90 INTEGRAL EQUATIONS

179. Let
b b

~ ~ K 2 (x, t) dx dt = 8 2,
a a
b b

~ ~ K~(x, t)dxdt=B~
a a

where Kn (x, t) is the nth iterated kernel for the kernel

K (x, t). Prove that if 8 2 = 8 2 , then for any n we will
have 8n = Bn.

15. Integral Equations with Degenerate Kernels.

Hammerstein Type Equation
The kernel K (x, t) of a Fredholm integral equation of
the second kind is called degenerate if it is the sum of a
finite number of products of functions of x alone by func-
tions of t alone; i.e., if it is of the form
n
K (x. t) = ~ ak (x) bk (t) (I )
k=l

We shall consider the functions ak (x) and bk (t) (k = 1,

2, ... , n) continuous in the basic square a~ x, t ~ b and
linearly independent. The integral equation with degenerate
hernel (I)

<p(x)-A.f[~1 ak(x)bk(t)]<p(t)dt=f(x) (2)

is solved m the following manner.

Rewnte (2) as
n h

<p (X)= f (x) +A. L ak (x) ~ bk (t) <p (t) dt

~=I a
(3)

and introduce the notaHon'

!,

~ bk(t)cp(t)dt =Ck (k= 1, 2, ... , n) (4)

a
FREDHOLM INTEGRAL EQUATIONS 91

Then (3) becomes

(5)

where Ck are unknown constants, since the function <p (X) is

unknown.
Thus, the solution of an integral equation with degene-
rate kernel reduces to finding the constants Ck (k = I,
2, ... , n). Putting the expression (5) into the integral
equation (2), we get (after simple manipulations)

n{ Cm-)bm(t)lf(t)+\~
m~l b r n
1 Ckak(t)
]
dt
}
am(X)=O

Whence it follows, by virtue of the linear independence of

the functions am (x) (m =I, 2, ... , n), that

Cm- f bm (t)[f(t)+\~1 Ckak(t)]dt ~0

or

For the sake of brevity, we introduce the notations

b b

akm = ~ ak (t) bm (t) dt, fm= ~ bm(t){(t)dt

a a
and find that

or, in expanded form,

(l-A.a 11 ) C1 -Aa 12 C2 - -A.a 1nCn = { 1, }

~ "-~n.c ~. +. (1. -. "-~22~ ~ 2~ ..... ~ "-~2~c~ ~ f 2.' (6)

-A.an1C1-A.an2C2- + (1-A.ann) Cn =fn

92 INTEGRAL EQUATIONS

For finding the unknowns Ck, we have a linear system of n

algebraic equations in n unknowns. The determinant of this
system is
1-A,a11 -A,a12 -Aa1n
-A,a21 1-Aa22 -A,a2n
fl {J.) = (7)
-/,ant -Aan2 1-A,ann
If fl (A,)* 0, then the system (6) has a unique sol uti on
Cl' C2 , , Cn, which is obtained from Cramer's formulas
1-A,an -A,atk-1f1 -l,a1k+i -A,a1n
-A,a21 -Aa2k-1f2-A,a2k+i -A,a2n (8)

(k = 1, 2, ... , n)
The solution of the integral equation (2) is the function
qJ (x) defined by the equality
n
<p(x)=f(x)+A, ~ Ckak(x)
k=l

where the coefficients Ck (k = 1, 2, ... , n) are determined

from formulas (8).
Note. The system (6) may be obtained if both sides of
(5) are consecutively multiplied by a 1 (x), a 2 (x), ... , an (x)
and integrated from a to b or if we put (5) into ( 4) for
qJ(x), replacing x by t.
Example. Solve the integral equation
:rt

qJ(x)-A, ~ (x cost+ 12 sin x+ cos x sin t) qJ (t) dt = x (9)

-:rt

Solution. Write the equation in the following form:

:rt :rt

{/J (x) =AX~ {/J (t) cost dt +A sin X ~ / 2<p (f) dt +

-:rt -:rt
:rt

+A,cosx ~ qJ(t)sintdt+x
-n
FREDHOLM INTEGRAL EQUATIONS 93

We introduce the notations

~ ~ ~

C 1 = ~ cp (t) cost dt; C 2 = ~ t 2 cp (t) dt; C 3 = ~ cp(t) sin t dt

-~ -~ -~

( 10)
where C 1 , C2 , C3 are unknown constants. Then equation (9)
assumes the form
cp(x)=C 1 A.x+C 2 A.sinx+C 3 A.cosx+x (11)
Substituting expression (11) into (10), we get
~

C1 = ~ (C 1 A.t + C 2 A. sin t + C 3 A. cost+ t) cost dt,

-~
~

C 2 = ~ (C 1 A.t+C 2 ASint+C 3 A.cost+t)t 2 dt,

-~

C3 = ~ (C 1 A.t+C 2 A.sint+C 3 A.cost+t)sintdt

-~

or

~ ~

- C 3 A ~ cos 2 t dt = ~ t cos t dt
-~ -~

:rt ~

-C 3 A. ~ t costdt= ~ t 3 dt,
2
-~ -:rt
:rt :rt

-C 11. ~ tsintdt-C 21. ~'sin 2 tdt+

-:rt -~
94 INTEGRAL EQUATIONS

By evaluating the integrals that enter mto this system we

obtain a system of algebraic equations for finding the un-
knowns C1 , C 2 , Ca:
C1 - A.nC 3 =0, }
C2 + 4AnC 3 = 0, ( 12)
- ~A.nC 1 - A.nC 2 + C3 = 2n
The determinant of this system is
I 0 -nA
~ (A)= 0 I 4nA. = I + 2A n2 2 =I= 0
-2nA. - A.n I
The system (12) has a unique 'olut ion
2n 2 A. lln 2 A 2n
c.= l-t-2A.2n2; C2=-1+2A.2n2; Ca= 1+2A2n2
Substituting the values of Cl> C 2 , C 3 thus found into (II),
we obtain the solution of the given integral equation
~
2An n (ll.ltX- 4~ .
<p (X)= l +2A 22 ll.lt Sll1 X+ COS X)+ X

Solve the following mtegral equations with degenerate

kernels:
2

180. <p (X) -4 ~ sin 2 x<p (t) dt = 2x-n.

0
I

181. <p(x)- ~ ercin<p(t)dt=tanx.

-I

182. <p (x)- A ~ tan t<p (t) dt =cot x.

n
4
I

183. <p(x)-A ~ cos(q lnt)<p(t)dt = l.

0
I

184. <p (x)-A. Sarc cos t<p (t) dt = -V-:-;I=x=- 2

0
FREDHOLM INTEGRAL EQUATIONS 95

5(1n 7 )
I
I' P
185. cp(x)-1. cp(t)dt=l (p >-I).
0
I

186. cp (x)-A. 5 In -tIn

(x t x) cp (t)dt = ~ (l-4x).
0

187. cp(x)-A.~ sinxcostcp(t)dt=sinx.

0
2"
188. cp(x)-A. ~ ln-tJsinxcp(t)dl=x.
0

"
189. cp(x)-A. ~ sin(x-t)cp(t)dt =cosx.
0
190. cp(x)-
2"
-A.~ (sinxcost-sin2xcos2t+sin3xcos3t)cp(t)dt=cosx.
I)

191. cp (X)-
1

-+ 5 lx--+(3t -i)++t(3x -l)] cp(t)dt= I

-I
2 2

Many problems of physics reduce to nonlinem integral

equations of the Hammerstein type (see (24], [28]).
The canonical form of the Hammerstein-type equation is
b

cp(X)= ~ K(x, t)f(t. q;(i)ldt

a

where K (x, t), f (t, u) are given function<; and cp (X) is the
unknown function.
The following equations readily reduce to equations of
type ( 1):
b

cp (X)=~ K (X, t) f (t, cp (/)) dt + \jJ (X) (1 ')

a
96 INTEGRAL EQUATIONS

where 'Jl (x) is the known function, so that the difference

between homogeneous and nonhomogeneous equations, which
is important in the linear case, is almost of no importance
in the nonlinear case. We shall call the function K (x, t)
the kernel of equation ( 1).
Let K (x, t) be a degenerate kernel, i. e.,
m
K (x, t) = ~ adx) bdt) (2)
I -=I

Then equation ( 1) takes the form

m b
cp (x) = .~ adx) ~ bdt) f (t, cp (t)) dt (3)
t=l a
Put
b

ei= ~ bi(t)f(t, cp(t))dt (i= l, 2, ... , m) (4)

a

where the ei are as yet unknown constants. Then, by virtue

of (3), we will have
m
cp (x) = ~ eiadx) (5)
i= I

Substituting into (4) the expression (5) for cp (x), we get m

equations (generally, transcendental) containing m unknown
quantities e1 , e2 , , em:
ei='de 1 , e 2 , ,em) (i=l, 2, ... , m) (6)
When f (t, u) is a polynomial in u, i. e.,
f(t, U)=p 0 (t)+p 1 (t)u+ ... +Pn(t)un (7)
where Po (t), ... , Pn (t) are, for instance, continuous functions
of t on the interval [a, b], the system (6) is transformed
into a system of algebraic equations in e1 , e 2 , , em.
If there exists a solution of the system (6), that is, if there
exist m numbers
e~. e~ ... , e~
such that their substitution into (6) reduces the equations
to identities, then there exists a solution of the integral
 F~EDHOLM INTEG~AL EQUATIONS 97

equation (3) defined by the equality (5):

m
cp (x) = ~ Gpai (x)
i= I

It is obvious that the number of solutions (generally,

complex) of the integral equation (3) is equal to the number
of solutions -of the system (6).
Example. Solve the integral equation
I

cp (x) =A~ xtcp 2 (t) dt (8)

0

where A is a parameter.
Solution. Put

(9)
Then
cp (x) = CAx (10)
Substituting cp (x) in the form (10) into the relation (9), we get
I

C = ~ iA 2C 2 t 2 dt
0
whence
c-~C
- 4
2 ( 11)

Equation (11) has two solutions

4
C1 =0, C2 =1:2
Consequ~ntly, integral equation (8) also has two solutions
_for any A =1= 0:
4
cp 1 (x) E: 0, cp 2 (x) = 1' x
There exist simple nonlinear integral equations which do not
have real solutions at all.
Consider, for example, the equation
I x+t
cp (x) = } 5e-
2 (1 + cp 2 (t)) dt (12)
0

4-1583
98 INTEGRAL EQUATIONS

Put

(13)

Then
X

<p (x)=Ce2 ( 14)

For a determination of the constant C we obtain the equation

(e: -1)C 2 -3C+3(e+-1)=o (15)

It is easy to verify that equation (15) does not have real
roots and, hence, the integral equation (12) has no real
solutions.
On the other hand, let us consider the equation
1

<p(x)= Sa(x)a(t)cp(t)sin (:gj )dt (16)

0

(a (t) > 0 for all t E [0, 1])

In order to determine the constant C, we arrive at the
equation
1

1 = ~ a2 (t) dt sin C ( 17)

1
If ~ a2 (t) dt > 1, then equation ( 17) and, hence, the original
0
integral equation (16) as well have an infinite number of
real solutions.
Solve the following integral equations:
1

192. cp (x) = 2 ~ xt cp 3 (t) dt.

0
1

t93. cp (x) = ~ (xt + x 2 t 2 ) cp 2 (t) dt.

-1
1

194. cp(x)= ~ x2t 2 cp 3 (t)dt.

0
FREDHOLM INTEGRAL EQUATIONS 99

1
xt
195. !p(X)= ~ l+qJ2(t) dt.
-1
1

196. qJ (x) = ~ (l +.!Jl 2 (t)) dt.

0

197. Show that the integral equation

1

!Jl (x) = f \a (x) a(t) (1 + !Jl (t)) dt

2

"0
(a (x) >0 for all x E [0, 1])
1

has no real solutions if ~ a 2 (x) dx > 1.

0

16. Characteristic Numbers and Eigenfunctions

The homogeneous Fredholm integral equation of the se-
cond kind
b

!Jl(X)-A ~ K(x, t)qJ(t)dt=O ( 1)

a

always has the obvious solution !Jl (x) = 0, which is called

the zero (trivial) solution.
The values of the parameter A for which this equation
has nonzero solutions qJ (x):;EO are called characteristic num-
bers* of the equation ( 1) or of the kernel K (x, t), and every
nonzero solution of this equation is called an eigenfunction
corresponding to the characteristic number A.
The number A= 0 is not a characteristic number since
for A= 0 it follows from (l) that q:> (x) 0. ==
If the kernel K (x, t) is continuous in the square Q {a~ x,
t ~ b} or is quadratically summable in Q, and thf: numbers
a and b are finite, then to every characteristic number A

* Some authors use "eigenvalues" in place of the term "characte-

ristic numbers". We use eigenvalue for the quantity cr= ~ , where f.. is
the characteristic number.
4*
100 INTEGRAL EQUATIONS

there corresponds a finite number of linearly independent

eigenfunctions; the number of such functions is called the
index of the characteristic number. Different characteristic
numbers can have different indices.
For an equation with degenerate kernel

q>(x)-J.. f L~1ak (x)bk(t)]q>(t)dt=O (2)

the characteristic numbers are roots of the algebraic equation

1-/..a11 -/..alz -f..aw
~(/..) = -/..a21 1-/..a22 ... -/..a2n =0 (3)

the degree of which is p ~ n. Here, ~ (/..) is the determinant

of the linear homogeneous system
(1-f..au)C 1 -Aa12C2 - -f..a1nCn = 0, }'
~~a:~~~.~(~ ~~~2 2! ~2.~ .' ~~a:n~~ =:= ~. (4)

-J..an1C1-J..an2C2- + (1-J..ann) Cn =0
where the quantities amk and Cm(k, m= 1, 2, ... , n) have
the same meaning as in the preceding section.
If equation (3) has p roots (1 ~ p ~ n), then the integ-
ral equation (2) has p characteristic numbers; to each cha-
racteristic number "-m (m = 1, 2, ... , p) there corresponds
a nonzero solution
Cp>, c~u, ... , c~u __,. "-~
q2>, C~2>, , C~2> __,. "-2

C (pl
1 '
C<Pl
2 ' '
C<P>
n ----r ""P
~

of the system (4). The nonzero solutions of the integral

equation (2) corresponding to these solutions, i.e., the eigen-
functions, will be of the form
. n n
q> 1 (x) = ~ qu ak (x), q> 2 (x) = ~ q 2> ak (x), ... ,
k=l k=l
FREDHOLM INTEGRAL EQUATIONS 101

n
CJlp (x) = ~ C),P> ak (x)
I<= I

An integral equation with degenerate kernel has at most n

characteristic numbers and (corresponding to them) eigen-
functions.
In the case of an arbitrary (nondegenerate) kernel, the
characteristic numbers are zeros of the Fredholm determi-
nant D (A.), i.e., are poles of the resolvent kernel R (x, t;
A.). It then follows, in particular, that the Volterra integral
X

equation cp (x)- A. ~ K (x, t) cp (t) dt = 0 where K (x, t) E L 2 (Q 0 )

0
has no characteristic numbers (for it, D(A.) =e-A,I., see
Prob !em 177).
Note. Eigenfunctions are determined to within a multi-
plicative constant; that is, if cp (x) is an eigenfunction cor-
responding to some characteristic number A., then Ccp (x),
where C is an arbitrary constant, is also an eigenfunction
which corresponds to the same characteristic number A..
Example. Find the characteristic numbers and eigenfunctions
of the integral equation
n
cp (x)- A. ~ (cos 2 x cos 2t +cos 3x cos 3 t) cp (t) dt = 0
0

Solution. We have
n n
cp (x) ==A. cos x ~ cp (t) cos 2t dt +A. cos 3x ~ cp (t) cos 3 t dt
2
() 0 .

Introducing the notations

n 11

C 1 = ~ cp {t) cos 21 dt, c2 = scp (t) COS 3 t dt (1)

we get
(2)
102 INTEGRAL EQUATIONS

Substituting (2) into (1), we obtain a linear system of ho-

mogeneous equations:
n n
C1 (1-'A ~ cos 2 t cos 2t dt)-C 2 'A ~cos 3t cos 2t dt = 0,}
n o n o (3)
-C 1 'A ~ cos"tdt+C 2 (1-'A ~ cos 3 tcos3tdt=0
0 0

But since
n n
Scos t cos2tdt = ~ , Scos3t cos 2t dt = 0
2

0 0
n n
Scos" t dt = 0, Scos 3 t cos 3! dt = ~
0 0

it follows that system (3) takes the form

(1-"'4:n;)c 1 =0,}
(4)
n) c
( I - "-8 2 = o
The equation for finding characteristic numbers will be
1-"':n;4 0
A.:rt =0
0 1--g
The characteristic numbers are 'A1 = n4 , 'A2 = n8 .
:4n;
For 'A=-, system (4) becomes

{ ~C 1 =0,
2. c2 =0
whence C 2 = 0, C1 is arbitrary. The eigenfunction will be
cp 1 (x) = C1 'A cos 2x or, setting C1 'A= 1, we get cp 1 (x) = cos 2 x.

:8n;
For 'A=-, system (4) is of the form
(-l)C 1 =0,
{
OC 2 =0
FREDHOLM INTEGRAL EQUATIONS 103

whence cl =0, c2 is arbitrary and, hence, the eigenfunc-

tion will be q> 2 (x) =C 2 A. cos 3x, or, assuming C2 A. = 1, we
get q> 2 (x) =cos 3x.
Thus, the characteristic numbers are
8
A.l = -4 :rt'
A.=-
2 :rt

and the corresponding eigenfunctions are

q> 1 (x) = cos 2 x, q> 2 (x) =cos 3x
A homogeneous Fredholm integral equation may, gene-
rally, have no characteristic numbers and eigenfunctions,
or it may not have any real characteristic numbers and
eigenfunctions.
Example 1. The homogeneous integral equation
I

q>(x)-A. ~ (3x-2)tq>(t)dt=0
0

has no characteristic numbers and eigenfunctions. Indeed,

we have
1

cp(x)=A.(3x-2) ~ tq>(t)dt
0

Putting
I

c = ~ t fP (t) dt (l)

we get
fP (x) =CA. (3x-2) (2)
Substituting (2) into (1 ), we get

[1-A.j (3t 2 -2t)dt}C=0 (3)

1

But since ~ (3/ 2 -2t) dt = 0, equation (3) yields C = 0 and,

0
hence, cp (x) = 0.
104 INTEGRAL EQUATIONS

And so for any ').., this homogeneous equation has only

one zero solution <p (x) = 0 and, hence, it _does not have
any characteristic numbers or eigenfunctions.
Example 2.- The equation
1

<p(X)-'A, ~ (Vxt- Vt x) <p {t) dt = 0

0 '

does not have real characteristic numbers and eigenfunc-

tions.
We have
( 1)
where
I I

ci = ~ tcp (t) dt, C2 = ~ VT <p {t) dt (2)

0 0

Substituting (1) into (2), we get (after some simple mani-

pulations) the system of algebraic equations

( 1- 2; )c+~c = 1 2 0' }
- ; ci + ( 1 + 2; ) c 2 = 0 (3)
The determinant of this system is
1 -~1 5 fl.
'A
3 'fv2
f.. (A)= t.. 2 A = 1 + 150
-2 l+s-
For real A, it does not vanish, so that from (3) we get
C1 = 0 and C2 = 0 and, hence, for all real 'A the equation
has only one solu!ion, namely, the zero solution <p (x) 0. ==
Thus, equation ( 1) does not have real characteristic numbers
or eigenfunctions.
Find the characteristic numbers and eigenfunctions for
the following homogeneous integral equations with dege-
nerate kernels:
n
4
198. <p (x)- A, ~ sin 2 x<p (t) dt = 0.
0
FREDHOLM INTEGRAL EQUATIONS 105

2n
199. cp (x)- 'A ~ s.in x cos tcp (t) dt = 0.
0
2n
200. cp (x)- 'A ~ sin x sin tcp (t) dt = 0.
0
n
201. cp(x)-'A~ cos(x+t)cp(t)dt=O.
0
I

202. cp (x)- 'A ~ (45x 2 In t - 9/ 2 In x) cp (t) dt = 0.

0
I

203. cp(x)-'A ~ (2xt-4x 2 )cp(t)dt=0.

0
I

204. cp (x) -'A ~ (5xt 3 + 4x 2 t) cp (t) dt = 0.

-I
I

205. cp (x)- 'A ~ (5xt 3 + 4x 2 t + 3xt) cp (t) dt = 0.

-I
I

206. cp (x)- 'A ~ (x cosh t- t sinh x) cp (t) dt = 0.

-I
I

207. cp (x) -'A ~ (x cosh t - t 2 sinh x) cp (t) dt = 0.

-I
I

208. cp(x)-'A ~ (xcosht-tcoshx)cp(t)dt=O.

-I

If the nth iterated kernel Kn (x, t) of the kernel K (x, t)

is symmetric, then it may be asserted that K (x, t) has at
least one characteristic number (real or complex) and that
the nth degrees of all characteristic numbers are real num
bers. In particular, for the skew-symmetric kernel K (x, t) =
=- K (t, x) all characteristic numbers are pure imaginary
'A= ~i, where ~ is real (see Problem 220).
The kernel K (x, t) of the integral equation is called
symmetric if the condition K (x, t) = K (t, x) (a:::;;; x, t :::;;; b)
is fulfilled.