1933.] APPLICATIONS OF MELLIN'S TRANSFORMATION.

495

APPLICATIONS OF MELLIN'S TRANSFORMATION TO

INTEGRAL EQUATIONS

By C. Fox.

[Received 18 July, 1933.—Eead 1G November, 1933.—Revised 3 March, 1934.]

1. Introduction.
1.1. The object of this paper is to solve certain types of Fredholms*
equation by means of Mellin's transformation. The methods are similar
to those used by Hardy and Titchmarsh f in their paper on Fourier kernels
and also to applications of Laplace's transformation made by Doetsch %
and Fock§.
The equation which we shall discuss is

(1.1.1) F{x) = G{x)+[ H(ux)F(u)du,

Jo
where G(x) and H(x) are known functions and F(x) is to be determined.
The method also applies to simultaneous equations of the type

(1.1.2) Fr(x) = Gr(x)+ \°Hr(ux) Fr+I(u)du,

Jo
where r = I, 2, ..., n; Fn+1(x) = F^x); the functions Gr{x) and Hr{x) are
known and Fr(x) are to be determined. The case n — 2 will be briefly
considered at the end of the paper.

1.2. Let s = o-\-it, where a and t are real. Then, as an illustration of

the method, multiply equation (1.1.1) throughout by a^~x and integrate

* Solutions of Fredholm's equation and references to the literature of the subject are
given by Whittaker and Watson, 1, Ch. 11, and Bocher, 2.
f Hardy and Titchmarsh, 3.
% Doetsch, 4 and 5.
§ Fock, 6.
496 C. Fox [Nov. 16,

with respect to x. Adopting the notation of Hardy and Titchmarsh. we

write
f{s)= f F(x)xs~1dx)
Jo
and similarly for other letters. We then have

dx [H{ux)F{u)xs-xdu.
o Jo
If we assume that we may change the order of integration, this becomes
f(s) = g(s)+h(s)f(i-s).
Changing s to 1—5 we also have

The elimination of/(I—s) leads to

-s)] = g(s)+h(s)g(l-s),
and so. when h(s) h(l—s) =£ 1, we may expect from Mellin's formula that
a solution of (1.1.1) is given by

(1.2.1)

Hardy and Titchmarsh have shown that, when h(s)h(l—s) = 1, H(x) is a

Fourier kernel and it is evident that in this case there can be no solution
unless
g(8)+h{8)g{l-8) = O,
a result which, as we shall show later, is equivalent to

(1.2.2) G{x)+\ H(ux)G{u)du = 0,

Jo
that is to say, if H(x) is a Fourier kernel, there is no solution of (1.1.1)
unless G(x) is a self-reciprocal function of the kernel H(x) with
characteristic number — 1. This can also be proved very easily as
follows:
F(x) = G{x)+\ H{ux)F{u)du
Jo

H{ux)G{u)du+\ \ H{ux)H{vu)F{v)dudv,
o Jo Jo
which, if H{x) is a Fourier kernel, reduces to ( 1 . 2 . 2 ) .
1933.] APPLICATIONS OF MELLESTS TRANSFORMATION. 497

In §2 conditions for Mellin's formula are stated, and in §§3 and 4 we

consider equations ( 1 . 1 . 1 ) and ( 1 . 1 . 2 ) respectively. I n § 4 . 2 we
illustrate the methods by solving the equation

F (x) = e~x+A e~ux F (u) du.

w

2. Mellin's formula.
The well-known formula of Mellin, which is a variant of Fourier's
transformation, is as follows:

(2.1) f{8)=[>F(x)afi-1dx,
Jo
1 rc+i*
(2.2) \
F{x)= f(s)x-°ds,
A™ Jc-ix
and corresponding to Parseval's* theorem we have

(2.3) ±- \f(s)g(l-s)x-°ds= CF(ux)G(u)du.

^ l Jc-ix, Jo
Conditionsforthe validity of these formulae are given by Bochnerf and
Pollard | . Those given by Bochner are
B. / /

(i) F(x)xs~1dx is absolutely convergent in the strip

Jo
a < R(s) = a < p.
(ii) F(x) is of bounded variation^ at the 'point x,
then ( 2 . 1 ) implies (2.2).

//, in addition,

(iii) G(x)xs~1dx converges absolutely in the same strip a<o-<j8.

Jo
then (2.3) is,also true.

* I am indebted to the Referee for pointing out that (2.3) is Parseval's formula and
also for the references to Doetsch.
t Bochner, 7, 35, p. 148 (10); 36, p. 151, (5).
t Pollard, 8, Theorem A., 462.
§ If F(x) in (2.2) is not continuous it must be replaced by %[F(x-{-0)+F(x — 0)].
SER. 2. VOL. 38. NO. 2020. 2K
498 C. Fox [Nov. 16,

The conditions given by Pollard are

P. / /
(i) for a particular value of c and all positive values of x, \f(s)\ is
bounded,
(ii) for some positive a,

exist,
(iii) F(x) has a finite integral in every finite interval consisting of
positive values of x,

then, for almost all values of s whose real part is c, (2.2) implies (2.1).

3. Theorem 1.
3.1. Suppose thath(s)h(l—s)^=l and let
i./M-. g(s)+g(i-s)h(s)

Then a solution of equation (1.1.1) is given by (1.2.1)*/ there is a common

strip a<cr= R(s) </? for which the functions g(s), h(s), k(s), G(x), H(x),
K{x) all satisfy conditions B and P above.
Ifh(s)h(l—s) = I, then there can be a solution of (1.1.1) only if G(x) is
s self-reciprocal function of the kernel H{x) with characteristic number —1.
Proof. The conditions enable us to apply Mellin's formula to ( 1 . 2 . 1 )
Avhen we have
g(s)+g(l-s)h(s)

Writing 1—5 for 5, we have

g(l-s)+g(s)h(l-s)

On multiplying this by h(s) and subtracting from (3 .1.1), we get

f(s) = g(s)+h(s)f(l-s).
Now multiply this equation by x~s and integrate along the line a = c,
where a < c < /?. Then, on using (2.3), we finally obtain

F(x) = G{x)+\ H{ux)F{u)du.

Jo
1933.] APPLICATIONS OF MELLIN'S TRANSFORMATION. 499

If h(s)h(l—s) = 1 there can evidently be no solution of (1.1.1) unless

g(s)+g(l-s)h(s) = O.
On multiplying this equation by x~s, integrating along the line a = c, and
using (2.3) again, we find that this condition becomes equivalent to

Jo

a result already proved in §1.2 by a different method.

3.2. As an illustration consider the equation

eruxF(u)du.
o
AVe have g{s) = T{s), h{s) = XT{s).
Hence, subject to the conditions of § 2 being satisfied, we have
i r r(«)+Ar(s)r(is)
v
,,
(T\ — I ' ' \-L i 1 o'-s•/Jo
{ ) x
~2m)c_iv i-\*r{S)r(i-8)
2 ^

The integrand has poles at sins7r = A2TT. If 0 ^ A 2 ? r ^ l and m is any

integer, then there are two poles on the real axis lying between 2m and
2m+2, at s = 2m+^> and s = 2m-\-l—p say, where 0 <p < l—p < 1.
All the poles are then given by
s = 2m-\-p-\-2n, s = 2m-\-l—p-\-2n,
"where n has any integral value, positive, negative, or zero. For
•0 < x < 1, F(x) can be evaluated by considering the residues of f(s) on the
left of the line o = c. For x>l,
1 [c+i=° r (s) sin 577- a ,
_ . I —: r» x too
2TT* J c - i x Sin«77 —A 2 7T

•can be evaluated as before by considering the residues on the left of tbe line
a = c and
1 fc+ic° ATT
2
x~s ds
-A 7T

by considering the residues on the right of the line a = c. The result is the
same whether 0 < r c < l o r a ; > l J but depends upon the value of c. There
are two cases according as
2m-{-p < c < 2ra+l— p or 2w+l— p <c
2K 2
500 C. Fox [Nov. If5,

Dealing with the second case first, Ave have

(3 .2.2) F{x) = ^ ~ - x-*»^YiI^-2

In order to determine the interval in which c lies we observe that, in virtue-

of the conditions of § 2. the integral

(3.2.3)
JO

must converge absolutely. For convergence at the loAver limit Ave must-
have m ^ 0. For convergence at the upper limit Ave can easily find the
asymptotic expansion of the series on the right-hand side of (3.2.2) by a
method such as Barnes's* method and it is then found that, for sufficiently
large values of x.

where .4 is a constant independent of x. Hence m > — 1 -j--J (1 — p) and so,

m = 0. The solution then becomes

(3.2.4) F{x)--

v <
1
cos pit „= 0 IF(1 — ])-\-2n) r(»-r'2n)) '
If we deal with the case when 2m-\-p <c<2m-\-l —p by the same methods,
we find that it is impossible to make (3.2.3) converge absolutely for any
value of m. Hence (3 . 2 .4) is the only solution of (3 . 2 .1).
We must still deal with, the case A2 77 ^ 1. When A2 77 = 1 the equation
sin srr = 1 has double zeros at s = 2n-\-h (n any positive or negative integer
or zero). The corresponding residues are

where AH and Bn are constants. These constants are not difficult to

evaluate, but it is laborious to do so and consequently Ave shall not
calculate them.
When A277 > 1. the equation sin577 = A277 has only complex roots. In
the interval 0 < a < 1 it has two roots, fx and v say, both of Avhich lie on the
line o=l. The general roots are then 2n-\-fi, 2n-\-v (n an3r positive or
negative integer or zero) and the solution of (3 . 2.1) is substantially the
same as that giA'en by (3.2.4).

* Barnes, 9.
1933.] APPLICATIONS OF ME LUX'S TRANSFORMATION.

It is worth noting that the solution of the equation

(3.2.5) F(x) = x\ e~ucF{u)du

Jo
also involves the consideration of the roots of the equation sinsTr = A-TT.
When A 2 77^:l. a solution of (3.2.5) is

(3.2.0 ^"-F|£:-.
where (i)in the case A2TT < 1, p is the smallest positive root of sin 77 = SX2TT
and (ii) in the case A2TT > 1, y) is a root on the line o = \. The expression
(3.2.6) with A — sec pv forms part of the right-hand side of (3 . 2 . 4-.)*.

4. Simultaneous equations.
The simultaneous equations

I Fx(x) = Gx(x)+ f'Hx(ux)F,(u)dit:

I Jo
F2(x) = 6?2(a) -j- I # o(ux) Fx(u) die.
Jo
•can be dealt with in the same Avay as (1.1.1). I state the results without
proof.
If hx(s)h2(l—s) ^--1. then the solutions of equations (4 .1) are

<4.2)

1-^(5)7^(1-5) '
The conditions for the validity of these solutions are easily determined with
the help of the statements of §2.
If hx(s) li2(l—5) == 1. then there can be solutions of (4 .1) only if

Gx(x)+ f *Hx{ux) G*(u)du = 0,

Jo
G2(x)+[°H2(ux) Gx{u)du = 0,

that is to say, if Gx(x) and — G2(x) are transforms of each other for the pair
of Fourier kernels Hx{x), H2(x).

* See Hardy and Titchmarsh, 10, where equation (3.2.5) is discussed and references
to other literature on the subject arc given.
502 APPLICATIONS OF MELLIN'S TRANSFORMATION. [NOV. 16.

References.
1. E. T. AYhittaker and G. N. Watson, Modern analysis (Camb. Univ. Press).
2. M. Bocher, An introduction to the study of integral equations (Camb. Math. Tracts.
Xo. 10, 1909).
3. G. H. Hardy and E. C. Titchmarsh, " A class of Fourier kernels ", Proc. London Math.
Soc. (-2), 35 (1933), 116-155.
4. G. Doetseb, "Bemerkung zu der Arbeit von V. Fock: Uber eine Klasse von Integral-
gleichungen ", Math. Zeitschrift, 24 (1925), 785-791.
5. G. Doetsch, "Die Integrodifferentialgleichungen vom Faltungstj'pus ", Math. Annalen,
SO (1923), 192-207.
6. V. Fock, "Uber eine Klasse von Integralgleichungen " Math. Zeitschrift, 21 (1924),
161-173.
7. S. Bochner, Vorlesungen iiber Fouriersche Integrate (Leipzig, 1932).
8. S. Pollard, " Identification of the coefficients in a trigonometrical integral ", Proc.
London Math. Soc. (2), 25 (1926), 451-468.
9. E. AV. Barnes, " A new development of the theory of the hypergeometric functions".
Proc. London Math. Soc. (2), 6 (1908), 141-177.
10. G. H. Hardy and E. C. Titchmarsh, " Solution of an integral equation ", Journal
London Math. Soc, 4 (1929), 300-304.

Birkbeck College,
London. E.G.4.