T
APPLICATIONS OF THE HILBERT
PROBLEM TO PROBLEMS OF
MATHEMATICAL PHYSICS
PROEFSCI-IRIFT
TER VERKRIJGING VAN DE ORAAD VAN DOCTOR IN
DE TECHNISCHE WETENSCHAP AAN DE TECHNISCHE HOCESCHOOL rE DELET, KRACHTENS ARTIKEL 2 VAN
HEI KONINKLIIK BESLUIT VAN 16 SEPTEMBER 1927, STAATSBLAD NO 310, OP GEZAG VAN DE RECTOR MAGNIFICUS DR O. BOTTEMA, H000LERAAR IN DE
AFDELINO DER ALOEMENE WETENSCHAPPEN, VOOR EEN COMMISSIE UIT DE SENAAT TE VERDEDICEN OP
WOENSDAG 5NOVEMBER 1958 DES NAMIDDAGS TE 4 UUR DOOR
JOHAN ADOLF SPARENBERO
Rau de nagedachtenis van mijn Ouders
Ran Paula
Content s
IntroductionChapter le The Hubert problem I
1 .1 The formulae of Plemelj I
1.2
The Hubert problem for an arc 31 .3 Singular integral equations 5
Chapter
2.
The Wiener-Eopf type integral equation 72.1
Transformation of the equations 72.2
Solution of the homogeneous integral equationof the second kind 8
2.3
The irthornogeneous equation of the second kindand the equation of the first kind 10
2.14 Examples
12
Chapter 3. A shrink-fit problem for a half infinite
range of contact 114
3.1
Formulation of the problem, determination ofGreen's function for the tube
ILl-3.2
Solution of the integral equation for thecontact pressure 17
3.3
Numerical calculation of the shrink-fitstresses 21
3.14 The limiting case of an infinitely thick tube 22
3.5
Discussion of the results 23Chapter Li.. The homogeneous first order
integro-differential equation of the Wiener-Hopf type 25
14.1 The general equation, case 1 25
14.2 Case 1 with a0 0, a = O and b1 # 0 28
14.3 The general equation tase 2 30
Li.Lt Example 32
Chapter 5. A Wiener-Hopf type integro differential
equation with fourth order derivatives 35
5.-1 Fourier transformation of the equation 35
5.2
The asymptotical expansion of E(A)39
Ohapter 6. The finite dock Li.6
6.1 Formulation of the problem Li.6
6.2
Derivation of the integral equation Lj.86.3
Determination of '(x) 506.L- Calculation of (x) 52
6.5
Construction of the solution for prescribedincoming waves 5L.
6.6
Numerical calculation of the reflected andtransmitted wave for the case that no
breaking occurs 56
/
Chapter7.
On the inf'ience Of the cross-section formof a ship on the added mass for higher order
vibrations 59
7.1 Formulation of the problem for the case of a
strip 59
7.2
Derivation of the integral equation6i
7.3
The cases a = O and a - O 637.14 The case a - 65
7.5
The case of finite a 677.6
The added mass for the strip 707.7
The added mass for a vibrating infinitecylinder 70
IN TR ODUCTIOI
Problems arising in mathematical physics,can in general
'ce classified from different points of view. From the
physical formulation a classification to several branches of physics and mechanics is obvious. From a mathematical
point of vìew, problems, originating in different parts of
physics,can sometimes be submitted to a uniform treatment by the same mathematical method. The problems dealt with in
this thesis have in conwion, that they all can be formulated
in terms of the Hilbert problem. This problem can be treated by the use of sectionally holomorfic functions (ref.1),which
concept is based on a set of formulae derived by Plemelj.
These functiors are regular in the whole complex domain with the exception of a discontinuity on a curve. In chapter 1 we give a short survey of the notations and results of this
theory.
The chapters
2-5
are concerned with Wiener-Hopf typeintegral and integro-diff'erential equations. Usually these equations are solved under the condition that there exists some strip of convergence for the Fourier-transformations (ref. 2,3).
Using
however, the theory of sectionally hob-morfic functions it becomes clear that a strip is not essentialand we need only to demand convergence on a line. In chapter 3
we shall discuss an application of the theory on a problem of shrink-fit stresses. The stresses are calculated by a method which is equivalent to the procedure for obtaining approximate
solutions described in an article by W.T. Koiter (ref. 24). In
the next chapter we discuss the homogeneous Wiener-Hopf type
integro-differential equation with first order derivatives.We
conclude the treatment of Wiener-Hopf type equations by
conside-ring an integro-differential equation with a fourth order
deri-vative which occurs under the sign of integration. This equation is a result of consideraequations about the anomalous skin
-effect of electrons in a metal (ref. 5) and is a generalisation
of an equation discussed by Reuter and Sond.heimer (ref. 6).
In the chapters 6 and 7 we discuss two problems on the motion of water with a free surface. First the two dimensional problem of the reflection and transmission of progressing waves, when a part of the surface of the water is fixed by introducing on the surface a rigid strip of infinite length. This is the so called finite dock problem. A proof of the existence of the solution
has been given by H. Ruhm (ref. 7), while Mac Carny (ref. 8)
dis-cusses the pressure under the dock. Here we have to solve a HUbert problem for a function which possesses some prescribed discontinuity on the strip. The second problem of this part
considers the forces exerted by the water when the strip executes
are compared with the forces at a half ininersed cylinder which executes a vibration with shear deformation. This has
been done in order to obtain a check on the three dimensional correction coefficient for the added mass, used in naval
architecture (ref.
9).
The question arose whether thiscorrec-tion coefficient, derived from the added mass of a vibrating ellipsoid of revolution, would be accurate enough for ships of shallow draught. We have to consider a singular integral
ecuation which can be reduced, by the theory of sectional
holomorfic functions, to an integral equation of the Fredhoim trpe.
Chapter
ITEE HILBERT PROBLEM
The formulae of Plemelj, the Hubert problem and singular
integral equations are discussed thoroughly in ref. i For
direct reference, however, we shall state some results. We
shall not enter into details but consider the theory to an
extent necessary for understanding the applications.
-1 .1 THE FORMULAE OF PLEMELJ
Let L be a smooth arc, in the complex plane, defined by
x=x(a),
y=y(s), SaSSb
(i.i.i)where s is a parameter and x(s) and y(s) have continuous first ordex derivatives which do not vanish simultaneously. Also we assume L to be simple; this means that never x(s1) = x(s2) and
zb
z=c
4_ (to)
.1 .1 Arc of discontinuity for s sectionally holomorfic function.
y(s1) = y(s2) for values
8a i s2 S-e.
On this arc we consider a function ç9=cp(t) which satisfies the Hölder condition
k(t2)-ç(t1 )I <A t2-t1 I , (i .1 .2)
where t=t(s) is a point of L which corresponds to the
para-meter value sand A and /L are positive constants. Then we form
the Cauchy integral
2.
From this definition we see that (z) Is an analytic function
in the entire complex domain with the exception of the arc L.
At L the values of (z) exhibit a jlxnp by passing from one
side of L to the other. Introducing on L a positive direction for increasing values of the parameter s we call the left hand side of L the positive "+" side and the right hand side the negative side. The limiting values of (z) are denoted
respectively by (t) and _(t) (fig. i.i.i). The values of
(z) are continuois up to L, with exception of the ends of L
for which q'(t) # o and satisfy, as Is proved in ref. 1, the following relations of Plemelj
- jt0) = (t0) (1.1 .L1.)
and
_j_
jrc(t)
+ _(t0) - t_t0 dt.
The integral is to be taken in the sense of Cauchy
Idt
= C
L-dt, (i .1 .6)
where C is a part of L with ends t1 and t2 in such a way that
< o < and 1t2-t01 = k0-t1
I.
Formulae (i .1 .L.) and (i .1 .5) can be verified directly in the case that ç(t) represents the values on L of a function
p(z) analytic in a neighbourhood of L. In this case we may
deform L slightly (fig. 1 .1 .2) in order to calculate for
instance the limit value
(t)
dt + q (t0).
= t_t0
/
r'i,1.i.2.
Deformation of L when (z) is analytic.- r
lnG(t
¶(z) = exp
- (t-z) dt. (i .2.Lia)
However by multiplying this function by an arbitrary ratioii
function P1(z), which may possess poles only at the ends of
L, we do nôt disturb relation (1.2.2). Hence
¶(z) = E exp
-dt P1 (z) (1 .2.L4b)
is a more general solution of (1.2.2).
In order to deal with the inhomogeneous equation,we write (1.2.1) in the form
_(t) (t)
_____ - w(t) - (1.2.5)
Analogous
= dt - (-t .1 .8)
Subtracting these formulae yields (i.i.L), adding yields
(-i I .5).
Now we Con-e to the definition of a sectionally holomorfic function. A function will be called sectionally holomorfic
when it is holomorf Ic in each finite region which does not
contain points of some smooth line L, while it is continuous up to L with possible exception of the ends of L.
It is proved in ref. I that (z) is such a function.
I 2 THE HILBERT PROBLEM FOR AN ARC
The problem is to find a sectionally holomorfic function
(z) which satisfies the relation
- c(t) (t) = g(t) (1.2.1)
on a smooth arc L, where G(t) and g(t) are given functions , which satisfy the H6lder condition and &(t) # O on L.
First we consider the homogeneous equation
= G(t) ¶(t). (1.2.2)
We solve this by trying to find a sectionally holomorfic
function ¶(z) without zero's in the whole complex domain.
Then in
F(Z)
is also sectionally holomorfic and has tosatisfy
Ein (t)1 - Ein r(t)_ = -in G(-t) (1.2.3)
on L. Comparing (1.2.3) and (1.1.Li) we find as a solution of
L1..
Hence again by (i .1 .Li.) we obtain
ç
f
g(t)
t
P ' ' " 6
Zj
-2ici ¶_(t)(t-z) d + 1.2.
where it is assumed that W(z) is chosen In such a way that
the integral converges, P (z) is again an arbitrary rational function with the same retriction as P (z) in (i .24b).
In ref. I lt is shown that (i .2.6) aid (1 .2.Lb)
are
thegeneral solutions of the probleme (1.2.1) and (1.2.2) when
the behaviour at infinity is prescribed to be algebraic.
When the arc L becomes an infinite line, for instance
parallel to the real axis,the sectionally holomorf Ic function '(z) in (i .2.L1.) is cut into two separate functions ¶(z) and.
F_(z) which are analytic in the half planes S
and S,
situated at the + and - side of L. Assuming tat the integral
in (1.2.L.) is convergent for this line we see that the
solution of (1 .2.2) yields the "factorisation" of a function G(t) defined on L
G(t) = ¶_(t)/ ¶(t) (1.2.7)
into two functions W(t) and _(t) which are boundary values
of functions W (z) and. W_(z) regular and without zero's in
and S respetively and continuous up to L. It Is clear
that we have to take z in S or S_ when we calculate W+(z)
or ¶_(z) with the use of (1.2.L).
The next step done in (1 .2.6) Is to form, under the assumption of convergence, the integral
(t) dt, z in S. (i .2.8)
This means for an infinite line L that we "split" the Í'uncti g(t)/'_(t) defined on L
=
e(t) - e_(t)
(1.2.9)
into two functions e(t) and e_(t) which are boundary values
of functions
8(z) and 8_(z) regular in S
ana
Sarid
continuous up o
L.
Sometimes it.is necessary to multiply G(t) by a simple
function in order to produce convergence of the integral in
(1.2.L). This will be demonstrated later on (para. 2.3).
The solution of (1.2.1) for an infinite line can now be
described very briefly as follows. First, factorize G(t),this
yields (1.2.5). Second, split the right
hand
side of (1.2.5).Third, compare the parts analytic in the same half planes
with the result
W(z)
1 .3 SINGULAR INTEGRAL EQUATIONS
At last we discuss in this chapter the relation between
singular integral equations and the Hubert problem. In
chapter 7 we shall have to treat a singular integxal equation of the form
i(-t0) + k(t-t0)1 (t)dt = f(t0), (1.3.1)
where (t) is the unknown function and k(t) and f(t) are
given functions, k(t) is bouzided at the interval Jt < 2.
First we consider the t'dominantt' part of this equation,defned
as
1
,-T 1' (t-t
dt = r(t0), (1.3.2)
o
which is closely related to the theory of sectional holomoific functions. IntroducIng
+1
«z) = iT
I
?i:t:dt (i .3.3)
-.i
we find from (1.1.5) and (1.3.2)
+ _(t) = r(t). (1.3.L)
Substitution of G(t) = -1 in (i .2.L.) shows that we may take as a solution ¶(z) of the homogeneous part of (i .3.L.)
¶(z)
(1_z2)2
The solution of (i .3.U) becomes by using (i .2.6)
«z)
2f
r(t)(-t2)dt
+ 2iri(1-z )
where we have fixed the values of (z) by taking
T(t) =
-I(i-t2)I,
t 1. Then by (1.i.L.)(t )= 2
7f(t(1;t2)
dt + AI. (1.3.7) ° ti(1-t)2 ¿-Here we have chosen for the arbitrary polynomial P(z) the
arbitrary constant
A
i f
ç(t)dt, (i .3.8)in order that (1.3.7) satisfies (1.3.2).
(i
6,
Now we return to (i .3.1) which we write in the form
t! dt =
r(t0)- f
(t)
k(t-t0)dt.(i .3.9)
Then apparently each solution of
+1
+1 fç(t)k(t_r)dt(1_r2)ciT
1i(i_t)*
(T-t)
where (t ) is a solution of the dominant part and B is sorne
arbitray Sonstant, satisfies also (i..9). This can be veri-fied by devidin,g both sides of (1.3.10) by (t-t ) and
inte-grating from -1 to +1 with respect to t0. Instead of (1.3.10)
we may consider +1 B ri(i-t3
(t01t0)+[ t(t) K(t,t0)dt
(1.3.11) where í'(t0) = o, and +1 k(t-T) (i 2) dT. (i .3.1 2) K(t,t0)=i(1_t)(1_t2)*_/
(T-t0)
This integral equation is of the second kind and the known
function and the kernel are quadratic integrable. Hence the
theories on the Neumann expansion of the solution and the
replacement of the kernel by approximating kernels of a
simpler type can be applied.
(i .3.10)
Chapter 2.
THE WIENER-HOPF TYPE INTEGRAL
E Q U A T I O N.
We now shall give a treatment of Vìener-Hopf trpe intel
equations, which resembles the classical procedure. It
deviates however at one point since we shall not demand a strip of convergence of the Fourier integrals to be used,but
only a line. This is an inmediate consequence of the fact
that we start from the concept of the. sectionally holomorfic function.
2.1 TRA.NSFORMATION OF THE EQUATIONS
The following three equations are considered , the
homogeneous equation of the second kind
f(x) -
k(x-) f()d
O,o
the inhomogeneous equation of the second kind f(x)
-
f
k(x) f()d
=and the equation of the first kind
fk(x) f()d
=where k(x-) is the kernel of the equations, h(x) is a
function known for X > O and f(x) is the unknown function. The equations are valid for all values
oo < X <
l-oo hence we shall have to cieterminein equation (2.1.3) also the values of h(x) for x(o.We introduce the notation Cor the Fourier transformation of a function f(x)
and for the one sided transforms
ff(x)e1 dx = F(X)
(2.1 .L)
-1
f(x)ei
= o I [f(x)e1 (= F(A).
,rj
-cIn order to apply a 1ourier - transformation to (2.i.i),
(2...2) anci (2.1.3) it is sufficient to assume that
e' k(x), e
h(x) and e f(x)belong to L (-, +oo). On these assumptions the Fourier transform K(A) of k(x)exists for Im A
= ¡
and the one sided transforms of k(x) andr(x) denoted by H(A), F(A) and H(A), F_(A) exist for
Im A
= ji
and are regular in S with Im A> ji
resp. S_ with Im A <ji
Sy the convolution theorem equations (2.1.i), (2.i.2) and
(2.
.3) are equivalent to the following Hubert - problemsi«Xfl F(A)
= O,F_(A)+I-
fE
K(A)F(A)=U(A)+H(A),
F(A)
K(A) - H_(A) =holding on the line L of infinite length
where F_CA), F
(A)and H
(A)are unknown
we consider F_tA) - H_(A7 as one unknown
vVe
assume that the known functions
K in (2,1,7), (2,1,8) and (2,1,9) satisfy tion (, 2) on L, in which case we canof chapter 1.
with Im A =
functions.In (2.1 .8) function.
())
and H
(X)
the Hölder
condi-apply the theory
2.2 SOLUTION OF THE HOMOGENEOUS INTEGPL EQUATION OF THE
SECOND KIND
We first consider the ifilbert problem (2.1.7) which
corresponds to the homogeneous eçuation (2.1 .1) and assume a
strip ¡9 of convergence for the integral (2.1 .Ll) for k(x).The
case of a line of convergence will be discussed at the end of
this paragraph. In ¡3 we choose a
line
L parallel to the realaxis in the A plane, on which no zero of
i-TE
K(A) lies.The line L with a positive direction (viz. Re A -+ +) defir
the half planes 8 and S_.
Of course we have no knowledge a priori whether the
Fourier-transformation of (2.1 .1) actually holds on L.However,
if
the transformation holds on some line L
in ¡3we may
de-termine this line afterwards and construct the solution to
w_(A) +
K(Afl
(Xa)n
(A-b)
where a is a point in 2 and b a point in S_, then ( 2.2.1) is satisfied. ive can write down at once the solution of
(2.2.3) with the aid of (1.2.4) and the discussion above
(1.2.7) 'n in[i- K(7fl IIJ exp - .- ¡ " diP(A),A in S.f. -
ii
(r-X)Hence also the functions
F(A)=(X-b)
¶(A),
F_(A)=(A-a) ¶_(X) (2.2.5)(2.2.6)are analytic in 2 and S respectively and are solutions of
the Hubert problem (2.1 .7)
We wish to obtain functions F (A) and F(A) which are
Fourier-transforms, this restricts P(A), because for Re A - ± F (A) and F_(A) must tend to zero.
Now suppose that, for the case of a strip
ß
of convergexof the transform of the kernel k(x), the function
i-f
K(X)1 has zero's A1; A2;.... in the S part of , arranged according to increasing imaginary parts. We investigate the functionsF(A)
F(A)
(A-A
F1JA)_
(x-A (2.2.7)(2.2.8)'
1' \ 1'
cc
These functions are regular in which are situated each on one parallel to L but lies between
F (A) is holomorfic in S
homorf Ic in S1 follows rorn
(2.2.3)
(2.2.L)
the regions S + and S
side of a lin L , which is
the zero's A anL A2. That
obvious and 1chat F (A) is
equation (2.1.7) bcause
In order to be able to apply the theory of para. 1 .2. it
is necessary that the integral in (.2.) converges for the
line L of infinite length. This is the case when
lin in {i K(Afl
= 0.
(2.2.1)
Re A-+±00
On first sight this seems to be true because K(A) is a
Fourier-transform and hence K(A) - O for Re A -+
.
Itre-mains however possible that
urn lni-.#T
K(A)}- um ini-4
K(Afl=-2rsii (2.2.2)Re A-+ Re A-+-cc
where n is an integer. If we consider instead of (2.i .7) the problem
lo.
F(X) and i-r K(Afl must have the saine zero's in the
part of ß. So it is possible, by moving u L parallel to the
real axis, to construct new functions F
+(x)
and Fwhich are of lower order for JA! an. which satisfy
equation (2.1 .7) on new lines L1 , L2 ...
In the case that F(A) and F(A) are not
Fourier-trans-forms, it is possible that F , F or F , F .... will be.
I-f-. Ir
¿t.
¿-,-.So the most direct way to obtain ail solutions is to take the line L above the zero with the largest imaginary part within
the strip of regularity of K(A) and to evaluate f(x) by the
inverse transformation. However it may be more convenient
to take for L, if possible, the real axis, in view of the
evaluation or approximation of the integrals in (2.2.Li) and
to translate L afterwards into the correct position. Vhen the transformation of k(x) is only permitted on a
line L, while {i-'T7z K(Afl has no zero on L we can use the
above theory. If in this case -T K(Xfl does possess zero's
on L we can also solve the problem, this will be discussed
in para. L..L1..
2.3 THE INHOIiOGENEOUS EQUATION OF THE SECOND KIND AND THE
EQUATION OF THE FIRST KIND
VVe are now in a position to solve the inhomogeneous
equation (2.1 .8). It is assumed that the line L, on which we consider the Hubert problem, is within the strip ß and above the zero of ti-.1r K(Xfl with the largest imaginary part within ß.
First we consider the case that the integer n defined in
(2.2.2) is positive or zero. We select a suitable solution '
the homogeneous equation (2.1 .7) from the set provided by (2.2.5) and (2.2.6) by taking for P(A) sorne polynomial of degree n without zero's on L. Denoting this solution by Y(A) and y(A) we may assume
hrn y(A)
= 1 .(2.3.1 )
Re A±o
-We can rewrite (2.1 .8) in the form
IF_(Ä)-H_(A)}
(A)H(X)
Y(A)
- Y(A) - Y(A)
This equation has by (1.2.10) the solution
Y(A)
F(A)=
- 2ai
/
Y_()(-X)
d, A in
d, A in S_.
In this case (n o) the solutions (2.3.3), (2.3.Li.)
can be
interpreted as Fourier-transforms and we find the solution of (2.1.2) by the inverse transformation.For n < O we cannot find functions (A) and *'(X) with
the property (2..1) and in general we arinot interpret in
this case (2.3.3) and (2.3.Li-) as a Fourier transform. Only for special functions H(7?) (2.3.3) and (2.3.L.) will tend to
zero for Re A
±,
We now shall treat equation (2.1.9) which reads
F(A)
K(X)-H(X) =iç(A). (2.3.5)This equation differs from (2.1.8) in a rather significant
way, viz, the function K(A) which is here the factor of
F+(A) tends to zero when Re A -+±oo. This mei that we
can-not use (i .2.Li.) because the integral will can-not exist for the
line L which is infinite. We assume the existence of a
function K*(X) with the properties
um K(A)/ K*(X)
= -i , (2.3.6)Re A -± and
IM(A)i I = IK(A)/K*(A)_1 I < e. < i, (2.3.7)
for A on L. Further we assume that the function K*(A) can be "factorized" by inspection in the following way
K*(X) = K(A)/K*(A), (2.3.8)
where K(X) and K*(A) are functions which behave algebraicly at IAl
-
and are regular and without zero's in S and S_ respectively. Then we consider the equation*[ x(A) M(X) - x_(A) = H(A) K*(X)
which is of the type (2.1.8),
while the integer is zero. Hence the problem can be treated in the way. The solutions of the original equation (2.3.b e corne
F(X)=X(X)/K(A)
, F_(X)=X_(A)/K(A). (2.3.10) When e, defined in (2.3.7) can be made sufficiently small we can use this method for obtaining approximate solutions.This will be demonstrated in chapter 3, where we discuss a
shrink-fit problem.
(2.3.9)
n (2.2.2)
indicated 5) then
1 2.
2.L- EXAMPLES
First we consider an equation which is solved in ref. 2 by the method of Viener and Hopf
f(x) = m
[e_1X
f()d.
(2..1)
We assume O < m < . After transformation we find
(p') 2 + F (X) o. (2.L.2)
+ (i+X)
The zero's of
- 2 are X1 = - (2m-1 ), X2= +(2m-1 ),
(i+X )
where we assume Im X > O. The strip of convergence for the
Fourier-transformatign of the kernel is -1 < Im X < + i In this strip we choose the real axis as the line L, on which w have to solve the Hubert problem (2.L.2). On this line
1
-
>0 and hence the number n in (2.2.2) is zero. The1 +X integral in (2.2.L) becomes +oo inh- 2m +,
f= ln (77-X) inh-
i+2
/
-r 7? iniiÄI
f
H-
ln (77-X)2m(i+î
)- m
-
j (1+772)(i+772_2m)This last integral can be calculated by residues, where we
have to take care that our contour does not enclose the branch point 77 = X. Substitution of the results in (2.2.L) yields
(X+i
= (-X1 . p(X) , X in S
F(X)
=(X-i)
. p(X) , X inThese two functions do not tend to zero for Re X -
±o
and.hence cannot be interpreted as Fourier-transforms. However,
by translating L upwards over a distance between Im X2 and
(7+i' p(Â
F(A)
=(X-1)(A-2 - (X-i)
It is obvious that P(x) must be a constant. Br the inverse
transformation we
find
the solution of (2.L.1).f(x) = C fcos ((2m-1) x)
x>O,
(2rn-1) (2.L.E
f(x)
=Ce
X,x<O.
As a second example we consider the equation co
f(x)-m
f
f(E)2 = h(x), m
. (2.L.7)
i+(x-)
This equation cannot be solved with the method of Wiener and Hopf because the Fourier-transform of the kernel converges
only on the real axis.
Fourier-transformation yields the equation
+
F(X)=H(X)
+ H_(X). (2..8)The function I-mze is positive on the real axis and
hence n (2.2.2) is zero. The solution of the homogeneous part of (2.L..8) is obtained from (2.2.L) and this solution
can not be interpreted as a Fourier-transform. However, by
taking P(A) 1, we may obtain the solution of the
inhomogeneous equation by the procedure outlined in para.2.3. In the present case
Y(X)=exp -
T1;f
,
A ins.
(2..9)
00
Then the solutions (2.3.3) and (2.3.Li.) may be interpreted as
a Fourier-transform. The solution of (2.L-.7) can now be
evaluated by the inverse transformation.
A SHRINK-FIT PROBLEM FOR A HALF
INFINITE RANGE OF CONTACT
We consider an in'inite elastic tube which is shrunk onto a semi infinite rigid shaft. In dealing with this problem we aim exclusively at the contact pressure between shaft and tube, for which an integral eauation of the Wiener-Hopf trpe is established. The integral representing the contact pressux is approximated numerically by a method equivalent to the one developped in ref. L1., which rests on approximating the kexl of the governing integral equation.
3.1 FOR1UITION OF TLIE PROBLEM, DETERMINATION OF GREEN'S
FUNCTION POR TI TUBE
Let be cylindrical coordinates such that the
axis_coincides with the axis of the tube (fig.3.1 .1). Let a
and b
be the inner ari outer diameter of the tube,respectiy. Assume that the uniform radial shrinkage of the tube isThe stress distribution in the tube for the case to be considered here is then governed by the following boundary conditions:
Chapter
3where Or() and
T are the normal and the tangentialstresses_rspectivey and u is the radial displacement. ie
seek o (a,x) for O < x < appropriate to the stress
distriEution governed by (3.1.1), (3.1.2) and (3.1.3). To
this end we note that this normal stress roust satisfy the
integral equation
for''
(3.i.L1.)for
r=b,
-
00X < + co,
Trx = O, 0r = O, for<O,
Trx=OOr=O
where () is a Green function which will be defined presently.
for
Fig.3. .1 Infinite elastic tube ahrk ontn a semi-infinite zigid shaft.
The function () is the radial displacement of the
inner wall of the tube corresponding to the following singular loading conditions:
for
co<<+oo5 T
=0,cr =0,
rx r -co < X < + co,
Trx 0,cr ò(x), (3.1.6) r forT)1
0r' x'°o
0,
(5.1.7)
in which á'(x) is the delta function of Dirac. The rotational
synmetry of this problem suggests an approach by means of
Love's stress function (ref.
io)
p(r,x) which satisfies thedifferential equation
2
22
d2ç1
(_+1__+)
rp =0.
Or
ròr
0xThe associated stress field is given by
Ox
= -4: [
(2- -Ox 2:
Or 2 Ox (s.l .5) (3.1 .8) (3.1 .9)16. = - [vLi - -_]
ÇD (:,3),
òx rôi' o2 - -. --- [ (i-v)- .-]
(r,x), Ox i iA -e (r,x)d, bL1< Im A
<co
exists and (3.1
.8)
is carried into(!
+ i - A ,A) =o.
&r
rÒr
(3.1 .9)
where
t'
designates Poisson's ratio, which we take equal to0,25 in numerical calculations.
The corresponding radial and axial dispiscements appear as
- - - (i+v)
o2)
(3.1.10) u(r,x)=-E --ô rO x = (1±v)[(1_2v) + + ] (3.1.11) a Òrròr
Thus we need to determine a function p which meets
(3.1 .8) and is such that the stresses 3.1
.9)
conform to(3.1.5), (3.1.6) and (3.1.7). The desired kernel in (.i .1i) is then obtained from
- -
- - - (1+v)o2()
u (X) = u(a,x) = - E - -
-o òrÒx a,x
We now establish with the aid of the Fourier
-transform. Let
e1X(,)
be absolutelr integrable withrespect to X in the interval
( , +)
l'or < Im A <then the Fourier transform
+00
(3.1 .1 2)
(3.1 .13)
(3.1 .1)4)
The general solution of (3.1 .1)4) admits the representetion
= . (A)K(A) + A2(A)AK1 (Ai) +
+ B1(A)I0(X) + B2(X)2cI1(A), (3.1.15)
Bessel-functions of first and second kind, respectively, Throughout this work we shall use the definitions for these functions
given in ref. 11. The arbitrary functions
A (X),
A2(2L),B (X)and B (X) are to be determined consistent with the transorms
of th boundary conditions (3.1.5) and (3.1.6). This process
yields four linear equations in the four unknowns A1 , A2, B1
Thus (,A) is comletely determined.
2 Upon in9ersion of and substitution in (3.1.12)
we obtain after some computation
-(i-v2D
f
N(,X)
(3.1.16)u0()_
nEXD(,,X)
00 with =X[K1
(À)I0(À)
+ K0(X)I1 (2)]2 -[2(1v) + X[(K1(x)I1(X)- K1(X)I1(X)]2-
, (3.1.17) 2Lb = [ 2(1-v)+][K1(x)I0(A)+K0()I1(x)]2
+ Aa + X { 2(1-v)+][K1 (A)I0(A)
+ K0(A)I1 (AS)]2-Ab
- [K0(xS)I0(x) -
K0(A)I0(AS)]2
-[ 2(1-V) +
A] [
2(1V)
+ Ag]Aa Ab
[K1 (XS)i (Xi)-
K1(AS)]2
[(Iu)
¡
+ .(3.1.18)
S
3.2
SOLUTION OF THE INTEGRAL EQUATION FOR THE CONTACTPRESSURE.
At this stage it is convenient to introduce the dimension-less variables
2
x=/,
b=S/,
p(x)_ 2(1-v )(3.2.1 ) Then (3.1
.L4) can
be writteni 8. co =
i_
rk(X)
P()
where k(x)= NÇ1 ,b A cIA. 00The integral equation (3.2.2) is of the Wiener-Hof type.
We first investigate whether the Fourier transform K(A) of
the kernel k(x) possesses a strip of regularity in the con)ex A-plane. From (3.2.3) we obtain
j NÇ1 .ò.X
K(A)
= X D(1 ,b,X
Using known expansions of the modified Bessel-functions, we
find and i b2(i+v)
+ (1-v)
-2(1-v2)
(b2-1)
K'(0)
= o
I«X) =+ o(A2)
as Re X'±co
(3.2.2) (3.2.3) (3.2.L) (3.2.5) (3.2.6) (3.2.7)The function K(X) has no poles on the real axis of the A -plane, because otherwise a cosine loading of the innerwall of the tube would cause infinite displacements. A point of singularity which can be expected, however, in view of the
branch points of K (A) and K (X) Is the origin A = O. This,
however, Is not thS case. Coisider for instance a typical
part of the function N(1 ,b,A),
K1(X)10(Ab) + K0(Ab)11(A). (3.2.8)
Using here a representation of K0(A) and K (X) in the
neighbourhood of the origin (ref. 11), we 'ind that the
logarithmic singularities of K (X) and K1 (A) cancel each
other. Hence K(A) possesses a Strip of règularity around
the real axis. This means that the kernel k(x) decreases exponentially for x -. ± co. Because the lefthand side of
(3,2.2) is a constant for
X >
O we take Im A > O and sufficiently small when we apply a Fourier-transformation to this equation. Then we obtain the Hilbert problemP(A) K(X) - O(A)
- rj
Avalid on a line L just above the real
axis
in the A-plane.This equation is of the form (2.3.5) and will be treated
by
the sanie procedure. First of all we have to find the function
K*(A). Because k(x) is an even function, it is clear that
K(A) is also even. We now consider in correspondence to ref.L. T (A2)
2
1
2 (3.2.10)
(A +s )2 T2(A )
where s is a rea constant and T (A2) and T2(A2) are
polynomials in A
with the same cerm
of the highest degree.From this we see that y*(A)_
TT
+ 0(IA!3) as Re A _+±co, (3.2.11)where fIni AJ < s. Hence by (3.2.7) the leading
term of
theasymptotic expansions of K(A) and K*(?) is the
srne.
Next wechoose s and the coefficients of T1 (A )
and
T2(A ) so thatIM(A)-i I
= ¡K(A)/K*(A) -1 I < e < 1, (3.2.12)where e is some prescribed positive quantity. That this is
possible follows from the theorem of Tschebyscheff
on
theapproximation of
continuous functions by rational functions (ref. 12, pages 55 and 65). Then we have to solve inaccordance to para. 2.3.
x(X) ií(A) - x_(A)
'
Krn(A). (3.2.13)A
The homogeneous part of (3.2.13) yields as a solution which satisfies condition (2.3.1 )
W(A)=
exp - d77, A in S.(3.2.1L)
Herewith the inhomogeneous equation has the solution
K*(77)
x(A)=
()-A)
di7, Am S.
(3.2.15) and the solution of (3.2.9) becomesP(A)
=x(x)/K(x),
(A) = x(A)/K*(A). (3.2.16)20.
?tg.3.2.I. Contour 01 integration.
indicated in fig. 3.2.1 . Because K*(77) and ¶_(77) are analytic in S_ we find with (3.2.16) Km(0) (X) X
-
+ +¶(0)
A . hence Ka(o)w(X)
p(x) -
(0)
A i(A)
(.2.1
8)Because of the occurrence of W+(A) in the integrand, this
integral is too involved for numerical calculations. It can
however, be expected (ref. 14) that, if e is sufficiently
small, we obtain a good approximation p*(x) of p(x) when we
put w(A) and
¶_(A)
equal to unity for all A, this yields-iAx
p*(x) =..i± K*(o)
f
e dA2ii
-A Even when we choose more specii'ally
K*(0) = K(0) we can easily show
p(x) - p*(x) 0(x°),
x -
O and p(x) - p*(x) - oX -'
00. Consicerp(x)_p*(x)
= 1 Ka(0) r(x)
2i -XK(X)
Ew(o)
1}d7t. (3.2.19) (3.2.20)(3.2.21
(3.2.22)(3.2.23)
Then it can be shown by estimating (5.2.1L) that the function
between brackets
under
the integral sign in (3.2.23) isabsolutely integrable. This implies (3.2.21) ana (3.2.22).
5.3
N1JhERICJL JJALCULATIOI OF THs:-INK-FIT
STRESSESAs has been suggested in para. 3.2. we shall use instead
of the exact function
K(A) (3.2.1)an approximate function
y*(A)_(A2+s2)
b0+b2X2hX.+b6A6
where the coefficients are given in table I for several values of the dirrensionless outer diameter b. The values or
s are assumed, while the values of an and b are computed
TAJ3LE I, coefficients of K'(X)
and.
by collocation. The relative deviation of K*(A) from K(7t)
is
less than L% for the whole real axis. By determining the roots of the polynomials in (3.5.1) we can easily obtain the factorization (2.3.8) of K*(7). Using these functions in the
integral (5.2.19) we find the values of table II for the approximate shrirJ-f it stresses.
TABLE II, p(x)/5.
From (3.2.19) we
obtain
by (3.2.2) and (3.2.22)p(x),
p*(x)
= -2(i-)
(b2-1) - x b2(1+v) +(i-(3.3.1)
+const.asx-O, (3.3.2) b s 2 a a6 b0 b2 b 66 3 1.25 1.00 -0.725 0.290 0.00553 1.00 -0.756 0.3314 0.00553 2 5.00 5.11 2.32 0.332 0 1.00 0.3142 0.332 0 1.5 6.00 9.12 0.2144 0.0265 0 1.00 0.0178 0.0285 0 1.2 10.0 30.9 0.0963 0.00303 0 1.00 0.00685 0.00303 0 b X0.03125 0.0625 0.125 0.1675 0.25 0.375 0.5 1 3 2.83 2.19 1.75 1.1i2 1.30 .25 2 2.140 1.80 1.39 1.01 0.921 0.9814 1.5 1.77 1.18 0.7143 0.538 0.619 0.658 1.2 1.1i1 0.716 0.2142 0.133 0.1149 0.253 0.308 0.3214 0.321422.
,->i 2 , 2
p(x), p*(x)
1V i
b -lj as x - co+ (ivfl
Formula (3.3.3) agrees with the result which we obtain from the elementary theory when an infinite tube is shrunk on to an infinite shaft of uniform diameter and contact occurs for their whole length.
.L. THE LIMITING CASE OF IINITELY THICK TUBE
We now consider the limit as b oo. We shall again have
to determine Green's function u0(x) of the inner wall. In
this case we shall have to take the unknown functions B (X) and B9(X) in (3..i5) equal to zero in order to avoid
singuíarity at infinity. This problem was discussed in ref. 13 so that we may write down at once the integral equation with the dimensionless variables (3.2.1)
00 I
k(x-)
p() d,
(3.14.1) where +0G -jAx 2(f7) d?.. ek(x) i
jt
X2K(IX) +[X2+2(1-v)
Ic(IAI)
c0 K(X)KA)
-A2K( ¡Xl) + [x2+2(1-v)}K(IAi) (3.L.2)Here a complication arises. The kernel k(x), for finite b, possesses a strip of regularity around the real axis in the X plane. This is no longer true in the limit as b -. c The function
(3.14.3)
is not analytic.
Proni the point of view of mechanics, however, it is not necessary to consider these difficulties in detail. For it is evident that the shrink fit stresses of an infinitely
thick tube are the limit of the stresses produced by tubes
of increasing thickness. Then, from continuity consideratis we can calculate the approximate stresses of the infinitely
thick tube by approximating the function KÇX) in (3.14.3).
Also the asymptotic relations (3.3.2) and (3.3.3) remain
valid.
Now the course of the calculation is the same as before; the coefficients of (3.3.1) are
s a0 a2 a a6 b0 b2 b b5
the error in this case being less than 1 .5%. We find for the stresses
x = 0.0625 0.125 0.25 0.5 1
p*(x)/
= 3.13
2.L41.97
1.73
1.58 1.50.3.5
DISCUSSION OF THE RESULTSThe contact pressures are plotted in fig. 3.5.1, which
shows p*(x)/ for various thickness ratios b/a. The ratios
chosen are b/a = 1
.2;
1 .5;2; 3 and.
. The stresses tend to iníinity as x + 0, the order of the singularity beingindependent of the thickness of the tube, as is clear from
(3.3.2). The interesting region for the variable x, inwhich
the stresses change rapidly, is approximately O x 1.
Here we see from fig.
3.5.1
that for b/a =2, 3
and c theshrink-fit pressure is a monotonically decreasing function
for increasing values of x ( 1. However, for b/a = 1
.2
and1
.5
there exists a minimum which becomes negative forsuf-ficiently thin tubes. Since negative shrink-fit stresses cannot exist, this means that our theory ceases to be
applicable. This behaviour is to be expected from the thecxy of beams on an elastic foundation.
16
.2
08
0,4
Flg.3.5.1. The contact pressux'e p(x)/8 for various thickness ratios b/a.
23.
2L1.
Finally we want to make a remark on the accuracy of the
numerical results given here, On the basis of observations made in ref. L the relative deviation of the approximate values p(x) fran the exact values p(x) ought to be
approxi-mately one half of the corresponding relative deviation of
K*(X) from K(A). For this reason the calculated shrink - fit
Chapter L
THE HOMOGENEOUS FIRST ORDER
INTEGRO-DIFFERENT IAL EQUATION
OF THE WIENER-HOPP TYPE.
We shall discuss an integro-differential equation with first order derivatives of the unknown function, which occur
both under and outside the integral sign. Equations of this
kind sometimes arise in physics (ref. 5). In this chapter the method of solution is exposed for this simple form,it is an extension of methods described in chapter 2. A solution can be obtained even when the kernel k(x) decreases
algebraicly for lxi -+co while at the same time the solution
does not tend to zero for x - + oo.
The following cases are treated separately:
Case 1. The kernel k(x) decreases exponentially ik(x)l<e with a > O, for lxi - co and the solution has the asymptotic1
behaviour lf(x)i =Û(eC),
with
y < a for x -. + coCase 2. The kernel decreases algebraicly k(x) =ixl n > i for lxi - co.
Whenever we need explicite information about the
singularity of the kernel
at
x O we shall assume that k(x)is an even function with respect to x (para. L.2.).
L.i
THE GEIthiRAL EQUATION, CASE iOur integro-differential equation has the following form co
a0 f(x)a1 fT(x)
=01
f()+b1 f'()
k(x-)de. (L.1.i)When we have found a function f(x) for x > O which satisfies this equation we have to solve a simple first order differen-tial equation in order to find f(x) for x < O. From the assumption of the exponential behaviour of k(x) at infinity
we find
la0 f(x)-fa1 f'(x)i < C x - - co. (LL.i .2)
Hence if we consider the left hand side of
(1.í .1) as
one
function h(x) we can apply a Fourier-transformation to this
26.
A remains
on a line L with Im A = , within the atri;ß,
y
< Im X < a.Applying a Fourier-transformation to (L..1) we obtain
H_(A)+F(X)a0-ia1A - 'T K(A)(b0-ib1A)}= 4a1-b1 K(Afl
(L.i
.3)where
H_(A) =
f
f(x)+aA f?(x)}eAX (L4..1 .L)00
First we consider the related homogeneous equation
y(X)+Y(X)
a-ja1X - K(X)(b-ib1
X)} = O.(L,.1.5)
This means that we seek solutions of (L..i.i) with
f(0) = O.We assume further that L has been chosen in such a way that
there lies no zero of the factor of
Y+(A)on L.
Analogous to the
procedure in para 2.2,we consider
the
expression(X-a)
in G(X)=ln {.-4a0-ia1A-
K(A)(b_ib1X)
(L1..i .6)
where a is in and b in S. The integer n can be
determined in sich a way that the principal value of in G(X) for ReX-p on L tends to zero.
We then
find,
for the solution of(L.i
.5)
Y+(A) =
L (A-b)'
(A) P(A),(L.i
.7)a1 +
Y_(A)
= (A-a)
A) P(A), (L.i.8)where
-i
rlflG)
exp
-
dï7, A in Sand P(X) is
anarbitrary
polynomial. Again we consider the zero's, A = A (m=1....
q) of- 'T
K(A)(b0-ib1A)
trip
ß and above the line L. Then also theY(A)
y(A)
q - (.1.11)(1.i.12) JI (A_Am) q Ji (A_Am) m=1 mxl within the sfunctions
(A) -(LL.l .9) (L..'!.i o)are solutions of (L..i.5), but now on a line L
with
Im X <Im A
< ß.
In the case (n+q) I we find a soiBtion of q(Lj.î .5) which can be interpreted as a Fourier-transform,while
an arbitrary polynomial
P(A) of degree (n+q-l) can be takenas a factor.
Now we return to the irthomogeneous equation (Lp.i .3) and
try to obtain a solution. The general solution can then be
found by adding solutions of the homogeneous equation. We
choose some solution Y(X) and
Y_(X),
assuming P(X) -i in(L.i
.7)
and (L4..i .8) in order to avoid zero's of Y(A) inS(A), from the set provided by (L.1.l1) and (L..1:12) and
wfite instead of (L-.i.3)
H_(X) (x)
f(o) a1-b.1 2n K(Afl
X - -
e (x)-.e
(A).(L..I.I3)
Y_( )
Y(A)
Y_(A) +Here we have introduced the functions e (A) and
e_(x)
whichare discussed in para. 1 .2 and which reult from splitting
the right hand side of (L-.1.13) into two parts regular in 5 and S respectively. They are determined within an additive
term consisting of an arbitrary polynomial
and
can easilybe
calculated without recourse to the general integral
representation (1.2.8),
in terms ofY(X) and
Y_CA). By definition we haveY(A)+Y(X) k0-ia1 A - i.f2 K(A)(b -ib1 A ) = O (Li..i .1L4.)
hence
K(A) (a-ia1 A)
Y(A)
=(b0-ib1A)Y(A)
+ Y_(A)}
(.i.i5)
Combining (L4..i.i3) andib
oin b, we find
I
(L..i
.1 5) and assumingfor instance
since Y(X) have no zero's in Sb1
y(-i
b0(.i.i6)
e (A)-
i(o)
+(b0-ib1A)
Y()
+Y(-i
) +Q(A) ,
(.1.17)
ib
where Q(A) is an arbitrary polynomial. The case
-
b°
in S_can be treated entirely analogous. The solution of (L.i.13)
andhence of
(L.i .3)becomes
r(o)
(a1b0-a0b1)28. f(o) a1b0-a0b1- b1 Q (A)Y_ (A) *f(b0-ib1X) (4.-i .1 8) b Y (X)
F(X)= -(A)y(A)=-
f + - Q(A)y(A). [r(b -ibi) o 1 + b1 (L.i.19)In this result we may take Q(X) O because this term only
furnishes a solution of the homogeneous equation. Aparent1 when Y_(X) is bounded at infinity we may interprete (1.i .1 8 and (L-.1.19) as Fourier-transforms.
L.2 CASE I WITH a0 # O, a1 = O ASD b1 * O
Equation (L..1.3) now reads
H(X)+F(X)1a0-T
K(A)(b0-ib fLO) b1 K(X) (L.2.1) and the homogeneous eauationY(A)+Y(A).a0-f
K(A)(b0-ib1A)= Y_(A)+Y(A)T(X)=O,(L.2.2)where we have introduced the abbreviation T(A). There is a
difference between the equations and (Li.2.2). In
(L..i .5) the factor of F (A) is of the order A when
Re A -' ± while in (14..2) the asymptotic behaviour of this function, T(A), depends on the behaviour of x(X) for
Re A -+ ± . We now remind the assumption that in such cases
we shall consider only kernels k(x) which are even functions of x, hence also K(A) is even in A.
The asymptotic behaviour of K(X) for Re X -. ± depends
on the behaviour of k(x) in the neighbourhood of X = O. We
assume k(x) to be continuously differentiable for O < e1 < x for each e1. We consider several possibilities for the
behaviour of k(x) in the neighbourhood of X = O.
Let k(x) be continuously differentiable for O x <
e1 < then
um
K(A)=IAim
, i < m. (L1..2.3)Re A-±oo
Let k(x) be the sum of a logarithm rd a continuously
um
AIAL1
Re
A-+±
e) Let k(x) be the sum of x
(o <
6' < i) and a continuouslydifferentiable function for O X < e2 then
um
IÇ(X) BIAI_1 (L4..2.)Re À-*±
The formulae (L4.2.3) and (L.2.5) follow directly from ref.
1L. while (Li.2.L-) can be deduced by subtracting from k(x) the function C K (x) where K (x) is a modified Bessel-function and C some agitable consant, then we arrive again at case a.
The difficulties which can occur in the cases a, b and e are concentrated in finding appropriate functions aniogous to (Li..i.6). We shall now state these functions.
a) In this case the factor T(?) of YÇA) in (Li..2.2) tendB
to a0 for Re X -. ± , hence instead of (L.i .6) we can use
in G(X) = in T(X) a0
(X-b'
(i-6')
T(A)(A-a)
2 in G-(A)=lnE (i-*6')ib1B
(A)r+
2 (L.2 6)T(X) tends to a0 ± i A b1 for Re
A ,
± co we takea
j*[
Ab n- i in G(A)=1n T(X) X-a2i
a-i2'Ab
(-)
,(L.2.7) a0+i Ab1T(A) tends to ± i b. BlAt6', O < 6' < i for Re
A-±co
we take
(L.2.8)
The values of the multivalued functions which are the factors of T(A) are to be fixed in such a way that the principal
value of in G(X) for Re A - ± co on L tends to zero.
The essence of the fornilae above is that we have multi-piled T(X) by functions which can be factorised by inspection
into functions reguiar and without zero's in S
and S.
Wefind the following soiutions for the homogeneos equation
( L1.. 2.2)
a) Y (A) =
(L.2.9)
+ (X b) n
(x)
P(A)30.
b)
(a0+i Ab1)
(L.2.11)
i
b
2ai
in
(ai
Ab1)PCA) a
L.2.i2)
y_(A) = e) = n(i±6')
(A b 2 ¶(A) P(A), y_(A) a'where ¶(X) is defined in each case by the integral (Li..i
.9)
with the corresponding G(A).
Following the treatment of the preceeding paragraph below
form (14.1
.9)
we arrive again at formulae analogous to (L.i.i8)and
(14.1.19).
The case a =0, b =0, a0 # 0, b * O is the homogeneous
Wiener-Hopf euatioi with exponenially decreasing kernel and
is discussed already in chapter
2.
14.3
THE E1AI EQUATION CASE 2
The treatment given in para. 14.1 is also largely applicable
to the case of equations with kernels which are only
trans-formable on a line L. However, there are differences, for in
the latter case we cannot translate L and hence (14.1.11) and
(14.i.12)
cannot be used. Further if there is a zero of (14.i.10)on L this cannot be avoided by translating L slightly.
We now discuss a method to obtain solutions of (14.1 .3)when
(14.i.10)
has zero's on L. To demonstrate the procedure we shall treat the case of one zero A = p of the first order. An extension to several zero's and zero's of higher order does not offer principal difficulties.From
(14.i.11)
and (14.i.12) we see that a zero X of (14.1.10)iXmX
m
introduces a term of the form c, e In the solution of the integro-differential ecuation. In our case with a zero y on the line of transformation it seems reasonable that the
solution of the integral behaves as c9e for x -+ + cc. Hence
we try to find a constant A in order that the function -ipx
v(x) = f(x) - A e -. 0, x - + co
o
(L1..3.i)
The integro-differential equation for v(x) can be derived easily from (14.i.i), 'e find
av(x)+av'(x)=
Îbov()+biv'(
k(x-)d
-O cc(14.3.2)
-
A(a0-ia1v)e1 +A(b0-ib1v)f
e' k(x-)d
, X > 0. (L..2.i3)
(L1.2.1L)
are
where
W_(X) a1 (Ä-v)V(X)
-n
Y(X)
(X b\a'
Under the assumption that V is a zero from (L..1.1O) we see
that the last two terms of the right hand side of (L..3.2)
cancel each other for x - + Applying a
Fourier-transfor-mation to (L1.3.2) we obtain
(A)-i-V(A) a0-ia1X - K(X)(b0-ib1 X)} =
v(O ) Al
= + ía -b. '1 K(A) ía -ja v- K(X)(b -ib vfl,
i
*[(Xv)
O I O Iwhere
v(O)
is the value of v(x) when x tends to zero throipositive values and
w(X)
f
(3.)
The solutions y(X) of the homogeneous equation
a0_ia-if K(X) (b0-ib1 X)
Y_(A)+Y(A)
(A-v)
-
°(L.3 3)
(L.3.6)
a0-ia117.-*Tr K(Îi)(b0-ib1ri)
(l?_a)n
1
in ii
i(n-v)
(n-b)
4.3.7)
Herewith (L.3.3) can be written in the form
[y(0)
ía -b
K(X)}+
Ai
íaY(A)
iY(A)
-
YA)
Ir(Xv)
-ia1v - K()(b0-ib1 y )}] =
0(X) -
(L.3.8)
where we have introduced G(A) and
O(A)
which result fromthe splitting of the right hand side. Under the assumption
that
p = i b ¡b ,for instance, in S
we find analogous tothe procedure ii para. Li..1
ív(o)ò
(A-v)--Ai(ò-ib1v)}E1-Y(X)/Y(p)
i. P(X)y(X)
1 +
.T
(À-v)(b -ib X)
a1 (A-v)32.
(a0b-a1b0) Y_(X)
w (A) (A_v(ofl(b_ibX)
(b0-1b1X)Y(p)
-.v(0)ib1 a1(A-v)-A a1(b0-ib1vfl+P(X) Y_(X). (L..3.iO)
When n > O we can interprete (.3.9) and (L.3.1O) as
Fourier-transforms while we can admit an arbitrary polynomial of the
degree (n-1). For the ease n = O we can consider these
functions as Fourier transforms when y(0) a
(L.3.11)
'
yC)
From (Li.3.9) we observe that we have to choose A in such a
way that the singularity X = V disappears, this yields
-'T p(v) y (y) y (p)
A -
(Y(p)Y(t'))
(.3.1 2)Hence we have determined a solution of (L1.3.3) and by the inverse transformation and (L.3.1 ) also a solution of (L.i .1) with in general r(o) # O. An analogous treatment can be given
for p = -i b,,/b in S.
If we want t solve an equation with several and higher
order zero's of (L4..î.iO) the analysis becomes much more
com-plicated. Assuming zero's X= y (r=î ... N) of the order p
we shall have to consider the ntegro-differential ecivatio
f or the function
N
r
-ivx
v(x) = f(x) - E ( E Arm xm)e r (L.3.13)
r=i m=1
By an analogous reasoning as before we can obtain relations for the coefficients A
The discussion of pa. L.2 can also be extended in the
way of para. L..3 to kernels which decrease algebraicly at
infinity, this we shall not do here. However, to show the
applicability of the ideas of this paragraph we shall discuss an example.
EXAMPLE
We consider the ordinary Wiener-Hopf integral equation
which we have discussedto some extent in para. 2.L1.
=
r
r(E)
f(x) =
m)
i(x-f)2
'riowever, the range of the parameter m is different we take here m > 1/it. Formal transformation of (Li..L.i) yields
F(X) + F(A)
i-meX
= O.The function Eirnze possesses two zero's on the real
axis which we call A = ±v The asmptotical behaviour of the
I solution is then assumed to be
where
f(x) Ae-ivx + BeÌVX, 11m x
Hence we consider instead of (Li.L4..i) the equation for the unknown function i -ivx ivx v(x) = f(x) - Ae + Be ), which reads -ive iv
v(x)_mL
v(E)dE
(Ae+
iVX (e +Be )-
1+(x)2
- Be )+mj1(x)2
o
Transformation of this equation yields
W(A)V(X)E1-nirre7'=
E Ai BiEimire
.fj(A-v) '.[t(X+v) o 1vX iVX)W (A)=
f
tv(x)Ae +Be dx.We now write (L.6) in the form
w_(X)
v(A)
(A2-v2)= -
Ai
+Bi
Y(A) - Y(X)
(A-v) r(A+v)
(L1..L.8)
where Y(A) are the solutions of
IAl
(x)+
1-mae
= o,- (A2-v2)
Y(A)
=(A+i) w(A),
Y_(A)=(A-i1
w_(x) (..1o)(L.L..ii)and
(L..i4.3)
34.
± exp
-f
+
ln{(1-e
(2)
From (L.L.8) we obtain with the use of (L..L.9)
w_(A)
v(A)
(A2v2) ± A(A+v) +
B(A-v). yiX)13)
Y_(A) -
Y(A)
He nc e
v(A)
f(A2-v2)
ABP()
y(Afl,
(L.t.iL)
w_(A) =
p(A)
y(A).
(4.L.15)In connection with
(L.)4.i1.)
we see that we have to takeP(A) C. Further we have to compensate the singularities
A = ± y in
(!.L.iLi.).
This yieldsA
= -
-
y(v) ,
(L.L.i6)B =
Y(-v).
(L.L.17)Hence the solution of
(Lj.Lj.i)
is determined.We have given two different discussions of eQuation (L.L.i), depending on the values of ni. For rn < 1/t the equation was discussed in para. 2.L,
m > The case m = i/z cannot be solved directly, however
it is possible to give an asjmptotic expansion of A
(L.L.i6
and B (L..L.17) for small values of g-'.
d?7, Ain
2)
Chap ter
5AWIENER-HOPF TYPE
INTEGRO-DIFFERENTIAL EQUATION WITH
FOURTH ORDER DERIVATIVES
We shall discuss an integro-differential equation, which
is an extension of an equation discussed by Reuter and
Sonciheirir (ref.
6)
which arises in considerations about the anomalous skixua effect (ref.5).
Although the derivativesoccurring in this equation are of the second and fourth oida' we can follow closely the line of thought of para.Li-.1 and para 11.2. It was asked to expand the quotient of the
solution and its first derivative for the value zero of the
independent v4riable in terms of a small parameter /3. The
case that this parameter is zero is discussed by Reuter and Sondheimer. It turns out that this quotient does not depend analytically on jS, there arises a term of the form /3 in /3.
5.1 !OURIER TRANSFORMATION OF THE EQUATION
The integral equation has the form
(2)
(Li.)f (x)=ia
j
f(t)-iß f (tflk(x-t)dt (5.1.1)
o (2) (Li.)
where a and /3 are positive real numbers. By f (x) and f (x
are denoted the second and fourth order derivatives of f(x The kernel k(x) reads
k(x)
=
f (
- -)
exp (-clxls)ds,
I
where o is in general a cplex number with a positive real part. We shall determine the physically important quantity
f(0)/f(0) ,
and obtain its asymptotical expansion for /3 - O.
(2)
(L1.)It will be assumed that f(x), f (x) and f (x)are bounded
(2)
for x -
+ .
From(5.1
.1) we deduce that f (x) is of the(5.1 .2)
36.
order e when X -* Then it is allowed to apply a
Fourier-transformation to (5.1 .1) when A remains on a line L above
and
sufficiently close to the real
axis
in the complex A plane.Wefind
H_(A)-F(A)tA2+ia'T K(A)(i-iß A =
(i)
(3)
(2)
=
Ef (o)-f(ofl+
K(A)-f (o)+f
(o)+A2f)_iA3f(o)
(5.1.L)
whee
°
(2)
H(A)=
.._ f
f(x) e1
dx,
-
O0 and--f
k(x)e" dx= {--- i(
+£)
in(_fl,
c
(5.1 .6)the logarithm is defined by
in (_j) - -ai,
Re A-+ co A on L. (5.1.7)First we consider the homogeneous part of (5.1 .)
Y (A)-Y (A)A2+ia*T
K(A)(1-ißA)
=Y_(A).-Y(X) T(A)=0,
-+
(5.1.8)
where, as in para. L.2, we have introduced the abbreviation
T(A). In the following we shall use the notations of that
paragraph. The asymptotic behaviour of K(A) is
L.
a A - +
-
-,Re A-+ 00,A on L,n=i n A A3 3X
hence
T(A)czßxA2IAI, ReA-*±c»onL.
The function in O(A) can be taken in the form
A2+ia K(A)(i-ij3 A) ln O(A) = ln
a132r(A2+p2) (2±2)
(5.1 .9)
(5.i.10)
(s.i.ii)
where p is an arbitrary real positive number. We find for
where
in G(ri)
= exp - Çi7-Xj d77.
+ Y_(-j)ÇA+.t)
Y_(-i)(A+1)
(A' -R (- -i i (-i u) + R - (ii')
+
-
S+
t Y_(A+
Y (-i)(X+i) y)2(A)
i R(i)
+ 2h
( (A) y (A)-. + , Y_(A)=W_(A)(X-ip)1i', (5.1.12) + c3i(X+ip)*IX+ip (5.1 .13) We now write (5.1 .L-)H_(A)
F(A)
(i)
(3)
Y(A)
Y(A) =
y(7)[
f (0)-iXf(0) +aß K(A)-f (0) +(2)
2
f) - j
f(0)fl (A)- e(A).(5.1.1)
+ iX ±
(o) +
AThe functions
e
(A) and G (A) can be calculated exactly inthe same way as1n para.
I.i
by splitting K(A)/Y_(A) withthe aid of (5.1.8). We find.
(3)
(2) (i)-x2j.r
(0)+ix3ßr (o)+ ir
(o)+Ar(ofl +
A -ii)
Y(A)
-R1 (-u) +1 R1 (-ip) + R2(1)i R(ij4
+ y(i)2(A-i)
(s.i .15)
1 (3) (2) 2-r
(0)+i(o)+A f)-iA r(o)}
+(5.1 .16)
where
()
(2) (i)38.
(3)
(2) 2R2(X)= - f
(o)iAf (o)
A f (0)-i xr(o)
(5.I.18)iit
and j = e is a zero of the function (1-ißA). The
solution, (5.1.15) and (5.1.16) are essentially the same as (4.i.18)
and
(4.1.19). Because wewa\to
obtain a functionf(x) whose fourth order derivative f") possesses a one
sided Fourier-transform for O x
()
(2)i iAx
c= -f
(o)+iA
f (o)+A2f)-1Af(0)A)
-If
(x)eo
(5.1.I) it follows that the second
(5.1.16) must tend to zero Re A - ± co In te general
to zero as IAI /2 Hence
following conditions
-R1 (-12)i
R1(-1/2)
R2(/2)i R(ií1)
Y(12)
Y(-i)
+ 2 + Y(j/2)122-R1 ()
R1(-1/2)
R2(bL)R2(i/1)
- 0.
Y_(-ii) +Y(/2)ìi2
Y(1/2)/22When (5.1.20) and (5.1.21) are satisfied, the second term of
the right hand side of (5.1.16) tends to zero asRr /2 for
ReX-i ± . ¶ÇJs, however, generates a singularity of the form
x for f'x) as x -+ 0+, which cannot be tolerated for
physical reasons, hence we have to go one step further and
demand
-R1(-/2) i R1 (-1/2) R2(/2) i R(iji)
=
0.
Y_(-i/2)
Y(/2)/22
Y(i/2)/22
These three linear equations, in the four unknowns f(0)
(i) (2)
(3)
f (o), f (0) and f (o), determine the ratio
f(0) 122 Y(i12)-i y(/2fl +122Y_(-i12)i y_(-12)
12
( )
-. (5-.1-.23)
f
(o) 12(j/2) 'Ç(12fl
12term of the right haTld side of
more quickly than A4 for
form in which it stands it
tends we shall have to satisfy the00
where we take instead of L the real axis in the A plane
as
line of integration.
5.2
THE ASYMPTOTICAL EXPÂiSION OF E(A)Instead of (5.1 .2L1.) we can write
i 172+ia
K(ii)(1-iTh)
l._I_n
a ß Zi-)
¿
(rj2-A2)
39.
From (5.1.12), (5.1.13) and (5.1.23) we see that the
asymp-totical expansion for ¡3 -+ 0 of r(0)/f (o) can be calculated
directly from the asymptotical expansion of
+00 *[
K(ii)(1-iß
E(A)=f
aJ3Jr(772+iD2) (i2+p2)
w,
(5.1.2Li.)
J
(ii-A)
(5.2.1)
where we have taken for p the value zero. This limit process
must also be made in (5.1.12). We now consider
(
ue
i7\_
1-e
i
u17{-ln(a7zli)-ln
277+ln1+ja
22i
x(ii)?72(1iuri4fl1
77 U e
(5.2.2)
where u
¡34
and
assumes the values
, ,or
and
it is asked to determine E(u e1) for large values of u.
We shall treat the three terms of the integrand of(5.2.2)
separately. First we consider
-2 e1u
ln(ai)f
22i
=ilnanu, uein S,
17 -U e
(5.2.3)
next
00 +00 0 d77in
d77--euI
ln ii
-2 euf
ii2_u2e2
-2-u2e21
di7f
2u2e2
O 00
The path of integration in the first integrai on the
right-hand side of (5.2.Li.) is the real axis with a small semicircle
above the point 17 = O. We find
-2
euf
2 =(±- )in in u,
ue
in S.
(5.2.5)
-Lo.
The last term in the integrand of (5.2.2) causes more troitie. We divide the interval of integration as follows:
JùlH
+ir
K(r,)772(1 -iu%77
=
f
...
d27 +f
...
dJ7 +f
...
d37 = I + II + III,(5.2.6)
where s = + ¿ and r = 3 + 6', . and 6' being arbitrary but
su'ficient1y small real quantities. We shall treat the integrals I, II and III separately
us - r
-772_U2e2h -us =_u_2e_2J [1n1+iafK(77)il2
+ o 2 2a1uK(r)
(IÌ_ e_2)nd77.(5.2.7)
+ ln 1±
nO u2
The expansion of the denominator is possible on account of
the special range of integration. The first logarithm of
the integrand in (5.2.7) multiplied by the general term of
the expansion gives
e -2(n+1)ìp U e 2(n+1)f
772n ln E1+iaJK(77)77 2d77. (5.2.8) oFor a further reduction of this form we need the expansion of the logarithm for 77 - +
a;
using (5.1 .9) we obtainin 1+ia'[K(77)7721
bi7', t
real (5.2.9)with b = jan ; b = iaL.c b5 = lanc2; Hence we
can wrte instead of (5.2.8S
_e2+1
f
772n Ein(i+jx()2)
b21
u
ni)
m=3(i+i)
us 2n +f
2n bm77_m + 772n(771)Li.
Go 2nb21
f
2n in(1+ia'[ K(17)172)- Z bml7_m-
17 m=3 172n(171) 5.2.10) uswhere we have supposed n 2. For n = O or n = I we have to
change (5.2.10) slightly, because then we need fewer conver-ence-producing ternis or none at all, see (5.2.13) and
(5.2.1L1.). The first integral is independent of u. After inte-gration we find for the second integral terms of the form
s s
i
I,
i i 2n - 2 and ln(77-fl) (5.2.11)
or
u
and
in (u5+1 ) = s in u - (-i )m (5.2.12)m=1
Because the final result
must
be independent of s, we canomit these terms from the beginning. If now we expand the
logarithm in the third integral of (5.2.10) and integrate,we
see that only terms of the type of (5.2.12) arise, hence we
can also neglect this integral. It willturn out that for our
purpose t is sufficient to consider quantities up to the
order u; then we obtain from (5.2.10) for the values n O
I and 2 Go A,
n=0,
-e2u2
f
in 1+ia K(17)2dî7 =-e2u2
(5.2.13) o 00-eu
f
ln1 1
2ivK(77)2-
=-e u B, n=1 (5.2.1L1)-e6u6 J [in
1+iaK(17) 2ia77+ic
--6ip -6
=-eu C,n=2.
2 1cx7Vc d -'7+1 (5.2.15)The second logarithm in (5.2.7) gives, with the generai term of the exransion of the denominator