LEGENDRE TRANSFORMS

By C. J. TRANTER (Shrivenham)
[Received 20 October 1948; in revised form 28 July 1949]

Summary
INTEGRAL transforms containing Legendre polynomials as kernels are
defined and inversion theorems set up. Two such transforms are applied,
in the manner now established for the solution of boundary-value
problems by integral transforms, to two problems with 'mixed' boundary
conditions. One of these problems was proposed by J. W. Nicholson
who failed to solve it by orthodox Bessel-Fourier analysis and the
second is a generalization of Weber's and Gallop's classical problems on
the electrified disk.

1. Introduction
Laplace, Fourier, Hankel, and Mellin transforms have recently
been applied to the solution of boundary-value problems of r ^.the-
matical physics. The application of such transforms often reduces
a partial differential equation in n independent variables to one in
n— 1 variables and it is often possible, by successive operations of this
type, to reduce the problem to the solution of an ordinary differential
equation.
Most of the work done has involved problems in which the range of
a variable has been (0,oo) but Doetsch* has extended the theory to
finite ranges in the case of Fourier transforms and, more recently,
Sneddonf has discussed finite Hankel transforms. Sneddon has also
suggested extensions to transforms with other kernels but he gives no
examples.
The object of the present note is to define transforms in which the
kernel is a Legendre polynomial, to set up appropriate inversion
theorems, and to apply the theory to two problems involving 'mixed'
boundary conditions. It seems appropriate that the transforms defined
below should be termed Legendre transforms and this terminology is
adopted throughout.
* G. Doetsch, Math. Annalen, 112 (1935), 52-68.
t I. N. Sneddon, Phil. Mag. (7) 37 (1946), 17-25.
Quart. J. Math. Oxford (2), 1 (1950), 1-8
3695.2.1 -„
2 C. J. TRANTER
2. Legendre transforms and their inversion theorems
If Pn(/i) is the Legendre polynomial of degree n (n& positive integer),
Whittaker and Watson* quote that

(1)
(m = n) }'
I define the odd Legendre transform by

(2)
0
V being supposed a function of ft and V2n+1 a function of n. Application
of (1) immediately gives as the inversion theorem for this transform
V = | (4n+3)V2n+1PZn+1(jt). (3)
Similarly the even Legendre transform is denned by

V2n = jVPMdn, (4)

0
and its inversion theorem is
PM. (5)
Other Legendre transforms could be denned. Thus we could take

Vn = jvPnMdr, (6)
—i
and, from the resultf

the appropriate inversion theorem would be

f F » . (8)
n-0
However, in the present work, I shall use only the odd and even
transforms denned respectively by (2) and (4).
It is evident that no result could be derived by using Legendre
transforms as defined above which could not have been obtained by a
direct use of expansions in Legendre polynomials. The advantage of
the use of these transforms is that they reduce the analysis involved
• E. T. Whittaker and G. N. Watson, Modem Analysis (Cambridge, 1927), 306.
t Whittaker and Watson, loc. cit. 305.
 LEGENDRE TRANSFORMS 3
in the solution of boundary-value problems almost to a 'drill' and
bring the work into line with recently published work of this type
involving Laplace, Fourier, and other integral transforms.

3. The transforms of 8/a/u{(l—/*2)aF/3/x}

In the applications given later, we shall require the odd and even
transforms of d/d(i{(l — /I 2 )8F/3/A}. These transforms are easily obtained
as follows. Using Legendre's equation in the form*

| M ) , (9)

and integrating by parts we find

/ L.- / ^'-
^ 0
1

f^] + Pm(O)[V]ll=o-m(m+1) J V
o
H +mP m _ 1 (0)[F] /I=o -m(m+l)F m , (10)
°MJo
where Fm = J VPJp) dp, (11)

and we have used the relation

Pm(0) = mP^fO). (12)

4. An application of the odd Legendre transform

Nicholsonf has proposed the problem of finding a function F which
satisfies Laplace's equation in the semi-infinite solid z > 0 such that,
on the surface z = 0, dV/dz = /(/>) inside the circle p = a, and V = F(p)
outside this circle. F and / are given functions of p, it being assumed
that they are such that the various integrals which arise in the solution
exist.J Nicholson has failed to find a solution by the usual Bessel-
Fourier analysis, and the problem is here solved by transforming from
cylindrical to oblate spheroidal coordinates and using the odd Legendre
transform.
* I am indebted to a referee for improvements in the analysis of this section
and for other helpful criticisms.
t J. W. Nicholson, Phil. Trans. Royal Soc. (A) 224 (1924), 303-69.
| See also § 7.
4 C. J. TJIANTER
We write z = a^, p = a(l- M 2 )i(l+£ 2 )*, (13)
and it is easy to show that the surfaces fi == constant, £ = constant are
respectively hyperboloids of revolution and oblate spheroids. The
spheroid £ = 0 is the circular disk p < a, z = 0, while fx = 0 is the
remainder of the plane z = 0. In the coordinates (/*,£)> Laplace's
equation becomes

and the boundary condition on z = 0 is

O* = 0). (16)
Multiphcation of (14) by Pm(/i), integration with respect to ft between
0 and 1, and use of (10), (16) gives

We do not know the value of dV/dfj. when p. = 0, but this term can be
eliminated by taking m = 2 » + l (n an integer) since -P2n+i(0) = °-
Thus we have for solution the ordinary differential equation
), (17)

where 2?2n+1(£) = -(2n+l)P2n(0)F{aJ(l+Z% (18)

After multiplication by P2n+i(fi) and integration over (0,1) the boundary
condition (15) becomes
i

(C = 0). (19)

The solution of (17) can be effected by the method of variation of

parameters and the use of the oblate spheroidal harmonics introduced
by Nicholson.* These harmonics, denoted by pjx), qn(x), are solutions
of the differential equation
^} o, (20)
• J. W. Nicholson, Phil. Trans. Royal Soc, (A) 224 (1924), 49-93. This paper
will be referred to subsequently by [N] followed by a page number.
 LEGENDRE TRANSFORMS 5
and their properties are summarized by Nicholson on pp. 52, 53 of the
paper referred to.
Thus we write, as the solution of (17),

so that % t l = A2n+ia)p'Zn+1(O+B2n+1tt)q'2n+1U), (22)

provided that
j W M , + i ( 0 + 8 W i « ; ) # i n + i t f ) = 0, (23)
dashes denoting differentiation with respect to £. Multiplication of
(22) by (l + £2), differentiation with respect to £, substitution in (17)
and use of (20) with (2n+l) for n, gives
(24)
Solving (23), (24) for -4'2n+1(0, #2n+i(0 a*"1 u se of the relation [N 78],
(25)
leads to ^ ^ = R a ) q { 0 , (26 )

- (27)
Since F is to vanish when £ ->oo, V2n+1 and therefore -42n+i(0 must
also vanish under this condition. Hence, from (26),
00

^2n + l(0 = - / £2n+l(*)<72n+l(*) **• (28)

From (22) and (19),
A2n+1(0)p2n+1(0)+B2n+1(0)q'2n+1(0) = aj
0
(29)
and, since [N 53]
'(0) g t g (i)»(2+i)P(0) (30)
we have

0 (31)
Finally, equation (27) gives
/ ) dx. (32)
6 C. J. TRANTER
The complete solution to the problem, by using (21) and the inversion
theorem (3), is therefore given by

V = | (4n+ZlA^+1(i)p9n+1(i)+-Ban+1(i}qtn+ia)]P^+1W, (33)
n-0
i(0 being given by (28), B2n+1(0) by (31) and (28) with £ = 0, and
S2n+1(0 by (32).
5. An application of the even transform
The similar problem of finding a function V satisfying Laplace's
equation in z > 0, such that on z = 0, V = F(p) inside the circle p = a
and dV/dz = /(/>) outside this circle, can be similarly solved. In terms
of oblate spheroidal coordinates we now have to solve (14) with
« = 0), (34)

^ O* = 0). (35)
at, o/jt,
We proceed as in § 4, but, since F is not now known on (j. = 0, we
take m = 2n and, without going into details of the analysis, the result is

», (36)
where A2rJ&), B2n(t) are given by
00

A2ntt) = - j Rin(x)q2n{x) dx, (37)

(38)
o
l
brB2n(0) = -A2n(O) + {-p}£- f (39)
0
and B2n(x) = P2n(0)aa;/{aV(l+a;2)}. (40)
6. Special cases
The result just given can be illustrated by considering the special case
of the steady temperature in a semi-infinite solid, the surface of which
is kept at unit temperature inside the circle p = a, the rest of the surface
being impervious to heat. This solution also gives the potential in z > 0
due to an electrified disk of radius a whose surfaces are at unit potential,
a problem considered by Weber.*
* H. Weber, Jour.f. Math. 75 (1873) 75-105.
 LEGENDRE TRANSFORMS 7
Here F(P) = l,f{P) = 0: equation (40) gives J?2n(z) = 0and (37), (38),
(39) lead to
A2n(0 = 0,
I

B2n(Z) = B2n(0) = t £ £ ?

\2/w (n = 0).
Hence (36) gives, since P0(/x) = 1,
V = (2/77)00(0, (41)
and using the results*
00

PniH-knd) = ( W J e-*°J0(\P)Jn+i(\a)\-i dX, (42)

o
4
and Jj(Aa) = (2/7rAa) sinAa, we have, in cylindrical coordinates,
CO

V = (2/TT) J e- As J 0 (A /3 )sinAaA- 1 dX, (43)

o
which is Weber's result.
By using a particular case (£ = 0) of the expansion of

in terms of Legendre polynomials given by Nicholson [N 56], the results

(36)-(40) are easily applied to the case of the electrified disk with
surface potential J0(cp) (c constant), a problem solved by Gallopf and
of which Weber's problem is a particular case.

7. Verification of the solution

The analysis given is strictly formal, and it should be verified that
the solutions do in fact satisfy the differential equation and boundary
conditions. If F = 0 in the problem of § 4, this is easy, but verification
in the general case seems a formidable problem. The problem of
assigning to F and / conditions sufficient to justify the solution given
is an interesting one; that the solution does hold at any rate in certain
cases when F does not vanish can be seen from the following example.
Suppose we take
F S ii n1
/ 2 1U (° < P < ^ (P) = - (
(P > )
7r(az—p*)* TX p
* J. W. Nicholson, Phil. Trans. Royal Soc. (A) 224 (1924), 320.
t E. G. Gallop, Quart. J. of Math. 21 (1886), 229-56.
 8 LEGENDRE TRANSFORMS
in the problem of § 4. Then, omitting details of the algebra, wefindthat

giving
F = -cot- 1
n-0

This is the correct solution. The example is, of course, the field due to
an electrified disk starting from the charge distribution on the disk and
the potential of the external field in the plane of the disk. Similar
remarks apply to the solution given in § 5.