

=
if

()

n\(p)n

'

we

first

in the numerator;

use Vandermonde's theorem f and then reverse the order of the factors and the last expression is the coefficient of fn in X X (a; p; *).

The
*

result required is therefore established

when a and p have general complex

values J.
Journal fur Math. xv. (1836), pp. 138

141;

see alsoJBach, Ann. Sci. de I'ticole norm. sup. (2)

in. (1874), p. 55.

t See, e.g. Chrystal, Algebra, n. (1900), p. 9. $ Another proof depending on the theory of contour integration has been given by Barnes,

Tram. Camb.

Phil. Soc. xx. (1908), pp.

254257.

4'42]

DIFFERENTIAL EQUATIONS
p
is

103

When

a negative integer, equation (l)is obviously meaningless unless

also o is a negative integer

and a < p
! \ |

j.

The

interpretation of (1) in these

circumstances will be derived by an appropriate limiting process.


First let a be a negative integer

(=
p

N) and
The
;

let

that the preceding analysis

is

valid.

series

^(N;

p not be an integer, so p\ ) is now a

terminating
sists of

series,

N+ 1

+ 1, p + p + with the later factors of the sequences

is an infinite series which conterms followed by terms in which the earlier factors p + N, + 2, ... in the sequences in the numerators can be cancelled

while

(p

+N

p,

1,

+ 2,

...

in the denominators.
X

When
iFt (p

these factors have been cancelled, the series for


;

is

are both continuous functions of p near p any of the integers N, + 1, + 2,


;

+ iV p

= M,

and where

. . .

Hence we may proceed


of (1)
(2)

to the limit

when p

-*.

if,

and the limiting form

may then be

written*

tJFxO-JV;
"1

-Mi Ol^e^F^N-M; -M;


means that the

-)1,
term in
*', i.e.

in which the symbol

series is to stop at the

the last term in which the numerator does not contain a zero factor, while
1 means that the series is to proceed normally as far as the term and then it is to continue with terms in f Jf+1 f->/+2 ... the vanishing factors in numerator and denominator being cancelled as though their ratio were one of equality.

the symbol
in % M ~N
,

With
(3)

this convention, it is easy to see that

Ai-N; -M;
+ ( - )M ~ N

?)1=i^,(-iV;

-M;

f)

1
1
;

N
MKM~Ay.
-

** +I

^ (^ " ^ +
,

3/

?).

When we
(4)

replace iV

by

M N and by
) 1

we have
;

^(N-M; -M;
+ ( ->"
If

= iJFi <tf -

Jf

-M; -?)1
1
;

k")! <- P'* (^ +


we have
f)

3/

2
;

t).

As an ordinary

case of (1)

^(M-N+l; M+2;
and from
(5)
this result

(A +l;
r

if

+ 2; -

),

combined with
;

(2), (3)

and

(4)

we deduce that

^ (-

iV;

- Jf 01 - rf

^ (JV-

if;

if;

) 1.

an integral
462],

This could have been derived directly from (1) by giving p value, and then making p tend to its limit.
* Cf. Cayley, Messenger (old series), v. (1871),

(instead of o)

pp.77 82

[Collected Papers,

vm.

(1895), pp. 458

aDd Glaisher, Messenger, vm.

(1879), pp.

2023.

104

THEORY OF BESSEL FUNCTIONS


We
next examine the equation

[CHAP. TV

(6)

0*J

(p

+1

2p

2cz)

& (p +

f;

\&z%
is also

which

forms the remainder of equation (1) in 4*41,

and which

due to

Kummer*. If we suppose

that 2p

is

not a negative integer, the coefficient of (cz) n in


left in (6) is

the product of the series on the

(-2r
1

(p

ix

m= o(n-m)\ml(2p
n

(-)*

2) m

(2p
l)m (-

+
n

2) n m=0

1 2m Q> + u(---2;>-i)n-w ml(n-m)l


is

Now

2W

Cm (^ +
l

- 2p - l)n-,
,

the coefficient of tn in the

expansion of (1

2t)~p~

(1

t) n+2p+1 and
1

so it
1

is

equal to
/>+)

2~
of the

f(o+)
.

(1

- 20"P_I (1 - t)n+2P +1 tr"-

dt

Y~\

0-

~ 0~p_1 w~ n~

dM,

where w

= t/(l t) and the contours enclose the origin but no other singularities integrands. By expanding the integrand in ascending powers of u, we
is

see that the integral


even.

zero if n

is

odd, but

it is

equal to v,,

(p

4- l)i
.,

when n

is

Hence

it

follows that

%
=o2

2n

m .n!(p + f) n
(cz)

'

and

this is the result to

be proved.

When we make p
lim
,F,

tend to the value of a negative integer,

N, we

find

by

the same limiting process as before that

(p +

2p + 2

- 2cz) = F (l-N;
l
l

- 22V; - 2cz) 1

It follows that

&Q-N;
we change the we find that
If
(7)

icV) = ^.

(l-iV; 2-22V;

- 2c*) 1

signs of c and z throughout and add the results so obtained,

2 tfx (f

- N icV) = e.
;

(1

- N;

- 2tf; - 2cz)
1

~1

+ e-^.

F (l-N; 2-2N;
1

2cz)~\,

* Journal fUr Math. xv. (1836), pp. 138 141. In connexion with the proof given here, see Barnes, Tram. Camb. Phil. Soe. xx. (1908), p. 272.

4*43]

DIFFERENTIAL EQUATIONS

105

course, the expression for

the other terms on the right cancelling by a use of equation This is, of (1) J_y+ j (icz) in finite terms with a different notation.

For Barnes' proof of Rummer's formulae, by the methods of contour


gration, see 65.

inte-

4*43.

Sharpe'8 differential equation.

The equation
(1 >

'g+g + C + ^-o.

which is a generalisation of Bessel's equation for functions of order zero, occurs in the theory of the reflexion of sound by a paraboloid. It has been investigated by Sharpe* who has shewn that the integral which reduces to
unity at the origin
(2)

is

y=C\
J

cos {z cos

+A

log cot \0)d0,

where
(3)
1

Jo

"cos

(4

log cot \0) dd.

This

is

the appropriate modification of Parseval's integral

( 2-3).

To

in-

vestigate its convergence write cos

= tanh

<f>,

and
*>)

it

becomes

(4)

=G

r CSW + * tanh<
Jo
coshrf> <f>
1,

dd> ^

It is easy to see from this form of the integral that

it

converges for (complex)

values of

for

which \I(A)\<

andf
2

C=The

cosh

hvA.

ir

integral has been investigated in great detail by Sharpe and he has given elaborate rules for calculating successive coefficients in the expansion of y in powers of z.

A simple

form of the solution (which was not given by Sharpe)

is

y- to ,^(jTi^;
The reader should have no
* Messenger, x. (1881), pp.

1;

2iz).

difficulty in verifying this result.


xn. (1884), pp.

174185

6679

Proc. Camb. Phil. Soc. x. (1900),

pp. 101136.

t See,

e.g.

Watson, Complex Integration and Cauchy's Theorem (1914), pp. 64

65.

106
4*5.

THEORY OF BESSEL FUNCTIONS


Equations of order higher than the second.

[CHAP. IV

The

construction of a differential equation of any order, which


of Bessel functions, has
;

is

soluble

been effected by Lommel* its possibility depends on the fact that cylinder functions exist for which the quotient <$ (z)/^ v (z) is independent of z.

by means

Each of the functions Jn (z) and Yn (z), of integral order, possesses (z), H property [ 2*31, 3*5]; and the functions of the third kind p

this
{2)

(z)

possess it ( 3*6 1),

whether v

is

an integer or

not.

Now when
C1 )

3*9 (5) is written in

the form
}( "- m,

gr^'^fr V)- (7)w *


is is

^,-m(7 V*)>
if

the cylinder function on the right

of order

m = 2v.
and

This
v

the case either

(i)

if v is

an

integer, n,

m = 2n,

or

(ii)

if

= n+\

and
if

m = 2n + 1.
,

Hence

^n denotes either Jn or Yn
we obtain Lommel's

we have

From
z^ n

this equation

result that the functions z* n

Jn (y \Jz)

Yn (y \lz) are

solutions off

,o\

where 7 has any value such

J"y_ (frry dz ~ zn that 2n = ( )n cm


<y

'

so that
(r

= ic exp (riri/n).
we obtain 2w

= 0,

1, 2,

. .

n-

1)

By

giving 7
r

all

possible values

solutions of (2),

and these

form a fundamental system.


Next,
if tfn+i

denotes

{1)

n+i

we have

#Hn+i) -*+*> **#+*, so that

and hence
\P)

z* n+

lH

{l)

n+i (7 Jz) is a solution of

dzm+1

zn+i

'

where 7 has any value such that 72n+1

= cm+1 e- (n+ *
{riri/(n

),ri
,

so that
(r

y
and the solutions
*

= -ic exp

+ )},

= 0,

1, 2,

. .

2rc)

so obtained

form a fundamental system.


;

Studien

tiber die

BesseUschen Functionen (Leipzig, 1868), p. 120

Math. Ann. u. (1870),

pp. 624635.

The more

general equation

has been discussed by Molins, MSm. de VAcad. des

Sci. de Toulouse, (7) vin. (1876), pp.

167 189.

4-5]

DIFFERENTIAL EQUATIONS

107

pp.

For some applications of these 317320.

results, see Forsyth, Quarterly Journal, xix. (1883),

In view of (1), which holds when m is an integer, Lommel, Math. Ann. n. (1870), p. 635, has suggested an interpretation of a "fractional differential coefficient." Thus he would
JgJ length by Heaviside in various papers.

exp(yV)

to

mean 90o

(yijz).

The

idea has been developed at

some

Lommel's formulae may be generalised by considering equation


4*31, after writing it in the

(3) of

form

(* + a) (^

+ a - 2j3i>) u = - 'ptftPu,
(<yz^).

the solution of the equation being u = zpv- a(@ v by induction that, with this value of u,
n-\
II

For

it is

easy to verify

(^

+ a - 2r/3) (^ + a - 2/3p - 2r)

r=0

= (-) n/92nc2ns 2n

<,

w,

and so solutions of
-i 4)

(^

+ a - 2r) (^ + a -

2/3v

- 2r/8) u = (-)n/3^cmz^ u
= 0,
n-

are of the form

where

u = z^"-^v ( 7*0), 7 = c exp (riri/n).

(r

1,

. . , .

1)

By

giving 7 these values, system.

we

obtain 2n solutions which form a fundamental

In the special case in which n


(Sr

= 2, equation (4) reduces to + a ) (^ + _ 2)(^ + a 2pv) (* + a - 2/3* - 2/3) u = Wu.


Sponge Spicules, namely
<P
[

This equation resembles an equation which has been encountered by Nicholson* in the
investigation of the shapes of
(5)

2 Au.d u\

that

is

to say

$(3-l)(S + 4/*-2)
j*,

($

+ 4p-3) u=*-i*u.
(4)

If

we

identify this with the special


v
:

form of

we obtain the following four distinct

sets

of values for a, ,

JL

1
2.

2
1

-1
* Proc.

~\

* 3

10

Royal Soe. xcm.

Soc. lxxxix.

(1917), pp. 573

587

(1917), pp. 506


;

519. See also Dendy and Nicholson, Proc. Royal the special cases of (5) in which /*=0 or 1 had been solved

previously by Kirchhoff, Berliner Monatsberiehte, 1879, pp. 815828.

[Ann. der Physik und

Chemie,

(3) x. (1880), pp.

501512.]

108

THEORY OF BESSEL FUNCTIONS

[CHAP. IV

These four cases give the following equations and their solutions
(6)

^=;

**{#! +^ 4 ()},
22tt;

(7)

~ S} =
4

{?

=^
J

{^2 (2^)+^2(2tV^},

(8)

{*V S} *

-*'*{^(l )+^t-(t il .
-^{^(~*) +*(-*)}.

(9)

t S} =2
,2
;

tt;

of Bessel functions

These seem to be the only equations of Nicholson's type which are soluble with the aid in the case fi 2, the equation (5) is homogeneous. Nicholson's general
is

equation

associated with the function


/ 3-2/x
3

2 + 2/x

1+2/i
'

z*-** \
'

V4-2/*'

4-2 M

4-2M

(4-2M )V'

4*6.

Symbolic solutions of differential equations.


solutions of the equation 4*3 (1)

Numerous mathematicians have given


namely

in symbolic forms,

when p
( 4*3)

is

a positive integer (zero included).

These forms

are intimately connected with the recurrence formulae for Bessel functions.
It has

been seen

that the general solution of the equation

is

and from the recurrence formula

39 (6) we have

***Wcu) - (- ays*** (Jf [r*Vh (ciz)}.


Since any cylinder function of the form
(ae
z
*<fj

(ciz) is expressible as

+ Pe-^ls/z,
may

where a and be written


<

/3

are constants,

it

follows that the general solution of (1)

2>

u=zP+1

[jdz)

ae + &ey .
6*

c*

r
6

A
(3)

modification of this, due to Glaisher*,

is

w
a'

= z*** (Ay^(a'** + P'e- *),


may be
seen by differentiating CLeP+ffe-*2 once.

where

= a/c, #'= -fife.


Tram, of
the

This

* Phil.

equation
p. 92.

Royal Soc. clxzii. (1881), p. 813. It was remarked by Glaisher that Earnshaw, Partial Differential Equations (London, 1871), See also Glaisher, Quarterly Journal, xi. (1871), p. 269, formula (9), and p. 270.
(3) is

substantially given by

4 * 6]

DIFFERENTIAL EQUATIONS

109

Note. A result equivalent to (2) was set by Gaskin as a problem* in the Senate House Examination, 1839; and a proof was published by Leslie Ellis, Camb. Math. Journal, n. (1841), pp. 193195, and also by Donkin, PhU. Trans, of the Royal Soc. cxlvii. (1857), pp. 4357. In the question as set by Gaskin, the sign of c 2 was changed, so that the solution involved circular functions instead of exponential functions.

Next we
(4)

shall prove the symbolic theorem,

due to Glaisherf, that

W4Y.JL
\zdz)

zp+1
it is

that the function


operators z3 (d/dz).
It
is

In operating on a function with the operator on the right, is multiplied by I/* *- 2 before the
5

supposed

application of the

convenient to write

and then
(5)

to use the symbolic formula

/('*) (e"Z)
.

e*

./(*

+ a) Z,

in

which a
The proof

is

a constant and

Z is

any function of z.

S, as is

Equations (1914),

of this formula presents no special difficulties when/(<>) is a polynomial in the case in the present investigation. See, e.g. Forsyth, Treatise on inferential jrwwm*
33.

It is easy to see from (5) that

= "-*<*- 2p + 2)<*when we bring the

2j,

+ 4)<* -

2p

+ 6). ..*,
(5).

successive functions er (beginning with those on the left) past the operators one at a time, by repeated applications of

reverse the order} of the operators in the last result, and by a reversal of the previous procedure we get

We

now

= <*- [<* + 2p - 2) (* + 2p - 4) (* + = e -(p+i) [(4*$) (e**>)... (eie *).e-w--> 9


. . .

2) * -<*-*>]
.

= J-[( 23 lYJL~\

z^

\_\

dz) z*-*]

'

Th e P bIem W<18 * he 8eCODd Part f uestion 8 Tu eday 1 afternoon, Jan. n Cambridge University Calendar, 1839, p. 319.

8,

1839; see the


t,ie

JJ:;iZ S
t
It

(1876)>

pp 240
'

~243

'

349

"350

and PhiL Tram of


'

Rmjal Soc

was remarked by Cayley, Quarterly Journal, xn.

(1872), p. 132, in a footnote to a paper


i.e.

by

Glaisher, that differential operators of the form *+'

s -,

-, obey

the commutative law.

'

110

THEORY OF BESSEL FUNCTIONS


this is the result to

[CHAP. IV
1,

and

be proved.

If

we

replace

p by p +

we

find that

When we

the general solution of


{1)
v /Q 8

transform (2) and (3) with the aid of (4) and (6), (] ) is expressible in the following forms

we

see that

~zp+A

dz)

*-

>

<

>

l / -dy+WtP + fftr * uss **\*di)

^
,

'

The

solutions of the equation


2 v d --

. 2pdv - -E-j- cH = 0,

[(3) of 4-3],
<9 >

which correspond to

(2), (3), (7)

(10)

-^Ub) v = ^ (-^J
+1

and (8) are

(aV*

+ 0'e-a

),

/nx
12 >
more

1^,

dya f + fier"

<

^s)

?!

(7) is due to Boole, Phil. Trans, of the Equations (London, 1872), ch. xvn. on Differential Royal IX. (1854), p. 281. pp. 423425; see also Curtis, Cambridge and Dublin Math. Journal, II. Journal, (1841), pp. 169, Ellis, Camb. Math. first Leslie given by The solution (9) was 193, and Lebesgue, Journal de Math. xi. (1846), p. 338; developments in series were

A different and

direct
;

method of obtaining
Treatise

Soc. 1844, pp. 251, 252

obtained from

it

by Bach, Ann.

Sci.

de

I'jtcole

norm, sup.

(2) in. (1874), p. 61.

Similar symbolic solutions for the equation

~-c

z%>~i

v=0

were discussed by Fields,

John Hopkins University

Circulars, vi. (1886


(9),

7), p. 29.

transformation of the solution


is

due to Williamson, Phil. Mag.

(4) xi. (1856),

pp. 364371,
(13)

/3
v==c

IV

^\dc'c)

ae"

+ 3e ~
'

*)-

r)

This

is

derived from the equivalence of the operators cz.

~-

if) - g-

when they operate on

functions of

We thus obtain the equivalence of the

following operators

-^'[(M^m-i}
it

being supposed that the operators operate on a function of

cz;

and Williamson's formula

is

then manifest.

4*7]
4*7.

DIFFERENTIAL EQUATIONS
Liouville's classification

111

of elementary transcendental functions.

in

Before we give a proof of Liouville's general theorem (which was mentioned 412) concerning the impossibility of solving Riccati's equation "in finite
"

except in the classical cases discovered by Daniel Bernoulli (and the we shall give an account of Liouville's* theory of a class of functions known as elementary transcendental functions; and we shall introduce a convenient notation for handling such functions.
limiting form of index -2),

terms

For brevity we write f


h (z) =
e 1 (z)

l(z)

log z,
,

l2

(z)

= I (I (z))

l3

(z)

= I (l2 (*)),
= e(ea (z)),
{*/(*)},

. .

e(z)

= ez

ea (z)

= e(e(z)),

e3 (z)

*if(*)-*f(*)-ff(*)dM,

9,/(#)-9{9/(#)},

*/<*)- 9

....

A
lr

function of z

if it is expressible
<f>(z),

e r y}r(z),

is then said to be an elementary transcendental function.% as an algebraic function of z and of functions of the types <irx( z )> where the auxiliary functions <f>(z), yjr(z), %(z) are

expressible in terms of z and of a second set of auxiliary functions, and so on; provided that there exists a finite number n, such that the nth set of auxiliary

functions are

all

algebraic functions of

z.

The order
(I)

of an elementary transcendental function of z

is

then defined

inductively as follows:

Any

algebraic function of z

is

of order zero.
r,

(II) If

fr (z)

denotes any function of order

then any algebraic function

of functions of the types

*M*\
(into

efr {z),

,fr (z),
first

fr (z),

/(*),.../,(,)
is

which at

least

one of the
is

three enters)

said to be of order r

1.

(III)

Any

function

possible order.

Thus elfr (z)

order

r,

not of order r

+ 2.

supposed to be expressed as a function of the lowest is to be replaced by r (z), and it is a function of f

In connexion with this and the following sections, the reader should study Hardy, Orders of Infinity (Camb. Math. Tracts, no. 12, 1910). The functions discussed by Hardy were of slightly more restricted character than those now under consideration,
i

since, for

his purposes, the

not required, and also, for his purposes, postulate the reality of the functions which he investigates.
r is

symbol

it is

convenient to

It may be noted that Liouville did not study properties of the symbol s in merely remarked that it had many properties akin to those of the symbol I.

detail,

but

T Journal de t It
is

Math. n. (1837), pp. 56105 in. (1838), pp. 523-547 supposed that the integrals are all indefinite.
;

iv.

(1839), pp.

423456

t " Une fonction fiuie For the purposes of

explicite."

this investigation, irrational

powers of

:,

not be regarded as algebraic functions.

such as z" of course must


',

112
4*71.

THEORY OF BESSEL FUNCTIONS


Liouville's first

[CHAP. IV

theorem* concerning linear differential equations.

The

investigation of the character of the solution of the equation

in

which

% (z)

is

a transcendant of orderf
:

n,

has been

made by

Liouville,

who

has established the following theorem

If equation

(I) has

solution which is

a transcendant of order

m+

1,

where

m > n,
(2)

then either there exists a solution of the equation which is of order\ n,

or else there exists a solution, u i} of the equation expressible in the form


U:

= <M*) */*(*),
<M

where

M (z) is

of order

ft,

and

the order of

(z) does not exceed

/a,

and

fi is

such that

n^p^m.
;

/m+i ( z )

m + 1, let it be fm+i(z) then an algebraic function of one or more functions of the types lfm (z) sfm ( z )> efm (z) as well as (possibly) of functions whose order does not exceed m. Let us concentrate our attention on a particular function of one of the three types, and let it be called 6, ^ or according to its type.
If the equation (1) has a solution of order
is

(I)

m + 1,

We shall first shew how to prove that, if (1) has a solution of order then a solution can be constructed which does not involve functions of
S-.

the types 6 and


For,

if possible, let /, )l+ i (z)

= F(z, 0), where F is an algebraic


itself) of

function of
in

and any function of z (other than is algebraically independent of 0.

order

m+1

which occurs

Then
/qx {6)

it is

easy to shew that

*F

/ n F X K) " Tz*~

W+
dz>
1

df (z)
dz

d*

fm

(z)

d6dz

+ (_!_
it

dfm {z) \*d*F


dz

\d_

f_l_ dfm (z)}


dz
]

\fm {z)

dP^ldz\fm (zj

W-'-xW

being supposed that z and 6 are the independent variables in performing

the partial differentiations.

The expression on the right in (3) is an algebraic function of 6 which vanishes identically when 6 is replaced by lfm (z). Hence it must vanish
identically for all values of 6
;

tor if it did not, the result of

equating

it

to

zero would express lfm

(z) as an algebraic function of transcendants

orders do not exceed

together with transcendants of order


6.

m+ 1

whose which are,

ex huputhesi, algebraically independent of


'

Journal de Math.
is

iv. (1839), pp.

435442.

t This phrase

used as an abbreviation of " elementary transcendental function of order u."


;

% Null solutions are disregarded


less

if

u were of order

less

than , then - --^ would be of order

than

w,

which

is

contrary to hypothesis.

4'71]

DIFFERENTIAL EQUATIONS
is is

113
replaced

by

In particular, the expression on the right of (3) vanishes when + c, where c is an arbitrary constant and when this change
;

made the

expression on the

left

of (3) changes into

d-F(z,0 +
which
is

c)
'-

-F(z,0 + c). x (z),


,

therefore zero.

That

is

to say

When we

differentiate (4) partially with regard to

c,

we
'"

find that

dF(z,
oc

+ c)
'

o-F(z,
dc>

+ c)
'

are solutions of (1 ) for all values of c independent of z. If performing the differentiations, these expressions become

we put

after

dF(z,

)
'

d3F(z,

d0
Fe>

d&
(I).

) "'

'"'
will

which are consequently solutions of


-Fee, ....

For brevity they

be called

Now

either

and

Ft

form a fundamental system of -solutions of (1) or

they do not.
If they do not,

we must have*

F = AF,
where

is

independent both of z and

On 0. F=<PeA0
,

integration

we

find that

where 4> involves transcendants (of order not exceeding + 1) which are algebraically independent of 0. But this is impossible because e Ae is not an algebraic function of and therefore and e form a fundamental system ; of solutions o/'(l).

Hence
where
in

Fee

is

expressible in terms of

F and F

by an equation of the form

and

B are

constants.

FM = AF, + BF, Now this may be


<t>2e0

regarded as a linear equation

(with constant coefficients) and

its solution is

F = O^* +
where
<I>,

or

/= e* {4>

+ 4>a 0},
4>,

and

3>2 are functions of

the same nature as

while a and

/9

are

the roots of the equation

a?-Aa;-B = 0.
The only value
a

=$=

F which is an algebraic function and then F is a linear function of 0.


of

of

is

obtained when

Similarly, if
it

fm+1 (z)
* Since

involves a function of the type ^,

we can prove

that

must be a

linear function of ^.

F mutt involve

0,

Ft cannot be

identically zero.

114

THEORY OF BESSEL FUNCTIONS


them
linearly, so that
1

[CHAP. IV
and ^,

It follows that, in so far as fm+x (z) involves functions of the types


it

involves

we may

write
>

fm+i(z)^%0
where the functions
of order

(z)02 (z)...0p (z).^ 1 (z)%(z)...^ q (z).yfrPiq (z)


order
,

yffp q (z) are of

m + 1 at most, and
which be
...

the only functions

m+1

involved in

them
...,
...

are of the type .

Take any one of the terms


function of
lt 0.,,

in

fm+l (z)
let it

is

of the highest degree, qua

...%,%,

and

Ox {z)0t {z)

P (z).* 1 (z)

* Q (t).irPiQ (z).
it

Then, by arguments resembling those previously used,


"

follows that

d_d_
d0!

d_

d_
1

d_
d*S 2

_a_

d02

'"

d0p'd&

"*

S^

/**.(*)

is

a solution of (1)

i.e.

yffp^iz) is a solution of (1).

But ^p
the types

Q (z) is either

function of order

m+1
case,

a function of order not exceeding m, or else it is a which involves functions of the type and not of
repeat the process of reduction to functions of lower

and ^.

In the former
order,

we

and in the

latter case

we

see that

some solution of the equation


a solution which
is

is

an

algebraic function of functions of the type .

We

have therefore proved


is

that, if (1) has


it

a transcendant
else it

of order greater than n, then either


solution which

has a solution of order n or

has a

$* ( z )> where /M (z)


(II)

an algebraic function of functions of the type ef^iz) and is of order /* and <M (z) is of an order which does not exceed ft.
next prove that, whenever (1) has a solution which is a n, then it has a solution which involves
it

We

shall

transcendant of order greater than

the transcendant ef^iz) only in having a power of

as a factor.

We concentrate
e/j, (z),

our attention on a particular transcendant

of

the form

and then the postulated solution may be written in the form where is an algebraic function of and any function (other than of order /* + 1 which occurs in G is algebraically independent of .
;

G (z,

),

itself)

Then

it is

easy to shew that

rf*G

&G

rPG

d*G

+ e[/"

{/,'(*))]

|g-G.x<*>.

The expression on the right is an algebraic function of which vanishes when is replaced by e/^iz), and so it vanishes identically, by the arguments used in (I). In particular it vanishes when is replaced by c , where c is independent of z. But its value is then

'-G(*,ce).x(),

4*71]
so that
/a\ (6)

DIFFERENTIAL EQUATIONS
d*G(z,c%) ;

115

~. , , -G(z,c&). X (z) =
.

0.

When we

differentiate this with regard to

c,

we

find that

dG(z,c&)
dc
'

&G (z,
9c

c0)
''
'

are solutions of (1) for all values of c independent of expressions become


ft

z.

If

we put

= 1, these

dG(z,S)

^ d*G(z,e)
*

d@
Hence, by the reasoning used in

a@ 2
we have

'

"

(I),

G9 = AG or else

&G*B = AGB + BG,


where

and

B are

constants.

In the former case


$>,*

G Q% A
or

and
1

in the latter

has one of the values


l

+ O,**

&{&

+ &t \og&} =

&{&

& f?(?)},'
t
;

where

4>,

Q>lt

<t>8

are functions of z of order jl+ 1 at most, any functions of

order fi+1 which are involved being algebraically independent of

while

7 and $ are the roots of the equation


x (x
In any
case,

1)

- Ax - B = 0.
is

either contains

only by a factor which

a power of

or else

is

the

sum

of two expressions which contain

only in that manner. In the

latter case*,

G(z, c%)-c*G(z, 0)
is

a solution of (1) which contains

only by a factor which

is

a power of 0.
r

By

repetitions of this procedure,


/*

we

see that, if 0,,

2 , ...

are

all

the

which occur in the postulated solution, we can derive from that solution a sequence of solutions of which the sth contains ... 0,; and the rth 0,, 0, ... 0, only by factors which are powers of n 2 member of the sequence consequently consists of a product of powers of 0i, r multiplied by a transcendant which is of order ft at most; this solution is of the form
transcendants of order
1
,

(*)exp|2 7log*V>
which
is

of the form

^M (z)

ef^ (z).
;

* If $j is not identically zero

if it is,

then

4> 0* is 2

a solution of the specified type.

116
4*72.

THEORY OF BESSEL FUNCTIONS

|CHAP. IV

Liouville's second theorem concerning linear differential equations.


if

We
(!)
[in

have just seen that,

the equation

;i-X<*)
is

which x ( z )

f order n] has a solution which


n,

is

an elementary tran-

scendant of order greater than

then

it

must have a solution of the form

where

/*

^ n.

If the equation has


/*

more than one solution of

this type, let a


it

solution for which

has the smallest value be chosen, and let

be called

,.

Liouville's theorem,

which we
is

shall

now

prove,

is that,

for

this solution, the

order of d (log uj/dz

equal

to n.

Let

d
and then
If
t

log

dz
is

C'

of order

at

most

let

the order of
is

be N, where iV < /*.

N=n,
by
t,

the theorem required

proved.

If

N>n,

then the equation

satisfied

namely

has a solution whose order iV

is

greater than

n.

Now
tys-i i z),

an algebraic function of at least one transcendant of the types s/jr-i (z), ef N _ x (z) and (possibly) of transcendants whose order does
t is

not exceed
If
t

N-

1.

We

call

the

first

three transcendants

0,

^,

@ respectively.

contains more than one transcendant of the type 0, we concentrate our attention on a particular function of this type, and we write
t

= F(z,0).
if

By arguments

resembling those used in 4*71, we find that,

N>

n,

then

F(z,0 + c)
is

also a solution of (2).

The corresponding
expJJF(*,

solution of (1)

is

+ c)dz,
z.

and

this

is

a solution for

all
c,

values of c independent of

Hence, by

differentiation with respect to

we

find that the function u^ defined as

^c
is

[ex V

jF(z,0 + c)dz]j^

also a solution of (1);

and we have

u2 = u JF
l

dz,

so that

diu
U

da,

*Tz-

*dz~

= U F *

4-72,4-73]

DIFFERENTIAL EQUATIONS
solutions of (1)
is

117

But the Wronskian of any two


where C
If
is

a constant*; and so

a constant.

C= 0, Fis independent of 0, which is contrary to hypothesis ;soG^ 0, and


m,

J(C/F,).
;

an algebraic function of and similarly it is an algebraic function of all the functions of the types and ^ which occur in t. Next consider any function of the type which occurs in t we write
m, is
;

Hence

&), and, by arguments resembling those used in 471 and those used earlier in this section, we find that the function u., defined as
3
ris
r

= G (z,

exp }&(z, c<d)dz

a solution of (1)

and we have

so that

__

i(j

_=

,,.eG 9

This Wronskian

is

a constant,

C and
lt

so

,-V{W(?.)l.
Consequently u
the types
! is

an algebraic function, not only of all the transcendants of and ^, but also of those of type which occur in t and therefore
x

is

of order

iV.

This

is

contrary to the hypothesis that u 1

is

of order

fi

l,

where f<L^N,ifN> n.

The
of

contradiction shews that


is n.

N cannot be greater than n

d (log Uj)/dz 4
#

And

this is the

hence the order theorem to be established.


;

73.

Liouville's theoi-emf that Bessel's equation

has no algebraic integral.

We

shall

now shew

that the equation

We

has no integral (other than a null-function) which first reduce the equation to its normal form

is

an algebraic function of

z.

by writing

y=uz~^, p

= v-%-

* See e.g. Forsyth, Treatise on Differential Equations (1914), 65.

t Journal de Math. iv. (1839), pp. 429435 ; vi. (1841), pp. 47. Liouville's first investigation was concerned with the general case in which x (z) is any polynomial the application (with various modifications) to Bessel's equation was given in his later paper, Journal de Math. vi.
;

(1841), pp.

113,

36.

118

THEORY OF BESSEL FUNCTIONS


is

[CHAP. IV

This

of the form

dhi

-"%<*>
,

where
(2)

X (.).(l)-i.
an algebraic integral then (1) also Let the equation which expresses this integral, u,
;

If possible, let Bessel's equation have

has an algebraic integral.

as an algebraic function of z be
(3)

64(u,z)
is

= 0,
;

where 64

a polynomial both in u and in z

and

it is

supposed that 64

ik

irreducible*.

Since u
(4)

is

a solution of (1)

we have
0.

64uu 64z* - 2>64uz 64u 64z + 64M 64U* + 64u3 uX (*) =


(3)

The equations
of (3) satisfy
(4).

and

(4)

have a common

root,

and hence

all

the roots

For, if not, the left-hand sides of (3)

have a highest common polynomial in u and in z.


hypothesis.

and (4) (qua functions of u) would than 64 itself, and this would be a factor other reducible, which is contrary to be would Hence 64
uM
Then,
if s is

Let

all

the roots of (3) be

,, ,
tt/

...

any positive integer,

+ W *+
2

+ Ujg*

is

a rational function of z ; and there for which this sum is not zerof.

is

at least one value of s not exceeding

Let any such value of

s be taken,

and

let

=i

Also

let

= s (s - 1)
Since

. . .

(8

- r + 1) j^ .-* (-^)'
are
all

where r =
prove that
<*\

1, 2, ... 5.

** 2

... m,,

solutions} of (1),

it is

easy to

dWo

-w
1

(6)

^=W^ + r(s-r+l) X (z)W .


r
is to say,

l ,

(r

1, 2, ...

- 1)

That

t If not, all the roots of (3) +

in z or in both u and -J^ has no factors which are polynomials in u or would be zero.

Because

(4) is satisfied

by

all

the roots of

(3),

qua equation in

u.

4 *73]

DIFFERENTIAL EQUATIONS

119

Since
so that

is

a rational function of

z, it is

W=
and

X A n z+2

expressible in partial fractions,

BW

'\H>
positive

where

An

Bn>q

are constants, k

integral values only in the last

and X are integers, n assumes summation and aq 0.

Let the highest power of l/(z


It follows
l

- aq)

which occurs in

be lj(z

aq ) p

Kz -

9 ) in

by an easy induction from (5) and p+r where r = r is lj(z - a g ) 1, 2,


,

(6) that the highest


... s.

power of

Hence there is a higher power on the left of (7) than on the right. This contradiction shews that there are no terms of the type 2? n n>9 (z - a q )~ in a and so

= X A n z\
n=
-it

We may now
have a
last

assume that
if it

term

because this expression for does not vanish identically.

A K ^ 0,

must

From (5) and (6) which occur in


,

it is
lt

W W W W
,

easy to see that the terms of highest degree in z ... are* 2 3


,

A K z\ XA K z K ~\ A K sz\ XA K (Zs-2)z k -\....


By
in

a simple induction

W,r h

shew that the term of highest degree A K z*.l.3...(2r-l).s(s-2)...(s-2r+2).


it is

possible to

An induction of a more complicated nature term of highest degree in v+l is

is

then necessary to shew that the

^z*->2.4,...(2r).(s-l)(s-3)...(s-2r+l). 2F (h-ls;l-sl

l) r+1

where the
If s

suffix

+1

indicates that the first r

+1

terms only of the hyper-

geometric series are to be taken.


of degrees
odd, the terms of highest degree on the left and right of (7) are X - 2 and X respectively, which is impossible. Hence vanishes whenever s is odd.
is

When
That
and
is to

s is even, the result of

equating coefficients of zx ~ in (7)


x

is

say

XAi.sl^-XAi.sl^d, -te i_to XA^sl^Q,- %s; _*; J) = 0,


XA K .sl
,

l)is

so,

by Vandermonde's theorem, 2.4.6... ..


,

The
t

expression on the

left

1.3.5...(-1) vanishes only when X is


iu
vi.

zerof.
is

* It is to be

remembered that the term of highest degree

seems to fail at this he apparently overlooked the possibility of \ vanishing. The failure seems inevitable in view of the fact that J* 2 (z) + J _ n _ h (*) is an algebraic function of *, by 34. The +{ subsequent part of the proof given here is based on a suggestion made by Liouville, Journal de Math. iv. (1839), p. 435; see also Genocchi, Mem. Accad. delle Sci. di Torino, xxm. (1866), Comptes Rendiu, lxxxv. (1877), pp. 391394. pp. 299362
(1841), p. 7,

The

x (2)

1.

analysis given by Liouville, Journal de Math.

point, because

120

THEORY OF BESSEL FUNCTIONS

[CHAP. IV
and
that,

We

s is even,

have therefore proved that, when n is expressible in the form

s is odd,

vanishes,

when

2 A ny8 z- n
where

0)g

does not vanish.


expresses the coefficients in an equation

From Newton's theorem which


in terms of the

sums of powers of the roots, it appears that and that the equation (u, z) = is expressible in the form
(8)

M must be even,

u M + 2 u M~"
r=l

4 if

(1/*)

= 0,

where the functions T$ r are polynomials

in \jz.
l/'z,

When we

solve (8) in a series of ascending powers of


is

we

find that

each of the branches of u

expressible in the form

m=0
a positive integer and, in the case of one branch at least, c does not vanish because the constant terms in the functions r are not all zero.

where n

is

And

the series which are of the form

2 cm z- m,n
wi=0

are convergent* for all sufficiently large values of

z.

When we

substitute the series into the left-hand side of (1),


,

we

find that

the coefficient of the constant term in the result is c and so, for every branch, c must be zero, contrary to what has just been proved. The contradiction thus obtained shews that Bessel's equation has no algebraic integral.

4*74.

On
are

the impossibility

of integrating Bessel's equation in

finite terms.

We

now

in a position to prove Liouville's

theorem f that Bessel's

equation for functions of order v has no solution (except a null-function) which is expressible in finite terms by means of elementary transcendental
functions, if
2i> is

not an odd integer.


Bessel's equation to its

As

in 4-73,

we reduce

normal form

where x ( z )

= ~ 1 + P (P + *)/**
d
(log u)/dz

Now

write

t,

and P = v ~ and we have

%'

* Goursat,

* + , + i-*k!>-a
Cours d* Analyse,
ii.

(Paris, 1911),

convergence of a series derived in this


t Journal de Math.
vi. (1841),

pp.

pp.273281. Many treatises manner from an algebraic equation. 113, 3G.

tacitly

assume the

4*74]

DIFFERENTIAL EQUATIONS
x iz)
is

121

Since

f order zero,

it follows

from

has an integral expressible in

finite terms,

4"72 that, if Bessel's equation then (2) must have a solution

which

is

of order zero,

i.e.

it

must have an algebraic

integral.

integral,

If (2) has an algebraic integral, let the equation which expresses this t, as an algebraic function of z, be

(3)

a(t,z) = 0,
is

where S4
Since
(4)

an irreducible polynomial in
a solution of
(2),

and

z.

t is

we have

S4z +{ x {z)-t^s4t = Q.
t

As

in the corresponding analysis of 4*73, all the branches of

satisfy (4).

more than two branches of t, and let three of them be called tlt t2 t^, the corresponding values of u (defined as exp jtdz) being uly v?, us These functions are all solutions of (1) and so the Wronskians
, .

First suppose that there are

dus
dz

du*

duj

dtu

du2
dz
2

dux

dz

dz
lf
,

dz
3.

dz

are constants, which will be called

G C2 C
duz
Ua

Now

it is

easy to verify that

n
and
t3

dus
rf

~d~

~ U* Ua

'

~"

^'

is

not zero, because,

if it

were

zero, the

equation (3) would have a

pair of equal roots, and would therefore be reducible.

Hence G

0,

and so
UtU3

= dfcU -

2 ).

Therefore u^u^ (and similarly m3 Mj and u^u^)

is

an algebraic function of z.

But
and therefore ux
is

Wl

=A

/^-^,
in 4-73,

cannot be the case, and so

an algebraic function of z. This, as we have seen t has not more than two branches.

Next suppose that t has two branches, so that S4 (t, z) is quadratic Let the branches be U V V, where U and V are rational functions, of z.
substituting in (2)

in

t.

By

we

find that

K)
Let

{V'
factorised so that

+ 4>UV=0.

V be

V=Az*Tl(z-a q y<i,
where

is

constant, Kq

and \ are

integers,

and

k,{

and a q are not

zero.

122

THEORY OF BESSEL FUNCTIONS


From
the second

[CHAP. IV

member

of (5)

it

follows that

\z

(z

aq

and then by substituting into the


<>

first

member

of (5)

we have
(,),o.
.

^ + f4i^ + {s +2 4(r^)F + ^n(,-a,r'- X


2,
and,
if
is

Now
is

consider the principal part of the expression on the left near a q

It

evident that none of the numbers x q can be less than

any one

of

them

greater than

it

must
Kq

satisfy the equation

T!/f q

=s

V/>

so that Kq is

or
tc

4, which

are both excluded from consideration.

Hence

all

the numbers

are equal to

2.

Again, if

power in

we consider the principal part near V must cancel with the 1 in % (z), so
xJV
is rational,

oo

we

see that the highest


.

that

\ = 2 Kq
9

It follows that
is

and consequently (t,

z) is reducible,

which

contrary to hypothesis.

Hence

cannot have as

many
x

as two branches and so


t

it

must be

rational.

Accordingly, let the expression for

in partial fractions be
7? -n, g

t= 2
where
If

A n zn + 2

r(*-o)*'
integers,

An
we

and 5 i9 are constants, k ancL.X are

n assumes

positive

values only in the last summation and aq


substitute this value of
t

0.
we
find that

in (2)

If

we

consider the principal part of the left-hand side near aq

we

see that

l/(z

aq )

cannot occur in

to a higher

power than the

first

and that

so that
Similarly, if

Bljq -&hq = 0, Bhq =l.


,

we consider the principal parts near and oo we find *=1, (A^y-JL-t-pip+l); X~0, 4,' = -l. Since p = v ^, we may take A- = p without loss of generality.
Y

that

It

then follows that

u
Accordingly,
if

= z-Pe i?U(z-aq ).
9

we

replace

u by z~ p e u w

in (1),

we

see that the equation

must have a

solution which is a polynomial in

z,

and the constant term

in

this polynomial does not vanish.

4'75]

DIFFERENTIAL EQUATIONS

123
the relation connecting

When we substitute 2 cmzm for w in (7) we find that


successive coefficients
is

m (m 2p 1) cm 2icm-! (m p - 1) = 0,
and so the
series for

w cannot terminate unless in p 1


is

can vanish,

i.e.

unless

is

zero or a positive integer.

Hence the hypothesis that Bessel's equation


of necessity to the consequence that one of the
positive integer;

soluble in finite terms leads


v \ is zero or a an odd integer.

numbers +
if,

and

this is the case

if,

and only

2v

is

Conversely we have seen ( 34) that, when 2v is an odd integer, Bessel's equation actually possesses a fundamental system of solutions expressible in The investigation of the solubility of the equation is therefore finite terms.
complete.

Some

applications of this theorem to equations of the types discussed in 4*3 have


xi. (1846), pp.

been recorded by Lebesgue, Journal de Math.

338

340.

4*75.

On

the impossibility

of integrating Riccati's equation in finite terms.

By means

of the result just obtained,

we can
brf

discuss Riccati's equation

r az + f
with a view to proving that
It has
it is,

in general, not integrable in finite terms.


is

been seen

( 4*21)

that the equation

reducible to

where

n=2q 2;

and, by 4-3, the last equation q = 0.

is

reducible to Bessel's

equation for functions of order l/(2g) unless

Hence
finite

the only poss-ible cases in

terms are thpse in which q are precisely the cases in which n

is zero
is

which Riccati's equation is integrable in or 1/q is an odd integer ; and these


to

equal

or to

4w 2w +

(i-0,
l

1,

2,

...)

Consequently the only cases in which Riccati's equation is integrable in finite terms are the classical cases discovered by Daniel Bernoulli (cf. 4'1 1) and the limiting case discussed after the manner of Euler in 412.
This theorem was proved by Liouville, Journal de Math. vi. (1841), pp. 1 13. It seems impossible to establish it by any method which is appreciably more brief than the
analysis used in the preceding sections.

124
4*8.

THEORY OF BESSEL FUNCTIONS


Solutions of Laplace's equation.
first

[CHAP. IV

The
seen
(

1*3) to

appearance in analysis of the general Bessel coefficient has been be in connexion with an equation, equivalent to Laplace's

equation, which occurs in the problem of the vibrations of a circular membrane.

We

shall

now shew how

Bessel coefficients arise in a natural

manner from

Whittaker's* solution of Laplace's equation


(1)

^ + ^ + ^=
solution in question
is

The
(2)

V
f denotes

f(z + ixcosu
n

-f

iy

smu, u)du,

in which

an arbitrary function of the two variables involved.


is

In particular, a solution

ek
J

(z

+ ixcosu + iysinu) cog

mu

^
by the equations

in

which k
If

is

any constant and

m is

any integer.

we take

cylindrical -polar coordinates, defined

x = p cos
this solution

<p,

y = p sin

<f>,

becomes
eiipcog(u-*)
|

0**
J

cosmM ^ M=::e*z

eik* CMV cos

m (v +

it

<f>)

dv,

= 2ekz

e^ 008 " cos mv cos m<f> dv,


.

J o

= 2iri m e** cos m<$> Jm (kp),


by
2-2.

In like manner a solution


I

is

gk
IT

(z

+ ix cos u + iy sin u)

sm mu

and

this

is

equal to 2vim e kz sin m<p

Jm (kp).

Both of these solutions are

analytic near the origin.

Again,
nates,
it is

if

Laplace's equation be transformedf to cylindrical-polar coordi-

found to become

d^V
dp'
*

ldV
p dp

1 p*

d*V
d<f>*

F
'

dz*

Monthly Notices of

the

R. A. S. lxii. (1902),

pp.

617620; Math. Ann.

lvii.

(1902),

pp. 333341. t

The

simplest

method

of effecting the transformation is by using Green's theorem.

See

W. Thomson, Camb. Math.

Journal,

iv.

(1845), pp.

3342.

4-8, 4-81]

DIFFERENTIAL EQUATIONS
this equation of

125

and a normal solution of

which ekz

is

a factor must be such that

lcTV
Vd<p*
is

independent of

to

m where
2

factor of

is to be one- valued, it must be equal Consequently the function of p which is a must be annihilated by
<f>,

and, if the solution

is

an integer.

dp 3

p dp

\
if it is

and therefore
line

it

must be a multiple of Jm (Jcp)

to be analytic along the

0.

We

thus obtain anew the solutions


e**
.

sin

m<b r

Jm (kp). r/
'

These solutions have been derived by Hobson* from the solution Maxwell's method of differentiating harmonics with respect to axes.

^*J

(kp)

by Clerk

Another solution of Laplace's equation involving Bessel functions has been obtained by Hobson (ibid. p. 447) from the equation in cylindrical-polar coordinates by regarding d/dz as a symbolic operator. The solution so obtained is
cos
.

sin

m+ 'V(pz)M>

is an arbitrary function but the interpretation of this solution when <@ involves m a function of the second kind is open to question. Other solutions involving a Bessel function of an operator acting on an arbitrary function have been given by Hobson, Proc. London Math. Soc. xxiv. (1893), pp. 5567 xxvi. (1895), pp. 492494.
;
;

where/(s)

4'81.

Solutions of the equations of wave motions.

We
K }

shall

now examine

the equation of wave motions

dx2
t

dy3

dz 3

~&

dt 2

'

in which

represents the time and c the velocity of propagation of the waves,

from the same aspect.


Whittaker'sf solution of this equation
(2)
is

V=
J -IT

JO

f(x sin u cos v + y sin u

sin v

+ z cos u 4- ct,

u, v)

dudv,

where

/ denotes

an arbitrary function of the three variables involved.


is

In particular, a solution

-r
where

gik(xainucosn + yainuainv+zcoiiu + et)

f,
i

J? /

dU fJv

F denotes an
London Math.

arbitrary function of
Soc. xxn. (1892), pp.

u and

v.

* Proc.

t Math. Ann.
(1904),

431449. 342345. See also Havelock, Proc. London Math. pp. 122137, and Watson, Messenger, xxxvi. (1907), pp. 98106.
lvii. (1902), pp.

Soc. (2) n.

126

THEORY OF BESSEL FUNCTIONS


The
physical importance of this particular solution
lies in

[CHAP. IV
the fact that
kc.
it

is

the general solution in

which the waves


(a?,

all

have the same frequency


(r, 0,
<f>),

yfr) be the which the passes through the polar axis is the direction (6, <f>) and the plane y\r = .s-axis. The well-known formulae of spherical trigonometry then shew that

Now let the polar coordinates of angular coordinates of the direction

y, z)

be

and

let (m,

(u, v) referred to

new axes

for

cos

a>

= cos

+ sin sin u cos (v = sin a> sin sin u sin (v


cos u
<f>)
yfr.
;

<f>),

take the arbitrary function F(u, v) to be notes a surface harmonic in (u, v) of degree n we

Now

8n (u, )sin u, where Sn may then write

de-

Sn (U,
where Sn
is

V)

= Sn (0,
(a>, yfr)

<f> l

O), i/r),

a surface harmonic* in

of degree n.

We

thus get the solution

Vn = e**
Since

(*
J

e*rc08 *- Sn (0,

<f>

to,

ty) sin wdtodty.

8n

is

a surface harmonic of degree n in

(to, sfr),

we may

write

Sn (0,

<*> 5

<0,

t)

- ^ (0,

4>)

Pn ( C0S )

+
m=l

{4 n<> (0,#) cos myfr


<)

+ BnM (0,
(0,
<f>)

<f>)

sin mi/rj

Pnm (cos

o),

where 4

(0, 0),

A< w) (0,

and

Bn

< TO)

are independent of
-^r,

w and

^r.

Performing the integration with respect to


ikel

we

get

Vn = 1ire A n (0, 4>) r^trmm Pn (cos o>) sin a dm


= (2*)Ue^^^A n (0,4>)
by

332.

Now the equation of wave motions is unaffected if we multiply x, y, z and by the same constant factor, i.e. if we multiply r and t by the same constant unaltered so that A n (0, <f>) may be taken to be inand factor, leaving dependent f of the constant k which multiplies r and t.
t
;

<f>

Hence lim (k~ n Vn )


to say, r n

is

a solution of the equation of wave motions, that


t)

is

A n (0,
is

<) is

a solution (independent of

of the equation of

wave
<f>)

motions, and

consequently a solution of Laplace's equation.


fact that Laplace's operator is

Hence

A n (0,

* This follows

from the

an invariant

for

changes of rectangular

axes.

t This

is

otherwise obvious, because

Sn may be taken independent

of k.

4-82]
is

DIFFERENTIAL EQUATIONS
n. If we assume it to be permissible any such harmonic, we obtain the result that

127
to take

a surface harmonic of degree


<f>)

A n (#,

to be

eikct ?-iJH+i (kr) *


is

Pnm (cos 6) C0S md> r


8111

of the equation of wave motions*; and the motion represented by this solution has frequency kc.
solution,

To

justify the

assumption that

A n (0,

<f>)

may

construct the normal solution of the equation of

be any surface harmonic of degree wave motions

n,

we

which has factors of the form e*et


the

sin

md>. r

The

factor
is

which involves 6 must then be of


annihilated by the operator

form Pnm (cos 0) and


;

the factor which involves r

*("*) -<++*'.
so that
if

this factor is to

be analytic at the origin

it

must be a multiple

of

Jn + {hr))^,:

4*82.

Theorems derived from solutions of

the equations

of Mathematical

Physics.
It is possible to prove (or, at any rate, to render probable) theorems concerning Bessel functions by a comparison of various solutions of Laplace's

equation or of the equation of wave motions.

Thus,

if

we take the

function

e^t/o [k>s/(p*

+a 2

lap cos

<f>)},

by making a change of origin


a factor and
expect
it to
it is

to the point (a, 0, 0),

we

of Laplace's equation in cylindrical-polar coordinates. analytic at all points of space.

see that it is a solution This solution has e** as

It is therefore natural to

be expansible in the form

AJ
Assuming the
consideration
in p
is

(kp)

+ 22 (A m cos m<f> + Bm sin m<f>) Jm (kp) =


Wl
l

and

a,

we observe that the function under an even function of 0, and so Bm = and, from the symmetry ^4 m is of the form c m Jm (ka), where cm is independent of p and a.
possibility of this expansion,
;

We

thus get
Jo {kx'Xp*

+a
2

lap cos

<)}

= 2 em cm Jm (kp) Jm (ka) cos m<f>.


m=0

If

we expand both

sides in powers of p,
<)

a and cos

coefficients of (k-pa cos

m we
,

<f>,

and compare the

get

* Cf.

Bryan, Nature, lux. (1909),

p.

309.

128

THEORY OF BESSEL FUNCTIONS


so

[CHAP. IV

and

we

are led to the expansion*

Jo
of which a

{W(ps + a*-2ap

cos

<p)}

= 1
m=

m Jm (kp) Jm (ka) cos


in 112.

m<f>,

more formal proof will be given


if

Again,

we take e ik{ct+z \ which is a solution

of the equation of wave motions, axis of z from


this expression

and which represents a wave

+
to

oc to

moving in the direction of the with frequency kc and wave-length 27r/k, we expect

be expansible + in the form


h

(^) \K)J
where cn
is

iket

c n i*

=o

JnH (kr)P n (cos 0),

a constant

so that

eikrco *

=
(If)*

CninJn+i {kr)
n

Pn (cos
side,

e)

If

we compare the

coefficients of (kr cos 6)

on each

we

find that

n \-^ and
so cn

1T)

2+*r(n + f)"2.(n!r

we

are thus led to the expansion}:

ikrc se

= fiE) i

\krj =o v
will

(n

+ l)iJn+ i(kr)P n (cos0),


11 5.

of

which a more formal proof


4*83.

be given in

Solutions of the wave equation in space of p dimensions.

The

analysis just explained has been extended by

Hobson

to the case of

the equation

dx?

dx

'"

'

dxp*

c2 dtis

normal solution of this equation of frequency kc which function of r and t only, where
r

expressible as a

V(#i

a^*

. . .

+ xp

-),

must be annihilated by the operator


a?2

9r

^
,

and

ikct so such a solution, containing a time-factor e

must be of the form

e
*

iket

<& ilp -*

(kr)l{kr)*^K

This

is

due

to

Neumann, Theorie

der Bessel'schen Functionen (Leipzig, 1867), pp. 59

65.
z.

harmonics do not occur because the function is symmetrical about the axis of X This expansion is due to Bauer, Journal filr Math. lvi. (1859), pp. 104, 106. Proc. London Math. Soc. xxv. (1894), pp. 4975.
t

The

tesseral

4-83]

DIFFERENTIAL EQUATIONS
f

129

Hobson
of rank p
;

describes the quotient ^\{ P-i)(kr)l(kr)^fh %) as a cylinder function such a function may be written in the form

<$(kr\p).

By

Hobson succeeded
integral order

using this notation combined with the concept of jp-dimensional space, in proving a number of theorems for cylinder functions of

and of order equal to half an odd integer simultaneously.


of such theorems
2

As an example

we
2

shall consider

an expansion

for

J {k v^r + o where
it is

2ar cos

<j>)
\

p),

xp by the equation xp r cos This function multiplied by e**ct is a solution of the wave equation, and when we write p = r sin it is expressible as a function of p, t and of
<f>

convenient to regard

as being connected with

<f>.

<f>,

<f>,

no other coordinates.

Hence
** J
is

{k

V(r2

+ a? -

2ar cos

<f>)

p)

annihilated by the operator

dp2
that
is

dp

dx

'

to say,

by the operator

&
,

J>~1
r

(ff-2)cos4> d
r*sin<j>
d<f>

dr*

dr

r*d<p*

Now normal functions which are annihilated by this operator are of the form
"

(hr)ir->

pn (cos <p\p),
+ cpy-*P.

where

Pn (cos

<p

p)

is

the coefficient* of on in the expansion of


(1

- 2a cos
we

<f>

By

the reasoning used in 4*82,


2

infer that

J \k V^r +

a*

2ar cos
1

<f>)

p)

00
1

~ (/fca^-^/fcr)^-

AnJn+ip~ ( kr ) ^h-Jp-i (ha) Pn (cos


1

<f>

\p).

Now
(k?ar cos

<f>)

expand all the Bessel functions and equate the n on each side we find that
;

coefficients of

2"
2*"+ iP-

r (n + |p)
(

=
{2

An
+fc>-

T (n + ip
1
).

2"T(n + ^-l) nlV($p-l)

'

so that
*

A n = 2**-

+ %p - 1) T (|p Pn (cos

So

that, in Gegenbauer's notation,


j

p) S3 C*

p_ * (cos

0).

130

THEORY OF BESSEL FUNCTIONS

[CHAP. IV

We

thus obtain the expansion


{fcy'( r2
2

Jfr-,
(r

+a

+ a -2arcos<ft)j - 2ar cos <)il>_ *


i!

An
all

analytical proof of this expansion, which holds for Bessel functions of


is

orders (though the proof given here

valid only

when p

is

an

integer), will

be given in 114.

4*84.

Batemav's solutions of the generalised equation of wave motions.


systems of normal solutions of the equation

Two
*

'

dx1i

dxf

2 9a;,

dx4*

c2 dt*

have been investigated by Bateman*, who also established a connexion between


the two systems.
If

we

take

new

variables p, a, %,

yjr

defined by the equations

xx p cos 2,

x3 a cos -^r,
Xi

x2 = p sin %,
the equation transforms into

a sin ^r,

'

{dp 2

p dp

p 3x

2
)

[do*

+
<r

'

d*

o*

dyfr\

c* dt*

A
where

normal solution of this equation with frequency kc

is

J
3> is

(kp cos 0>) / (&<r sin 4>) a* (Mc+^+fc*),

any constant.
if

Further,

we

write

p=*r cos
so that(r, ^,
-^r,

0,

<r

= r sin <,

<)

form a system of polar coordinates, equation (2) transforms

into

O
Now

&V SdV
dr*

r dr

l&V
r* d<p*

c ot<t>-tfm<p
'

tV
dtp

r* 1

r cos <

&V +
9x
2

92

ld*V
'

r2 sin2 4*

W*
+ * c<)

c2

^'

normal solutions of this equation which have

e* (MX +"*

as a factor

are annihilated
3

by the operator
d_

dr3

r dr
182188; Proc. London Math.
Soc. (2) hi. (1905), pp.

* Messenger, xzxiu. (1904), pp.

111123.

4-84]

DIFFERENTIAL EQUATIONS

131

and since such solutions are expressible as the product of a function of r and a function of they must be annihilated by each of the operators
<f>

y
dr2
d

3 9

4X(X+1)
r*
'

rdr
l

9<

+ 4 \(\+l)+ (cot<t>-t*n<l>)^r \ ^ 2
3^>

** cos2 ^>

*
sin 2
'

where X

is

a constant whose value depends on the particular solution under

consideration.

The normal

solutions so obtained are

now

easily verified to

be

of the form
(kr)-1

J^ +1 (kr) cos$ sin"<f)


x
2

F (* - X,
X

^- + \ +
<f>

+ l

sin 2

*ox+*++**>.

It is therefore suggested that

J^ (kr cos
is

cos

<I>)

Jv (kr sin

<f>

sin

<I>)

expressible in the form


l

2 a A (kr)~

J^ (kr)

cos*< sin"<

2JP,

(~^ - \ ^^ + X +

sin 2 <f>)

where the summation extends over various values of X, and the coefficients a* depend on X and <J>, but not on r or <j>. By symmetry it is clear that aA where

= 6 X cos*<I> sin" 2 F
.

(^-J^ -

X,

^^ + X + 1

1;

sin 2 <l> \

b K is

independent of

<.

It is not difficult to see that

X = Ha* +
and Bateman has proved that

")

+ .

(n

= 0,

1, 2, ...)

We

shall not give

Bateman's
sin

proof,

which

is

based on the theory of linear


expansion of

differential equations,* but later ( 11*6)

we

shall establish the

J (kr cos

<f>

cos 4>)

Jv (kr sin

<f>

<I>)

by a

direct transformation.

CHAPTER V
MISCELLANEOUS PROPERTIES OF BESSEL FUNCTIONS
5*1.

Indefinite integrals containing a single Bessel function.

In this chapter we shall discuss some properties of Bessel functions which have not found a place in the two preceding chapters, and which have but

one feature in common, namely that they are


definitely elementary character.

all

obtainable by processes of a

We

shall first evaluate

some

indefinite integrals.

The recurrence formulae


(1)

3.9 (5) (*)

and (6) at once lead to the

results

V>%

dz

= *H- #v+i (z)> = -z-"+> r(? v_


y

(2)

[" 2-" +1

# (z) dz

(z).

To

generalise these formulae, consider

^z^f{z)%%(z)dz;
let this integral

be equal to
z"+>

{A

(z) <$ v (z)

+ B (z)

<&v+1 (z)},

where

(z)

and

B {z)

are to be determined.

The

result of differentiation is that

z"+>f(z)

V, (z) =

jH-i {
I

>

(z ) < v (z)

+ A (z)

2l

^
z

(z)-A(z)<@ v+1
(jr)

(z)\
)

+ z^
In order that
take

[B' (z) <@ v+1

B (z) <$

<*)}.

A (z)

A (z) =
it

B'

(z),

and B (z) may not depend on the cylinder function, we and then
/(s) .5 A' (z) +
2

^
z

A(z)

+ B (z).

Hence
(3)

follows that

** +1

\b"(z)

^~

B' (z)

J"

B (*)| # (*) dz
{B' (z) <@ v (z)

= *+

+ B (z) <@ v+1 (*)}.

This result was obtained by Sonine, Math. Ann. xvi. (1880), p. 30, though an equivalent formula (with a different notation) had been obtained previously by Lommel, Studien fiber die BesseVschen Functionen (Leipzig, 1868), p. 70. Some developments of formula (3) are due to Nielsen, Nyt Tidsshrift, ix. (1898), pp. 7383 and Ann. di Mat. (3) vi. (1901),
pp.

4346. For some associated


integrals

which involve the functions ber and

bei, see

Whitehead,

Quarterly Journal, xlii. (1911), pp. 338

340.

5-1, 5-11]

MISCELLANEOUS THEOREMS
is

133
(3),

The

following reduction formula, which

an obvious consequence of

should be noted:
(4)
I

V+

ir

^ 0) ds

= - (ft - v*)f'* - &r (z) dz


2 L

li

+ |> +1 ^v+1 (z) + (/*-) **# (*)].

5*11.

Lommel's integrals containing two cylinder functions.


simplest integrals which contain

The

derived from the Wronskian formula of


r J
(z)
/

312
'

two Bessel functions are those (2), namely


\

J'_ (z)

r / \ T - Jv (z)Jv

(z)

I/7T 2 sin -

which gives
K

J
/"*

zJ*(z)
dt

2sini/7r
7T
.

'

(z)
/_,, (.g)

9 Kl)
)

zJv (z) J_ <>) ~

2 sin

10g
*tt

'

(z)

and

similarly,

from

3*63 (1),
z

,ox

dz

_ir

Fy (>)

^
K
}

['
J

dx _ zYv\z)~

it

Jv (z)

(z)'

The reader should have no difficulty in evaluating the similar integrals which contain any two cylinder functions of the same order in the denominator. The formulae actually given are due to Lommel, Math. Ann. iv. (1871), pp. 103 116. The reader should compare (3) with the result due to Euler which was quoted in 1*2.

Some more
It is at

interesting results, also

due to Lommel*, are obtained from

generalisations of Bessel's equation.

once verified by differentiation that,


d*y

if

y and y
rt

satisfy the equations

d?n

then

j\P -Q)ynd^y^-v^.
*

Math. Ann.

xiv. (1879), pp.

520536.

134

THEORY OF BESSEL FUNCTIONS


Now
apply this result to any two equations of the type of
fi

[CHAP.

V
If

431

(17).

*^V> ^i,

denote any two cylinder functions of orders

and v

respectively,

we have

where

<f>

(z)

and

i/r

(2) are arbitrary

functions of

z.

This formula
<f>

is

too general to be of practical use.


Iz.

As a

special case, take

(z)

and
Z

yjr

(z) to

be multiples of z, say kz and


v

It is then found that

(7)

{{^-P)z-^^\^{kz)^
dz

(lz)dz

= ,fa(kz) d ^-^(lz)

^
dz
further in two special cases (i)fiv,
right.
is

= z {k^ +1 (kz) W. (Iz) - 1&* (kz) # r+1 (lz)} -(/*-) ^M (kz) <@ v (Iz).
The expression on the left simplifies
(ii)k=*l.
still

If

we take
,

fi

v, it is

found that

(8)/

(*.)

i. { u) d* =

'<ww>yww (Wi
by differentiating the expression on the
I,

This formula

may be

verified

It becomes nugatory numerator is a constant.

when k =

for

the denominator

then zero, while the

If this constant is omitted,

an application of

l'Hospital's rule

shews that,

when
(9)

I -*- k,

z<@ (kz)

J"

V (kz) dz^-^

{kz <^M+1 (kz)

#/ (kz)

- ks&r (kz) W^(kz) - ^V (kz) ^ M+1 (kz)}.


remove the derivates on the
is

The

result of using recurrence formulae to

right of (9)
Z

(10)

l z<&t (key* (kz) dz

= \z* {2^ (kz) & (kz) - ^_

(A*)

#M+1 (kz)

5*11]

MISCELLANEOUS THEOREMS

135

Special cases of these formulae are

(11)

fs <$,? (kz) dz = %z* {%* (kz) -

(for)

#+,

(**)}

- \* {(l " ^) *V <*) + ^m'


(12)
[*?<&,, (kz)

<**)}

LM (kz) dz = *

{2^M (kz) iLM (fa) +

^
^
,

(fa)

f _M _x (fa)

+
the latter equation being obtained by regarding
function of order
e~'
t 'rt

+1

(fa)iU +1 (fa)},
as a cylinder

^ M (kz)
k=
I

/*.

To obtain a
is

different class of elementary integrals take

in (7)

and

it

found that

i3)

/y <*.)*.
result of

(*)

- - fat^w^y^w^wt
ff,(fa)ff,(fa)
,

The
(14)

making

j/

-*- /*

in this formula is 8
+1 (fa)

| V, (fa) tf,(fa)

^ = g j^
-

-^|

(*)

(fa)| 3^ M+1 g-

^(fa)^M (fa)
2.

The

last equation is also readily obtainable

by multiplying the equations


(z)

V^(,)0, V^ d
by \<f*(z)

^^ = 2f^

li

-^^(z)
z

respectively, subtracting

and integrating, and then

re-

OI&

placing z by kz.

As a
(is)

special case

we have

jj(kz)^ -g{/

w (fa)gKfa)-^(fa)gu
this result will

+1

(fa))+~j

|l

-(fa)L

An

alternative

method of obtaining

be given immediately.

Results equivalent to (11) are as old as Fourier's treatise, La Theorie Analytique de la Chaleur (Paris, 1822), 318 319, in the case of functions of order zero but none of the other formulae of this section seem to have been discovered before the publication of

Lommel's memoir.
Various special cases of the formulae have been worked out in detail by Marcolongo, Napoli Rendiconti, (2) in. (1889), pp. 9199 and by Chessin, Trans. Acad. Sci. of St Louis, xii. (1902), pp. 99108.

136
5"12.

THEORY OF BESSEL FUNCTIONS


Indefinite

[CHAP.

integrals containing two cylinder functions;

Lommel's

second method.

An

alternative

of the integrals just discussed.

method has been given by Lommel* for evaluating some By this method their values are obtained in a
adding the two results
{<&> (z)

form more suitable for numerical computation.

The method

consists in

^
so that
(p

[z

V, (z) K. (z)} = - z>

tf r+1 (z)

+9

(z)

(,)}
1

+ (p +
{zo S? M+1 (z)

/M

+ V) Z>~ # {Z) V$ v (Z),

& v+

>

CO}

=*

[V, (z)

W v+X (z) + ^M+1 (z) (z)}

+ (p- H,-p-2)z'>-> <@+1 (z) W v+1 (z), + ft + v) j'z'1

(z) <@ v

(z)dz

+ (p-p-v- 2)

z"- 1 <@r+l (z)

+1

(z)

dz

= Z>
and then giving
special values to p.

{<&

(jf)

v (jr)

+ ^M+1 (0

+1 (,)},

Thus we have
e

(1)

z-->-^+1 (z) & p+1 (z) dz


2(^ + y +l)
l

(2)
J

^
if

^
)

(z)

*'

{z)

+l ( *>

<*>!

^()?r Wdf- f

~7^
v+

i^(f),(f) +

^(f)?rH (i)i,

As
(3)

special cases of these

jV*- 9 V+1 <#) df


1

^g
=

{, (*) + *V
(*)

<*)}.

rz

(4)

+1 r

(*) <fc

^_ W

+ *V+i (*)}

Again,

p be made zero,

it is

found that

- ^M (*) ^, (*) + ^M+l (*) H1 (*)>


so that,

by summing formulae of this


rz

type,

we get
rz

(5)

0*

+
)J

J~
(,) V (,)

M + + 2n) J

^M+n (*) *f,+ (*) ~


,

A*

- ^M () ^r (#) + 2 2 V,^ ()
* Math. Ann. xiv. (1879), pp.

^+

(#)

+r ^+ (*) ^*+ (*)

530536.

5'12-5*14]

MISCELLANEOUS THEOREMS
if fi

137

In particular,
(6)

= v = 0,

/'*.(#)*.(*)*

where n

1, 2, 3, ....

But there seems


/<

to

be no simple formula

for

%(z)<& (z)^,
Mag.
(5) xi. (1881), p. 217.

For a
I.

special case of (1) see Rayleigh, Phil.

[Scientific Papers,

(1899), p. 516.]

5*13.

Sonine's integrals containing two cylinder functions.

analysis of 5-1 has been extended to the discussion of conditions that

The

by Sonine, Math. Ann.

xvi. (1880), pp.

30

33,

may be expressible

in the

form
M {0 <)}W W {+ ()>+* <*)

A (*)

^M+1 ft

(*)}

f\,

{*(*)}
1

+^(*)^{*(*)}fr+1 {^(*)}+^(*)IH4 {*(*)>^ {^W}.


but the results are too complicated and not
here.
sufficiently

important to justify their insertion

5*14.

Schafheitlin'8 reduction formula.

A reduction
which

formula for
\' *<&*() dz,

is

a natural extension of the formula

5*1 (4),

has been discovered by

Schafheitlin* and applied by


<&9 {z) as v varies ( 156).

him

to discuss the rate of

change of the zeros of

To obtain the formula we observe that

= [- ***<&. (z) <@: (*>] +


Now, by a
(fi

*
1

{*+#/ (*)+(,*

>

**+

%%

(z) 'w:

(* dz.

partial integration,

+ 3) f *#*<&* (z) dz |>+ #/ (*)]

+2

V <@: (z) {*&:

(z)

+ (* - *>

(^)j cte,

* Berliner Sitzungsherichte, v. (1906), p. 88.

138

THEORY OF BESSEL FUNCTIONS

[CHAP.

and so
(fi

1)
J

V+

<&,'* (z)

dz

= |>+

W
2

(z)]

J'**

(z*

2
i/

# (z) <&,' (z) dz.

Hence, on substitution,
(/*

1) [

V {? 2

1/

# (z) dz
1)

= |>+3 ^V
+

(*)

- (ji +
3

^+

(*)

#/ (*)]
{(/a

2 [
'

V+
(*)

<@v (z) <@: (z) dz


1)

l)2

M j"V+#
2

(*)

^V(s) ek

= |>+

- (/*+
Z

^+

^(*)

#/ (2) + z* +3 * (z) + {(/* + 1)2 -"'l* *


4

^ ^)]
3

-(p+3)j
By
(fi

z+*'@vi (z)-(ti

1) {

(/*

+ 1) -

v*\

J'rW(t)dz.

rearranging

we

find that

+ 2)

** #,
J*

(5) (2c

= (n+ 1)

(i/

-\

(jjl

1 )2 } [

V
2

#2 (*)
2

<2e

+ * |>+'
and

{*#.' (*)

- * (^ + 1 ) # (*)} + *+ <* - v + J (a* +


2

1 )}

^8 (*)],

this is the reduction formula in question.

5*2.

Expansions in
shall

series

of Bessel functions.

We

now

discuss
2*7.

some of the simplest expansions of the type ob-

tained for (\z)m in for Chapter xvi.

The general theory

of such expansions

is

reserved

The
(1)

result of

27

at once suggests the possibility of the expansion

to Y .l<**p!$ j^.1
is

which

due to Gegenbauer* and

is

valid

when

ji is

not a negative integer.

To

establish the expansion, observe that

n=

ni

is a series of analytic functions which converges uniformly throughout any bounded domain of the 2-plane (cf. 3'13); and since

-^ {(\z)- J+ M (z)}
it is

= ^7^ {nJ^+m-i (z) - (A* + ") ^m+2h+i (*)},


now under
w
consideration
r
is

v-m

evident that the derivate of the series

s r(u + w)
n
-

,.

r(At+n +
-o
(2),
-

i)

1
J

Lm-o
*

Wiener Sitzungsberichte, lxxiv.

(1877), pp.

124130.

5*2, 5*21]

MISCELLANEOUS THEOREMS
a constant.

139
the constant

and so
is

the' sum is
is

When we make z--0, we see that

unity; that

to say

and the required


The
this is

result

is

established.
it is

reader will find that

not

difficult to verify
z,

right in (1) is rearranged in powers of

all

that when the expansion on the the coefficients except that of z* vanish ; but

a crude method of proving the

result.

5*21.

The expansion of a Bessel function as a

series

of Bessel functions.

The expansion
(i)

($z)-'J (z)

= az)-r( v +
v

i-fi)

5
.to

Qi+2n)r(g+n)
nir(F + l-/*-)r(ir + n +
/j.

l)

T jM+2n

W
.

is

is

a generalisation of a formula proved by Sonine* when the difference v ft are not negative a positive integer ; it is valid when fi, v and v

integers.

It is

most

easily obtained

by expanding each power of z

in the expansion

of i\zY~"J,(z) with the aid of 5'2, and rearranging the resulting double series, which is easily seen to be absolutely convergent.
It is thus found that

(*zY

*'

\z )

n/

TT\
*
l) pto
(fj,

(-) m!r(i, +m+ ~i: o

_ 2

+ 2m + 2p)r(n + 2m+p)

/*-h*W

_ v
- I

(~) m v Qt + 2n)r(/i-hm + w ) J , (^^T)! ",tom!r(ir + m + l).t.


i

W
/\

(-)Tfr + m + n)
V(p + n)r(v+l-fi)
;

v
".?.

+ 2 ")^' !I> + l--,0I> + n+l) ("


,

,Q\r

by Vandermonde's theorem
If

and the
find that

result is established.

we put

v fj.

+ m, we

which

is

Sonine's form of the result, and

is

readily proved

by induction.

* Math. Ann. xvi. (1880), p. 22.

140

THEORY OF BESSEL FUNCTIONS


By
a slight modification of the analysis, we

[CHAP. V
that, if

may prove

is

any

constant,
(8)

(**)-*(*)-*.?.,,#&
x 2F,(/x

+ n, -n;

+ 1;

h?) (fi

+ 2n) J^+^iz).

This formula will be required in establishing some more general expansions


in 11-6.

5*22.

Lommel's expansions of(z

+ h)^ v Jv [\/{z + h)}.

It is evident that (z

+ h)~^ v Jv {\/(z + h)},

for all values of the variable,

qua function of z + h, is analytic and consequently, by Taylor's theorem combined


ao

with

3-21 (6) ;

we have
*

(1)

(,

+ h) -4.Jp y (z +h)] . J

km

tJm

{g

-* Jw Wz)]

Again, (z

+ h) "Jv {<J(z+ h)}


| j

is

analytic except

when

z+h = 0;

and

so,

provided that h
(2)
(,

<

,
\

we have
m-0

+hpj

W(z + h)}=Z

^ !*- [zJ
"*J ***

(V*)}

(2)

These formulae are due to Lommel*. If we take v \ in (1) and = ^ in we deduce from 3*4, after making some slight changes in notation,
i/

) COsV(^-2^)= 2 ~,Jm- h {z),

equation (4) being true only when J*|<i|*i- These formulae are due to Glaisherf, who regarded the left-hand sides as the generating functions associated with the functions whose order is half of an odd integer, just as exp {\z (t ljt)\ is the generating function associated with the Bessel coefficients.

Proofs of (3) and (4) by direct expansion of the right-hand sides have been given by Glaisher the algebra involved in investigations of this nature
;

is

somewhat formidable.

* Studien iiber die Bessel' schen Functionen (Leipzig, 1868), pp. 11 16. Formula (1) was given by Bessel, Berliner Abh. 1824 [1826], p. 35, for the Bessel coefficients. t Quarterly Journal, xn. (1873), p. 136 ; British Association Report, 1878, pp. 469470. Phil. Tram, of the Royal Soc. clxxii. (1881), pp. 774 781, 813.

5-22]

MISCELLANEOUS THEOREMS
shall

141

We
(5)

now enumerate

various modifications of (1) and (2).


z*

In (1) replace z and h by

and kz z and then


,

jv {W(i + *)} = a +
/ (z V2)

wi

(-) w (i^) ?

m=0

V
1,

y+w (^)

>

and, in particular,
(6)

- 2* 5 fcJ^if^J^ (*).
and then make k -*

If

we

divide (5) by (1

+ &)*"
(i/

we

find that

1)

TO=0

m!

In like manner, from


(8)
|

(2),

(W" Jv-m (*), Jv [z<J{\ + *)} - (1 + ky+ S K TO! m=0 -^f|

provided that k
If

<
-*-

1.

we make k
lim

+ 0, we

find,

by Abel's theorem,
(

[(1+*)KMW0 *-- 1+0


provided that the series on the right

~ )W(

)W>

+*)}] i

m=0

w^
-

,/y _

w (^),

when

is

an integer. If v
./_

is

is convergent. The convergence is obvious not an integer, then, for large values of to,

m (*)

sm (to - *) 7t
l '

{1

(1/w)}

= _( 4f^sin^ +1
7TTO*

Hence the condition


satisfied,

for
is

convergence
absolute.

is

R (v) > 0,

and

if

the condition

is

the convergence

Consequently,

when

R (v) > 0,

and

also

when
(9)

is

any

integer,

I
m=
if

- )m{h * )m
ml

Jv- m (z) = Q.

In like manner,
(10)

R(v) >

1,

and

also

Jv (zJ2) = 2-i X
m-0
kind in

when

v is

any

integer,

we have

r W!

Jv -m (z).
may be substituted provided that h < z (8)
j \ \

It should be observed that functions of the second kind


for functions of the first
(1), (2), (5)

and

and k <
\ J

so that

(11)

(*

+ A)-*'

K,

M* + h)} -

2 N-L*l

-^m Yv+m (^z),

(i2)

(*

+ &) rr {v(r+ A)} = i <**>%<-> r^^/*),


m=0
Wi!

da)
(14)

{*

vd + *>] = (i +
V(l

<-rl^>

Yv+m (zl

Y,

\z

+ k)} = (1 + jfc)-* 5
m=0

&* F_
TO!

lrt

(*).

142

THEORY OF BESSEL FUNCTIONS

[CHAP.

These may be proved by expressing the functions of the second kind as a linear combination of functions of the first kind by proceeding to the limit when v tends to an integral value, we see that they hold for functions of
;

integral order.

By combining the first kind, we

(11)

(14) with the corresponding


we may

results for functions of

see that

substitute the symbol

^ for the

symbol

Y throughout.
of

formulae were noted by Lommel, Studien, p. 87. Numerous generalisations be given in Chapter xi. It has been observed by Airey, Phil. Mag. (6) xxxvi. 242, that they are of some use in calculations connected with zeros of (1918), pp. 234

These

last

them

will

Bessel functions.

When we combine
that,

(5)

and

(13),

and then replace V(l


1)W(

+ k) by

X,

we

find

when |\-1|<1,
9. (\m) = X* 2
(

~ )m(V

(15)

m-0

", m.

^h ,

and, in particular,

when X

is

unrestricted,

d)
Bessel functions.
It

/.(x.)

H-fr'-^fr^ w
for
(8) is

These two results are frequently described* as multiplication theorems

may be
v is

(when

observed that the result of treating (14) in the same way as taken equal to an integer n)

that

(17) x
'

-(n-l)l

(2/ 2 )=tt 2 ra=0

i-Jfm Yn . m {z).

alternative proof of the multiplication formula has bee.n given by Bohmer, Berliner Sitzungsberichte, xiii. (1913), p. 35, with the aid of the methods of complex integration

An

see also Nielsen, Math.

Ann. lix. (1904),

p.

108,

and

(for

numerous extensions of the

formulae) Wagner, Bern MittheHungen, 1895, pp. 115119; 1896, pp. 5360. special case of formula (1), namely that in which v = l, was discovered by [Note.

Lommel seven
(1861), p. 356.

years before the publications of his treatise

see Archiv der Math,

xxxvn.

His method consisted in taking the integral

The
i

co (r cos
1,

+ ijr sin 6) ddt)


it

over the area of the circle * + $=

and evaluating
t)

by two

different methods.

result of integrating with respect to


j

is

fir 2ttJ_i|_

(r cos sin vs

M + r)r sin 0)
.

-\JQ-P>
-

d
a rsin.0
.

_W0-f)
\

" \ /_, C0S

^ OS 6) 8i " {V(1 ~^' r SiD &) ^~6


COfi{$rcoa6h{1
J -i

_i
*r

(-rtrnw* n
(2i+l)!

m=o
(

)m , id$

^
TO =o

- )m ($r sin 6?m Jm + i(r cos 0) (rcos0) m + 1 ml

'

* See, e.g. Schafheitlin,

Die Theorie der Besselscken Funktionen (Leipzig, 1908),

p. 88.

5*23, 5*3]

MISCELLANEOUS THEOREMS
<f>)

143

and the

result of changing to polar coordinates (p,


fn
/

is

5~

f
I

4bW J

v J

cos {rp cos (0 $)} pdpdUp

*W

/""
/

f
I

ir

cos (rp cos ^) pdpd<f>

=
If

^ ff
we

cos tfr)

^d^^-jjl- ?)* cos (r) <*=./, (r)/r.


i/

we compare
and r2

these equations
sin 2 *.]

obtain (1) in the case

=l

with

and h replaced by

r2 cos2

5*23.

The expansion of a Bessel function as a

series

of Bessel functions.

5*22 (7), Lommel has deduced an interesting series of Bessel functions which represents any given Bessel function.

From formula
ft

If

and v are unequal, and

fi is

not a negative integer,

we have

The repeated
arrange
it

series is absolutely convergent; consequently

we may

re

by replacing p by

m n, and

then we have

m=o U-ow! ym and hence, by Vandermonde's theorem,

n)l

(v

+n+

l)j

a)

j,

| n.-^^ w = r^+1) M r + m+
1

(v

o^
m!
,

ft)

(v

1)

* +TO

'

This formula was given by Lommel, Studien

iiber die Bessel'schen

Functionen (Leipzig,

2223, in the special case M =0; by putting *=0, it is found that


1868), pp.
(2)

differentiating with respect to v

and then

*, Fo (,)- Ji
,

(.)

log <*)

-tel>

C^> (^~^M+m W
*{V' (l +i)+V''(-/*)-'<Kn-M)K

and,

when /i=0, we have Lommel's formula

This should be compared with Neumann's expansion given in 3'57.

53.

An

addition formula for Bessel functions.


is

An
(1)

extension of the formula of 2*4 to Bessel functions of any order


./r

(*

+ <)- 2 J^iftJ^z),
This formula
is

where z
|

<

{,

being unrestricted.

due

to Schlafli*

and

the similar but more general formula


(2)
is

%\{z + t)=
to Soninef.
*

2
= 06

^(0/.(i)
( JWd.
xvi. (1880), pp.

due

MatA. Ann. hi. (1871), pp. 135137.

78.

144

THEORY OF BESSEL FUNCTIONS


is

[CHAP.

It will first be shewn that the series on the right of (1) convergent series of analytic functions of both z and t when

a uniformly

|*|<r,

R^\t\^A,
numbers
v

where

r,

R,

are unequal positive

in ascending order of magnitude.

When

m is large and positive, J ^m (t)Jm {z) is comparable m-V sin vtt {\Ry {r/Ry* V(
)
. .

with

and the convergence of the


series for (1

series is
is

r/R)".

When m

large

comparable with that of the binomial and negative (= w), the general

term

is

comparable with

(-)"(*A)'(jAr)"

T(v + n+l).n\
and the uniformity of the convergence
the test of Weierstrass.
follows for both sets of values of

m by

Term-by- term differentiation

is

consequently permissible*, so that


'
,
,

- (t) Jm {z) == J "m {t)Jm {z) Qt " l~z)ji/v m -J m?

J*-m {t) J m {z)]


'

=H 2

*
{

2 J -m(t){Jm- (z)-Jm+1 (z)}, -^ *


m= -oo
v
1

" m= ao " 1

J,--! (0

- Jv- m+1 (t)} Jm (Z)

and
that

it is

seen, on rearrangement, that all the terms

on the right cancel, so

d-a)J./~< >'-<'>- a
,
30

Hence, when z
|

<

\t \

the series

m=

2 Jv -m (t) Jm (z)
-oo

is

an analytic function

of z

and

which

is expressible
t

as a function of z

+t

only, since its derivates

with respect to z and

are identically equal.

If this function be called

F {z +
If

1),

then

we put
if

= 0,

m= oo we see that F(t) = Jv (t), and the truth of (1) becomes

F(z + t) =

Jy-nit)

Jm (z).

evident.

Again,

the signs of v and


J-. v (z

m in
l =

(1) be changed,

we have
(z),

+ t)= X
-oo

(-)m J- v+m (t)Jm


(1),

and when
(3)

this result

is

combined with
Y,(z + t)=
* Cf.

we

see that

X
1=-00

Y _m (t)Jm (z).
v

Modern Analysis,

5*3.

5*4]

MISCELLANEOUS THEOREMS
this is

145

When

combined with

(1),

equation (2) becomes evident.


that,

The reader
(4)

will readily prove

by the same method

when

<

|,

Jv (t - z) I Jv+m (t) Jm (*),


^ (*-*)= I Wr+miftJmiz),
= oo
T

(5)

(6)

(<-*)= X

Yv+m (t)Jm (z).

Of these

results, (3)

(Leipzig, 1868),

when

v is

was given by Lommel, Studien iiber die BesseVschen Functionen an integer while (4), (5) and (6) were given* explicitly by Graf,
;

Math. Ann. xliii. (1893), pp. 141 given in Chapter xi.

142.

Various generalisations of these formulae

will

be

5*4.

Products of Bessel functions.


J. (z) Jv (z) has been given by various sometimes stated to be due to Schcuholzerf, who in 1877, but it had, in fact, been previously published (in 1870) More recently the product has been examined by Orr, while

The ascending series for the product


writers
;

the expansion
it

is

published

by

SchlafliJ.
||

Nicholson has given expansions

(cf.

5'42) for products of the forms

J(z)Yn (z) and

Ym (z)Yn (z).
satisfied

In the present section we shall construct the differential equation

by the product of two Bessel functions, and solve it ( 541) obtain the expansion anew by direct multiplication of series.
in series.

We

shall

then

Given two

differential equations in their

normal forms

if

y denotes the product vw, we have


y['

= if'w + 2v'w' + vw" = -(/ + J)y + 2v'w

/
,

where primes indicate

differentiations

with respect to

z.

* See also Epstein, Die vier Reehnungsoperationen mit BeueVsehen Functionen (Bern, 1894), [Jahrbuch tiber die ForUehritte der Math. 18931891, pp. 845846]. t Ueber die Auswerthung bettimmttr Integrate mit Hillfe von Ver&ndetungen det Integrationsweget (Bern, 1877), p. 13. The authorities who attribute the expansion to SchSnholzer include Graf and

Oubler, Einleitung in die Theorie der BesseVschen Funktionen, u. (Bern, 1900), pp. 8687, and Nielsen, Ann. Sci. de Vltcole norm. sup. (3) xvin. (1901), p. 50 ; Handbueh der Theorie der Cylinder/unktionen (Leipzig, 1904), p. 80. According to Nielsen, Nouv. Ann. de Math. (4) u. (1902),

obtained some series for products in the Iserlohn Programm, 1862. % Math. Ann. hi. (1871), pp. 141142. A trivial defect in Schlafli's proof is that he uses a contour integral which (as he points out) converges only when B(/t+r + l)>0.
p. 396, Meissel

Proe.
||

Camb. Phil.

Soc. x. (1900), pp.

93100.

Quarterly Journal, xliii. (1912), pp. 78100.

'

146

THEORY OF BESSEL FUNCTIONS


It follows that

[CHAP. V

dz

d ~ \y" + (J + J) y] = 2v'V + 2v'w"

and hence
Hence, in
(1)

= - 2Ivw' - 2Jv'w y'" + 2 (/ + /) y + (/' + J') y = (/ - J) (v'w - vw'). the special case when IJ,y satisfies the equation y + 4/y' + 2/>=0;
/'/

but, if

IJ,

it is

easy to shew by differentiation that

(2)
This

*
is

qr+w+wv+J-),
}

_ _ (/ _
;

^
(1),

Appell, Comptes

the form of the differential equation used by Orr Rendm, xci. (1880), pp. 211 214.

in connexion with

see

To apply these
to a

results to Bessel's equation, the equation has to


;

be reduced
it

normal form
(z) as

both Orr and Nicholson effect the reduction by taking


but, for purposes of solution in series,

z
is

9!> v

simpler to

new dependent variable, take a new independent


-j> ~e Z
, Z
>

variable

by writing

dz~dO~*'

-^

so that

$%$ + (* - *) / (*) = 0.
satisfied

Hence the equation


d_

by J^ (z) Jv (z), when


8

ft*

^ v*,
2

is

dd
that
is

0+2(2^-^-i, )^| + 4^y| + (^-^)

y = 0,

to say

(3)

[&-2(p* + v*)** + ( fi*-v>Y']y + 4e(*r + l)(% + 2)y =


satisfied

>

and the equation


(4)

by

(z)

J (z)

is

*(*-4)y + 4"(ar + 1)0 = 0.


{-)m cm

Solutions in series of (3) are


*

where a
m

/*

and
4 (a

+ 2m !)( + 2m) ctn _i (a + ft + + 2m)(a + p -v + 2m)(a- p+ v + 2m)(a- fi-p + 2m) If we take a = /* + v and
1>

^-2^+"r(/*+i)r(i/ +

i)'

we

obtain the series


o

(_)w

w?o m!r(/* + v + m + l)r( A* + w+l)r(i/+w+l) and the other series which are solutions of (3) are obtained by changing the
signs of either
/i

Qgy,+p+m, Y(
/*

fl

+ y+ 2m + 1)

or v or both

and

v.

"

5*41]

MISCELLANEOUS THEOREMS
considering the powers of z which occur in the product
if 2/*, /M (z)

147

By
is
fi*

Jv (z)

it if

easy to infer that,

2v and 2
(~)m

(/* + v)

are not negative integers and

^ ,

then
r t .\ T u\- v J|lW,/rW

m w W
and,

(^Y ++im r (^ + + 2m + 1)
series,

"".r iir(fi + ir+m+i)r(/+iii + i)r(+t+i)"

In like manner, by solving (4) in


negative integer, then

we

find that,

when

2i> is

not a

'

W -wr /!r(2
2
=0 (w!)
2

+ m-Hl){r(j/ + m+l)}
(-)m (zYm (2m)\

'

when

>

is

not a negative integer, then

(7)

J(s)J_(s)=

r (i/ + m + 1) T(- v + m +

1)

reasoning which resembles that given in 4'42, it may be shewn that (6) holds when v is half of an odd negative integer, provided that the quotient T (2v + 2m + l)/r {2v + i + 1) is replaced by the product (2v + + l)m

By

5*41.

Products of series representing Bessel functions.

It is easy to obtain the results of 5*4

by

direct multiplication of series.

This method has the advantage that special investigations, for the cases in which fi* = v* and those in which y. + v is a negative integer, are superfluous.

The

coefficient of (-)"*

(yY +v+sm

in the product of the two absolutely

convergent series

(_y ( fr)M m =o! r((i + m+ 1)


m
is
tt

n=nn\T(v + n + l)

(-)w (**)"+w

= n!

Y(v

+n+ l).(m-n)! T(/i-m-n + l)


w+ Ii\ r<\ x,xn (v + m + l,)*=o tR m 1) I
l)

_l r (/* + w!ir/

Cw

(-'-)w_n .(-/A-m)

fl

m!

T(/*

+ m+

p (i> + w +
l)m

l)

(n

+v+m +

m!r(/t + w+l)r(i/+m + l)'

when Vandermonde's theorem


Hence,
for all values of
/*

is

used to
v,

sum the

finite series.

and

n\ (1)
and

g t /.-\ t {z)-^ (.\ j^z)J

(-)m

m]

(i^ +,,+aw (^ + + m+l)TO r{fJt + m + i )r ( v + m + iy


i>

this formula comprises the formulae (5), (6)


'

and
.

(7) of 5*4.
----H--'
'

*~
S

148

THEORY OF BESSEL FUNCTIONS

[CHAP.

earlier writers; it

This obvious mode of procedure does not seem to have been noticed by any of the was given by Nielsen, Math. Ann. lii. (1899), p. 228.
series for

cos z and J {z)e,mz were obtained by Bessel, Berliner Abh. 1824, and the corresponding results for Jv (z) con z and Jv {z)smz were deduced from Poisson's integral by Lommel, Studien uber die BesseVschen Functionen (Leipzig, 1868), pp. 16 18. Some deductions concerning the functions ber and bei have been made by Whitehead, Quarterly Journal, xlii. (1911), p. 342.

The

{z)

[1826], pp.

38

39,

More generally, if we multiply the series for J^(az) and an expansion in which the coefficient of ( w orb" (fcY +v+am
m
,o!
Qitn2n A2n

J (bz),
v

we

obtain

is

T (y + n + 1) .(m - n)l T (ft + m-n + l) 2 m a _ a tB\(-m,-iA-m\ v + 1 6 /a ) ml r<> + m + l)r(i/ + l)


;

and so
(2)

J, (az) Jv (bz) -

V2

'

r ^ ^J (~)w QaiY" . fi (- m, - /i - m x | t! r(/i+TO+l) m=0

i/

+1

fr/q')

and

this result can

be simplified whenever the hypergeometric

series is

expressible in a compact form.

One
another

case of reduction
is

is

the case b

the case

b = ia,

provided

= a, which that ft* = v*.

has already been discussed

In this case we use the formula*

and then we see that


/o\

r/
v

w/
;

\
'

v
wt=0

r(|m
-

(-)m (^azy" +m cos fymr + i)r(i/ + |w + i)r(i/+in (-)m (\azf" +im

i)

- %

=0 w! T (v +

m + 1) r

(i/

+ 2w +

1)'

(5)

/fa*)/

fart- v
in (3)
a

(-f (ifliycM(i + ht)T


we
find that

If
(6)

we take a = el"*
ber (z)

+ bei

(z)

- 2

ptt

it***"*
3*8.

an expansion of which the leading terms were given in


* Cf.

Kummer, Journal fUr Math.

xv. (1836), p. 78,

formula

(53).

5*42]

MISCELLANEOUS THEOREMS
(3), (4), (5)

149
(5)

The formulae

xv. (1906), pp. 490 497 and Monatshefte fiir Math, und Phys. xix. (1908), pp. 164170, from a consideration of the differential equation satisfied by Jv (az)Jv (bz).

were discovered by Nielsen, Attidella R. Accad. dei Lined,

Some
types

series

Jy3 (z) and J% (z) t/_ (z),


giving
-
fi

have been given, Quarterly Journal, xli. (1910), p. 55, for products of the but they are too cumbrous to be of any importance.

By
(7)

the special values | in

(2), it is

easy to prove that


6).

J^ (z sin 6) = iig (2 sin 0)-*^ r *2y + n) <V (cos


this formula in

The special case of by Hobson*.


5*42.

which 2v

is

an integer has been given

Products involving Bessel functions of the second kind.


series for the products

The

J {z)Yn {z), Jm (z)Yn (z),


IL

and

Ym {z)Yn {z)
is

have been the subject of detailed study by Nicholson f; the following outline of his analysis with some modifications.

an

We

have
irJ, (z)

Yn (z) = 1
to

J* (,) Jw ()}

- (-) i {/, (z) J. (*)},

where

i/

is

to be

made equal

n after the differentiations have been performed.

Now
l~
{

JM (,) J, (*)} - log {\z). J (z) J (z)


v
I"

(~)r (^>t+,,+8r

T(p + v + 2r + l)

and

r=0 Lr:

r (fi - v + 2r + 1) rO-i/ + r+l)r(/* + r+l)r(-i/ + r + l) fi/r (p - y + 2r -H 1) - ^ (/* - v + r + 1) - yfr (+ r + 1)}


(-)" (^y4-"*"*
1/

We

n-l

oo
.

divide the last series into two parts,

2 and X

In the former part we

have

* Proc. London Math. Soc. xxv. (1894), p. 06 8 je also Cailler, ilfm. d ia Soc. de Phys. de Geneve, xxxiv. (19021905), p. 316. t Quarterly Journal, xliii. (19i2), pp. 78 100. The expansion of Jq(z) Yq (z) had been given previously by Nielsen, Handbuch der Theorie der Cylinderfunktionen (Leipzig, 1904), p. 21.
;

150

THEORY OF BESSEL FUNCTIONS


is

[CHAP.

while in the latter part there


r is replaced in this part

no undetermined form to be evaluated.


it is

When

by n + r,

seen that

(1)

wJ,(b)Yu <*)%
9

r!r</*
n +* r

+ r+l)
l) r

(A* + n + r + | (-Y {hzY+ r!(n+r)ir(^ + r+l) r -o

{2 log (*)

2^r (p

+ n + 2r + 1)
l)

-^r(^ + n + r+l)-^r(/* + r +

- -^ (?i + r+l)-yjr(r + 1)}.


The expression on the
vn

right

is

a continuous function of

fi

at

/tt

=m

where
fi

= 0,

1, 2, ...,

and so the
(1).

series for

irjm

(z)

Yn {z)

is

obtained by replacing

by

m on
The

the right in
series for

Ym (z) Yn (z)
|_

can be calculated by constructing series


~

for

[ d-J h (z)Jv (z)

dfldv

jn-m, v-n

in a similar manner.

The details of the analysis, which is extremely laborious, have been given by Nicholson, and will not be repeated here.

5*43.

The integral for J^ (z)

Jv (z).
integral ( 2*6) for

A generalisation of
applying the formula*

Neumann's

Jn

(z) is obtainable

by

eo**-+Jo
to the result of 5
-

cos fr

41 ;

+ m + 1)r(v + m + 1) provided the integral has this value when m 0, 1, 2,


r(fJ
, . .
.

v)

0d0 = 2iL++M

that

R (jM + v) > - 1.
then evident that

It is

J(*)J(*) =
so that

when

J?

(/u.

- 2 m -o.'o + *>) > 1,


7r

^s(fi-v)dd0, ,w .. + m ..n m!r(/*+v + l)

/M (s) J (z) = -

V,^, (2* cos 0) cos (fi-v)0d0\

TJfl

the change of the order of summation and integration presents no serious


difficulty.
* This formula is due to
p. 263.

Cauchy

for

a proof by contour integration, see Modern Analysis,

5-43-5-51]

MISCELLANEOUS THEOREMS

151

If n be a positive integer and


(2)

R n) > 1, then JM (z) Jn (z) = *i~L P'j^ (2z cos 6) cos + n) 0d0, T
(jj,

(fi

Jo
/*

and

this formula is also true if

and n are both integers, but are otherwise


when nv

unrestricted.

Formula

(1)

was given by

Schlafli,

Math. Ann. in. (1871),

p. 142,

are both

integers; the general formula is due to Gegenbauer, Wiener Sitzungsberichte, cxi. (2a),
(1902), p. 567.

5*5.

The expansion of (\zY+ v as a


that

series

of products.

A natural generalisation of the formulae of Neumann (2*7) and Gegenbauer


(

5 2)
(1)

is

(**r

r{/M

+ v+1 y
m-0

m.
v are

The formula

is

true if
if

fi

and

not negative integers, but the following

proof applies only

R(fi

+ v + l)> 1.

From

5*2

we have

(.>-.- 1 0>+>+)ro. + > + m) J


If

^^
;

we multiply by

cos

(fi

v)0 and
to-0

integrate, it is clear from 5*43 that

2^+

:p\*".co.-)M- 2 0>+ + )ro.+>+)


Jo

ml

of

and the result p and v the


The formula
beriehte,

follows
result

by evaluating the integral on the left for other values may be established by analytic continuation.

is at

once deducible from formulae given by Gegenbauer, Wiener Sitzungs-

lxxv.

(2), (1877), p. 220.

5*51.

Lommel's

series

of squares of Bessel functions.

An

expansion derived by

Lommel* from

the formula

i8

j>

+h)<J,yW, H (.).i"' Z dz n=0


=
J

so that
(*) <fc 2 J'xJK-i
*
p.

(*

+ 2) JV2 (*)1
ii.

The
;

results of this section will be

532

MUnchener Abh.

xv. (1886), pp.

found in Math. Ann. 548549.

(1870), pp.

632633

xiv. (1878),

152

THEORY OF BESSEL FUNCTIONS


5*11 (11),

[CHAP. V

Hence, by
(1)

we have

**

{./

_ x (z)

- J^ (z) J (z)} = 2
n-0

(i;

+ 2n) JV (*),
2
;

beginning of the
fluous.

by adding on terms at the on taking zero as the lower limit when R (v) > is superseries, it may be seen that the restriction R,(v) >
If we take in turn we have ( 3*4)
(2)

= \, v = f

and add and subtract the

results so obtained,

-= 2
n=0

(n

+ *)JV(*).

(3)

*.
by taking i>=
1,

(-r<n + *)/+(*).

while,
(4)

we

see that

\z* {J *

(jr)

+ J (5)} = I
is

(in

-0

1) An+x

(*)

Another formula of the same type

derived by differentiating the series

2
-o
for it is evident that

J\ + (s)

a
^-

2 n=0

2 eJ\+n (*) = 2 2
00

e /"+ (*)/'+(*)

=0
n
t/"

= 2
t=0

+n (s) (Jv+n-i (^)

Jv+n+i (*)}

= ivJ?(z)\z,
and so. when iJ(v) >0, we obtain a modification of Hansen's formula (2 namely
00
-

5),

(5)

2
=o

en

J\+n (z) = 1v\ Jf(i%.


Jo
*

f*

fit

An

important consequence of this formula, namely the value of an upper


for
j

bound

Jv (x)

\,

will

be given in
found that

13 42.

By

taking

v \,
v 2
m=o

it is

, \ t e J n+i (*)

2 dt = - P sin** 1 "

tJo
f
|_

2 sin2 *]*
7T
t

2 f*
7T.I

_ ,cft
*

Jo

and

so

(6)

2
n=0

/'

n+J (*)

= -i(2*), T

where, as usual, the symbol Si denotes the "sine integral." This result is given by Lommel in the third of the memoirs to which reference has been made.

5-6]
5*6.

MISCELLANEOUS THEOREMS
Continued fraction formulae.

153

Expressions for quotients of Bessel functions as continued fractions are deducible immediately from the recurrence formula given by 32(1); thus,
if

the formula be written

/,_,(*)

%zJ v+x (z)j v'

Jv (z)
it is

at once apparent that

*/_,(*)

\z-j{(v+m-l)(v\-m)}

\z

J + m+1 (z)/{v + m)
v

This formula
is

J v+m\ z )

easily transformed into

'

J ^ (z)
v

2v/z

-2(v + l)/z-...-2(v + m)lz - Jv+m (z)


for general values of v\

'

These results are true


Bessel*
for integral values

(1)

was discovered by
is

of

v.

An

equivalent result, due to Schlomilchf,

that, if Q {z)

= Jv+l {z)l{\z Jv (z)},

then

Other formulae, given by LommelJ, are

Jv+1 (z)
(4)

z
2(i/

z*

Jv (z)
/<

+ l)-2(i/ + 2)-2(i/ + 3)-...-2(/ + m)*'


i
i

J*

z2

zJv+m+1 (z)
J, +m (z)
x
'

W
The

,k\

^+2 (*) _

zJv+m + {z) i! + 2fel) 2(i/ + l)-2(* + 2)-...-2(> + m)- J +m (s)


all

'

Bessel functions in

these formulae

may

obviously be replaced by any

cylinder functions.
It

was assumed by Bessel

that,

when

m -*-

<x>

the last quotient

may be

neglected, so that
,x (b)

AA^L-^ll
J_i(*)~"
1

i^l\ V (v

+ l)}

$*/{(+ !)(+
1

2)}

* Berliner Abh. (1824), [1826], p; 31. Formula (2) seems not to have been given by the earlier writers; see Encyclopedic des Sci. Math. n. 28, 58, p. 217. A slightly different form is used by

Graf, Ann. di Mat. (2) xxin. (1895), p. 47. integral values of t Zeitschrift fiir Math, und Phys. n. (1857), p. 142; Schldmilch considered
v only.

J Studien

fiber die

Math, und Phys. xxx. pp. 292297.


W. B.
F.

BesseVschen Functionen (Leipzig, 18G8), p. 5 see also Spitzer, Archiv der (1858), p. 332, and Gunther, Archiv der Math, und Phy*. lvi. (1874),
;

154

THEORY OF BESSEL FUNCTIONS


It is not obvious that this assumption is justifiable,

[CHAP.

though it happens to and a rigorous proof of the expansion of a quotient of Bessel functions into an infinite continued fraction will be given in 9'65 with the help of the theory of " Lommel's polynomials."
be
so,

[Note.
the fraction

The reason why the assumption

is

not obviously correct

is that,

even though

pm /qm tends
;

to a limit as m--oo

it is

not necessarily the case that

aw ^">

^m+1

tends to that limit

this

may be

seen by taking

pm =m+amm,

q m =m,

a,=-l.]

The reader will find an elaborate discussion on the representation of Jv (z)/Jv -i (z) as a continued fraction in a memoir* by Perron, Munchener Sitzungsberichte, xxxvn. (1907), 504; solutions of Riccati's equation, depending on such a representation, have pp. 483

been considered by Wilton, Quarterly Journal, xlvi. (1915), pp. 320323. The connexion between continued fractions of the types considered in this section and the relations connecting contiguous hypergeometric functions has been noticed by Heine, Journal fiir Math. lvii. (1860), pp. 231247 and Christoffel, Journal fur Math, lviii. (1861),
pp. 9092.

5*7.

Hansen's expression for J,(z) as a limit of a hypergeometric function.

It

was stated by Hansen f that


J.

(i)

W -iin r a^
the

..i'1
J

>

M;+i;-^).
\ and

We shall prove
fi

this result for general (complex) values of v and z when tend to infinity through complex values.

If

1/8, fi

= 1/t),

(m -f
K

l)th term of the expansion on the right

is

This
is

/,

=-,

[(1

+ rh) (1 + rv )\
,

a continuous function of 8 and rj and, if 8 r) are arbitrary positive (less than 2 z j"1 ), the series of which it is the (wi + l)th term converges uniformly with respect to 8 and 77 whenever both 8 ^ 8 and 17 ^ i For the term in question is numerically less than the modulus of the (m + 1 )th
;

numbers

term of the (absolutely convergent) expansion of

r^,+\)

& W*., ^
is

>

+ 1 K^o),
>

and the uniformity of the convergence


Since the convergence
* This

follows from the test of Weierstrass.

uniform, the

sum

of the terms

is

a continuous

(1908), pp.

memoir 8588.

is

the subject of a paper by Nielsen, Milnchener Sitzungsberichte, xxxvni.

t Leipziger Abh. n. (1855), p. 252 ; see also a Halberstadt dissertation by F. Neumann, 1909. [Jakrbueh ilber die Fortschritte der Math. 1909, p. 575.]

5-7, 5*71]

MISCELLANEOUS THEOREMS
($, rj)

155
is

function of both the variables

at (0, 0), and so the limit of the series


;

the

sum
/

of the limits of the individual terms

that

is

to say

^o\r(v + l)

\B

r}'

m . m!r(i/ +

m+l)

and

this is the result stated.

5*71.

Bessel functions as limits of Letfendre functions.

known that solutions of Laplace's equation, which are analytic near the origin and which are appropriate for the discussion of physical problems connected with a sphere, may be conveniently expressed as linear
It is well

combinations of functions of the type


rn

Pn (cos0),

rn

Pnm (cos 0) cos m<f>; sin


.

these are
coordinates

normal solutions of Laplace's equation when referred to polar


(r, 0,
<f>).

Now

consider the nature of the structure of spheres, cones and planes

associated with polar coordinates in a region of space at a great distance from

the origin near the axis of harmonics.

The spheres approximate to planes and


;

the cones approximate to cylinders, and the structure resembles the structure
associated with cylindrical-polar coordinates

and normal solutions of Laplace's


( 4*8)

equation referred to such coordinates are of the form


e k*Jm

(kp)^m<f>.

It is therefore to be expected that,

when r and n

are large* while

is

small

in such a

way

that r sin

(i.e.

p) remains bounded, the Legendre function

should approximate to a Bessel function; in other words,

we must expect

Bessel functions to be expressible as limits of Legendre functions.

The

actual formulae by which Bessel functions are so expressed are, in

effect, special cases of

Hansen's

limit.

The most important formula


(1)

of this type

is

limPn (cos^)=/ (4
result,

This

which seems to have been known to Neumannt in 1862, has been

investi-

gated by Mehler, Journal fiir Math, lxviii. (1868), p. 140; Math. Ann. v. (1872), pp. 136, 141 144; Heine, Journal fur Math. lxix. (1869), p. 130; RayJeigh, Proc. London Math.

Soc. ix. (1878), pp.


i.

(1899), pp.

pp. 236 and Rayleigh.

6164 Proc. Royal Soc. xcii. A, (1916), pp. 433437 [Scientific Papers, 338341 vi. (1920), pp. 393397]; and Giuliani, Giorn. di Mat. xxn. (1884), 239. The result has been extended to generalised Legendre functions by Heine
; ;

It has usually been

assumed that n tends


it is

to infinity through integral

values in proving (1); but

easier to prove

it

when n tends
would be

to infinity as

a continuous real variable.


* If
?i

were not large, the approximate formula for

Pnm (cos 0)

(sin

m 0)/m

!.

t Cf. Journal fUr Math. lxh. (1863), pp.

3649.

156

THEORY OF BESSEL FUNCTIONS

[CHAP.

We

take Murphy's formula

Pn (cos z/ri) =
modification that

sjFx

n,

+1

1
is

sin2 \z\n)\

and the reasoning of the preceding section

applicable with the slight

we use the

inequality

|rin(f/)|*f|(*/n)|,

when
|

\z

$ 2 n
|

,
|

and then we can compare the two


1
;

series
;

& (where B
is is

n,

n+

sin2 \z\n\

,F, (1/S

1/S6

+1
|

^$

2
1

2
!

),

an arbitrary positive number less than z _1 atid the comparison made when \n\>lj8 The details of the proof may now be left to the reader.
|

When
the proof.

is restricted to it is

be a positive integer, the series

for

Pn (coaz/n)

terminates, and

convenient to appeal to Tannery's theorem * to complete This fact was first noticed by Giuliani the earlier writers took for
;

granted the permissibility of the passage to the

limit.

In the case of generalised Legendre functions (of unrestricted order m),


the definition depends on whether the argument of the functions
is

between

+1

and

1 or not

for real values of


v

x (between
n,
;

and

ir)

we have
sin" **/).

Pw- (cos -J =
so that
(2)

'

'

^_

& (-

n+ 1 m + 1;
(as),

lim n m

Pn~m (cos -) = Jm

but otherwise, we have


z\ tanh (iz/n) ( ,, Pn~m (cosh = F (~ n,n + l; m + 1; ~ sinh \z\n), r m ^i) -J
r>
i

_,

so that
(3)

lim n m

Pn-m (cosh -) = Im (z).


for functions of the

The corresponding formula

second kind

may be deduced
;

from the equation which expressesf


,,,

Qnm
~

in terms of
,

Pnm and P~ m

it is

(4)

w _.ooLsm

hm

,.

m sinri7r rn -^
(

m+

z\ r Qnm ( cosh n)7r nj \

Km

(z).

This formula has been given (with a different notation) by Heine +; it is most easily proved by substituting the integral of Laplace's type for the Legendre
function, proceeding to the limit

and using formula

(5) of

62 2.

* Cf. Bromwioh, Theory of Infinite Series, 49. t Cf. Barnes,' Quarterly Journal, xxxix. (1908), p. 109

the equation
1

is

2r(-m-n)
in Barnes' notation,

BinVlir

nnir

P~" Q n= r(l-m +

P-"*
?j)

r(l + i + n)

which is adopted in this work. + Journal fur Math, i/xix. (1868), p. 131.

5*72]

MISCELLANEOUS THEOREMS
slightly different from those just discussed, is

157

Another formula,
(5)
this
is

K
.

"/(il*)" /,(
it

);

due to Laurent*, and

may be proved by
.

using the second of Murphy's

formulae, namely

Pn (cos 6) = cos" %6 F
2

(-

n,

-n

- tan

\6).

[Note. The existence of the formulae of this section must be emphasized because it used to be generally believed that there was no connexion between Legendre functions and
Bessel functions.

Functions, Lame's Functions


[i.e.

in his Elementary Treatise on Laplace's BesseVs Functions (London, 1875), p. vi, that "these Bessel functions] are not connected with the main subject of this book."]
it

Thus

was stated by Todhunter

and

5*72.

Integrals associated with Mehler's formula.

section was given

completely different method of establishing the formulae of the last by Mehler and also, later, by Rayleigh this method depends
;

on a use of Laplace's integral, thus


1 Pn (cos $) f*

(cos

i sin

I TT.'O

V cos $) n d<j>
fifo

en Iok (cos* + < sin cos $)


7T T.'o
' (

Since

n log
uniformly as n
-- oo

{cos (z/n)

+ i sin

(z/n) cos

<} -*-

iz cos

<f>

when
lim

$ <f> ^ ir, we have at once


d<f>

Pn (cos z/n) - - f *

=J

(z).

Heine f and de Ball J have made similar passages to the limit with integrals of Laplace's type for Legendre functions. In this way Heine has defined
Bessel functions of the second and third kinds
results in 6*22
;

reference will be

made
(z).

to his

when we

deal with integral representations of F

Mehler has also given a proof of his formula by using the Mehler- Dirichlet integral
2 Pn (cos6) = If n<f>=yjr, it

; o

*<&4*L
V{2 (cos <p -COS0)}

may

be shewn that

- I' zln\ i n (cos (cos*/*)*-J


o

-^

C0S

but the passage to the limit presents some proper integral.

little difficulty

because the integral

is

an im-

Various formulae have been given recently which exhibit the way in which
* Journal de Math. (3) i. (1875), pp. 384385; the formula actually given by Laurent is trioneouB on account of an arithmetical error. t Journal fur Math. lxix. (1868), p. 131. See also Sharpe, Quarterly Journal, xxiv (1890)

pp.

383386.
X Attr. Nach. cxxvm. (1891), col. 1

1.

158

THEORY OF BESSEL FUNCTIONS


its

[CHAP. V
infinity.

the Legendre function approaches its limit as

degree tends to

Thus, a formal expansion due to Macdonald*


(1)

is

Pn -(cos0)
= (n + )"' (cos 0)- [Jm (x)
+ sin
2

\0 {^x

J^ (x) - Jw (x) + ^x' Jm+l


1

(x)}

.],

where x = (In

1) sin \0.

Other formulae, which exhibit an upper limit for the error due to replacing a Legendre function of large degree by a Bessel function, aref
(2)

Pn (cos

rj)

iir- 1

Qn (cos rj)

= ^(sec rj)

e <* + i>,- t*n '>> [Jo {(

+ h) ^n v}

iY

{(

+ ) tan rj\]
4^i-\/(sec v)

+ 7r{R(n)

+ ^'

(4)

Qn (cosh f) = e -(+i)(f-tanh ^( S ech f)

K
+

{(n

+ \) tanh f
V(sechg).e-(+*>*

ffl3

R{n)+$
; ,

and (4), f > the numbers lt 2 < t) < where, in (2), 3 are n may be complex provided that less than unity in absolute magnitude, and its real part is positive. But the proof of these results is too lengthy to be
%tt, and, in (3)

given here.
5*73.

The formulae of Olbricht.


a Bessel function
is

by Hansen's formula as a limit of a hypergeometric function has led Olbricht j to investigate methods by which Bessel's equation is expressible as a confluent form of equations associated with Biemann's P-functions.

The

fact that

expressible

If

we take the equation


dz*

dz

V
is

z*

'

of which a fundamental system of solutions

the pair of functions

z~*Jv (z),
fa,
b,

z-Y
c,

v (z),

and compare the equation with the equation defined by the scheme
1

Pio,
l',
* Proc.

,
ff,

y,

z\,
-I

7.

London Math. Soc. (2) xm. (1914), pp. 220 221 eome associated results had been obtained previously by the same writer, Proc. London Math. Soc. xxxi. (1899), p. 269. t Watson, Trans. Camb. Phil. Soc. xxn. (1918), pp. 277308 Messenger, xlvii. (1918),
; ;

pp. 151160.

X Nova Acta Caes.-Leop.-Acad. (Halle), 1888, pp. 148.

5*73]

MISCELLANEOUS THEOREMS

159

namely
d* y

Q- a -a
\
(

'

\-fi-fi'

l-y-yl
z-c
(b
)

dy
dz
yy' (

dz*

z-a
z-a

z-b
+ fifi'

aa'(a-b)(a-c)

-c)(b-a)

-a)(c- &)

z-b
(z

z-c
y

) )

-a)(z -

b) {z

c)

= 0,

we

see that the latter reduces to the former if

a
while
b, c, fi, fi', 7,

= 0,

= v fi, + 1)
0,

= v fi,
way
that
2

y tend

to infinity in such a
2/*

fi

remain

finite (their

sum being

while

yy - $b fifi'
-2ifi,
y,
\

+ fi' and 7 + and b + c = 0.

7'

We

thus obtain the scheme


(

2ifi,
fi,

limPJ
~*

v-p,
2

z\,
J

l-v-fi,
where
7,

-fi,

7,

- M + i V{0* + ) + tf
is
(

2
}-

Another similar scheme

0,

ifi,
fi,

oo,
J

limP<
"*"

v-fi,

y,

z\
)

i-v-fi,

-fi,

y,

with the same values of 7 and y

as before.
is

scheme

for

(z)

derived directly from Hansen's formula


(

0,

qo,

-4a&
0,

lim

pj

\v

a-\v,
fi-\v,

zA.
J

ilZl)
those which have
Note.
It

[-% v>

v+l-a-fi.

Olbricht has given other schemes but they are of no great importance and

now been

constructed will be sufficient examples.


fiir

has been observed by Haentzschel, Zeitschrift

Math, und Pkyt. xxxi.

(1886), p. 31, that the equation

whose solution

( 4*3) is xS'Wv (hit),

may be

derived by confluence from Lamp's equation

when the
zero.

invariants

g2 and g3

of the Weierstrassian elliptic function are

made

to tend to

CHAPTER VI
INTEGRAL REPRESENTATIONS OF BESSEL FUNCTIONS
6*1.

Generalisations of Poisson's integral.


shall study various contour integrals associated with

In this chapter we

Poisson's integral ( 2*3, 3"3)

of the contour of integration, large

and Bessel's integral ( 2-2). By suitable choices numbers of elegant formulae can be obtained

also be applied in

which express Bessel functions as definite integrals. The contour integrals will Chapters vu and VIII to obtain approximate formulae and

asymptotic expansions for

Jv (z) when

z or v is large.

It happens that the applications of Poisson's integral are of a more elementary character than the applications of Bessel's integral, and accordingly

we

shall

now study
which we

integrals of Poisson's type, deferring the study of integrals

of Bessel's type to 62.


integral
shall

The investigation of generalisations of now give is due in substance to Hankel*.

Poisson's

The simplest

of the formulae of

33

is 3'3 (4),

since this formula contains

a single exponential under the integral sign, while the other formulae contain
circular functions,
shall

which are expressible in terms of two exponentials. We therefore examine the circumstances in which contour integrals of the type
r

e izt
J a

Tdt

are solutions of Bessel's equation;

it is

not of
of
z.

z,

and that the end-points, a and

supposed that T is a function of t but b, are complex numbers independent

The

result of operating

on the integral with Bessel's

differential operator

V,, defined
b

in 3*1, is

as follows:

V,

izt

Tdtl

= z v+i
izt

lzt

T{\h

dt
b

(2i/

izt +1 1) iz j e

Ttdt

= iz"+
*

T (t* -

1)1

+ iz+*

e izt \{2v

1) Tt

- ^ {T (i 2

1)}1 dt,

Math. Ann. i. (1869), pp. 473 485. The discussion of the corresponding integrals for Iv(z) is due to Schlafli, Ann. di Mat. (2) i. (1868), pp. 232242, though Schlafli's results are expressed in the notation explained in 4-15. The integrals have also been examined in great detail by Gubler, Zurich Vierteljahrsschrift, xxxm. (1888), pp. 147 172, and, from the aspect of the theory of the linear differential equations which they satisfy, by Graf, Math. Ann. xlv. (1894), pp. 235 2C2 lvi. (1903), pp. 432 444. See also de la Vallee Poussin, Ann. de la Soc. Sci. de

and

K v (z)

Bruxelles, xxix. (1905), pp.

140143.

6-1]

INTEGRAL REPRESENTATIONS
integration.

161

by a partial
if T, a, b

Accordingly we obtain a solution of Bessel's equation

are so chosen that


%- {T(t>-l)}
t

= (2v + l)Tt,

izt

T(t2 -l)

= 0.

The former of these equations shews that


(P

is

a constant multiple of

I)" - *,

and the
t

latter

either so that
initial

it is

shews that we may choose the path of integration, a closed circuit such that e izt (t2 l) v+i returns to its
e
izt

value after

has described the circuit, or so that

(&-

l) v + i vanishes

at each limit.

A
t

contour of the

first

type

is

a figure-of-eight passing round the point


t

=1

counter-clockwise and round

clockwise.

And,

if

we suppose

temporarily that the real part of z

is

positive, a contour of the second type is

one which starts from

+ coi and

returns there after encircling both the points

1,

1 counter-clockwise (Fig. 1

and Fig.

2).

If

we take

a, 6

1, it is

Fig. 1.

Fig. 2.

necessary to suppose that


integral.

R(v +

> 0, and we
(t
2

merely obtain Poisson's

To make the many- valued


of
t

function

1)""* definite*, we take

the phases

and

1 to

vanish at the point


t

where the contours

cross the real

axis on the right of

1.

We

therefore proceed to
/(1+,

examine the contour integrals


/-(-i+.i+j

-1-)

z"

e (a - 1)"-* dt,

z"

eizt

(t*

- 1)--* dt.

at

* It is supposed that v has not one of the values \,\,\\ 1, and both integrals vanish, by Cauchy's theorem.

...;

for then the integrands are analytic

162

THEORY OF BESSEL FUNCTIONS

[CHAP. VI

It is to be observed that, when R (2) > 0, both integrals are convergent, and differentiations under the integral sign are permissible. Also, both

integrals are analytic functions of v for all values of

v.

In order to express the

first

integral in terms of Bessel functions,

we

expand the integrand in powers of z, the resulting series being uniformly convergent with respect to t on the contour. It follows that
/"(1+.-1-)
00

zu

Ja
2

J*(P-l)'-*dt= 2
m=o
or

;m z Vrm,

^-r-jm+, -1-) V*(P-l)*dt.


ml
Ja

an odd function of t according as m is even or odd; and so, taking the contour to be symmetrical with respect to the origin, we see that the alternate terms of the series on the right vanish, and we are
then
left

Now tm (t I)" - * is an even

with the equation

t"
J

J* (t* a

1)"-* dt = 2

= 2
on writing
u
is
t

(2m)! m-o w (\m z v+2m


K

;a

t* (t2

1)"-* dt

Jo
,-(1+)

'

w -o

(2m)

,
!

u"-i(u-l)-*du,
Jo

= <Ju;

in the last integral the phases of


1.

u and u

vanish

when

on the real axis on the right of u =

To evaluate the

integrals

on the right, we assume temporarily that


into the straight line
to
1,

R(v + %)>0;
from
to
1

the contour

may then be deformed


first part,

taken twice; on the


,

u 1 (1 u)e~ vi u 1 = (1 ii) e +ni We thus get


fd+>
I

going from and on the second part, returning from where, in each case, the phase of 1 u is

1 to 0,

we have we have

zero.

*-* (u

1)"-*

Jo

du

= =

{-(-
. 2t COS

- e<"- '}

n
/

um -l (1 - )"- du
'
.

Jo
Z>7T

Now

r(m + )I> + ) Z \,/ - r(m + /;+l)

both sides of the equation


"

(1+

V- ( -iy-iau- u cos ri + r(+


r (w + v +
i)

are analytic functions of v for all values of v) and so, by the general theory of analytic continuation*, this result, which has been proved when (v + ) > 0,

persists for all values of


*
Ii ("

v.

Modern Analyst*,

5-5.

The

reader will also find

it

possible to obtain the result,

when

+ )~-0,

by repeatedly using the recurrence formula

Jo
which
is

++J Jo
(>,

obtained by integrating the formula

l{ u '+i( U -l)-'+'+i. =(Ml + v + n + l )Mw-i (u _ ir + M +l +


the integral
is

+ n + i)um-*

{M

1)

v+ M -i.

then expressed in terms of an integral of the same type in which the exponent of

u - 1 has a positive real part.

6-1]

INTEGRAL REPRESENTATIONS
for all* values of v,
oo

163

Hence,

r(l+, -1-)
J

(\m z v + *m Y( m
</ (z).

a.

1\
l)

m-o (2m)!r(j/+m +

= 2" +i i r (i) r (i + $> cos i/tt


Therefore,
(i)
if v
-f-

is

not a positive integer,


,

j.

w - ^"ty /r
|

" ** ((> - i)"

and

this is Hankel's generalisation of Poisson's integral.

Next let us consider the second type of contour. Take the contour to lie wholly outside the circle 1 = 1, and then (t 2 l)" -i is expansible in a series of descending powers of t, uniformly convergent on the contour thus we have
1

m=o

(|

and

in the series the phase of

lies

between

it and + ^tt.
i')

Assuming^ the
/(-i+,i+)
J<*>i

permissibility of integrating term-by-term,


<*>

we have

zy

TVA

A-m\ r(-i+.i+)
V)
J
oot

m -Q Wil (-

But
J

oo i

aa

exp ia

where a

is

the phase of z (between + \tr)\ and, by a well-known formula*, the

last integral equals

1iri\Y (2m 2v + 1).

Hence

,J

r
*><

'

1+)

*- <* - 1)-'

s * - ,-o

"

->7;%7r " ( - r (2m i~; 2^


!

r(

v)

-h

1 1)

when we use the duplication formula T(%-v + m) and T (- + m + 1).


1/

to express

T (2m

2i>

1) in

terms of

* If v

-J

is

a negative integer, the simplest

way

of evaluating the integral is to calculate the

residue of the integrand at

u= 1.
/
'
|

t
its

To

justify the
A'.

term-by-term integration, observe that


(t
2

tzt

dt

is

convergent

let

are given a positive

- l)" it follows that, wheu we independent of t, such that the remainder an integer We then terms of the expansion does not exceed e/K in absolute value when M ^ after have at once
value be

Since the expansion of

-^

J oot converges uniformly,

number

e,

we cau

find

cot

jnir{-i>) J aci + > ,izt f(-l+.l1 '\e tzt dt\=e, V <eK--1


=o
'

aud

J cot the required result follows from the definition of the

sum
Cf.

of

an

infinite series.
12-15.

X Cf. Modern Analysis, 12-22.

Modern Analysis,

164
Thus,

THEORY OF BESSEL FUNCTIONS


when

[CHAP. VI

R (z) >

and v + %

is

not a positive integer,

This equation was also obtained by Hankel.

Next consider
(-1+.1+)
eizt

{t*-iy-*dt,

J oo
of z which
is

exp ( - 1*>)

where co is an acute angle, positive or negative. This integral defines a function


analytic

when

\ir +
and,
if

a>< arg z <

fyrr

+ co;
| J

arg 2 < 77-, the contour can be deformed into the second of the two contours just considered. Hence the analytic continuation of J- V (z) can be defined by the new integral over an extended range of values of arg z; so that we have
z
is

subject to the further condition that

(3)

J-*(')~
giving
co

r^-\,>" n7V/'(-i+,i+)
f

\j?rm
.1Tl

X \-%)

f ociexp(-Ju)) J

**V-iY-*dt,
\ir

where arg z has any value between

\ir + to tr

and

co.

By
for

a suitable value*,

we can

obtain a representation of J_ (z)

any assigned value of arg z between

and

it.

When R (z) > and ^( + J)>0we may

take the contour to be that shewn in Fig.

3,

\/ /\

Fig. 3.

in

which

By

it is supposed that the radii of the circles are ultimately taking each straight line in the contour separately, we get

made

indefinitely small.

J-> {z)=

T{

K^n\)

ei

[/I
+

"

' sni(v ~

h)

(1

-^" *
i

Je * e-3rC4
1

J)

(1

- #)-\

dt

izt e -Hi>-l)

(\-t*y-l dt

e*t

e^C-V (l-Py-h dt
(1

+
* If
I

f*

e* **("- i)

- t*y~i dt~]
*-),

w be
I

increased in a series of stages to


valid for

sentation of

J- v (z)

an appropriate value (greater than any preassigned value of arg z may be obtained.

a repre-

6-11]

INTEGRAL REPRESENTATIONS
bisecting the third path of integration
t,

165
by
it,

On

and replacing
is

in the various integrals

-t, t,

it

respectively,

we

obtain a formula for

J- V (z), due

to Gubler* which corre-

sponds to Poisson's integral


(4)

for

Jv (z)
S

the formula
" f jo

J_ v (s)=

\f W 7Y-J r (v+#) "($) |_


,

2{

mvn

e-'til

+ tty-hdt+f'cosizt + v^.il-fiy-idt],
Jo
it is

and,

if

this be combined with Poisson's integral,

found that

'V-Tb$&l[fl'*<*< 1

- f>rl

*-K'
it

rm,:l+ *r ~ >

*l

a formula which was also discovered by Gubler, though "Weber t in the case of integral values of p.

had been previously stated by

After what has gone before the reader should have no difficulty in obtaining a formula
closely connected with (1),

namely

(6)

*w- r -*k^/>-ir.~>.*
it is

in which

supposed that the phase of

t2

vanishes

when

is

on the

real axis
i

on the

right of t\.

6*11.

Modifications of Hankel's contour integrals.


6*1 into the contours

Taking R (z) > 0, let us modify the two contours of shewn in Figs. 4 and 5 respectively.

Fig. 4.

Fig. 5.

By making

those portions of the contours which are parallel to the real xxxm.
(1888), p. 159.

* Ziirich Vierteljahrsschrift,

See also Graf, Zeitschrift fur Math, und

Phys. xxxvin. (1893), p. 115.


pp. 8690.

t Journal fUr Math, lxxvi. (1873), p. 9. Cf. Hayashi, Nyt Tidsskriftfor Math. xxni. b, (1912), The formula was examined in the case v=0 by Escherich, Monatshefie fiir Math,
(1892), pp. 142, 234.

und Phys. m.

166
axis

THEORY OF BESSEL FUNCTIONS


move
off to infinity (so that

[CHAP. VI

the integrals along them tend to zero),

we

obtain the two following formulae:

0)

J.{z)

2ttiT(i)
d+)
e
izt

(t

l)"-i dt

+
J

e izt (t2
-1+ooi

iy-i dt

_J l+i

(2)

J _{*)-

2 7nT()
r(i+)
e
.'

rzt

(t

f(-i+)

I)""* dt

+
J

eizt (t2
-1+ooj

- 1)*-* dt

l+ooi

In the

first

result the
t
2

the phase of

1
is

to

be

many-valued functions are to be interpreted by taking at A and to be + ir at B, while in the second the

phase of

t-

at

and

is

it

at B.

in
t
1

To avoid confusion it is desirable to have the phase of t2 1 interpreted the same way in both formulae; and when it is supposed that the phase of
is

1
(3)

+ 7r at

B, the formula (1)

is

of course unaltered, while (2) is replaced

by

J-.(*)1

2r(i)
/(i+) d+)
e
J 1+ooi
izt

(t

1)"-* dt

f(-i-)
e
J -1+ooi
izt

-i

+ e~ vin

(t

1)"-* dt

J
2

In the

last of these integrals,

the direction of the contour has been reversed


t

and the alteration in the convention determining the phase of


necessitated the insertion of the factor e~ 2{y ~^ )H
.

1
(2),

has

On
(4)

comparing equations (1) and (3) with

3'61

equations (1) and

we

see that

T(\-v).(\z

y[^

(5)

unless v

is

an integer, in which case equations

(1)

and

(3) are not independent.

We
value

can, however, obtain (4)

and

(5) in the case


all

when

v has

an integral

from a consideration of the fact that continuous functions of v near v n. Thus


(n),

the functions involved are

HJ

(z)

lim

HJV

(z)

dt

and similarly

for

Hn

(z).

6-12]

INTEGRAL REPRESENTATIONS

167

(5)

As in the corresponding analysis of 6*1, the ranges of validity of (4) and may he extended hy swinging round the contours and using the theory of
Thus,

analytic continuation.
if

\nr < a> < f ir, we


J5Tr

have
Z)V
*""

(6)

(z)

T(

*~rn?

f
" 1_)

(*

W*

dt '

while, if

f 7r < a>< \ir, we have

(7)

gw(#)B E fl-^-y):
7r * * <.)

^(^-i)"-*^
*exp(-t)
lies

provided that, in both (6) and

(7),

the phase of z

between

\ir + <u

and

Representations are thus obtained of

H^ (z) when
s has

arg z has any value

between ir and 2tt and jr.

2tt,

and of

{z>

(z)

when arg

any value between

If co be increased beyond the limits stated, it is necessary to make the contours coil round the singular points of the integrand, and numerical errors are liable to occur in the interpretation of the integrals unless great care is taken. Weber, however, has adopted this procedure, Math. Ann. xxxvn. (1890), pp. 411 412, to determine the for-

mulae of 362 connecting


Note.

HM
V

- z),

HW
V

- z)

with

HJ (z),
it

V V) (*).

The formula

2i*F

(z)

= ffvW(z) Hv V)(z) makes

possible to express

Yv (z) in

terms of loop integrals, and in this manner Hankel obtained the series of 3'52 for Y(); this investigation will not be reproduced in view of the greater simplicity of Hankel's other method which has been described in 3-52.

6*12.

Integral representations offunctions of the third kind.

In the formula 6'11 (6) suppose that the phase of z has any given value between tr and 2tt, and define by the equation
arg z
so that

a)

\ir<fi<

+ @,

\tr.

Then we

shall write

so that the phase of

= e-l z- (- u), u increases from tt + /3 to


t

ir

+ fi
/

as

describes the

contour
(1)

and

it

follows immediately that

*,"(*)-

iTilt

^x*) -.Ur^-K
+ \iujz
has
its principal value.

vSe* I'-l**

t>

f o+i

in \ _ i
1

^)
if /3

du

where the phase of

Again,

be a given

acute angle (positive or negative), this formula affords a representation of

H w (z) valid over the sector of the s-plane in which

\ir

<

arg z

<

f ir

+ 0.

168

THEORY OF BESSEL FUNCTIONS


611
(7),

[CHAP. VI

Similarly*, from

where

is

any acute angle (positive or negative) and


3-61 (7),

Sincef, by

by restricting

v so that

f 7r 4- /S < arg z <%ti- + /3. #_W (*) = "* # (*), it follows that we lose nothing R (v + |) > and it is then permissible to deform
;

the contours into the line joining the origin to

oo

exp i/3, taken twice

for the

integrals taken round a small circle (with centre at the origin) tend to zero with the radius of the circle J.

On
(3)

deforming the contour of

(1) in the specified

manner, we find that


*,.

^-(i)
/3

'tf/.

r-^(l + )
.

where

may

be any acute angle (positive or negative) and

R (v + \) > 0, - |tt + j3 < arg z < f v + 0.


In like manner, from
(4)
(2),

ff.(,).(l.)

T?

_r
-fir

Jt

^^ (!_)

*,,

where

# may

be any acute angle (positive or negative) and

R(v + $)>0,

+ <arg*<lir + .

The results (3) and (4) have not yet been proved when 2v is an odd positive integer. But in view of the continuity near p = n+$ of the functions involved (where n = 0, 1, 2, ...) it follows, as in the somewhat similar work of 6'H, that (3) and (4) are true when v \, f f .... The results may also be obtained for such values of v by expanding the integrands
,
,

in terminating series of descending powers of

z,

and integrating term-by-term; the formulae

so obtained are easily reconciled with the equations of 3*4.

The general formulae

(3)

and

(4) ar_e of

fundamental importance in the


for large values of \z\.

discussion of asymptotic expansions of

Jv (z)
is

These

applications of the formulae will be dealt with in Chapter vn.

A useful

modification of the formulae

arg z = /3 (so that arg z is u 1z cot 0, it follows that


<5 >

due to Schafheitlin. If we take restricted to be an acute angle), and then write

^*)~
H<*(,) "

r(y+1)r(|)

W
( 6) *

W=
t

eoe-lg.ar**-* ^ sin T(ir+J)r(i)Jo


2 " +1

^
z" 1
{

s^h,,

r"eW.
e
dtf >

To

obtain this formula, write

+ 1 = 6"^

- u), ~ U( ~

- l = 2e

w (1

- iujz).

f There seems to be no simple direct proof that

T{h - v)
is

2
vi,

u)V

~^ i+i^

hdu

an even function of v. % Cf. Modern Analysis,

12-22.

Journal fur Math. cxrv. (1894), pp. 3144.

V
6*13]

INTEGRAL REPRESENTATIONS

169

and hence that


(7)

J "W-r(

y +|)T(i)Jo

sin+^
cos-*
f**
.

**'

to

-*-

cos (

- vO + \6)

^ cntede
were

These formulae, which are of course valid only when


( 15'32-*-15*35).

R (v + 1) > 0,

applied by Schafheitlin to obtain properties of the zeros of Bessel functions

They were obtained by him from the consideration

that the

expressions on the right are solutions of Bessel's equation which behave in the

appropriate manner near the origin.

/ e-'" uv ~i (l + tt)'1- * du, which


occurring in (3) and (4)
lix. (1904), pp.

is

reducible to integrals of the types

when

fi

= v, has

been studied in some detail by Nielsen, Math. Ann.

89102.
from the aspect of the theory of asympby Brajtzew, Warschau Polyt. Inst. Nach. 1902,

The
nos.
1,

integrals of this section are also discussed

otic solutions of differential equations

2 [Jahrbuch iiber die Fortschritte der Math. 1903, pp. 575677].

6*13.

The generalised Mehler-Sonine

integrals.

of a positive variable of a suitably restricted order.

Some elegant definite integrals maybe obtained to represent Bessel functions To construct them, observe that, when z is positive (= x) and the real part of v is less than \, it is permissible to take a = ?r in 6*11 (6) and to take to = - \ir in 611 (7), so that the contours are those shewn in Fig. 6. When, in addition, the real part
is

of v

greater than
of

it is

permissible to deform the contours, (after the

manner
to

612)

so that the first contour consists of the real axis from

taken twice while the second contour consists of the real axis from 1 to oo taken twice.
oo

^A-'J
Fig. 6.

<rS,\

r,^< l'^
" e-^-^) f>< (* 2

We

thus obtain the formulae

HP (w) = r
#<*> (as)

-.{?7fr^- (1

1)"-* dt,

=-

^~$;$^

(1

- e2 <"-i>)

te < (

- l)r~* dt,

the second being derived from

611

(7)

by replacing thy t.

170

THEORY OF BESSEL FUNCTIONS


In these formulae replace v by

[CHAP. VI

given by 361

(7).

It follows that,

when x >

and use the transformation formulae and \<R(v)<\ then


t

(2)

bj* (x) - r
"
/
v

iT{ i-

v)

r(i).($xy
f"
J i

11

'-ir*
- i)- + *

so that

w
(4)

,o\

w " r(i-.or(*).(iy
2
(x)

ain (xt).dt
(*
2

=-

f <*<*>*
.'i

r<*-iOr(*).tt*) F

(*-ir**

Of these results, (3) was given by Mehler, J/a*A. Ann. v. (1872), p. 142, in the special case v==0, while Sonine, Math. Ann. xvi. (1880), p. 39, gave both (3) and (4) in the same special case. Other generalisations of the Mehler-Sonine integrals will be given in & 621.
6 "14.
Hargreave and Macdonald.
it is

Symbolic formulae due

to

When R (*) > and R (v +) > 0,

evident from formula 6-11

(6)

that

where the phase of


If

t2

lies

between

and

\n.

D denotes (d/dz) and /is any polynomial, then


f{it).e<=f(D)e",

and

so,

when

v+i

is

a positive

integer,

we have

r(v+i)r(i)

(1+

When

+\

is

representation of HJV>

not a positive integer, the last expression may be regarded as a symbolic (z), on the understanding that/ (D) (e^'jz) is to be interpreted as
Afci
i
I

M /(it)dt.
*

J i+i

Consequently
(1)

(1)

(?)=

r(v+i)r(i)
,

(l+^)
(i

and similarly
(2)

#<

W - r(+J)r(j) m

+i^- 5Elti22L'
'

so that
<

3>

jr

'-f5lrIB (1+1 ^-..

<

4>

r'Vfi$fmO +Br**T-

6-14, 6'15]

INTEGRAL REPRESENTATIONS
from
(4)

171

The
unless

series obtained

by expanding in ascending powers of

does not converge

it

terminates; the series obtained in a similar manner from (3) converges only

when R(y)>\.
The expressions on the right of (3) and (4), with constant factors omitted, were given by Hargreave, Phil. Trans, of the Royal Soc. 1848, p. 38 as solutions of Bessel's equation. The exact formulae are due to Macdonald, Proc. London Math. Soc. xxix. (1898), p. 114.

An

associated formula, valid for all values of

v, is

derivable from 6'11

(4).

If

is

any

positive integer,
v

we

see from the equation in question that


z)
"

v) H W(z) = T ^~ 7rtr ;^
C?/

U+)

f
J
l

(<2-l)"--l(-).(l + Z)2)e*<fc

+ o

= (-)"r(W)(^
so that
(5)

p
(1+i)2)W

{ti

_ iy - n . iiztdtf

Bv ,_ r(.+t-.).) F (1 +i)J).

m^ H^
^
',

w)

A similar equation holds for the other function of the third


(6)

kind, and so
(z)} -

^^ = r(y+rV+i) (K
result,

"y

^-"

This
v

proved when
;

R (z) > 0,

is

easily extended to all values of z

by the theory of

analytic continuation

it

was discovered by Sonine, Math. Ann.

xvi. (1880), p. 66,

when
in the

= r and
when

used by Steinthal, Quarterly Journal,


v

xvm.

(1882), p. 338

when

= n+%;

case

= n+%

of Bessel functions)

the result was given slightly earlier (without the use of the notation proof by Glaisher, Proc. Camb. Phil. Soc. ill. (1880), pp. 269271.

based on arguments of a physical character has been given by Havelock, Proc. London Math. Soc. (2) II. (1904), pp. 124125.

6*15.

Schlafii's* integrals of Poissons type for


<o =

Iv (z) and
iz,

v (z).

If
i

we take
|

\tt in 6*1 (3)

and then replace z by

we

find that,

when

arg z

<

\tt,

and the phase of

at the point where

crosses the negative real axis is

2tt.

-1

Fig. 7.

4
of the circles

If

we take

R (v + ) >

to secure convergence, the path of integration

may be taken to be the contour of Fig. 7, in which the radii may be made to tend to zero. We thus find the formulaf #)r r( (* 2
T-" {2) =

*2^r<i)

(1

" <^> i?~* ( ' ~


+
i (e

dt

- ""

+ e-3vni )

e~zt (1

- <2)"-* dt

* Ann. dt Mat. (2) i. (1868), pp. 239241. Schlafli obtained the results (1) and (2) directly by the method of 6-1. t Cf. Serret, Journal de Math. ix. (1844), p. 204.

172

THEORY OF BESSEL FUNCTIONS


t
2

[CHAP. VI

in which the phases of

and of

are both zero.

Now, from

3'7l (9),

we have
(2)

and so
(

-T<MmL'*<
/_, (.)
to say*

-*r-

- irt

3)
is

_ /.

W = r <*-*?*"<* .c*"
v

((a

that

(4)

K
K
v

(,)

^j^ffi J"*-*

(*

- I)""* dt,

whence we obtain the formula


(5)
(z)

T = ^) ^iT [V* c08h< sinn^A, v


>

a result set by Hobson as a problem in the Mathematical Tripos, 1898. The and args < \ir. The reader will formulae are all valid when R (v + \) > find it instructive to obtain (4) directly from 6*11 (6).
j

6*16.

Basset's integral for


is

K
(

(xz).

When x arg * < i


i

positive

and z

is
v

a complex number subject to the condition

77"'

tne integral for

H ^ (xze^) derived from


r( + *)-(i*>(1+)

611

(6)

may be

written

in the form
v> _ (xze )^xze n_ H
v

*
(^-l)-+i-

^T(^)

f J +00

Now, when R(v)^

\,

j"-t-*arg Z) tends to zero as p p e

the integral, taken round arcs of a circle from p to <x> by Jordan's lemma. Hence, by Cauchy's

theorem, the path of integration

may be opened
zt

out until

it

becomes the
2
2

line

on which

R (zt) = 0.

If then

we write
(8) that

= iu,

the phase of

- (u /z ) -

1 is fl-

at the origin in the it-plane.


It then follows from

37

(xz) = \Ttie-l H% (xze^)

-*" r (v + ) _ = 2TQ)

(%xz)~ v

r xi z
'

)i,z
f

e~ xzt dt * (* -l)" +
2

r (V 4-

1)

00 (2z) v f*

_e e-" du

(i)

J->(^

and so we have Basset's formula


Kl)

^
when

\xz)a;

V (\)
1

] o

(u2

+z

v+ i'
)

valid

R (p + |) ^ 0,

>

0,

arg 2

< \tt

The formula was obtained by

Basset f,
*

for integral values of v only, by regarding

K (x)

as the limit of

Chemie,

was examined in the case i=0 by Biemann, Ann. der Physik und 130139. (Cambridge, 1888), p. 19. t Proc. Camb. Phil. Soc. vi. (1889), p. 11 ; Hydrodynamics, n.

The

integral on the right

(2) xcv. (1855), pp.

6*16, 6-17]

INTEGRAL REPRESENTATIONS

173

a Legendre function of the second kind and expressing it by the corresponding limit of the integral of Laplace's type {Modern Analysis, 1533). The formula
for

Kn (xz)

j
is

obtainable by repeated applications of the operator


{xz),

zdz
jis

Basset also investigated a similar formula for /


in his result.

but there

an error

Among
some

on the right in (1) was studied by numerous mathematicians before Basset. these investigators were Poisson (see 6'32), Journal de V&cole Polytechnique, IX. (1813), pp. 239241; Catalan, Journal de Math. v. (1840), pp. 110114 (reprinted with
integral
Serret,

The

Mem. de la Soc. R. des Set. de Liege, (2) xn. (1885), pp. 26 31); and Journal de Math. vm. (1843), pp. 20, 21; ix. (1844), pp. 193210; Sohlonrilch, Analytischen Studien, n. (Leipzig, 1848), pp. 96 97. These writers evaluated the integral in finite terms when v+$ is a positive integer.
corrections,

(1841), pp. 65

Other writers who must be mentioned are Malmsten, K. Svenska V. Alcad. Handl. lxii. -74 (see 7'23) Svanberg, Nova Acta Reg. Soc. Sci. Upsala, x. (1832), p. 232 Leslie Ellis, Trans. Camb. Phil. Soc. vm. (1849), pp. 213215 Enneper, Math. Ann. xi.
;

Glaisher, Phil. Trans, of the Royal Soc. clxxii. (1881), pp. 792 815 J. J. Thomson, Quarterly Journal, xvnr. (1882), pp. 377-381 ; Coates, Quarterly Journal, xx. (1885), pp. 250260; and Oltramare, Comptes Rendus de V Assoc. Francaise ' xxiv. (1895), part n. pp. 167171.
(1873), pp.
;

360365

'

The

last

named

writer proved
.

by contour integration that


1

/ cos xu du
o

(u*+z*)*

_ (-) n-1 jr r d"' ~ 2**-. (n-1)! [dp^

(~)w
~ iff
i

/e~ xs ^P\~\

\~^fr~)] p=1
~

The former

of these results

r d*~ e xtp i ~^-i.(n-l)! U/>B - 1 (l+jo)J p =r may be obtained by differentiating the


C

equation

cos

Jo
and the
latter is

xu du u*+z2 p ~
.

ire~ xz ^P

2zjp

'

then obtainable by using Lagrange's expansion.

6' 17.

Whittakers* generalisations of Hankel's integrals.


3'32

Formulae of the type contained in

suggest that solutions of Bessel's

equation should be constructed in the form


z*

fa J

eia Tdt.

It

may be shewn by

the methods of 6'1 that

and

so the integral is a solution if T is a solution of Legendre's equation for functions of order v - \ and the values of the integrated part are the same at each end of the contour.
* Proc.

London Math. Soc. xxiv.

(1903), pp.

198206.

174
If

THEORY OF BESSEL FUNCTIONS

[CHAP. VI

and end at

taken to be the Legendre function Q_j (t), the contour may start is an acute angle (positive or negative) provided that z satisfies the inequalities

T be

+ ootexp( ia>), where w

^7r + o>< arg Z<^TT + CD.


If

T
it

be taken to be P_j()> the same contour


{t)

is

possible;
is

but the

logarithmic singularity of P_j

at

= 1

(when

not an integer)

makes

impossible to take the line joining

to 1 as a contour except in

the special case considered in 3*32; for a detailed discussion of the integral
in the general case, see 10'5.

We now proceed
First consider

to take various contours in detail.

oo

iexp (tui)
t

where the phase of

t is

zero at the point on the right of


lie

=1

at which the

contour crosses the real axis. Take the contour to


| |

wholly outside the circle


It is thus found, as in

and expand

Q v -{t)

in descending powers of

t.

the similar analysis of


(1)

6*1,

that
}

(*)

P Kt) 7rl
,

xiexp(-iw)

#* Q.-i (0 dt,

and therefore
(2)

J_ r (,) = "*>

S* Q__j (I)
^f^
J o&iexp(-i<o)

<fc

vr 1

If

kinds of Legendre functions,


#,<*> (*)

we combine these formulae and use the we find that

relation* connecting the two

(3)

^f

4* P,_i (0

(ft.

Again, consider

this is a solution of Bessel's equation, and, if the contour

right of the line


z
_.
i.

R (t) = a,
that

it is

clear that the integral is


is

+ oo Hence -* oo i, we find

the integral

a multiple of

be taken to lie on the [z* exp (a\z ))} as


Similarly by

(l i

(z).

making

J
*

oo *

exp ( i)

The

relation, discovered

by

Schlafli, is

cf.

Hobson,

Phil.

Tram, of

the

Royal Soc. clxxxvii.

(1896), p. 461.

6-2]
is

INTEGRAL REPRESENTATIONS
(z).

175

a multiple of Zf(2)
(4)

^
9

a,

<*)-

(5)

H
J5Tr

(z)

WJrm - W ra
'
v

From a consideration
I

of (1)

it is

then clear that

**Q.-Mdt,
e'*

&- (*> dt

>

and hence, by 361 combined with


(6)

Schlafli's relation,

(#) from

ffl (I) cos vir J ,- P


(3).

^^
(l*\i

**<" + >

r(-l+)

(-M

4* P,_ x <fc; V * (0

this is also obvious

The
of

integral which differs from (6) only

by encircling the point

+1

instead

is

zero since the integrand

is

analytic inside such a contour.


real axis

In (5) and (6), arg (t+ 1) vanishes where the contour crosses the on the right of 1, and, in (5), arg (t 1) is ir at that point.
6*2.

Generalisations of Bessel's integral.

We shall next examine various representations of Bessel functions by a system of definite integrals and contour integrals due to Sonine* and Schlaflif. The fundamental formula which will be obtained is easily reduced to Bessel's integral in the case of functions whose order is an integer.
We
integral
1
1 (v

take Hankel's well-known generalisation J of the second Eulerian


1 1)
/(

+>
'

+m+

Ittx) _,

in

which the phase of

increases from
(|,) 2iri

- ir

to

ir

as

describes the contour, and


)2wt
>t

then

{-r(\ z y+ _ m= om\r(v + m+l)

no + ) _)m

,t,J

m!

* at"

Consider the function obtained by interchanging the signs of summation

and integration on the

right; it is

This

is

an analytic function of z

for all values of

z,

and,

when expanded

in

ascending powers of z by Maclaurin's theorem, the coefficients may be obtained by differentiating with regard to z under the integral sign and making z zero
after the differentiations^

Hence

(Moscow, 1870) ; Math. Ann. xvi. (1880), pp. 929. t Ann. di Mat. (2) v. (1873), p. 204. His memoir, Math. Ann. m. (1871), pp. 184149, should also be consulted. In addition, see Graf, Math. Ann. lvi. (1903), pp. 423432, and Cbessin, John*

* Mathematical Collection, v.

Hopkins University Circulars, xiv. X Cf. Modern Analysis, 12-22.

(1895), pp. 20

21.
Cf.

Modern Analysis,

532,

4-44.

176

THEORY OF BESSEL FUNCTIONS

[CHAP. VI

and so we have at once


(i)

*W-^n '-pI'-sI*
,

This result, which was discovered by Schlafli, was rediscovered by Sonine;

and the

latter writer
|

was the

first

to point out its importance. until

When
it

arg z

< \ir, we may swing round the contour about the origin

passes to infinity in a direction

making an angle arg z with the negative

real axis.

On
< 2>

writing

\zu,

we then
1

find that,

when arg z <


|

\tr,

J' = 2S
= e w we
,

f(+) 1\) .L """ e*P |** [U ~ i)\ du


(

This form was given in Sonine's earlier paper


Again, writing u
(3)

(p.

335).

have
j
r oo

+ B-t
e
first
z8inh '-

/., (z)

= =-.
is

dw,
by
Schlafli.

valid

when arg^ <


|
|

\ir.

This

the

of the results obtained

In this formula take the contour to consist of three sides of a rectangle, as in Fig. 8, with vertices at oo tri, tri, vi and x> + iri.

ni

-ni
If

Fig. 8.

4
w
-

we

write

+
l

iri for

w on
iri,

the sides parallel to the real axis and id for

on the
(4)

lines joining

to

we

get Schlafli's generalisation ofBessel's integral


z sin 0) d$

Jv (z) = - J'cos {yd when arg z <


|

^^ J"
known

-t-z**t

dt>

valid

\tr.

If
if

we make arg z
true

~-

\tt,

the

first

integral on the right

is

continuous and,

R (y) > 0, so also is


when
z
is

the second, and

{z) is
if

to be continuous.

So

(4)

is still

a pure imaginary

R{v)

is positive.

The

integrals just discussed were

examined methodically by Sonine in his


:

second memoir; in that memoir he obtained numerous de. te integrals by be an acute angle appropriate modifications of the contour. For example, if

(positive or negative)

and

if

- tt + ^ < arg z < \ir - yjr,

6'21]

INTEGRAL REPRESENTATIONS

177

the contour in (3)


oo

may be replaced by one which goes from oo (w yfr) i to + (V + yjr) i. By taking the contour to be three sides of a rectangle with corners at oo {ir ^) i, (ir ^jr) i, (ir + ^r) i, and oo + (tt + yfr) i, we obtain,
as a modification of (4),
(5)

J(z) =

efc-^^cos^tf-scos^sintf)^
Jo
e -"*sin>7r r~

_, ginh ,,^_,

if

JO

Again, if we take ifr to be an angle between and ir, the contour in (3) may be replaced by one which passes from oo {\tr 4 ^r)i to oo + {\tr -4- >fr) t, and
so

we
(6)

find that

J,

(*)

=1

+
provided that
|

f cos(i/0-*sin0)d0 ""Jo 1 f g-zrinht *.*-,,(


I

gin (s^gjj

C08

^ - $ 1/W - 1^)
take

(ft,

arg z

is less

than both
positive

<ty

and

When
(7)

J?

(i)

>

and z

is

(= x),

^. we may
tt

^=
cosh
*

in the last

formula, and get*

J (#)=-!
I? J

cos

(i/0

- x sin 0) d0 +

I
|)

*->* sin

(a?

- Ai/tt) dt.

TT J

Another important formula, derived from (1), is obtained by spreading out the contour until it is parallel to the imaginary axis on the right of the origin; by Jordan's lemma this is permissible if R(v) > - 1, and we then obtain the
formula

in which c

may have any

positive value

this integral is the basis of

many

of

Sonine's investigations.
Integrals which resemble those given in this section are of importance in the investigation of the diffraction of light
(1899), pp.
;

by a prism see Carslaw, Proc. London Math. Soc. xxx. 121161 W. H. Jackson, Proc. London Math. Soc. (2) I. (1904), pp. 393414; Whipple, Proc. London Math. Soc. (2) xvi. (1917), pp. 94111.
;

6*21.

Integrals which represent functions of the second and third kinds.


Schlafli's integral

If

we substitute

62 (4)

for

both of the Bessel functions

on the right of the equation

Y
we
ir

(z)

= J (z) cot vir J_ v (z) cosec vir zamd)dd- cosec vir


cosv7r/
Jo
I

find that

(z)

= cot vir

cos (vO

cos (v$

+ z sin 8) dO
/>

J o

Jo
/"oo

e- t

-ZBinht

dt-l e**-""** dt.


Jo

Cf. Oubler,

Math. Ann. xlii. (1897), pp. 583584.

178
Replace

THEORY OF BESSEL FUNCTIONS


by
it

[CHAP. VI
it is

in the second integral

on the right, and

found on re-

duction that

(1 )

(z)

=-

! "sin (z sin

0-v0)d0--

7TJo

vt ( (e TTJo

+ e~

yt

cos vir) e"* 8inh *

dt,

a formula, practically discovered by Schlafli (who actually gave the corresponding formula for Neumann's function), which is valid when arg z < \ir.
| \

By means

of this result

we can

evaluate
tjzsinhw-i-w

when arg z <


|
\

r<x>
.

+iri

dw>
9,

TTIJ _oo

%ir

for

we take

the contour to be rectilinear, as in Fig.

and

ni

Fig. 9.

write

t,

i0, t

+
is

7rt

for

on the three parts of the contour

we then

see that

the expression
_
J

equal to

Too
.

yt-zsmhtfa

+_

fir
/

eHzsiiiO-ye)

0 +

g vni _
7TI

foo

e -v-sinh

<ft,

7rl.'o

'"Vo

Jo

and this is equal to Jv (z) + 1 F (s) from formula Hence, when arg z < \tr, we have
|
|

(1)

combined with

62 (4).

(2)

{1)

(*)

A
1

feo+iri

e*

Mnh W-' M dw,


'

(3)

#<*>(*)=-

"V *""""^.
811

pp. 327

Formulae equivalent to these were discovered by Sommerfeld, Math. Ann. xlvii. (1896), 357. The only difference between these formulae and Sommerfeld's is a rotation

of the contours through a right angle, with a corresponding change in the parametric
variable; see also

Hopf and Sommerfeld, Archiv der Math, und Phys.

(3)

xvm.

'{1911),

pp.

1-16.

By an obvious change
(4)
J5T.M (z)

of variable

we may

write (2) and (3) in the forms

=A
|

u--> exp ^z (u
o

- -^ du,

(5)

HJ* (*)

-.
I

Wo

<*P \\z u -u)\ du {

6-21]

INTEGRAL REPRESENTATIONS

179

the contours are those shewn in Fig. 10, emerging from the origin and then bending round to the left and right respectively results equivalent to these were discovered by Schlafli.
;

Fig. 10.

[Note.

There

is

no

difficulty in

proving these results for integral values of


;

v,

in view

of the continuity of the functions involved

cf.

6*11.]

proceed to modify the contours involved in (4) and (5) to obtain the analytic continuations of the functions on the left.
If
o> is

We

an angle between

tr

and

rr

such that

to

arg z < %ir, we have


|

(6)

HW
r

(*)

= -.
1
/>

exp hz ( u - )
exp (-)*
(

du,

and
/
is")

(7)

^"W-sL,,.

^p{*.(-J]
left

du,

the contours being those shewn in Fig. 11 and Fig. 12; and these formulae
give the analytic continuations of the functions on the

over the range of

Fig. 11.

Fig. 12.
to

values of z for which

\tr< args< m + ^rr;


difficulties

and

o>

may have any

value

between* v and
|
I

ir.

* If u were increased beyond these limits, the phase of u.

would

arise in the interpretation of

180

THEORY OF BESSEL FUNCTIONS


Modifications of (2) and (3) are obtained

[CHAP. VI

by replacing

why w \iri

it is

thus found that*


(8)

H^
v

(Z)

g^vwi foo+Jirt
re TTl J _ao-Jiri

i*cosh-vu>

fa

/"oo+Jiri

TTl

Jo
j'ao

eizcosh

v>

cogh vw dW)

(9)

(z)

= -^\
Wl

p\vni

^iri
e-izcoshu>-*w

dw

J -ao+tni

=
provided that arg z
| \

Qghrxi
:-

( Jtt*
e

-izcosh w

^h vW

^ Wj

7rt

Jo

< \tt.
by taking z
positive

Formulae of

special interest arise


1.

(=#)

in (6)

and (7)

and
take

1 < R (*>) <


a>

double application of Jordan's

lemma

(to circles of

large

and small radius respectively) shews that, in such circumstances, we may = %-rr in (6) and to = \tt in (7). It is thus clear, if u be replaced by
that
HJti (x) =
p\vai

ie l ,

(10)

^7T?

foo

eixcoaht-H fa
J_oo e

=
<ft

OgJnrt
.

Too
/

Tt

ecosh t

cogh

<ft,

JO
^-r"TTl
/

(11)

H (x) = -r\ Wl
v
a;

- ixcoah -
f

<r tecosht cosh

itf

<ft,

J -oo

7o

and hence, when


(12)

>

and

1 < R (v) <


<

1,

we have
cosh
i/tf .

Jv (x) =

2 r - j sin (x cosh 2 r
I

i>7r)

dt,

(13)

(x)
(cf.

cos (# cosh

vtt)

cosh vt.dt;

TTJo

and, in particular
/ta\ (14)
/, -.

6*13),

*<*>-; j, v^T)'
TT
,
.

2 fsinarf.<ft

(15)

f " cos #

<fo

when we
The

^'"iJ.W^U'
replace cosh
t

by

two formulae are due to Mehler, Math. Ann. v. (1872), p. 142, and Sonine, Math. Ann. xvi. (1880), p. 39, respectively ; and they have also been discussed by Basset, Proc. Camb. Phil. Soc. "in. (1895), pp. 122128.
last

slightly different

form of (14) has been given by Hardy, Quarterly Journal, xxxn.


in (14)

(1901), pp.

369384;

if

we

write

x=2

y/(ab),

xt=au+b/u, we

find that

(16)

f"
The reader

n (+)
/"

^= "A {2 J{ab)}.
from the formula
,.
rf

Note.

will find it instructive to obtain (14)


i>

n i cos0)=as (
(1).

sin (

combined with the formula 5*71

)} j v{2(co This was Mehler's original method.

^^
+ )*

* Cf. Coates, Quarterly Journal, xxi. (1886), pp.

183192.

6-22]
6*22.

INTEGRAL REPRESENTATIONS
Integrals representing

181

Iv (z) and

(z).

The
due to

modifications of the previous analysis which are involved in the dis-

cussion of

Iv {) and
t)

(z) are of sufficient interest to

Schlafli*,

function F(a,

though he expressed of 415.

his results

be given fully; they are mainly in terms of the

The

analysis of

62

is

easily modified so as to prove that

/-<>- (
|

&L'-*(' + s)*
\ir,

and hence, when arg z <


<2 >

''W- s?/.T u"" exp


1

*(*
f

1)}

du-

foo -Hri

(3)

/ (z)=er-.
(2)

^coshw-wr fa

ITTIJ aa-ni

The formulae
at 00
(4)

and

(3) are valid

when arg 2 = $-rr

if

R(v) > 0.

If in (3) the contour is taken to be three sides of a rectangle with corners

in, tri,

in, 00

+ iri,

it is

found that

/,(*)*

ifV^'coBi/tfcW-^^f
ttJo

-***--cft,

jo

so that

/_.(*)-/,(*)

= ^52^ f V^'coshitf.cfc, T Jo
=
I

and hence, when arg .z <


|

\tr,

(5)

jBT.,

(z)

e-"osh * C osh

i><

<ft,

a formula obtained by Schlafli f by means of somewhat elaborate transformations.

From

the results just obtained,


1

we can evaluate
ezcoshw-vw

foo+irt

fa

when arg z <


j

\ir.

For

it is

easily seen that


1
f
I

/*oo+irt

/ iri
J

roo+irtl

escoshto-^afo^
^"n"* I

+
-0O-1T*

gzcoshw-rv, J

ATTIJ -OO-Trt

(../

JoO-lrt)

raoiri

= 5

ezeoshw-vw
Too

ATTIJ -a>- w i
gviri

^w + /^ (^

= 5.

e-*cosh*- dt

+ jv (^

* 4nn. di Jlfot. t Ann. di Mat.

(2) v. (1873),

pp.

199205.

(2) v. (1873),

pp. 199

201

Math. lxix. (1868),

p. 131, as the definition to

this formula was used by Heine, Journal which reference was made in 5*72.
;

fiir

182

THEORY OF BESSEL FUNCTIONS

[CHAP. VI

and hence
(6)
2ttc

2i sin vtt

Again,

we may

write (5) in the form

(7)

K
1 j

(z)

= ^r

e -*cosht-,.t dt}

and hence, by the processes used in


yooexp(-i*>)

621,
exp-J

(8)

K, <*)
7r

^
I / Oexpt'u

\z (u +

-\\ du,

when 7r < &><


Similarly
(9)

and

\ir + a> < arg z <

\tt

+ a>.
- " -1
f
/

e"" /_ (*)
valid

- e~ vwi Iv (z)

sinv7T /""expOr-ult

Jx")

J Oexp (-+)

exp -Us ( u
(

+ - )\ du
tt/J

this

is

when
for

< a>< 2ir and $tt + &> < arg z <

^7r

+ eo.

The contours
respectively.

the formulae (8) and (9) are shewn in Figs. 13 and 14

Fig. 13.

Fig. 14

Further,
tion in (8)

when

is

positive

(= x) and

1 < R (v) <


it

may
;

be swung round until


it is

1, the path of integrabecomes the positive half of the

imaginary axis

thus found that


v-"-1 exp

K* 0*0 = i r_ *"**
so that

- \ix

(v

dv,

(10)

(x)

= le-fc"* f
j'*

e~ i* 9inht -' t dt,

J -ao

and, on changing the sign of


(11)
JSrr

v,

<

(*)=*J*'

<ri*sinht+,t

(fc i

6*23]

INTEGRAL REPRESENTATIONS
these results

183

From
(12)
so that

we

see that
.

2 cos | mr

K
^

/oo

(x)

e-**sinh t cos h vt

^
.

(13)

(x)

=
all

cos

(a;

sinh

t)

COS|l/7Tj

cosh

j/

dt,

and these formulae are In particular


(14)

valid

when x >

and

- 1 < R (v) <

1.

jr.<.)-(*.<ho*-["!=|2fc* .'o Jo v(f + -U


=T

a result obtained by Mehler* in 1870.


It

may

be observed that

if,

in (7),

we make the substitution

^ze t

we

find that

(15)

jr.W-K^^expj-r-^}^,
R (z2 > 0.

The integral on the right has been studied by numerous mathe) among whom may be mentioned Poisson, Journal de VEcole Polytechnique, ix. (cahier 16), 1813, p. 237; Glaisher, British Association Report, 1872, pp 1517; Proc. Camb. Phil. Soc. in. (1880), pp. 512; and Kapteyn, Bvll. des Sci. Math. (2) xvi. (1892), PP 4144. The integrals in which v has the special values \ and f were discussed by Euler, Inst. Calc. Int. iv. (Petersburg, 1794), p. 415; and, when v is half of an odd
provided that
maticians,
integer, the integral

has been evaluated by Legendre, Exercices de Calcul Integral,

I.

(Paris,

1811), p. 366; Cauchy, Exercices des Math. (Paris, 1826), pp.

5456; and Schlomilch,

Journal fur Math. xxxm. (1846), pp. 268280. The integral in which the limits of integration are arbitrary has been examined by Binet, Comptes Rendus, xn. (1841), pp. 958
962.

6'23.

Hardy's formulae for integrals of Du Bois Reymond's

type.

The

integrals

f"
I

a?

sin

sin

P"
t

f*
dt,

a?

cost. cos

f.

dt,

Ja

Jq

in

of

which x > 0, 1 < R(v) < 1, have been examined by Hardy f as examples Du Bois Reymond's integrals
f(t)*
/, o

m t.t~*dt,
By constructing a differential equation

cos

in

which f(t)

oscillates rapidly as

* 0.

of the fourth order,


functions
;

Hardy succeeded in expressing them in terms of Bessel but a simpler way of evaluating them is to make use of the results
*

of

^621,

6-22.

Math. Ann. xvm.

(1881), p. 182.

t Messenger, xt. (1911), pp. 44

51.

184
If
f"
I
.

THEORY OF BESSEL FUNCTIONS


we
t

[CHAP. VI

replace
.

by x&,
dt

it is

clear that
(a?e*)

2 a?

sin

sin

f'
.

= x" __

f
I

sin

JO

_
1

sin (xe~ l ) e vt dt
.

in

fgitzcosht

g 2&ccosht

girfzsinht

g2tassinhtJ

git

fa

./

00

= - \af [Trie-*!! (2a?) - Trie* H^\ (2x) - 2er** Kv (2a?) - 2^"'* #_ (2a:)],
and hence we have
(1)
/

sin

sin

JO

%
t
"K2

--

?\ A = 4sin|i/7r
.

(2a?) [/ L

- /_ (2a?) + /_ (2a?) - / (2a?)],

and similarly
(2)

lob
f

cos

r-

eft

"ITT"
,
,

(X (2a?) -

Jv (2a?) + /_ (2a?) - / (2a?)].


become
(2a?),

When
(3)

v has the special value zero, these formulae

sin*sin^.^ = 7ry
Jo

it
t

(2a?)

+ if

(4)

r Cos<cos^.^=-iTrF
Jo
t

(2a?)

Jfir

(2a?).

6*24.

Theisinger''s .extension of BesseFs integral.

curious extension of Jacobi's formulae of 2*2 has been obtained in the case of J {x) and Jx (x) by Theisinger, Monatshefte fur Math, und Phys. xxiv. (1913), pp. 337 341 we shall now give a generalisation of Theisinger's formula which is valid for functions of order

where
If

-\<v<\.
is

any positive number*,

it is

obvious from Poisson's integral that


(
" >-aii. C os (x cos 0) sin2 " 6dd
**

j9

<

X) =

flf)"

+
Now
2

(y

(l

- e-enn#) cos (# cos 0) sin 2 " 0tf0.

(l-e _ !,in *)cos(;FCos0)sin8 "0cW


ir

_ g oxsinfl

-/.o sinh (x sin ff) sinh (a? sin


'

- ix cos 0)
sml1

sin 2 " 0cW

ft
'

1-expJicm^-l/i)} -i~~5nh{i (*-V)}

(*jzl!*f* "^
2f

*'

where the contour passes above the origin. Take the contour to be the real axis with an indentation at the origin, and write z= tan < on the two parts of the contour; we thus
find that the last expression is equal to

exp(-cmcot<6) y ^
Jo
sin(.rcot<)
,

,.

sin (x tan id>) x "^

,_..

ml cot2" <h "*


.,
,

+ .[**
J
o

- eXl (a
sin
;

(.r

cot
j.

^
<)

-^-r ^sin<p

d<b

</>)

sin

<*tan v
i.

W ^ .e-cot<j>\ ^sin0
2
.

=4
,

r**
I

sm

cot (iax cot <6) vs f/ cos (Max Vi

/i

j\

/i

f vir)
<f>

sin
.

2 \ r-^V cot " <f -r-1-^ , ^sin#' sin(#cot</>)

(.r

tan

i<A)

rf<

* In Theisinger's analysis, a is

an even

integer.

6'24-6'3l]
and therefore

INTEGRAL REPRESENTATIONS

185

r(y+
(1)

^ J
2 +
/

A.) [

**

e -.in C08

(-p

cos 6) sin*"

0d0

/**"

Jo

cot sin (iax 2

/i

d>)

j\

.sin (a? tan Ad>) ,,, /i AJ cot8 " d> cos (4<m; cot <b - vn) r-^-.

vi

sm (j;cotd))

^~

^sind)

^-^

dd>

The transformation fails when v >, because the integral round the indentation does not tend to zero with the radius of the indentation. The form given by Theisinger iu the case v = \ differs from (1) because he works with 3-3 (7) which gives
r(-j)r(j)
(2)

Z-r a

{*)

( ** e-"infl sin (x cos

2 6) sin

""2

6 cos 6d6
,,_.,,
'

+4
provided that \

f**
/

,1 j x jx /i sm (lax cot <6) cos (*a# cot <A - vir)


i.
.

sin 2 (Aar tan Ad>)


r-f
,

2 ?Z cot "

d>

-r-~

cW>

<v<
the integral representations

6'3.

2%e equivalence of

of

K
v

(z).

obtained in

Three different types of integrals which represent 6'15 (4), 622 (5) and 616 (1), namely

(z)

have now been

'

(,,

'i|frr
=
Jo
f

e~*

{t *~

iy ' h dt

e zcoaht

cosh vt.dt,
.

_ !> + ). (25)" f cos as* cfo A 'W" (u> + /+*


,
x

^T(J)

The

equality of the

first

and second was

directly demonstrated in 1871

by

Schlafli*; but Poisson proved the equivalence of the second and third as early as 1813, while Malmsten gave a less direct proof of the equivalence of the

second and third in 1841.


in question.

We

proceed to describe the three transformations

6*31.

Scklq/li's transformation.

We
pp.

first give an abstract of the analysis used by 199201, to prove the relation

Schlafli,

Ann. di Mat.

(2) v. (1873),

r
which
arises

t e-'^'coahvddB ffl*y I" tr*(fi-l)"-idtK r ' Jo ("+i) J


i

from a comparison of two of the integral representations of


to suppose that

v (2),

and which
convenient

may
at

be established by analysis resembling that of 2*323.

We have, of course,
first

R(z)>0 to secure

convergence, and

it is

to taket

-<R(v)<l.
(1837), pp.

* An earlier proof is due to Kummer, Journal filr Math. xvn. much more elaborate than Sclilafli's investigation.

228

242, but

it is

The
W.

result is established for larger values of

R {)

either

by the theory of analytic continua7

tion or by the use of recurrence formulae.


B. P.

186

THEORY OF BESSEL FUNCTIONS


<S

[CHAP. VI

Now define

by the equation
Ji
'

(*-0 r

'

where x >

and then,

if

t=x - (x -

1) ,

we have

on expanding the
Replacing

last factor of the

integrand in powers of u and integrating term-by- term.

x by

cosh

6,

we

see that

Ji
so that, by a partial integration,

(coah 6

-ty

v r($)

'

2"

-* ooBh^cM-?!^^ ("c-'^flsinhdsinh^dd w"il^ f" r (l) jo r(J)


J

= 7^

, r(l-i') Jo

tf-'w'MtfsuihdcM

dtdx

-w=i)l'!'"^r*r(i-r)Ji

J/

'

1}

*(*-<)*
1}

r(i-0 J, J.*

('

the inversion of the order of the integrations presents no great theoretical difficulty, and the transformation is established.

6*32.

Poisson's transformation.

The

direct proof that

-~^-

ft_

I>+*)-(2*)"

r cos(*M

)rftt

due to Poisson* Journal de VEcole Polytechnique, ix. (1813), pp. 239241. The equation true when |argz|<r, x>0 and R(v)> -, but it is convenient to assume in the course of the proof that R (v) > | and arg z < w, and to derive the result for other values of z and v by an appeal to recurrence formulae and the theory of analytic continuation.
is is
| |

If

we

replace

by ajnew variable defined by the equation v=a**'~ w, we see that

it is

sufficient to

prove that

,.

f coa(xu)du

4r(A)

,,

,,

See also Paoli, Mem. di Mat.

e di Fit. della Soe.

Italiana delle Sci. xx. (1828), p. 172.

6-32, 6-33]

INTEGRAL REPRESENTATIONS
is

187

Now
r

the expression on the left

equal to
s"-i exp

Jo Jo

s 2

<v,..n )* + i +*2
.

dtdu=

{-sfvP+z2)} cos xit.dsdu


*"- J

Jo Jo
/OO
o
/"OO
I

{exp (

- 2) cos #m dv)
.

exp ( ss2 )

ds,

Jo
of the integrations*.

when we

write

ts (uP+z*) and change the order


/

Now
and so we have

exp( **) cos #u.efa=r()*-i exp ( #*/*),

y o

r(v + ^ )

/o"^Spi ==ir(*
result.

'"

IeXp{

/o"'

-^- i

<i8

v(2z])"J

which establishes the


[Note.

It is evident that

8=\xe~ tlz=\v1 lvlz.


i

The only reason

for modifying

r
is

by taking v as a parametric variable

to obtain an integral which is ostensibly of the


;

same form as the integral actually investigated by Poisson


exp
(

with his notation the integral

is

jf - a2 .*~ B

dx.]

/;

6*33.

Malmsten's transformation.
in proving that,

The method employed by Malmstent


then

when R(z)>0 and

R {?)>

-i,

m coB(xu)du r(i)(^)- m r(+*) - v*, * _ [ [ c {l t (**)T(J);. (*+^)" + i~ ro+*) Ji


is

1Y X)

- idt

ah

not so direct as the analysis of 6*31, 6*32, inasmuch as it involves an appeal to the theory of linear differential equations. It is first shewn by Malmsten that the three
expressions

qua functions of x, are annihilated J by the operator

*d^-V v - l) dx- XZ
and that as x-*~
that

2 <*

^ d

9
'

+ oo

the third

is

(**)

while the

R (v) > 0.

It follows that the second

first and second are bounded, provided and third expressions form a fundamental system

of solutions of the equation

S- <*->-*'-*
t

Bromwich, Theory of Infinite Series, 177. Svenska V. Akad. Handl. lxii. (1841), pp. 6574. X The reader should have no difficulty in supplying a proof of
A'.

* Cf.

this.

188

THEORY OF BBSSBL FUNCTIONS

[CHAP. VI

and the first is consequently a linear combination of the second and third. In view of the unboundedness of the third as a?-^+oo, it is obvious that the first must be a constant
multiple of the second so that
(cos xu)

du
C,

/;
where

V is

independent of x.

To determine

make x-*-0 and then

so that

r()r(i)

^cryv)
Z**
'

2z*T(v+$)
follows,

and the required transformation


for the

when R(v)>0,

if

we use the

duplication formula c a ?

Gamma function.
consequence of Malmsten's transformation
;

An immediate

gtt
.'

'

is

that

7o
expressible in finite terms
for it is equal to

(*+82 r

is

n-i

(2^)(2n-m-

1)
*

(2s)"-*(-l)! m=0

m.'(ra-ro-l)!

This method of evaluating the integral is simpler than a method given by Catalan, Journal de Math. v. (1840), pp. 110 114; and his investigation is not rigorous in all its stages. The transformation is discussed by Serret, Journal de Math. vnr. (1843), pp. 20,

21;

ix. (1844),

pp.
i.

193216;

see also Cayley, Journal de Math. xn. (1847), p. 236

[Collected Papers,

(1889), p. 313.]

6*4 Airy'8

integral.

The

integral

cos

(<*

Jo o

act)

dt

which appeared in the researches of Airy* "On the Intensity of Light in the neighbourhood of a Caustic" is a member of a class of integrals which are

The integral was tabulated by Airy by quadratures, but the process was excessively laborious. Later, De Morgan f obtained a series in ascending powers of ar by a process which needs justification either by Stokes' transformation (which will be explained immediately) or by
expressible in terms of Bessel functions.

the use of Hardy's theory of generalised integrals^.


*

Tram. Camb.

Phil. Soc. vi. (1838), pp. 379

402.

Airy used the form

cos |t (tc* - mw) du>,

/:
but this
is easily reduced to the integral- given above. t The result was communicated to Airy on March 11, 1818
;

see Trans.

Camb. Phil. Soc. vni.

(1849), pp.

595599.
;

X Quarterly Journal, xxxv. (1904), pp. 2266

Trans. Camb. Phil. Soe. xxi. (1912), pp. 1

48.

6'4]

INTEGRAL REPRESENTATIONS
satisfies

189

Stokes observed* that the integral

the differential equation

as!

**-<>,

and he

also obtained the asymptotic expansions of the integral for large


x,

values of

both positive and negative.


(foe. cit. p. 187),

As was

observed by Stokes
;

this differential equation can be reduced

to Bessel's equation

cf.

4*3 (5) with 2g

= 3. The expression
first in

Bessel functions of orderst


Berichte dea natur.-med.

was published

of Airy's integral in terms of a little-known paper by Wirtinger,


(1897), pp. 7

Vereins in dnruibruck,

xxm.

15,

and

later

by

Nicholson, Phil. Mag. (6) xvn. (1909), pp.

617.

Subsequently Hardy, Quarterly Journal, xli. (1910), pp. 226 240, pointed out the connexion between Airy's integral and the integrals discussed in 6*21, 6*22, and he then

examined various generalisations of Airy's integral

(.10*2

10*22).

To evaluate

Airy's integral J,

we observe
exp (***

that

it

may be written in the form

ixt) dt.

\r.

Now
The

consider this integrand taken along two arcs of a circle of radius p with
p,

centre at the origin, the arcs terminating at

pe** and pe**1 pe*% respectively.


,

integrals along these arcs tend to zero as p

-*-

oo

by Jordan's lemma, and

hence,

by Cauchy's theorem, we obtain Stokes' transformation


cos
(t*

(<*>

xt) dt

co

expert

Jd

exp (?

ixt) dt

Zjooexpfiri
Ti

=\
~

I* {^
Jo

exp (- t **'T) + -** i exp(- t 3 e-^xr)} dr;

the contour of the second integral consists of two rays emerging from the
origin
rays.

and the third integral

is

obtained by writing re*", re**1 for

on these

Now,

since the resulting series are convergent, it

may be shewn

that

exp(-T +
/,
* Trans.

^a!T)dT=

2
w=0

; -

T*exp(-T)<*T,
.'o

Camb. Phil. Soc.

ix. (1856), pp.

166187.

[Math, and Phys. Papers, h. (1883),

pp. 329349.] See also Stokes' letter of May 12, 1848, to Airy, Sir G. O. Stokes, 160. Scientific Correspondence, n. (Cambridge, 1907), pp. 159

Memoir and

+ For other occurrences of these functions, see Rayleigh, Phil. Mag. (6) xxvm. (1914), 609619; xxx. (1915), pp. 329338 [Scientific Papers, vi. (1920), pp. 266275; 841349] on stability of motion of a viscous fluid; also Weyl, Math. Ann. Lxvm. (1910), p. 267, and, for
pp.

approximate formulae, 8*43 infra.


X The integral

is

convergent. Cf. Hardy,

toe. cit. p.

228, or de la Vallee Poussin, Ann. de la

Soc. Sci. de Bruxelles, xvi. (1892), pp.

150180.

Bromwiob, Theory of Infinite

Series, 176.

190

THEORY OF BESSEL FUNCTIONS


= 2

[CHAP. VI

and so
|

cos

(<*

Jo

+ ~

art) eft

'?

m =o

ml

Jo

(- t*) dr rm exp r\
/

= i 5 (x)sin(m + l)7r.r(m + )/m!


o m =o

"*
This
is

U-omir(m + |)
De
Morgan.

+ **

mt.iti!r(m + |)J
the series on the right

the result obtained by

When

are expressed in terms of Bessel functions,


a;

we

obtain the formulae (in which

is

to be taken to be positive)

due to Wirtinger and Nicholson


.

(i)

/;oos ( , - **> = j, v(w

I/-,

d$) + a $%)]

(2)

.(,+*)*- w<*.>. [/., (f$H(|$)]


V# v
(1x\lx\

6*5.

Barnes' integral representations of Bessel functions.


Mellinf, Barnes J

By using integrals of a type introduced by Pincherle* and


proof of

has obtained representations of Bessel functions which render possible an easy

Kummer's formula

of 442.

Let us consider the residue of

- T (2m - s)
at s

{izy
is

- 2m + r, where
is

= 0,

1, 2,
.

. . . .

This residue

(iar)

lm+r/r

!,

so the

sum

of the residues

( ) m zim e~ iz

Hence, by Cauchy's theorem,

J ^ Z)e
if

27ri\ntoJoo

2.m\r(p + m +
1, 2,

aS
l)

'

the contour encloses the points 0,

It

may be

verified,

by using

Stirling's formula that the integrals are convergent.

Now suppose that R (v) > \, and choose the R (y + s) > $. When this last condition is satisfied r ( 2 m - g> 5 .m!r(i/ + w + l) =o2 m
2r'l

contour so that, on
the series

it,

is

convergent and equal to

r <-*)

wt l.l

l..

r(-)l> + j + t)
559562
;

* Rend, del R. Istituto Lombardo, (2) xix. (1886), pp.

Atti delta R. Accad. dei

Lincei, ser. 4, Rendieonti, iv. (1888), pp.

694700, 792799.

t Mellin has given a summary of his researches, Math. Ann. lxviii. (1910), pp. 305 337. X Gamb. Phil. Tram. xx. (1908), pp. 270 279. For a bibliography of researches on inteRraU of this type, see Barnes, Proc. London Math. Soe. (2) v. (1907), pp. 59 65.

'

6*5]

INTEGRAL REPRESENTATIONS
If therefore

191

by the well-known formula due to Gauss. of summation and integration* we have


t (,\ *-*

we change the

order

-_<*

l0+)

r(-)r( y + +

|).(irr<to

The only poles of the integrand inside the contour are we calculate the sum of the residues at these poles, we

at 0,

1, 2,

When

find that

so that
(1)

^
is

( ^ )e

- =

_(ML- F
i

(I

i; -2iz),

which
(2)

Rummer's

relation.
{

In

like

manner, we find that


*-,

M*) *= r *+ x yK ( +

2"

l;

2iz).

These formulae, proved when B(v) > , are relations connecting functions of v which are analytic for all values of v, and so, by the theory of analytic continuation, they are universally true.
It is also possible to represent Bessel functions

by

integrals in which no

exponential factor

is

involved.

To do

this,

we

consider the function

r(-*->r (-)(!)'+
qua function of s. It has poles at the points
5

= 0,
is

1, 2, ...;

- v ,-v+l,-v + 2,
(~)m (^Y +Sm

....

The

residue at s

iri

sin vnr

'

ml r

(v

+m+

1)

while the residue at s

= v + in
iri-"

is

(-)w (|s)- y+OTt


'

sin V7r

ml T

(v

+m+
v

'

1)

so that
(3)
7re-4<" +1 >'rf

H (z) = - 2^.
v

T (-

- s) T (- s) (^iz)"^ ds,

and, in like manner,


(4)

irew H

w ( z)

-~lr(-v-s)r (- *) (- \izy*" ds,


and return
to

where the contours


in (4) the contours

start from

+ oo
\

after encircling the poles of


\ir in (3) or arg ( iz)
|
\

the integrand counter-clockwise.

When

arg iz

<

<

\tt

may

be opened out, so as to start from

oo i

and end at

oo

i.

If

we

reverse the directions of the contours


f" H (z) = -U I'm
J

we

find that

c+ l

(5)

7re-*<" +1 "*

r ( _ _ ,) r (- *) (ii>)' +M ds,
Infinite Series, 176.

-c-aoi

* Cf

Bromwich, Theory of

; ; ,

192
provided that argiz
|

THEORY OF BESSEL FUNCTIONS


I

[CHAP. VI

< \ir and


;

(6)

vJW H
|

o)

( z)

= -L r
frm]

e+x>%

r (_ _ S ) r (- s) (- \%zy*~ ds,
is

-c-oot

provided that arg ( iz) \<\nr; and, in each integral, c

any positive number

exceeding jR

(v)

and the path of integration

is parallel

to the imaginary axis.

There is an integral resembling these which represents the function of the and the argument first kind of order v, but it converges only when It (v) >
of the function
is positive.

The

integral in question is

'.-K,$g&*>
and
it is

obtained in the same

way

as the preceding integrals

the reader

will notice that,

when

is

large

on the contour, the integrand

is

0(\s |~* -1 )-

6*51.

Barnes' representations offimctions of the third kind.

By

using the duplication formula for the

Gamma

function

we may

write

the results just obtained in the form

m UJ
Consider

(z) ** j >rW 2ty*rJ.

r( -M+j).(2uy* ^- r r^ + ^r^ + s+^sin**-*


y

now

the integral
1

_ *l.
Ziyir
in

I"
J _aot

r(-8)r(-2p-8)r(p + 8 + i).(2igyd8,

periodic in

which the integrand differs from the integrand in (1) by a factor which is It is to be supposed temporarily that 2v is not an integer and s. that the path of integration is so drawn that the sequences of poles 0, 1, 2, ... 2v, 1 2v, 2 2i/, ... lie on the right of the contour while the sequence of poles v \, v~\, v \, ... lies on the left of the contour. In the first place, we shall shew that, if arg iz < f ir, the integral taken round a semi| \

circle of radius
for, if s

= pe**,
v

p on the right of the imaginary axis tends to zero as p * we have


v

<x>

'

'

r (s) T (2v + * -f l)sin*7rsin(2i' + *)7r

and,

by
1

Stirling's formula,

T(v + s+i).(2izy g r(s + l)JT(2i/ + s+l)

~ pe

i9

log (2iz)

- {v + pe") (log p + id) + pe ie -

^ log (2tt)

and the real part of this tends to co when ^7r<6<\ir, because the dominant term is p cos 6 log p. When is nearly equal to |7r, sin sir is comparable with \ exp [pir sin 6 and the dominant term in the real part of the logarithm
|

j)

of s times the integrand

is
.

p cos 6 log 2z
|

and

this tends

p sin $ arg 2iz pcosd log p + pd sin 6 + p cos 6 - 2p7r|sin &\ arg iz < \w. to - <x as p -* oo
if*
j j

6*51]

INTEGRAL REPRESENTATIONS
s times the integrand tends to zero all along the semicircle,
if

193

Hence
to pass

the integral round the semicircle tends to zero

the semicircle

is

and so drawn so as

between (and not through) the poles of the integrand.

It follows from Cauchy's theorem that,


integer, then

when arg iz < \tr and 2v


j J

is

not an

- ^rJLi^/tr j
may be
contour.

Pr
-oof

(-

r (-

2*

- ) r (* + + j)

^y

ds

calculated by evaluating the residues at the poles on the right of the

The

residues of

r(-s)r(-2v-s)r(v+8 + i).(2izy
at s

and

= 2v + m

are respectively

7T

ein2inr

r(v+m + $).(2iz)m m\T{2v + m + l)

_
'

tt_ r (sin2*"7r

v + m + \) (2iz)-+m mlV(2p + m + l)
.

and hence

_(2z

p r(-s)r(-2v-s)T(v +

+ l).(2iz)ds
+

= ( 2,M r(, + i) sin 2vir I (2j/ + 1)

sin 2v7r

'

It follows that,
(2)

when

j.argiz

< f ir,

B.*to-

rn~' mi r>-Mr
rrrt

x ["* r(-)r(-2v-)r(i +
J -oe

f).(2w)<fo

>

and

similarly,

when arg ( iz) < f tt,


| |

(3)

*,(*)--

i*->n)

cog

(,.)

^2^)"

7T*

*r

r(-)r(-2v-*)r(i/+s + ^).(-2wyrf*.

The restriction that v is not to be an integer may be removed in the usual manner by a limiting process, but the restriction that 2v must not be an odd integer cannot be removed, since then poles which must be on the right of the contour would have to coincide with poles which must be on the left.

CHAPTER
Approximate formulae for
ill

VII

ASYMPTOTIC EXPANSIONS OF BESSEL FUNCTIONS


7"1.

Jv {z).

In Chapter
cases

various representations of Bessel functions were obtained

in the form of series of ascending powers of the

z1

is

argument z, multiplied in some by log z. These series are well adapted for numerical computation when since the series not large compared with 4 (v + 1), 4 (v + 2), 4 (v + 3),
.

converge

fairly rapidly for

such values of

z.

But,
initial

when

is large,

the series

converge slowly, and an inspection of their

terms affords no clue to the

approximate values of Jv (z) and Yv (z). There is one exception to this statement when v + \ is an integer which is not large, the expressions for J v (z) in finite terms ( 3*4) enable the functions to be calculated without difficulty.
;

The

object of this chapter

is

the determination of formulae which render

possible the calculation of the values of a fundamental system of solutions of


Bessel's equation

when

is large.

There are
gation
large.

really

two aspects of the problem to be considered

the investi-

when v is large is very different from the investigation when v is not The former investigation is, in every respect, of a more recondite
it is

character than the latter, and


It must, however,

postponed until Chapter


first

vm.
before Poisson 's -J"

be mentioned that the

step towards the solution of

the more recondite problem was


published.

made by Carlini* some years

investigation of the behaviour of

(x),

for large positive values of x,

was

The formal expansion obtained by Poisson was


/o( *)
r
z N

/ 2

\T
COB
[

U)

<

12

.3 2

"^>-t -2l(8^

+
(

l2

.3 8 .5 2 .7 a
4!(8*)'

-"j
3 2 52

+ sin(*-i7r).j
when x
is

l2

IT

^-3^
l2

..

large and positive.

But, since the series on the right are not con-

vergent, and since Poisson gave no investigation of the remainders in the


series, his analysis (apart

to be regarded as suggestive

from his method of obtaining the dominant term) and ingenious rather than convincing.

is

* Iticerche sulla convergenza delta serie che serva alia soluzione del problema di Keplero (Milan,

1817).

An

account of these investigations has already been given in

1*4.

t Journal de VEcole Polytechnique, xn. (cahier 19), (1823), pp. 350 352 ; see l 6. An investigation of Jv (x) similar to Poisson's investigation of J (x) has been constructed by Gray and
-

Mathews, A Treatise on Bessel Functions (London, 1895), pp. 34

38.

7-1]
It will

ASYMPTOTIC EXPANSIONS
-

195
;

be seen in the course of this chapter that Poisson's series are asymptotic has been proved by Lipschitz, Hankel, Schlafli, Weber, Stieltjes and Barnes.

this

It must be mentioned that Poisson merely indicated the law of formation of successive terms of the series without giving an explicit expression for the general term such an
;

expression was actually obtained by

W.

R. Hamilton*
for

(cf.

1*6).

The analogous formal expansion


written in the form

(x) is

due to Hansenf and a few


;

years later, JacobiJ obtained the more general formula which

is

now

usually

Jn (x)~(
(

COS(#

|W7T \tr) -3)


2

(4n2 - I s) (4rt2

f
.

2UMf
,
,
,

~ +(

4n2 -

(4n*

- 3 ) (4>n* - 5
2

(4n 2

7 2)

_
Y\
.

4!(8*)<

-sin^-^Tr-i^.j-yj^

(42 -l 2

(4n 2

-l 2)(4n2 -3

)(4n2 -5 2)

gy^^

L +

...JJ

These expansions for J (x) and Jx (x) were used by Hansen for purposes of numerical computation, and a comparison of the results so obtained for isolated values of x with the results obtained from the ascending series led Hansen to infer that the expansions, although not convergent, could safely be used for purposes of computation .

Two years before the publication of Jacobi's expansion, Plana had discovered a method of transforming Parse val's integral which placed the expansion of / (x) on a much more satisfactory basis!". His work was followed by the
||

researches of Lipschitz**,

who gave the


(z)

first

rigorous investigation of the

asymptotic expansion of Jo
tion
;

with the aid of the theory of contour integraLipschitz also briefly indicated how his results could be applied to Jn (z).
for

The general formulae


plex) value

Jv (z)

and

v (z),

and z is Hankelft, written in 1868.

large and complex, were obtained in the great

where v has any assigned (commemoir by

* Some information concerning W. R. Hamilton's researches will be found in Sir George Gabriel Stokes, Memoir and Scientific Correspondence, i. (Cambridge, 1907), pp. 130135. + Ermittelung der absoluten Storungen [Schriften der Stermoarte Seeburg], (Gotha, 1843),

pp. 119123.

is

t Astr. Nach. xxvm. (1849), col. 94. [Get. Math. Werke, vn. (1891), p. 174.] obtained by making the substitutions

Jacobi's result

n+1>cosx+(- l^-^sinx, J2 cos(*- iir-iir) = - l)*<


.

^/a.8ln(*-iii-J)=:(-l)*" (
in the form quoted.

" +1) Bin*-(-l)*" (, '- 1, cos*,

a note by Niemoller, Zeitsehrift fiir Math, und Phys. xxv. (1880), pp. 4448. della R. Accad. delle Sci. di Torino, (2) x. (1849), pp. 275292. expansions of Jv () and IV (z) by If Analysis of Plana's type was used to obtain the asymptotic JVlcMalion, Annals of Math. vin. (1894), pp. 5761. ** Journal fur Math. lvi. (1859), pp. 189196.
See
II

Mem.

+t Math. Ann.

I.

(1869), pp.

467501.

196

THEORY OF BESSEL FUNCTIONS


for

[CHAP.

VH

The general character of the formula


iiber die Bessel'schen

Yn (z)

had been indicated by Lommel, Studien

memoir

Functionen (Leipzig, 1868), just before the publication of Hankel's and the researches of Weber, Math. Ann. vi. (1873), pp. 146 149 must also be

mentioned.

was investigated (and proved to be this result was reproduced, with the addition of the corresponding formula for / (z), by Kirchhoff f and a littleknown paper by MalmstenJ also contains an investigation of the asymptotic
of

The asymptotic expansion

(z)

asymptotic) at an early date by

Kummer*

expansion of

(z).

close study of the remainders in the asymptotic expansions of

(x),

(x),

(x)

and Kq{x) has been made by Stieltjes, Ann. Sci. de VEcole norm. sup. (3) in. (1886), pp. 233 252, and parts of his analysis have been extended by Callandreau, Bull, des Sci. Math. (2) xiv. (1890), pp. 110 114, to include functions of any integral order; while results concerning the remainders when the variables are complex have been obtained by Weber, Math. Ann. xxxvu. (1890), pp. 404416.

The expansions have


1906, pp. 239

also been investigated

by Adamoft, Petersburg Ann.

Inst, polyt.

265, and by Valewink|| in a Haarlem dissertation, 1905.


is fixed,

/ (z) and Yv (z), when seem to be most simply carried out with the aid of integrals of Poisson's type. But Schlafli1[ has shewn that a large number of results are obtainable by a peculiar treatment of integrals of Bessel's type, while, more recently, Barnes** has discussed the asymptotic expansions by means of the Pincherle-Mellin integrals, involving gamma-functions, which were examined in 65, 6*51.

Investigations concerning asymptotic expansions of


large while v

is

7'2.

Asymptotic expansions of
shall

H^ (z) and i/
|

(2)

(z) after

Hankel.

We

now

obtain the asymptotic expansions of the functions of the

third kind, valid for large values of


modifications, will follow that given

;
|

the analysis, apart from some slight

by Hankel ff.

Take the formula

6*12 (3),

namely

valid

when -\ir<fi<\ir and


of the factor (1

|7r + j8<

args <

|7r

+ jS,

provided that

The expansion

+ $iu/z) v -l in descending

powers of

z is

(v-j)iu
"*"

2z

+ (,,-fr)(-f).(^ +

270

*"'

* Journal fUr Math. xvn. (1837), pp. 228242. f Ibid, xlviii. (1854), pp. 348376. t K. Svenska V. Akad. Handl. lxh. (1841), pp. 6574. See the Jahrbuch Uber die Fortschritte der Math. 1907, p. 492.
Ibid. 1905, p. 828. IT Ann. di Mat. (2) ** Trans. Camb. Phil. Soc. xx. (1908), pp. 270279.
II

vi. (1875),

pp.

120.

tt Math. Ann.

i.

(1869), pp.

491495.

7-2]

ASYMPTOTIC EXPANSIONS

197

but since this expansion is not convergent all along the path of integration, we shall replace it by a finite number of terms plus a remainder.

For
1

all

positive integral values of p,

we have*
{1
J

+ 2z)

-J -^TWz)
=0

+ (p-l)\{2iz)

tf
I
;

dt
2iz)

It is convenient to take

so large that
satisfies

R {v p \) ^

and we then choose

any

positive angle 8

which

the inequalities

||^|7T-S,
The
effect of this choice is that,

|arg*_(|7r + )| $ 7r-$.

when

8 is given, z is restricted so that

7r + 28 ^ arg z % 2?r 28.


When
the choice has been made, then
ut
1

>sin8,

|arg(l-^)

<7T,

2iz
for the values of
t

and u under consideration, and so P~i < e'l 'wi (sin 8)*<"-p-*> (i _

~Y~

= Ap

say,

where

A p is independent

of

z.

On
term,

substituting its expansion for (1


find that
1

+ liu/*)*"*

and integrating term-by-

we

**-**> Pi - (AY e H *'w (z) w-UJ


where

(*-*>- r ( ,> + , +*>

U-o ir(v + i).(Kf)-

1L1 + jKp
J'

where

2?p is

a function of v,

p and

8 which

is

independent of z.

Hence,
(1) v 7

when R {v - p - f) <
v '

and

R (v + J) > 0, we have
(#->)]
'J
,

**- 1" Pi' ^""hi^^ H (*) - (--Y + m!(2t)m \irzj m=0


[

when
angle
;

is

such that

it + 28 ^ arg z ^ 2tt 28,


is

8 being any positive acute

and the symbol

function

Bachmann-Landau symbol which denotes a of the order of magnitude^ of z~* as z -* oo


the
| \

The formula

(1) is also

valid

when

R (v - p |) >

this

may be

seen by

* Cf. Modem Analysis, 5-41. The use of this form of the binomial expansion seems to be due to Graf and Gabler, Einleitung in die Theorie der BesseFschen Funktionen, i. (Bern, 1898) pp. 8687. Cf. Whittaker, Modem Analysis (Cambridge, 1902), 161; Gibson, Proe. Edinburgh Math. Soc. xxxvm. (1920), pp. 6 9 ; and MaoBobert, ibid. pp. 10 19.

t Cf.

Modem Analysis,

2-1.

198

THEOBY OF BESSEL FUNCTIONS

[CHAP. VII

supposing that

R {v p J) >
if

and then taking an integer q so large that


is

R(? q \) <0;

the expression which

contained in

[ ]

in (1)

is

then

rewritten with q in place of p throughout, followed by q p + 1 terms each of which is


of these q

it

may be

expressed as

terms

(z~ p ) or o (z~ p ) ; and the

sum

p + 1

terms

is

therefore

(z~ p).
i

In a similar manner (by changing the sign of


work) we can deduce from
(2)
6*12 (4) that

throughout the previous

Hi* (.)-)'.----

g;<i^|^ +
tr

(*-,)]
is

provided that

R (v + |) >
we

and that the domain of values of z


2tt

now given

by the inequalities

+ 28 % arg z ^

25.

If,

following Hankel,

write

{4>v*

I s } {4* 2

-3

2
}
.

..

{tv*

- (2m -

l)2 }

2 2m .w!

these expansions become

(4)

H,

W = (A)*e-<-^g |^ +
o

<)(*-*)]

For brevity we write these equations thus


(5)

HP (z) ~ f A) V^-w

m>
t

-=-.<f'

(6)

fc^}. jJ(,)^,(A y e--<S-i.or-i*> m y ' ' WzJ . (2iz)m


(7),
(j>, m) is an even function of v, it follows from the formulae of which connect functions of the third kind of order v with the corre-

Since

361

sponding functions of order v, that the restriction that the real part of v exceeds - \ is unnecessary. So the formulae (1) (6) are valid for all values
of
v,

when z

is

confined to one or other of two sectors of angle just less than

Sir.

In the notation of generalised hypergeometric functions, the expansions are


(7)

HP

(z)

~ (A)* ^*~-w

.f, (i

+ ,

_^
.

(8)

HP (z) ~ (A)* e-uz-ir-iu)


is

(^ + v>i
(8)

_ v _ _1_
2-ir

y
arg z

of which (7)

valid

when

tt < arg z <

2-tt,

and

when

<

< ir.

7*21]
7*21.

ASYMPTOTIC EXPANSIONS
Asymptotic expansions of Jv {z), J- V (z) and
the formulae of 7
2,

199

v {z).

If

we combine

we deduce from the formulae of 3*61


first

(which express Bessel functions of the


functions of the third kind) that
/-n (1)

and second kinds


(-)w -(">2m)

in terms of

/ i Tt\ / VT i s [coe^-W-iir).^ *(*)-[-)


^

{2gym
(-)w -(^,2m

/ i \ v i -sin^-Ji/w-iir).^

+ iy

J2zj^i
.(f,

(2)

n W ~(iy

i sin(z \ vtt

i x v $ir). 2.

(-)w

2w)

m-0
i

(2zfm
(-)'*('.

/i

2w+l)"

'

- Sin (* + |7T - iTT)


( 4)

^.(ij^^w-wyt^-)
n
only),

" (-)m .(^,


(

2m +

1)"
'

^y w+1

and

(in the case of functions of integral order

(5)

T.(.)~(

T)

Ln(*-W-i).jS o
+ cos(^-^7r-|7r).^
o

(2 , )m
v (

^ )2m+1
|

J.
|

These formulae are

all valid for

large values of z provided that arg z


| |

<

tt;

and the error due to stopping at any term is obviously of the order of magnitude of that term multiplied by \\z. Actually, however, this factor \\z may be replaced by ljz2 this may be seen by taking the expansions of HJ$ (z) and {z) to two terms further than the last term required in the particular v combination with which we have to deal.
;

(2)

As has been seen

in 7'2, the integrals


{l)

which are dealt with when

R (v) > - \ represent H

(z)

and

(z),

but,

when R(p)% ^

the integrals

from which the asymptotic expansions are derived are those which represent {2) .ff (,) _ (z) and -y (z). This difference in the mode of treatment of J, (z)

and

{z) for
is

such values of v seems to have led some writers to think* that


not valid unless jR
(v)

formula (1)
* Cf.

> \.
;

Sheppard, Quarterly Journal, xxm. (1889), p. 223 Searle, Quarterly Journal, xxxix. The error appears to have originated from Todhunter, An Elementary Treatise on Laplace's Functions, Lamfs Functions and Bessel's Functions (London, 1875), pp. 312 313.
(1908), p. 60.

1 l 1

200

THEORY OF BESSEL FUNCTIONS


The asymptotic expansion
(1
i.

[CHAP. VII

grating

e***

and
f
J

+ oc

of J (z) was obtained by Lipschitz* by interound a rectangle (indented at 1) -with corners at + 1 Cauchy's theorem gives at once

8 )-*
2

eizt (1

- < )-* dt + e**

-1

f V-">*
.'o

-* (2
rct>

+ tu)-* du
M -i (2

- l
o

e-(f+)z

- tu)-* du = 0,
but in order to
it

and the analysis then proceeds on the

lines already given;

obtain asymptotic expansions of a pair of solutions of Bessel's equation

seems necessary to use a method which involves at some stage the loop integrals discussed in Chapter vi.
It

may be
m v (-)

convenient to note explicitly the

initial

terms in the expansions

involved in equations
.

(1) (4); they


(4i/
2

are as follows
2

(v,

2m)

) (4i/

-3

m-o

(2*)*

2!(8*)

+
(-)
.

(4i/

(4y

- 32) (4y2 - 5) (4/a - 7 8 )


4!(8s)
4>

(y,

2m + 1)
2+ 1

4^ l!8s

(4g

(4i^

mt

(2^)

3!(8s)T

- 32) (4 - 5 8 ) +....

The reader should

notice that

a formula given by Lommel, Studien,


Note.

p. 67.

The method by which Lommel endeavoured to obtain the asymptotic expansion of Y n (z) in his Studien, pp. 93 97, was by differentiating the expansions of J v (z) with respect iX> v but of course it is now known that the term-by-term differentiation of an

asymptotic expansion with respect to a parameter raises various theoretical


It should

difficulties.

be noticed that Lommel's later work, Math. Ann. iv. (1871), p. 103, is free from the algebraical errors which occur in his earlier work. These errors have been enumerated by Julius, Archives Nderlandaises, xxviit. (1895), pp. 221 225. The asymptotic expansions of Jn (z) and Yn (z) have also been studied by McMahon, Ann. of Math. vni. (1894), pp. 57 61, and Kapteyn, Monatsheftefiir Math, und Phys. xiv. (1903), pp. 275282.

A novel

application of these asymptotic expansions has been discovered


:

in recent years

they are of some importance in the analytic theory of the

In such investigations the dominant terms of the exThis fact combined with the the of Bessel functions that theory forms only a trivial part of consideration has made it seem merely to mention in question desirable the investigations and more by Hardy . Wigert the recent papers the work of Voronoif and J
divisors of numbers.

pansions are adequate for the purpose in view.

* Journal far Math. lvi. (1859), pp. 189196. t Ann. Sci. de Vticale norm. sup. (3) xxi. (1904), pp. 207'86b. Math. Kongresses in Heidelberg, 1904, pp. 241 245.

459534

Verh. des Int.

% Acta Mathematica, xxxvn. (1914), pp. 113


Quarterly

140.

Journal, xlvi.

(1915), pp.

263283

Proc.

London Math.

Soc. (2) xv. (1916),

pp. 125.

w
7*22]
7*22.

ASYMPTOTIC EXPANSIONS
Stoked phenomenon.
for values of z

201

The formula 7-21 (1) for Jy (z) was established and arg z < 7r. If we took arg z to lie between between tr and tt) we should consequently have
| |

such that
lies,

2ir (so that

arg ze~ ni

Jv {z) = e vwi Jv (ze~


oo e v
/

1ci

V* T

-i
,-i

m v (-)
x

(*.

2m ) 2m +
)

5
m-o

(-)"*

(v,

1)"

\&ze

so that,

when

<

arg z

<

27r,
}

*w~ **<(-)
and

[cos <,

+ i^)j

v2 ;r
/.

-sin(* + i.wr + iw) 2


JH-0

2 , Vj^h V/^

this expansion
(1).

is

superficially quite different close

from the expansion of

721

We

shall

now make a

examination of this change.

The expansions

of 7*21 are derived from the formula

and throughout the sector


the asymptotic expansion

in

which

ir < arg z <

2rr,

the function

HP (z) has

W
(1)
is,

\irtj
for

TO -o

(2iz)m

The corresponding expansion

HP
i

{z),

namely

HM(z)co(--)
,irz)

e-i-l<>-i*

X m.
it.

("'

w)
'

(2iz)m

however, valid for the sector

27r

< arg z <

To obtain an expansion
3'62 (6),

valid for the sector

< arg z < 2tr we


.

use the formula of


VTd

namely

HP (z) = 2 cos vtt


and
this gives
( 2)

HP (zer*) + e HP (ze-),

iw>~)'.^^.sj$
VtTS/
TO _o

(2t^) TO
tt
]

difference between

The expansions (1) and (2) are both valid when < &rg z < them has the asymptotic expansion
2COBV7T.

now the

()'*<#*"+* 5
Kir* J

m-o

<7^J2>, m
(2t*)

and, on account of the factor e

iz

of lower order of magnitude (when

which multiplies the series, this expression -is z is large) than the error due to stopping
j

202
at

THEORY OF BESSEL FUNCTIONS


definite

[CHAP. VII
(e~ iz z-P-l)
(1)

any

term of the expansion (1);


tt,

for this error is

when
which

we

stop at the pth term.

Hence the discrepancy between


is

and

(2),

occurs

when 0< arg*<

only apparent, since the series in (1) has to be

used in conjunction with


Generally

its

remainder.

we have

where the constants c,, c2 have values which depend on the domain of values assigned to arg z. And, if arg z is continually increased (or decreased) while \z\'\% unaltered, the values of c x and c have to be changed abruptly at various 2 stages, the change in either constant being made when the function which multiplies it is negligible compared with the function multiplying the other constant. That is to say, changes in c, occur when I (z) is positive, while changes in c2 occur when I(z) is negative.
It is not difficult to prove that the values to be assigned to the constants
Ci

and
Cl

g2 are as follows
C3

= e*p(*+iW +1) +V"\ <h = i^


{v

= $a**"4) c, = i**+iM

[(2p

1)

7T

< arg z < (2p +

1)

tt],

[2P

7r

<

arg 2

< (2p + 2) tt],

where p

is

any

integer, positive or negative.

This phenomenon of the discontinuity of the constants was discovered by Stokes and was discussed by him in a series of papers. It is a phenomenon which is not confined to Bessel functions, and it is characteristic of integral
functions which possess asymptotic expansions of a simple type*.

The fact that the constants involved

in the asymptotic expansion of the analytic function

Jv (z) are discontinuous

was discovered by Stokes in (March?) 1857, and the discovery was apparently one of those which are made at three o'clock in the morning. See Sir George
Gabriel Stokes, Memoir and Scientific Correspondence, I. (Cambridge, 1907), p. 62. The papers in which Stokes published his discovery are the folio wing t: Trans. Cavib. Phil. Soc. x. (1864), pp. 106128; xi. (1871), pp. 412 425; Acta Math. xxvi. (1902), pp. 393 397. [Math, and Phys. Papers, iv. (1904), pp. 77 109 ; 283298 v. (1905), pp. 283-287.]

The

third of these seems to have been the last paper written

by Stokes.

7*23.

Asymptotic expansions of Iv
7*2 (5)

(z)

and

{z).

The formula

combined with equation

3*7 (8),

which connects

(z)

and if

(l)

(iz),

shews at once that

-(2z)
* Cf.

~nr +
( 6-4).

2i(8^

-J'
and
^

Bromwich, Theory of Infinite Series, 133. t Stokes illustrated the change witb the aid of Bessel functions whose orders are
the latter being those associated with Airy's integral

7*23, 7*24]

ASYMPTOTIC EXPANSIONS
\

203

when arg z <


|

tt.
,

And

the formula I (z)

= e* vni J
<

(e

-* z) shews that

<o\

(-) w ( y

m)

r'+(,,+ *"''

(",

"0

provided that

\ir < arg s <

f 7r.

On

the other hand, the formula / (z)


TO
(>,

er*"*
,

Jv (e* ni z)
|
TO

shews that
(y,

<*\

* V (-) ) t t \ ^~(2,r*)i m%> (2*) +


|7r < arg z <
between
\ir.

e-'-<"+*>"

m)

(2tt^

to(2^r'
z has a value for

provided that

The apparent discrepancy between


which arg z
lies

(2)
is,

and

(3)

when

\tt and

|7r

of course, an example of Stokes'

phenomenon which has just been


The formulae of this
(1837), pp.

investigated.

section were stated explicitly by Kummer, Journal far Math. xvn. 228242, and Kirchhoff, Journal fur Math, xlviii. (1854), pp. 348376, except that, in (2) and (3), the negligible second series is omitted. The object of the retention of the negligible series is to make (1) and (3) formally consistent with 37 (6).

p. 135,

The formulae are also given by Riemann, Ann. der Pkysik unci Chemie, (2) xcv. when v = 0. Proofs are to be found on pp. 496 498 of Hankel's memoir.

(1855),

A number of extremely interesting symbolic investigations of the formulae are to be found in Heaviside's* papers, but it is difficult to decide how valuable such researches are
to be considered

when modern standards


is

of rigour are adopted.

A
x
f

remarkable memoir

due to Malmste'n f, in which the formula

cos

ax dx
.

tre~ a
1

jo (1

+ a*) M+1 ~2+

.M!

x[(2a)n
is

+ nC

(n

+ l).(2a)n- + n C.
1

(n-rl)(n+2).(2ay-*-r...]

obtained

(cf. 6*3).
f

This formula
cos

is

written symbolically in the form


f

ax dx
.

ire~ a

Jo (1

+x

2
)

n+1

2 :W+1 .?i!(

\2a

1
_1

|>]
,

the

[ ]

denoting that \n\~ m

is

to be replaced

by (n)_ TO and

this, in

Malmsten's

notation,

means
l/{(w+ l)(+2)...(n

m)}.

It will

be observed that this notation

is

different from the notation of 4*4.

7*24.

The asymptotic expansions of her (z) and

bei(z).

From the formulae obtained in 721, 7*23, the asymptotic expansions of Thomson's functions ber (z) and bei (z), and of their generalisations, may be written down without difficulty. The formulae for functions of any order have been given by Whitehead^:, but, on account of their complexity, they will not
* Proc.

Royal Soc.

lii. (1893),

pp.

504529; Electromagnetic Theory,

ii.

(London, 1899).

My

thanks are due to Dr Bromwich for bringing to my notice the results contained in the latter work. t K. Svenska V. Akad. Handl. lxii. (1841), pp. 65-74.
X Quarterly Journal, xlii. (1911), pp.

329338.

204 be quoted here.

THEORY OF BESSBL FUNCTIONS


The
it

[CHAP. VII

functions of zero order had been examined previously by

Russell*; he found

convenient to deal with the logarithms! of the functions

of the third kind which are involved, and his formulae

may be written as follows

n ^ K
'

ber <*) - exP a (*) cos r (,\ \bei(z)~ V(27r^)sin^


)

'

(9\
V

'

ker (*)

_ ex P(-^) c OS o
V(2*/tt)
\

(kei(s)

sin^
13
128*4

;'

where
. .

a (z)

z ~ -r= +
'

25

-v/2

8^V2
7r

384^ V2
*

'

^^~V2~8~~8772"T6V~384*V2
The ranges
|

R(

25

+ ""
in the case of (1)

of validity of the formulae are


in the case of (2).

arg^

< \ir

and

arg z

< |tt

These results have been expressed in a modified form by Savidge, Phil. Mag.
(1910), p. 51.

(6) xix.

7*25.

Hadamard's
which
is

modification of the asymptotic expansions.


of considerable
theoretical

result

importance

is

due to

Hadamard J; he has shewn

that

it is

possible to modify the various asymptotic

expansions, so that they become convergent series together with a negligible re-

mainder term. The formulae will be stated for real values of the variables, but the reader should have no difficulty in making the modifications appropriate to complex variables.

We
u,

take

first

the case of / (x)

when

> \. When we

replace sin

\B by

we have
I' ( * )=

uMml?
2
(2a?)" e*
J
i

a"' ,An'' em
.

i>+i)r<i).Jo
2
(2a?)" e*

exp (- 2u*x) uiv (1 - u*y~i du


{\

- v)m
m!

-I> + 1)1^)^0
We may
C\\ {)

Wa +2m eXp(

,~ , 2Ma?)rfW '
is

io

the last result being valid because the series of integrals


write this equation in the form

convergent.

T (t\ K{a!)

-r( V + l)V(i)^2x)J: ml(2xr]


e*

(J

v )m

rzx

fv+m-hf-tfif l dt e

V(27r) TOto

(l-v)m .y(v + m + ^,2x) I> + i).ro!(2aOm

'

where 7 denotes the "incomplete Gamma-function" of Legendre.


* Phil. Mag. (6) xvn. (1909), pp. 531, 537. t Gf. the similar procedure due to Meissel, which will be explained in 8-11. Of. Modern Analysis, 16-2. t Bull, tit la Soc. Math, de France, xxxvi. (1908), pp. 7785.

7*25, 7*3]

ASYMPTOTIC EXPANSIONS
x,

205

For large values of

the difference between

y
is

(v

+n + \,
r( + $)

2x)

ana

tt

+v >

0(x v + n + i.e~ Sx ) which

is

o(l) for each integral value of n.


for

In the case of the ordinary Bessel functions, we take the expression


the function of the third kind

*' w

H, 1+ <>-U)T>+tfJ.' "^( s)

WxJ r(v+t)
so that

OT

m!(2#) w

and

similarly

(3 )

^w

-( i) r
;

^- M
\
is

From

i^K^ ^
)+o
;

>

these results

it is

easy to derive convergent series for the functions

of the first

and second kinds.


for functions of order zero only

Hadamard gave the formulae


functions of any order exceeding

but the extension to

obvious.

7*3.

Formulae for

the

remainders in the asymptotic expansions.

In

7*2

we gave an

investigation which shewed that the remainders in

(z) are of the same order of and v magnitude as the first terms neglected. In the case of functions of the first and second kinds, it is easy to obtain a more exact and rather remarkable theorem to the effect that when v is real* and x is positive the remainders
the asymptotic expansions of
v

Hw

(z)

after a certain stage in the asymptotic expansions of

J v (x) and Y v (x)

are

numerically

less

than the

first

condite investigation ( 7*32), it same sign as the first terms neglected.

terms neglected, and, by a slightly more recan be proved that the remainders are of the

Let us write

i).f;--{(^ir-(-r }*=^->'
*

We may

take

v^Q

without losing generality.

206
so that

THEORY OF BESSEL FUNCTIONS


\*

[CHAP. VII

(2
(2)

[cos (x

T \vtt - {-n) P (x,

v)

- sin (x *\vir-\ir)Q (x, v)\

Yv (x) = (^) yrrx/

[sin (x*\vTr-Tr)P (x, v)

+ cos (x + \mr

tt)

Q (x,

v)\

Now I(v)

and, in the analysis of 7*2,

we may take

B to be \ir since

the variables are real, and so A^,


It follows that, if

1.

p be taken

so large that 2j>i/

J,

there exists a

number

0,

not exceeding unity in absolute value, such that


(i 1 I

2^

<*-"> f !?Y" + *'<*-"> f 2!\* Y~* _ ~ TOto m! V2tW "^ (2^)! ^2**/

'

and, on adding the results combined in this formula,


/i

we have*
2* (-*> / (2p)I

+
_,_
1

2x)
|

Y~* + fi
j.

**)

-2 Y~* "
O

"v

(WW
(2m)!

to

UW

JLY" 4.

M* W

'

where 6

^1

and, since

is

obviously real,

- 1 ^ $ $ 1.

It follows

on integration that

P( *'* )=
and since

mto
"
f
9

(2,n)!(2*)-

(2p)!(2^r(, + |)J

^
P

du

>

e~

W +2p-* cZw U

e~tt u+w-* dw

= T (/ + 2p +

(x, v) does not we see that the remainder after p terms in the expansion of exceed the (p + l)th term in absolute value, provided that 2p>v \.

From

the formula
(i

V-2W
we
of

j.

*U

Y~*

V"* _ t (*-"> f Y" 4. ?iftr y W' f * ",* 2**/ (2p+~l) ml \2**/ \

1
'

find in a similar

manner that the remainder after p terms

in the expansion

Q (x, v) does 2p>v-.

not exceed the (p

1 )th

term in absolute value, provided that

These results were given by Hankel, Math. Ann. I. (1869), pp. 491 494, and were reproduced by Gray and Mathews in their Treatise on Bessel Functions (London, 1895), p. 70, but small inaccuracies have been pointed out in both investigations by Orr, Trans.
Camb. Phil.
Soc.

xvn. (1899), pp. 172180.

In the case of

(x)

we have the formula

* This result
(1859), pp.

was obtained 189196.

in a rather different

manner by

Lipschitz, Journal filr Math. lvi.

7-31]

ASYMPTOTIC EXPANSIONS
p

207

and
(i V

+ iX' = * (-lJLz2!k ( u y
h

2x)

mZ

ml

\2x)

and,

when

p>v \,

the last term

may be

written

(p-D!
where

<

<$

1,

and

so,

on integration,

*.()where
This

()V
v

v -\(
,t

(2ar)

(2a )"pJ'
f

$ #2
is

<$

when p >

- \.

a more exact result than those obtained for


;

P (x, v)
is

and Q (x,

v)

by the same methods

is, of course, the fact that (1 + %ut/x)"-P-l is positive and does not oscillate in sign after the manner of (1 + \iutjx) v -P-* liut/asy-P-*. (1

the reason

why

the greater exactness

secured

7*31.

The researches of Stieltjes on


-

(x),

(x)

and

(x).

The results of 7 3 were put into a more precise form by Stieltjes*, who proved not only that the remainders in the asymptotic expansions associated with J (x),Y (x) and (x) are numerically less than the first terms neglected,

but also that the remainders have the same sign as those terms.
Stieltjes also
it+.

It is only to
all

the terms

(x), but his result is complicated and we shall not reproduce be expected that I is intractable because in the dominant expansion have the same sign whereas in the other three asymptotic expansions the
(..)

examined I

terms alternate in

sign.

It is evident from the definitions of 7 '3 that

P (*' 0) =
Q (X
'

2^r

~ UX
~ UX

U~k K 1

+ ** M) "* +

(1

** M >"*J
~'
]

du

>

0)

2J^So

^
1

{(1

+ url " (1 ~

hiu)

du

In these formulae replace (1 +

j'w) -i
/'*

by
d<f>

_
</>'

"".'o
*

|iw

sin 2

Ann.

Sci. de

VEcole norm. sup.

The function Iv (x) has


in Taylor's

also been

Math.-Vereinigung, xix. (1910),

pp. 233252. examined by Schafheitlin, Jahresbericht der Deutschen pp. 120129, but he appears to use Lagrange's form for the
(3) in. (1886),
it is

remainder

theorem when

inapplicable.

208

THEORY OF BESSBL FUNCTIONS

[CHAP. VII

It is then evident that

[*"{!

-} w*sintf> + ... +(-)*-

($u*Bm*<i>)r-1

+ (-)p ($ sin* 4>/{l + \u* sin4 0)} (ty,


where

is

any positive integer (zero included).

Now,

obviously,

Jo

l+Ksin
and
1
;

2
</>

Jo

K%

*'

where 8

lies

between

and hence

\ (a + \iu)-i + (i

- \wr*} - 1 -

~f ({uY + ^jf^- ) w_i and

If

we multiply by

M the positive function e" *

integrate, it is evident

that
(1)

P(*.0)-l-g^ s ^,+

..+(-)

(2p-2)!(8a>)*-

+ ^ rUi
and jj is any positive integer result which had to be proved for P (a?, 0).
where

i.y.y...(4p-iy
(2p)!(8<r)v

<

0,

<

1,

(zero included);

and

this

is

the

Similarly, from the formula


-

i ((1

4 , n ii + iiu) 4 _ (i _ itU )-i - - j o


,
.

fi"

6 *^&
ittsin9
dd>

we

find that

m
(2)

n/ n\ c^,o)_-

la
(ai),

i'-

3*- 5'
1

8! (8*)

"

l'.ff.5...(4p-8y (2p-l)!(8)- 1

and p is any positive integer (zero included) and result which had to be proved for Q {x, 0).
where

< 6t <

1,

this is the

In the case of

Stieltjes took the formula

K
and replaced
(1

(a)

* C/ e- u-i (1 +!)"* du,


i
.

+ $w)~ J by -

i,

t ^ie

Proced ure then

follows the

method just explained, and gives again the

result of 7*3.

By an ingenious device, Callandreau* succeeded in applying the result of Stieltjes to obtain the corresponding results for functions of any integral order ; but we shall now explain a method which is effective in obtaining the precise results for functions of any real order.
* Bull, det Sci.

Math.

(2) xiv. (1890),

pp. 110114.

"

7-32]
7*32.

ASYMPTOTIC EXPANSIONS
The signs of

209

the remainders in the asymptotic expansions associated

with

(x)

and

(x).

been seen that Jv (x) and Yv (x) are expressible in terms of two functions P(x, v) and Q{x, v) which have asymptotic expansions of a simpler type. We shall now extend the result of Stieltjes ( 731) so as to shew that for any real value* of the order v, the remainder after p terms of the expansion of P (x, v) is of the same sign as (in addition to being numerically
It has already
less

thanf) the (p
(x,

holds for

on

which these conditions Q enable the theorem to be stated in the following manner:
.

+ l)th term provided The v) when 2p > v - f

that 2p

>

- \:

a corresponding result
lay

restrictions

In

the oscillatory parts

of the series for

P {x, v) and
7 3,

Q {x,

v),

the

remainders

are of the same sign as, and numerically less than, the first terms neglected.

By

a slight modification of the formulae of

we have

p {x

>

v)

2T^ry)/

" e~ ux " v-i


~ ux

{{1

* iu)v

~h

+(1 "
(1

iu)v

~ i]

du

'

Q{x, p)=

2iT^Y) \1
;

"""' {{l

+ ** M) ""*

~ * iuy ~ i] du

'

and, exactly as in 7 3

we may shew

that

IRi + nay-* + (i - iiuy-i]


(

Ji
(1

trAtz^itr
\&ni).

wi=0

)P(i

ur~7
(2p~2)?
o

- typ

"

{(1

^y^ - & - wr**)

dt.

The reader will see that we can establish the theorem if we can prove that, when 2p > v \. the last term on the right is of fixed sign and its sign is that of

It is clearly sufficient to
1
[

shew that
{(1

is

v $] positive. Now this


2p

\\ _ t?P-* U
expression

+ \%vty-** - (1 - liuty-v+l)
equal to J
-

dt

is

1 f (1 - 1)**-* ii \w-*-i \e~ K {i+i * Jo (2p-v-$)T(2p-v-$)] K '


.

iut)

- e~ x

<1

- i M 1 d\ dt
'

=r7K-
1

f {2p-v + $)JoJ 9

I "( 1

- m-2 \v->-* sin ($ Xut)


2

e~x d\dt

= --_^ ,- f"\*--^ P(l -t)*1 {2p v + $)Jo Jo


*

sin ($\ut)dtd\.

As

in 7-3

we may take

v^O

without loss of generality,

t This has already been proved in 7*3. t Since isin^Xwt) |^$Xuf, the condition
infinite integral.

2p>-b

secures the absolute convergence of the

210

THEORY OF BESSEL FUNCTIONS


Now
(1

[CHAP.
t;

VH

t)w~*

is

a monotonic decreasing function of


f,

and hence, by the


exists such that

second mean- value theorem, a number

between

and

1,

|(1Jo

t)

2?-*

sin ($\ut) dt

( sin (l\ut) dt

> 0.

Jo
is positive,

Since

T (2p v + )

we have succeeded
+

in transforming

2F-T^Ti

1 (1 C ~ O^"*** K

W-*

- (1 " ImO'-*^} dt

into an infinite integral in which the integrand is positive, and so the expression

under consideration

is positive.

That

is to say,

i{(i

+ 4y-*+(i-iwy-}
"
mto
(2wi)!
it

(2p)!
( 7*3)

where

0^0 when 2p> v \. And


|

has already been seen

that in

these circumstances

6 ^
|

1.

Consequently

the last equation by e~ux u v ~* and integrating,


stated for

^1 we at once
;

and then, on multiplying


obtain the property

P (x, v).
v) follows

The corresponding property for Q (x,


\.

from the equation

i + 1 r. - (i

_ })-*) =

-g

<->"-^ -;^ft">""
(

the details of the analysis will easily be supplied by the reader.


Note.

The
lies

analysis fails

when

\<v<\

if

we take p=0, but then the phase


\

of

(l$m) r -*

between

and \{v-\)ir, and so


like
v)

{{l+\iu) v -\ + (\ -iiw)"

*}

has the

same sign as unity, and, in \{v-\)u, and hence P{x,

and

manner, {(1 + %iu) v ~h (1 - \iu) v ~ i}/t has the same sign as $(j-, v) have the same sign as the first terms in their
still

expansions, so the conclusions are

true

and the conclusion

is

true for

Q (x,

v)

when

7*33.

Weber's formulae for the remainders in the expansions of functions

of the third kind.

Some

inequalities which are satisfied


ll)

by the remainders

in the asymptotic

expansions of H,

have been given by Weber*. These inequalities owe their importance to the fact that they are true whether z and v are real or complex. In the investigations which we shall give it will be supposed for
(z)
(2

and

>

(z)

simplicity that v

is real,

though

it will

pe obvious that modifications of detail

only are adequate to

make the mode

of analysis applicable to complex values


404416.

Math. Ann. xxxvn. (l|890), pp.

,,

7-33]

ASYMPTOTIC EXPANSIONS
There
is

211

of

v.

no further

loss of generality in

We

shall write \z\

= r,

and, since large values of z


j

assuming that v^O, R(z)^0. are primarily under con|

sideration,

we

shall

suppose that
have*, by

1r^v \.

If v

- \ > 0, we

612 (3),

j(iy y- -'-'.r e
,

-.,

>->f 1

y-'

2\i e'G-W-1")
\TTz)

;iy
If

w
r(- +
i)

/ *oxp {-
(l
j.

(l

- ^*)} - A*
-

_^p
1)

^ v<

we use

the recurrence formula

// (z) = (2/*) ( +
It is thus (1)

and apply the inequality just obtained


found that

(*)

- f>, (z)
on the right.

to each of the functions

HM (z)^G\ (%Trz)-i

e*(-W-

and similarly
(2)

H (z)^0\ ($ir*)-*e-*fr-*"-W

where

(3)

("<t)

The

results

and v (z). more refined inequalities from them in the following manner
'

may be called Weber's crude inequalities satisfied by Hv (z) By an elegant piece of analysis, Weber succeeded in deducing
{1)

Take the first p terms of the series involved in Hankel's two expansions and denote them by the symbols 2 (z p), 2 2 (z p), so that
(1)
( '

u>

(z

v)

V
m=0 dz

(-)m -(v>)

2 m

(z

o)

V (W)

It is easy to verify that

Yd
dz*

Z%

+
z>

z " {z,p)

-z*{-2izy-i

We

regard this as an equation to be solved by the method of variation of para-

meters; we thus find that

(z;p) = {\ttz)* e -^-w-H |4 (^ jyr

,i,

^+B ^^
(

(2)

* In the third line of analysis the inequality e x

^ 1 + x (x^O)

has been used,

t When !< we take 2r>y + $.

212

THEOBY OF BESSEL FUNCTIONS

[CHAP.

Vn

where

A (z)

and

B (z) are functions


(z)

of z so chosen that

(A' (z) H v v

(z)

+ B' (z) H (z) 0,


+ B (z) H.
(z)

|4' (z) H."


It follows that

= - (*x#)-* e^M

/^^
9

A' (z)

= * <*)r* *<- W

^
P

Jfr

(,),

and so
il (#)

- 4 - 1 Tr/^ {tT (5 + t)}-> 6**-*~-W

h 2 :^'f

H (Z +

t) dt,

t)]p

where

A is a constant. We obtain a similar expression


(*
;

for

B (z), and hence

it

follows that

2, m

p)

- {-4#

(1
>

(*)

+ ##,

,2)

-'(*-*--i') (*)} (**)* e

-**P-toP)J. \^t)

F*F+0|5
-*-

*
+ oo
,

By considering
is

the behaviour of both sides of this equation as z


that

it

not

difficult to see

A=l

and

B = 0.
in the forms

Hence we may write Hankel's formulae

Hu
y

(z)

\ttz)
(z)

J*-*"-**) {2 r (z; p)
e -*- W-i*> {2<*>

+ Rp

<

H,
where the remainder

= (~\

(z;p) +

Rp %

Rp w may
j,

be defined by the equation

" ***

'

{V

'

P)

\JTV
\z

\-2i(z

+ t)}*
so,

dt

Sinee

R (z) ^ 0,
we

we have

+ t\^ ^/(r2 + t'), and Q3


|

by using the crude

inequalities,

see that. the modulus of the last integrand does not exceed
2*-p 7T"1
.

(r3

+ t*)-* lp+1)
|

Hence
|

Rp w
1,

^ 2 1-* G*p

(v,

p)

f
.'

(t*

+ )-*<p+ dt,

and

so,

when p >

we have

(4)

IVI<l(y)l^f|^r
I-p

and similarly

r(Jp + J).|(2*)r and it will be observed that in Weber These are the results obtained by the relative values of v concerning made the analysis no hypothesis has been results obtained by from the differ and p\ in this respect Weber's results
;

K^w

,p;i

other writers.

7'34]
7*34.

ASYMPTOTIC EXPANSIONS
Approximations
to

213

remainders in

the asymptotic expansions.

When
expansion

the argument of a Bessel function


is

term in

it

is not very large*, the asymptotic not well adapted for numerical computation because the smallest (with the remainder after the smallest term) is not particularly

small; at the same time the argument

may

be sufficiently large

for

the

ascending series to converge very slowly.

An
by

ingenious method for meeting these numerical difficulties was devised

Stieltjesf;

we
(x)

shall explain the

method

in detail as applied to the function

K (x) and state the results which were obtained by Stieltjes by applying the
method
to

and

(x).

We
so that

apply the transformation indicated in

7*31 to

the formula 6"15 (4),

\- e

~x

~mi du

e~ xu dOdu _ ejV2 f * J** ~ it J Jo m*(! + |wsin2 0)

V
7T

2
m = oJo
JO

*-?-(lu V
M*

m Bin* 0) '

d0dn

JO

That

is

to say,

n"^"(-i-^^ds*."
.'( 'o

*(1

+ |usin

0)

where

D Rp=~i\

at f"

f*
.'o

e~ xu (hi sin8 0)*

t'.'o

w*(l
(0,

w,

+ usm
,

r~-^rd0du. 2
0)
is

Now
when x

the value of
is

m for which

m)j(2x) m

least is nearly equal to

2x

large

accordingly, in order to consider the remainder after the

smallest term of the series for

(x),

we choose p hp +
;

so that

x=
where a
is

<r,

numerically less than unity

and then
.

Rp=-i
e _1

tt*Jo Jo

2 (1 -*" sin 0)p

m*(1

..,

$usm +\
:

2
-

0)

m d0dn.
up
to a

Now, as u increases from to qo \ue~* H increases from (when u = 2) and then decreases to zero so we write
,
;

maximum

$-*

= e~

~S\

where increases from

to oo

and, for similar reasons,

we

write

sin 2

= ef.
J
(x),

* The range of values of .r under contemplation for the functions from about 4 to about 10. t Ann. Sci. de VEcole norm. sup. (3) in. (1886), pp. 241252.

(x)

and

A' (x) is

214

THEORY OF BESSEL FUNCTIONS


of integration becomes the whole of the
(f, rj)

[CHAP. VII
plane
;

The domain
found that

and

it is

R
where
a o,o

2f^r
7T*

f-

pM)
2o- 2

a r>s

J -oo J -oo

? v>}dtdV
)

lr%=0

= i

2,o

^j,

ao,2

= i>

by some rather tedious arithmetic.


the asymptotic expansion of

It follows* that the

dominant terms of

Rp for large values

of

are

so that

It is easy to verify

by

Stirling's

theorem that
K

(2xy
term omitted.

'

W
if

'

so that the error due to stopping at one of the smallest terms

is

roughly half of the

first

In

like

manner

Stieltjes

proved that,

P (x, 0) and Q (x, 0) are defined as


K

in 7-3, then
(3)

{>

~Jlo

(2*)

'

where
(5)

(6)

provided that

is

chosen so as to be nearly equal to

x,

and t

is

defined to

be x

p.

Results of this character are useful for tabulating Bessel functions in the critical range;

some similar formulae have been actually used for that purpose by Airey, Archiv der Math, und Phys. (3) xx. (1913), pp. 240 244-; (3) XXII. (1914), pp. 30 43 and British Association

Reports, 1913, 1914.


It would be of some interest to extend the results, which Stieltjes has established for Bessel functions of zero order (as well as for the logarithmic integral and some other

functions), to Bessel functions of arbitrary order.


* Cf.

Bromwich, Theory of
8*3.

Infinite Series, 133, 137,

and 174, or the lemma which

will

be

proved in

7*35, 7'4]
7'35.
If

ASYMPTOTIC EXPANSIONS
Deductions from Schafheitlin's integrals.
replace a by 2 tan
in the formulae of
f*
]

215

we

732, we deduce that


t

p/

(2x) v+ i
1

sin-* 6 cos (v
cos 2 " +1 6
sin"-*
sin
(i;

- V) 6

(v

+
+

i)

n,

(2x) v+ l
A

[**
o

- 4)

(v

f).

cos 2 " +1

which resemble Schafheitlin's integrals of


It is obvious

612.

from these results that

P (X,
Q (*,

V)>0,
*)

(_<<

3)

>

o,

(!<"<!)

Q(ar,i')<0.

(~\<v<\)
that

An
is

interesting consequence of these results

is

we can prove

that

Q(x,v)/P(x,v)
an increasing function of x when

< y < and that

it is

a decreasing

function of #

when

<

i>< f

For we have

Q'(x,v)P(x,v)-P'(x,p)Q(x,v)
(

(2x) v+ l

2 )

ft* ft*

where

F (e
>

*>

l^^
(cos0cos<)
:

<

tan *

- tan *> c s <* (tan

*>

e sin ("

*)

<*>>

so that

!(, *)

+ *<*

0)

g=|^|^
;

tan

</,)

sin (i

-,,)(-

*).

If we interchange the parametric variables 0, $ in the double integral and add the results so obtained we see that, when | < v < f the double integral has the same sign as \ - v and this proves the result.
,

7"4.

Schlafli's investigation

of th e asymptotic expansions of Bessel functions.

its

In a memoir which seems hardly to have received the recognition which importance deserves, Schlafli* has given a very elegant but somewhat

elaborate investigation of the asymptotic expansions of the functions of the third kind.

The

integral

formulae from

generalisations of Bessel's integral

which he derived these expansions are although Bessel's integral is not so well

adapted as Poisson's integral


*

for

constructing the asymptotic expansion of


in

Ann. di Mat.

(2) vi. (1875), pp.

the importance of this

memoir

is

recognised

120. The only standard work on Bessel functions is the treatise by Graf and Gubler.

which

216

THEORY OF BESSEL FUNCTIONS


when
z
is

[OHAP.

VH

(z)

large

and

v is fixed, yet Schlafli's


it

method not only succeeds

in obtaining the expansion, but also

expresses the remainders in a neat and

compact form.
Schlafli's

procedure consisted in taking integrals of the type

and selecting the contour of integration in such a way


Ire* (u
is constant.

that,

on

it,

the

phase* of

+ l/)
is

He
it
;

took two contours, the constants for the respective contours

being
(I)

and

and

it is

supposed that r

positive

and a

is real.

Let us

first

take the phase to be


u

ir

write
,

+ pe a

where p

is

positive

and 6

is real,

and then

reia p*e2i0 l(l


is

+ pe *)
1

negative,

and

is

consequently equal to

its

conjugate complex.

Hence we have
(1) {L)

sin(a

+ 2fl)

sm(a + 6)'

U
variable

sinfl

sin(a
<f>

+ 0)

Next choose a new parametric


<f>

such that

= 20 + a - 7T,
.

and then
' ( v2)

/^

w=
as
<f>

cos \ (a T7 cos |(a

- <b)
+
<p)

,,

(u

- l)
u

TT e *.

re

~=

T7 cos | (a

- r sin JX cos i/ a ~ 9) f( +
2
,

.x

9)

(ir a) to (ir a), u traces out a contour emerging from the origin at an angle {ir a) with the positive real axis and passing to < a < 2v. infinity at an angle (ir a) with the positive real axis, provided that
Now,
varies from If this restriction is not laid

on a the contour passes to

infinity

more

than once.

We
value
a,

shall

now lay

this restriction

on a

specified for formula 6*22 (9), provided that

and then the contour is of the type we give a> and arg z the same

as is permissible.

It follows that

2i sin vrr

^bC'-'M-DI-J*
<f>

where u
*

is

defined in terms of

by equation

(2).
this section with the

The reader
viii.

will find it interesting to

compare the general methods of

"method
Chapter

of steepest descents" which

is

applied to obtain various asymptotic expansions in

7*4]

ASYMPTOTIC EXPANSIONS
<f>

217

Changing the sign of is equivalent to replacing u by 1/u, and so, replacing the expression on the left by its value as a function of the third kind, we have
(8)
e*""

H (re'>-**>) = TV"' +
.

i*")

exp

\reu (u

+ 1)1

^~to

d<f>.

From
as
<

(2) it follows that

varies monotonically from

re* (u Yf/u increases steadily* from to it a and, if we write


;

oo

- re u (u so that
t is

iy/u

t,

positive

when u

is

on the contour, we have


dt
'

du
~u

dt

- reu (w - 1/u) ~ e -t(T-)i ( wi + u-*) *J{rt)


ir.

the range of values of arg w being less than

Next, by Cauchy's theorem,

it is

outside the contour because the origin


It follows that
( 4)

convenient to take the point = 1 inside the contour, but f = is a branch-point.

must be

n. +u-. = <*+j

-f

,jr?(.t-}W
Jo
J

Hence

W
Now
1

'

'"
p- 1

27rirM"

(f-iy +
(-)p
tP

W^)

'

it is

evident that
(-)
g>

" w = (C- l)""* (r+r + (- l)v (re-)P {(?- 1)' + #/(re*)} (f - 1)' + #/(re*)
where

ft>

is

any

positive integer (zero included).

It will be convenient subse-

quently to suppose that

exceeds both

R(p )

and

R ( v $).
and observing that

On making

this substitution in the last integrand


,

j_

h+ w+
,

1+

2-Tri]

*
-

^
2),

a*

I>-m + ).(2m!)
I

I> + m + i)

= m\(v,m)
(2m)!
*

(with the notation of 7


(6)

we deduce

that

H.

(<-.-.)

= (^)*exp {nf //(+, l/w+.n-)

) [ <=^> + V>]
<?-#>-* "+p-*

where
(_)p

ddt

"*
, .

2mV(2V); 27rtV(27r)io
d
dtp cos

J i
sin 2

(- V)^ (re{(r1

1)'

+ ?*7(ia )}

a + cos ft

_ ~

sin

ft

(1

+ 2 coa a cob ft + cos" ft) (cos a + cos ft) 2


*

W.

B. F.

218
First consider
I

THEORY OF BESSEL FUNCTIONS


r(+,i/+,i+)

[CHAP.Vn

" +*>-*dZ

When p
circles

(c-iy^-M^-iy+w^M is so large that it exceeds both R (v ) and R{v ^), we take the
ton]

contour to be as shewn in Fig. 15; and

when the

radii of the large

and small
to zero.

tend to

oo

and

respectively the integrals along

them tend

If

now we

write

=e ,ri (l-a;)Joo
on the two rays (which are
1
r (+, l/t*+,

all

that survives of the contour),

we

find that

i+)

+p-i

d
(-)* cos
vir
1

ton)

(-l)*p->{(f_l)*+#/(re*)}
f

aP-* (1

- xY+*-*dot;
'

tx(l-x)/(reia )

Fig. 15.

Now

the numerator of the integrand is positive (when v is real), and the modulus of the denominator is never less than 1 when \nr< a< tr\ for other
values of a
it is

never less than

sin

Therefore

where
is

is 1

or

cosec a

according as cos a

is

negative or positive.

When

complex,
(7)

it is

easy to see that

Vk

cosi/tt

\(R(v),p)
|

cos

(i/tt)

(2r)*>

7'4]

ASYMPTOTIC EXPANSIONS
finally,

219

Hence,
(8)

when

*.<,

w.

$v~ [s fcg^>
>
| |

ir < arg z < w,

+ * fcggtfj
is

where
fied

0j

does not exceed 1 or sec (arg s)

according as I(z)
|.

positive or

negative, provided that v is real

and p + $ >

When v is complex, the modisee that, in (8),


X

form of the remainder given by (7) has to be used.


Since

has

its real

R{1 -te(l - x)jfrd*j\ > when ^(e^^O, we part positive when v is real and l(z)^0.
by
iz in (8)

If z be replaced

we

find that,

when arg z\<ir,


|

and,

when

v is real,
(i)
(ii)

i2(0,)>Oand|08 j<l, ifR(z)>0,


|0,|<|cosec(args)|,ifl*(s)<O.

The

modifications necessary for complex values of v are left to the reader.

(II)

We

next discuss the consequences of taking the phase of

ire<-(M-2
to be zero.

+ l/tt)

As

before,

we

write

u=

+ pe**,
therefore equal to its conjugate
(1).

and then reu />9 **/(! +pei9) is positive, and complex, so that we obtain anew equation
preceding analysis by writing
<p

We

then diverge from the

= -(20 + a)
(-iy =
u
rsin'ft

so that

(10)

tt-" *^**) sin(a-0)


11
<f>

rCfa
'

sin(a-)sin(a +

4>)'

Now,
angle

as

varies from

to

a,

traces out a contour


real axis

origin at

an angle a with the positive


with the positive real
axis,

emerging from the and passing to infinity at an provided that a lies between ir and it.
6'22 (8)
if,

The contour is then of the type specified for formula missible, we give a> and argz the same value a.
It follows that,

as

is

per-

when tr < a <

tt,

K
where u
(11)
is

(re**)

= cos vir

u-"' 1 exp 1

- ^re* (u +
;

-) f

jt d<t>,

defined as a function of
<

<f>

by equation (10)
1

and therefore

(/*'<-*>)

-^1T( m-k + u) exp |- Ire * (u + =1

d log u
1)

220

THEORY OF BESSEL FUNCTIONS


if

[CHAP. VII

and hence,

now
t

= reu (u-lflu,

we

find that

have consequently expressed a second solution of Bessel's equation in a and the analysis its asymptotic expansion can be deduced (z), the final result being that, when proceeds as in the case of v
form from which
;

We

{1)

- tt < &rg z <


(13)

fa,
pz 1

H 0) = (}
|

e-*

<*-*'*-*>

(m)

(v,p)

L-o(2i*)"
|

(2**)M

argz according as I{z) ^ or I(z) > 0, and .ft (02 ) ^ when I(z) ^ 0. If v is provided that v is real and p + 1 > u complex the form of the remainder has to be modified, just as in the case of (8).
where
2
|

does not exceed 1 or

sec
|

It should be observed that, since the integrands in (3)

and (11) are even

functions of

v, it is
,

must exceed |
Poisson's type.

unnecessary in this investigation to suppose that R (v) as was necessary in investigations based on integrals tof

7'5.

Barnes' investigation * of asymptotic expansions of Bessel functions.

The asymptotic expansions of functions of the third kind follow immediately from Barnes' formulae which were obtained in 6*5, 651. Let us consider " t " '~p r (- s) r (- 2v - s) r ( v + s + j) (- iizy ds
,

/.

ao

ivp

= (- 2u)-*-*
If arg
|
|

["'
J -TCi

T(-s + p+p) r(-s-v +p) V(s-p + 1) (- 2iz)ds.


8,

( iz) ^ f it

we have

r(-s+v+p)r(-s-v+p)r(s-p + i)(-2izyds\
I/:
1

< j"
J -coi

\r(-s + v + p)r(-s-p+p)r(s-p + $)el*- s)


convergent and so the
first

'

ds\,
is

and the

last integral is

integral of all

0{(-2i*)-'-p}.
But, by the arguments of
6'51,

the

first

integral

is

2iri

times the
it is

sum
at

of the residues at the poles on the right of the contour, and so

equal to

7T- Hv w {z)j\ei(z~ v " s = - p , p f,

cos pit (2z) v ] plus


...,

27ri times the sum of the residues p p + \. The residue at v m \ is


r (p + m + ) T(- p + m + $)
m\(-2izy +m+ *
(1908), pp.

(-)"'

* Trans.

Camb. Phil. Soc. xx.

273279.

7'5, 7*51]

ASYMPTOTIC EXPANSIONS
j

221

and

so,

when arg ( iz) < \ir $


|

M
and

- 1Fr(,

:yS +
i

' ti)

^-H
investigation of

this is equivalent to the result obtained in 7*2.


(z)

The

HJ

may be

constructed by replacing

by - i throughout.

The reader should notice that, although the determination of the order of magnitude of the remainders by this method is transparently simple, it is not possible to obtain concrete formulae, concerning the magnitude and the sign
of the remainders, which are ultimately supplied

by the methods which have

been previously considered.

7'51.

Asymptotic expansions of products of Bessel functions.

It does not seem possible to obtain asymptotic expansions of the four products Jn(z) J,(z) in which the coefficients have simple forms, even

when
ifM
<2>

fi

= v. The

reason for this

(z)

H^ (z)

have asymptotic expansions

exists for the general

term

(z) (z) and which no simple expression the leading terms in the two expansions are
is

that the products


for

H^

2 eg**ri0+>+i).ri
trz

2/^+2^-1
4dz
...}.

The productsiTM (*)HW * (z) and J? M (z)H^(z), however, do possess simple


asymptotic expansions ; and from them
for

we can deduce asymptotic

expansions

j;(i)/P (#)+r,(i)F,(i)
and
for

J, (z)
simplest

(z)

- FM (z) J (z).
is

The

way of constructing the expansions


7*5.

by Barnes' method,

just explained in 541 suggests that

consideration of series of the type obtained in

we should examine

the integral

is to be chosen so that the poles of T (2s + 1) lie on the left of the contour and the poles of the other four Gamma functions lie on the right of the contour and it is temporarily supposed that /*, v and fiv are not

the contour

integers, so that the integrand has

no double

poles.

The

integral is convergent

provided that

arg(t>)

< f 7r.

222

THBOBY OF BESSEL FUNCTIONS


First evaluate the integral

[CHAP.

VH

by swinging round the contour

to enclose the

sequences of poles which


is

lie

to the right of the original contour ; the expression

equal to minus the

sum

of the residues at these poles, and the residue at

m + JO* + v) is
ggH+ww
~~

sin fiir .sin vtr

r (fi + v + 2m + 1) (-)m (^zy+* +am ) sin (/a + v)ir' m\T + m + 1) T (v + ra + 1 ) T (ft + v + m + 1


.

(jj.

It follows that

^D<^ +1 r (H-H r
>

ft

?--)

xr (^- a r (-^-^) (
)

)s,<fo

~ sin fiir sin pit \


e*"->- J_^ (z)
sin (v

sin (ji

+ v) tr
sin
(/*

sin

(ji

- v) ir
| )

J (z)

g-(M+")X JL (*) /_, M

(,g)

fi) w

+ v) ir

sin (^

+ v) 7r sin (/* i>) ir


(^

2lnT^ ^
C

)J' (^)+

(^

)F ' (

- cot | (/* - *) ir {/M (*) Fr (s) - FM (z) J (*)}]


"*

2 COS J

(/LI

+ V) IT

.[{/(*)/,(*)+ r,(*)F.Cr)}

+ \,*&\(r-v)Tr{J {z)Y {z)-Y


ti

tl

{z)J (z))l
if

By
| |

writing
|

for i
\

arg iz and arg ( iz) are

throughout the analysis we deduce that, less than $ w, i.e. if arg z\<ir, then
|

both

- COt J (/. " ")

(J (*)

F, (*) - 1 , (*) J. Wl]

r (=*-) r(-^-*)cos*7r.(|*)<fc,

7*51]

ASYMPTOTIC EXPANSIONS

223

and
(2)
-

M . t +>))r [|J">WJ.W+F,(,)r.(.)
+ tan i 0 - ")

[J* (') Y. (*) 1

F () /, (*))]

-BD^^er-'-) "^-)
xr(J- s) r (-^-) sin, w .(i,)-<&.
These results hold for all values of p and v (whether integers or not) provided that, in the case of the former p + v and p p are not even integers, and. in the case of the latter p + v and p p are not odd integers.

We
(1)

now

obtain the asymptotic expansions of the functions on the

left

of

and

(2) after the

manner of

7*5.

We first

take

to be

tegrands on the

left

of the line R(s)


/oof

an integer so large that the only poles of the in= -p-\ are poles of Y (2s + 1) ; and then
raoi-pl
J

-x*

-xj p J

(when either integrand is inserted) is equal to 2wt times the sum of the residues at the poles between the contours. Since
, x.-p-J J

-xt-p-J

^ ^ ^ ^ ^_^
when arg z <
j i

we deduce
(8)
.

that the asymptotic expansions,

ir,

are

Jh (z) Jv (z) + F (z) F, (#)] - cot \{p - v) 7T


+

/*

[J, (,) F Cr) -

(z)

F (*)]

7T

.=o

"

(2m +

1) !($*)*+

ipi-rini^-Oir-^l
and

+1

'

+^ -- +

1,

g"

2'"W

2
4
2

At +
\

|/

+1

/A-I/

+ 1 I/-/A+1
2

1-/A-1/
2
'

_1\
?/*

~7rzcosi(/*-v)7r"

2'

224

THEOEY OP BESSEL FUNCTIONS


In the special case when fiv, the
last

[CHAP. VII

formula reduces to

(5)

JS(*)+7S(s)~2- 5 {1.3.5...(2m-l)}^,

and, in particular,
(6)

Jo

(,)+I.

<*)~~J& o
I.

(2m)l(&)-

Formula
formulae
(3)

(5)

seems to have been discovered by Lorenz, K. Danske Vidensk, Sehkabs


[Oeuvres scientifcpies,
(1898), p. 435], while the

Skrifter, (6) vi. (1890).

more general

were stated by Orr, Proc. Camb. Phil. Soc. x. (1900), p. 99. A proof of (5) which depends on transformations of repeated integrals was given by Nielsen, Handbuch der Theorie der Cylinderfunktionen (Leipzig, 1904), pp. 245 247 the expansion (5)

and

(4)

is,

however, attributed to Walter Gregory by A. Lodge, British Association Report, 1906,

pp.

494498.

It is not easy to estimate exactly the magnitude or the sign of the remainder after any number of terms in these asymptotic expansions when this

method
of J,2
(z)

is

used.

An

alternative

F2 (z) will be given in

method of obtaining the asymptotic expansion 13'75, and it will then be possible to form

such an estimate.

CHAPTER
8*1.

VIII

BESSEL FUNCTIONS OF LARGE ORDER


Bessel functions of large order.

The subject of this chapter is the investigation of descriptive properties, including approximate formulae, complete asymptotic expansions, and infunctions and the probe examined are of primary importance when the orders of the functions concerned are large, though niany of the results happen to be, true
;

equalities of various types connected with Bessel

perties

which

will

for functions of all positive orders.

We

shall first obtain results

which are of a purely formal character,


Next, we shall obtain certain

associated with Carlini's formula (1*4).

approximate formulae with the aid of Kelvin's* "principle of stationary phase." And finally, we shall examine the contour integrals discovered by

Debyef

these will be employed firstly to obtain asymptotic expansions


real,

when

the variables concerned are

secondly, to obtain

numerous

inequalities of

varying degrees of importance, and thirdly, to obtain asymptotic expansions of Bessel functions in which both the order and the argument are complex.

In dealing with the function J (as), in wfiich v and x are positive, it is found that the problems under consideration have to be divided into three
classes,

according as xjv

is less

than, nearly equal to, or greater than unity.

Similar sub-divisions also have to be

made

in the corresponding theorems

concerned with complex variables.

The
and

trivial

problem of determining the asymptotic expansion of J,

(),

when

v is large

z is fixed,

may be noticed
by applying

here.
Stirling's

It is evident,

theorem to the expansion of

3*1,

that

^(*)~exp{v+vlog(^)-(v+|)logv}.[co+^+^ +
where
p. 369.

...l,
lii.

1/J(2n)',

this result has been pointed out

by Horn, Math. Ann.

(1899),

For physical applications of approximate formulae for functions of large order, may be consulted: Macdonald, Proe. Royal Soc. lxxi. (1903), pp. 251 258; lxxii. (1904), pp. 5968; xc.A (1914), pp. 5061; Phil. Trans, of the Royal Soc. cxx. A 144; Debye, Ann. der Physik und Chemie, (4) xxx. (1909), pp. 57136; (1910), pp. 113 March, Ann. der Physik und Chemie, (4) xxxvii. (1912),, pp. 2950; Rybczynski, Ann. der
[Note.
the following writers

* Phil. Mag. (5) xxiii. (1887), pp. 252255. [Math, and Phys. Papers, iv. (1910), pp. 303306.] In connexion with the principle, see Stokes, Trans. Comb. Phil. Soc. ix. (1856), p. 175 footnote.

[Math, and Phys. Papers, n. (1883), p. 341.] t Math. Ann. txvu. (1909), pp. 535 558; MUnehener Sitzuvgsberichte, xl.

[5], (1910).

226

THEORY OF BESSEL FUNCTIONS


;

[CHAP. VIII

Phyrii und Ckemie, (4) xli. (1913), pp. 191208 Nicholson, Phil. Mag. (6) xix. (1910), Love, Phil. Trans, of the Royal Soc. ccxv. A (1915), pp. 105 131 Watson, pp. 516 537 Proc. Royal Soc. xcv. A (1918), pp. 8399, 546563. The works quoted all deal with the problem of the propagation of electric waves over the surface of the earth, and are largely concerned with attempts to reconcile theoretical with experimental results.]

8*11.

Meissel's first extension of Carlini's formula.


( 1*4)

The approximation

obtained by Carlini

is

the
;

first

term of the

asymptotic expansion of a Bessel function of large order


in the expansion were formally calculated
(1892),
col.

subsequent terms

281284,

It is clear that Bessel's equation


)

(1)
if

we

define a function

^
us>
...

in the following

by Meissel, Astr. Nach. cxxix. manner:


be written

may

+^-^(i-^.W=o;
u

u (z) by the equation

^( y*>* r(iT+l) exp {/


then equation
(2)
(1)

^ dz
2

transforms into
* [it' (*)

{u (s)} 2 ]

+ zu (z) -

i/

(1

- 5 ) = 0.
(z) is

If

now we assume

that, for large values of


v,

v,

expansible in a series

of descending powers of

thus

u (z)

vu +

it,

+ v^/v + u

3 /v*

...,

where

i*o,

i, Ma,

denote functions of z which are independent of

v,

by

substituting in (2) and equating to zero the coefficients of the various powers
of v on the
left,

we

find that

,{V(i

-)}/*,
s

^=20^).
Ui
'

"2== 3

Ws=
iu

45+105* + 8(1 -i-y


I62
8

_
s

64^ + 560a

+ i28(i-^)V
9

8 (i!^r 4565 5 + 25s 7


'

+ 3685 + 9245 + 374* + lSz


7

32(l-5 )
it is

2 7

Hence, on integration,

found

(cf.

1*4) that

J'u (z) dz -

v {log
1
1

+
(

^
(

_ 2, +
)

V(l

*)
2

- l} - i log (1 - z*)
2

+ 353 _ 24i/|(i_^
2

\
j

4>z

+ z*
|

16r> 2

(l-5 )3
3

+ 57607
+
3252

16

-1512s 2 - 3654^-3755" (l-*8 ) 1

2 88-g4

23 25s

13-g8

128^(1

-5 )

2 6

'

8*11, 8*12]

FUNCTIONS OF LARGE ORDER


Meissel's formula

227

Hence we have
(3)

( pz)=s

<?*y ex p y

&T (v + 1) (1

- * ^ *)* {1
4>z*

exp (- F*>

V(l

- *s )}'

where
<*>

(1-*)t 16i^(l-^) *-tS-}j

+ z*

1__

!6

- 1512.? - 3654g - 375s _


8

J
J

5760^1
32**

(i_*.)*

+ 288** + 232* + ISz* + ""


128v*(l-^)8

It will appear in 8*4 that the expression given for

is

the

sum

of the

four dominant terms of an asymptotic expansion which is certainly valid

when

lies

between

and 1 and v

is large.

It is stated

by Graf and Gubler* that the first approximation derivable from

(3),

namely

j
is
;

(ve)

*pW(r-/
(2irv)i(l-**)i{l+V(l
-*)}''

due to Duhamel but a search for the formula in Duhamel's writings has not been and it seems certain that, even if it had been discovered by Duhamel, his discovery would have been subsequent to Carlini's.
successful,

Note.
(5)

The reader should


t/, v (v

observe that (3)

may

also
.

be written in the form


.

= seen a) '

y/(2trv

' , '

tanh o)

where
(6)

^=^^(2 + 3secha)-^^(4secha+sech*a)
- gsLrJ (16 - 1512 sech8 a - 3654 sech4 a - 375 sech8 a)

-C

Ma
4

(32 sech* a+ 288 sech* a + 232 sech8 a + 13 sech*- a)

+ ....
8*12.

Meissel's second expansion.

The expansion obtained


z
is real

in 8*11 obviously fails to represent


for

Jv (vz) when
may

such values of z, Meisself obtained two formal solutions of Bessel's equation and, if we write z = sec /3, the reader
;

and greater than unity;

will see,

by making some modifications

in

81 1

(5),

that these solutions

be written in the form

/(*-*) P{f Attr. Nach. exxx. (1892),


col.

A tM.
i.

* Einleitung in die Tfieorie der BeueVsehen Funktionen,

(Bern, 1898), p. 102.

363368.

228

THEORY OF BESSEL FUNCTIONS

[CHAP. VIII

where*
(1)

P, = ^(4sec*,8

+ sec</3)
+
s

^^

(32 sec8

288 sec4

+ 232 sec

+ 13 sec8 )
8

+ g^T (768 sec + 41280 se&0+ 14884 sec/S + 17493 sec

+ 4242 sec
(2)

10

+ 103 sec

12

Q,

= ,(tan-/3)-^^(2 + 3sec
I

/3)

5 yg

^ (16

- 1512 sec 8

3654

sec*

- 375 sec

0)

+
to far,

3 225(30i;
....

< 256

+ 7872 sec2 3 + 18 9!200 sec4 + 4744640 sec6 + 1914210 sec8 /3 + 67599 sec )
'

10

To determine

Jv (vsec 0)
;

in terms of these expansions,

we take

to tend

and compare the

results so obtained with the expansions of Hankel's

type given in 7'21

we see that, as -*- \tt, P* 0, Q v ~v (sec -

%tt),

and we
(3)

infer that

JT. (v sec 0) =

J(~~)

e-^+^-K
e- p-- iQ-+ini

(4)

#<"

(i/

sec

0)

= J(r^!~)
/

'

It follows that
(5)

J, f> sec

/S)

^/

(^~

- p" cos (Q
- p " sin
(Q,

- ^r\

(6)

F, (v sec 0) =

^/(~^)

- ,

where P and Q are defined by (1) and (2). It will appear subsequently ( 8 41) that these formulae are actually asymptotic expansions of Jv {y sec 0) is any assigned acute angle. and F (y sec 0) when i> is large and
#

of

Formulae which are valid for small values of 0, i.e. asymptotic expansions valid when z and v are both large and are nearly (z) and F (z) which are equal, cannot easily be obtained by this method but it will be seen in 8*2 that, for such values of the variables, approximations can be obtained by rigorous methods from Schlafli's extension of Bessel's integral.

The reader

will observe that the

approximation has been carried two stages farther than in

the corresponding analysis of 8*11.

8*2]
Note. form
(7)

FUNCTIONS OF LARGE ORDER


The dominant terms
in the expansions (5)

229

and

(6),

which

may be written

in the

Jv (x)=Mv coa(Qv -br),


*'-

Yv {x)=Mv &va{Q v -ir),

Wh6re

(**-*) )*>
(v/x),

Qv ~ */(*? v 2 ) - \vir + v arc sin

had been obtained two years before the publication of Meissel's paper by memoir on Physical Optics, K. Dantke Videnskabernes Selskabs Skrifter, [Oeuvres Scientifiques, L (1898), pp. 421 436.]

L. Lorenz in a
(6) vi. (1890).

The procedure of Lorenz was to take for granted which has been proved in 7*51,

that, as

a consequence of the result

-* Jf *'

t*L

Til + 8'
y2

y2

-*,

1-3 (*>-*). (*-f) ** 2.4

"I

-J

-JLTi
*#!_

-*T*
J
'

tf*

and then to use the exact equation


(8) V
'

SL'._L_
dx
-nxM*'
(1),

which

is easily

deduced from the Wronskian formula of 3*63

to prove that

whence the approximation stated

for

Qv follows without difficulty.

Subsequent researches on the lines laid down by Lorenz are due to Macdonald, Phil. Trans, of the Royal Soc. ccx. A (1910), pp. 131144, and Nicholson, Phil. Mag. (6) xiv.
(1907), pp.
ix. (1911), pp.
is closely

697707 (6) six. (1910), pp. 228249; 516537; Proc. London Math. Soc. (2) 6780; (2) XL (1913), pp. 104126. A result concerning Qv+1 -Q which connected with (8), has been published by A. Lodge, British Association Report,
;

1906, pp. 494498.

8*2. The principle of stationary phase. Bessel functions of equal order and argument.

The principle of stationary phase was formally enunciated by Kelvin* in connexion with a problem of Hydrodynamics, though the essence of the principle
is to be found in some much earlier work by Stokes f on Airy's integral ( 6'4) and Parseval's integral ( 2*2), and also in a posthumous paper by Riemann %.

The problem which Kelvin propounded was


integral
1

to find

an approximate expression

for the

2irJ

cos [to {x tf(m)}] dm,

which expresses the effect at place and time (x, t) of an impulsive disturbance at place and time (0, 0), when /(to) is the velocity of propagation of two-dimensional waves in water
corresponding to a wave-length 2n/m.

The

principle of interference set forth

by Stokes

* Phil. Mag. (5) xxra. (1887), pp. 252255. [Math, and Phys. Papers, rr. (1910), pp. 303306.] t Comb. Phil. Trans, a. (1856), pp. 175, 183. [Math, and Phys. Papers, n. (1883), pp. 341, 851.] t Ges. Math. Werke (Leipzig, 1876), pp. 400-406.

230
and Rayleigh
of values of

THEOBY OF BESSEL FUNCTIONS


in their treatment of group-velocity

[CHAP. VIII

that, for large values of

and wave-velocity suggested to Kelvin x-tf(m), the parts of the integral outside the range (ft -a, /* + ) are negligible on account of interference if /x is a value (or the value) of m

which makes

^[{*-*/(i)}]=0.
In the range (ji-a, /a+a), the expression m{x-tf(m)} three terms of its expansion by Taylor's theorem, namely
is

then replaced by the

first

M {*-<f (m)} + 0. (m
and
it is

-/0-iW .00 + 2/'

0*)}

(m-tf,

found that, if*

W2
"n/[-Mm/"(m) + 2/'(m)]'
cob
{*//' 0)

then

f ~7r^/[-2t{hf"( H.) + 2f'(H.)}]

+ !*} Ar

{/' Qi) + jir}


V["2ir<{f*/"(M) + 2/'0*)}]-

In the last integral the limits for


replaced by
It will
the case

a,

which are large even though a be small, have been

oo

and

+ oo

due

to

be seen from the foregoing analysis that Kelvin's principle is, effectively, that in of the integral of a rapidly oscillating function, the important part of the integral is that part of the range of integration near which the phase of the trigonometrical
is

function involved

stationary \.

It has subsequently been noticed J that it is possible to give a formal mathematical proof of Kelvin's principle, for a large class of oscillating functions,

by using Bromwich's generalisation of an

integral formula
will

due to
for

Dirichlet.

The form of Bromwich's theorem which

be adequate

the applica-

tions of the principle to Bessel functions is as follows

let

ybea function
i
jo

Let F{x)be a function of x which has limited total fluctuation when x^O; Then, ifl<fi<l, of v which is such that vy -* oo as v -* oo
.

v*

a?- 1

F (x)smvx

dx-~

(+ 0)

tr~ l sin t dt
.

= F (+ 0) V (ft) sin \fitr

Jo

and,

if0<fi<l,the

sines

may

be replaced by cosines throughout.


will

The method which has just been explained


*

now be used

to obtain
at =/*;

an

This

is

the appropriate substitution

when m{x-tf(m)} has a minimum

for a

maximum
e.g.

the sign of the expression under the radical is changed.

t A persistent search reveals traces of the use of the principle in the writings of Cauohy. See equation (119) in note 16 of his TMorie de la propagation des Ondes, crowned Sept. 1815, Mem. prisent&s par divers savants, i. (1827). [Oeuvres, (1) i. (1882), p. 230.]

t Proc. Camb. Phil. Soc. xix. (1918), pp. 4955. Bromwich, Theory of Infinite Series, 174.

8*2]

FUNCTIONS OF LARGE ORDER


is

231
This formula,

approximate formula for J (v) when v which was discovered by Cauchy*, is


(1)

large

and

positive.

J9 (v),

r(*)
2^.3*7T^'

This formula has been investigated by means of the principle of stationary phase, comparatively recently, by Nicholson, Phil. Mag. (6) xvi. (1909), pp. 276 277, and Rayleigh,

Phil.

Mag.

(6) xx. (1910), pp.

10011004

[Scientific

Papers, v. (1912), pp. 617620]; see

also Watson, Proc.

Camb. Phil.

Soc. xix. (1918), pp.

4248.

From

6'2 (4) it is evident that

J( v) =
and obviously

tJo

1 f'cos

[v

(0-

sin 0)}

d$-^^ Jo
ff

( V-ie-fsinh*) dt,

smw
Hence

"
e

^ {t+Anh() dt
[v

["^t dt=Q ^i v)
d$

J
Now
let
<f>

()

f T JO

1 'cos

(6

- sin 6)}

(1/v).

= sin 0, and
f 'cos { v (0 Jo

then
I'

- sin 0)\ d0 l

cosv<l>

Jo 1

-cos

d<f>.

But
and hence,
it

im

_i_
-cos0

2
6*
(0, v),

*-ol
if
<f>*/(l

cos 0)
['
I

has limited total fluctuation in the interval

follows from Bromurich's theorem that


cosi/0
,. -* dq>

Jol-costf

6 r ~ -= r 6* Jo
|

/*

...

* cos v<bd<b

,.

r ^
'

=S

ra)C Si7r

and then
It
still

(1) follows at once.

this result

has to be proved that 0*/(l we observe that


.
4>*
f

- cos 0)
1

has limited total fluctuation

to establish

<# (1-costfJ
where
so that

_ <fr~*sinflg(0)
3(1 -cos 6)*

'

o ^ * v '

2(1-008^ _
sin ^

3 $ _ 8m 0)

^(0)=0,

gr(,r-0)=+a>,

^ (0) (1 - cos 0)*/(l + cos 0) > 0,


and therefore, by integration, g(0)^0 when O$0<tt. Consequently <*/(l-cos0) monotonic and it is obviously bounded. The result required is therefore proved.
* Compte* Eendus, xxxvm. (1854), p. 993. Cauchy's methods will be given in 8-21.
is

[Oeuvret,

(1)

xn. (1900), p. 663.]

proof by

232

THEORY OF
By means
of

BjSSSEL FUNCTIONS
it is

[CHAP. VHI
possible to obtain

some tedious integrations by parts*,

a second approximation, namely

and

it

may

also

be proved that

(3)

J/W-lr^ + o (*"*);
is

an associated formula
(4)

F.(,)~-?
2*7Tl/*

The asymptotic
in 8-42.

expansions, of which these results give the dominant

terms, will be investigated with the aid of

more powerful analytical machinery

8*21.

Meissel's third expansion.

The
It will

integral just discussed has

obtain the formal asymptotic expansion of

now be explained how this a more complete form) by Meissel the theoretical cesses employed will be investigated in 8*42.
;

been used by Cauchyf and Meissel $ to Jn (n) when n is a large integer. expansion was obtained by Cauchy and (in
justification of the pro-

Taking the formula

Jn (w)
let

=-

cos {n (0

sin

0)} d0,

us write

sin $*

it

then follows that, for sufficiently small values of t,

m=0

and

\q =

1,

\ = $,

\2 = TlW

^* =

26&00>

^<

~ T7 llooo

>

It follows that

Jn (n) - -

f( I

(2m +

1)

X. *"l cos Qnfi)


limit,

d0.

When n

is large,

\ni? is large at the

upper

and Meissel inferred that


(\n&)dt,

/n(w)~- 2 (2m
7Tt=0

+ l)*. O .'0
\

t*" cos

* BeeProc. Camb. Phil. Soc. xix. (1918), pp. 4248. t Comptet Rendu*, xxxvm. (1854), pp. 990998, 11041107. [Oeuvret, (1) xu. (1900), pp. 161164, 167170.] t Astr. Nach. cxxvn. (1891), col. 359 362; cxxviu. (1891), col. 145154. Concerning formula (1), Meissel stated "Sohon vor dreissig Jahren war ich zu folgenden Forme! gelangt."

8*21, 8*22]

FUNCTIONS OF LARGE ORDER


the sign indicating a "generalised integral"
( 6*4)
;

233

where

is

and hence, by

integrating term-by-term and using Euler's formula, Meissel deduced that

(1)

I I A T (fm + i) Jn (*) ~ 7T (l ) \/
in _

,m+}
cos

(*m +

*) .

Meissel also gave an approximation for X m , valid

when
(1).

m is large;

and

this approxima-

tion exhibits the divergent character of the expansion

The approximation is obtainable by the theory developed in the memoir of Darboux, "Sur l'approximation des functions de tres grands nombres," Journal de Math. (3) iv. (1878), pp. 556, 377416.

We consider the singularities of 6 qua


be monogenic) are the points at which

function of

the singularities (where 6

fails

to

dirn and

t=(12rir)*, where

r +1, 2, 3,

...

and near*

**=

(12w) the dominant terms in the expansion of 8 are


2*- + (36r)*{l 1 J

* + -41'

By the theory of Darboux, an approximation to X m is the sum of the coefficients of fr** 1 in the expansions of the two functions comprised in the last formula ; that is to say
that

Xw ~2.(36,r)i

*-M-<.->
(2w + l)!(12ir)*" +
,
'
1

2r(2m+)
3^r(|)r(2m + 2).(127r)*m
and
so,

by

Stirling's formula,

(2)

Xm

~
;

(18)*r(|)(m+J)*(12ir)*
This is Meissel's approximation by Cauchy, loc. tit., p. 1106.

an approximation of the same character was obtained

8*22.

The application of Kelvin's principle

to

J
i.

(v sec 0).

The

principle of stationary phase has been applied


for

by Rayleighf

to obtain

an approximate formula and v is large J.

J (v sec 0) where

is

fixed positive acute angle,

As

in 8'2

we have

J, (v sec 0)

= - (cos [v (0 - sec

sin 0))

d0+O (1/v),
= 0.
in-

and

sec

sin

is

stationary (a

minimum) when
<f>,

Write sec creases from to

sin

tan +

so that

<f>

decreases to zero as
to
tr.

and then increases as

increases from

* These are the singularities which are nearest to the origin,

t Phil. Mag. (6) xx. (1910), p. 1004. [Scientific Papers, v. (1912), p. 620.] % See also Macdonald, Phil. Traru. of the Royal Soc. ccx. A (1910), pp. 131144; and Proc. Royal Soc. lxxi. (1903), pp. 251258; lxxii. (1904), pp. 5968.

234

THEORY OF BESSEL FUNCTIONS


Now
cos [v (0

[CHAP. VIII

sec

sin 0)}
r/-0

dd
f-/5

Jo

+ tan/5-]

dd

and 0*
Hence,
i/^

as

0-/8.

V(2 tan 0)
<f>*

(d0/d<f>)

has limited total fluctuation in the range


tftat

0^0 ^ tt,

it

follows from Bromwich's theorem

cos \v (0

- sec

J*

sin 0)}

d0 ~ 2

cos (v (<

+ - tan 0)}

//

2 <f>tan)

^i/

tan 0,

and

so
(v / "\

n\ (1) v '
The formula
/0 \ (2) '
is

ti

o\ sec r'/ 0)

{ir) ~ cos {y tan 0-0)m


*

V(i*"rtanJ)

( 777

sin{*>(tan/9-#)-?r} v/ o\ 7 * ^ Y (y sec #-/ 0) ~ i-v os v(i/7rtan/8)


v
-

derived in a similar manner from 6 21

(1).

The reader
expansions

will observe that these are the


(6).

dominant terms in Meissel's

8*12 (5),

To complete the

rigorous proof of these formulae

we have

to

shew that 0*

(d$/d<f>)

has

limited total fluctuation.

Now

the square of this function, namely

<f>

(dO/chf))

2
,

is
.

equal to

- sec ft sin - ft + tan ft _ ... _AW cos tf)* (1 - sec


say.

'

But
.,
^

'~

cos

ft

cosecd

(1

- sec ft cos fff- 2 (fl-secftsin - ft+tanft)


cosftcosecd(l-secftcos0) 3

The numerator, k (0),

of this fraction has the differential coefficient

- cos ft cos $ cosec2 0(1- sec ft cos 0) 2


and so k(0) decreases steadily as $ increases from
as $ increases from \ir to
to

ir, and then increases steadily since k(0)=O wben 0=ft<r, it follows that h! (0)^0

when

0^0 <ft
|

and

A'(0) changes sign once (from negative to positive) in the range

ft^d^JT.

Hence
point (a

yjh

(ff)

is

monotonic (and decreasing) when


ft

< 6<
|

minimum)
it

when
proved.

< 6 <

slh (6) consequently has limited total fluctuation


in the range
ir
;

<6<

since

and it has one stationary bounded and continuous when ^ 6 ^ r, as had to be


ft,

is

8'3]
8*3.

FUNCTIONS OF LARGE ORDER


The method of steepest
descents.

235

A development of the theory of contour integration, called


steepest descents*, has been applied
tions

the method of by Debyef to obtain integral representaof Bessel functions of large order from which asymptotic expansions are
If,

readily deduced.

in general,. we consider the integral


e"/<"
J,
<f>

(w) dw,

v is supposed to be large, the contour is chosen so that it passes through a point w at which /'(w) vanishes and the whole of the contour is then determined by the assumption that the imaginary part of f(w) is to be constant on it, so that the equation of the contour may be written in
[
|

in which

the form

If(w)=If(w ).
are real

To obtain a geometrical conception of the contour, let w = u + iv, where u, v and draw the surface such that the three coordinates of any point
;

on

it

are
u,
v,

Rf(w).

If

absolute
all

Rf(w) = z, and if the z-axis be supposed to be vertical, the surface has no maxima or minima except where /(w) fails to be monogenic; for, at
other points,
dv 2

du*

The

points [u

R/(w )]

are saddle points, or passes, on the surface, so that


is

the plan of a curve on the surface which goes through one of the passes on the surface. This curve possesses a further property derived from the equation of the contour for the rate of change of the contour of integration
;

f(w), at any given value of w, has a definite modulus, since /(w) is supposed to be monogenic and since If(w) does not change as w traverses the contour,
;

it

follows that
is

Rf(w) must change

as rapidly as possible

that

is

to say, that

the curve
it,

characterised

by the property that

is

so chosen that it is

any point of the steepest curve through that point and on the
its direction, at

surface.

It may happen that we have a freedom of choice in selecting a pass and then in selecting a contour through that pass our choice is to be determined from the consideration that the curve must descend on both sides of the pass for if the curve ascended, Rf(w) would tend to + oo (except in very special
;

cases) as

left

the pass, and then the integral would diverge

if

R (p) > 0.

* French "Methode du Col," German "Methode der Sattelpunkte." t Math. Ann. lxvii. (1909), pp. 535 558; Miltichener Sitzmigsberichte, xl. [5], (1910). The method is to be traced to a posthumous paper by Biemann, Werke, p. 405 ; and it has recently been applied to obtain asymptotic expansions of a variety of functions.

236

THEORY OF BESSEL FUNCTIONS

[CHAP. VIII

oscillate rapidly

The contour has now been selected* so that the integrand does not on it and so we may expect that an approximate value of
;
:

the integral will be determined from a consideration of the integrand in the

neighbourhood of the pass


the interference effects
(cf.

8*2)

from the physical point of view, we have evaded which occur with any other type of contour.

The mode of derivation of asymptotic expansions from the integral will be seen clearly from the special functions which will be studied in 8*4 8*43,

8*6, 8'61

convenient to enunciate at this stage a lemma f which will be useful subsequently in proving that the expansions which will be obtained
;

but

it is

are asymptotic in the sense of Poincare.

Lemma. Let F{r)

be analytic

when r ^ a
\

8,

where a

> 0,

>

and

let

F(t)= I am ^ m^~\
when
|

^ a,

r being positive ; also,


t,

let

F (r)

<

lTe*

r
,

where

and

b are
the

positive

numbers independent of

when t

is positive

and r > a.

Then

asymptotic expansion

is

1 am T (m/r) v~ m lr m=l valid in the sense of Poincare when \v\is sufficiently large and
e- VT F{r)dT~
/,

where

is

an arbitrary positive number.

It is evident that, if

M be any fixed integer, a constant K


m=l

can be found

such that

whenever t

whether r ^ a or t
j
J

> a and
;

therefore

e- vr F{r)dr=

2
\.

m=l JO

e^a^^^dr + RM

where

l^jfl^f

\e~T

K^i^^dr

Jo
x

= K r (M/r)/{R (v) - b) M* = 0(v-x'r


),

provided that

R (v) >

b,

which

is

the case

when

> b cosec A. The

analysis

remains valid even when b is a function of v such that compared with v. We have therefore proved that
f JO

R (p) b is not small


{v~ M'r ),

erF (t) dr = X am T (m/r) it-**" +


m-l
is

and

so the

lemma

established.

an account of researches in which the contour is the real axis see pp. 1343 Burkhardt's article in the Encyclopadie der Math. Wits. n. 1 (1916). t Cf. Proe. London Math. Soc (2) xvu. (L918), p. 133.

* For

1350 of

8*31]
8*31.

FUNCTIONS OF LARGE ORDER


The construction of Debye's contours* when

237

the variables are real.

It has

been seen in 6*2, 6*21 that the various types of functions associated

with J, (x) can be represented by integrals of the form

taken along suitable contours.

On

the hypothesis that v and x are positive,


of the contours appropriate for the
in 6*2, 6*21.

we

shall

now examine whether any

method of steepest descents are of the types investigated

explained in

In accordance with the principles of the method of steepest descents, as 8*3, we have first to find the stationary points of

x sinh w vw,
qua function of w,
(1)
i.e.

we have

to solve the equation

a;

cosh

w v 0;
consider,

and it is at once seen that we shall have three distinct cases to in which xjv is less than, greater than, or equal to 1, respectively.
sider these three cases in turn.
(I)

We con-

When xjv < 1, we

can find a positive number a such that

(2)

x v sech
is

a,

and then the complete solution of (1)

w = a + 2mri.
It will

be sufficient to confine our attention to the stationary points + a; at f


a?

these points the imaginary part of


of the contour to be discussed
is

sinh

w vw

is zero,

and so the equation

/ (x sinh w vw) 0.
Write

w = u + iv,
= 0,
or

where

u, v are real,

and

this equation

becomes

cosh u sin v
so that v
/ox (o)

v cosh a = 0,
v cosh siny
a
.

coshw =
,

The contour

gives a divergent integral.


(3).

We

therefore consider the

contour given by equation

and ir, correspond pairs of values of u which are equal but opposite in sign and as v increases from to n, the positive value of u steadily increases from a to + x
;
.

To

values of v between

* The contours investigated in this section are those which were discussed in Debye's earlier paper, Math. Ann. lxyii. (1909), pp. 535 558, except that their orientation is different; cf. 6-21. t The effect of taking stationary points other than a would be to translate the contour parallel to the imaginary axis.

238

THEORY OF BESSEL FUNCTIONS

[CHAP, vin

is unaltered by changing the sign of v and so the contour is symmetrical with regard to the axes the shape of the part of the contour

The equation
between v

= tr

and v = tr

is

shewn in
TTl

Fig. 16.

>
-77
Fig. 16.

If
it is

= sinh a a cosh a (sinh w w cosh a),


is real

easy to verify that t (which

increases in the directions indicated

on the curves shewn by the arrows.

in the figure)

As

travels along the contour from oo

mi
since,

to oo

+ iri,

t decreases from

+ oo

to

and then increases to

+ oo

and

by

6*2 (3),

IlTlJ ao-vi

we have obtained a curve from which we can derive information concerning Jv {x) when x and v are large and x\v< 1. The detailed discussion of the
integral will be given subsequently in 8*4, 85.

iri give information concerning a second but this problem is complicated by Stokes' phenomenon, on account of the two stationary points on the contour.

The contours from

oo

to oo

solution of Bessel's equation;

(II) (4)

When x/v>l, we

can find a positive acute angle


a;=i'sec)9,

/3

such that

and

the relevant stationary points,


cosh

which are now roots of the equation

w cos /8 = 0,

are

w= i#.
take the stationary point
if$,

When we

the contour which

we

obtain

is

I (sinh
so that, replacing
/

w w cos /3) = sin ft fi cos ft,


the equation of the contour
sin ft
is

why

+ iv,

5\

cos h

u=

+ 1^_J^ __
(v
ft)

cos

ft

8*31]

FUNCTIONS OF LARGE ORDER


for values of v

239

Now,

between
sin ft

and

tr,

the function

+ (v ft) cos ft sin v


is

has one

minimum

(v

= ft) at which the value of the function


and
ir,

zero

for other

values of v between

sin ft

+ (v ft) cos ft > sin v.


and
ir,

Hence, for values of v between


of

equation (5) gives two real values

(equal but opposite in sign), and these coincide only

when

= ft.

They

are infinite

when

v is

or

ir.

The shape

of the curves given by equation (5)

is

as

shewn

in the upper

half of Fig. 17; and if

tsj (sin ft ft cos ft) (sinh w w cos ft),


it is

easy to verify that r (which

is

real

on the curves) increases in the

directions indicated
to oo

by the arrows. As

travels along the contour from


to
-f-

+ iri,

r decreases from

+ oo
\

to

and then increases

oo

and so we

^^

ni

S^3

-"""*^

-ni
Fig. 17.

have obtained a curve from which [ 6'21 (4)] we can derive information concerning (#) when x and v are large and x/v > 1. The detailed discussion v of the integral will be given in 8*41, 15'8.
{1)

If

we had taken the

stationary point

ift, we

should have obtained the

curves shewn in the lower half of Fig. 17, and the curve going from
oo

oo

to

7rt

gives an integral associated with

{i)

(x); this also will be discussed

in 8*41.

The two

integrals

solutions of Bessel's equation, so that there

now obtained form a fundamental system of is a marked distinction between


1.

the case x/v

<

and the case xjv >

240
(III)

THEORY OF BESSEL FUNCTIONS


The
case in which v = x

[CHAP.Vm
to be con-

may be

derived as a limiting case either


0.

from

(I) or

from (II) by taking o or

equal to

The curves now

sidered are v
(6)

and
cosh u
18.

v/sin

v,

and they are shewn in Fig.

Fig. 18.

We

obtain information concerning

H,

ll)

(v)

and

{i)

(p)

by considering the

curves from

oo

to oo

iri,

while information concerning


oo

Jv (v)

is

obtained

from the curve which passes from oo mi to tion will be given in 8'42, 853, 8*54.
Geometrical properties of Debtee's contours.

+ vi. The

detailed investiga-

8*32.

An interesting result which will be found to be important in dealing with zeros of Bessel functions ( 15*8), and which is also used in proving certain approximate formulae which will be stated in 8*43, is associated with the second of the three contours just discussed (Fig. 17 of 831).

oc

to

+ jrt

is

positive

The theorem in question and does not exceed *JZ.

is

that the slope* of the branch

from

It is evident that, for the curve in question,

smh u -T- =
.

du
dv

sin( /3) (v ff)convcos3

sin 8 v

But sin(; /3)secv-(i>-/3)cos/3 has the positive derivative cos /3 tan 2 v, and hence
follows that
sin

it

(v- /9) - (v f3) cos

v cos

has the same signt as v /3. Therefore fur the curve under consideration, dv/du

since v /3 and u
is positive.

are both positive or both negative

* Proc. Comb. Phil. Soc. xix. (1918), p. 105.

Since, in the limiting case (Fig. 18) in


is

which

3=0,

the slope

is

on the

left

of the origin

and

^3

immediately on the right of the origin, no

better results of this type exist.

f This

is

obvious from a figure.

8*32, 8*4]

FUNCTIONS OF LARGE ORDER


</3,

241

Again, to prove that dv\du does not exceed


1

we

write

Wr) = ^

8iDg + (i, ~ i3)COS


sin v
'

and then

it is sufficient

to prove that

3^'*(*>)-iP(t;)+l>0.

Now
2*'

the expression on the


()

left

(which vanishes when v=j8) has the derivate

W
:

[*" ()(v)
3

+ ()]
{sin'p+3 cos2 } cos 0+ sin 2 v sin
,

- [(*-)

0- 3 cos v sin

(i>

- /3)J.

v, But
.

/o\ o (v - 0) cos 0+
x
'

8in 2

vsin-3cost>sin(v-/3) -' i
sin2 r +3 cos* v
--5
s

has the positive derivate


positive

*.

T~\%> an<*

^ ^ nce
8i
(v)

it is positive

when =0,
it

it is

when

<v<

ir.

Therefore, since

has the same sign as

- 0,

follows that

2*' [3*'>) -*()]

has the same sign as v 0, and consequently

3^'()ip()+l
has =/3 for its only minimum between This proves the result stated.

v0 and

t>=r

and therefore

it is

not negative.

8*4.

The asymptotic expansion* of Jp {v sech a).

the results obtained in 8'31 we shall now obtain the asymptotic first kind in which the argument is less than the order, both being large and positive.

From

expansion of the function of the

We retain the notation of 8*31 (I) and it is clear that, corresponding to any positive value of t, there are two values of w, which will be called wx and Wj the values of wx and w% differ only in the sign of their imaginary part, and it will be supposed that
; ;

l(w )>0,
1

J(>8)<0.

We

then have

where x v sech a.

Next we discuss the expansions of /i and w% in ascending powers of t. Since t and dr/dw vanish when w = a, it follows that the expansion of t in powers of w a begins with a term in (w a)* by reverting this expansion, we obtain expansions of the form
;

Wl
*

_= i

_^! Ti(+D =o> + l

Wa _ a=

2
=0

(-)'n+1

ni+1

-^

T*<"+1 >,

The asymptotic expansions contained


lxvii. (1909), pp. 535

by Debye, Math. Ann.

558.

in this section

and

in 8*41, 8*42 were established

242

THEORY OF BESSEL FUNCTIONS


Moreover
1

[CHAP. VIII

and, by Lagrange's theorem, these expansions are valid for sufficiently small

values of Ti.

f(Q+.o+)/d t i;A

am

dr
>

~2wJ
2vi]

Ut/t*< +1

dw
Ti(m+i)
*

The double
fractional

circuit in the T-plane is necessary in order to dispose of the


;

powers of t and a single circuit round a in the w-plane corresponds to a double circuit round the origin in the T-plane. From the last contour
integral it folio ws that

am

is

the coefficient of l/(w

o)

in the expansion of

T-j(m+i)

m ascending
wa=
t

powers of

w a; we

are thus enabled to calculate the

coefficients a m .

Write

= - sinh a (cosh
c,

W and we have W 1) cosh a (sinh W W)


= cosh a,
c8
{c -t

where

= | sinh a,

coefficient of

in the expansion of

= ^ sinh a, Therefore a m c W + c F + ...}-* {m+l)


. . . .

is

the

The coefficients and so we have


fa
(wi)

in this expansion will be called

a (m),

a, (m), Oa (m), ...,

= c -* (wt+1)

ai(w)
a l(TO)
/
\

= Co-H-,|_^.|},
(m + l)(m + 3) m + 1 .Ca f _ <v *i*n|__ + s
i.

<,j.t>

2!

a,(m)
(1)

= c -Hw|-||
Co

(m + l)(m + 3) 2dc,
2s 21
.
'

c2

(ro+l)(i + 3)(wi
2*. 3!

+ 5)

c^
'c*

(m +

1)

(m +

(m+l)(m + 3)(ra + 5)
2s 3!
. '

W 3^
c
8

3) /2cj c8

CoV

(to

+ 1) (m + 3) (wt + 5) (m + 7)
2*. 4!

c^)

'cA

On

substitution

we find that fa =a (0) + (-! sinh o) *, (- | sinh a)-1 {J a, = a, (1) =


a,
2

coth

a},

(2)
| I

= a, (2) = - (- J sinh a)"* {$ - ft coth* }, 8 a, a, (3) = - (- 1 sinh a)" {-& coth a - ft coth a}, coth'a 4 fljfr cotha}, 4 - a* (4) - + (- i sinh a)"t T ^ {

ff

8*4]

FUNCTIONS OF LARGE ORDER


->-i
is sufficiently

243

Now
when t
|
j

^= 2

a, m

T"-

small

and since

-7it

aw

= cosh o cosh w,
to zero as r tends to

follows that

d (u\ w^jdr tends

+ oo
and
so

Hence the conditions stated


Jo

in the

lemma
dr)

of 8*3 are satisfied,

\dr
"

has the asymptotic expansion

o^IVm + l )

when x

is large.
{(a/,

Since arg

-o)/t*}-- \ir as t

-*-

0, it follows that, in

($5),

the phase of a

has to be interpreted by the convention arg a


(8)

= + ^7r,

and hence
>

Jr ( y secha)^
v(2

^ tenha) J
!

-^^--. (||;taDha)TO

where
(4)

(A =l, ^^-^cotn'o, 1 ^ 2 = TlF-tVVcoth 'a + ^^coth*a,


(v sech o) valid (3) gives the asymptotic expansion of number and v is large and positive.

The formula

when

a is any fixed positive

The corresponding expansion


contour from
,_, (5)

for the function of the second kind, obtained

by taking a

oo

to

oo

+ irt,
. ,

is

v , r,(,secha)~-

K-tnha)

V(iffytanha)

^-T^T -(^vtanhard^-w^ldr,
qua function of the

r(m+i)

(~)m A m

The

position of the singularities of

complex variable t, should be noted. These singularities correspond to the points where w fails to be a monogenic function of t, i.e. the points where dr/dw vanishes. Hence the singularities correspond to the values a + 2ntri of w, so they are the points where
t = 2mri cosh
a,

* = 2 (sinh a - a cosh a) + 2mri cosh ,

and n assumes

all

integral values.
;

It is convenient to obtain
(W( To)

a formula for dw/dr in the form of a contour integral be a pair of corresponding values of (w, r), then, by Cauchy's theorem,
'

if

fdw\ = J_ /"('+) dw dr Jl_ f (*+> dw_ tt 2tri J dr t-t \drJo 2irt J where the contour includes no point (except w ) at which t has the value t

244
8*41.

THEORY OF BESSEL FUNCTIONS

[CHAP. VIII

The asymptotic expansions of Jv (v sec 8) and F (v sec 8).

In 8*4 we obtained the asymptotic expansion of a Bessel function in which the argument was less than the order, both being large ; we shall now obtain the asymptotic expansions of a fundamental system of solutions of
Bessel's equation

when the argument is greater than the order, both being large.
;

We

retain the notation of 8*31 (II)

it is

clear that, corresponding to

any

positive value of t, there are

two values of

lying on the contour which

passes from oo to be supposed that

oo

+ tri

these values will be called Wj and

it

and

it will

R(w )>Q, R(w )<0.


l

We

then have

HJm (v sec 8) =
where x =
third kind.
It is thus found that
v sec 8.
is

wi

XT
\

Jo

[dr

~-

-r- \ dr,

dr)

The

analysis

now proceeds
by

except that a

replaced throughout

exactly on the lines of 8*4 %8, and the Bessel function is of the

Jo

\dr

dr)
of o
,

m=0
that
is

xm+*

To determine the phase


arg {(Wi

of ( i sin )""*,

we

obseive that

i8)/r} -* + \w

as t -* 0,
a

and so

= el7V(sin/8).

Consequently
(1)

H
like

{1)

(v sec

8)

we

*JQvrr tan 8) M =
find that

T ()

1,

L-*l
'

*
. *

(| pi tan )*

In

manner, by taking as contour the reflexion of the preceding contour

in the real axis of the w-plane,


(2)

HjHv*e*e)~^T*~;: 5 T tt h) V(Wtan/8) '(-i^tan/S)**


.-

w=0

T(i)

In these formulae, which are valid when /8 is a fixed positive acute angle and v is large and positive, we have to make the substitutions
:

(3 >
j

^2 = Th
and
(2),

+ ih cot2 y3 + $& *,
we
find that

If
(4)

we combine

(1)

J (v8ec8)rsj
\pwtaa 8/ L
_i_

^
*>

^
i

*
n

=o

T(i)

(i/tan/S)

m ? (-)

r(2m + 8)

Atn+r

"I

8'41, 8-42]
(5)
(

FUNCTIONS OF LARGE ORDER

245

F(i/sec)~
/

w r(2m l) ilw, / \*T sin(>tan-i//3-7r) a a x v ?'.- 2 (-) =-r + *-'.,-. 3/ Vi^rtan^/ L r<$) (|itanJ* =o

*.

'-

-cosCi/tan^-iz/g-iTr)
*
'

|
)W =o

0-)"-r(2m + f)
r(i)

(^tan/9)2 ' +1

The dominant terms

in these expansions are those obtained

by the principle

of stationary phase in 8*21.

8*42.

Asymptotic expansions of Bessel functions whose order and argument

are nearly equal.

to give adequate approximations

The formulae which have been established in 84, 8'41 obviously fail when a (or ) is small, that is when the

argument and order of the Bessel function concerned are nearly equal. It is, however, possible to use the same method for determining asymptotic expansions in these circumstances, and it happens that no complications arise by
supposing the variables
to be

complex.

Accordingly we shall discuss the functions

#.(*),

HJ*(i)

where z and v are complex numbers of large modulus, such that \z v\ not large. It will appear that it is necessary to assume that z v *= o (z$), order that the terms of low rank in the expansions may be small.

is

in

We shall write
v

= z(l-

),

and

it is

convenient to suppose temporarily that

We
(1)

then have

H^
v

(z)

= .j

exp

\z

(sinh

w w) + zew] dw,

where the contour and negative.

is

that shewn in Fig. 18; on this contour sinh

w w

is real

We write
t=

w sinh w,
Y

and the values of w corresponding to any positive value of t will be called w and wt of which wx is a complex number with a positive real part, and w2 is a real negative number.
,

We
(2)

then have

H& (z) = i [ V" iexp (zew,)

^-

exp (zews )

^l dr.

246

THEORY OF BESSEL FUNCTIONS


The expansion
of t in powers of

[CHAP. VIII
its*,

begins with a term in

and hence we

obtain expansions, of the form

exp (zeWi)

-j-i
*"

= t~ 3 2 m T Jm
b.

m=

exp (zewt)

^p = t-3 2
is sufficiently

e <"*+

bm t*

and these are valid when t


|

small.

To determine the

coefficients bm
1
/-(0+.0+.0+) 0+.0+) /(<>+,

we

observe that

/dw

rf

<+!)

_j_r(o+)

~
As
is

6wt J

6XP (

^w ^W) (w - sinh /)*

<' +1 >

in the analogous investigation of 8*4, a single circuit in the r-plane


is

inadequate, and the triple circuit

necessary to dispose of the fractional

powers of t; a triple circuit round the origin in the T-plane corresponds to a single circuit in the w-plane.
It follows that bm is equal to Je* ,m+1,w * multiplied

by the

coefficient of

wm

in the expansion of

exp (zew)

{(sinh

w w)/w*}~l

("*+1 >.

The coefficients in this expansion will be called b (m%

&i(w), b2 (m),

...

so that

It is easy -to

shew that

(b,(m) = &*"**,
bx

(m) = &<*+*&,

w.).a- {-*}.
(3)

w=

61(m+l)

tT--V-|'
--

o*{m)

o
|

24

12Q

504()0

j.

For brevity we write


&w ()

= 6*"+m (e*),

8-42]
so that*

FUNCTIONS OF LARGE ORDER

247

5 B
(4)

= l, - fa % (m) - \ <?z*
(e*)
(e)

4()-e* B (es) = *V - ^,
t

^5

- ^J- " ifo*

*3

+ **,
4.(0)

[3,(0)

We

then have

exp

-^ ^=
(jreto,)

ft<0)-T7*flfanr.

^jb.}
TO (e*) t**
t

ttT

ir~*

2
=0

ei<+u 6* <M+1

>

ezp(#e^)^*-iT-< 2
\

e^

m+1

(IT
.

^Q^m+li Bm (e2)T^n
lemma

TO=0

and [exp (zew) (dw/dr)]


It follows

satisfies

the conditions of the

of 8*3.

from the lemma

of 8'3 that
(ez) S

(5)

H^(z)^^^i^^Bm
h *{*)
9

mUni + l)^.^^,

and

similarly

(6)

AJ^-i( ^
+

5TO(

^)sini(m + i )w .r|i^i|)

We
(7)

deduce at once that

j^z
F,

^Lh

Bm{ez)

^ Hm+l)v '~m^
821
(2), it is to

(8)

W ~-^2

(->.^Min4(m + l).5jg^).
f

From
<)

the Cauchy-Meissel formula

be inferred that, when

m is large,
^m(0)~ r() (w + i)J(127r)Jw
,

but there seems to be no very simple approximate formula

for

Bm (ez).

The dominant terms in (7) were obtained by Meissel, in a Kiel Programing 1892; and some similar results, which seem to resemble those stated in 8'43, were obtained by Koppe in a Berlin ProgrammX, 1899. The dominant terms in (8) as well as in (7) were also investigated by Nicholson, Phil. Mag. (6) xvi. (1908), pp. 271 279, shortly before the

appearance of Debye's memoir.

The values of B (0). Bt (0), ... B w (0) were given by Meissel, Astr. Naeh. czxvii. (1891), 359362; apart from the use of the contours Meissel's analysis (of. 8*21) is substantially the same as the analysis given in this section. The object of using the methods of contour integration is to evade the difficulties produced by using generalised integrals. The values of B (tz), Bj (cz) and JB9 (ez) will be found in a paper by Airey, Phil. Mag. (6) xxxi.
*
col.

(1916), p. 524.

t See the Jahrbueh liber die ForUchritte der Math. 1892, pp. 476 178. I See the Jahrbueh itber die ForUchritte der Math 1899, pp. 420, 421.

248

THEORY OF BESSEL FUNCTIONS

[CHAP. VIII
|

We
The

next consider the extent to which the condition

arg z

<

\tr,

which

has so far been imposed on formulae (5)

(8), is

removable.

qua function of t, are the values be a monogenic function of t, so that the t 2 singularities are the values of r corresponding to those values of w for which
singularities of the integrand in (2),

of t for which

(or

w)

fails to

drjdw

= 0.

They are

therefore the points

Intri,

where n assumes

all

integral values.

It is consequently permissible to
less

swing the contour through any angle w than a right angle (either positively or negatively), and we then
analytic

obtain the

continuation of

(l)

(z) or

H
v

(z)

over the range

V < arg z < \ir t\. By giving v suitable values, we thus find that the expansions (5) (8) are valid over the extended region

w -

7r <

arg z <tt.

If we confine our attention to real variables, we see that the solution of tbe problem is not quite complete ; we have determined asymptotic expansions of Jv (x) valid when x and
v are large

and

(i)

xjv

<1,

(ii)

are transitional regions between

x/v>l, (iii) x- v not large compared with #*. But there (i) and (iii) and also between (ii) and (iii), and in these
|

transitional regions xjv is nearly equal to 1 while

a?

is large.

In these transitional

regions simple expansions (involving elementary functions only in each term) do not exist.

But important approximate formulae have been discovered by Nicholson, which involve Bessel functions of orders + J. Formulae of this type will now be investigated.

8*43.

Approximate formulae valid in


failure of the formulae of 8*4

the transitional regions.

The

8*42 in the
n,

transitional regions led

Nicholson* to investigate second approximations to Bessel's integral in the


following

manner

In the case of functions of integral order

Jn () = and,

cos (n0

x sin 0) d$,
it

when x and n
is

are nearly equal (both being large),

follows from Kelvin's

principle of stationary phase ( 8*2) that the important part of the path of

integration

the part on which


consideration,

is

small

now, on this part of the path,

sin 6 is approximately equal to

\ 0*.

It is inferred that, for the values of

x and n under

Jn (*0 ~ - I'gos (n0


ffJQ
f

-x0 + %x0) d0

OS -

"cos (n0

-x0 +

\x0*) d0,
und Phys.

ir J a
* Phil.

Mag.

(6) xix. (1910),

pp.

247249

see also Erode, Archiv der Math,

(3)

xxiv. (1916), pp.

239250.

8*43]

FUNCTIONS OF LABGB ORDER


last expression is

249
It follows that,

and the

one of Airy's integrals (6*4).

when x<n,
(1)

Jn (x) ~ - {-^^-} *1 [
a?

>

and,

when

> n,

(2)

^(^^Ij^iOp^H.^,
{2

where the arguments of the Bessel functions on the right are

(x- n)] }/#*by

The corresponding formula


Nicholson
(3)
;

for

Yn (x)

when

a;

> n was
it is

also found

with the notation employed in this work

r.(.)~-p^(V-i-^-

The chief disadvantage of these formulae is that it seems impossible to determine, by rigorous methods, their domains of validity and the order of
magnitude of the errors introduced in using them.
remedying this defect, Watson* examined Debye's integrals, and discovered a method which is theoretically simple (though actually it is very laborious), by means of which formulae analogous to Nicholson's are obtained together with an upper limit for the errors involved.

With a view

to

The method employed x(~


j/sech o)

is

the following
its

Debye's integral for a Bessel function whose order v exceeds

argument

may be

written in the formf


g(tanha a)
v

J
where
t

(v sech a)

~
1)
is

rx>-\-ni
/

e~XT dw,

= sinh a (cosh w

cosh a (sinh w w),


on
it.

the contour being chosen so that r


If r
is

positive

when we approximate by -\wn sinh a; and when a =


'

expanded in ascending powers of w, Carlini's formula is obtained neglecting all powers of w save the lowest,
all 0, Cauchy's formula of 8 2(1) powers of w save the lowest, \wi
.

is

similarly ob-

tained by neglecting

first

These considerations suggest that it two terms, namely \vP sinh o -

is

desirable to

examine whether the

w3 cosh a,
iirst

may

not give an approximation valid throughout the


integral which

transitional region.

The

we

shall investigate is therefore

\e-**dW,
where
* Proc.

r
Camb.
is

= - \ W* sinh a - W

cosh

a,

Phil. Soc. xix. (1918), pp.

t This w.
B. F.

deducible from 8-31 by

96110. making a change

of origin in the u-plane.

250

THEORY OF BESSEL FUNCTIONS

[CHAP. VIII

positive

and the contour in the plane of the complex variable is so chosen that t is on it. If U + iV, this contour is the right-hand branch of the

W=

hyperbola

tftanha
and

+ Jtf^F

2
,

this curve has contact of the third order

with Debye's contour at the origin.

It therefore has to

be shewn that an approximation to


+ni
/ocexp(Jn-i)

r<x>
I

e~ ZT
oo

dw

is

e'^dW.

-ni

J ooexp( Jirt)

These integrals

differ

by

i/:+/;H-^K
and so the problem
|

is

reduced to the determination of an upper bound


it

for

{d(w W)fdr)
will

j.

And

has been proved, by exceedingly heavy analysis

which

not be reproduced here, that

d (w dr
and so

<

Sir sech a,

ifj>j:H-i*
Hence

<

6-7T

~
ZTTl J ao-ni

e-*r

dw = J_l
ATI J
,

e-*rdW
ex p (- Jiri)

+ -l,
V

where 6l
j

<

1.

To evaluate the
6*4),

integral on the right (which is of the type discussed in modify the contour into two lines starting from the point at which

W = tanh a and making angles + %ir with the real axis. If we write W= - tanh ^e ini on the respective rays, the integral becomes
4-

e*

ni

exp (^ tanh3 a) [ exp {- & v?


Jo
<

- \ v%e$ ni tanh2 a}
a)
|

dl-

e~ ini exp (Xv tanh


Expand the integrands
procedure which
is

exp

\-%v?- |i>e-* tanh

powers of tanh 2 a and integrate term-by- term and we get on reduction easily justified
in

7ri tanh a

exp (1/ tanh3 a)

[7~_j (1/

tanh 3 a) - /j {\v tanh 3 a)],

and hence we obtain the formula


(4)

Jv {v sech a) =
IT

75-

VO

exp

{v

(tanh a

-f

3 ^ tanh a

a)} if
exp
[v

j (

i>

tanh3 a)
a)},

-I-

3#i

i.-

-1

(tanh o

whore d
]

<

1.

This

is

the more precise form of Nicholson's approximation

(1).

8*43]
It can be

FUNCTIONS OP LARGE ORDER


shewn
that,

251

whether \v tanh8 o be small, of a moderate size, or a smaller order of magnitude (when v is large) than the approximation given by the first term on the right.
large, the error is of

Next we take the case x (= v sec 0).

in

which the order v

is less

than the argument

We

then have

H (v sec 0) =
where
T

gvi(tanj30)
:

foo+Hir-p)

Tn

e~*r div,
J -co-ip

= - i sin

(cosh

w - 1) cos
is

(sinh
it.

w - w),

the contour being so chosen that t

positive on

The

process of reasoning already employed leads us to consider the integral

/
where
r

e-^dW,

= -\i W* sin

cos 0,

and the contour in the plane of the complex variable is such that t positive on it. If WsU+iV, this contour is the branch of the cubic

is

(U*-V*)ta,n0 + $V(3U*which passes from

r*)

=
oo

oo t'tan/9
foo+t(ir-/3)
I

through the origin to

exp^Trt.

It therefore has to

be shewn that an approximation to


roeexpini

e-^dw

is

e~XT dW.

The

difference of these integrals is

and

it

has been proved that,

when*
UiT

\ir,

then

d(w- W)
Hence
it

<

127r sec 0.

follows that

-.
TTIJ

e-^dw^.
-*-ip
irXJ- oo-i tan
fi

e -xr

dW + zl^ v

where

& <
j

1.

To

evaluate the integral on the right, modify the contour into two lines

meeting at
real axis.

W=

tan

and inclined at angles


tan

Kir

and

ir

respectively to the

On

these lines, write

W= *

-f

- i tan

+ %&\
If
/3

The important values


8'32 is

of

/3

are, of course, small values.

is

not small, Debye's formulae

of 8-41 yield effective approximations.

The geometrical property

of Debye's contour

which was

proved in

used in the proof of the theorem quoted.

252

THEORY OF BESSEL FUNCTIONS


in

[CHAP. Vin
it is

expand the integrands


found that
e~XT d

powers of tan a

j3,

integrate term-by-term, and

W=f
=

iri

tan # exp (- \ vi tan3 )

[-**

J_j

(4 v tan8

+ e**' Jj (^ v tan
3

#)]
3

7re<rttan

^ exp (- J i* tan
it is
3

#j<'>

(^ tan

).

On
(5)

equating real and imaginary parts,

at once found that

Jv (p sec /3) = tan cos {* (tan /3 - tan -)}. [J"_j + /j]


+ 3 -* tan sin
[v

(tan

(6)

F (i/ sec )

= | tan /3 sin

{i>

- 3"* tan cos


;
! \

{*/

- tan - )} [J. j - Jj] + 240 (tan - tan - )} [J"_ j + J] (tan - tan -)}. [7"_ - </j] + 240 M
3
.

2 /i/,

where the argument of each of the Bessel functions J>. on the right is %v tan 3 and and are both less than 1. These are the more precise 2 3 forms of Nicholson's formulae (2) and (3); and they give effective approximations except near the zeros of the dominant terms on the right.
|
|

It is highly probable that the

upper limits obtained

for the errors are

largely in excess of the actual values of the errors.

8*5.

Descriptive properties* of
integral,

Jv (vx) when <

a?

< 1.

which was obtained in 8*31(1) to represent an asymptotic expansion of the function. v But the contour integral is really of much greater importance than has hitherto appeared for an integral is an exact representation of a function, whereas an
(v sech a)

The contour

was shewn

in 8*4 to yield

asymptotic expansion can only give, at best, an approximate representation.

And
is

the contour integral (together with the limiting form of

it

when x =

1)

peculiarly well adapted for giving interesting information concerning


v is positive.

Jv (vx)

when

In the contour integral take v to be positive and write


/

= log {reie

so that

u = log r,

},

= d.
selected,

With the contour

x sinh w w is equal to its conjugate complex, and the path of integration flexion in the real axis. Hence
1

is its

own

re-

roo+n-t

(vx)

= rJ
i __
rw
I

e"

,a

" inh -"

dw

gv(as8inhww) fly^

ttJo
*

The

results of this section are investigated in rather greater detail in Proc.

London Math. Soc.

(2) xvi. (1917), pp.

150174.


8-5]

FUNCTIONS OF LARGE ORDER


find that the equation of the contour is

253

Changing the notation, we

20

r
so that

sin

and,

when

this substitution is
,

made
2

for r, the value of

(w x sinh w)
a 2

is

log

e + y/ifr-X* sin 0)
-.

x sin

.a a at cot y (tr
.

a;

sin 2 0).

*m

This last expression will invariably be denoted by the symbol* F(0,x),


so that

(1)

J {vx)=^{'<r'>de,
(a procedure

and by differentiating under the integral sign


justified) it is (2) K '
/o\

which

is

easily

found that
l \ Ti J (vx) =
w
f

e- *<.*<

fib*

ttJo

x */(&* x* sm* 0)

- Ja
fl- a'Binflcosfl .

dv.

This

is also easily

deduced from the equation


1

raa+iri
.

JJ (vx) =

(x*inhu>-w)

sinh

^
it

Before proceeding to obtain further results concerning Bessel functions,


is

convenient to set on record various properties! of F(0,x). The reader will

easily verify that

<

3>

m^'^

w
so that
(5)

4^.,-^a^
fl-g'sinflcosfl

^-M)*^-^' ^
6
n

F(0,x)>F(O,x)>F(O,l) = O;
d
V(<* \

and

also

ia\

Next we
(7)

shall establish the

more abstruse property


a? sin 2 0)/V(l

F(0, x) >F(0,x) + $ {& it,

+ x>).

To prove

we

shall first
i

shew that
O aPsmO cos 6
..,
.

t It is

This function will not be confused with Schlafli's function defined in 4-15. supposed throughout the following analysis that O<.c^l,O^0<ir.

254
It is clear that

THEORY OF BESSEL FUNCTIONS


g(0,x)

[CHAP. VIII

= s'(\-x*)<<J(\+x*),

g(Tr,x)=\
so that, if

<J(\+x%
0,

g (0, x), qua function of

attained its greatest value at


2

or

tt,

that value would be less than


greatest value

^(1+x
O

).

If,

however, g(0,x) attained

its

when
1

had a value

between
2

and
O

ir,

then

a;

cos

2fl
2

_ (0 - x
O 2

si n

cos
o

f_

(tfo'-^sin ^)*

(0o -a?sin 2

)i

'

and therefore

g {0, x)^g (8
so that, no matter

x)

= V(l - a? cos 20 ) ^ V(l + *),


o

where g(0,x) attains


2
)-

its

greatest value, that value does

not exceed ^(l

+#
a

Hence

^W-*w^^
1

sin

cos
2

V(l+a )

and

so

whence

(7) follows at once.


is

Another, but simpler, inequality of the same type


(8)

F (0, x) > F(0,


this,

x)

+ * 0* V(l - *").

To prove

observe that

dF
e f

'

X)

>

\/(0*

- " sin
is

0)

> V(l - &),

and integrate

then the inequality

obvious.

From

these results

we

are

now

in a position to obtain theorems concerning


v.

J (vx) and

JJ (vx)

qua functions of

Thus, since
OV

IT JO

the integrand being positive by

(5), it follows

that

creasing function of v, in like manner,

JJ {vx)

is

Jv (vx) is a positive dea positive decreasing function

of v.
Also, since d {**<*>

JAva:)\ = _ 1

n
(

SF{0> x)

OV

_ F ^ 0> x)}

*-*>. *>-"*.*>

d0 <0,

7T.' o

the integrand being positive by

(5), it follows
e"

that e vF({t x
'

>

Jv (vx) is a decreasing

function of v; and so also, similarly, is

F,0,x)

'

(vx).

8*51]

FUNCTIONS OF LARGE ORDER

255

Again, from (8) we have

Jv ( vx) ^
<
so that

Jo

exp {- i v0- V( 1

- a )} dd
8

eF(0,x) roo

exp {- i/0\/(l
p -vF(0,x)

- x*)\ d0,

(9)

<L(vx)^
(1
is

x*)l

^(Zttv)'

The

last expression

easily reduced to Carlini's

approximate expression
is

( 1*4,

811)

for

J(px); and so Carlini's expression


v.

always in error by

excess, for all

* positive values of
result for
0*

The corresponding
and replace G(0,x) by

JJ {vx)
2

is

derived from

(7).

Write

- a? sin

=(#,#),

for brevity.

Then

2xj;{vx)=
fi

/V-.*
fir
I

dG ( >^ {G(0,x)}-id0

-vF(0,x)
7T

expi-^GyVa+a^.G-MG
.'o

e -F(Q,x) fao
IT

Jo
(vx)

Texp {- J^(l+^)j
^ -'*<. )
2 a? ) (

G-idG,

and so
(10)

xj;
factor

+ *)VV(2 w).
is

The absence of the

^(1

from the denominator

remarkable.

It is possible to prove the formula f

(11)
J

*<*>*~ (2ir*y (1 -*)

+ V (l

-W-

in a very similar manner.

This concludes the results which we shall establish concerning a single


Bessel function whose argument
is less

than

its order.

8'51.

Lemma

concerning F(d,x).

We
(1)

shall

now prove
dF{0,x)

^ x ^ 1 and ^ 0-^sintfcos0 ^, ^ A >0. je--{F(0,x)-F{O,x)}


the
that,

lemma

when

< ir, then

y,^^
is

The lemma

will

be used immediately to prove an important theorem con,/

cerning the rate of increase of


* It is

(vx).
in error by excess for sufficiently

evident from Debye's expansion that the expression


v.

large values of

t Cf. Proc.

London Math.

Soc. (2) xvi. (1917), p. 157.

256
If

THEORY OF BESSEL FUNCTIONS


V(^ - a
2

[CHAP. VIII

sin2 0)

= H{0, x), we

shall first prove that


,

dF(0,x) dH(0,x) d0 d0 /
is

a non-decreasing function of

that

is
i

to say that

(\-0cot0)2 +

-a?sm*0

a? sin
is

cos

a non-decreasing function of

0. is

The

differential coefficient of this last function of

(0 - x* sin

cos 0)-* [(0* cosec2

- 1 - sin 0) (1 - x*) + 2 (0 cosec - cot cosec - % sin* 0) (1 - ar ) + 2a- (1-0 cot 0) (0 cosec - cos 0) + sin 0(12 2 2

2 2

a,-

) ],

and

every group of terms in this expression is positive (or zero) in consequence

of elementary trigonometrical inequalities.

To

establish the trigonometrical inequalities,


(i)
(ii)

we
1

first

observe that,

when

< 6 ^ *r,

d+sindcosd-2^sin

sin

d>0,

6 + sin 6 cos 6 - 26* cot 6

> 0,
positive differential

(iii)

- 6 cos - sin 3 0^0,


wheu 60 and have the
(iii)

because the expressions on the left vanish


coefficients
(i)

2(cos0-0 -1 sin0) 2

(ii)

2 (cos 6 - 6 cosec 6)\

sin 6 (6

- sin

6 cos 6),

and then
2

cosec2 6 - Q3 cot 6 cosec 2

sin 2 6
)

= (6* cosec2 6 - 1) (1 - 6 cot


2

+ cosec

(sin

- 6 cos - J sin 3 6) ^ 0,
cos 6 - sin 3 6)

cosec2

0-l-$sin 2
(5

= 6 cosec2

+ sin 6 cos - 2d"

sin 2 0)

+ cosec 6 (sin

>

0>

so that the inequalities are proved.

It has consequently been

shewn that

(}><>
where
the. variables are

understood to be
0.

and

x,

and primes denote

differ-

entiations with regard to

It is

now

obvious that

ifiM-*{S>*
and,
if

we

integrate this inequality from

to 0,

we

get

Since F' and /7/i/' vanish

when = 0, F>(0,x)H{0x) _
if (0, #)

this inequality is equivalent to

>Q

and the truth of the lemma becomes obvious when we substitute the value
of

(0,

x) in the last inequality.

8*52]
8*52.

FUNCTIONS OF LARGE ORDER


The monotonia, property of Jv (vx)jJv {v).

257

We
is fixed,

shall

and

now prove a theorem of some importance, to the effect that, if x ^ x ^ 1, then Jv (vx)/Jv (v) is a non-increasing function of v, when

is positive.

be valid only when S^.r^l, (where 8 is an some expressions introduced in the proof contain an x in their denominators; but the theorem is obvious when ^ x < 8 since e vF > z Jv (vx) and e -vF{o, x)/jv () are non-increasing functions of v when x is sufficiently small moreover, as will be seen in Chapter xvn, the theorem owes its real importance to the fact that it is

[The actual proof of the theorem

will

arbitrarily small positive number), since

>

true for values of

x in

the

neighbourhood of unity.}

It will first

be shewn that

(1

Jv{vx) *M"*)_ dJ*W dJAvx)


Ovox
ox
Cv
establish this result,

To

we

observe that, with the usual notation,

and,

when we

differentiate

under the integral sign,

OV

7T

L\/^
W

> <*

<* *))-*
>

or

' e vn$,x)

de
"

\\C((l -rVh

dFie X ^

IP/A \in(A

^W^1^L

A,)

x e-J?ie,x)d0,
if

we

integrate by parts the former of the two integrals.


it

Hence

follows that

where

by using the inequality F(yfr, x) > ^(0, x) combined with the theorem of 8*51.

'

258
Since CI
that
is

THEORY OF BESSEL FUNCTIONS


(0,
yfr)

[CHAP. VIII

is

not negative, the repeated integral cannot be negative


that

to say,

we have proved

J
so that

( vx

^H^} _ dJAvx) dJ (vx ) > "


dvdx
dx
dv

'

\dJv (vx)

Integrating this inequality between the limits x and

1,

we get

~dJv (vx)
so that

,"|*

dJv (vx)
dv

r -I J v (VX)

*Z

dJv (v) J* (") fo~ J


may be
written

Since

(vx)

and

(v) are

both positive, this inequality

in the form
(2)

^{J
v.

(vx)/Jv (v)}^0,
to

and
is

this exhibits the result

which was

be proved, namely that

(vx)jjv (v)

a non -increasing function of

8*53.
If,

Properties of Jv (v)

and

J '(v).
v

for brevity,

we write F(0)

in place of F(0,

1),

so that

(1)

F (0) ^ log +
J
v

^
v

~ Sin2 0) - cot J V( are

^n*

B),

the formulae* for

(v)

and

J '{v)

The

first
5

4^/(9 V3)

and we

is term in the expansion of F (6) in ascending powers of shall prove a series of inequalities leading up to the
s

result that

F (0)1

is

a non-decreasing function of

0.

We

shall first

shew that

To prove

this

we observe
6-*

that
cot *)/*}
sin* 0)

*"(*>

IP-* ~ ^~ V(^ -

^
*

Vi &2

,,

_ siR2 ,.

'

* It is to

be understood that

JV (v) means

the value of

dJv (x)/d.v when x has

the particular

value

v.

^
FUNCTIONS OF LARGE ORDER
d^
1

8-53]

259

and that
j

dd
d_
f

-0 cot 0) _ ~
fr~~~)
a

fl

cosec 2

fl

+ flcotfl-

2
'

p
"'"

vW-sin
0*

fl))

d&\
Hence
it

= _ (ffcosec'fl + flcot 0-2) sin "^VO^^-sin^) J"


. ,

2
'

follows that

d (F'(0)}
<** I

#- | - * (* - sin

0-sin0cos0
2

W <*

//te

C0Sec2 *

0/1

+ ecot0 ~
x (0 + sin

v 2>
cos

- 20* cot 0)

by inequalities proved in
Consequently
(3)

851.

0F"(0)-2F'{0)>Q,
to say

that

is

^ [OF'

(0)

3F (0)} > 0.
to

If
(4)

we integrate

this inequality

from

we get

0F'(0)-SF(0)>O,
F(0)/03 should be a non-decreasing function of
0.

and

this is the condition that

It follows that

*m
\

9V3'

and therefore

(v)< -

exp j-

(*
"SIalways in error by
is

<

ii."

9lp |-

T(i)

2*3**'
so that Cauchy's approximation for
/

(v) is

excess.

An
(5)

inequality which will be required subsequently


2
(tf*

- sin 2

8)

F'

{&)

- 3 (0 - sin 6 cos 0) F{6) >

0.

The truth

of this

may be

seen by writing the expression on the

left in

the form
%

{&*

- 2 sin 2 6 + 6 sin 6 cos 6) F' (0) + (0 - sin 6 cos 6)


is

{BF'

{$)

- ZF{fi)\

in

which each group of terms


[Note.
(6)

positive

(cf.

8-51).

A formula resembling those which


l

have just been established

is

Jv

( vt )

dt

~ J- - T

see

ZVtiV.

Mag.

(6)

xxxv. (1918), pp. 364370.]

260
8*54.

THEOBY OF BESSEL FUNCTIONS


Monotonic properties of Jv {v) and JJ
(i>).

[CHAP. VIII

It has already

been seen
v.

( 8'5)

decreasing functions of

It will

and JJ (v) are now be shewn that both v*Jv {v) and v*Jv '(v)
that the functions
v

(v)

are steadily increasing* functions of

v.

To prove the
dv

first

result

we

observe that

dj^ioi _

Sir J

r
o
|_

e vFi e )de

^r
tJo
f
./

F(d)e-.F W dd

= *Ll \$e -FW~Y + il


07T

'
o

Jo

oTT

fl

'

(0)

_ 3F(0)}

e~' F

^ dd

>0,
since the integrated part vanishes at each limit
positive.

and

8*53) the integrand

is

Hence v*Jv (v)


(1)

is

an increasing function of v; and therefore

v*J9 {v)< lim {v*J9 (i>))


it

= r(J)/(2*S*ir)- 0-44731.
noted that
2/8 (8) =0-44691.

In connexion with this result

may be

J
To prove the second
o7T

(1)

= 0-44005,

result,

by following the same method we

find that

dl'

|_

JO

>o,
by

8*53 (5), and so

v*Jv'(v)

is

an increasing function of

v.

Hence
(2)
v*

J: (v) < lim

{*i

JJ (v)} - 3* T (f )/(2*w) = 0-41085.

It is to

be noted that

J
8*55.

'

(1 )

= 0-32515,

4J6

'

(8)

= 0-38854.

The monotonic property of v*JJ {v)jJv (v).


is

A
is

theorem which

slightly

more recondite than the theorems just proved


{!*/, (*)]

that the quotient

W(V)} +
15

a steadily increasing function of v.


* It is

as v-*-<z>

not possible to deduce these monotonic properties from the asymptotic expansions. If, /(*)~ </>(), and if ${) is monotonic, nothing can be inferred concerning monotonic

properties of f() in the absence of further information concerning/^).

8-54, 8-55]

FUNCTIONS OF LARGE ORDER


use the integrals already mentioned in

261

To prove this result we 854 for the four functions

853,

U,
place of
0,

vKTAv\

Ti/

lv
to be
>/r

Taking the parametric variable

in the first

and third integrals

in

we

find that

" dv {v*J (y)

ir

J o J o

where
n,
(0,

t) =

r (0> </(*-

si,,'

0)

^01 ^rtf F <*>


V(0"-sin'0)

uT

yT '

KY "

>

^^
;

{* (*) - *<*)) -

^#

+*" <*> - *<+>

by 8'51. The function fi x (0, i/r) does not seem to be essentially positive (cf. 8*52) to overcome this difficulty, interchange the parametric variables B and yjr, when it will be found that

Now, from the inequality just proved,

^V^-sin'^^-s^
+

^ + VrgJnVrC0B^-2gin'Vr U
^V(^
2

-sin2 ^)

_
'

"

fl

+ flsinflcosl-2sinl
&)/(&- sin* 0)
l5r
J

vr '

vr/;

Since 0"1 ^(0*


of
6,

- sin

0)

and

&F'(0)

- F(0) are both ( 8'53) increasing functions

term in the sum on the right are both positive or both negative; and, by 851, 853, the second and third terms are both positive. Hence fl! {6, yfr) +- llj (yfr, 6) is positive, and therefore
the factors of the
first
x
'

dv

Jv {v)

which establishes the result stated,

262
8*6.

THEOBY OF BESSEL FUNCTIONS

[CHAP. VIII

Asymptotic expansions of Bessel functions of large complex order.


results obtained
{x)

and

(8'31 8*42) by Debye in connexion with J (x) where v and x are large and positive were subsequently extended* to the case of complex variables. In the following investigation, which is, in some respects, more detailed than Debye's memoir, we shall obtain asymptotic

The

expansions associated with


It will first be

(z)
|

when

and z are large and complex.

supposed that arg z\<\Tr, and we shall write


v

= z cosh 7 = z cosh (a + i/3),

where a and # are real and 7 is complex. There is a one-one correspondence between a + ifS and vjz if we suppose that ft is restricted to lie between^ and 7r, while o may have any real value. This restriction prevents z/v from lying between 1 and 1, but this case has already ( 8'4) been investigated.

The

integrals to be investigated are

HJV

(z)

TTIJ

e-^ w

>

dw,

H
where f(w) =

(z)

=-

e-'/M

*.'-

dw^-~\ mJ
is

_.+

*/(* dw,

w cosh 7 sinh w.
at 7,

stationary point of the integrand

and we

shall therefore in-

vestigate the curve whose equation is

If(w) = //( 7 ).
If

+ iv, this equation may be written in the form (v - fi) cosh a cos & + (u a) sinh a sin - cosh u sin v + cosh a sin /8 = 0.
replace

we

by u

The shape
{(u

of the curve near

(a,

#)

is

- ay - (v -fif} cosh a sin /3 +


\tt

2 (u

- a) (v - 0) sinh a cos /3 =
yS),

so the slopes of the

two branches through that point are

arc tan (tanh a cot

i if +
creases as

i arc tan (tanh a cot /8),


;

where the arc tan denotes an acute angle, positive or negative Rf(w) inw moves away from 7 on the first branch, while it decreases as w moves away from 7 on the second branch. The increase (or decrease) is steady, and Rf{w) tends to + 00 (or - x ) as w moves off to infinity unless the curve
has a second double-point t.
* Mnnchener Sitzungsberickte, xl. [5], (1910) ; the asymptotic expansions of Iv were stated explicitly by Nicholson, Phil. Mag. (6) xx. (1910), pp. 938943. f That is > say 0<yS-cw. X As will be seen later, this is the exceptional case.
'

(x)

and

A" (x)

8-6, 8-61]

FUNCTIONS OF LARGE ORDER

263

If (i) and (ii) denote the whole of the contours of which portions are marked with those numbers in Fig. 19, we shall write

S,v

(z)

f
.

-*/'">
(i)

dw,

TTI

S* (z) = -.( 7Tt


J

e?fM dw,

(il)

and by analysis identical with that of 8 41 (except that i# is to be replaced by 7), it is found that the asymptotic expansions of Sv w {z) and 8^ (z) are
given by the formulae
( i)

s.{.)~
s' a,
.

m
V ;

2)

W ~ ^_
\

\'(- \ viri tanh 7) g~ y(tanhy "

/7
^

h> :"- 1 "-

TO

s ~

r( ",+ * )

T ()
j)
o

'

(h

A"
tanh 7) m
'

yHW

tanh y))

r(t + s i

am

r (i)

_ ^ y tanh 7)w (

= arg z + arg ( i sinh 7), and the value of arg ( i sinh 7) which lies between \tr and \ir is to be taken.
where
arg ( | i/7ri tanh 7)
(i)

Fig. 19.

The values of

A A
,

1}

A.2}

...

are
1

(A =l,
(3)

A = $-ji coth*y, * = ih ~ Mr coth* 7 + Aft coth4 %

n) remains to express v w (z) and (z) in terms of v and to do this an intensive study of the curve on which

It

Sv n)

(z)

and #

( -'

(*)

//()- //(7)
is

necessary.

8'61.

The form of Debyes contours when

the variables are complex.


is

The equation
(1
)

of the curve introduced in the last section

(v

/8) cosh o cos y8 + (u a.) sinh o sin fi cosh u sin v + cosh a sin /8 =
are current Cartesian coordinates and
is

0,

where

(u, v)

<

/S

<

it.

Since the equation

unaltered by

a change of sign in both u and


;

a,

we

shall first study the case in

which a ^ and since the equation is unaltered and it # are written for v and #, we shall also at first suppose that 0</9^^7T, though many of the results which will be proved when y8 is an acute angle are still true when /9 is an obtuse angle.

when

it


264

THEORY OF BESSEL FUNCTIONS


For brevity, the expression on the
d<f>(u,v)
left in (1) will
.

[CHAP. VIII
<f>(u, v).

be called
v,

Since

= sinh
.

o sin

p sinn
.

u sin

it

follows that,

when
u,

v is given, d()>/du vanishes for only

one value of

u,

and so

the equation in

<f>

(u, v)

= 0,
is infinite

has at most two real roots


Of
7T.

and one of these

whenever v

is

a multiple

When

<v<

7r,

we have*
,

<f>

<(+ oo ,v) = oo (a, v) = cosh a {(v /3) cos /3 sin v + sin /3} ^ 0,
<f>(- oo

v)

oo

and so one root of the equation


than a and the other

in u,
< (u,

v)

= 0,
a.

is less

is

greater than

both becoming equal when v =

/3.

By
it is

considering the finite root of the equations


<

(u, 0)

= 0,
0, and

<f>

(u, it)

= 0,
a,

seen that, in each case, this root


as v tends to
ir is

is less

than

so the larger root tends

to

+ oo

or to

ir

for values of v just less

than

or just

greater than
of the curve

the equation

<f>

(u, v)

has a large negative root. The shape


lines in Fig. 20.

therefore roughly as

shewn by the continuous


v lies

Next consider the configuration when


TTi

between

and

ir.

\
)

<z
y-2iri
Fig. 20.

^
TTi

When

is

/3,

d(j>(u,

v)/du vanishes at

= a,

and hence <f>(u,@) has a

minimum
at

value
2 cosh a sin
13

(1

/Q cot y8 a tanh a)

u o. There

are

now two
1

cases to consider according as

yS cot j8 o tanh a
4> (a,

is (I)
*

positive or (II) negative.


d<f> (a,

Since

u)/3t'=cosha(cos/3-coBv), and this has the same sign as v-/3,


at v

r)

has a

minimum value zero

= .

8-61]

FUNCTIONS OF LARGE ORDER


of values of the complex
1

265

The domains

y=a

ift for

which

>8 cot ft a tanh a


are

is

positive (in the strip

0^#<7r)
complex

numbered

1, 4,

5 in Fig. 21

in the

domains numbered responding domains


in Fig. 22.

2, 3, 6a, 66, 7 a, 76 the expression is negative; the cor-

for the

viz

cosh (a 4- ift) have the same numbers

TTl
-1>

7b
1

66
5

la
3
2

6a

Fig. 21.

Fig. 22.

(I)

so that the curve never crosses the line v


is

that

When 1 ft cot ft a tanh a is positive, (a, ft) is essentially positive, = ft. The only possibility therefore the curve after crossing the real axis goes off to 00 as shewn by the
<f>

upper dotted curve in Fig.

20.
<f>

(II) When 1 ft cot ft a tanh a is negative, the equation ( a, v) = and ft 27r, for has no real root between cos v). o<f>( a, v)/dv = cosh a (cos Therefore has a ( a, v) single maximum at #, and its value there is so ( negative, that and $ 27r. a, v) is negative when v lies between
<f>
<f>

Also

(f>

(u,

/S

2tt)
<j>

has a

maximum

at u

a, and its value

there

is

negative,

so that the curve

(u, v)

does not cross v


00

= /S 2<rr
as

hence, after crossing

the real axis, the curve must pass off to

in,

shewn by the dotted curve

on the right of Fig. 20.


This completes the discussion of the part of the curve associated with
S, m (z)

when a>0, O<0^$tt.


to consider

Next we have
line v

what happens

to the curve after crossing the

= + ir.
<f>

Since
the line
<f>

(a, v)

= cosh a {(v (3) cos ft sin v 4- sin ft},


is

and the expression on the right

positive

when

^ ft,

the curve never crosses

u~a;
(u,

also

mr) = (u a) sinh a sin ft + (mr

ft) cosh a cos ft + cosh a sin ft,

266

THEORY OF BESSEL FUNCTIONS


this is positive

[CHAP. VIII
which go
off to

and

when u >
lie

a,

so that the parts of the curve

infinity

on the right must


1

as shewn

in the north-east corner of Fig. 23.

When
i.e.

- a tanh a + (ir - ft) cot ft > 0,

when

(a, ft) lies in

found that the curve does not cross v


v

= 7r

passes off to

any of the domains numbered 1, 2 and 3 in Fig. 21, it is = 2v ft, and so the curve after crossing ac + iri as shewn in Fig. 23 by a broken curve.

Fig. 23.

We now
numbered 6a

have to consider what happens when


in Fig. 21.
1

(a, ft) lies

in the

domain

In such circumstances

- a tanh a + (?r - ft) cot ft <


at v

and
u

^ ( a,

v)

has a

maximum

negative.

The

curve, after crossing v

= 2tt ft, the value of ( a, 2ir ft) being = tt, consequently remains on the right of

=a

until it has got above v


<f>

= 2rr ft. - ft),


it

Now

( a,

v) is increasing in the intervals (2tt

(/3,2rr-0),
let

+ ft,

4tt

(4>Tr

+ ft,

6ir

- ft),

. .

the

first

of these intervals in which

becomes positive be

(2Mtt

+ ft, 2Mtt +2-7T- ft).


minimum
at u

Then
positive,

<f>

(u,

2Mir +

2ir

ft)

has a

therefore go off to infinity on the

and so the curve cannot cross the line left, and consequently goes

o, at which its value is t>= 2Mir + 2tt ft; it must


to

-oo +(2M+l)iri;
it

cannot go to infinity lower than


line in

this, for

then the complete curve would

meet a horizontal

more than two

points.

8-61]

FUNCTIONS OF LARGE ORDER


(a, ft) is in

267

When

6a, the curve consequently goes to infinity at

-oo +(2M+l)Tri,
where

is

the smallest integer for which

tanh o +

{(M

1) tr

/9}

cot

is positive.

We can now construct a


for & (1) (z)
(2)

table of values of the end-points of the contours

and $ (z), and thence we can express these integrals in terms of (z) and (z) when (o, fi) lies in the domains numbered 1, 2 and 6a in v v Fig. 21 and by suitable reflexions we obtain their values for the rest of the complete strip in which < yS < ir. The reader should observe that, so far as the domain 1 is concerned, it does not matter whether /S is acute or obtuse.

{1

>

(i)

If

is

the smallest integer for which

l-o tanh a +
is

{(M

1)

ir

} cot

fi

positive

when

cot

is positive,

and

if

is

the smallest integer for which

l-o tanh o - (Nir + ) cot


is

positive

when cot#

is

negative, the tables of values of

Sv

(1)

(z)

and Sv

(z)

are as follows

Regions

End-points

S0>( 2 )

1,3,4
2,6a
5,

00
oo

iri,
,

76

- oc

+ 7Tl + iri - oo + 2m'


OC

HW(z)
2J
(z)

oo

66 7a

oo
-oo,

2Niri, oo +- iri

2e-^iJ_ v z ) X*iH e v ){ze-if*i)


{
i

cc

+ (2M+l)iri

e-Mvniffv V)(ze-Mni)

Regions

End-points

SyW (z)

1,2,5
3,

oo + iri, oo + ir,
oo

oo

#<*>(*)
iri

7a

oo
oc

2J.(z)
Zevwi

4,66

+ 2ri,
+
(2J/

J_ v ( 2 )
( ze

6a
76

-ao

+ l)ri,
oo

oc

eMvwi ffjM (zeMni)

ao + rt,

2Niri

e-Wvwi ffJP)

Nwi)

From

these tables asymptotic expansions of any fundamental system of

when v and z are both arbicomplex numbers, the real part of z being positive. The range of validity of the expansions can be extended to a somewhat wider range of values of arg z by means of the device used in 8*42.
solutions of Bessel's equation can be constructed
trarily large

268
The reader
pass from

THEORY OF BESSBL FUNCTIONS


will find it interesting to

[CHAP. VIII

oo
1

to oc

+ iri

and from - oc

prove that, in the critical case /3=r, the contours + ni to oo , so that the expansions appropriate to

the region

are valid.
differences

Note.

The

between the formulae

for the regions

regions 7a and 76 appear to have been overlooked by Debye,


Soc. xcv. A, (1918), p. 91.

da and 66 and also for the and by "Watson, Proc. Royal

8*7.

Kapteyris inequality for

Jn (nz).

extension of Carlini's formula (811, 8'5) to Bessel coefficients in which the argument is complex has been effected by Kapteyn* who has shewn that, when z has any value, real or complex, for which z- 1 is not

An

a real positive numberf, then


,g

(1)

\Jn(n*)\*

exp{w\/(l--g2 )}

{l

+ V(l-* )} n
a

2 This formula is less precise than Carlini's formula because the factor (2jrn)* (1 -s )* does not appear in the denominator on the right, but nevertheless the inequality is sufficiently powerful for the purposes for which it is required %.

To obtain the

inequality, consider the integral formula

Jn {nz) = ^.
in

r
J

exp { \nz
w
,

(t

1/t)} dt,

which the contour

is

circle of radius e

where u

is

a positive number to

be chosen subsequently.
If

we

write

eu+i8 we
,

get
[n [\z

Jn (nz) Now,
if

f exp
|

(eV - e^e~ i9) -u-

id}] dO.

M be the maximum value of


exp {\z (eV*

- e- u e~

i0

-u- id]

on the contour,

it is

clear that

\jn (n*)\*M:

But

if

- pe ia

where p

is

positive

and a
u e-*)

is real,

then the real part of

\z (eu eu>
is

\p
this attains its

w {e cos (a

- er -u-id u cos (a e~ + 0) $)} - u,


when

and and
*

maximum

value

tan 6
its

coth u tan a,
+
sin2 a)

value

is

then
s p V(sinh u

- u.

Ann. Sci. de VEcole norm sup. (3) x. (1893), pp. 91120. when z approaches the real axis it follows that the t Since both sides of (1) are continuous sign may be given to the inequality is still true when * - 1 is positive: for such values of z, either radicals according to the way in which z approaches the cuts.
X See Chapter xvii.

' '

8-7J

FUNCTIONS OF LARGE ORDER


for all positive values of u,

269

Hence,

Jn (npe

im

< exp [np

v/(sinh 2

+ sin8 a) nit].
may be
by
as small as
this

We

now choose u

so that the expression on the right

possible in order to get the strongest inequality attainable

method.

The expression
2 p V(sinh u

sin3 o)
is

has a

minimum, qua function

of u,

when u

chosen to be the positive root of


1

the equation*
sinh u cosh u
\/(sinh 2

-f

sin* o)

With

this choice of u it

may be proved
s
"8

that

2 V(l
and,

sinh u cosh u

(cosh 2

g2la ),
must be taken
in

by taking z

to be real,

it is

clear that the positive sign

the ambiguity.

Hence
2 {1

V(l
%
!

- z*)) sinh u cosh


,

u=

e2"

.
|

*,

and

so
z

exp

\/(l

log

T+

-z) _

2 V(sinh a u

+
I

sin2 a)
e
3*

exp
I

\/(l

-z)
a

s \~ V(l - z*) V(i-* )

- e**

= 5V(l-z')-
sinh 2 m

sin2 a

sinh u cosh

w
sin 2 a)

= p \/(sinh 2 u +
and
it is

u,
j
j

now

clear that

Jn (nz) $
I

y rexpV(l-^) fcexp V( 1-^) |"

An
as both

interesting consequence of this inequality


j

is

that

Jn (nz)

1 so

long

'.

and
I

s exp

-v/(l

z')
is satisfied,

To

construct the domain in which the last inequality

write as

before z

= pth,

and define u by the equation


sinh u cosh u

_
a)
"~

V(sinh 2 w

+ sin

The previous

analysis shews at once that,

when

gexp y'(l l+y/(l-Z>)

z*)
2

1'

then
*

2 p \/(sinh a

+ sin

a)

u = 0.

This equation

is

a quadratic in sinh 3 u with one positive root.

270
It follows that

THEORY OF BESSEL FUNCTIONS


2u
sinh 2 m
'

[OHAP. VIII

P*

sin*

sinh u (u cosh u sinh


sin2 a increases from
is

u).

As u

increases from

to

T1997

...,

to 1

and p dea

creases from 1 to* 0-6627434....


inside

It

then clear that

* ex PV(l-^2)
|

+ V(i-* )

$1

and on the boundary of an oval curve containing the

origin.

This curve

Fig. 24.

The domain
it

in

which

J*n (nz)

certainly does not exceed unity.

is

shewn

in Fig. 24

will

prove to be of considerable importance in the

theory of Kapteyn series (Chapter xvn).

When the order of the Bessel function is positive but not restricted to be an integer we take the contour of integration to be a circle of radius e* terminated by two rays inclined + it arc tan (coth u tan a) to the real axis. If we take 1 = ev on these rays, we get
1 1

, ,, T \W)\<X'+\\) % **l-*P
,

\amvir\

/""

cosh (u + v) cos 2a cosh (v-u)

\/(sinh^ + sin^)

, - vV d

"1

j^/l +
[

s
|
I

2!L!^|

w
\\
I

r
J u

e xv{-v(v-u)}dv\
)

and so
|Jr(v*)|< \l

+
is

l/it

pex P V(l- g) )' 1 + ^(1 -*) j

This value

given by Plummer, Dynamical Astronomy (Cambridge, 1918), p. 47.

CHAPTER

IX

POLYNOMIALS ASSOCIATED WITH BESSEL FUNCTIONS


9*1.

The

definition

of Neumanns polynomial

On (t).

is the discussion of certain polynomials which occur in various types of investigations connected with Bessel functions.

The

object of this chapter

The

first

of these polynomials to appear in analysis occurs in

Neumann's *

investigation of the problem of expanding an arbitrary analytic function f(z) into a series of the form 2an n (z). The function n (t), which is now usually

called

Neumanns
l/(t

polynomial,

is

defined as the coefficient of

en

Jn (z)

in the

expansion of

z)

as a series of Bessel coefficients f, so that

W
From

t=z~ Jo (Z) (<) + 2Jl (2)


- S
n~0
this definition
en

(<)

+ 2J* < 2 >

* <*)+

Jn (z)On (t).

shall derive an explicit expression for the function, and it will then appear that the expansion (1) is valid whenever \z\< \t\. In order to obtain this expression, assume that z < t and, after expanding l/(t - z) in ascending powers of z, substitute Schlbmilch's series of Bessel coefficients ( 27) for each power of z.
j \ j j

we

This procedure gives

- =- + t-z
1

2
s=1

<

** +1

7*

1
fr

s
m=0

(,\a.

v
*=1 &

v
=

(s

+ 2m).(s + m-l)l
nl
-

(,i

Assuming
*

for the

moment that the repeated series is absolutely convergent*.


Function en (Leipzig, 1867), pp. 8

Theorie. der Bessel' sch en

Math,
series.

lxvii. (1867), pp. 310

314.

Neumann's procedure,

derive the differential equation which will

15, 33 see also Journal fur assuming the expansion (1), is to be given subsequently ( 9 12) and to solve it in
;

after

t In anticipation of 1611, we observe that the expansion of an arbitrary function by substituting for l/(t - z) in the formula
. ,= f(z)r

is

obtained

/(*+)/()"
(

si/
1st.

+ Cf. Pincherle's rather more general investigation, Rcndiconti R.


pp.

Lomlardo,

(2)

xv. (1882!,

224225.

272

THEORY OF BESSEL FUNCTIONS


a rearrangement by replacing
;

[CHAP. IX
series

we
is

effect

by n 2m, and the rearranged

a series of Bessel coefficients

we thus get
(<i(-D 2"- 2"1" 1 * n^ , n_ m+1
V

i r- = 7 c
*

n/(')+

S
n=l

n.(n-m

m=<\

=0

m!

1)1)
J

/.(*)

r/x

Accordingly the functions


/ox (2)

n (<) are defined 1

by the equations

n u\
o,()

^"i,V ml(i^
-iyt

^"n.in-m-l)!
-

{n>l)

(3)
It is easy to see that

(4)

en

O(0*-^iT

|i

2(2n

2)

+ 2i4t(2n-2)(2M _ + ...|, 4)

and the

series terminates before there is

any possibility of a denominator factor

being zero or negative.

We
l/(*

have now to consider the permissibility of rearranging the repeated series for

- 1).

A sufficient condition is
.i
t
i

that the series


.

2* % % (<+2m).(+m-l)i * iTTTi J 1 * TTi ' *

*/
.

U-o

. \ , +2> \'/ If

should be convergent.
(4),

To prove

that this

is

actually the case,

we observe

that,

by

2' 11

we have

m=0

'MI

m=0

(*+

1)!

+ <3(J|*|) -{exp(i|*P)}/(2m)!

<(il*l)*e*P(il*l

)-

Hence
s

_J1_

(+2m).( + m-l)l
,

,,

||

_,.,_,*

||eipQi|)
'

||(|*|-|*|)

The absolute convergence of the repeated series is therefore established under the hypothesis that \z\ < \t\. And so the expansion (1) is valid when \z\< \t\, and the coefficients of the Bessel functions in the expansion are defined by (2) and (3).
It is also easy to establish the uniformity of the

pansion (1) throughout the regions

1
1

^ R,

convergence of the ex\z\^r, where R > r > 0.

9*1]

ASSOCIATED POLYNOMIALS
these inequalities are satisfied, the
x

273

When

sum
( /)*

of the moduli of the terms does not


+2ro

^_

(8

+ 2m).( + m-l)!
"!

exp

.lo^^lio

*"
(*

(fr*) \

exp

(frr )

+ 2m)!
z

Jand
t,

fl-r

Since the expression on the right is independent of convergence follows from the test of Weierstrass.

the uniformity of the

was called by Neumann a Bessel function of the second now used (cf. 353, 3*54) to describe a certain solution of Bessel's equation, and so it has become obsolete as a description of Neumann's function. The function n (t) is a polynomial of degree n + 1 in 1/t, and it is. usually called Neumann's polynomial of order n.

The function

n ()

kind*; but this term

is

If the order of the terms in

\n m
/k\ (5)

or \{n

\) m
/*\

for

in (2), according as

Neumann's polynomial is reversed by writing n is even or odd, it is at

once found that


r\

* *4

n.fjw
a

+ ma
s

1)1

Q'O-ijSt dn-m)!^^ w n* n ( -2 1 _ J.
r=
1

"

*~

)
i

(n2
-

-4-

-2 )(w
2

-4)
'-

+ ...
(n odd)

(6)

n (t)

j JE o (1

__ 1Jiara=s
3

_n n(n*-l) nfo - l')(w'-3) +.... + -p + ^


, t

These results may be combined in the formula


(*7\
1

'

nnW (t\-"4 mv t
(5), (6)

w T (j n

+ j m) r <>8

j ( t w ) w
'

ran-|m + l).(|Om+1
it is

The equations

and (7) were given by Neumann.


easily proved that

By
(8)
(9)

the methods of 2*11,

lO,(0k*.(n!).(*|*!> ^xpCil*!^
enOn (t) = l.(n\).(lt)--*(l
|

+ 0),

(n>l)

where

<;

[exp (

*
|

- l]/(2 - 2).
Sa n Jn (z)
;

From

these formulae
series

it

follows that the series


is

n (t) is convergent

when z is outside the circle of convergence of the latter series. a n Jn (z) n (t) does not tend to zero as n -- oo and so the former series does not converge. Again, it is easy to prove that, as n -*- oo
absolutely convergent

whenever the

lan (zjt)n

and,

Jn <*)

(t)

^
(2n

{l

<-)}

By analogy

with the Legendre function of the second kind,

Qn (t), which

is

such that

t-z

r.

-=, n=o

+ l) Pn {z)Qn (t).

Cf.

Modern Analysis,

% 15*4.

274

THEORY OF BESSEL FUNCTIONS

[CHAP. IX

and hence it may be shewn* that the points on the circle of convergence at which either series converges^ are identical with the points on the circle at which the other series is convergent. It may also be proved that, if either
series is uniformly convergent in

any domains of values of z and

t,

so also

is

the other

series.
is

Since the series on the right of (1)


analytic functions

a uniformly convergent series of

when
J

<

1
1

\,

it

follows

by

differentiation! that

(10)

(-)g.(p r -^ y

(t-zf + * +1

+ g)!: = T y_ 2

dPjn (z) diQn (t)


dzP

w=0

dV

'

where p, q are any positive integers (zero included).


It

may be

convenient to place on record the following expressions:

o.(0i/.

o
8 5
,

= l/t + 4/f, t (t) = l/t + 16/* +


2 (t)

192/*

= i/< = (t) 3/< + 24/P, = (t) 5/t* + 1 20/t* +


()
2
,

1920/**.

The
Otti,

coefficients in the

polynomial

On (t),

for

=0,

1, 2, ... 15,

have been calculated by

Bern Miltheilungen, 1898, pp.

4, 5.

9*11.

The recurrence formulae

satisfied by

On (t).
l)
n (t)
,

We
(1)
(2)
(3)

shall

now obtain the formulae


n+1 (0

(n

- 1)

+ (n + 1)
1

W _,(0

- 2 (n \~

2n --

n?r
,

(>1)
(n^l)

On_ (0-On+1 (0=2On


-0,()0.'(*)

(<),

The first of these was stated by Schlafli, Math. Ann. in. (1871), p. 137, and proved by Gegenbauer, Wiener Sitzungsberichte, lxv. (2), (1872), pp. 33 35, but the other two were proved some years earlier by Neumann, Theorie der BesseFschen Functionen (Leipzig, 1867),

p. 21.

Since early proofs consisted merely of a verification, we shall not repeat

them, but give in their place an investigation by which the recurrence formulae are derived in a natural manner from the corresponding formulae for
Bessel coefficients.

Taking

<

1,

observe that, by
n

91

(1)

and

2-22 (7),

(t-z)l
7i=0

Jn (z)

n (t)

=1= 2
n=0
if 2fr n is

2 M COS \niT

Jn (z),
is

* It is sufficient to use the

theorems that,

convergent, so also

2&/n,

and that then

Sbjri*

is

absolutely convergeut.
\i) in.

t This was pointed out by Pincherle, Bologna Meviorie,


Cf.

(1881

2), p. 160.

Modem

Analysis, 5-33.

9'11]

ASSOCIATED POLYNOMIALS

275

and hence
z

I
n=0

en

Jn (z)

n (0

=I
=0

en

Jn (z) {tOn (0 - cos' \ntr)


- cos2 \ntr],

=X
since tO (t)=l.
If

/ (z) {tOn (t)

now we

use the recurrence formula for

Jn (z)

to

modify

the expression on the right,

we get
{Jn- X
(z)

2
If

=0

en

Jn (z)

n (t)

=2
n=l

+ Jn+1 (z)}

[tOn (t)

- cos

mr}/n.
it

we

notice that

Jn+1 (z) {On (0 cos2 n7r }/n


x

tends to zero as n-*>oo,

is clear

on rearrangement that
(t)

Jo (z) {0

- tO, (t)} + J
I

(z) {20, (t)

- \t0, (t) + }

= 0. + XJn (Z) \20n (t) - *!*L - *2^> + ?^!l?5l 2 n+1 n1 re 1 j n =2

Now
all

regard

zasa

variable, while
left

the Bessel functions on the

remains constant if the coefficients of do not vanish, the first term which does
;

not vanish can be made to exceed the sum of all the others in absolute value, by taking \z\ sufficiently small. Hence all the coefficients vanish identically* and, from this result, formula (1) is obvious.

To prove

(2)

and

(3) observe that

/d
\dt

+ dz) t-z~

d\ _i

'

and

so, \z\

being

less

than
j

t
j

we have
n' (t)

2
By

Jn (Z)

+2
we

en

Jn

'

(z)

(t)

= 0.

rearranging the series on the

left

find that

5 *nJn (Z) On' () =0

-J

(z)

(t)

- I \J^ (,) - Jn+1


n=l n=l

(jr)}

On (t)

= - J (,) 0, (*) - 2
that
is

Jn (Z) {0 n+1 (0 - On-, ()},

to say,

J. CO {Oo' (0

+ 0, (t)} + i Jn (M) {20 n

'

(t)

n+l

(0 "

n _, (0}

= 0.

On
*

equating to zero the coefficient of


(1),
is

Jn (z)

on the

left,

just as in the

proof of
This

we

obtain (2) and

(3).
if

then

a convergent power series vanishes identically, 3-73). Ttie argument is valid here because the various series of Bessel coefficients converge uniformly throughout a domain containing z = 0.
all its coefficients

the argument used to prove that,

vanish

(cf.

Modern Analysis,

276

THEORY OF BESSEL FUNCTIONS


By combining
(1)

[CHAP. IX

and

(2)
1)
1
)

we

at once obtain the equivalent formulae


'

(4) (5)

ntOn^ (t) - (w2 ntOn+1 (t) - (n2 -.

0 it) = (n - l)tOn
n (t)
(

(t)
'

+ n sin

\mr,

= - n + 1 ) tOn

(t)

+ n sin 2 \mr.

If
(6)
(7)

be written
(n

for

{d/dt), these

formulae become
2

- 1) (^ + n + 1) n (0 = n \tOn-* (t) - sin %rnr}, (n + 1) (^ - n + 1 On (t) = - n {tO n+1 (t) - sin |wtt}
)

The Neumann polynomial


by the equation
(8)

of negative integral order

was defined by Schlafli*

0_(<)
this
n.

= (-)n On (0.
(1)

With
values of

definition the formulae

(7)
n

are valid for

all

integral

9'12.

The

differential equation^ satisfied by

(t).

From the
(*

recurrence formulae 9*11 (6) and


n (t)

(7), it is clear

that

+ n + 1) (* - n + 1)

-L;- (^ + n + 2 1) {- ntOn+1 (t) + n sin \nnr}

=-

Tit

-j

(^

+ n + 2)
2

n+ , (0

+ n sin \mr
8 2

= t {tOn (t) cos


and consequently O n (t)
(%

\n-jr)

+ n sin

\nir

satisfies the differential equation

l)2

n (t)

(<

- n ) On (t) = t cos \mr + n


2 2

sin2 \nir.

It follows that the general solution of the differential equation

is

y = On (0 +
(1)

^(0,

and so the only solution of is On (t).


It is

which

is

expressible as a terminating series

sometimes convenient

to write (1) in the

form

where lere
(w even)

( odd)
*
t

Math. Ann. in.

(1871), p. 138.

Neumann, Theorie der

Bessel'schen Functionen (Leipzig, 1867), p. 13

Journal fur Math.

livii. (1867), p. 314.

9' 12, 9*13]

ASSOCIATED POLYNQMIALS
differential equation is to observe that

277

Another method of constructing the


2

f
and so

z?

+z

l +zj Jo 6-*4 {z) {t)= i

enw2,7n {z)

* (0

'

.**<*> W={*2 ^+4z + Z


2z*
z

t^z t-z

(t-zf

(t-zf

00

00

Now
and hence
Therefore

t^J^ (z), z= 2
00

fin +

(2 + 1)

J2n +

l (z),

+ z = t2

n=0

2 en,gn

(t)

Jn (z).

2jn Jn (z) {< ~ + 3t I + On


as in 9*11,

+ 1* - *} On (t) - V ()] ^ 0.

equating to zero the coefficient of Jn (z) on the left-hand side of this identity, just we obtain at once the differential equation satisfied by On (t).

9*13.
It has

Neumann's contour
been shewn by

integrals associated with


that, if

O n (z).

Neumann*
Jc

be any closed contour,

(1)

Om (z) On (z) dz = 0,
J m (z) On (z) dz =
0,

(m = n and

mn)

(2)

(ma n>)
,

(3)

f Jn (Z) On (Z) dz c

= 27TlA;/n

where k

is

the excess of the

number

of positive circuits of the contour round

the origin over the

number

of negative circuits.

The
larity of

first

result is obvious from Cauchy's theorem, because the only singu-

(z)

n (z) is at the origin,

and the residue there

is zero.

The
grand
is

third result follows in a similar

manner

the only pole of the inteis l/ea .

a simple pole at the origin, and the residue at this point


result,

To prove the second


by zO n (z) and

multiply the equations

V m Jm (Z) = 0,

V n [z0 n {Z)\ = *9n {Z)


and subtract. If

Jm (z)

respectively,
./

U (z)

be written in place of

W iifO|<f)i-, W ^),
BesseVschen Functionen (Leipzig, 1867), p. 19.

the result of subtracting assumes the form

*U' (z) + zll{z) + (m2 - w2 ) zJm (z) On (z) = z*g n (z) Jm (z),
* Theorie der

278

THEORY OF BESSEL FUNCTIONS

[CHAP. IX

and hence
[z

U (z)] c + (m - *) f Jm (z) O n {z)dz=\


2

Jo

Jo

z*gn (z)

Jm (z) dz.

The integrated part vanishes because U(z) is one-valued, and the integral on the right vanishes because the integrand is analytic for all values of z and hence we deduce (2) when 2 n 2
;

Two
(4)

corollaries,
1

due to
/(o /> +
/
)

Schlafli,

Math. Ann.

m.

(1871), p. 138, are that

J* {x+y)
)

m (y) dy = Jn . m (x) + ( -)m Jn + m (#),

/"<-* +
/

(5)

-
first is

Om {x+y) Jn (y) dy=Jm _ n (z) + (-)n Jm+n (*)


(2)

The

obtained by applying

and
00

(3) to

the formula 2*4

(1),

namely

Jn {*+y)=

2 /, +p (i)/.j(y), p 00
variable.

and the second follows by making an obvious change of

9*14.
It

Neumanns

integral for

On (z).

was stated by Neumann* that

(i)

o, (.)
shall

J"

+**+

= /;

^i+i:
2

^
;

+ *

^du.

We
(2)

now prove by induction the

equivalent formula

On (z) =
is

U Jo

/*>

exp ia
[{t

+ V(l +
|

)}

+ {t- V(l + * )} w] *~* dt,


a

where o

any angle such that a + arg z <


j

\ir

on writing

u/z,

the truth

of (1) will then be manifest.

A
(3)

modification of equation (2)


roo+ta

is

0 (z) = \\
(2)

w {e *

+ (-)""} -*** cosh ^0.

Jo

To prove

we observe that
/"ooexpia

racexpia

O
and
so,

(s)

=
Jo

e-*dt,

0,(z)=
Jo

te-*dt;

by using the recurrence formula


oo

911
<f>

(2), it follows

that

we may write

exp ia
gt n (t)e~ dt,

0. <*)-["
Jo

where
(4)
<f> +1

(0 -

2t<j> n (t)

- <_, (0 = 0, =
*.

and
(5)
*

<M0=1>

*!<)

Theorie der BtsseVtchen Functionen (Leipzig, 1867), p. 16; Journal filr Math, lxvii. (1867),

p. 312.

9*14]

ASSOCIATED POLYNOMIALS
solution of the difference equation (4)
<f>

279

The

is

n (t)

= A{t + V(* + l)} n + B {t - V(l + t*)} n


2

where

and

are independent of n, though they

The

conditions (5) shew, however, that

A B=\

might be functions of t. and the formula (2) is

established.

This proof was given in a symbolic form by Sonine*,


/<*> exp ia
<t>n

who wrote

<p n

(D)

(1/z)

where we

(0

*"*** dt>

D standing for (djdz).


is due to Kapteynf, which we now write in

completely different investigation of this result


is

whose analysis the form

based on the expansion of

9*1 (1),

--=X n Jn (S)O n (z).


s

*=o

When

<

s
j

we have

= IT\ 2 9
2" '">
.

oo (n=-<* V.n=
1

Jn (0\ e-du,
j

if

p be so chosen that

u
z
It follows that

We shall now shew that the


will
W

interchange of summation and integration


z

is justifiable; it

be sufficient to shew that, for any given values of ( and


f
\

(such that
|

<

|),

fu + J(u i 4-z )\ n
2

n=N+\
can be made arbitrarily
3^-all

by taking

N sufficiently large J

now

\uJ{u*+z*)\^2(u + \z\),
andso

Jn{0

r\^^p^l\
J
2
|

e -u

du

^J^ fV(+|,|)
z
I \

-*

1*1"

J LI

<|(C/*)"|exp{|| + iim.
*

Math. Ann.

xvi. (1880), p. 7.

For a similar symbolic investigation


(3) x. (1893), p. 108.

see 6 "14 supra.

t Ann. Sci. de VEcole norm. sup.

X Cf. Bromwich, Theory of Infinite Series, 176.

280
Therefore, since
|

THEORY OF BESSEL FUNCTIONS

[CHAP. IX

<
I I

|,

we have
|

r-

.-l+i J.

{uJW+f^jn() (z)
*=
left

If
*

*P

{!!

+*

<f}
'

\*\{\'\-\i\rby taking

and the expression on the when z and ( are fixed.

can be

made

arbitrarily small

N sufficiently large

Hence, when

<

j,

we have
/(f)
.'0

= 2
w = - ee

{w+^+f!)} e~"rfu _
*

y)

tti

'

= 2en Jn (00n (z)


where
n (2;) is defined

by the equation
\*

0 (,) = j"
and
it is
1.

+
n

" + Wj- </(* + ** ^ du


e
(z),

easy to see that

so defined,

is

a polynomial in 1/z of degree

n+

When
term,
it is

the integrand

is

expanded*

in powers of z

and integrated term by

easy to reconcile this definition of

n {z) with the formula 9*1 (4).

9*15.

Sonine's investigation of

Neumann's

integral.

An

extremely interesting and suggestive investigation of a general type

of expansion of l/(a

z)

is

due to Soninef; from


is

this general expansion,

Neumann's formula
difficulty.

(9'1) with the integral of

91 4 can be
(w)

derived without

Sonine's general theorem

as follows

Let
so that

yjr

<fr

(w) be an arbitrary function of is the function inverse to ty.

w;

and, if

yfr

= x,

let

w = ^ ()>

Let

Zn

and

A n be defined
1

by the equations%

z =
Then
it

f<+>

ti)

** m

CL

k> ^=/o
= Zn A n
n =-a>

din

*-"*{*(*)}"<**

being assumed that the series on the right is convergent.

Suppose that
curve

for

C surrounding

the origin and the point


(1918), 264.

any given positive value of x, z, and

w > (x) w < ^ (x)


\ \

-fa

on a closed

on a closed

* Cf.

Hobson, Plane Trigonometry

t Mathematical Collection (Moscow), v (1870), pp. 323 modified slightly, but the symbols ^ and ^ are his.

382.

Sonine's notation has been

J This
Wist.
ii.

is

connected with Laplace's transformation.

See Burkhardt, Encyclopadie der Math.

(Analysis) (1916), pp.

781784.

9-15, 9*16]

ASSOCIATED POLYNOMIALS
z.

281

curve

surrounding the origin but not enclosing the point

Then

+
2iriJ

1
1

oo

/*oo

y%n

gg(w)-

dwdx

\Jc
J
-" x

Jc

w-^{x)

gZ^(W>) a*

2iriJ

=
provided that

Jo

e (z
;

dx =

l/(o

z),

B(z)<R (a)

and the

result

is

established

if it is

assumed

that the various transformations are permissible.

In order to obtain Neumann's expansion, take


y}r

(w)

\ (w

l/w),

vf (x)

= x V( +
2

1)

and then

n= oo

= i KJ(*){4 n + (-)A_n
=

}.

Since

4 n + (-) ^4_ n = ( V'* [>


Jo

vV + l)} n + (-)n {* V(* + 1)}" W

] <*#,

we

at once obtain
Sonine notes
{jp.

Neumann's

integral.

328) that

Jn (z)~(hj) n ln\, *nOn (a)~n\($a)-*, and in the later part of his a\ of the expansion so that lfta-z) converges when J z memoir he giyes further applications of his general expansion.
\

<

9*16.

The generating function of On


series

(z).

The

(-^n^On (*),

which

is

a generating function associated with

On (z),
the

does not converge for any value of t except zero. series after the method of Borel, in the following manner

Kapteyn* however, has

"summed"

n.(n + m-\)\t2n

_
_1

(n

+ $).(n+m)l

2n +

nlomlo
1

(n-)!(i*)* | n.(n + m-l)l

"

t*

n Ion=m

_1

(2w)
2

~ 2 JE

(-)! (^) 2m+2 <^ (!+<) _


!

(2m+l)l
2"* + 2

<* + 1
(1

(l+<^
2m + 2
)

"(**)*" +

(1

-'

2
)

2m+1
'

w=

(i*)

- <a

l+<
*

(-).m! tm

*(l-<*)m=o{*( 1 -< , )} m
Nieuw Archief voor Wiskunde

(2), vi. (1905),

pp. 49

55.
10

282

THEORY OF BESSEL FUNCTIONS


/>
7j

[CHAP. IX
^

e~"du
tot

>

an d this integral

is

convergent so long as

(1

1'*)

z\t is

not negative.

There

is

no great

difficulty in verifying that the series

2
n=o
t

n(

nt

On (z)
|

is

an asym-

ptotic expansion of the integral for small positive values of


integral

may be

regarded as the generating function of

(z).

when arg z < n, and so the Kapteyn has built up much


|

of the theory of

Neumann's function from

this result.

9 '17.
It
is

The inequality of Kapteyn 's type for


possible to deduce from

(m).

Neumann's

integral an inequality satisfied


satisfied

by

n (nz)

which closely resembles the inequality

by

J n (nz)

obtained

in 8-7.

We

have
1 n < n *) = 2i^+i

r*
J

H + V(w + z )] n +{u>- V(/


2

z*)} ]

e~nw dw,

the path of integration being a contour in the w-plane, and so


n (nz)

^
I

~
i

["
|

[{,

y/(vfi

+ *)} e-] | dw
!

,
j

J o

where that value of the radical


greater modulus.

is

taken which gives the integrand with the

Now
is

the stationary point of

[w

+ V(w + z )} e~ w
2

V(l
(!)

-z ),

and

so

On (nz) ^
|

JL_
|

Lt^^p
is

Jl {"

+ VK + z>)} e-\.\dw\,
is

where the path of integration


stationary point.

one for which the integrand

greatest at the

If a surface of the type indicated in 8*3

is

constructed over the w-plane,


;

the stationary point

is

the only pass on the surface


if

and both

w=

and

w = + oo
(2)

are at a lower level than the pass

Vd-*

Hence, since a contour joining the origin to infinity can be drawn when (2) is and since the integral involved in (1) is convergent with this contour, it follows that, throughout the domain in which (2) is satisfied, the inequality
satisfied,

(3)
is satisfied for

O n {nz)<.~
z
i

-z) z exp V(l s )


1 4-

V(l

j"" 1

some constant value of A


8'7.

and

this is

an inequality of the same

character as the inequality of

9-17, 9-2]
9'2.

ASSOCIATED POLYNOMIALS

283

Gegenbauers generalisation* of Neumanns polynomial.


z"/(t

If
z

we expand

z)

in

ascending powers of z and replace each power of

by the expansion

as a series of Bessel functions given in 5*2,

we

find

on

rearrangement that
zv

* z v+s

2 "+*

(v

+ s + 2m).T (v + s + m )
(i/

"

" (<i 2" +n 2m

J iio
no greater theoretical

* -

+ n)

1"

(v

+ n - m)

A
;

~~^.

Jv+n {z)
\

the rearrangement has been effected by replacing s by n


difficulties

2m, and

it

presents

than the corresponding rearrangement in 91.

We are

thus led to consider Gegenbauer's polynomial

A ntV {t),

defined by

the equation

this definition is valid


|

whenever

v is not zero or a negative integer

and when

z\<
(2)

\t\,

we have
t

. =2 A Z =0
M

n>v

{t)Jv + n

{z).

The reader should have no


formulae
(3)
(v

difficulty in

proving the following recurrence


2
\(v
V* + n1
-

+ n-VAn+^M + iv + n + VA^.it)--*
2" (v

1
[

J t

An

(t)

+ n)

\(v

+ nY -1}
t\\\n +

1"

(v

+ \n- \)

Sinz *W7T,

\)

(5)

(i/

+ n) tA n_ 1<v (t) - (n +

+ n - 1) 4,,,, (0 2"(v + n)(v + n-l)r(v + ln-$) , ,, ., I^ 8in2l = ( y + w _l)^' i". n> () + rfln + i)
1) (v
v

(6)

(1/

+ w) ^+1 , (<) - ( + n +

1) (2

+nn)(i

1)

A n>v (t)
,

^ /x + 2"(i/ + = -( + + -v 1)^', ,,(*)


(7)
*

+ n + l)r (i + ln + i) ... ^sin'iuTr, r(in + f)

4^)2T(+l)/t
Wiener Sitznngsberichte, lxxiv.
(2),

(1877), pp. 124

130.

284

THEORY OF BESSEL FUNCTIONS


The
differential equation of

[CHAP. IX
is

which

A n>v (t)

is

a solution
)

(R\

d*y

J>-2vdy

(n+l)( 2in -w-l)


J

where
(9)

^(

.-A
r(J

L__.l

WW
+^ _I ^+n (0also

The

general solution of (8)

is -4 ,,()

Of these
proved that
(10)

results, (3), (4), (8)

and (9) are due to Gegenbauer; and he

_.
J

A n>v (t)e<*dt = 2in T{v).(v + n)Cn (z),


2az + a*)'*
is
;

where

Cn v (z)
is

is

the coefficient of an in the expansion of (1

this

formula

easily proved

by calculating the residue of (izt) m A n> (t) at the


for

origin.

The corresponding formula


(11)

Neumann's polynomial
izt

1
I

/"< 0+)

Lttx J

(t) e

dt

= in cos

n arc cos

z)

The
(12)

following formulae

may

also

be mentioned
v

A m v (z)A n
c
'

(z)dz

= 0,

(m = nandm^n)

(13)

z-*Jy+m (z)A n>v (z)dz = 0,


z~ v

(m2 ^fen2 )

(14)

I
.'

Jv+n (z) A n
n = 0,

(z)

dz =

2irik,

c
1,2,
. . . ,

where

G is any closed
first

contour,

and k

is

the excess of the

number

of positive circuits over the

number

of negative circuits of

G round the origin.


;

The

and third of these


V +m J v+m (z)

last results are

proved by the method of 913

the second

is

derived from the equations

= 0,

V+n

{*-

An

(z)}

= z*-> gnt (*),


z2

whence we

find that

(m - n)

(2v

m + n)

Jo

z~ v

Jv+m (z) A n,,( z ) dz =


Sn (t).

~v

gn<v {z) Jv+m (z) dz

0.

9*3.

Schlafli's

polynomial

polynomial closely connected with Neumann's polynomial


it is

O n (t) was

investigated by Schlafli.
properties,

In view of the greater simplicity of some of its frequently convenient to use it rather than Neumann's poly-

nomial.

9'3]

ASSOCIATED POLYNOMIALS
polynomial
is
1 \

285

Schlafli's definition* of the

(1) (2)

Sn (t) = 2
S.()-0.

<n {n (n

_m_

A)

(iO-"

+MW
,

(n

>

1)

On comparing
(3)

(1) with 9*1 (2),

we

see at once that

n Sn (t) = *0n (0 - cos8 \nir.


n (t) in

If we substitute for the functions

the recurrence formulae 9*11 (1)


cos2 mr,

and

(2),

we

find from the former that

(4)

+!
latter,
x

(t)

+ _! (0 - 2n~ Sn (t) = U~
1

and from the


(n
If

- 1) #n _ (0 - 1 (n +
this

1) +,

(0 = nSn

'

(t)

- nt- 8n (t) - 2*"


1

cos2 fnw.

we multiply
(5)

by 2 and add the

result to (4),

we get

Sn^{t)-Sn+1 (0 =
(4)

25/(0.

may, of course, be proved by elementary algebra by using the definition of Sn (t), without appealing to the properties of
(5)

The formulae

and

Neumann's polynomial.

The
(6)

definition of Schlafli's polynomial of negative order is

S_n (0 = (-)w+1 >SU0,


all

and, with this definition, (4) and (5) are true for

integral values of n.

The
(7)
is easily

interesting formula, pointed out

by

Schlafli,

Sn^(t) + Sn+1 (t) = 40n (t),


derived from (3) and
(4).

Other forms of the recurrence formulae which may be derived from (4)

and

(5) are

(8) (9)

nSn (0 - tSn (0 = 2 cos8 ^nir, tSn+1 (0 - nSn (0 + tSn (0 = 2 cos8 |n7r.


*n_, (0 * '

If

we write ^

for

(d/dt), these

formulae become
2 cos^wtt,

(10) (11)
it follows that

(S

+ n) Sn (0 = tSn ^ (0 -

(^ - n) Sn (0 = - tSn+1 (0 + 2 cos2 wr-

(^

- n ) Sn (0 = t (^ + 1 - n) #n_, (0 + In cos8 \nir = - Sn (t) + 2t sin fnir + 2n cos \nir,


8 t 2
8

and so
(12)

Sn (t)

is

a solution of the differential equation

du - + t-~ + (< - n
d?v
8

)y =

2 sin 2 /wr

+ 2n cos

\rnr.

Math. Ann.

m.

(1871), p. 138.

286
It

THEORY OF BESSEL FUNCTIONS


may be S
1

[CHAP. IX
:

convenient to place on record the following expressions

(t)

= 2/t,
16/t
3
,

&(*)-4/*,

S3 (t) = 2/t +
5

S
5
,

S (t) = 2/t + 4,8/t + 768/*


3

= 8/ + 96/i = S (t) 12/t + 384/f + 7680/tf


4

(t)

1
;

The general descending

series,

given explicitly by Otti, are

-2 2n(n-- 2 )(w -4 = 2n + 2n(n + + t P


i 2
2

) '

'

t-

2 (n2

l2)

2 (wf -

(w 2

-3)
2

polynomial (<)> for n = l, 2, ... 12, have been calculated by Otti, 14 Otti's formulae are reproduced (with some obvious errors) by Graf and Gubler, Einleitung in die Theorie der Bessel'schen Funktionen, II. (Bern,

The

coefficients in the

Ztera Mittheilungen, 1898, pp. 13

1900), p. 24.

9*31.

Formulae connecting

the

polynomials of Neumann and

Schlafii.

have already encountered two formulae connecting the polynomials of Neumann and Schlafii, namely

We

\nSn (t) = t0n (t) - cos \ntr, 8n- (t) + S (t) = 40n(f),
2
l l

of which the former is an immediate consequence of the definitions of the functions, and the latter follows from the recurrence formulae. A number of other formulae connecting the two functions are due to Crelier * they are easily derivable from the formulae already obtained, and we shall now discuss
;

the more important of them.

When we
find that
(1)

eliminate cos2 |/wr from

93 (3) and

either

93 (8)

or (9),

we

(2)

Sn _ (t)-Sn(t) = 20 n (t), Sn+1 (t)+Sn(t) = 20n (t).


1

Next, on
(3)

summing

equations of the type 9*3

(5),

we

find that

Sn (t) = -2

<(i-D

2
w=
(t)

S'n^mr-i(t)

+ Bin*lnir.S

(t),

and hence
(4)
*

Sn (t) + Sn _

= -2 2 fi^-, (*) + &(*).


TO

Comptes Rendus, cxxv. (1897),

pp. 421423, 860863; BernMittheilungen, 1897,

pp.

6196.

9-31, 9*32]

ASSOCIATED POLYNOMIALS

287

Again from
4

93 (7) and (5) we have {<?_, (0 + O n+J ()} = n _ 2 (t) + 2Sn (t) + S n+2 (t) = (Sn_ 2 (*) - Sn ()} - {Sn (t) - Sn+2 (t)\ 4Sn (0 = 2S' n_ (t)-2S\l+1 (t) + 4 Sn (t) 4Sn" (0 + 4^(0,
1
!

so that
(5)

Sn" (t) + Sn (t) - O n _


is

(t)

+ O n+1 (t).
911
(2),

This

the most interesting of the formulae obtained by Crelier.

Again, on
(6)

summing formulae

of the type of

we
2

find that
(t),

On (t) = -2

I
5/1=0

0'n -2m-! 0)

+ 8^4^.0,(0 + cos $mr.O


O'n-^-,

and hence
(7)
n (t)

0^

(t)

= -2 "%
m=0

(t)

0,

(t)

{t).

9*32.

The
(1)

peculiar

Graf's expression of Sn (z) as a sum. summatory formula

Sn (z) = Tr

m= n

[Jn (z)Ym (z)-Jm (z)Yn (z)}

was stated by Graf*


Gubler's treatisef.

Graf and most readily proved by induction it is obviously true when n = 0, and also, by 363(12), when n=l. If now the sum on the right be denoted temporarily by n (z), it is clear that
This formula
is
;

in 1893, the proof being supplied later in

<f>

<n+i (Z)

<n-i (Z)

~ (%n/z)
2

<f>

n {Z)

= 7T Jn+1 {Z) + irJn_


1

m+1

m= n-1

Ym (Z) - 7T

n+1

Fn+j (z)

2
m=n-l

Jm (z)
Jm (z)
m=-n

(z)

2 m= n+l

Ym (z)-7rYn_

(z)

m= +l

-(2n7rjz)Jn (z)

2
m=n

Ym (z) + {2nir[z) Yn (z) 2 Jm (z).

Now

modify the summations on the right by suppressing or inserting terms at the beginning and end so that all the summations run from n ton; and

we then
vanish.
<f>

see that the complete coefficients of the


It follows that

sums 2 Jm (z) and 2

Ym (z)

n+1 (z)

+ <_, (z) - (2n/z) n (z) = 7rJn+1 (Z) Yn+1 (Z) + Y_ n ^ (Z)} - 7T Fn+1 (z) {Jn+1 (z) + J_n_, ( Z)\ - -rrJ^ (z) n (z) + F_M (Z)} + 7T n_ (z) {Jn (z) + J_ n (z)\ _ - {l + (_ l){ \Jn _, ( Z ) Yn (z) - Jn (z) Fn _ (,)}
<f>
{

= 4.2 -1 eos

^wr,
is satisfied

by 363(12); and so n (z) satisfies the recurrence formula which by Sn (z), and the induction that n {z) = Sn (z) is evident.
<f>
<f>

Math. Ann. XLin.

(1893), p. 138.

t Einleitung in die Theorie der BesseVschen Funktionen, n. (Bern, 1900), pp. 34

41.

288
9*33.

THEORY OF BESSEL FUNCTIONS


Greliers integral for

[CHAP. IX

Sn (z).

If

we take the formula 9*14(2), namely


r ao

exp

ia

On (e)

=h
.'

[{t

V'(l

+ t*)} n +

\t

- V(l + )} w ] e~* dt,


2

and integrate by

parts,

we

find that

eft

_ cos'^wtt + " 5
Hence
it

00

_^ 2zJ

i"

\t

+ \/(l

4-

P)}

V( 1

+ * )} w
2

_rf

V(l+**)

follows that

This equation, which was given by


(2)
is

Schlafli,

Math. Ann. in. (1871),


f

p. 146, in

the form

Sn (z)= j"** {?*-(-)*-'*} e-'**dO


Crelier's researches*, of

fundamental in

which we

shall

now

give an outline.

We

write temporarily

Tn =
and then
so that

[t

+ V(l + *2)} n -

\t

- V(l + ?)}'\

y"+! _ ot 4T
and therefore
7'
C

IT
*

'

+ 2 + 2*+...+2e'

the continued fraction having w elements. It follows that quotient of two simple continuants^ so that

Tn ]Tn
x

is

the

Tn+ l= K(2t,2t,...,2t)n

Tn
the suffixes
n,

K{2t,2t,...,2t) n

^
and since

denoting the number of elements in the continuants.

It follows thatj

Tn /K(2t) n-i is independent of n; T = 2V(1 + I'), #(20o=l,


1

we have

Tn = 2</(l + t*).K(2t)n_
* Comptes Rendus, cxxv. (1897), pp.
;

1 ,

t Chrystal, Algebra, n. (1900), pp.

% Since

all

the elements of

421423, 860863 Bern Mittheilungen, 1897, pp. 6196. 494502. tbe continuant are the same, the continuant may be expressed by

this abbreviated notation.

9*33, 9*34]

ASSOCIATED POLYNOMIALS
fee

289

and hence
exp ia

(3)

Sn (z) = 2
Jo

K(2t)n _ e~^dt,
1

From

this result it is possible to obtain all the recurrence formulae for

8n CO by using properties of continuants.

9'34.

Schlafli's expansion

of Sn (t

+ z)

as a series of Bessel

coefficients.

We

shall

now

obtain the result due to Schlafli * that,

when

<

1
1

Sn (t 4- z)
(1)

can be expanded in the form

Sn (t + z)=

m= 00

I Sn _n (t)Jm (z).

is

The simplest method of establishing this formula for positive values of n by induction f. It is evidently true when n = 0, for then both sides vanish
1,

when n =

the expression on the right

is

equal to

S, (0 Jo (*)

+ X

{Si-m (t)

Jm (*) + S^ (0 J_ m (z)}
2 {Sm^W + S^ityJ^-z)
*=1

= 2Oo (t)J
=
2

{-z) +

I e m Om (t)Jm (-z)

=
by 91(1) and

2/(*

+ z) =

S,

(t

+ z),

93 (7).
truth of (1) for Schlafli's polynomials of orders

Now,

if

0, 1, 2, ... n,

we assume the we have


(t

Sn+i

+ z) = S^ (t + z)- 2Sn

'

(t

+ z)
2
S'n _,n (t)Jm (z)

m=

X
2

Sn-r-i(t)Jm

(z)-2

wi= 00
30

\Sn. m-,{t)-2S'H . m (t))Jm (z)

m= *>
and the induction
is

established

to obtain the second line in the analysis,

we have used the obvious

result that

S'(t
*

+ z) = ^SH (t + z).
;

Math. Ann. in. (1871), pp. 139


the reader
(cf.

141

the examination of the convergence of the series

is

left to

9-1).

The extension

to negative values of follows

on the proof

for positive values,

by

9*3 (6).

290

THEORY OF BESSEL FUNCTIONS

[CHAP. IX

The expansion was obtained by Schlafli by expanding every term on the right of (1) in ascending powers of z and descending powers of t. The investigation given here is clue to Sonine, Math. Ann. xvi. (1880), p. 7 ; Sonine's investigation was concerned with a more general class of functions than Schlafli's polynomial, known as hemi-cylindrical functions
< 10-8).

When we make
(2)

use of equation
n (t

9*3 (7), it is clear that,

when

<

1
1

\,

+ z)=

= 00

O n.m {t)Jm (z).


Wiener Sitzungsberichte, lxvi.
(2),

This was proved directly by Gegenbauer,


pp. 220

223, who expanded


[cf.

(1872),

n (t+z) in ascending

powers of

by Taylor's theorem, used

the obvious formula

$ 9-11 (2)]

(3)

2P

^r-)==

* (-) m pCm .On - p+2m


series.

(t),

and rearranged the resulting double


It

is

easy to deduce Grafs* results (valid

when \z\<

\t\),

(4)

Sn (t-z) = On (t-z)=

I Sn+m^Jmiz),

(5)

Wl =

i O n+ m (t)Jm (z).
oo

9*4.

The

definition of

Neumanns

'polynomial ft n (t).

The problem
formulae of

of expanding an arbitrary even analytic function into a

series of squares of Bessel coefficients


2*72,

was suggested to which express any even power of z as a

Neumann f by

the

series of this type.

The preliminary expansion, corresponding

to the expansion of l/(t

z)

given in 91, is the expansion of l/(F z2 ); and the function ft n () will be defined as the coefficient of n Jnz (z) in the expansion of l/(tfa z*), so that
C1)

ir-r, t*-z

= # (*) Oo (0 +
n=0

<*) fli

(0

W
tft

<*)

n. (0

= 2 c/.'Wfl.^
To obtain an
panding
*

explicit expression for fi n (0>

ke

<

*
I

aQ d, after ex-

l/(s

z})

in ascending powers of
141

z,

substitute for each power of z the

Math. Ann.

xliii. (1893), pp.

142;

see also Epstein, Die vier Rechnung*operationen


iiber die Fortschritte der

mit Bessel'schen Functionen (Bern, 1894).


pp. 845846.]

[Jahrbuch

Math. 18931894,

t Leipziger Berichte, xxi. (1869), pp. 221256. [Math. Ann. in. (1871), pp. 581610.]

9-4]

ASSOCIATED POLYNOMIALS
given by

291
( 2-72).

series of squares of Bessel coefficients


9*1,
1

a 2

Neumann

As

in

we have

z 2*

-s ~,=oM+a

when we rearrange the


presents no

series

by writing

ns

for

this

rearrangement

greater theoretical difficulties

than the corresponding rearrange-

ment

in 9*1.
is

Accordingly the function fl (t)

denned by the equations

V>
(3)

Un

W ~ 4 ,r

(n

- *)! (2s)! (i
we

M+a

'

(n>l)

n i (o-i/'.
the order of the terms in (2)
find that

On reversing
<*)

nW ~4 mto
"nW-i+oir+o ^3.4 2
**
-

* n.(2n-tn-l)l{(n-m)!} ~ m!(2n-2/rc)!(02M 2w+2


full, it

(n>l)
'

while, if (2) be written out in


1

assumes the form

14na

1.2 4n'(4n-2')
/L

1.2.3 4n a (4na - 2 2 )(4tt2 -4a )


+4, s.

(5)

/6

Also

A*
(6)

@44

2.(2w-l)(2n-2)
Hen*
2"

+
)
'

+ 1.2...n.(2n-l)(2n-2)...nf
where

.= ~

2 n-l
2/i

@ = (2n-l)(2n-3) 2n(2n-2)
4

'

_ (2-l)(2n-3)(2n -5) "' W6_


(7)

2n(2n-2)(2n-4)

0 =
Since
(8)

(2m-1)(2w-3)...1 2n(2w-2)...2
211
),

< Bm <

1, it is

easy to shew by the methods of

that

n it)

<*

2*
j

- 2- 2 (n 2 exp !) (
2 2

2
|

and,

when n > 0,
en

(9)

nw (0 2W <- "|

(n!)a (l

+ 0),

where

0\

S {exp|*| a

- l}/(2n- 1).

292

THEORY OF BBSSBL FUNCTIONS


By
reasoning similar to that given at the end of 91,
it is

[CHAP. IX
easy to shew

that the domains of convergence of the series


are the same.

1an Jn* (z) iln (t) and %an (z/ty*


the curious formula, due

The reader should have no


to

difficulty in verifying

Kapteyn*,

ao)

-(')- s

5/r^(a^)Qn
(t).

9*41.

The recurrence formulae for


to

The formulae corresponding


2
(1) K)
t

911

(2)

and

(3) are

n'(t) lln{t)

= n -^

**<*>

n-1

n+

2n 'Q n-l'

{n>4 < n >2) >

(2)

(2/<)rV(o
(2/0

= *n (0-**MO,

(3)

'

(t)

= - 2n, (t) + 2n

(t).

There seems to be no simple analogue of

9*11 (1).

The method by which

Neumann f

obtained these formulae

is

that described in 9*11.

Take the fundamental expansion

9*4 (1),

and observe that

and

that,

by Hansen's expansion of
2 Jo ()
J' (z)

2*5,

= - z 2 JV, (*) - ./Vi (*))/


{

We find by

differentiations with regard to


en

t,

and with regard

to

z,

that

-2t/(t*-z>Y= 5
n=0
a/(ff

Jn*(*)fV(0>
j; to
flo

- *y = 2 j

(#>

(0

J'n., (*)

- Jn+1 (*)} n n (o/n

=* I
n=l

{/*.,

to -/Vito){n(*)-n. (*)}/
it is

On comparing
*-

these results,

clear that
2

en

Jn*(z)a n \t)+ i {Jv to-^


1

+ ito}.{(0-n(o}/n=a
left

On

selecting the coefficient of

Jn

(z)

on the

and equating

it

to zero

(cf.

91),
*

we

at once obtain the three stated formulae.


Sci. de

Ann.

VEcole norm. sup.

(3) x. (1893), p. 111.

+ Lnpziger Berichte, xxi. (1869), p. 251. [Math. Ann.

m.

(1871), p. 606.]

9-41, 9*5]
9*5.

ASSOCIATED POLYNOMIALS

293

Gegenbauer's generalisation of Neumanns polynomial


;L+ "l(t z)

Q n (t).

in ascending powers of z and replace each power If we expand z of z by its expansion as a series of products of Bessel functions given in 55, we find on rearrangement (by replacing s by n 2m) that
gfl

+v

* gH + v + S

2*i+ " + *(

^ r(/j.+s+l)r(v+%s + l)(fji+i>+s+2m)r(iJ. + v + s+m)


\z)

t/

M+ j+i

J v+ ig +m

(z)

=0

i=0 (,m=0

*n swi + i
&

(/A+y + n)r(/i,

+ ^w-m +
m!r(/tx

1) r(i/
i/

+ |n-m + + + w-2i + l)

l)r(/A

+ y + M -w)

it is

supposed that z
|

<

|,

and then the rearrangement presents no greater

theoretical difficulties than the corresponding rearrangement of 91.

We

consequently are led to consider the polynomial

Bnillt (t),

defined by

the equation

m
(-

1)

R ;**(*) -n

+ i>+w) = 2+'+(m ^T+l


X
<J
,t
r(fjL

+ %n-m + l)r(v+$n-m + l)r(fi + v+n-m) m!rV + v + n-2wi + l)

W
,

'

This polynomial was investigated by Gegenbauer*; it recurrence formulae, none of which are of a simple character.
It
(2)

satisfies various

may be noted

that
,;
0,0

(0

= **" (*)
functions in ascending

The
on
series

following generalisations of Gegenbauer's formulae are worth placing

record.

They are obtained by expanding the Bessel


residues.

and calculating the


2^.

(3)

t-"J
J

(2t sin

<f>)

B2n +
dt

1;

>v (t)dt

= 0.

(4)

2^. J

t~ >

J, (2t sin
/

<f>)

Bm

M>

(t)

In the
1

_ 2* + "((j + v + 2n)r(fi, + l)r(fj, + v + n)8m''<b n\T(fi + v+l) x aFa (-n,/ji + l,fj.+ v + n; f* + $* + $, \f- + l v + special case in which ft = v, this reduces to
r(o+)

2 *J sin

<f>).

(5)

t-*Jp {2tam4>)Bm

v, v

(t)dt=2*(v+n)r( v )sm*<l>Cn*(co82<t>).
<f>

This formula

may be still further

specialised by taking
(2),

equal to tt or tt.

* Wiener Sitzungtbcrichte, lxxv.

(1877), pp.

218222.

294
9*6.

THEORY OF BESSEL FUNCTIONS


The genesis of Lommel's* polynomial

[CHAP. IX

RmiV (z).

The recurrence formula

Jv+1 (z) = (2v/z) Jv (z) - /_, (z)


may
v

obviously be used to express

Jv+m {z)

linearly in terms of

Jv (z)

and
to

J ^{z)\

and the

coefficients in this linear relation are

polynomials in \jz

which are known as Lommel's polynomials.


obtain explicit expressions for them.

We

proceed to shew

how

The

result of eliminating

Jv+1 (z), Jv+3 (z),... Jv + m_


x

(z)

from the system of

equations

Jv+P+1 (z) is easily

{2 (v

+ p)/z] Jv+P (z) + Jv+P_

(z)

0,

(p = 0,

1,

. . .

m - 1)
=0.

seen to be
-2z- 1
(v
1,

^ + m(2),
0>

+ m-l),
-2z- l

1,

(v

+ m-2),

0,

0,

1,

0,

0, 0,
(z),

0,
0, 0,

Jp('),
Jv .
1

-22- 1
1

("

+ l)

(z)-(2v/z)Jy

0,

By expanding

Jv+m (z)

is

in cofactors of the first column, we see that the cofactor of unity; and the cofactor of (-)"*- 1 Jv {) is
i.

-2?- 1 (v + m-l),
1,

0,
1,

0,
0, 0,

-23~ 1 (v+m-2),
1,

-2-i(i/+i-3),

0,
0,

0,

_2*-J( + l),
1,

0,

-22-1

The

cofactor of (-)m

~1

/I^i (z) is this

determinant modified by the suppression

of the last

row and column.


denoted by the symbol (-)m RmtV (z); and called Lommel's polynomial. It is of degree m in 1/z
1

The
and
it is

cofactor of (-) w,
is

(z) is

Rm, (z), thus defined,

also of degree

in

v.

The
which
unity
;

Rmiy (z) is defined


it

suppressing the last row and column of the determinant by is equivalent to increasing v and diminishing by and so the cofactor of (-)m ~ ./_, (z) is (-) mr " 1 iC-i.r+i (z).

effect of

Hence

follows that

Jv+m (z) - J (z) Rmt


*

(z)

J_
v

( z)

R m_

+1 ( z )

= 0,

Math. Ann.

iv. (1871), pp.

108116.

9*6, 9*61]

ASSOCIATED POLYNOMIALS

295

that

is

to say

(1)

Jv+m (z) = Jv (z) Em>


is

(z)

- J^ (z) R m-h +1 (z).

It

easy to see that*

Rm

(z) is

the numerator of the last convergent of

the continued fraction

2*rl(v

+m~

1>

>-2z->(v+m-2)-2z-*(v + m-3)-...-2z->v-

-R,,(z) was denned by Lommel by means of equation (1). then derived an explicit expression for the coefficients in the polynomial by a somewhat elaborate induction ; it is, however, simpler to determine the coefficients by using the series for the product of two Bessel functions in the

The function

He

way which
It

will

be explained in

9'61.

had been observed by

Bessel, Berliner Abh., 1824, p. 32, that, in consequence of the

recurrence formulae, polynomials

A n . x (2),
2
2)

Z?n _i (2) exist

such that

'.<)-

-2

'

'^- r"

^n-l (*) JiW- *-!

M*

<*.

where

[cf.

962

(8)]

^n_i

(2)

BH (2) - A n (2) 5B _

(2)=

2K^...(2n-2f2n

'

It should be noticed that Graft and Crelierf use a notation which Lommel's notation ; they write equation (1) in the form

differs

from

Ja + >

W-^CT

<*>

Ja <*) " K>-1 <*> '.-1 (*>

9*61.

The

series for

Lommel's polynomial.
(z),

m It is easy to see that (-) J- r -m


shews that
(1)

qua function of the integer m,

satisfies

the same recurrence formulae as J,+m (z); and hence the analysis of

96

also

(-)* J_,_ (z)

= J.

(z)

Rm, v {z)

+ J.v+1 (z) R m- ltV+1 (5).


(z),

Multiply this equation by


It follows that
(2)

J^ (z) and 96 (1) by J"_+1

and add the

results.

J+m (z) J_+1 (z) + (-)m J- -m (z) /.-1 {z) = Rm (z) {Jy (z) J-^ (z) + /_ (z) /_, (z)}
,

2 sin vtt
Rm,v{z),
TTZ
* Cf. Chrystal, Algebra,
11.

(1900), p. 502.

t Ann. di Mat. (2) xxm. (1895), pp. tionen, n. (Bern, 1900), pp. 98109.

4565; Einleitung
131163.

in die Theorie der Bessel'sehen Funk-

t Ann. di Mat.

(2) xxiv. (1896),

pp.

296

THEORY OF BESSEL FUNCTIONS


32 (7).
But, by

[CHAP. IX

by

541, we have

j v+m {z)

j_ +1
X

w(z)

opir{v +

m+p+
+

l)r{ _

+ 2+p y
n)
1

J- v - m (z) Jv -

= 2
n~l

n\T(v- m + n + 1 V (v +
)

_ S
n =o n\

(~) (-

-)n

{\z)- m+tn

T(

m+n

1)

r(y + n)

+ i
when we
that
(jo

P =o(r/i

replace

??

in the

(|*y+g+* (-v+*- P + i) + p + l)!r(-i/ + 2+p)r(i/ + w+p + l)' last summation by m-f jo + l. Now it is clear
!

mw

(m-\p+l)\
and
so,

+ l)m+p+i = (m + 2p + 1) = (m + p + p\(m+p+\)\ p\

2)p

when we combine the

series for the products of the Bessel functions,

we

find that

2 sin vir

__

(-) m+w (-

m + n) n {\z)- m+m-

_ smvir

^~

<* (-)n (m
Mr

-n)\r(v + m- n) (%z)-m + m w!(m-2n)! r(v + w)

'

the terms for which n

> \m

vanish on account of the presence of the factor

in

+ n)n in the numerator. When v is not an integer, we


p
M =o

infer that

<* (-) {m

- n)\ r (p + m n!

w) {\z)- +

(i-2)!
-

r(i/
-

n)

But the
that

(\ n

drZ\~ m +-n

R m>v (z) is
so,

and
for

m (z), by means of a determinant, shews a continuous function of v for all values of v, integral or not by an obvious limiting process, we infer that (3) is a valid expression
original definition of

even when v is an integer. be necessary to replace the quotient

R mv (z)

When
T(-

is

a negative integer

it

ma}'

T(v
in part of the series.

+ m-n)
y
y

-n + l)

T(v+n)

'

r(-v-m + n + l)

The series (3) was given by Lommel, Math. Ann. iv. (1871), pp. 108111; an equivalent result, in a different notation, had, however, been published by him ten years earlier,
Archiv der Math,

und

Phys. xxxvn. (1861), pp. 354

355.

An
(4)

interesting result, depending on the equivalence of the quotients just


first

mentioned, was

noticed by Graf*, namely that

RmA z = (~)m Rm, -,-.+!


) *

(*)

Ann.

di

Mat.

(2) xxin. (1895), p. 56.

9'62]

ASSOCIATED POLYNOMIALS
Pochhammer
.

297

In the notation of
(5)

(cf. 4*4,

4*42),
;

we have
z*).

Rm (z) = (v) m {% z)~m J\ (\-\ m, - \ m v,~ m, l-v-m; Since R mv {z)/z is a linear combination of products of cylinder functions of
+
2
{(i;
2
2
}

m and v 1, it follows from 54 that it is annihilated by the operator + m y + (v~ l) ^ + + my -{v- \Tf] + 4s (^ + 3^ + 2) where ^ = z (d/dz) and so Rmv (z) is a solution of the differential equation [(* + m) (* + 2v + m -2)($-2v- m) (^ - m - 2)] y (6)
orders v

O* -

}(i/

+ 4^-^(^+l)3/-0.
equation equivalent to this was stated by Hurwitz, Math. Ann. xxxm. (1889), p. 251; and a lengthy proof of it was given by Nielsen, Ann. di Mat. (3) vi. (1901), pp. 332 334 ; a simple proof, differing from the proof just given, may be obtained from formula (5).

An

9*62.

Various properties of Lommel's polynomial.

We

proceed to enumerate some theorems concerning

Rm<v (z),

which were

published by

Lommel

in his

memoir of 1871.

In the first place, 9*6 (1) holds if the Bessel functions are replaced by any other functions satisfying the same recurrence formulae and, in particular,
;

(1)

Yv+m (z)= Yv (z)R m v {z),

Y^^R^^iz),

whence
(2)

it

follows that

Yv+m (z) J (,) - J^ m ( Z ) Y^ (z) = R m {z) Yv {z) J _, (z) - Jv (z) Yv ^(z)}= - 2R m<v (z)l(7rz).
,

961 (2), take m to be an even integer replace m by 2m, and v by v - m. The equation then becomes (3) Jv+m (z)Jm+1 _ v (z) + J_ v _ m (z)J_m _ i+v (z) = 2 (-)m sin vir. R^m v - m {z)\{TTz\ and, in the special case v = , we get
Next, in

;
,

(4)

J'^ (z) + J3_ m _


to say

(z)

= 2 (-)- R2m>h ^n (z)/(ttz),


n)!} 8 .n!
_

that

is

(5)

j Wi ( + ./._
is is

2 5 vr~v-)i{*-w W - t^w=o
|(ra

This
order

the special case of the asymptotic expansion of 7*51 when the

half of an odd integer.

In particular, we have

()

J>

M + y_ W - 2

(i

+ ),

298
Formula
(5)

THEORY OF BESSEL FUNCTIONS


was published in 1870 by Lommel*, who derived
#

[CHAP. IX
it

at that time

by a

direct multiplication of the expansions ( 3 4)

followed by a

somewhat lengthy induction to determine the

coefficients in the product.

As

special cases of

96 (1) and

9"61 (1),

we have
cos z
/
.

Jm+i (*)
]

Y sin z.RmA (z) /

(J

# w_

li

(z),

(-)'"
[

J-m -i (z)

2 \4

cos

z.Rm ,i (z) + ^

2 \*
sin *
.

i^

lf

(*).

By
(8)

squaring and adding

we deduce from

(4) thatf

R\
if,

(z)

+ R*m-,, i (z) = (-)'" R, i-m (4


we
replace

Finally,

in 9*61 (2),

m by the
v

odd integer 2ra

and then

replace v by v
(9)

m, we

get
-m-i (f) = 2 (-)" sin vTrR%m+Ai _m (z)/(ire).
(1901), p. 23, is that

Jv+m+1 (z) J- v+m+1 (z) - J-y-m-i (z) J


interesting result, pointed out

An
if

by
n

Nielsen,
v+

.4 ran. rfi J/a*. (3) v.

we have any
z,

identity of the type 2

algebraic in

we can

at once infer

m=o the two

fm (z)J
2

m (z) = 0, where

the functions

fm (z)

are

identities
/, (*) i2m _!, +

2 fm {z) m=0

(*)s0,
in

by writing the postulated identity


s

m=0 the form

(*)

= 0,

m=0 and observing that, by 474 combined with


algebraic function.
pp. 331

/.w (/,(i)Ji;,()-/,.i(i)ii..i, t i(*)}^


3*2 (3), the quotient

Jv -i(z)lJv ()

is

not an

Nielsen points out in this memoir, and

its sequel, ibid. (3) vi. (1901),

340, that this result leads to many interesting expansions in series of Lommel's polynomials; some of these formulae will be found in his Handbuch der Theorie der Cylinderfunktionen (Leipzig, 1904), but they do not seem to be of sufficient practical

importance to justify their insertion here.

9*63.

Recurrence formulae for Lommel's polynomial.

In the fundamental formula

Jv+m 0) = J*
replace

(*)

Rm, v (?)

~ J*-\ (*) Rrn-1, H-l (z),


for

m and
we

by

m+1

and

on comparing the two expressions

Jv +m (z),

see that

,/_! (Z) {i? m _ liH -i (*)


*

+ Rm+1,,-1 (*)} =

{/",

(*)

+ /,-(*)}
(1871), pp.

Rm,u(z).

Math. Ann. n. (1870), pp. 627632. t This result was obtained by Lommel, Math. Ann.

iv.

115116.

9-63]

ASSOCIATED POLYNOMIALS

299
it is

Divide by
(1)

J ^ (z), which
v

is

not identically zero, and


w - l <*)

apparent that

Rm-1,,+1 (*)

+ Rm+l

^- Rm
}

v (z).

To
and

obtain another recurrence formula,

we

replace

m in 9 62 (2)
-

by

w+1

m 1, and use the recurrence formula connecting Bessel orders v + m 1, v + m and v + m + 1 it is then seen that 2 iC->, (*) + R m+ (s) = ^^1 Rm> (z\ (2) z
;

functions of

i,

and hence, by combining


(3)

(1)

and

(2),

R m-i

(z)

+ R m +i, v (s) Rm-i, v+i (t) Rm+i, v-i (*) =

(m +
_

Rm, v (*)

Again, write

962 (2)

in the form
\z-i F_,

-z-*-*R mtV
7T

(z)={z-- m Jv+m (z)}


it.

(*)}-{*--'

Y9m (s)}[r-*J w

(z)},

and

differentiate

We
=

deduce that
~

(4)

R'm, v (^)

z
(2),

Rm,
,

(*)

+ Rm+i, *-i (*) ~ Rm+i, v (^)

and

so,

by

(3), (1)

and

(5)

R'm>

(s)

= i2 TO
Z

,,

(z)

+ Rm (z)

lt

(z)

- Rm-i, v+i (z)>


y

(6)

R' m< v (z)

=
z

Rmt

R m-h +1 (z) Rm+


)

lt

(z),

(7)

R'mA Z) = ~

T"" Z

RmA Z

+ Rm+hv-i (z ) + Rm-h* ( z

)-

of these formulae were given by Lommel, Math. Ann. IV. (1871), pp. 113 due to Nielsen, Ann. di Mat. (3) VI. (1901), p. 332; formula (2) has been used by Porter, Annals of Math. (2) in. (1901), p. 66, in discussing the zeros of Rm> v (z).

The majority

116, but (6) is

It is evident that (2)

may be used

to define
is

Rmv (z),
JLv

when the parameter

m is zero or a negative
of m,

integer; thus, if (2)

to hold for all integral values

we

find in succession from the formulae

R%v\ z) =
that
(8)

4,v(v

+ l)
1>

~i

p 3 )= ^i,A
.

0<

y(z)=l,

R_ltV (z) = 0, R^y(z) = -1,


y

and hence generally, by induction,


(9)

R- m
we compare

(Z)

= (-)* Rm-^y (z).


(2) xxiii. (1895), p. 59.

This formula was given by Graf, Ann. di Mat.

If

(9) with Graf's other formula, 9*61 (4),


1

we

find that

(10)

Rm,y(z) = (-)" R-r^^y (z) = - R-.m^ v+m+1 {z) = (-)" Rm ,.y-m +i (z).
the functions of negative parameter are defined by equation
(9), all

When

the

formulae (1)

(7) are true for negative as well as positive values of m.

300
9'64.

THEORY OF BESSEL FUNCTIONS


Three-term relations connecting

[CHAP. IX

Lommel

polynomials.

It

is

possible to deduce from the recurrence formulae a class of relations

which has been discussed by Crelier*. The relations were obtained by Crelier from the theory of continued fractions.

shews that Jv+m (z) and Rmv (z), qua functions same recurrence formula connecting three contiguous functions; and so a repetition of the arguments of 9*6 (modified by replacing the Bessel functions by the appropriate Lommel polynomials) shews that
First observe that 9*63 (2)
satisfy precisely the

of

in,

( 1)

Rm+n

(z)

R m> y(Z) R n< +m (Z) J m


in
is

it

\Z)

Kn

i,

v+m+i

\Z)'

Next in 9*63(2) replace m by (m + v)/z from the two equations it


;

and v by v+1, and eliminate

then seen that

**t, v

\Z) -tim.v+i

z)

~~ -^"in+i.v z \ )

Rm-i,v+i \ z )

Rm1, v \Z) Rm~l


and so the value of the function on the left m 1. It is consequently independent of m is unity, we have Crelier's formula
(2)
is
;

v+1 \ z )

Rm, v\ z

Rm2, v+1 \ z )>

unaffected by changing

and, since its value

when

m into m=

Rm< (z) Rm< v+1 (z) - R m+hv (z)


(cf.

m-

lt

v+1 (z)

1,

a result essentially due to Bessel

96)

in the special case v

= 0.
+ n,

More generally, if in 9*63 (2) we had we should have similarly found that
R,n,
v

replaced

m by m n

and v by v

\Z) -tim-n+i, v+n \Z)

~ Rm+i, v \ z ) Rmn, v+h \ z ) = ivm _ \Z) xv m_ n + n yZ) iim


li

v\Z) timni, v+n \z h

and so the value of the function on the left m 1. It is consequently independent of m is R n - i<v (z\ we find from 9*63(10) that
(3)

is
;

unaffected by changing

and since

its

value

when

m into m=n

Rm>

(z)

Rm

-.

n+ i +n (z)
t

Rm +i, v (z) Rm-n, v+n \ z ) = -RjiLommel +.


1 in this equation,
v

1,

\Z),

a result given in a different form by

Replace
(4)

m and wbym-1
v

and n +

and

it is
,v

found that
J-

Rm-\

(z)

#m _n-i, K+M+i (z) Rm,


p

(z)

Rm-n-2, v+n+1 \ z )~

If

we

rewrite this equation with

in place of n

and

R n (Z eliminate Rm ^hv (z)

be-

tween the two equations, we see that

n>

(z) R m -P - v+p+i (z) RPt (z) rt m _w _j v + n+i \z) Rm, v (z) [ Rm-p-2, v+p+i (z) Rm-n-i, v+n+1 ( z ) ~~ Rm-n-l, v+n+i ( z ) Rm-p-^ v +p+i \Z)] Rm, v \Z) Rn-pi, v+p+i \ z
Jt
i

)>

by

(3).

If we

transform the second factor of each term by means of 9*63 (10),


cit.

we obtain
(5)

Crelier's result (loc.

p. 143),
,

ni

(z) jRp_ m_i,+ m +i (z)

Rp
136

\Z)

Rn-mi, i/+m+i \ z )

= Rfn v \Z) Rpni,v+n+i \ z


t

)'

Ann. di Mat.

(2) xxiv. (1896), p.

et seq.

Math Ann
-

Iv<

1871 )'

P-

U5,

9'64]

ASSOCIATED POLYNOMIALS
is

301
;

This

the most general linear relation of the types considered by Crelier

it

connects any three polynomials

R m>v (z) Rn>v (z), RPiV (z) which have


)

the same

parameter v and the same argument symmetrically


(6)

z.

The formula may be written more

Rn, v {z) Rp-m-i, v+m+i (z)


to say

+ Rp, v (z) Rm -n-\, v+n+i (z) + Rm v ( z ) Rn-p-i, K+p+i (z) = 0,


,

that

is

(7)

2 Rn,Az)Rp- m -hv+m+ i(z) = 0.


m, n, p

similar

result

may be
by

obtained which connects any three Bessel


integers.

functions whose orders differ

If

we

eliminate

Jv+m -i (z)

between

the equations*

"c+n \Z)
^Jy+p \Z)

= J+m \Z) K>nm,v+m \Z) Jy+mi \ z ) -">n m-i, p+m+i (*)>


Jy+m \Z)
ltpm,v+m, \ z )

J v+m-i \Z) Rpm-\,t>+m+\ \Z),

we
/

find that

M-n (*) Rp-m-i,v+m+i (z)

Jv +p (z) R n -m-i,v+m+i (z) ~ J +m \Z) L-ftn-m, v+m \Z) -ftp-m-i, v+m+i \Z) -Hp-m, f+m {Z) Rnmi, v+m+i (*)] *= J+m (Z) Rp-n-i, f+n+i (z)
'

the last expression

is

obtained from a special case of (5) derived by replacing


respectively.

m,

n, p,

by

0,

m, p m, v + m

It follows that
(

8)

2 Jv+n (z) Up-TO-i,


m,n,p

+m+1 (*)

= 0,

and obviously we can prove the more general equation


m,n,p

where

^ denotes any cylinder function.

The last two formulae seem never to have been previously stated explicitly, though Graf and Gubler hint at the existence of such equations, Einleitung in die Theorie der BmeVschen Funktionm, n. (Bern, 1900), pp. 108, 109.
[Note.
If

we eliminate J^-i
i

z)

from the equations


1

J+m () = J* (*) Rm, v () - Jv we

()

Rm (*)

1.

p+

1 (*),

Wv + m-l () s= ^k (*) Rm-1, v ()/_!


and use and
(2) to simplify the resulting equation,

Rm-2, v + 1

(*)i

find that
1 (*)>

Jv(z)= -Jv + m (?) Rm-2, r + l() + ^ + -l (*) ^m-1. k +


so, replacing v

by

- m, we have
(')

A--m()= -Jy(t) Rm-2,v-m + l(z)+Jv-i(*) ^w-l,v-m + l

By
that

using 9-63 (10),

we deduce that
()

Jv-m i?)=Jv (*) R-m, v


is to

-Jv -l () R-m-h v +

(*),

say that the equation 9*6 (1), which has hitherto been considered only for positive values of the parameter wi, is still true for negative values.]
* It is supposed temporarily that result is symmetrical, this restriction

m is the

smallest of the integers m, n,

may be removed.

p ; but since the final See also the note at the end of the section.

302
9*65.

THEORY OF BESSEL FUNCTIONS


Hurwitz' limit of a

[CHAP. IX

Lommel polynomial.

We
(1) V '

shall

now prove

that
lira
<

*-.

**)^.^(*> = j
r(i/

+ m + l)

(,).

This result was applied by Hurwitz, Math. Ann. xxxm. (1889), pp. 250 252, to discuss Jv (z) when v has an assigned real value ( 15*27). It has also been examined by Graf, Ann. di Mat. (2) xxiii. (1895), pp. 49 52, and by Crelier, Bern " Mittheilungen, 1897, pp. 9296.
the reality of the zeros of

From

961

(3)

we have
l)

r(v + m +

n= on\V{v

(m-n)\r(v + m-n + l) + n + 1)' (m-2n)\r(p + m + l)


K+an

'

Now

write

(m n)\T
so that

(v + m n + 1) _ (m-2n)!I> + m + l) =

Q ( ' n)

>

(m, n)

7^

(m

(i/

+ ra)(i' +

w)(ra r-P

1)

...

(m - 2n +
v

m-l)...(i/

-.v

+ m-n +
|

1) TV 1)

>

If

now

N be the greatest integer contained in


ri) is

numerator of 6 (ra,

numerically less

then each factor in the than the corresponding factor in the


,
\

denominator, provided that n

> N.
\0{m,n)\<l,

Hence, when

n> N,

and

m > 2N,
value,

while,

when n has any fixed

lim

(m, n)

1.

Since
is

(->;(*'>+"

absolutely convergent,

lim

it

follows from Tannery's

theorem* that

hZY+"
is

e <m, n)

= i

Jf^Tuniform in any bounded domain t

and the theorem of Hurwitz

established.

Again, since the convergence of 2

=o!

ru++i)
,

rr is

of values of z (by the test due to Weierstrass),

it

follows that the convergence of

$z)" +m
to its limit
is

/^

+ 1 (2)/r(

+ m+l)

also uniform in

any bounded domain of values of z.

* Cf. Bromwich, Theory of Ivjinite Series, 49. f An arbitrarily small region of which the origin excluded from this domain when R (v) < 0.

is

an internal point must obviously be

9*65, 9*7]

ASSOCIATED POLYNOMIALS
the theorem of Hurwitz
it is

303

From

easy to derive an infinite continued


=jfe

fraction for

Jv-

(z)/Jv (z).

For,

when

Jv (z)

0,

we have

^# =
by

lim

r^t^l
L^-l/I^M-l,
m_
,

lim

963 (1). On

carrying out the process of reduction and noticing that

R* v+m-Az)
^i,+m(*)

=2

_
2z
l

1
(v

+ m)

we

find that

^Hi^-2^
ttm,, + i (z)

(i/

+ 1) z- 1

(i/

1 2) r-

. . .

-2

+ m) z~*

and hence
(2 )

^W-J
v

(z)

2(v

+ l)z- -2(v + 2)z1

-...
last

This procedure avoids the necessity of proving directly that, when m-*-ao, the element of the continued fraction

Jv {z)
may
be neglected
;

2( + l)*--...-2>+)*-iis

Jv + m (z)
p. 52.

the method

due to Graf, Ann. di Mat.

(2)

xxnr. (1895),

9*7.

The modified notation for Lommel polynomials.

In order to discuss properties of the zeros of Lommel polynomials, it is convenient to follow Hurwitz by making a change in the notation, for the reason that Lommel polynomials contain only alternate powers of the variable.
Accordingly we define the modified
equation *

Lommel

polynomial gm>v
l)*"
*

(z)

by the

m
(2)

(1)
so that

n ^'

M-^ W- :
tt

ro-

C ntB

(-)*r(

+ m-n + I> + n+l)

Rm

,r +1

(z)

- (hz)~ gmA*) 9*63,

By making
(3)

the requisite changes in notation in

964, the reader

will easily obtain the following

formulae
[

0+i,r (*)

(4)
(5)

= (" + m + l)gm>v (z) - zgm_hv (z), #,+,, r-i (z) = vgm (z) - zg^, +1 (z),
,

963 (2)] [ 963 (1 )]


9-63(7)]

IF*

bTz

P^'W = *9^A*) +gm+h


( z ))

,- 1 (z),

[J

(6 )

^m ^ (,-m-i gmv
gm
,

<,

m+1 _, (*)
+1

gm+1<v (z),
0<

9-63 (4)]

(7)

(z) gm+i< +1 (z)

- gm+2t (z) gm _

1%

(z) = z m g

(z) glt +m+1 (z).

[A
* This notation differs in

special case of 9*64 (5).]

unimportant details from the notation used by Hurwitz.

304

THEORY OF BESSEL FUNCTIONS


will

[CHAP. IX

These results

be required in the sequel


of
all

it will

not be necessary to write

down the analogues

the other formulae of

9'6

964.
(3) is of
2>

The result of eliminating alternate functions from the system some importance. The eliminant is
(f

+ m) gm+2> (z) = cm (z) gm% (z) - (v + m + 2) z*gm _


cm (z)

(z),

where

(v

+ m + 1) \{v + m) (v + m + 2) 2z).

We

thus obtain the set of equations


f

(v

+ 2)gi<v (z) = ca (z)g2>y (z)- (v + 4>)z'g0tV (z), (z) = c4 (z) gitV (z) - (v + 6) *g%w (z), (v + 4) g
St

(8) (v

2s) gv+ % y (z)

= Ca, (z)

<7M)

(z)

- {v + 2s + 2) z g^ {z),
3
t

(y

+ 2m - 2)gvn,* (z) = <hm-* (*) 3W-9... (*) - (v + 2m) z g2m-*, , (z).


2

9*71.

The reality of the zeros of g^^iz) when v exceeds

2.
> 2,
the

We

shall

now

give Hurwitz' proof of his theorem* that when v

zeros of gtm,v{z) are

1 > v > 2,

a ^ real; and also that they are in which case one of them is negative.
is

all positive, except

when

After observing that gmt,v(z)

shew that the

set of functions g*m,*(z), 3W-a.*(s)>

a polynomial in z of degree m, we shall ... g2iV (z), g^ v {z) form a set


(8),

of Sturm's functions.

Sufficient conditions for this to

existence of the set of relations

97

be the case are (i) the combined with (ii) the theorem that

the real zeros of gvm-^vi?) alternate with those oi gimv {z).

To prove that the


9*m,( z )l9im--2,*(z )
is

zeros alternate, it

is sufficient

to prove that the quotient

a monotonic function of the real variable z, except at the where the quotient is discontinuous. denominator, zeros of the

We
where

have

f^ %9
,,

(z)

*M - j dz { {gtm-^Kz))
%

w(

= - 389**, *-*,

Wi r - 9r.* (z) g\ (*) - 9:* (*)#V (*);


9*7 (3) it follows that

and from

B9m-i.m-8 so that

* SK&m-s,*^ + iv + 2m - 2) g\n^y (z),


m
X

i {v + 2r) z- tf^ m,^* - g>*n-*, (z) + (v + 2m) r=l


and
therefore, if

9 (z),

m > 1,

5KK2IW,*W_8

is

expressible as a

sum

of positive terms

when

> 2.
* Math. Ann. xxxin. (1889), pp.

254256.

9-71, 9-72]

ASSOCIATED POLYNOMIALS

305

The monotonic property is therefore established, and it is obvious from a graph that the real zeros of g^^, (z) separate those of g2mv (z).
It follows from Sturm's

theorem that the number of zeros of g^n^iz) on

any

interval of the real axis is the excess of the

number of alternations

of sign

in the set of expressions g*m tV {z),

g^^

(z),

...,g0l/ (z) &t the right-hand

end of

the interval over the number of alternations at the left-hand end.


of zeros is the excess and not the deficiency in that the a decreasing function, and not an increasing function of z, as in the usual version of Sturm's theorem. See Burnside and Panton, Theory of Equations,

The reason why the number


(z)/2,_2, (z) is

quotient gr^

i.

(1918), 96.

The arrangements

of signs for the set of functions

when

z has the values

oo

0, oo

are as follows

2m

2m- 2

2m -4

00

+
+

+
+
~ (~)m l
(

+
+
_)m-2

...

...

+ +

00

(->*

...

The upper
negative
;

or lower signs are to be taken according as v


is

+1

is

positive or

and the truth of Hurwitz' theorem

obvious from an inspection

of this Table.

9*72.

Negative zeros of g<un, (z) when

p< 2.

Let

be

less

than

2,

and

let

the positive integer * be defined by the

inequalities

-2s>v>-2s-2.
It will

now be shewn

that*,

has no negative zero; but

that,

when v lies between 2s and 2s when v lies between 2* 1 and 2s


that, in

1, 2,

gsm ,(z)
t

has one negative zero.


that v

Provided

each case,

m is taken to
any) of

g-mtV (z) be so large

+ 2m

is positive.

It will first be

shewn that the negative zeros

(if

g^^z)

alternate

with those of g*m -%,y{?)'


* This proof differs

from the proof given by Hurwitz

see Proc.

London Math.

Soc. (2) xix.

(1921), pp.

266272.

306

THEORY OF BESSEL FUNCTIONS


By means
of the formulae quoted in 9*7,
it is

[CHAP. IX

clear that

= ^2m-2, zg*m-i, + 9*m+i, v-i( z )} - 9*m, ( z ) l^am-i, 1(2) ~ S'am-i, -(*)} = (1/ + 2m)0 tM -i,r() + gmi~2, v {z)g*m,v-\(z )-g*m-i,v-i(z)gvmAz) = (l/ + 2m)5r 2m_ (^) - ^nrWft.rtW-lW
{

1>

>o,
provided that
i>

+ 2m

is

positive

and z

is

negative.

Therefore, in the circum-

stances postulated, the quotient

is

a decreasing function, and the alternation of the zeros

is

evident.

The

existence of the system of equations 9*7 (8)

now shews

that the set

of functions
g*m,v(z), g*m-2, v(z)>

g+t, *( z )>

.g*,

*( z )>

-g**-*A z)>
form a set of Sturm's functions.

+g*-4,v( z ),

(~Ygo,v( z)

The

signs of these functions +, +,

when

is

oo

are
(-)*,

....

+,+,-,+,

...,

and

there are s alternations of sign.

When

is

zero, the

signs of the

functions are
, ,
-.,

+,-,+,

,(-)*,

the upper signs being taken when 2s > v > 2s 1, and the lower signs being taken when 2s l>v> 2s 2; there are s and s + 1 alternations of sign in the respective cases. Hence, when 2s >v > 2s 1, #,() ^ as no negative zero but when 2s 1 > v> 2s 2, gzm,v(z ) nas one negative
;

zero.

The theorem

stated

is

therefore proved.

9*73.

Positive

and complex zeros of g^^^iz) when v< by the

2.

As

in 972, define the positive integer s

inequalities

- 2s >
It will

v>

2s

2.
between

now be shewn*

that when v
but that,

lies between.
lies

has m 2s positive zeros;


<Jvm v( z )

when v

2s and - 2s 1, gtm, v (z) 2s 1 and 2s 2,


each case,

has

m2sl positive zeros.

Provided

that, in

mis so

large

that

m+v is positive.
;

* This proof la of a more elementary character than the proof given by Hurwitz paper cited in 9*72.

see the

9-73]

ASSOCIATED POLYNOMIALS
first place, it

307

In the

follows from Descartes' rule of signs that, in each case,

9tm,v{z) cannot have more than the specified number of positive zeros. when v lies between 2s and 2s 1, the signs of the coefficients of
1, z,

For,

z\ ...,z*>, z+\ z-+\ *+, +, +, +,


...,

...,

zm in

ffsm v (z)
,

are

+, -, +, -,

...,

(-);

and since there are m 2s alternations of sign, there cannot be more than m 2s positive zeros. When v lies between 2* 1 and 2* 2 the corresponding set of signs
is

_ _ _
and since there are
than

_ _

a.

( V-

m 2sl

alternations of sign there cannot be

more

w 2s

positive zeros.

that there are as

Next, we shall prove by induction from the system of equations 9*7 (8) many as the specified number of positive zeros.

When

v lies

between - 2* and

alternations of sign (being all

the coefficients in gVtV (z) have no +) and so this function has no positive zeros.
1,

2s

On

the other hand

0+ 2 ,(0) > 0, gu+3f (+


,

ao )
;

=-

and so gu+2lV (z) has one positive zero, aM say and, by reasoning already given, it has no other positive zeros. Next, take <7+4,i,(s); from 9*7(8) it follows that its signs at 0, a M + ao are +,, + hence it has two positive zeros, and by the reasoning already given it has no others. The process of induction (whereby
, ;

we prove
function)

that the zeros of each function separate those of the succeeding


is

now

evident,

and we

infer that

gm

y (z)

has

m 2s positive zeros,
2,

and no more.
Again, when v
-

5 4+s,i'(^) no positive zeros.

lies between 2s 1 and have no alternations in sign (being all

2s
),

the coefficients in

and

so this function has

On

the other hand

#+4,r(0)

< 0,

5T +4(K (+ 00 )

=+

CO

and so g4a +4t ,(z) has one has no other positive zero.

positive zero,

and by the reasoning already given

it

By

appropriate modifications of the preceding reasoning

we prove

in suc-

cession that gu+^viz),

gv+8tV (z),

...

have

2, 3, ...

positive zeros,

and in general

that g2m,r(z) has

m 2s 1
v<

positive zeros.

By combining

these results with the result of


2,

972, we obtain Hurwitz'

theorem, that, when

and mis so large


is the integer

that ni

v is positive, gsmfV (z)

has 2s complex zeros, where s

such that
2.

- 2s > v > - 2s -

CHAPTER X
FUNCTIONS ASSOCIATED WITH BESSEL FUNCTIONS
10* 1.

The functions J(#) and E,

(^) investigated

by Anger and H. F.

Weber.

In this chapter we shall examine the properties of various functions whose


definitions are suggested
shall first investigate functions defined

by certain representations of Bessel functions. by integrals resembling Bessel's

We
inte-

gral

and Poisson's

integral, and, after discussing the properties of several

functions connected with

Yn (z) we shall study a class of functions, first defined


case.
(z), is

by Lommel, of which Bessel functions are a particular

The

first

function to be examined, J,

suggested by Bessel's integral.

It is defined

by the equation
J (z)
-=

(1)

7T J o

f 'cos (v$

- z sin 6) d6.
when
v has the integral value n.

This function obviously reduces to


It follows from 6*2 (4) that,

Jn (z)

an integer, the two functions are distinct. A function of the same type as J, (z) was studied by Anger*, but he took the upper limit of the integral to be 27r; and the function J,(z) is
v is not

when

conveniently described as Anger's function of argument z and order

v.

A
(2)

similar function

was discussed

later

by H. F. Weberf, and he also

investigated the function

E (z)

defined by the equation

E (z) = - ('sin (v0 - z sin 0) dd.

In connexion with this function reference should also be made to researches by Lommel, Math. Ann. xvi. (1880), pp. 183-208.
It

may be

noted that the function

Jo cussed by Anger is easily expressible in terms of J (2) and E, 2tt 6 in the right-hand half of the range of integration, we get

2m
1

/"2"-

cos {?6

- z sin

6)

dd which was actually


(z)
;

dis-

for, if

we

replace 6

by

*T

/"2t
/

cos(v0-2Sin0)cW=

fn
/

*w J
2

cos (p6 - z ain 6) dd + 7T.

If' cos(2vir-vd+zain8)dd
I

2,1t

cos

vit

(z)

+ sin

vit cos

vn

E (z).

* Neueste Schriften der Naturf. Ges. in Danzig, v. (1855), pp. 1

29. It was shewn by Poisson that

Vv
to give Anger's

cos (vO - z sin 6) d0 = (z


(cf.

v)

sin v jt,

Additions a la Conn, des Temps, 1836, p. 15 name to the function.

10*12), but as

he did no more it seems reasonable


factor 1/x in his defi-

t ZUrich Vierteljahrsschrift, xxiv. (1879), pp. 33


nition of

76.

Weber omits the

Ev (z).

10-1]

ASSOCIATED FUNCTIONS
in ascending powers of
:

309
z,

To expand J (z) and E (z)


in the integrals
sinm 6 sin

write %tr

<f>

for

and proceed thus

vdd$ =

cosm

<f>

sin (^vtt

+ v<f>) d(f>

JO

J-Jir

2 sin \vir
-'o

cos"* < cos i><d</>

it

sin ^yn-

by a formula due

to

Cauchy*.

In like manner,
Sin TO

0COSl>0d0=_-=7-;

But, evidently,

J(*)=- 2
Tr

v
;

/c m =o (2m)! Jo

:,

sin2m

0cosv0d0+ - 2 hr 7r m=0 (2m +

2* 1 rrr sin l)!Jo

sin vddd,

so that
(3)

J r (^) = cos^7r 2

jTT

\:l
co

Tx

(_)m/l^)m+i
=r.

+ Sill &V7T 2
and similarly

T Ti

(4)

B(^) =

sinii/7r

2
TO -

"7 ^ r(m-| /+l)r(m +


.->I

'

i/

+ l)
1

vVi/
i

-cos*i/7r

2 =-

5--

These results may be written in the alternative forms


,-.
1
'

( AZ) ~

._ sini/7r
inr

I"

z*

|_

+ -i^

z^

z6
2

"1

(2

-^)(4

-i/

)~(2

-i/ )(4 -i/ )(6


2
5

-y )
2

"'J

+
(6)

sin

w["
L
12

7T

-;2_ (l 2 -'2 )(3 2 -va ) + (l 2 -i'2 )(3 a -i/2 )(5 a -y2 )~'

.2?

1 s E,(,)= -i:^
VTT

^[l-^L+ r 2 -i>
"

^
2
i/

2 2 2 (2 -i/ )(4 -i/2)

,-...1
*"j
z*

cos virY
tt"

z
2 2

3
2

[l 2 -!/2

~ (l - v ) (3 -

+ (l

-i/2 )(32 -i;a )(5 8 -^)

'

Results equivalent to these were given by Anger and Weber.

memoir) in a
17,

The formula corresponding to (5) was given by Anger (before the publication of his letter to Cauchy which was communicated to the French Academy on July
1854
*
;

see Comptes Rendu*, xxxix. (1854), pp. 128

135.

M&m.

sur les integrates difinies (Paris, 1825), p. 40.

Cf.

Modem Analysis,

p. 263.

310

THEORY OF BESSEL FUNCTIONS


For a reason which
will

[CHAP.

be apparent subsequently
Z3
iy
(

( 10*7), it is

convenient

to write
(7)
So,,

(*)

p^ ~ (p _ y
z2
yi
a ^2
2

3a _

^+

18

_ ^ (32 _ ^(52 _ ^ ~
2 2
i/

2?_

' ' >

(8)

_,,(*)

1 =-- +

z*

_ ^) ~

a(2

_ *) (4 ,

+
)

'

"

and, with this notation,


(9)
,,~x

we have
,
.

J,

(z)

sin vrr

v sin vir

s 0t v (z)

*_,, (*),

(10)

E (z) =
/
^

l+cosj/ir

,
So,

(^)

i/(l

cos i/7r)

*-i. , (z).

It is easy to

deduce the following formulae from these results


cos vB
.

(11)

cos (z sin

6)d6 =
dd

v sin vir

s_hv

(z),

Jo

(12)
Jo

sin vd

cos (z sin 6)

= - v (1 - cos

vjt)

s_i

(z),

(13)

sin v 6

sin (z sin 6)

d6

= sin vtt

s0< (*),

Jo
(14)

f cos Jo
Mir
I

i>0

sin (* sin 0)

d0 =(1

+ cosi/7r)
.

s0> (z),

(15)

cos

v<f>

cos (z cos <f>)d(f>

= v sin ^irr
= cos i/7r.s
,

S- h {z),

Jo
/in-

(16)

cos v<. sin (z cos <)d<

.(*).

Jo
Integrals

somewhat resembling the


[e

integrals discussed in this section,

namely

COS
sin

(ne-cos8)dd,
Sitzungsberichte, lvii. (2), (1868), pp.

have been examined by Unferdinger, Wiener


/oc
I

611620.

Also, Hardy, Messenger, xxxv. (1906), pp.


sin

158166, has investigated the integral


flO

(vO-zsin
equal to

6)

-r=-,

and has proved


according as v

that,

when

v is real, it is

\it

rj

Jn (z),

where

tf n

is 1,

or -

11

is positive, zero,

or negative.

10*11.

Weber s formulae connecting


from the formulae

his functions with


(9), (10),

Anger s functions.

It is evident
(1)

101
/

(15) and (16) that


<f>) d<f>,

J (z)

J_ (z) =

cos \vtt fi" cos


Tr

(2)

J (s)

- J _, (s) =

A.

gin ftvir

v<f>

cos (z cos

f*
I

cos

v<f>

sin (2 cos <)

d<f>,

1011, 10-12]

ASSOCIATED FUNCTIONS
E (2;) + E_ (.s) =
a

311

(3)

T T

Jo
.'0

cos

v<f>

sin (s cos

<f>) cUf>,

(4)

E (3)

E_ (s) =

^^

cos

v<f>

cos (2 cos

<) d<f>.

It follows on addition that

J, (z)
so that

= I cot |i/tt {E (*) - E_ (*)} - tan |i/tt {E (z) + E_ (5)}


sin vir
.

(5)

J (z)

= cos vir E (2) - E_ (z),


.
.

and similarly
(6)
sin vir
.

E (z) = J_v (z) cos vir J (2).


due to Weber.

The formulae

(5)

and

(6) are

10*12.

Recurrence formulae for J (z) and E (s).


functions of

The recurrence formulae which are satisfied by the Weber have been determined by Weber.
It is evident from the definite integrals that
_! (z)

Anger and

+ Jr+i (z)

- J (z)

=-

(cos

0--J

cos {v0

- z sin 0) d0

\* ^-{sm(v0-zam0))d0
irz J

d0

_
and
B_, (s)

2 sin vir
irz

+ E, +1 (z)

2i>

2 f'/ cos E (s) = f


J

iA -sin (v0 -zain0)d0


J

= _
It is also very easy to prove that

-J* {cos (i>0

z sin

0)}

d0

2(1 cos vir)


irz

fJ-,(*)- J+1 (*)-2J/(*) = 0,

]_, (z) - E+1 (z) - 2 E/ (*) = 0.


From
(1 )
/i
\

these results

we deduce the
t / \ +J r+1 (0) =
.

eight formulae
(z)
/ \

t / \ J,-! (*>

t J, Z
2l/

2 sin

w
TTZ

(2)

l v^{z)-J v+l {z) = 2J v'{z),


(^

(3)
(4)

+ v) J (z) = zJ v _, (z) + (sin w)/ir, (^ - v) J (z) = - zJ v+ (z) - (sin V7r)/7r,


i

312

THEORY OF BESSEL FUNCTIONS


L m(.) + miM)-m.(M)- *<

[CHAP.

(5)
(6) (7) (8)

-"\

B r _ (#)- r+1 (jr)-2


l

r (*),

(^ + v) E (z) = Z E_, (z) + (1 - COS 1/7T)/7T, O - v) E (z) = - ^B+ (z) - (1 - cos i/7r)/7r,
i

where S, as usual, stands

for z(d/dz).

Next we construct the


(^
2

differential equations; it is evident that

- v>) J (z) = O - v) \z J v _! (s) + (sin w)/ir} = z(^ + l v) J v _! (*) {v sin vji)\ir = z J v (z) + (z sin i>9r)/w - (v sin tnr)/7r,
2

so that
(9)

v.J.(,)-^~.
also

We

have

(^

- v*) E (z) = (*-v) {*_! (z) + (1 - cos i*r)/w} = Z (^ + 1 - v) E,., (2) (1 COS V7r)/7T = Z E (2) Z (1 + COS I/7r)/7T (1 cos v7t)/tt,
J/

1/

so that

(10)

v >B , ( ,)
(9)

p.

iJ!
7T

<'')""
.

7T

Formulae equivalent to
(1879), p. 47, respectively
;

and

(10)

were obtained by Anger, Neueste Schriften der


Vierteljahrsschrift, xxiv.

Naturf. Ges. in Danzig, v. (1855),

17 and by Weber, Zurich

formula

(9)

had been discovered

earlier

by Poisson

(cf.

10"1).

10*13.

Integrals expressible in terms of the functions of

Anger and

H. F. Weber.
It is evident

from the definitions that


i

(1)

J, (*)

E (z) = 7T

r exp {i(v$-z sin 0)} d$.


Jo

By means
is

of this result, combined with formulae obtained in 6'2 622, it possible to express numerous definite integrals in terms of the functions of

Bessel,

Anger and Weber. Thus, from

6*2 (4)

we have

(2) V '

-^ T e-"-i"h< dt = Sin V7T


result
is

{
l

J (z)

- Jv Or)},
jarg.r|

when jarg0|<^7r; the R(v)>Q.


Again, we have
(3) '

valid

when

= |7r,

provided that

f"
J

e"

-"inht dt =

-^ \Jsini/7r

(z)

- J- (*)},

10*13, 10*14]
so that, (4)

ASSOCIATED FUNCTIONS
(2)

313

when we combine

and

(3),

(* e~** inht cosh vtdt=\ir tan \ vir {J {z)


Jo

Jv (*)} - \ ir {E (z) +
J (z)}

F (a)},

(5)

f" e-zaiaht sinh


Jo

itf

d<

|tt cot i/tt [Jv (z)

- %w

{E (*)

+ Yv (z)\.
but

The

integral [

e-zcosht cosh

^^ has already been evaluated (63);

Jo

[" e-zco* ht smhvtdt


Jo

does not appear to be expressible in a simple form; its expansion in ascending powers of z can be obtained from the formula of 6*22 (4), 2siny7r M -z^ht sinh vtdt} I_ v (z) + Iv (z) = - PVcose cos v d6 + ( e
7

7TJo

7r

'O

but, since
[*
-

..

cosm
j

(-)m sinv7r

* co8 ***

- V(v + m)
(5) are

*
'

Fl {- m>

+ "2~

m_
;

i/

-f

ra

2-'

"V'
6*21,

*. the integral under consideration cannot be evaluated in any simple form

The formulae 9*33 we have


(6)

(2)

nugatory when v

is

an

integer,

but from

rent Jo

z * inht

dt

= l{Sn (z)-TrE n (z)-TrYn (z)},


{Sn (z)

(7)

I" e- M
Jo

- zainht

dt

= \ (-)n+1

+ 7rE n (z) + ir Yn (z)}.

The

associated integrals

08 tC l" e-* *(x cosh t)dt (xaiuh t)dt, ("e-^ am 8m v Jo Jo xx. Journal, (1885), p. 260. have been noticed by Coates, Quarterly

Various integrals of these types occur in researches on diffraction by a prism Whipple, Proc. London Math. Soc. (2) xvi. (1917), p. 106.

see, e.g.

10' 14.

Asymptotic expansions of A nger- Weber functions of large argument.

It follows

from

1013 (2) that,


and

in order to obtain the


|

asymptotic expansion

of

J v

(z)

when

\z\ is large

args| <%tt,

it is

sufficient to obtain the

asymptotic expansion of the integrals

Jo

out this investigation we shall first expand cosh sinh i/t/cosh t in a series of ascending powers of sinh t.

To

carry

i;/cosh

and

* See Anding, Sechsstellige Tafeln der Besselschen Funktionen imaginaren Arguments (Leipzig, Takeuchi, Tdhoku Math. 1911) [Jahrbuch Uber die Fortschritte der Math. 1911, pp. 493494], ard

Journal, xviii. (1920), pp. 295296.

11

W.

B. V.

314
If en

THEORY OF BESSEL FUNCTIONS


= v, we
have, after the
1

[CHAP.

manner of

7*4,
(

M+,l/+)

ui

i/M *

so that

cosh vt COshy<

cosh*

_ 1 f<+. !/+.!+) "~ fi"-*(f-l)d ~ 2tti 2w7 (^l)2 -4fsinh J


:

<

(+, i/+,

i+)

^ r j>-i 2i*

gm sjn h t
2*p

2^7

L-o

(?-l)2m+1

p sinh *
2

*
2

(f-l^-iftf-l^-^sinh

]dr*}

Now
(i+)f*+m-*d
$7rt J 27rtJ

r(i/

+ m + f)
r(m-t-.^
'

(^-l)am+1

r(|i/-m+).(2m)!
OT _ (-) cos^7r

^)r(m+|-^)
(2m)!

7T

and,

if

we take p
shewn
/(+, l/t+,l + )

so large that

R(p + $$v) > 0,


we
find that

and then take the contour

to be that
1

in Fig. 15 of 7 '4,

fp
1

+ * v ~irf^

2m

(f- 1)^" {(- iy

- 4fsinh 2 }
7r

~
Jo
1

+ 4<c (1 - x) sinh

'

If v

and

are real, the last expression

may be
<2p)I

written in the form

where

dx <

1, since 1

+ 4#

(1

x) sinh <^ 1.
2 j/)

It follows that,

when

R (p + \ ^

^ 0, we have
(2m)!
,

COSh
COSh

I/*

COS V7T
7T

m=o

+ ,, (zmi+M<irW (2 sinh
(2p)!

r]

For complex values of v and < this equation has to be modified by replacing < X ^ 1 by a less stringent condition, in a way with which the reader will be familiar in view of the similar analysis occurring in various sections of Chapter vn.
the condition
Similarly

we have
u h

_ u -* = _L

f( + ,l/M+)

2-iri-J

z *

k-w

1 J - _L_
\

-1/4

d,

10-14]
so that

ASSOCIATED FUNCTIONS

315

sinh vt

~ 2iri J sinh 2t
1

+ ,i+) 1_ r(u+,ilu

$.#
( - I) - 4f sinh
2

r(+, i/+, i+)

"as J
whence
sinh
vt
it

r L:
1

r p-i 22w* w sinh 2"1 t f

(t-ir

+a

2 p sinn 2* <r- 1>* {<r- ly -*r ***}]

"I

,*.

follows that, if

we take p

so large that

R (p + 1

\v) > 0, then

= sin^irn- p7T

(-)w

cosh*

|_t
2 ^^

r(m + 1 + \v)Y(m + l -jiQ ginh tym+1 (2 v (2m + 1)!

(-)^( P +
it

ui y)r( P + i-^)
(2p

+ l)!
- )" r(m+ *
1

sinh
'

+1 1

On

integrating these results,

follows that
< o

cosh

* r^.a~qgJg:
.

J
2

;^
2m+2

(ro+

"^
& y)

sinh vt

e-* 8inh * cfe~

sin^Tr * (-)OT r (w +

2
27T

+ jQ r (m + 1 -

OT= o

(f2)

If

i;

is real

and z

is positive,

these asymptotic expansions possess the


is

property that the remainder after p terms


cally less than,

of the

same sign

as,

and

is

numeri-

the (p

l)th term
(2)

when p
(3)
I2

is

so large that

R(p + l

$v) ^ 0.

It follows from
_
,

1013
v

and
1

combined with

10*11 (6) that

(1)

J r (*Wr (*) +

sin vir
trz

-"2

(I 2

~ v*) (32
v (22
a

L.....1
2

sin vir \v
trz
|_i

y(2-y)

?
r, 1

+
2
,

- v ) (4a - y )

?
(!*-*){-*)
;

"'J'
1

/ox (2)

- r (*)~-F(#) x v/n
/

+ CQBinr

^
[z

l
2

-"
2

ii-

-J
2

-cosier Vv
irz

y(2 2

-y )

y (2 2
'"

v*)
z*

(4

- i/ ) _
8

"I

z*

"J

These results were stated without proof by Weber, Zurich Vierteljahrstchri/t, xxiv. 48 and by Lommel, Math. Ann. xvi. (1880), pp. 186188. They were proved as special cases of much more general formulae by Nielsen, Handbuch der Theorie der Cylinderfunktionen (Leipzig, 1904), p. 228. The proof of this section does not seem to have
(1879), p.

been given previously.

Since the only singularities of cosh vtjcosht and sinh vt/cosht, qua functions of sinh t, are at sinh t = i, it is possible to change the contours of integration into curves in the tf-plane on which arg (sinh t) is a positive or negative acute
angle; and then
(1)

we deduce

in the usual
|

manner

(cf.

6*1) that the formulae

and

(2) are valid over the sector

arg z\<ir.

316
10*15.

THEORY OF BESSEL FUNCTIONS

[CHAP.

Asymptotic expansions of Anger- Weber functions of large order and

argument.

We
when
j

shall

now obtain asymptotic


|

expansions, of a type similar to the

expansions investigated in Chapter


v
|

viii,

which represent J (z) and E

(z)

and z are both


|

large. 10*13, it will

In view of the results obtained in

be adequate to obtain

asymptotic expansions of the two integrals


1 -

f*
/

e T M-*Bnh

fa

TTJo

As

in

Chapter

viii,

we write
v

z cosh (a + ift) = z cosh 7,


tri.

where
(I)

^ ft ^

it

and 7

is

not nearly equal* to

We

first

consider the integral

Wo
in which
t

lf vt ~za niit If* g ('<8hy+8inht)^ e~ dt =


^

""Jo

it is

supposed temporarily that v\z is positive.


t

When cosh 7 is positive,


to oo
;

cosh 7 + sinh t steadily increases from to 00 as shall take this function of t as a new variable t.
It is easy to

increases from

we

shew that

is

a monogenic function of t, except possibly

when

= (2n + 1) id cosh y sinh y + y cosh y,


is positive,
rri

where n
positive

is

an integer; and, when coshy


;

number
{

imaginary

iff),
is

none of these values of r is a real cosh y does not vanish, and, when y is a pure the singularities are on the imaginary axis and the origin is not one of
for,

when y is real, (2n + 1)


iti.

them

since

not equal to

The expansion

of dt/dr in ascending powers of t

is

T
where

m=
(ft
,
'

a-=2^j

f<

0+ >

^^
1

~ &S J

If

(0+) <fc
.jjm+i

and so a is the coefficient of I ft in the expansion of t-2-1 in ascending powers of t. In particular we have

a _

~l+cosh7'
8,3

ai

_ ~

2(1
225

cosh 7 )4

'

* ~ 24 (1 + cosh 7
cosh2 7

cosh 7

'

_
From

~"

54 cosh 7
we
are

4-

720(1+ cosh 7)10


8'3,

the general theorem of

now

in a position to write

down

the expansion

* Expansions valid near

y = iri are obtained at the end of

this section.

10-15]

ASSOCIATED FUNCTIONS
is

317

This expansion

valid

when
|

v/z is positive
j

it

has, so far,

been established
to cover the

on the hypothesis that

arg z

<

\tr,

but,

by

a process of swinging round the

contour in the T-plane, the range of validity

may be extended

domain in which arg


|

<

ir.

Next, we consider the modifications caused by abandoning the hypothesis that cosh 7 is real. If we write t = u + iv, the curve on which t is real has
for its

equation

u sinh a sin ft + v cosh a cos

#+

cosh

u sin

0.

The shape
of 8'61.

of this curve has to be

examined by methods resembling those


v).

For brevity we write

u sinh a
Since 4>
(u, v) is

sin

+ v cosh a cos /3 + cosh u sin v = <P (u,


> 0.

unaffected by a change of sign of both u and

a,

we

first

study the curve in which a


its centre.

It is evident that the curve has the origin as

Since
it

33> (u, v)ldu

= sinh a sin /9 + sinh u sin v,


value, 34>/3tt vanishes for only

follows that,
w,

value of

any assigned and so the equation in u


v has
<I>

when

one

(m, v)

=
is infinite

has, at most,

two

real

roots; and one of these

whenever

v is a

multiple of

it.

When
and,

> v > tr, we have


3>

oo

v)

= oo

4>

(+

oo
<1>

v)

= oo

when v = - rr,
the value of
<J>

the

maximum
{

value of

(u, v),

qua function of

u, is

at

= a,

(u, v)

then being
I

- cosh a sin fi
If this is negative,

contour does not


v

- a tanh a 4- (it - ft) cot ft}. = has no real root, and so the (u, fi the equation = meet the line v @ ir or (by symmetry) the line
<I>
-rr)

= ir j3.
Hence provided that the point
(a, /3) lies

in

one of the domains num-

v)= lies as in Fig. 25, bered 1, 2, 3 in Fig. 21 of 861, the contour when a is positive contour the continuous curve indicating the shape of the
<
(it,

Fig. 25.

and the broken curve the shape when a increases is marked by an arrow.

is

negative; the direction in which t

318

THEORY OF BESSEL FUNCTIONS


It follows that the expansion (1) is valid

[CHAP.
lies in

when

(a,

0)

any of the

domains

I, 2, 3.

(or, /8) does not any of these domains. To effect our purpose we have to determine the which passes through the destinations of the branch of the curve 4> (u, v) =

Next, we have to consider the asymptotic expansion when

lie in

origin.
first the case in which a is positive and /3 is acute. The function maxima at v (2n + 1) ir $ and minima at v = (2w + 1) tr + fi, tt) is now each minimum being greater than the preceding; and since (a, positive when v is is greater than positive, it follows that <p (a, v) tt.

Consider
has

<S>

(a, v)

<f>

Hence the curve cannot

cross the line a

a above the point at which

v = 7T. and similarly it cannot cross the line u = a below the point at which v tr. The branch which goes downwards at the origin is therefore confined to the strip a < u < a until it gets below the line v 2Kir + tt #, where

is

the smallest integer for which


1

- a tanh a +

{(2K +
v

1 ) ir

- 0}
,

cot J3

>

0. it

The curve cannot cross the line line u = a and goes off to infinity
Hence,
if

(2/'+l)
acute,

7r+j8,

and so

crosses the

in the direction of the line v

= 2Kir.

is

positive

and

is

we

get

TTJo

7m=0

while, if a is negative

and

is

acute,

we get
1

w. + .
(3)
6

ttJo

- (2m)l

a.
1
'

*m-o

J-*

By combining
If,

these results with those obtained in 8*61,

we

obtain the

asymptotic expansions for the domains 6 a and 7 a.


the branch which goes below the = u o below (a, ir /3) and it does not axis of u ao along the line v = (2L + 1) tt, go to must so it again, cross the w-axis for which integer where L is the smallest

however,

is

obtuse and a

is positive,

at the origin cannot cross the line

- a tanh o and

{(2L
is

+ 1) ir +

j3\

cot

/8

> 0.

Hence,

if

is

positive

obtuse,

we get

IT

Jo

~ m=0
and
/8 is

while, if a is negative

obtuse,

we get

(5)

if

r'--^

S.

'

10*15]

ASSOCIATED FUNCTIONS
these results with those obtained in 8*61,
4, 5,
is

319

By combining

we obtain the

asymptotic expansions for the domains


Since formula (1)
is

66 and

76.

the only one which

of practical importance,

we shall

not give the other expansions in greater detail.

An approximate

formula for

am when

m is large and y is zero, namely


(->mr <*>

-~
(II)

was obtained by Cauchy, Comptet Rendu*, xxxviii.

(1854), p. 1106.

Next consider the

integral

If j

00

grt-tnnht fa

If* g-(-tcosh y +mnh* _ fa


I

The

only difference between this and the previous integral


lies

is

the change in
1, 4,

the sign of cosh 7; and so, when 7 5 in Fig. 21 of 861, we have


(6) K '

in any of the regions

numbered

ir

1 rV-iainht^^l X J*)***' ir mZo 2am+1 Jo

'

where a^'

is

derived from a, by changing the sign of cosh


1
/

7,

so that

_,_

2
(1

l-cosh7'

-cosh 7)*'

,_

+ cosh 7

24 (1

- cosh 7)'

This expansion

fails to

be significant when 7

is

small, just as the previous ex-

pansion (1) failed

when 7 was

nearly equal to
write

iri.

To

deal with this case


v

we

= z{\ e),

r = t sinht,

after the

method of 8*42.
00

It is thus found that

If

e,t-zinht

dt= -l
ttJ
=s

If -00

e" e~ tzt

dt

7rJo

dr

-T-

dr

__Lj~V
oir.'o

6**+*

m =o
<->"

B m (-ez).(~.y*-*dT
i;"-<">

and hence

O)

l fv-,.,b((ft
TT'Q
result equivalent to this

1 37T

1
m=

r
(f'"+t> (m+l)
(if)*

has been given by Airey, Proc. Royal Soc. xciv. A, (1918),

p. 313.

320
10*2.

THEORY OF BESSEL FUNCTIONS


Hardy's generalisations of Airy's
integral.
( 6'3)

[CHAP.

The
If s

integral considered
in the following
<f>,

by Hardy*

by Airy and Stokes manner:


2 3

has been generalised

= sinh

then

= 4s + 2 2sinh3=8s + 6s 2cosh4< = 16s4 +16s + 2 5 1 2 sinh 5<f> = 32s + 40s + 10s,


2 cosh
2<f>
2 3
.

and generally

the cosh or sinh being taken according as n

is

even or odd.

Now
so that

write
r(,

a)- *.,*,(- i,*-in;l-n;-4q/*),


7 8 (,o) = e2
T

+ 2a
2 2 2

T
T

(t,a)

= t* + 3at
4 6 s

= e +4at + 2a T (<,a) = t +5at + 5a


Z 4 (,a)
6

Then the
(1)

following three integrals are generalisations f of Airy's integral

Cin ()

= Jcos Tn (t, a)dt,

(2)

flK (a)

-J" sin T, (*,)<**


J"
exp {-

(3)

Ein (a) =

Tn (t, a)} dt

may be shewn J that the first two integrals are convergent when a is But the third integral positive or negative) if n = 2, 3, 4 (whether real fairly obvious that Ein (a) is indeed and it is complex is when o converges
It
;

an integral function of

a.

an even integer, the three functions are expressible in terms but when n is odd, the first only is so expressible, the function of H. F. Weber. the other two involving

When n

is

of Bessel functions

Before evaluating the integrals,

we observe
is

that integral functions exist


;

which reduce to Cin (a) and Sin (a) when a


Gin (a)

real

for take the


a)} dt.

combination

+ iSin (a) =

[" exp {%Tn (t, Jo

* Quarterly Journal, xli. (1910), pp. 226240. + The sine-integral in the case n=3 was examined by Stokes, Camb. Phil. Trans, pp. 168182. [Math, and Phys. Papers, n. (1883), pp. 332349.]

ix. (1856),

J Hardy,

loe. cit.,p.

228.

10*2, 10-21]

ASSOCIATED FUNCTIONS
when taken round an

321
arc of a circle of

By

Jordan's lemma, the integral,

with centre at the origin (the arc being terminated radius with complex coordinates R, Re*11"), tends to zero as R-~cc
.

by the points

And

therefore
/ooexp(}irt/)

Cin (a) + iSin (a) =

exp [iTn (t,


Jo
f exp {-

a)} dt

= e W
;

Jo

Tn (r, T*)) dr,

where r = te~^ln and the last integral is an integral function of o. The combination Cin (a)-iSin (a) may be treated in a similar manner, and the
result is then evident.

10*21.

The evaluation of Airy-Hardy integrals of even order.

To

evaluate the three integrals Ctn (a), Sin (a),


is positive,
t

suppose temporarily that a

and then,

Ein {a) when n is even, we making the substitution

2a* sinh (ufn)

in the integrals,

we

find that,

by

621

(10),

Gin (a)

+ iSin (a) = n

2a* f

exp (2a*n i cosh u) cosh (u/n) du


1

J o

= iria* wthat
is

****

1/n

(2a* w)>

to say

Cin (a) + iSin (a) If

nsi

{e*>

/w)

J. Vn (2a*) - rW^ JVn <*)}

we equate

real

and imaginary

parts,

we have

^(.)- E

3=S^ 5

jlJ-*(W-)-^(^}.

In a similar manner,

Ein (a) =
so that,
(3)

2et* f

00

exp(-2a*n cosh u) cosh

(u,'n)du,

J o

by

622 (5),
#iw (a) = (2a*/n) # (2a*).

These results have been obtained on the hypothesis that a is positive; and the expressions on the right are the integral functions of a which reduce to Cin (a), Sin (a) and Ein (a) when a is real, whether positive or negative. Hence,

when a is negative the we have


Gin (a) =

equations

(1), (2), (3)

are

still valid,

so that, for example,

2n sin flw/n) | j!o m! T (m + whether a be positive or negative.

- 1/n) " " Jo \ T (m + 1 + 1/)J

'

"

322

THEORY OF BESSEL FUNCTIONS


Hence, replacing a by

[CHAP.
and n
is

/9, we

see that,

when

ft is

positive

even,

then
<*>

Ci <"

" 2n

inU) [J~ " (2


l

^ + '
)

(2/Si0)1 '

(5)

& (-0) = 2);


Ein (-/3) =

J^

/ra)

J-lln (*-) - J-1M (20*)),

(6)

^g^
(9) that,

(/-. (2/3*) + /. (2/?


is

It follows from

431

when n

even, the functions Ci* (a)

and

Sin (a) are annihilated by the operator

and that Ein (a)

is

annihilated by the operator

In the case of the

first

two functions

it is difficult

to obtain this result*

directly from the definitions, because the integrals obtained

by differentiating

twice under the integral sign are not convergent.

10*22.

The evaluation of Airy-Hardy integrals of odd order.

To

evaluate Cin (a)

when n

is

odd,

we suppose temporarily

that o

is

positive, and then, by 6*22 (13),

Cin (a)

= 2a*

00

cos (20*" sinh u) cosh (u/n)

du

Jo
2ftin)

= 2a*cosOWn)
n
That
(1)
is

to say,

CTn(g)

= 2a*o8 (^/n) Jfi/<|(2at, )


n
7ra*

2nsin(^7r/n)

{J_ 1/n

(2a*)-/1/n (2a*)}.
see that,

Using the device explained


<2 >

<-

n& W
(J-'

in 10*21,

we

when

/9 is

positive,

+ *' <**"
by the operator

It follows that the equation 10*21 (4) is true

whether n be even or odd

and, whether n be even or odd, Cin (a) is

annihilated

^+
for all real values of
* It
a.

(-)n 2 n

has been proved by Hardy,

loe. eit., p.

229, with the aid of the theory of " generalised

integrals.

10-22]

ASSOCIATED FTTNCTIOHS
;

323

we

find that,

Next we evaluate Ein (a) when a is positive making the usual substitution, by 1013 (4),
Ein (a)

=
-

exp (- 2a*n sinh


jo
{tan (*7r/n)

u) cosh (u/n)

du

J 1/n (2a*) - B 1/n (2a*)}

Hence the
any value
(3)
is

series

which represents Ein (a) when n

is

odd and a may have

Ein (a)

n cos (i7r/r?)
7T
[

J
g

r (m + 3 _ i/n) r (m + f + 4/f|)
(-)^ttmw

+ sin;(ir/n)
and hence
(4)
it follows

|m =

w!

T(m + 1 - 1/n)

M=0 *! T(m + 1 + l/n)J

(-)m amM

)
'

that

jj^ +

na on

-2
|

Ein () = wofK""3'.
is

Next consider Cin (a) + iSin (a), where a positive. From 10"13 (4) we deduce that
ta'n (a)

temporarily assumed to be

+ % Sin (a) = n
=

2a* f"
.'

exp (2a*" * sinh w) cosh (u/n) du

(tan

Qwln) J 1/n (- 2a*t) - B 1/n (- 2**i)}


J-un ("
2a*t)

.**
, { n sin (7r/w)
l
,

- J-1/n (- 2a*i)}
amn

7ra*<

n+1 U"

n cos (t>>) m=0 T (m


7ra* + n sin ., /x (M* 7_ (-Tr/n)

+ f - */n) r (m + f + /n)
(2o*)

1/n

- -

/1/B (2a*)},

and therefore
(5)

Sin (a) -

7ro* (w+1)

"

<*

mn

wcos(Ww)

r(m + t-i/n)r(TO + f + i/n)

2wcos($7r/w)

whence
(6)

it

follows that,

when # > 0,

Sin (- 0)

^ ^^

r (m + f _ i/n) r (^ + | + */)
W
)}>

W 2 + 2/iCOs(j7r/) ^fl /.x (^/ ( 2# ) " ^/ ( ^ 9

324

THEORY OF BESSEL FUNCTIONS


for all real values of a,

[CHAP.

and hence,
(7)

-1 - n

an

~2
j

Si n (a)

= - not

<

n -3

>

This equation was given by Stokes in the case n =


It should
(8)

3.

be noticed that

Sin (a)

+ (-)*<+ Ei n (a) =

^^
n sin (717/1)
different

{sin

(**/)

+ (- 1 )*<+}

x{/_1/n (2a*) + /1/n (2a*)}

Z^-j-^ {sin (*,r/n) + (where


ft

1 )+>} J

x{/_ 1/n (2*)-J1/n (2/3*)},

= a,

and a and

/9

are real.

The formulae
his

of the preceding three sections are due to Hardy, though

in

methods of obtaining them were the special case n = 3.

and he gave some of them only

10*3.

Cauchy's numbers.

In connexion with a generalisation of Bessel's integral which was defined (see 10*31), it is convenient to investigate a class of functions known as Cauchy's numbers.

by Bourget, and subsequently studied by Giuliani

The

typical

number, N-n,k,m,
t

is

defined by

Cauchy*

as the coefficient of

the term independent of

in the expansion of
/

1\* 1\*/

IV
and

in ascending powers of

t.

It is supposed that n, k,

m are integers of which

the last two are not negative.


It follows from Cauchy's theorem that
1
/'

(0+

1\*

l\ m

(i)

JU*--5RJ
2tt
,

<~'H)
\*\ \e- ni0 Jo
cos (^rriTT

(")
6d6

dt

e~ nie cos* 6 sin -v

+ (-)m eni0

cos*

sinw

6d6

7T

2+*
7T

- n6) cos* 6 sinTO 0d6.


is

Jo

It is evident from the definition that


if it is

N-n^m

zero

if

n+k+m

is

odd or

a negative integer.
473475, 510511; xn.
(1841), pp.

* Comptes Rendu*, xi. (1840), pp. pp.

9293; xm.

(1841),

682687, 850854.

10-3]

ASSOCIATED FUNCTIONS
(1) it is seen that

325

From
(2)

#-*,*,

= (~)m

Fn,k.m

= (-)"-* &n.k,m-

These results, together with recurrence formulae from which successive numbers may be calculated, were given by Bourget*.

The recurrence formulae


(3)
-^-n, k.m
t

are

(4)

+ " -n-i, *-i, m N-a t,m = N-n+i,k,m-i ~ "ni,*,i-i>


^-n+i,
Jfc-i,

to

>

and they are immediate consequences of the identities


t~(t

+ 1/0* (t - l/0m - *- (* + 1/0*" (t - 1/0 + 1-~* (t + 1/0*- (* - l/0~ " m_1 - t-n ~ 1/0* (* - I/O" (* r- (* + 1/0* (* - i/0m - p- (* + 1/0* (* - V0
1

By means

of these formulae any Cauchy's

number

is

ultimately expressible in

terms of numbers of the types

N-n

r
t

t,<n -^ -,o,m-

different class of recurrence formulae, also

due

to Bourget,

owes

its

existence to the equation

It follows that

^.*--2^TTi)/
2w(ro

rn

K)
/
i

dA'-i)
d*l
V

dt

l)J

*/

by a

partial integration.

On
t

performing the differentiation we see that


1 ) iV_, *_2, m+2 ,

(5)

(m +

1)

N-n k,m = nN-n> *_!, m+ - (k -

and similarly
(6)

(k

+ 1) iV_n>i m = nN^ k+h m-i - (m - 1) iV_ *+2 m _2


,
p

Developments due to Chessin, ^nnafa o/ Math.


8

x.

(18956),
k-$,

pp.

12, are

(7)

-^-n,k,

2 Cr m= r=Q
t

2r- n+s -.2r,

m>

(8)

#-,*,m = 2 (-y,Cr .N- + ,-tr.


r=0
(3)

*.

m-,-

These may be deduced by induction from


Another formula due to Chessin
(9) s
is

and

(4).

^-.
This
is

*.

2 m" r=0
(

r
)

Cp - r m Cr
.

where p=\ (k+mn).

proved by selecting the coefficient of n in the product

(t+l/0*x(-l/r Journal de Math. (2) vi. (1861), pp.

3354.

326
10*31.

THEORY OF BESSEL FUNCTIONS


The functions of Bourget and Giuliani.

[CHAP.

The function
<i)

Jn<k (z)

is

defined

by the generalisation
{*

of Bessel's integral

J,.
is

W = ^ P *~
is

))'

P }'* (' - j)} *

where w

an integer, and k

a positive integer.

It follows that
A..

* (*)

f*
I

2,r

ex P {~

* (

nB ~ z sin

0)\

(2

cos

W de

>

and therefore
(2)

/.*(*)

= - r(2cos0)*cos(n0-ssm0)d0.

T Jo

The function /,*(*) has been studied by Bourget, Journal de Math. (2) vi. (1861), pp. 42 55, for the sake of various astronomical applications ; while Giuliani, Oiornate di Mat. xxvi. (1888), pp. 151 171, has constructed a linear differential equation of the fourth

order satisfied by the function.

[Note. An earlier paper by Giuliani, Giornale di Mat. xxv. (1887), pp. 198202, contains properties of another generalisation of Bessel's integral, namely

i'

cos (nd

- z*> sin p 6) dB,

but parts of the analysis in this paper seem to be incorrect.]

If
<8>

we expand the integrand

of (1) in powers of z,

we deduce from
;

10*3 that

/,*(*>

= I
=o

i\U,*, m ^Jf ml

and

it is

evident from (1) that

(4)

Jn
10*3 (2)

(z)
,

= Jn (z).

Again from
(5)
(6)

and

(3) it is evident that

J-nA*) = i-)n- k Jn,k{z\


Jn,k (*)

= /-!, *-i (z) + Jn +i,k-i (Z)

',

and,

if

we take k

1 in this formula,

(7)

Jn ^z) = ^Jn (z).


proving that
2J'n<k (z)

These results were obtained by Bourget; and the reader should have no
difficulty in

(8)

= Jn .hk (z) - Jn+1 ,t(z).


*
{/-!, *

Other recurrence formulae (due to Bourget and Giuliani respectively) are


(9)

-A*+2 (*) =

~ A*+i

(*)

(?)

- /+,,*(*)},

(10)

4/\*_ CO = /,*(*) " *Jn,k-, {Z\

10-31]

ASSOCIATED FUNCTIONS
differential equation is
;

327

The

most simply constructed by the method used

by Giuliani

thus

V n Jn *(*)=ir

f"
.'o

du
C

4 {- (n + z cos 0) sin (nd - z sin


n

0)} (2 cos

0f d$

2k
7T

(n

+ z cos 0) sin (n0 z sin 0) (2 cos 0)* -1 sin 0d0


(2

Jo
cos 0)*"1 sin

= - 2kz J' nk (z) + [\ffi cos (n0 ~ z sin *)}


= - 2kzJ'n k (z) -
'

ed6

cos (n0
./

- * sin 0)

7T

<W

{(2 cos

0?-* sin

0} d0,

and so

V Jn.* (*) = - 2kzJ'n<k (z) - k?Jn<i (z) + 4k (k - 1) Jn<k^ (z).


Operating on this equation by
d* -r9

+
'

and using

(10), it follows that

G& + *) {Vn Jw
.

"*

W + 2kzJ ^ + ^Jn
n k
'

W -*(*-!)
a
}

</.*

CO.

and hence we have


(11)
z*

Giuliani's equation

J\

t (z)

+ (2k + 5) zJ'\ k (z) + {2z* + (k+ 2) - n8 J\ k (z) + (2fc + 5) zJ' n>k (z) + (z* + k + 2 - n2 ) Jnk (z) = 0.
by Giuliani that

It

was

also observed

(12)

*" (2 cos

0)*-

5
n=0

*n J*n,k(z)<sos2n0
ao

+i t
=o
this is verified
right.

M+1

Jm+hk (z)sm(2n + l)0;


on the

by applying Fourier's rule

(cf.

2*2) to the function

A somewhat similar
(1883), col. 1

function J{z\
is

v,

k) has

been studied by Bruns, Atr. Nach. civ.


series

8.

This function

defined

by the

(13)

J(z;

*)-J^ w!r (, + 2 * +m )( v + 2*-2)( + 2A)(+2*+2 + 2)'


of this function is that

The most important property


(i4)

j<
it

,,

q-Jto

..

*+V= {y +_%\*+l + 2y
2* /* + (*)

whence
(16)

follows that

J(z; v,*)=

=t (i/ + 2m-2)(i+2m+2)*

328
10*4.

THEORY OF BESSEL FUNCTIONS


The
that
definition

[CHAP.

of Struve's function H(s).


the functions defined by integrals

Now

we have completely examined


it is

resembling Bessel's integral,

natural to investigate a function defined by

an integral resembling Poisson's integral. This function is called Struve's function, although Struve investigated* only the special functions of this type of orders zero and unity. The properties of the general function have been examined at some length by Siemonf and by J. Walker*.
Struve's function

H(4
-

of order

v, is

defined by the equations

(1
>

<*>

r$m

(1

f.

8in

" *

"
provided that R(v)

r^hfel) .C

sin ( " cos $) sin*

> - |.
we have

By

analysis similar to that of 3*3,

r (v + 1) r (f) ,-o (2m + 1)! Jo


r (-)m zim+1 .ml _(i*) | r(|),-o(2w + l)!r(i/ + m + f)'

so that
(2)

H"^ )= J
is

r(m + f)r(, + m + f)all

The function H (z)

defined by this equation for


It is evident that

values of

v,

whether

R (v) exceeds $
tegral function of
It
is

or not.

H (z) is an

integral function of v
is

and, if the factor (\z) v be suppressed, the resulting expression


z.

also

an

in-

easy to see

[cf.

211

(5),

3121

(1)] that

< 3)

H'<*>=r<#rW)< 1+
ii*# P AIit -i},
{T
!

>'

where

w
and
j

+f

is

the smallest of the numbers

+f

I,

*>
I

+f

|>

^
I

+$

(3)

* Mim. de VAcad. Imp. des Sci. de St P4tersbourg, (7) xxx. (1882), no. 8 ; Ann. der Physik, xvn. (1882), pp. 10081016. See also Lommel, Archiv der Math, und Phys. xxxvi. (1861),

p. 399.

t Programm, Luisenschule, Berlin, 1890.


pp. 340342.]
this section, with the exception of (3), (4), (10)

[Jahrbuch uber die Fortschritte der Math. 1890,

J The Analytical Theory of Light (Cambridge, 1904), pp. 392 395. The results contained in and (11), are there given.

'

'

))

104]

ASSOCIATED FUNCTIONS
obtain recurrence formulae thus

329

We can

H dz ^
and similarly

"

5 W = Jo

(-)TO (2u
2" +2m+1

+ 2to + 1) z2v+im r (to + 1) T (i/ + m + 1)

d
<
t*

" (

* )}

_ S =
,

.o

2" +

- +1 r (m + f

(-)m (2m

+ l)g2w
)

r (

to

+*

=
,r_i

sF*"* 5 r (to +

1)
,'

r ( v + to + 4

=
On
(5)

2*r( + f)r(iT*~

H +l( *
''

)-

comparing these

results,

we

find that

HU
H^

(*)

+ H +1 () =

- H, (5) +
z

Y^\Wl\)

'

(6)

(,)

- H +1 (,) = 2H/ (z) 1

p (7^|yf(J)

(7) (8)

(^+ v )IH v (z) = zIl^


(* - V ) h (*)
(

(z),

* H " +1 ( * } = r( y +t)r(f ) "

In particular we have
(9)

^ {^
(*
2

(,)}

= *H

(z),

{H

(z)}

= | - H, (*).
(z)}

Again, from (7) and

(8),
2

we have
{*!!_,

- v ) H (*) = (* - v)

so that

H (z)

satisfies

the differential equation

(10)

V.H.
Ii
(a)

W - r(

^
> J.

(t)
(z)

The function
to

which bears the same relation to Struve's function as /

bears

Jv {z) has been


(11)

studied (in the case


is

v=0) by* Nicholson, Quarterly Journal,

xlii. (1911),

p. 218.

This function

denned by the equation

L ()-

r(. + | )P(r+(l+t)

= _2(|^__ f^sinh^cos^sin 2 "^^,


the integral formula being valid only

when R

(p)

The reader should have no


function.
* See also Gubler,

difficulty in obtaining the

fundamental properties of this


p. 424.

Zurich Vierteljahrsschrift, xlvii. (1902),

g
330
10*41.
It

"

THEORY OF BESSEL FUNCTIONS


The loop-integral for
-

[CHAP.

H (z).
H (z)
fails

was noticed in

10 4 that the integral definition of

when

R(v)^ \, because
integral.

the integral does not converge at the upper limit.

We

can avoid this disability by considering a loop-integral in place of the definite

Let us take
(V
Jo

iy-i sin

zt

dt,

where the phase of


*

t* 1 vanishes at the point on the right of t 1 at which the contour crosses the real axis, and the contour does not enclose the point

= -l.
If

we suppose
r(i+)
(t
2

that

R (v) > \
.

ment

(0, 1) of the real axis, taken twice, and

we may deform the contour we find that


ri
I

into the seg-

1)"-* sin zt
1

dt

2i cos wr

(1

2 )"-*

sin zt

dt,

Jo

Jo

where the phase of


Hence,
(1)

t* is zero. when R (v) > \, we have = r <* h, ( .)


in
1
(. )

yr
Jo

v - D-* *. ^
. ;

and
all

Both sides of this equation are analytic functions of v for all* values of v so, by the general theory of analytic continuation, equation (1) holds for
v.

values of

From
(2)

this result,

combined with

6*1 (6),

we deduce
(1+)
ei

that

(z)

+ <*) =
lie

^rVi)*^ /

'

lr

dt

and

a> be any acute angle (positive or negative), between \ir + a> and \ir + o>. We then deform the contour into that shewn in Fig. 26, in which the four parallel lines make an angle a> with the imaginary axis. It is evident that, as the lines

To transform
let

this result, let

the phase of z

move off to infinity, the integrals along them tend to The integral along the path which starts from and returns to 1 + oo ie~ iu is equal to ffw ( z ) and on the lines through the origin we write t = iu, so that on them
parallel to the real axis
zero.
;

(P
It follows that

_ i)*-4 = e =F (r- (1 + u*y-l.

Jv (z) + ill. (z) *

(z)

+r (y + 1) r

e (1 +^)"-*^,
|} j o
is

The

isolated values J, f

|,

...

are excepted, because the expression on the right

then an

undetermined form.

10-41]

ASSOCIATED FUNCTIONS
1

331

where the phase of


(3)

2
it

has

its

principal value;
>
)

and hence

< H.()-r. W+r(>+


result,

er^il
r(i) /t

+ u*y-*du.

This

which

is

true for unrestricted values of

z for which

ir

<

arg z

< ir,

will
|

v, and for any value of be applied immediately to obtain the asym-

ptotic expansion of

H (z) when

is large.

Fig. 26.

was obtained by J. Walker*, who assumed that R(z)->0, so that a> might be taken to be zero. In the case = z/ 0, the result had previously been obtained by Rayleighf with the aid of the method of Lipschitz ( 7 "21).
result equivalent to (2)

R(v)> \,

If,

as in 6*12,

we

replace

a>

by args /S,

it is

evident that (3)

may be

written in the form

(4)

H.().y.(, )+

<*)
r(|

j;

e-(i

+7.)

<fe,

where

\tt <

/8

<

\ir

and \ir +

$< arg z < \ir +

(3.

This equation gives a representation of


for unrestricted values of

H (z) when

arg z\<tt. To obtain

a representation valid near the negative half of the real axis,

we

define

H (z)

arg z by the equation

(5)

H (zem' ) = *("+ H (z),


{

and use

(4) with z replaced

by ze*.

* The Analytical Theory of Light (Cambridge, 1904), pp. 394395. t Proc. London Math. Soc. xix. (1889), pp. 504507. [Scientific Papers,

m.

(1902), pp.

4446.]

332
If

THEORY OF BESSEL FUNCTIONS


we write z=ix
in (3),

[CHAP.

where x

is positive,

we

see that,

when R(v)<%>

and, by considering imaginary parts,

we deduce

that

(6)

L (*) = /_ (*) -

r(y

+yr

( |)

siu (** )

^+

tt2 )"" i

<**'

a result given by Nicholson, Quarterly Journal, xlii. (1911),

p. 219, in

the special case in

which i>=0.
10'42.

The asymptotic expansion of

H (z) when

\z\is large.

We
lating

now obtain an asymptotic expansion which may be used for tabuStruve's function when the argument z is large, the order v being fixed.
shall

Since the corresponding asymptotic expansion of

(z)

has been completely


it is sufficient

investigated in Chapter vn, it follows from 10*41 (4) that determine the asymptotic expansion of
/ooexpi/3
/

to

t***\-*

As

in | 7*2,

we have

(p-.l)!*w

V
28.
1

'

We
for

take

so large that

22 (v p ) < 0, and

take $ to be any positive angle

which

|/3|^tt-S,

|args-/3|^7r-o\
which

so that z is confined to the sector of the plane for

7T +

28 ^ arg z ^

ir

We
so that

then have
(

J
1

^sinS,

arg

--

<

tt,

(l
\

^V" "

gftrl/WI Sin ^)2iJW-2p-l (

= Ap>

say,

where

A p is independent of
on integration that

z.

It follows

where

B.

=f

v*
l

;*

e *dtt

(z-w).

10-42, 10-43]

ASSOCIATED FUNCTIONS
that,

333

We
(1)

deduce

when arg z <


j \

tr

and z
|

is large,
j

H (,) - Yv (z) +

^ ^
may
r

^-^

{z

^
may
be

provided that

R(p v +^) > 0;

but, as in 72, this last restriction

removed.
This asymptotic expansion
( 2)

also be written in the form

b.

w - y.

(.)

n + ^S*r +

(z

"^-

It may be proved without difficulty that, if v is real and z is positive, the remainder after p terms in the asymptotic expansion is of the same sign term neglected, provided that as, and numerically less than the first ~& established This be by the method used in 7 32. 0. may R (p + \ v)

p.

The asymptotic expansion* was given by Rayleigh, Proc. London Math. Soc. xix. (1888), 504 in the case v Q, by Struve, Mem. de I' Acad. Imp. des Sci. de St Pe'tersbourg, (7) xxx. (1882), no. 8, p. 101, and Ann. der Phys. und C/iemie, (3) xvn. (1882), p. 1012 in the case v = l; the result for general values of v was given by J. Walker, The Analytical Theory of Light (Cambridge, 1904), pp. 394395.
If v has any of the values \,\,
...,

then

(1

+ u /z ) v ~l
2

is

expressible as a
It follows

terminating series and


that,

(z) is also expressible in

a finite form.
is

an odd positive integer, of elementary functions. In particular


v is half of

when

H (z)

expressible in terms

Hj(*)=(^)*(l-cos*),
(3)

.-(s)
10*43.

K)-V*~0vm, which
represent Struve's function
large.

The asymptotic expansion of Struve s functions of large order.

We
H (z)
As

shall

now
|

obtain asymptotic expansions, of a type similar to the

expansions investigated in Chapter

when
usual,

and z are both


j
|

we

shall write

= z cosh (a + lyS) = z cosh y

and, for simplicity,

we

shall confine the investigation to the special case in

which cosh y is real and positive. The more general case in which cosh y is complex may be investigated by the methods used in 86 and 1015, but it is of no great practical importance and it involves some rather intricate analysis.
*

For an asymptotic expansion

of the associated integral

/;s(-r:r.
see Rayleigh, Phil.

Mag.

(6)

vm.

(1904), pp. 481

487. [Scientific Papers, v. (1912), pp.

206211.]

334

THEOEY OF BESSEL FUNCTIONS


The method of steepest descents has
and
not, as in the previous investigations, to
(3),
v

[CHAP.

to be applied to an integral of Poisson's

type,

one of Bessel's type.

In view of the formula of 10*41


f
/-

we

consider the integral

dw

which we write in the form

where t =

w cosh 7
7
is

log (1

-f

2
).

It is evident that t,

so that, since

qua function of w, has stationary points where w e iy equal either to o or to ifi, two cases have to be considered,
the stationary points
(I) e a ,

which give

rise to

(II)

e*
is less

Accordingly
(II) in
(I)

we

consider separately the cases (I) in which zjv


is

than

1,

and

which z\v

greater than

1.

When 7

is

increases from

to 00
e*

a real positive number a, t is real when w is real, and, as w -2a to e~a cosh a log (I + e t first increases from ),
,
.

then decreases to

cosh

a log (1
.

+ e**) and

finally increases to

00

In order to obtain a contour along which t continually increases, we suppose that w first moves along the real axis from the origin to the point e~, and
angles,

then starts moving along a certain curve, which leaves the real axis at right on which t is positive and increasing.

To

find the ultimate destination of this curve, it is convenient to

make a

change of variables by writing

w sinh f
where
,
i\

it],

e~ a

= sinh f
has for

and are

real.
is real,

The curve in the -plane, on which t


cosh f sin
t\

its

equation

2 cosh a arc tan (tanh f tan


-

17),

and

it

has a double point* at |


write
v>

We now

V) -

2 arc tan (tanh g tan


cos yi

tj)

sm ^
whose corners

and examine the values of F(g, 17) as % 0, A, B, C have complex coordinates


0,

traces out the rectangle

arc sinh
to

1,

arc sinh 1

\tri,

\tri.
,

As tges from
increases from

A, F(^,

i/)is

equal to 2 sinh /cosh 2

and

this steadily

to 1.
* Except

when a=0,

in which case

it

has a

triple point.

10-43]

ASSOCIATED FUNCTIONS
f
is

335

When

on AB, F(g,

rj) is

equal to

V2
and

arc tan

i>

cosec
nj

17,

this steadily increases from 1 to trj^l as

increases from

to

%-rr.

Note.

To

establish this result, write tan

*J2

and observe that


arcten

dt

{r

aroten

=
<v(i+2<*)
t,

li+^

7 ^'

because

-=

-5-

arc tan t, which vanishes with

has the positive derivate

When f is on BC, F(g, ij) is equal to ir sech , and this increases steadily from tt/V2 to 7T as f goes from to C; and finally when is on CO, F(^, rj)
is zero.

Hence the
the rectangle

curve, on which F(^,

rj) is

equal to sech

a,

cannot emerge from

OABG,

except at the double point on the side

OA

and so the

part of the curve inside the rectangle


singular point G.

must pass from

this double point to the

The contours in the w-plane for which a has the values broken and continuous curves respectively.

0,

\ are shewn in

Fig. 27

by

Fig. 27.

Consequently a contour in the w-plane, on which t


the singular point
to
i;

is real,

consists of the

part of the real axis joining the origin to e~* and a curve from this point to and, as

traces out this contour, t increases from

00

It follows that, if the expansion of d/dr in

powers of t

is

l-

Tm

336
then

THEORY OF BESSEL FUNCTIONS

[CHAP.

* m!&
^

2.

1=0 *

and hence, by 10 4
#

(1),

we have

(1)

h.(.)~-<J,(.) +

T7

^^ 2=&.
>

It is easy to prove that

6,-1,
(II)

61

=2coshy,
is

&2

= 6cosh2 y-,

6,

= 20 cosh8 7 - 4 cosh 7
r
is real

When 7
to 00 as

a pure imaginary (=

i0),

from

travels along the real axis from

to 00

and increases steadily and so


;

dw I *~ <* + "^ " j.


Hence, from 1041 (8)
(2)
it follows

~"

{TOT*) 3FJ

dT

that

H,(,)^Fy (,) +
j

r(y + i)r(i)

jS
o

r>

provided that arg z\<\ir. This result can be extended to a somewhat wider domain of values of arg z, after the manner of 8*42.

From the corresponding results in the theory of Bessel functions, it is to be expected that these results are valid for suitable domains of complex values of the arguments.
In particular, we can prove that, in the case of functions of purely imaginary argument,
(3)

1* v
v
I |

(vx)~Iv (vx)

when

is large,

arg v

< \ir, x is fixed, and the error is of the order of magnitude of

i[ -4^w w(i +
i
(

*>}j

times the expression on the right.


a [Note. If in (I) we had taken the contour from w=0 to w=*e~ and thence to w= - i, ufv {). This indicates (z) in place of containing v we should have obtained the formula that we get a case of Stokes' phenomenon as y crosses the line /3=0.]

10*44.

The relation between


the order n
is

H (z)

and

B B (z).
we can deduce from when n

When

a positive integer (or zero),

(z) by a polynomial in z; and n (z) differs from 10-1 (4) that in 1/z. is a negative integer, the two functions differ by a polynomial

10-44, 10*45]
For,

ASSOCIATED FUNCTIONS
is

337

when n

a positive integer or

zero,

we have

p (lz) = v = _r(|m+l)r(m + w + l)'


\mtri
,

n +m

and

BW
and
therefore, since

M=0 r(|w+l)r(n + n+l)' Jn (z) Jn (z), we have


n

^w

gi(ml)tri/l^\nm
'

m=\ " (1 - Jm) T (n + 1 - |/n)


that
is to

say

n \ (i)
In
(2)
like

^= 1 e E-^>
manner, when
/., _ (~) B -(*)

^ ^i

<n "

1>r

,,i

ili<if)r! I _ Hw(2> (z) r(n + i-t)

is
1

a negative integer,
'

n+1

Ic*'""

r (n -

m - j)

(js)-"*-^

r(m + |)

H"w( ^

10*45.

TAe

si^ra

of Struve s function.
interesting result that H(a;)
is

We
positive

shall

now prove the

positive

when

a?

is

and v has any positive value greater than or equal to \. This result, which was pointed out by Struve* in the case v = 1, is derivable from a definite integral (which will be established in 1347) which is of considerable importance in the Theory of Diffraction.

To

obtain the result by an elementary method,

we
J,

integrate 10*4(1) by

parts and then


/
^

we

see that, for values of v exceeding


1

(h)"' "-W- rfr'tW

[*"
.I.

d cos (x cos 6)

..,-,)jgi*~edB

l> tf)r(i) {[eos(xcos9)sin-J


- (2i/ 1)
|

'cos (x cos 0) sin 2'- 2

'*?

cos

0dd

Jo

*Y ^"p
V

1 1

- (2v - 1)
*- 2 -'?

^cos (* cos 0) sin2"- 3 cos 6dd\


II

= ( i*?
^0,

1 (i>

i^iV ./ + )r()

1(

f'sin2

cos

- cos (x cos

-9)1

<W

since the integrand is positive.


*

Mem.

proof given here

de VAcad. Imp. des Sci. de St Ptersbour<j, (7) xxx. (1882), no. is the natural extension of Struve'- proof.

8,

pp. 100101.

The

338

THEORY OF BESSEL FUNCTIONS


v is less

[CHAP.
;

When
have

than

J,

the partial integration cannot be performed

and,

when

i/

= J, we

Hj(*)-(^.) 4 (1 -cob *)>(>,


and the theorem
is

completely established.

A comparison of the asymptotic expansion which was proved in 10*42 with that of Yv (x) given in 7*21 shews that, when x is sufficiently large and positive, H (x) is positive
K (x) is not one-signed when v \ and that asymptotic expansion of H (x) is if v

>

<
.

for the

dominant term of the

(ir)

F -

r(,+|)rd)
according
that Struve's function
is

or

/2\*

U>

sin

^-ii-)
x when
v

asi*>ori<. The theorem

of this section proves the more extended result

positive for all positive values of

>\

and not merely

for sufficiently large values.

The theorem
functions
;

indicates an essential difference between Struve's function

and Bessel

for the asymptotic expansions of

Chapter

vn shew

that, for sufficiently large

values of x,

Jv (x) and Yv (x) are


Theisinger's integral.

not of constant sign.

10*46.
If

we take the equation


*j

{/o()-l(*-"f

j^r'dt

and choose the contour to be the imaginary +ttan |0, we find that

axis,

indented at the origin*, and then write

j {h (x) - Lo (*)} = f
and so
(1)

cos (* cot

<f>)

log tan (i*r+|tf>)

^^
, >

7 (*)-I*(jf)

=^ J

*cos(#tan<}i)logcot(^)
(

a formula given by Theisinger, Monatshefte fiir Math, und Phys. xxiv. (1913),
If

p. 341.

we

replace

x by *' sin 6,

multiply by sin

6,

and

integrate,

we

find,

on changing the

order of the integrations in the absolutely convergent integral on the right,


iW

t
so that
(2)

E, (* tan

<f>)

log cot (<) ^$-r

*'

| f

{/ (x sin 6)

-L
1

(x sin 6)} sin

6d6

/o*'E (^tan0)logcot(^)^ = |. -^1

X
,

on expanding the integrand on the right in powers of x.


Theisinger.
*

This curious result

is also

due to

The presence

of the logarithmic factor ensures the convergence of the integral round the

indentation.


10-46, 10-5]
10'5.

ASSOCIATED FUNCTIONS

339

Whittaker's integral.

The

integral
a*
I

izt

v _i(t)dt,

is a solution of Bessel's equation only when 2v been studied by Whittaker*.

which

is

an odd integer, has

It follows

from
*

61 7

that, for all values of v,

(1)

V, L*

<$*

P_x

(t) dt\

=-

lira

0*e* (1

- P) P;_j(0]

=
If

cos
7T

* vir .z*e

-&
.

we expand the integrand (multiplied by


it is

e") in ascending

powers of s and

integrate term-by-term +

found that
.

Ji
(2)

2*1

, e^P^CO^-a^e-" S fn rr (r/ _l * ^ * v + f + y) i-i M = r(m+|-p)r(w


,
v ,

.,_

(2r)m .ro!

The formula

of 3'32 suggests that

we

write

and then
the series

it is

easy to verify the following recurrence formulae, either by using

(2), or

by using recurrence formulae

for

Legendre functions
2ii-"z^e- iz

(3)

W_
v

(z)

+ -

W ^{z)
v

2v
z

W
,

'

(Z)

j(2Tr)r(-v)T(%+v)\'
2ii- v z~le- iz
'

(4)

W_

(z)

r+1 (z)

= 2W

(z)

^2ir)r^-v)r^ + v)
2i*~ y z*e~ iz
),

5)

p + ,) w, (,) = ,w... (.) +


()w.w

r(t . t)r(i + ,)
2i*~ y zi e~ u
.

(6)

iw

H ,(.) + V(a|r) r(t _ >)r(t+y)


_

An

asymptotic expansion of

W (z)

for large values of z


|

may be

obtained by

deforming the path of integration after the manner of Lipschitz


* Proc.

( 7 21).

London Math.

Soe. xxxv. (1903), pp.

198206.

t By a use of Legendre's equation the recurrence formula

may
*"

be

verified;
1

and we
4

can prove

that

P^ (t)dt= p
1,

by expanding

(4

- *. 4 + "

- 4')

in ascending powers of 1

and integrating term-by-term.

340

THEOBY OF BESSEL FUNCTIONS


The
function
is

[CHAP.

thus seen to be equal to

-355-/,
Now
it is

e"

M p-<-'>*-^-/
-

M *-.>*

known

that*, near

<=
i;

1,

iV^(*)-t*i(*
i

-** +
8

_ fcos*7r\ P ^<-*> ^-^"j


*\

i-10. r(m-y + )r(m+i/ + )


.;

/l

- \m

.*

(ml?

VY)
>

x log
|

\T~) ~ 2^ (m + 2 ) + ^ (w ~ v + ) + ^ ( + * + $)}

and since

we

obtain the asymptotic expansion

(7)

W,(#)~ *#(*)
g-i <*+**+*)

COS V7T T

(if,

m)

+ ^(w + ^ + y) ^r(m+
Some
noticed

l)~log2^-^7rt'}

functions which satisfy equations of the same general type as (1) have been by Nagaoka, Journal of the CoU. o/Sci. Imp. Univ. Japan, iv. (1891), p. 310.

10*6.

The functions composing


will

Tn (z).
kind, of

The reader
integral order,

remember that the Bessel function of the second


written in the form ( 3*52)
1

may be

%
+

(n-m- 1)1
"*
j

WM M
,

=0

?.

^(^)r

<

10

<*'> ~ * <ro +

>

- * < + m

The

series

on the right may be expressed as the sum of four functions, each of


fairly

which has
(1)

simple recurrence properties, thus


{log

nrYn (z) = 2

<!)-* (1)} Jn (z) -

n (z)

+ Tn (z) -2Un (z),

* Cf. Barnes, Quarterly Journal, xxxix. (1908), p. 111.

10*6]

ASSOCIATED FUNCTIONS

341

where
(2)

Tn (z) = -

"2

^-'--f^1 (^)-

n+2wl

w =o ra!(n + ro)!
and
(cf.

3*582)

<3>

g'<')-.!, ( ;i(,7m)

<< + ' +
l

>-< 1 "-

The
pp. 142

functions 7^(2) and


147,

Un (z) have been studied by Schlafli, Math. Ann. m. (1871), though he used the slightly different notation indicated by the equations
Sn (z)=-2Gn (z),

Tn (z) = 2fn (z), Un (z)= -En (z);

more recent

investigations are

due to Otti* and to Graf and Gublert.

The
(4)

function

Tn (z)

is

most simply represented by the


f'(J 7T

definite integral

Tn (z) = -

0) sin (z sin

nd) dd.

To

establish this result, observe that

(-ra*r*~ 1 r ( ,\ - [l I jfnW ~La m >riw _ i r(m+i+6)r(n+m+i-6)J e o _j_ra | (-)w (|^)w+wt /+> (i + ty***
-&r* [&>-*-*
i_
(n

i
J,.o
,

2w)!
/

J_,
(o+)

m+1+e
*

(-)w (jg)w+2OT

(i

w+2w log t tj)

e | __2 f' ~iriJom>-in-i

nie

(- 1> sin
(n
.

fl)

w+2w

(0

- tt)

+ 2m)\

where

has been replaced by e *""r,i


(

It follows that

Now
3
_ ttt . t m>in~i

ta"

sin 0)n +*l

(n

+ 2m)\

=
and
so

( 1> sin 0) (- iz sin 0) sinfl - + (_)n eiz sin |{ e


(cosh

(n even) (n odd)
}
(

(sinh

-L P(0 _ 1^) yn (s) = 7rt


Jo

-8in#

+ em(e-r)+izsin} <&
4269. Lommel'a

* Bern Mittheilungen, 1898, pp. 156. t Einleitung in die Theori-e der Bessel'tchen Funktionen, iv (Bern,
treatise, pp.

1900), pp.

77

87, should also be consulted.

342
If
is

THEORY OF BESSEL FUNCTIONS


replaced by
ir

[CHAP.

in the integral obtained


(4),

second of the two exponentials, the formula


obtained at once.

which

by considering only the is due to Schlafli, is

The corresponding

integral for

Un (z)
!

is

obtained by observing that


i

un ^ W = -fi
-and
so,

(-rq*)w+2w r (i + 6) Ide Zo m T (n + + 1 + e) J'.?


{<**)-

[I

r (i +

Jn+t (*)} j^,

from

6*2 (4),

we deduce that

(5)

Un (z) = {\og$z)-,fr(l)}Jn (z)


+ - r$ sin (n0 - z sin
7T

0)

d$

+ (-) n

" e-nc-*nhf

<ft.

Jo

Jo

10*61.

Recurrence formulae for


10*6 (4) (z)

Tn (z)

and

Un (z).

From

we

see that

Tn_

+ Tn+1 (z)-(2n/z)Tn (z)

=-

7T J o

I "(h

~ 0) sin (z sin - n0)

{2 cos

- 2n/z} d0

=-
8

irz J o

P(W -0)4b {cos (z sin a"


4
z

- n0)} d0

4 =cos hmr 2
z

Jn (z), '
v

on integrating by parts and using Bessel's integral.

Thus
(1)

Tn.

(z)

+ Tn+1 (z) = (2n/z) Tn (z) + 4 {cos \mr - Jn (z)}/z.


2

Again
and so
(2)

Tn

'

(z)

=7T

!"($ ir-0) sin

cos (s sin

(9

- n0) d0,

Jo

r-iW-ZV+ito-MV^).
these formulae
it

From
(3)
(4)

follows that
2

C* (*
(cf.

+ n) Tn (z) = zT^ (z) - 2 cos

\nir
2

+ 2/ (s),
- 2Jn (z),

- n) T (z) = -z Tn+1 (z) + 2 cos


find that

\nir

and hence
(5)

1012) we

2 2 ^n Tn (z) = 2{z sin \mr + n cos \mr) - 4??

Jn (z).

10*61, 10-62]

ASSOCIATED FUNCTIONS

343

for

Jn (z),

With the aid of these formulae combined with the corresponding formulae Y(s) and Sn (z), we deduce from 10*6 (1) that

(6) (7)

(8)
(9)

(10)

Un_, (z) + Un+l {z) - (2ft/*) Un (z) - (2/z) Jn (z), Un _, (z) - Un+l (z) = 2 Un (z) - {21z) Jn (z), (Sr 4 n) Un (z) = zUn ., (z) + 2Jn (z), (*-n)Un (z) = -zUn+1 (z), [cf. 358(1), V n Un (z) = -2zJn+l {z).
'

358(2)]

The reader may


It
is

verify these direotly

from the

definition, 10*6 (3).

convenient to define the function

T_ n (z),
ir

of negative order, by the

equivalent of

106 (4).
(z)

If

we

replace

by

in the integral

we

find that

T_ n

=vr

f "(Itt

7r

6) sin (z sin

nO) dd

Jo

=
and so
(11)

2 [*
I

it

Jo

(1

- 0) sin

(z sin

n0 + mr) d0,
(z).

T_ n (z) =

(-)"+*

Tn

We
it is

now

define

U-H (z) by
U-n (z)

supposing 106 (I) to hold

for all values of

then found that

(12)

(-)
{

Un (z) - Tn (z) + Sn (z)}.

10-62.

Series for

Tn (z)

and

Un {z).
- Jn _ im (z)}

We
(1)

shall

now shew how

to derive the expansion

Tn (z) = I I
10*6 (4).

{Jn+2M

{z)

from

The method which we


,

shall use is to substitute


2,

2 7r

= Z

sin

2m0

in the integral for

Tn (z), and then integrate term-by-term. This procedure needs justification, since the Fourier series does not converge uniformly near
=
To

and
0.

= ir,

and, in fact, the equation just quoted

is

untrue for these two

values of

justify the process*, let 8

and

be arbitrarily small positive numbers.


find

Since the

series converges uniformly

when
I

8
..

% $ ^7r-o\ we can
...

an integer m^ such that

M
2
ro=l

(*tr-d)I

sin2m0| |<c,

The analysis immediately

following

is

due to D. Jackson, Palermo Rendiconti, xxxn. (1911),


is 1-8519...

pp.

257262. The value

of the constant

344

THEORY OF BESSEL FUNCTIONS


all

[CHAP.

of 8 between

throughout the range 8 ^ 8 ^ v 8, for and n, we have

values of

M exceeding m^.

Again, for all values

Ljt-02 2 m =i

=
i

ye

{l+2cos2< + 2cos4* + ... + 2cos2Jft}(fc

_
.

i1* sin(2i/"-H)f J
9
t
'

_J_
sin
.

= Anfor some value of between 8 and a monotonic (increasing) function.

(2JT+ l)t f i* sin 1

a=

/"(if+J)T sin
/

Jjt

eta:

^r,

by the second mean-value theorem, since

/sin

is

By drawing the graph of .r -1 sin# f n sin x exceed \n dx in absolute value * x / o


I

it is

easy to see that the last expression cannot

if this

be called \ rrA, we have

\Tn {z)
i

| *>i

siu

(z

sin 8

n8) dd
I

",=iyo

21
,

["{{ln _ 6 )- 2 -*^lsin(* S intf-tf)^|

<

-{/:

/r*aiK--i
|

!!

^i-'---'-

where

5 is the upper

bound of

sin (z sin 8

n8)

Since

2(AB + e) Bis
that*

arbitrarily small, it follows from the definition of

an

infinite series

2 mn (z) = 2 T

(* sin
J

2m6

, sin (z sin n
.
.

- w0) dd

= 2
m=l
''*

{t/n + fc)

/n-2m (^)}

and the result


It will

is

established.

be remembered that

Un (z) has already been

defined

3*581) as a

series of Bessel coefficients

by the equation

and
as a

that, in

3582, this definition was identified with the definition of

Un (z)

power

series given in

106 (3).

10*63.

Graf's expansion of

Tn (z + 1)

as a series of Bessel

coefficients.

It is easy to obtain the expansion


(1)

Tn (z + t)= 2 T^^Jmiz),
Wl= on
* This expansion

was discovered by

Schlafli,

Math. Ann.

m.

(1871), p. 146.

10*63, 10'7]
for,

ASSOCIATED FUNCTIONS
it is

345

from 10-6 (4),

evident that

Tn (z + t)=- P\\tt - 6) sin(* sin 6 -n0 + z sin 6) dO


TT J

J_
771

\
'

(\ir

- 6) Jm (z) {e'(t8in-n9+me) _ e -*(^in9-n+)J


i=-oo
/,

ffl

=
by using

? ["(!.
;

_ 0) I

(0) sin

{*

sin

- (n - w) 0} d0,

under the integral sign is uniformly convergent, the order of summation and integration may be changed, and the

21

since the series

result is evident.

The
plicated

pi-oof of
;

it

the formula given by Graf, Math. Ann. xliii. (1893), p. 141, depends on the use of the series of 10*62 combined with 2*4.
for

is

more com-

There seems to be no equally simple expression

Un (z +

t).

10'7.

The genesis of Lommel's functions

$M> (z) and

sM) (2).

which includes as special cases the polynomials z0 n (z) and Sn (z) of Neumann and Schlafli, was derived by Lomniel, Math. Ann. IX. (1876), pp. 425 444, as a particular integral of the equation
function,

(1)

V v y = kz+\
fi

where k and

are constants.

It is easy to

this equation, proceeding in ascending


,

shew that a particular integral of powers of z beginning with z** is


1
,

z*

+i

z*+*

(2)

[> + If - S
,

{(/T+

If.

1/

2
}

[i>

+ 3)* - v*\

]
'

air

=0 (i> -$V + m=

(-)m

(^r
(\p

|),+i

+ P + |)m+i

wt

r^-^+m+Dr^+^+w+i)
ks^ v {z).

For brevity the expressions on the right are written in the form

The
ft

function sMt 0)
is

is

is evidently undefined when either of the numbers an odd negative integer*. Apart from this restriction the general

solution of (1)
(3)
In
like

evidently

y -#,(*)
manner
the general solution of

+ ***(*)

is

(5)

y = *-*<-!>
*
.

{<$ v

()

+ *V (2)}.
and
v is discussed in 10*71.

The

solution of the equation for such values of n

F.

12

346

THEORY OF BESSEL FUNCTIONS


Next
let

[CHAP.

parameters."

us consider the solution of (1) by the method of "variation of We assume as a solution*

y=A(z)J
where

(z)

+ B(z)J_ v {z),
v

A (z) and B (z)

are functions of z determined by the equations

= 0, J\ (z) A'(z) + J'_ v (z) B' (z) = kzr-\

Jv (z) A'(z) + J.
we
see that
v

(z)

B'

(z)

On

using 3*12
v

(2),

A (z) =
'

2 sin

^ V Jw
f J
\_

(z) dz,

B(z)

= - ^~i^2 sin

vir J

VJ

(z) dz.

Hence a

solution f of (1)

is

it

sin vir

where the lower

limits of the integrals are arbitrary.


is

Similarly a solution of (1) which


integers or not,
(7)
is

valid for all values of

i>,

whether

= \1ctt

lY, (z)

[V J

(z)

dz

- Jv (z)

\* z*

Y v (z) dz\

numbers fi v + 1 have positive real and (7) may be taken to be zero. If we expand the integrands in ascending powers of z, we see that the expression on the right in (6) is expressible as a power series containing no powers of z other than ^ +1 z +3 z***, .... Hence, from (3), it follows that, since neither of the numbers /jlv is an odd negative integer, we must have
It is easy to see that, if both of the
parts, the lower limits in (6)
, ,

(8)

,^ r

\*z*J_ (,)*_[,/,(,) l sin vir


|_

{z)dz-J_ v (z) FzJ,(z)dz\


jo
v is not

j o

In obtaining this result

it

was supposed that

an integer

but

if

we

introduce functions of the second kind,


2

we

find that

(9)

8^A*) =
this formula

\v\yAz)\ zJAz)dz-Jv {z)^ z*Yv (z)dz\,


we may proceed
to the limit in

and in

making

an

integer.

It should be observed that, in

Pochhammer's notation
l)

( 4*4),

(10)

^ %(,)

tl)(^> +

x/,(i; i^-ii' + |,i/* +


* Cf. Forsyth, Treatise on Differential Equations (1914), 66;
v is
it is

|i'

+ f; -i*

2 )-

supposed temporarily that

not an integer.

The

of

z,

generalisation of this result, obtained by replacing zn+i in (1) by an arbitrary function was given by Cbessin, Comptes Rendus, cxxxv. (1902), pp 678679 and it was applied by
;

him, Comptes Rendus, cxxxvi. (1903), pp. 1124


Bessel's equation.

1126, to solve a sequence of equations resembling

10-71]

ASSOCIATED FUNCTIONS
associated function

347

The

Mi (2)

is

derived from a consideration of a solution

of (1) in the form of a descending series. solution and investigate its properties.

We now

proceed to construct this

10*71.

The construction of the function S^tV (z).


of the equation 10'7(1), proceeding in descending
,

A particular integral
(1)

powers of z, beginning with z*~ l y

is
>

= kz^

- 1)* - v (fl + P

j (/

- 1)* _

,/>}

|(

- 3)* - V>
;

a-.].
if it

it is

This series, however, does not converge unless it terminates but a solution of 10*7 (1), and it will be called kS^iz).

terminates,

The

series terminates if
in

ft

v
//,

is

an odd positive integer, or


case.
1,

if /*

v is an

odd positive integer, and

no other

In the former case we write q


(r \

= v + 2p +

and then we have

K
'

(-rr(ij*-iF+|)r(jj + ii' + l)

m=o

r(i

l)r(i;

+ m+l)

When

fi

v=2p +

l,

the function

vanishes,

and

so,

when

fi

v is

an odd positive integer, we have

(2)

s> ,(*)^,(*) + x ' f. \ /


'

2'~ir(i

''-*"

+ * )r(i '' +
Sln V1T
.

i ''

+ l)
V) IT
.

[COS %(fJL

v) IT

J_ (z)

COS |(/Ll +

Jv (z)].

Since both sides of this equation are even functions of v, the equation is true also when /x + v is an odd positive integer, so that it holds in all cases in

which
of

Sllv (z) has, as yet, been defined. We adopt it as the general definition $M( (z), except that, when v is an integer, we have to use the equivalent form
S^(z) = *,(*) +
shewn
2"" 1

(3)

r (/* - I v

4-

\) r(|A*

+ * +

|)
v) ir
.

x [sin \{yk v)ir.Jv (z)


It will be
is

cos % (fi
;

(*)].

an odd two functions


latter.

in 1073 that S^iz) has a limit negative integer, i.e. when sMi (z) is undefined
sM)V (a)

when fi+
and
so,

v ov

fi-v

of Lommel's

and S^

v (z), it is

frequently more convenient to use the

348

THEORY OF BESSEL FUNCTIONS


It will appear in 10*75 that the series (i),

[CHAP.

defined

when

either of the

of significance
yields, in fact,

by means of which S^iz) is numbers ft v is an odd positive integer, is still when the numbers fi v are not odd positive integers. It an asymptotic expansion of S^ v (z) valid for large values of

the variable
10*72.

z.

Recurrence formulae

satisfied

by Lommel's functions.

It is evident

from

107 (2)

that

**'Wthat
is

fr

+ iy-*
8

to say

(1)

W.., (z) = **
it is

+1

- {i> + 1) -

:*}

M (4

Again,

easy to verify that


j

TZ^ S
so that

*' "

^=

(/*

+ V ~ X ) ^ #-l. r-1 (*).


1) V-1..-1
(*)>

(2)

8\ w (Z) + (v/z) SM> (z)=(fl + Vs' ilLiV

and similarly
(3)

(z)-(v/z)s^ v (z)

= (fi-v-l)s
we

ll

_ltV+1

(z).

On

subtracting and adding these results


{2v/z) SrtV (z)
,

obtain the formulae

(4)
( 5)

= (/M + v-l) s^ _j (z) - (fi - v - 1 ) $_i. m-i (*)> 2s'M (0) = (/i + v - 1 ) v-j, .-i (*) + (/* - v - 1 ) Sp_ ,+1 (z).
lt

The reader

will find it easy to

the type SntV (z) the type #Mi (z)


(6) (7)

deduce from 10*71 (2) that the functions of may be replaced throughout these formulae by functions of
;

so that

(8)
(9)

(10)

= **+1 - {(fi + 1) - #M( , (*), S^ v (z) + (*/,) SM, <) - 0 + * - 1) flMlPU4 (#), 6\, (#) - (v/z) ,(*) - 0 - v - l)8^ v+l (*), (2*/*)^ (*) = (,* + /- 1)^^_ ,,_ (^)-(m-i'- l)iSU.r*(*). 2S'^(z) = (M + v -l)S^. (z) + <ji-v-l)S^+1 (z).
(*)
2
1/

These formulae may be transformed in various ways by using (1) and (6). They are due to Lommel, Math. Ann. ix. (1876), pp. 429432, but his methods of proving them were
not in
all

cases completely satisfactory.

Lommel's functions Sfl>v (z) when fiv is an odd negative integer. The formula 1071(2) assumes an undetermined form when fi v or fi + v
10'73.
is

an odd negative integer*. We can easily define Sv-i,(z) by a repeated use of 10*72 (6) which gives

Sv-vp-\ >v (z)


(-)*&.,,,(*)

in terms of

n\ (1)
*

*-i
<?

J.
o
(z) is

>*+*.,(*)SMl
integer.

2^(- J
of

(-fi^
,)m+1 ( v

-p)tn+1

2^! (1-,)/
which
ix

Since

an even function

v, it is

sufficient to consider the case in

-v

is

an

odd negative

10*72, 10*73]

ASSOCIATED FUNCTIONS

349

We

next define

Sv-h ,(z) by

the limiting form of

1072 (6), namely

(2 )

^w^,[

x>

.-Cifey+ i)]fi

The numerator (which is an analytic function of when /* = i/ 1, and so, by L' Hospital's theorem*

near

/*

= v 1)

vanishes

Now
L

it is

easy to verify that


1

pW
3/*

_U,-i

-i-iw n I U + 1,

.> +

(-)"(^)"
(p

l)ir( + m + 2)

x {2log z + yfr(l) + yfr


Also

+ 1) - ^ (m + 2) - yfr(v + m + 2)}.

[^
and

{2m+i

r (i/* - ** + f) r (^ + 1 * + f) cos j i> + v) }]^

r_ x

= 2' r (v + 1) sin i/7T {log 2 + |^ (1) + |^r (v + 1) + tt cot vv},

[^{2^r(^-^+f)r(^ + ^+f)cosH/*-^)^]^_ = 2^7rr( v +i),


and hence
(3)
it

follows that

s^y-i'-rwsJrlfSli)
x
{2 log

\z - ty (p + m +

1)

yjr

(m +
for

1)}

- 2"v

ir

V (v) Y (*),
a negative
(where n
is

and

this formula,
is,

which appears to be nugatory whenever v

is

integer,

in effect, nugatory only

when

when

=n

a positive integer) we define the function by the formula

^n-i,n \8) = >__! (#),


in which the function on the right
is

defined by equation 10-73(1).

To discuss

the case in which v=0,

we take the formula


(/t+1)2

-VWwhichgives
'

Sr

Since

-'.o(*)=^[p{^ +1 -^ +S.o()}]

Mss

_i

()-* {r (J, + f

)}Jo

^t^
.

+2* + i {r (*/*+$)}* (cos &**


it follows,

(z)

- sin &**

()},

on reduction, that
<8L,,o(*)=5 *
* Cf.

(4)

^&#^[{log(^)-f (m+l)}2-4^'(i + l)+i^].


Bromwich, Theory of Infinite
Series, 152.

350
10*74.

THEORY OF BESSEL FUNCTIONS


Functions expressible in terms of Lommel's functions.

[CHAP.

From the descending series given in 1071 (1) it is evident that Neumann's polynomial On (z) is expressible in terms of Lommel's functions by the equations
(1)
0.im (z)

= (1/*) S^ n i, Sn (z)
is

0^ 0) = {(2m +

1)/*}

S0tWl+1 (z),

and

Schlafli's

polynomial
flU (*)

similarly expressible

by the equations
(*).

(2)

= 4 S-i. (*)

&t*n (*) = 2S0>2TO+1

It is also possible to express the important integrals

#J
;

(z) dz,

z"

(z)

dz

in terms of

Lommel's functions
.

thus we have

dz
d_

{z

Jv (z)

zl ~>

5M_
*-*

]>

_ 1 (*)}

= zJv^ (z) /SM _ ,_ (g)+(jt-v- 1) zJv (z) SM_2,, (*)>


1 1

l {Z ~" /"_! (Z)

dz

SM> (*)} = - Zjv (Z) S^iz) + (fl+V-l) zj^ (Z)

S^^

(z).

On

eliminating
find

8^^-^z)

from the right of these equations, and using

1072 (6), we
(3)

by integrating that
(ii,

P ** Jv (z) dz =

+ v-l) zJv (z) 8^,h _, (z) - zJv-x {z) Sp {z\


t

and proofs of the same nature shew that


(4)
[* z

Yv (z) dz = (p + v-l)zYv (z) /S^^-i (z) - zY^ (z) ,(*),

and, more generally,


(5)

j* z

(z)dz

= ( i + v-l) z<@v (z) 8^ _, (*) - *#_, (*) 6V, (z).


f

Special cases of these formulae are obtained by choosing

p and

i>

so that

the functions on the right reduce to

Neumann's or
l

Schlafli's polynomials,

thus

(6)

f z9g^
f Sf

(*)

dz = Z*

L* j K,n (Z) Otm(*)

(*) -

#_, (*) Om
(*)J

(7)

9Bl+ x (*)

dz = * {#+,
and
(6) are

&m (*) - #m (*) 8am+l (*)}.

(1876), pp.

contained in Lommel's paper, Math. Ann. ix. were given by Nielsen, Handbuch der Theorie der Cylinderfunktionen (Leipzig, 1904), p. 100, but his formulae contain some misprints.

Of these

results, (1), (3), (4)


(6)

425444;

and

(7)

It should

be noticed that Lommel's function, in those cases when

it is

10*74, 10*75]

ASSOCIATED FUNCTIONS
is

351

expressible in finite terms,

The

equivalent to Gegenbauer's polynomial of 92. formulae connecting the functions are*

/n
(8)
.

2r

r(y + m) v+ 2m Q

..

2 v+1

T(v+m + l) v+2m+l
III
.

It follows that the most general case in which the integral (5) is expressible in terms of elementary functions and cylinder functions is given by the formula

(0)

/'

^r-wm

(*)

dz ~

ml z*- v tvT (y + m) I

>M-2m \ z ) -2mi,v\ 2; )

+ 2m-l
(

_
The function denned by the
series

&,+tm-l (z)
v

A^

{z)

+ 2m

r(v+2m+i)

r(v-i)
;

,+l,lw

has been studied in great detail by W. H. Young t this Function possesses many properties analogous to those of Bessel functions, but the increase of simplicity over Lommel's more
general function seems insufficient to justify an account of

them
is

here.

The

integral v

f* I Jo

J
,

(t)
'.

t(z t)

dt has been studied (when v

an integer) by H. A. Webb,

Messenger, xxxin. (1904), p. 58; and he stated that, when /=, its value is On (z). This is incorrect (as was pointed out by Kapteyn) ; and the value for general values of v is J

{S1 ,u(-z)-pS ,y(-z)}/z,


()

when

It (v)

>

and arg ( - z)
|

<

ir.

10*75.

The asymptotic eocpansion of S^iV (z).

We
a)
when
|

shall

now shew by Barnes' method

that,

when

ft

y are not odd

positive integers, then

fll<

y(z) admits of the asymptotic expansion


i >'--'M0'-) ,

^. w ~^-[i-^-y-" + tz
j

--, i-...],

is

large

and arg z < w.


|
j

Let us take the integral

27rtJ_ooi- i> +i

r(|-^/A + |i/) rd-i/A-iv)


drawn by taking p
left

'

sins7r

The contour
poles of the
*

is

to be

to

be an integer so large that the

only poles of the integrand on the

of the contour are poles of cosec sir, the

Gamma

functions being on the right of the contour.

Gegenbauer, Wiener Sitzungsberichte, lxxiv. (2), (1877), p. 126. t Quarterly Journal, xian. (1911), pp. 161177. % Of. (rubier, Zurich Vierteljahrsschrift, xlvii. (1902), pp. 422428.
Proc.

London Math.

Soc. (2) v. (1907), pp.

59118;

cf.

6-5, 75, 7 '51.

352

THEORY OF BESSEL FUNCTIONS


The
integral
is

[CHAP.

convergent when arg z


j

<

ir,

and

it

may be

seen without

difficulty that it is

(a*-2?).

It

may be shewn from

the asymptotic expansion of the

Gamma

function

that the same integrand,


centre at

when

integrated round a semicircle, of radius

R with

p \, on

the right of the contour, tends to zero as

R -*- oo

provided

that

tends to infinity in such a manner that the semicircle never passes

through any of the poles of the integrand.


It follows that the expression given above is equal to the

sum

of the

residues of
2

T(j-1ui +lv-s)r(l-l H,-$v-s) nrQir sin^TT r(i-i * + jior<j-*A*-*i')


' i

at the points

0,-1,-2,. ..,-<_p-l),
1,2,3,...,

When we

calculate these residues

we

find that

r(|-|^i + |i')sinj'7r w = oW T(l -v + m)


,
1 -

+
so that

_-i (-yr(i-^-i-iir+m)r(i-^-|y + w)

sini>7r

x [cos i 0* -

v)

/_ (*)

- cos (/* + v) it Jv (z)] =


.

(^~2p),

and

so,

by

1071

(2),

we have the formula

and

this is equivalent to the asymptotic expansion stated in (1).

10*8]
10*8.

ASSOCIATED FUNCTIONS
Hemi-cylindrical functions.

353

Functions
(1)

Sn (z) which

satisfy the single recurrence formula

S-* (*) - Sn+1 (z) = 2S' (z)

combined with
(2)

B.W

'(*)
will

have been studied in great detail by Sonine*. They


cylindrical functions.
It is evident that

be called hemi-

S (z)

is

expressible in the form


(*),
;

Sn (s)=/(D).S
where
fn (?)
satisfies

D = d/dz and / (D) is a polynomial in D of degree n


the recurrence formula

and the polynomial

combined with
It follows

by induction

(cf.

91 4) that
tt

n a a /(l)=H(-f + V(l +i)} + {-f-V(l +i)}

],

and therefore
(3)

Bm {M)-ll[-D + J(I* + l)}+[-D-</(l> + l)}*].B,(s).


it is

If
(4)

supposed that (1) holds for negative values of

w, it is

easy to see that

S_ (*)=(-)"Sn
tt

(*).

To obtain an

alternative expression to

(3),

put = sinh

t,

and thenf
(n even)
(re

t Mi)-

jcoshntf

{-sinhn*

odd)

1+ 2! ? + ~4i
]

*
fc '-.
..

(even)
odd)
,,

re

,.

re (re

-jjf
Hence
2
I

l2)

^-31

(re

So <*)

S" (#) |j
*

(5)

Sn (^)=

+ \, n(n2 _ 12)
ft

re

Cre

22 )
'

S (,)+...

(re

even)

It is to be noticed that

but

8n (z), Un (z)
It should be

n (z) Tn (z) and E (*) are hemi-cylindrical and H(s>are not hemi-cylindrical functions.
y

functions,

remarked that the single recurrence formula


Z

gives rise to functions of no greater intrinsic interest than Lommel's polynomials.


* Math. Ann. xvi. (1880), pp. 19, 7180. t See e.g. Hobson, Plane Trigonometry (1918), 264.

354
10*81.

THEORY OF BESSEL FUNCTIONS


The addition theorem for hemi-cylindrical functions.

[CHAP.

We
(!)

shall

now

establish Sonine's important expansion*

Sm (z + t) = 2 J n (t)Sm _n (z);
n=
is

the expansion
at the point
z,

valid when z + 1 lies inside the largest circle, whose centre is which does not contain any singularity of the hemi-cylindrical

function under consideration.

inside or

Take as contour a circle G with centre z such that S on the circle. Then

(f) has

no singularity

,, ( , +0

ZlTl] c

ifiM %z
t

=^fc smxo[ ^nO


n

a-z)jn (t)^dc

The

series converges uniformly

on the contour, and so we have

Sw (s + t)=

i
cn

Jn (I)

B* (0 On (t- s)

= $

Jn (t)fn

(-^8m (z)dz

-.!,^>/-(-a)/-(5) (')*
But
so that
it is

easy to verify that


2/n (- *)/

(0 -/_ () + (-)"/-+ (IX


.(

(* + *)-./.(*)*(*) + 2 ^(0{S*-n(^) + (-rS M+n


n=l

2)},

whence Sonine's formula

is

obvious.

S (z) denotes a function of a more general type than a hemi-cylindrical function, namely one which merely satisfies the equation Sn, 1 (z)-8n+l (z)^2Snf (z),
without satisfying the equation

It should be noticed that, if

(z)

=S

'

(z),

we

still

have

f*

V di) Sm W = Sm_n
(1) is
still valid.

( z)

+ ( ~ )n &m+n ( ^'

and so the formula


the formulae of
* Math. Ann.

We

thus have an alternative proof of

53, 91,

934 and 1063.


v. (1872), pp.

xvh. (1880),

xn. (1880), pp. 48. See also Konig, Math. Ann. pp. 8586.

310 340;

ibid.

10-81, 10-82]
10*82.

ASSOCIATED FUNCTIONS

355

Nielsen s functional equations.

The

pair of simultaneous equations

1(2)

*U (*) + F

v+1 (*)

- (2v/z) Fv (z) =

%gv

(z)/z,

where fv {z) and g (z) are given arbitrary functions of the variables v and z, form an obvious generalisation of the pair of functional equations whereby cylinder functions are defined. It has been shewn by Nielsen* that the
functions / (z)

and g v (z) must

satisfy the relation

and

it

has been proved f that,

if this relation is satisfied,

the system can be

reduced to a pair of soluble difference equations of the first order.

For brevity write


f (*)

+ 9 (*) = - (*).
is

f (z )

~ 9* (*) - & (*)

and the given system of equations


J
(3)

equivalent to

(^

1(4)
It is

+ v) Fv (z) = zFp -x (z) - a, (z), (* - v) Fv (z) = - zFv+l (z) - ft, (z).


that
x

now evident
<*>*

- tf) Fv {z) = (*-v) [zFv .

(z)

a v (z)]

so that

= - z*F, (z) - z/3^ (z) -&-p) av (z), V V F (z) = - z$ v _ (z) ~(**-v) (z).
x

or,

Again
<*

*2 )

v (jr)

- <* + *) [- zFv+l (z) - fi m (*)] = - z*Fv (z) + za v+1 (z) -(* + p) &, (z).
V,F,(z)

We
(5)

are thus led to the equation

= zv,(s),
-(*-p) (z),
- (* + we
v) /9 (z).

where
|(6)
1(7)

zm v (z) = - *_,
z*r y (z)

(z)

= + zap+1 (z)

On
(8) It

comparing these values of

vr v (z),

are at once led to Nielsen's

condition

/_, (*) +/ +1

(*)

- (2-/*)/r (*) = 9.-i (*) ~ 9*+i (*) " 2S^' (*)


is sufficient for

now has

to be

shewn that Nielsen's condition

the exist(8) to

ence of a solution of the given system.


*

To prove

this,

we assume

be

Ann. di Mat.

(3) vi. (1901), pp.

t Watson, Messenger, xlviii. (1919), pp.

5159. 4953.

356

THEORY OP BESSEL FUNCTIONS


by
(6)

[CHAP.

given, and, after denning ta v (z)

and
is

(7),

we

solve (5)

by the method

of variation of parameters.
(9)

The

solution
1-k

(m)

= / (*)

L - jy

v (t) r (t)

dtl

+ Yv (z) \dy + |ttJ V (t) vt v (t) dtl

where a and b are arbitrary constants; and c and dy may be taken to be independent of z, though they will, in general, depend on v.
remains to be shewn that c and dv can be chosen so that the value of given by (9) satisfies (1) and (2), or (what comes to the same thing) that it satisfies (3) and (4). If (3) is satisfied, then
It

{z)

*/,_, (z)

|c - tt
J

Yv (t) vf (t) dtl - \ttz


e

(z)

F (z) r (z)

+ zYv ^ (z) \d + %tt


= zJr-, (z) c^ - \tt
-j

J (t) (t) dtl + faz Y (z) J (z) TB V (z)

Y^ (t) _, (t) dtl


J

+ zY _
that
is to say,

(z)
I

<*_,

+ |tt

l"

J_, (t) w^, (0 dtl (z),

*/"_! (z)

- c w -WJ"{ Y* (t) v v (t) - Y^ (t) <*_, {t)\ dtl


dv - d_, +
\tt
J

+ zY^
But
Tz
since (6)

(z)

[Jv (t)

r (*)

- J^ (t) r^ (t)}

dt\

+ a (z) = 0.

it is

easy to verify that


( jr)

{^U (z) &_,


and

- V. (z) a (jr)} ;

r (z)

<& (s)

- r r _, (z) V^ (z),

(7) are satisfied

and so (3)

is satisfied if *

zJ^ (z) jc - c,_y - \tt


+ zY,_
and
1

F^ (z) _, (z) - Y (z) (z)

(z)

|^-d-i + iw r^by
3'63-(12),

(^^

(^)-^(^)^(^)l*|

+ a,(z) = 0,

this condition,

reduces to

zj_, (z)

- c_, + \tt [ Fr _, (a) &_, (a) - F, (a) a (a)]} + * F_, (*) {d - d _i - \ir \JV_ (b) &_, (b) - J (b) a, (6)]} = 0.
[c
X

Consequently, so far as (3) is concerned, to satisfy the difference equations


f (10)
{

it is sufficient

to choose c

and d

\(11)

c - c_, - - tt F,_ (a) &_, (a) - F, (a) a, (a)}, d, - <*_, _ | {J^ (6) _, (6) - J, (b) a (6)}
x

10*82]

ASSOCIATED FUNCTIONS
will

357

and the reader


the value of

have no

difficulty in verifying that, if these

same two
then

difference equations (with v replaced by v

throughout) are
(4).

satisfied,

v {z)

given by

(9) is

a solution of

These difference equations are of a type whose solutions may be regarded

known*; and so the condition (8) is a sufficient, as well as a necessary, condition for the existence of a solution of the given pair of functional equaas
tions (1)
If,

and

(2).

as z -* *>

/(#)-0
where B >
0,

(**-),
,

5r,(*)

=
,

(**-'),

then we

may make a -* oo
may be
(v)

b -* x>

and we have

so that the general solution

written

(12)

F. (z)

= J9 {z)

|ir,

+ tt

(J

Yv (t) w (t) dt}

+
where ir^v) and
7r 2 (i/)

Y, (z)

(v)

- |tt

rJ

v (t)

v. (0^4

are arbitrary periodic functions of v with period unity.

Note. Some interesting properties of functions which $ati*fy equation (2) only are to be found in Nielsen's earlier paper, Ann. di Mat. (3) v. (1901), pp. 1731. Thus, from a set of formulae of the type

F_
v

(*)

(z)

- (2v/) *T () = 2g y ()/,

it is

easy to deduce that

(13)

Fv+n (z)=F

{z)Rn

(z)-Fv _ l (z)Rn . l

v +i(.z)

n-l

+ (2/)
the
first

2 SWw()-ffn-m-i, + +i(); m=0

two terms on the

right are the

complementary function of the difference equation,

and the

series is the particular integral.


is

* An account of various memoirs dealing with such equations Math. Soc. (2) n. (1904), pp. 438469.

given by Barnes, Proc. London

CHAPTER XI
ADDITION THEOREMS
11*1.
It has

The general nature of addition theorems.

been proved

( 4*73)

that Bessel functions are not algebraic functions,

and

obvious from the asymptotic expansions obtained in Chapter vn that they are not simply periodic functions, and, a fortiori, that they are not
it is fairly

doubly periodic functions.


to Weierstrass*,
it is
v

Consequently, in accordance with a theorem due

not possible to express

JV (Z + z) as an algebraic function
do not possess

of

,/

(Z) and

(z).

That

is

to say, that Bessel functions

addition theorems in the strict sense of the term.

There
coincide

are,

however, two classes of formulae which are commonly described

as addition theorems.
;

and the formula

In the case of functions of order zero the two classes for functions of the first kind is

J y(Z* + i - 2Zz cos


which has already been

<f>)}

= 5 em Jm (Z) Jm (z) cos m<f>,


ro-0

indicated in 4'82.
is

The simplest

rigorous proof of this formula, which


;

due

to

Neumann f,

depends on a transformation of Parseval's integral another proof is due to Heine J, who obtained the formula as a confluent form of the addition theorem
for

Legendre functions.

11*2.

Neumann's addition theorem^.

We
(1)

shall

now

establish the result

J
we

(*0

= 2 m Jm (Z) Jm (z) cos m<f>,


wi-0

where, for brevity,

write
zr

= V( + * - 2* cos
2

0),

and
*

all

the variables are supposed to have general complex values.


stated in 1

The theorem was


;

of

1893)

see

Phragmen, Acta Math. vn.

(1885), pp.

Schwarz' edition of Weierstrass' lectures (Berlin, 3342, and Forsyth, Theory of Functions (1918),

Ch.

xiii for proofs of the theorem. t Theorie der BesseVechen Fuvctionen (Leipzig, 1867), pp. 59 70. XHandbuch der Kugclfunctionen, i. (Berlin, 1878), pp. 340343;

cf.

5*71

and Modern

Analysis, 15*7.

xvi.

In addition to Neumann's treatise cited in (18801881), pp. 201202.

11 -1, see Beltrami, Atti della

R. Accad. di Torino,

11-1-1 1-3]

ADDITION THEOREMS

359

We

take the formula (Parseval's integral)

which

is

valid for all (complex) values of vr


2tt.

and

a,

the integrand being a periodic


is

analytic function of 6 with period

We

next suppose that a


a

defined

by

the equations
ts

sin

ol

Zz cos

<f>,

ta cos

= z sin

<f>,

and

it is

then apparent that

()

= jj-

*
f

exp

{i

- z cos

0) sin

+ iz sm

<

cos 6)

dd

* [* j 27T J _ (

m=-x

JM (Z)e w l
)

efe " in( *"' )rftf

Jm (Z)\
Jm (Z)
"

mie+izain

<*

6)

dd

= J_

%
X

mi <+> - 8in<>

m=

Jm {Z).Jm {z)e*,
<*>

the interchange of the order of summation and integration following from the uniformity of convergence of the series, and the next step following from
the periodicity of the integrand.

we group the terms for which the immediately obtain Neumann's formula.
If

values of

differ

only in sign,

we

The corresponding formulae


Journal fiir
formula; see 11*4.

for Beasel functions of order | were obtained by Clebsch, Math. lxi. (1863), pp. 224227, four years before the publication of Neumann's

11

3.

Grafs gen eralisation of Ne um arm's formula.

Neumann's addition theorem has been extended to functions of arbitrary The extension which seems to be of more immediate importance in physical applications is due to Graf*, whose
order v in two different ways.

formula
(1)

is

*<)

{^-^Y - J_J'+<
pp. 142

Z)Jm

{Z)

*"*'
ze f*

and

this formula is valid provided that

both of the numbers

are less

than \Z\.
*

Math. Ann.

xliii. (1893),

144 and Verhandlungen der Schweiz. Naturf.

Get. 1896,

pp. 59 61. p. 241.

special case of the result has also

been obtained by Nielsen, Math. Ann. ui. (1899),

360
Graf's proof
is

THEORY OF BESSEL FUNCTIONS

[CHAP. XI

published, an independent proof was given


;

based on the theory of contour integration, but, two years after it was by G. T. Walker, Messenger, xxv. (1896), pp. 76

80 this proof is applicable to functions of integral order only, and it may be obtained from Graf's proof by replacing the contour integrals by definite integ