Professional Documents
Culture Documents
Third Edition
Ruel V. Churchill
Professor Emeritusof Mathematics
Universityof Michigan
I
Operational Mathematics
This book was set in Times Roman, and was printed and bound
by the Kingsport Press. The designer was Jo Jones; the
drawings were done by John Cordes, J. & R. Technical Services, Ine.
The editors were Howard S. Aksen and Madelaine Eiehberg. Matt
Martino supervised produetion.
Contents
Preface xi
Bibliography 456
Appendixes
AppendixA Tables of LaplaceTransforms 458
x CONTENTS
Index 477
Preface
1 INTRODUCTION
The operation of differentiating functions is a transformation from functions
F(t) to functions F'(t). If the operator is represented by the letter D, the
transformation can be written
D{F(t)} = F'(t).
The function F'(t) is the image, or the transform, of F(t) under the transforma-
tion; the function 3t2, for example, is the image of the function t3.
Another transformation of functions that is prominent in calculus is
that of integration,
In each of the above examples inverse images exist; that is, when the
image is given, a function F(t) exists which has that image.
A transformation T{F(t)} is linear if for every pair of functions Fl(t)
and F2(t) and for each pair of constants C 1 and C2, it satisfies the relation
(1)
Thus the transform of a linear combination of two functions is the same
linear combination of the transforms of those functions, if the transformation
is linear. Note the special casesof Eq. (1) when C2 = Oand when C 1 = C2 =
1. The examples cited above represent linear transformations.
The class of functions to which a given transformation applies must
generally be limited to some extent. The transformation D{F(t)} applies to
all differentiable functions, and the transformation 1{F(t)} to all integrable
functions.
Linear integral transformations of functions F(t) defined on a finite
or infinite interval a < t < b are particularly useful in solving problems in
linear differential equations. Let K(t,s) denote some prescribed function of
the variable t and a parameter s. A general linear integral transformation of
functions F(t) with respect to the kemel K(t,s) is represented by the equation
Ir a function F(t), defined for all positive values of the variable t, is multiplied
by e-sI and integrated with respect to t from zero to infinity, a new function
f(s) of the parameter s is obtained; that is,
{O e-SIF(t) dt = f(s).
As indicated in the preceding section, this operation on a function F(t) is
called the Laplaee transformation of F(t). It will be abbreviated here by the
symbol L{F}, or by L{F(t)}; thus
object function by a capital letter and its transform by the same letter in
lowercase. But other notations that distinguish between functions and their
transforms are sometimes preferable; for example,
tfJ(s) = L{ f(t)} or .9(s) = L{y(t)}.
For the present, the variable s is assumed to be real. Later on, we shall
let it assume complex values. Limitations on the character of the function
F(t) and on the range of the variable s will be discussed soon.
Let us note the transforms of a few functions. First, if F(t) = 1 when
t > O, then
1 <X>
];
<X>
L{F} = f
o
e-sI dt = --e-st
s o
1
hence, when s > O, L{I} = -.
s
Ir F(t) = elctwhen t > O,where k is a constant, then
1 <X>
]
<X>
k s
L{ sin kt} = S2 + . '), L{cos kt} = S2 + ., ')
when s > O; but soon we shall have still simpler ways to obtain those trans-
forms.
It follows from elementary properties of integrals that the Laplace
transformation is linear in the sense defined by Eq. (1), Seco1. We can illus-
trate the use of this property by writing
PROBLEMS
1. Use the linearity property and known transfonns to obtain these transfonnations,
where a, b, and e are constants:
a + bs
(s > O);
(e) L{asint + bcost} = S2 + 1
s
(s> lel);
(d) L{cosh et} = S2 - e2
3. Show that the linearity property (1),Seco1, can be extended to linear combinations
of three or more functions when all the transfonns existo
4. If for all functions of some class and for every constant C a transfonnation T
satisfies the two conditions
Thus the values of such a function may take at most a finite number of
finite jumps on the interval (a,b). The class of sectionally continuous functions
includes functions that are continuous on the closed interval a ~ t ~ b. The
integral of every function of this class, over the interval (a,b), exists; it is the
sum of the integrals of the continuous functions over the subintervals.
The unit step function
Sit) =O when O < t < k,
=1 when t > k,
f o
Sk(t)e-SI dt =
k
e-sI dt = --e-sI;J
S k
o (k,O) (T,O)
Fig. 1
THE LAPLACE TRANSFORMATION [SECo 4 7
{T e-S'F(t) dt.
For example, when F(t) = Ct, its transform can be written, after the
substitution of x for fit, in the form
'" 2 '"
(1)
fo
t-te-s, dt = -
JS o l e-x2 dx (8 > O).
The last integral has the value fi/2 (Prob. 10, Seco5); hence
(8 > O).
4 TRANSFORMS OF DERIVATIVES
By a formal integration by parts wehave
Let F(t) be of order of elZlas t approaches infinity. Then whenever s > oc,the
-
bracketed term becomes F(O),and it follows that
(1) L{ F'(t)} = sf(s) - F(O),
where f(s) = L{ F(t)}.
Therefore in our correspondence between functions differentiation of
the object function corresponds to the multiplication ofthe result function by
its variable s and the addition of the constant - F(O).Formula (1)thus gives
the fundamental operational property of the Laplace transformation, the
property that makes it possible to replace the operation of differentiation by
a simple algebraic operation on the transformo
As noted above, formula (1) was obtained only in a formal, ormanipula-
tive, manIler. It is not even correct when F(t) has discontinuities. The
following theorem will show to what extent we can rely on our formula.
L{F'(t)} = lim
T oo o f T e-SIF'(t) dt,
ifthis limit exists. We write the integral here as the sum ofintegrals in each of
which the integrand is continuous. For any given T, let ti' t2," ., tn denote
those values of t between t = O and t = T for which F'(t) is discontinuous
(Fig. 2). Then
T ll l2 T
f o
e-SIF'(t)dt =
fo
e-stF'(t)dt +
f
I1
e-SIF'(t)dt +oo. +
fIn
e-SIF'(t)dt.
.F1t)
I
I
I
I
I
I
I I
01 (tl,O) (t2,OI (T,O)
Fig.2
THE LAPLACE TRANSFORMATION [SECo 4 9
After integrating each of these integrals by parts, we can write their sum as
tl t2 T T
e-stF(t)
]o
+ e-stF(t)
] tI
+ ... + e-stF(t)
]tn
+s
f
o
e-StF(t)dt.
(s > O).
(a > O).
PROBLEMS
1. State why each of the functions
- 1 when O< t < 2
(a) F(t) = te2t and (b) G(t) = t+ 1
{1 when t > 2,
is sectionally continuous on every interval.O < t < T and (!)(e"t)as t -+ 00, where a > 2
for F and a ~ Ofor G.
2. State why neither of the functions (a)(t - 1)-1 or (b)tan t is sectionally continuous
on the intervalO < t < 3.
3. Ir Sk(t)is the unit step function (Sec. 3),draw graphs of the following step functions
-
and find their transforms when s> O: (a) F(t) = 1 Sl(t); (b) G(t) = Sl(t) - S2(t).
AIso, note that both F and G vanish except on bounded intervals; thus their Laplace
integrals become definite integrals that exist for all s ( - 00 < s < 00). Show that
/(0) = 1 and g(O) = 1,and that
(a)les) = (1 - e-")/s (s:;' O);
-
(b) g(s)= (e-S e-2s)/s (s:;' O).
4. Obtain these transforms with the aid of Theorem 2:
(a) L{coskt} = S/(S2 + p) (s> O);
(b) L{sinh kt} = k/(S2 - k2) (s> IkJ).
5. Given the transform of e"', use Theorem 1 to show that
1
(s > k).
L{te'"} = (s - k)2
THE LAPLACE TRANSFORMATION [SECo 5 13
-
6. If G(t)= OwhenO;;; t ;;;k and G(t) = t k whent ;;;k, drawgraphs of Gand G'.
Giventhe transformof Sit), (a)apply Theorem 1to prove that g(s)= s-2e-ks (s> O);
(b) show that G(t) = f~Slr) d! and find g(s) from formula (4), Seco5.
7. Prove that the function F(t) = sin (eI2)is of exponential order (IX ;;;O)and that its
derivative F' is not of exponential order. Show that Theorem 1 ensures the existence
of the Laplacetransformof F' when s > O,in this case whereF' is not of exponential
order.
8. (a) Derive Eq. (4),Seco4. (b)Illustrate that equation by using it to find the transform
of the unit step function Sk(t).
9. (a) If a function F and its derivative F' are both (J)(e"')and continuous when t ;;;O
except possibly for finite jumps at a point to, and if F" is sectionally continuous on each
intervalO < t < T, apply Eq. (4), Seco4, to derive the formula
(r >0).
r(r) = t'" x,-Ie-x dx
(a) Integrate by parts to show that the function has the factorial property
f(r + 1) = rf(r).
(b) Show that f(1) = 1,and hencethat f(n + 1)= n! whenn = 1,2,....
(e) From the value of the integral J found in Probo 10, show that r(t) = .;:;r.
Then use the factorial property to find nt)
and formula (2)to show that L{Jt} = t.Jir/si.
14. Improper integrals Consideronly functions that are sectionally continuous on
each intervalO < t < T.
14 SEC.6] OPERATIONAL MATHEMATICS
(a) If O ;;;g(t) ;;;h(t) whenever t > O and if S~ h(t) dt exists, use the definition of
the improper integral
Let the symbol L - 1 {f(s)} denote a function whose Laplace transform is f(s).
Thus if
L{ F(t)} = f(s),
then F(t) = L -l{f(S)}.
Using two of the transforms obtained in the foregoing sections, we can write,
for instance,
k
L -1 ~ = ekl, L -1 = sinkt:
{s - k } { S2 + k2 }
This correspondence between functions f(s) and F(t) is called the inverse
Laplace transformation, F(t) being the inverse transform of f(s).
In the strict sense of the concept of uniqueness of functions, the inverse
Laplace transform is not unique. The function F1(t) = ekt is an inverse
transform of 1/(s - k); but another, for instance, is the function (Fig. 3)
F2(t) = ekt when O < t < 2, or t > 2,
= 1 when t = 2.
For the transform of Fz{t) is
FZ(tI!
../
I
f(2,1)
I
I
I I
O
Fig.3
also, for each individual function F of the set, let constants M and IXexist
such that IF(t)1< M ea.twhen t > O. As we have seen, F then has a transform
f(s) defined on some half line s > IX.
The theorem states that no two functions of class tff can have the same
transforms. Thus if f(s) is the transform of some function F(t) of c1ass tff,
then F(t) is the unique inverse transform L -1 {J(s)} in tff.
For'example, the only function L -1{e-S/s} ofc1ass rf is the unit step
function S 1 (t) defined in Seco3 if S 1(1) is defined to be t. As another example,
-
the only function L - 1 {1/(s k)} of c1ass tffis ekt.
. It is well to note here that not everyfunction of s is a transformo The
kind of functions f(s) that are transforms of functions F(t) of broad c1asses
are limited, as we shall see (Chap. 6), by conditions of regularity that include
requirements that f be continuous on a half line s > IXand that f(s) -+ Oas
s -+ oo.
We have noted that if L{ F} and L{ G} exist, then
7 A THEOREM ON SUBSTITUTION
Let a function F(t) be such that its Laplace integral converges when
s > IX.Then, replacing the argument of the transform f(s) by s - a,wherea
is a constant, we have
m!
Hence (s > a).
's - a)m+ 1 = L{ tme"t}
As another iIlustration,
s
L{coskt} = S2+ (s > O),
s+a
and therefore (s> -a).
L{e-a, cos kt} = (s + a)2 + k2
THE LAPLACE TRANSFORMATION [SECo 8 17
L-
S + 1 =! ! - 2/
1
{ S2 }
+ 2s 2 + 2 e .
The procedure can be shortened for such a simple fraction by
writing
s + 1 = !(s + 2) + !s,
s+l 1 1 11
and hence +--
S(S+ 2) - 2 s 2 s + 2'
Example 2
Find L -lL(s : a)2}'
1 1 -1 O
s
- 1
2 e'" a
s-a
n!
3 t" (n = 1,2,...) O
5"+1
n!
4 t"e'" (n = 1,2,...) a
(5- a)"+ 1
k
5 sin kt O
S2 + k2
s
6 cos kt O
S2 + k2
k
7 sinh kt 1'"
S2 - P
5
8 coshkt Ikl
S2 - k2
k
9 e-al sinkt -a
(5 + a)2 + k2
. 5+a
10 e-al cos kt -a
(5 + a)2 + k2
11 Ji fi O
2.j?
1
12 O
Ji
13 r(k + 1) O
t1(k > -1) 7+1
14 r(k + 1) a
t1e"'(k>-1)
(5 - a)1+1
18 cos at - cos bt
(b2 a2)s - O
(S2 + a2)(52 + b2)
THE LAPLACE TRANSFORMATION [SEC.8 19
we find that
a2 1 1 a
s(s + af = -;- s + a - (s + a)2.
Referring to Table 1, we can now write the result
a2
L-l = 1 - e-al - ate-al.
{ s(s + a)2}
Example 3
Since
f(s) =. Ss + 3
~. I '"
1 s+l 3
=-- +-
s- 1 (s + 1)2+ 4 (s + 1)2+ 4.
Referring to Table 1, or to Theorem 3, we see that
F(t) = el - e-'(cos 2t - ! sin 2t).
20 SECo 9] OPERATIONAL MATHEMATICS
(2)
Y(t) = A cos kt + ~ sin kt,
or y(s) =
S2 - s+2 A
=-+-+-.
B e
s(s - 3)(s + 2) s s- 3 s+ 2
Evaluating the coefficients A, B, and e as in the preceding section, we
find that
11 8 1 4 1
y(s) = ---+--+--.
3s 15 s - 3 5 s + 2
Hence Y(t) = -j + /se3t + !e-2t.
It is easy to verify that this function Y satisfies the differential equation
(3) and both conditions (4).
Example 3 Find the functions Y(t) and Z(t) that satisfy the following system
of differential equations:
Y"(t) - Z"(t) + Z'(t) - Y(t) = et - 2,
2 Y"(t) - Z"(t) - 2Y'(t) + Z(t) = - t,
Y(O) = Y'(O)= Z(O) = Z'(O) = o.
22 SEC.9] OPERATIONAL MATHEMATICS
Y(O)= o, Y'(O)= 1,
so that
C-2+s
y(s) = S(S2 - 2s + 5)
C- 2 1 s - 1 C+ 3 2
= S- [-:;- (s - 1)2 + ]
4 + 10 (s - 1)2 + 4'
C - 2 c + 3 C- 2
Thus Y(t) = --s- + e'( 10 sin2t - S- cos 2t) .
Since Y(n/8) = 1,it followsthat
C- 2 e"/8
1 = -+-(C + 3 - 2C + 4),
5 10j2
or that C = 7. Hence the solution is
Y(t) = 1+ e'(sin2t - cos 2t).
PROBLEMS
1. Use Theorem 3 to (a) obtain entry 9 from entry 5 in Table 1 and (b) show that
L -l{(S - a)-t} = e"'(1tt)-t, given L -l{S-t}.
2. Use partial fractions to find the inverse transforms of
a a2 a3
(a) s(s + a); (b) s(S2+ a2); (e) s(s + a)3'
Ans. (a) 1 - e-a,; (b) 1 - cos at; (e) 1 - (1 + at + ta2t2)e-a,.
3. Use partial fractions to obtain the inverse transforms shown in (a) entry 16 of
Table 1; (b) entry 17 of Table 1.
Solvethe followingproblemsand verifythat your solutionssatisfythe differential
equatios and any accompanyingboundary conditions.
4. Y"(t)- k2Y(t)= O(k # O). Ans. Y(t) = Ce'"+ C2e-kt.
5. Y"(t)- 2kY'(t) + k2Y(t) = O. Ans. Y(t) = e"'(C+ C2t).
6. Y"(t)+ k2Y(t) = a. Ans. Y(t) = C sinkt + C2coskt + a/k2.
7. Y"(t)+ 2Y'(t) + 2Y(t) = O,Y(O)= O,Y'(O)= 1. Ans. Y = e-' sin t.
8. Y"(t)+ 4Y(t)= sint, Y(O)= Y'(O)= O. Ans. Y(t)= sint - i sin2t. t
9. Y"(t)+ Y'(t) = t2 + 2t, Y(O)= 4, Y'(O)= -2. Ans. 3Y(t) = t3 + 6e-' + 6.
10. Y"'(t)+ Y'(t) = lOe2',Y(O)= Y'(O)= Y"(O)= O.
Ans. Y(t) = e2'- 5 - 2 sint + 4 cos t.
11. y(4/(t) = Y(t), Y(O) = Y'(O) = Y"(O) = O, Y"'(O) = 2.
Ans. Y(t) = sinh t - sin t.
12. X'(t) + Y'(t) + X(t) + Y(t) = 1, Y'(t) - 2X(t) - Y(t) = O,X(O)= O,Y(O)= 1.
. Ans. X(t) = e-r - 1, Y(t)= 2 - e-'.
24 SECo 10] OPERATIONAL MATHEMATICS
13. Y"(t)+ 2Z'(t)+ Y(t) = O,Y'(t) - Z'(t) - 2Y(t) + 2Z(t) = 1 - 2t, Y(O)= Y'(O)=
Z(O)= O. Ans. Y(t) = 2(1 - e-' - te-'), Z(t) = Y(t) - t. .
14. Y"(t)+ Y(t) = t, Y'(O)= 1, Y(n) = O. Ans. Y(t) = t + n cos t.
15. y"(x) + y(x) = 1,y(O)= 1,y(tn) = O. Ans. y(x) = 1 - sinx.
16. y"(x)+ 2y'(x)+ y(x) = O,y(O)= O,y(l) = 1.
(1)
T {F(t)} = L CX)K(t,s)F(t) dt = f(s),
(2)
T {F'(t)} = SoCX)
K(t,s)F'(t) dt
= K(t'S)F(t)J~ - SoCX)K'(t,s)F(t)dt
= - K(O,s)F(O)- SoCX)
K'(t,s)F(t) dt,
provided that K(t,s)F(t) -> O as t -> oo. The kernel K must involve some
parameter s if the transform f is to correspond to a unique function F of
some large class; otherwise, f is merely the numerical value of the integral
(1) of the product K(t)F(t), a number that is the same for many functions F.
In order that the final integral in formula (2) will representf(s) except
for a factor A.(s),where A.is some function of s, we require that
and formula (2) is the basic operational property of the Laplace transforma-
tion:
(6)
T{F(t)} = f K(t,s)F(t)dt = /(s)
is a generalization o/ linear vector trans/ormations. Our transformation (1) is
a special case of (6) in which a = Oand b is infinite.
A generalized vector, or function, F has an infinity of components
consisting of the values F(t) for each t; that is, the components are all the
ordinates of points on the graph of F(t) over the interval a ~ t ~ b. The
generalized scalar product, called the inner product, of two such vectors F
and G is the generalized sum of products of corresponding components,
namely
r F(t)G(t) dt.
Thus for each fixed s, the integral (6) is the inner product of the generalized
vectors K(t,s) and F(t).
We begin with ordinary vectors F in three-dimensional space having
rectangular cartesian components F1,F2,F3, which may be written F(l),
F(2), F(3). A general linear transformation of F into a new vector f can be
written in terms of components and constants Kij (i,j = 1,2,3). For the
first component we write
/1 = KllF1 + K12F2 + K13F3 = K1 . F
where K1 is a fixed vector with components Kll, K12, K13, and the dot
denotes the scalar product ofvectors. Similarly, K2 and K3 are fixed vectors
and
/2 = K2 F, .
/3 = K3 F. .
In space of n dimensions the general linear vector transformation is
determined by a familyof n fixed vectors K:
(7) ; = K . F (i = 1,2,..., n).
1 Churchill, R. V.:"Fourier Series and Boundary Value Problems," 2d ed., chap. 3, 1963.
26 SECo 10] OPERATIONAL MATHEMATICS
(8)
f(s) = f K(t,s)F(t) dt.
This is the linear integral transformation (6).
Functions of certain elasses,as well as vectors, are elements of linear
veetor spaees. The abstract theory of linear spaees is independent of the
kind of elements involved. In that respeet the treatment of funetions as
veetors is not merely an intuitive approach or an analogy. But the abstraet
theory specifies properties of inner products without suggesting formulas
for them. The theory will not be needed here.
2
Further Properties of the
Transfor.mation
11 TRANSLATION OF F(t)
There are several further operational properties of the Laplace transforma-
tion that are important in the applications. Those properties whose deriva-
tions and applications do not necessarily involve the use of complex variables
will be taken up in this chapter.
We begin with an analog of Theorem 3 of the first chapter. According
to that theorem, the multiplication of the object function by an exponential
function corresponds to a linear substitution for s in the transformo Now
let us note the correspondence arising from the multiplication of the trans-
form by an exponential function.
Let F(t) ha ve a transform,
Then
27
28 SECo 11] OPERATIONAL MATHEMATlCS
we see that
f(s)e-bS = Lco e-S'Fb(T) d'C.
The following property is therefore established.
provided that F(t - b) is defined where t > O; that is, provided that F(t) has
numerical values for those values of t in the range t > - b. For example,
the function F(t) = sin kt is defined for all t, so in this case we can write
k -bs
(t > O,b ~ O).
Fb(t) = Sb(t)sin k(t - b) = L- 1 { S2 e+ k2 }
Rt)
o o
I
L
~
(b,O
Fig.4
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 12 29
f(es) = ~foooe-~p(~) dt
in terms of the transform of F(t/e), to establish this theorem:
(3)
12 STEP FUNCTIONS
When t ~ O,the bracket symbol [t] is used in mathematics to denote the
. . . , that does not exceed the number t. Thus [n] = 3
greatest integer, O, 1,2,
= [3]. The function [t]is therefore a step function of the type that is some-
times called a staircase function with unit rise and run :
30 SECo 12] OPERATIONAL MATHEMATlCS
[t] =O (O ;;;;
t < 1),
=1 (1 ;;;;
t < 2),
=2 (2 ;;;;
t < 3),.. ..
The staircase function with an arbitrary positive ron h, and with a unit rise
=0 (t < O).
The function Y(t - h) is then the same as the translated function Yh(t) and,
in view of Theorem 1, the transform y(s) of the function Y(t) satisfies the
equation
- 1
y(s) = e h'y(S) + -
s (s > O).
=L l+ !. = !--.! >
(2) y(s)
{[ h] } s1 - j (s O).
1
(3)
(h, t - to) = h when to < t < to + h,
10,1)
O Ih,OI
Fig.5
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 12 31
l(h,t-fo)
11
I '1
II I
I
'hl
1-:J
o (fo,O)
Fig.6
(5)
AIso, the area under the graph of the function 1 is unity for every positive h;
hence
(8)
h
lim
O f -et)
et) I(h, t - to) dt = 1.
The order of the positions of the limit and the integral here is important, for
in view of Eq. (7)
f et)
-et)
lim I(h, t
h O
- to) dt = O.
It is interesting to note the behavior of the transform of the finite unit
impulse function as h tends to zero. By evaluating the limit ofthe right-hand
member of Eq. (6), we find that
(9) limO L{I(h, t
h
- to)} = e-sto (to ~ O,h > O,S > O).
L{fim
h OI(h, t - to)} = L{O} = O.
32 SECo 12] OPERATIONAL MATHEMATICS
Equation (9) presents the exponential function e -sto, or exp ( sto), not as the -
transform of a function, but as a limit of transforms of functions of the set
I(h, t - to) consisting of one function for each positive value ofthe parameter
h. Thus exp(- sto)approximatesthe transformof the impulsefunction
I(h, t - to) for small positive values of h, when to ~ O.
and therefore
(12)
that is
1 2
(13) when O ~ t ~ h,
X(h,t) = 2mh t
t2 - (t - h)2 1 1
= 2mh = ;;; ( t - 2.h) when t ~ h.
fa>
- a>I(h, t - to)F(t) dt =h f
lo F(t) dt = F(to + Oh).
(1) lim
h"'O f a>
-a>
I(h, t - to)F(t) dt = F(to),
sometimes called the sifting property of selecting that value F(to) of F(t).
We abbreviate that property by writing
(2)
f:a> c5(t- to)F(t) dt = F(to),
where the entire left-hand member is a symbol that denotes the operation
represented by the left-hand member of Eq. (1). That symbol is selected so as
to suggest an integral of a function, because if c5(t- to) is replaced by
-
I(h, t to), the symboI does represent an integral whose value is approxi-
mately F(to) when h is a small positive number.
If to ~ O,then I(h, t - to) = Owhen t < Oand
(3) lim
h"'O of a> I(h, t - to)F(t) dt = F(to) (to ~ O,h > O)
(5)
. f~a> c5(t- to)dt = 1,
34 SECo 13] OPERATIONAL MATHEMATICS
the symbolic form of Eq. (8), Seco 12. When F(t) = e-sI and to ~ O, the
symbolic integration formula (4) can be written
(6)
to signify that X is the limit as h -+ Oof the solution X(h,t) of problem (10)
there, the problem with f>(t)replaced by [(h,t). In the transformed problem
(11),Seco12,we may let h tend to zero and assumethat x(h,s)-+ x(s) where
x(s) = L{ X(t)} ; theresultis
(8)
PROBLEMS
1 . With the aid of our theorems and known transforms find the inverse transforms
F(t) tabulated below and draw a graph of F(t).
I(s) F(t)
e-bs
(a) ----(b > O) 0(0 < t < b);t - b(t > b)
s
(b) fis-te-2'
e -,..
(e) S2 + 1 - S.(t) sin t
sint(O < t < n);O(t > n)
se -s/2
(e) S2 + n2
4
(f) (2s + 1
4
(g)--
2. From the Maclaurin series that represents (1 - X)-l when Ixl < 1, or from the
sum of an infinite geometric series, show that
1 1 1 e-los e-2h. 00 e-.Ios
-s 1 - e-Ios=-+-+-+"'=
s s s L- s
.=0
(h > O, s > O).
Apply the inverse transformation term by term to this infinite series, formally, to obtain
the result shown in Eq. (2), Seco12; namely,
-1 1 1 t
(h > O).
L {-;1 - e-los
} = 1 + h] [
3. Show that Eq. (2), Seco12,can be written in the form
Draw the graph of F to see that F can be written in terms of the bracket symbol [t] in
the form
F(t) = 1 + (-1)1'1.
36 SECo 13] OPERATIONAL MATHEMATICS
(b) Show that the function F in part (a) is described by the following difference
equation of the first order, when t > O:
F(t) + F(t - 1) = 2 where F(t) =O when t < O.
Transform that difference equation to verify the formula given in part (a) for j(s).
5. Apply Theorem 2 to the transformation found in Problem 4 to show that
Y(t) I c"F(t -
= n=O nh);
but for each fixed value of t this series is a finite series because F(t - nh) = O when
t - nh < O. Thus an altemate form of the solution is
where F(t) = b when t > Oand F(t) = Owhen t < O,under the ~ondition that Y(t) =O
when t ;;;; O.
a
Ans Y(t) =b t + -(t - 1)2 +... + -(t a" - n)"+l
. [ 2! (n+1)! J'
where n ;;;; t ;;;;n + 1 and n = O,1,2,.. . .
10. Initially a particle of mass m moving along the X axis is at the origin X = O with
a velocity Vo. The only external force that acts on that particle is an instantaneous
impulse Poat time to, in the X direction. Thus the displacement X(t) satisfies the follow-
ing symbolic equation and initial conditions:
mX"(t) = Poc5(t- to), X(O) = O,X'(O) = Vo.
Find the formal solution (writing L{X"} = S2X - vo)
X(t) m -
= vot + PO(t to)So(t - to) (t ~ O,to > O)
and show X(t) graphically. To verify the solution, show that X"(t) = O when t # to,
that X(O) = Oand X'(O) = Vo,and that the momentum mX'(t) undergoes a jump Po at
the instant to.
11. Replace the instantaneous impulse Poc5(t- to) in Probo 10 by the finite impulse
PoI(h,t - to) and then find the displacements X(h,t). Show that the limit of X(h,t) as
h Ois the function X(t) found before.
12. Problem 10 can be stated in terms of functions as folIows:
14. Show that the formal solution of the symbolic problem (if we write L{ X'} = sx
here)
15. Let Al denote the change in /(h, t - to) when !J.t = -h. Draw the graph of the
function of t represented by the difference quotient
If F and F' are continuous on an interval that includes the points to and to + 2h, show
that .
Al 1
f a)
-F(t)
-a) h
dt = --
h 'o f'OH F(-r: + h) - F(-r:)
h
d-r:
1 'O+h
-Al
lim
a)
f
, 0 -00 h
F(t) dt = - F'(to)'
17. Since /(h,t) = l/h when 0< t < h and vanishes when t < O and when t > h,
describe the function /(h, - t) and show that
when F is defined for all t and continuous over some intervalltl < a. Symbolically,
o
f
(l(X)
a(x) a
(2)
g'(x) = X f(x,t)dt + f[x,fJ(x)]fJ'(x) - f[x,a(x)]a'(x).
The foregoing two properties of the integral (1) are established in
advanced calculus from fairly elementary considerations. Let us indicate
here how the continuity requirements on the integrand can be relaxed in
establishing certain properties.
Let a, b, and e denote constants, and let the integrand of the integral
aX+b
(3)
f o f(x,t) dt = h(x)
(O ~ x ~ e)
where f(x,t) is continuous over each subregion up to its boundary; that is,
for each subregionf can be definedon the boundary in such a way that it is
continuous in x and t together over the closed subregion. We then say that f
is continuous by subregions.
40 SECo 14] OPERATlONAL MATHEMATICS
e x Fig.7
01
exists. It can be written as the sum of iterated integrals over each of the
subregions in which f(x,t) is continuous. When the order of integration is
interchanged in each subregion, the sum of the integrals represents the
iterated integral, with order interchanged, over the entire region.
Thus our condition of continuity by subregions permits the interchange
of order of integration with respect to t and x. In particular, when a = O so
that the limits are constants,
(6)
5: s: f(x,t) dt dx = s: 5: f(x,t) dx dt.
These conclusions can be reached by the same procedure for regions of more
general shapes. The argument used above applied, for instance, when the
lower limits ofthe integrals (5) and (6) are constants other than zero.
II
(1)
when T> log (1M, a value that is independent of x, so that the integral
converges uniformly with respect to x when x ;; 1. In the same way we can
see that it converges uniformly when x ;; Xl whenever Xl > O.
But the convergence is not uniform in the range x ;; Oor x > O,because
the remainder (3) when x > O is small only if xT is large. Since x can be
arbitrarily small, the value of T that makes R(x, T) small must depend on x.
A consequence of this lack of uniform convergence in the range x ;; Ois the
discontinuity of the function p(x) at x = Othat is exhibited by Eq. (2).
The Weierstrass test for uniform convergence of the integral
(4) (O~ x ~ e)
{X' f(x,t) dt = q(x)
42 SECo 15] OPERATIONAL MATHEMATICS
and such that the integral f;;>M(t) dt exists, then the integral (4) is uniformly
convergent with respect to x(O ;;:;x ;;:;c). This test is established by first
noting that the integral (4) exists according to the comparison test (Probo 14,
Seco5),then that the absolute value of the remainder for the integral, for all x,
in the interval, do es not exceed the remainder f,;?M(t) dt in the integral of M.
The latter remainder is independent of x, and it tends to zero as T -. 000
The Weierstrass test sta tes useful sufficient, but not necessary, con-
ditions for uniform convergenceo Under those conditions the integral (4) is
also absolutely convergent.
Ir the integral (4) is uniforrnly convergent with respect to x and if its
integrand f(x,t) is continuous by subregions over the rectangle O ;;:;x ;;:;c,
O ;;:;t ;;:;T for each positive T, then the integral represents a sectionally
continuous function q(x); moreover
(5)
I: q(x) dx = I: {" f(x,t) dt dx = foooI: f(x,t) dx dt;
that is, the order of integration of the definite and improper integrals here can be
interchanged.
In particular, let f(x,t) be continuous in each rectangle O ;;:;x ;;:;c,
O ;;:; t ;;:; T, except possibly for finite jumps across a set of lines t = t(i =
1,2,. . o, n). Then q(x) is continuous ifthe integral (4) is uniforrnly convergent.
Suppose that Jf/Jxas well as f satisfies those continuity requirements and
that the integral f;;>(of/Jx)dt converges uniformly and the integral (4) exists.
Then the derivative of the latter integral exists and the order of differentiation
with respect to x and integration with respect to t can be interchanged:
d oo o
dxf f oo
(6) (O < x < c).
o f(x,t) dt = o oxf(x,t) dt
The continuity of q(x) can be seen by writing
(O ;;:; x ;;:;e).
q(x) = foTf(x,t) dt + R(x,T)
For each fixed T the integral on the right is a sum of integrals of the type
"
f"-1 f(x,t) dt
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 16 43
(7)
I: fooo : dt dx = fooo I:: dx dt
= fooof(r,t) dt - fooof(O,t)dt,
where we have used property (5) to invert the order of integration. Since
f:;' (8f/8x) dt is a continuous function of x, differentiationofthe first and last
membersof Eqs. (7)with respect to r leads to the equation
f_ ]
oo 8 d oo
o 8x
f(x,t)
x='
dt = - f
dr o f(r,t) dt
(O < r < e),
16 CONVOLUTION
We now determine the operation on two functions of t that corresponds to
multiplying their transforms together. This convolution operation, which
gives the inverse transform ofthe product oftwo transforms directly in terms
of the original functions, is one of primary importance in operational
mathematics.
Let F(t) and G(t) denote any two functions that are sectionally con-
tinuous on each finite intervalO ;;;t ;;;T and (9(elXt),
and write
f(s) = L{F(t)}, g(s) = L{ G(t)} (s > a).
It will be convenient to define G(t) to be zero when t < O; but our final result
depends only on the values of F(t)and G(t)when t > O. According to Theorem
1, for each fixed -r('r;:;O),
that is,
T
(1) f(s)g(s) = lim
ff
T-+oo o o
OO
F(T)e-srG(t - T) dt dT.
Since f(s) and g(s) exist when s > cx,the limit here exists.
The integrand ofthe inner integral in Eq. (1)is continuous by subregions
(Sec.14),over the rectangular region O ;;;;T ;;;;T, O ;;;;t ;;;;R, for each pair of
positive constants T and R. For if the jumps of F(T)and G(t) occur at
T = T~i = 1,2,..., m) and t = tJ{j = 1,2,..., n), then the jumps of the
integrand occur at the lines T = TIand t =T+ tj' In view of Eq. (5), Seco15,
the order of integration in Eq. (1) can be interchanged, provided that the
improper integral there is uniformlyconvergentwith respectto its parameter
T(O ;;;;T ;;;; T).
The uniform convergence is seen by noting that, owing to the exponen-
tial order of F and G, a constant N exists such that
(2)
where M(t) = N exp [-(s - cx)t]. The function M(t) satisfies the conditions
ofthe Weierstrass test (Sec.15); that is, M(t) is independent of Tand integrable
from zero to infinity.
Equation (1) can now be written in the form
= T-+oo
lim [11(T)+ lz(T)],
where
11(T) = foT e-sr IoTF(T)G(t - T)dTdt,
11z(T)1< N
fT ooe-(S-",)r
fo dTdt = s-cx
T NT e-(s-",)T;
therefore lim001z(T) = O.
T-+
FURTHER PROPERTIES OF THE TRANSFORMATION [SEC. 16 45
(5)
F(t) * G(t) = f~ F(1:)G(t - 1:)d1:,
Theorem 3 1ff(s) and g(s) are the transforms oftwofunctions F(t) and G(t)
that are sectionally continuous on each interval O ~ t ~ T and of the
order of eal as t tends to infinity, then the transform of the convolution
F(t) * G(t) exists when s > IX; it is f(s)g(s). Thus the inverse transform of
the product f(s)g(s) is given by the formula
(12) L -1
~
{ s,s+l }
= erf(Ji).
17 PROPERTIES OF CONVOLUTION
By substituting the new variable of integration A = t - 1"in the convolution
integral (5), Seco 16, we find that the convolution operation is commutative;
that is,
(1)
F(t) * G(t)= G(t)* F(t) = {F(t - A)G(A)dA.
FURTHER PROPERTlES OF THE TRANSFORMATION [SECo 17 47
F*(G*H) = f~ F(-r:)
f~-t G(t - T - A.)H(A.)dA.dT
and observe that the iterated integral here represents an integration over tbe
triangular region shown in Fig. 8. When the order of integration is reversed,
the integral becomes
Fig.8
o
48 SECo 17] OPERATIONAL MATHEMATICS
(t > O).
(7)
L -1Hf(S)} = I~ F(T)dT,
(8)
L -1 {s12f(S)} = {I: F(A.) dA. dT,
etc., for division by sn, provided F(t) is sectionally continuous and of the
order of eal(ex> O),wheres > ex.
FURTHER PROPERTIES OF THE TRANSFORMATION [SEC. 17 49
PROBLEMS
1. Use transfonns to show that
tS
(a) 1 * 1 * 1 = tt2; (b) t * t * t = 5!;
L -1
{ --d-
v' s - 1}
= el+ ~ + e'erf(Jt).
v' 1[t
4. Generalize fonnula (10), Seco 16, by replacing s by sla2 to obtain transfonn 40,
Appendix A, Table A.2.
5. Generalize fonnula (12), Sec. 16, to obtain transfonn 44, Appendix A, Table A.2.
6. Find L -I{(S + 1)-1s-t} and generalize to obtain transfonn 41, Appendix A,
Table A.2.
7. Obtain the inverse transfonn
L-I{SS~\} =e'erf(Jt)-2~.
8. With the aid of transfonnation (9), Seco16, show that
[ q dx = [s: f dt dx + [R(X, T) dx
and hencethat, correspondingto each positivenumber E,a number T.existssuch that
when T > T,
12. Extend property (5),Sec. 15,for the interchange of order of integration of definite
and improper integrals, to the following case:
where b 6; Oand <jJis continuous by subregions over the region b < x < e, x < t < T,
whenever T> e, and f:><jJ dt is unifonnly convergent with respect to x(b < x < e).
Suggestions: Show the region graphically. Introduce a functionfto which fonnula
(5), Sec. 15, applies by writing
f(x,t) = <jJ(x,t) whenb < x < e and t > x,
f(x,t) = O wheneitherx < b or t < X.
or Z"(t) = -k2Z(t) + F(t). Therefore Z(t) satisfies Eq. (1), and the
function (2) is verified as the general solution of that equation for all
values of t.
The function Y(t) given by Eq. (2) is continuous, together with
Y'(t), when F(t) is sectionally continuous, in view of the above ex-
pressions for Z(t) and Z'(t). The condition of exponential order on
52 SECo 18] OPERATlONAL MATHEMATICS
F(t) was not used in verifying the solution (2); the condition that F(t)
be sectionally continuous, on an interval containing all values of t to
be considered, was sufficient.
whose solution is
y(s) = a(s~ + s~}.
Therefore Y(t) = a(t + it3),
which can be verified directly as the solution of the above integral
equation.
(3)
Y(t) = F(t) + { G(t - -r)Y(-r)d-r,
where the functions F(t) and G(t) are given and Y(t) is to be found. Since
the transformed equation is
y(s) = f(s) + g(s)y(s),
the transform of the unknown function is
f(s)
(4)
y(s) = 1 - g(s)"
PROBLEMS
Solve the following differential equations.
1. Y"(t) - k2 Y(t) = F(t), if Y(O)= Y'(O)= O(k # O).
Ans. 2kY(t) = e'''f~e-k'F(r)dt - e-k'f'o~rF(t)dt.
2. Y"(t) - 2k Y'(t) + k2 Y(t) = F(t).
Ans. e-kIY(t) = CI + C2t + f~ (t - t)e-krF(t)dt.
3. 2Y"(t) - 3Y'(t) - 2Y(t) = F(t), if Y(O)= Y'(O)= O. Verify that your function
satisfies the differential equation and the initiaI conditions.
4. Y"(t) + 4Y/(t) + 5Y(t) = F(t), if Y(O)= Y(7t/2)= O.
Ans. e2IY(t) = sin t f~/2 F(t)e2rcos t dt - COS t f~F(t)e2' sin 1:dt.
5. Y///(t)- Y'(t) = F(t). Verifyyour result.
6. Y///(t)- Y"(t)+ Y'(t) - Y(t) = F(t),if Y(O)= Y'(O)= Y"(O)= O.
Ans. 2Y(t)= f~ F(t- t)(e' - cost ~ sint) dt.
7. Show that the solution of the system of differentiaI equations
X/(t) - 2Y'(t) = F(t), X"(t) - Y"(t) + Y(t) = O,
under the conditions X(O) = X'(O) = Y(O)= Y'(O)= O,so that F(O)= O,is
{ Y(t) dt - Y'(t) = t,
Y(O)= 2.
Ans. Y(t) = 1 + cosht.
10. Derive the solution (7) of Abel's integral equation (6).
11 . Solve the integral equation
= ate-'.
Y(t) = a sin t - 2 { Y(t)cos (t - t)dt.
Ans. Y(t)
Ans. Y(t) = 1.
2Y(t) + {Y(t)Y(t - t)dt =t+ 2.
14. Write Y"(t) = V(t). (a) Ir Y(O)= a and Y'(O)= b show that
Y'(t) = 1* V(t) + b, Y(t) = t* V(t) + bt + a;
(b) then convert the initial-value problem
Y"(t) + a(t)Y'(t) + P(t)Y(t) = F(t), Y(O) = Y'(O) = O,
19 DERIVATIVESOF TRANSFORMS
When the Laplace integral
(1)
f(s) = 1'" e-S'F(t)dt
(t > O),
where N 2 represents the maximum value of the function tne-" when t > O.
Thus the function tnF(t) is of the order of ellOI,where CXo= cx + .
Also let F(t) be sectionally continuous on each intervalO < t < T.
Then tnF(t) has that property. The absolute value of the integrand of the
Laplace integral of tnF(t)satisfiesthe condition
(2) (n = O, 1,2, . . . ),
(3)
(n = o, 1,2,. . . )
IX) ~F(t)e-SI dt
o
e-s'cosktdt = ~s +
converges, and the integral
(8)
L{ :k cos kt} = fooo:k(e-st cos kt) dt
= - 100e-S't sin kt dt
converges uniformly with respect to k for all real k, according to the
Weierstrass test. AlI integrands here are continuous functions of k and t.
. Hence formula (6), Sec. 15,applies to show that the integral (8) represents
the derivative, with respect to k, of the integral (7); that is,
a S 2ks
o
-L{tS1Okt} = ak( ? .
)
.? = -,? . .?,?,
and this is the transformation (6).
FURTHER PROPERTIES OF THE TRANSFORMATlON [SEC. 20 57
20 SERIES OF TRANSFORMS
A useful method of finding inverse transforms L -1 {J(s)} is that of repre-
senting the functionfby an infinite series ofknown transforms, then applying
the operator L -1 to the terms of the series. For certain types of series in
powers of l/s, conditions for the validity of the procedure will now be
established.
Theorem 7 Let f(s) denote the sum of an infinite series of positive integral
powers of l/s which is absolutely convergent when s > IX,where IX~ O:
1 ao al
(1) f(s) =
ex>
n=O
an-
~+l
= -
S
+ - + ...
S2
(s > oc~ O).
the numbers on the right are independent of t, and they are terms of a
convergent series. The series of the continuous functions ane-slr'fn! can
therefore be integrated term by term over the bounded interval (O,T)
(cf. Probo 15, Seco21); thus
T
a
I
T
o e-SIF(t) dt L~
00
= n=ono I o e-sIr' dt
= n=O
I: anrL{r'}
noL
- f T
ooe-str' dt
]
= I: - n. fT ooe-sIr' dt .
n=O( S'a: 1 a~
)
By subtracting the last series from series (1) which converges to f(s), we
find that
oo
a
(5) f(s) - Io
T
e-SIF(t)dt =
00
L ~ fT
n=on. e-sIr' dt.
We are to prove that the difference (5) vanishes as T -+ 00; that is,
to each positive number we are to exhibita correspondingnumber T.
such that
f.
lanl 00 -st.,. d ::::;lanl L {r'} = lanl
(~
n., T e, - n.I
"
t ..n + 1
~
(8)
I
f(s) - I T
o
e-stF(t)dt
I
~ Ni:.l1a~1
n=O n. iooe-sIr' dt
T
+ RN(T)
I: la~1
RN(T) = n=N n. T n=NS' i
ooe-str' dt ~ I: la:'l'
Since series (1) is absolutely convergent, it now follows that to each ( > O)
there corresponds a number N.. independent of T, such that
RN(T) < ! wheneverN > N..
Now consider the sum from n = O to N - 1 in condition (8). Let N
be some fixed integer greater than N. By successive integrations by parts
o
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 20 59
we find that
1
-
n! Tf co
e-stl"dt = -
e-ST
- + +
Tn
s [ n! (n - l)!s (n - 2)!S2
Tn-l
+ ... + - .
snJ
Tn-2 1
n! T f
lanl coe-Sltn dt < Me-ST
[
(ST)n
n!
+ (sT)n-l +... + 11
(n - 1)! J
{ } (O< k ;; 1).
L g(s) = r(k) t + r(k) * F(t)
By writing the convolution here as an integral from t = O to t = plus an
integral from to t, we can modify the proof in Seco 17 to show that
rc-l * F(t), and hence L -1{g(S)}, is 19(e"")if (Xl> (X.
In view of Eq. (11) we can write
(12)
rc-1 * F(t) = E (t - t)k-l F(t) dt
'
L~
)
= n=On.
a
Io (t - tt-1tndt
) a
= L ...!!...tk-l* tn,
n=On!
where the term-by-term integration is valid because the power series for
F(t) converges uniformly with respect to t over the interval O < t < t
(Prob. 15, Seco21). By using known transforms, we find easily that
..k-l * tn =
tU
r(k)n! ("+k
r(n + 1 + k)
and therefore
rc-l
- )
r(k) * F(t) L an-
= n=O tn+k .-'
1
G(t) = - L) -(-l)n 2n+ l("+t
and therefore we can write the transform of the product of t" and any
derivative of Y(t) in terms of y(s); for instance,
d2
L{t2Y'(t)} = ds2[sy(s) - Y(O)] = sy"(s)+ 2y'(s),
d
L{tY"(t)} = - dS[S2y(S)- sY(O)- Y'(O)]
= -s2y'(S) - 2sy(s)+ Y(O).
A linear differential equation in Y(t) whose coefficients are polynomials in t
transforms into a linear differential equation in y(s) whose coefficients are
polynomials in s. In case the transformed equation is simpler than the
original, the transformation may enable us to find the solution of the
original equation.
If the coefficients are polynomials of the first degree, the transformed
equation is a linear equation of the first order, whose solution can be
written in terms ofan integral. To find the solution ofthe original equation,
however, the inverse transform of the solution of the new equation must be
obtained.
or
y'(s)+ (~ - s)y(S)= -~,
62 SECo 21] OPERATIONAL MATHEMATICS
Integrating, we have
l e
y(s)= -S2 + -ets2
S2 '
=~ ~ -t = ~ I - ~~ - ~ -~ ~ - ...
y(s)
. S
I+
( S2 ) S [ 2 S2 2 2 2 !S4 ( ) J
- e - I) -
[ (- 1)
- -S 1 + n=1
ex> n(I)(3)...
2nn.s,
1 2n
(2n
. J - e n=O
ex> (-I)n(2n)!
L (2nn.1)2?H'"
s "
where O! = 1. The ratio test shows that series in positive powers
of l/s are absolutely convergent when s > 1. Theorem 7 therefore
applies to show that the operator L - 1 can be applied term by term
to that series to represent the continuous function whose transform is
y(s) as a convergent series in powers of t:
PROBlEMS
1. Use the transformation (6) and Theorem 2 to find L{J o(at)} listed in Appendix A,
Table A.2 (transform 55).
2. If Y, Y', and Y" are of exponential order and Y, Y' are continuous (t ~ O)while
Y" is sectionaJly continuous, and if L{ Y(t)} = y(s), show that
L{t2Y"(t)} = s2y"(S)+ 4sy'(s) + 2y(s).
Solve the following differentiaJ equations for Y if Y and its derivatives are to
have transforms.
3. Y"(t)+ atY'(t) - 2aY(t) = 1, Y(O)= Y'(O)= O,a > O. Ans. Y(t) = t2/2.
4. tY"(t) + (t - l)Y'(t) + Y(t) = O,Y(O)= O. Ans. Y(t) = Ct2e-'.
5. tY"(t) + (2t + 3)Y'(t) + (t + 3)Y(t) = 3e-'. Ans. Y(t) = (C + t)e-'.
6. t2Y"(t) - 2Y(t) = 2t, Y(2) = 2. Ans. Y(t) = t2 - t.
7. When k is a constant, show that the equation
t2Z"(t) + 2tZ'(t) + kZ(t) = O
leads to the same differentiaJ equation in the transform z(s).
8. With the aid of transformation (6) show that
s
L{J(t)} = ~- 1.
S2 + l
Use formula (8) to show that J(t) = -J l(t), and hence that
L{J(t)} = p+t- s.
p+t
9. Use the transforms of Jo and J (Prob. 8) to show that
(a)
io -
' Jo(-r)Jo(t- -r)d-r= sint;
L -1 He-l/.} = Jo(~,
and thus obtain the transformation 75, Appendix A, Table A.2.
Apply Theorem 8 to verify the following transformations:
1.
11. L-l{)se-l/.} = ~CoS(~.
(Cf. transformation 76, Appendix A, Table A.2.)
{ Js } .fii
i
(Cf. transformation 77, Appendix A, Table A.2.)
FURTHER PROPERTlES OF THE TRANSFORMATION [SECo 22 65
-1 1 f'J .(t)
13 )
. L { (S2 + 1).+t} = (1)(3)(5)...(2n - 1) (n = 1,2,... .
(Cf.transformation57,AppendixA, TableA.2.)
1 1""t"k
14. L -1 { 1- 1} = t);:l r(nk) (k> O,s > 1).
15. Given that a power series L':=oA.-r. converges uniformly over an interval
O ~ -r ~ t to a sum F(-r),that is, the series converges to F(-r)there and
N-l
F(-r) L A.-r. +
= "=0 R~-r) (O ~ -r~ t)
a result that was used to obtain Eq. (12), Seco20. Suggestion: Write
N-l
t-C * F(t) - L A.t-C * t. = t-C * RN(t)
n=O
and prove that the right-hand member vanishes as N 00, when t and e are kept
fixed.
22 INTEGRATIONOF TRANSFORMS
When a function F(t) is sectionally continuous and of the order of 't, then
its Laplace integral
f(x) = La>e-xtF(t) dt
is uniformly convergent with respect to x in every interval x;;; OCI,where
OCl> oc,according to Theorem 5. It follows from formula (5), Seco 15, that
when r > s > oc,
(1)
fs f(x) dx = Jofa>F(t)e-st
a>
t
dt (s > oc)
66 SECo 23] OPERATIONAL MATHEMATICS
(2)
Sufficient conditions for the validity of formula (2) are that F(t) be
sectionally continuous and of the order of eaJ,that s > exin formula (2),
and further that the limil of F(t)/t exists as t -. + O.
The function F(t) = sin kt, for example, satisfies the above conditions
when ex = O;in particular,t-l sinkt -. k as t -. O. Hencewhens > O,
sin kt k dx 1t S k
= arctan-.
""
(3) L{ - t } = fs x2 + k2 = -2 - arctan-
k s
Recalling how integration with respect to t corresponds to division
by s, we can now write the transform of the sine-integral function
'Sint
(4)
.
SI t = -dt.
o t i
This function is of some importance in applied mathematics. Its values are
tabulated in the more extensive mathematical tables. If k = l in Eq. (3),
it follows from Eq. (4) that
L
{ t }
= f s
---
( x+a x+b )
dx=log-
x+bs ]
,
when s > -a and s> -b. Hence
e-al - e-bt S+ b
(6) L = log-.s + a
{ t }
When a = O and b = 1, we have the special case
23 PERIODIC FUNCTIONS
Let a function F be periodic with period a over the half line t > O and
sectionally continuous over a period O < t < a. That periodicity can be
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 23 67
and introduce a function Fo that is the same as F when O < t < a and zero
elsewhere; thus
Mle,t)
(0,1) I 1
I I
I I
.I 1
o (2e,0) I
I I
I I
I L-
Fig.9
1 es
(3) (s > O).
L{M(e,t)} = stanh"2
The integral ofthe function M from Oto t is the function H(e,t) defined
as follows:
This function, whose graph is the triangular wave shown in Fig. lO, has the
transform
1 es
(4) (s > O).
L{ H(e,t)} = S2tanh"2
Hle,t)
(O,e)
Fig.10
jf
i
~
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 23 69
G(t + 2e)= - G(t + e) = G(t); that is, G is periodic with period 2e. M(e,t)
is an example of sueh a funetion, as is sin t when e = n, 3n, Sn, . .. ,
Ir Go(t) = G(t) when O < t < e and Go vanishes for aIl t outside that
interval, and if we write G(t) = Owhen t < O,we can see that G is deseribed
by the difference equation
(5) G(t) + G(t - e) = Go(t).
The transform g(s) therefore satisfies the equation
(6)
f~ e-s'G(t)dt
g(s) =- (s > O).
o (271',0)
Fig.11
70 SECo 24] OPERATlONAL MATHEMATICS
therefore
24 PARTIAL FRACTIONS
-
FURTHER PROPERTlES OF THE TRANSFORMATION [SECo 24 71
where we have evaluated the limit of (s - a)/q(s) as the limit of the quotient
of the derivatives of s - a and q(s), since q(a) = O and q'(a) ::j:O because
s = a is a simplezero of q(s). ConsequentIy 4>(a)= p(a)/q'(a).
When aIl the factors of q(s) are linear and not repeated and when
q(s) is written in the form
(3)
where all the constants an are distinct, we can write the inverse transform of
f(s) in fuIl. Let qn(s) denote the product of aIl the factors on the right of
Eq. (3) except the factor s - an, so that the function 4>(s)corresponding to
that factor is p(s)/qn(s). Then
(4) L- 1 P(S) = I
p(an) ~nt
{ q(s) } n= 1 qn(an) = I p(an) eant
n= 1 q'(an) .
The second sum here is sometimes caIled Heaviside's expansion.
The principal results are stated in the foIlowing theorem.
( )
'f' a t:: or q'(a{ ,
where 4>(s)is the quotient of p(s) divided by the product of all factors
of q(s) except s-a.
Theorem 12 is valid when the constant a is any complex number.
72 SECo 24] OPERATIONAL MATHEMATICS
PROBLEMS
1. Show that Theorem 9 applies and find the transfonn of: (a) (1 - cos at)/t;
(b) (1 - cosh at)/t; (e) (e' - cos t)/t.
Ans. (a) (b) See transfonns 105 and 106,Appendix A. Table A.2;
t
(e) log [(s2 + 1)/(s - 1)2](s > 1).
2. The condition that F(t)/t has a limit as t -+ +0 was used in proving Theorem 9.
Ir it is replaced by the condition that for some positive constants k and M, IF(t)1< Mr<
over an interval O < t < T, then g(s), the transfonn of F(t)/t, exists and vanishes as
s -+ oo. Thus fonnula (2) in the theorem is still valido Use this fact to establish trans-
fonn 36, Appendix A, Table A.2:
el" ti"
L ---=--- =~-~ (s > a and s > b),
{ zF }
by showingthat 11ft- i!"1/0 < M00n an interval (O,T).
3. (a) Use Theorem 11 to derive the transfonn (3), Seco23, of M(c,t).
(b) Apply fonnula (7), Sec. 23, to obtain the transfonn 68, Appendix A, Table A.2,
of the half-wave rectification of M(k,t).
j I (e) Use fonnula (9), Seco23, to verify that the transfonn of the full-wave rectifica-
tion of M(c,t) is l/s.
4. Sketch the graph of the periodic function F for which F(t) = t when -1 < t < 1
and F(t + 2) = F(t) for all t, and show that the transfonn of F is
l l
f(s)=--~ (s > O).
i S2 s smh s
I
,
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 25 73
5. Let F be the periodic function such that F(t) = t when 0< t < 1 and F(t + 1)= F(t),
and let G be this antiperiodic function: G(t) = t whenO< t < I and G(t + 1) = - G(t).
Show graphically the functions F, G, and the half-wave rectification G1 of G, and
obtain the transforms
le-s
(s > O),
(a) f(s) = SI - s(1 - e-
I 2 I ~ (s > O),
(b) g(s) = SI - (SI+ s) I + e-s
f(s)
(e) gl(S) = I + e-s (s > O).
6. Use Theorem 12 to find the inverse transforms tabulated below, where a, b, and e
are distinct constants.
f(s) F(t)
I -e"'-e'"
(a) (s - a)(s - b) a-b
s ae'" - b"
(b) (s - a)(s a-b
-1 (b - c)e'" + (e - a)e'"+ (a - b)e<'
(e)
(a - b)(b - c)(c - a)
7. Use Theorem 12 to find the inverse transforms of the following functions.
(a) f(s) = . .
4s + I
. ... .. . Ans. F(t) = e'/z- e-'/z + e-' - l.
s s
(b) f(s) = SZ- kZ' (e) f(s) = SZ + kZ'
3sz
(el)f(s) =.. .. .
We now consider partial fractions for the case in which the polynomial
q(s) contains a repeated linear factor (s - a)n+ 1. We write
cp(s) Ao AlAr An
(2) n+l =-+
(s-a ) s-a (s-a )2 +... + (s-a )'+1 + ... + (s-a )n+l +h(s),
where the numbers Ar are independent of s, and h(s) is the sum of the partial
fractions corresponding to the remaining factors of q(s). It follows from
Eq. (2) that
This equation can be simplified by recalling the formula for the derivative
of order n of the product of two functions u(s) and v(s), namely,
dn n ,
-(uv) = L n. u(n-r)(s)v(r)(s).
ds" r=o(n - r)!r!
When u = cp(s)and v = tT',then arv/as' = t'tT' and Eq. (5) reduces to the
form
;,
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 26 75
As an example, if
1
f(s) = (s - l)(s - 2)2'
then the term in F(t) correspondingto the factor s - 1 is . To correspond
with the factor (s - 2)2, we have 4>(s)= (s - 1)- 1 and 4>'(s)= - (s - 1)- 2
so that 4>(2)= 1 and 4>'(2)= - 1. In view of formula (8) the term in F(t)
is (-1 + t)e2t. Hence
F(t) = + (t - l)e2t.
Since the number a may be imaginary and since a factorization of
every polynomial into linear factors, real or imaginary, exists, Theorems
12and 13providea systematicway offinding inversetransformsof quotients
of polynomials in all cases where the factors of the denominator can be
determined. Ir, however,imaginaryfactors are present,the results are given
in terms of imaginary functions. The reduction of the results to real forms
is sometimes tedious. To obtain the real form of the inverse transform
directly, and to observe the character of that function in general, we may
proceed as follows.
26 aUADRATIC FACTORS
In this section we assume that the polynomials p(s) and q(s) have real
coefficients.The imaginaryzeros of q(s) then occur in pairs, each pair
consisting of some complex number a + ib and its conjugate. The corre-
sponding linear factors of q are
(1) ql = s - a - ib, q2 = s - a + ib (b ::/=O),
where a and b are real numbers. The product of those factors is the real
quadratic factor (s - a)2 + b2.
,~
Let q have the factors (1), not repeated,and let <Pland <Plbe the
quotients of polynomials obtained by removing the factors ql and ql in
turo from the fraction p(s)/q(s); thus
(3)
This is the component of F corresponding to the quadratic factor
(s - a)l + b1. We shall represent it in terms of real-valued functions.
The rational function <Pl(S)is continuous at the point s = a + ib.
In a neighborhood of that point excluding the point itself
(4) <Pl(S)= (s - a - ib)f(s).
Similar1y, <Plis continuous at the point a - ib and
(5) <P2(S) = (s - a + ib)f(s)
throughout a neighborhood of that point with that point deleted. From
elementary properties of complex conjugates we can see that the conjugate
of f(s), a rational function with real coefficients, is f(S). Consequently
<Pis) = (s - a + ib)f(s) = (s - a - ib)f(s) = <Pl(S).
The factor in brackets is twice the real part of the first term, 2 Re[<P
1(a+ ib)eib'J.
So if we let r 1 and 81 represent polar coordinates of the point representing
the complex number <Pl(a+ ib),which is never zero,
(7) <Pl(a+ ib)= rleiOI = rl(cos81 + isin(1),
we can write the component G in the real form
(8)
G(t) = 2eQtr1
cos (bt + (1) (b =FO,rl > O).
Ir a = O, the component G(t) is the simple periodic function
2rl cos (bt + 8), and if a < O, it is a damped periodic function. Those
components, corresponding to cases a ~ O,represent stable oscillationsin
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 26 77
Th A..
( 1 2)
' -4 + 4i 1 - i .J2 -i,,/4
en '1'1 - + I = 2i(4i) = ~ = le
and G(t)= .J2e-t cos (2t - nJ4)= e-t(cos 2t + sin 2t). After adding the
term corresponding to the linear factor s + 1,we find that
F(t) = e-t(cos 2t + sin 2t - 1).
In case q contains the square of the quadratic factor (s - a)2 + b2,
we write
(9) o exp
.j;.e../SY1(s) [- (2).
- ) J ).2'
By first adding corresponding members of equations (8) and (9), then
substitutingx fort0). -
1/)" we can write
2 .fieo 2
.fi
.
2.j;.e-Isy f[ (
(s) = .fi o exp - 2).
1
1
- J (2 +
)
1
).2 ) d)'
~
eo
2
=v r: fs -eo
exp (_X2) dx = 2 -.
S
.--
Therefore
L -l l -k-.s = - k
-e
{s } 2.;:;c o I
' e -k2/( 4f) . -i..- d. = - 2
fi f a:>
e -J.2 dJI.
k/(2../i)
1
Therefore
where erf(x) is the error function defined in Seco 16. Equation (10) can be
written
o,
FURTHER PROPERTIES OF THE TRANSFORMATION [SECo 27 81
PROBLEMS
1. Find the inversetransformstabulated below:
f(s) F(t)
s+a a -b -b' a- e
a - b -c'
(a) (s + b)(s + e)2 (b- e)2e + [ b - et- (b- e)2e
J
s+3
(b) (2s + l)(s2 + 2s + 2)
2b4
(e) (S2+ b2)
s2-b2
t cos bt
(d) (S2 + b2)2
2Ss2
(St - 8)e-' + 6 sin 2t + 8cos 2t
(e) (s + 1)2(S2+ 4)
2. Without finding F(t), determine whether the oscillations of that function are stable
or unstable when its transform is
s
(b) f(s) = .- . -~~. Ans. Unstable.
.
6. Solve for Y(t):
-
(a) (t t2)Y"(t)+ 2Y'(t) + 2Y(t) = 6t, Y(O) = Y(2) = O.
Ans. Y(t) = t2 - 8t-l(t - 1)380(t - 1).
. (b) Y(t) + 2 J~ Y(x) cos (t - x) dx = ge2'.
Ans. Y(t) = Se2'+ 4e-' - 6te-'.
7. Draw the graph of the ramp function Y such.that
Y(t) =t- Y(t - 1) and Y(t) =O when t ;;;! O,
and show that Y(t) = t - n when 2n;;;!t ;;;!2n + 1, Y(t) = n + 1 when 2n + I ;;;!
82 SECo 27] OPERATlONAL MATHEMATICS
(x > e),
~(x) = focof(x,t) dt
as x - 00, is the same as the integral ofthe limito (a) Letf(x,t) be continuous over each
rectangle e ;;:; x ;;:; e, O ;;:; t ;;:; T, except possibly for finite jumps across a finite number
=
oflinest ti in each rectangle,and (b)let the above integralconvergeuniformlywith
respect to x when x> c. Also, (e) letfhave a limit
F(t) = x~co
lim f(x,t)
uniformly with respect to t(t ,p ti) on each interval O < t < T; that is, for each positive E
there is a number N. independent of t such that
IF(t) - f(x,t)1 < E whenever x> N. (O < t < T),
and (d) let f~ F(t) dt exist. Then prove that
lim q(x) =
X-CIO lo
co F(t) dt.
Note: The difference of the last integral and q(x) can be written
~dt
o t + x
. is not uniformly convergent with respect to x on the half line x > l (cf. Probo 10).
Show that lim q(x) = rr/2 while the integral of the limit of the integrand, as x
is zero. x~co
00, -
12. If F(t) ~ G(t) when t > O,then f(s) ~ g(s)for every real value of s for which the
transformsf and g existo Prove that property, and iIIustrate it with some particular
functions. Note that the case F(t) = 2 sin t, G(t) = sin 2t shows that the converse of
the property is not always valido
FURTHER PROPERTIES OF THE TRANSFORMATlON [SEC. 27 83
13: Inilial-value theorem Ir F is continuous over the half line t ~ O except possibly
for a finite number of finitejumps, while F' is sectionally continuous over each bounded
interval there, and if F and F' are of exponential order, then the limit of the product
sf(s), as s - 00, exists and equals the initial value of F:
s~oo
lim sf(s) = F(O).
17. Illustrate the modified initial-value theorem presented in Probo 16 in case F(t)
is (a) sin kt; (b) at + bt2; (e) 1 - cosh t.
18. Ajinal-valueIheorem Let both F and F' be bounded over the half line t ~ Oand
continuous there except possibly for a finite number of finite jumps; also let 1F'(t)1
be integrable from t = O to t = oo. Then
lim sf(s)
3-+0
= t-Q'J
lim F(t).
Prove that theorem. In the formula for L{F'} note that our conditions ensure the
continuity of that Laplace integral with respect to s when s ~ O (Sec. 15).
19. Illustrate the final value theorem (Prob. 18) in case (a) F(t) = A + Be-';
(b) F(t) = So(t - to); (e) F(t) = t - (t - I)So(t - 1).
20. Generalizedconvolulion For a function F(t;r) of two variables, defined over the
quadrant t > O,-r > O,letf(s,-r) denote its transform with respect to t. Using the same
parameter s, let](s,s) denote the iterated transform:
(t ~ 0)0
24. When. is fixed and positive, let Z(t,.) denote the solution of the differential
equation with constant coefficients (independent of t and -r)
m dn
sponding equation for the iterated transform z(s,s) of Z(t,.) (Probo 20) and compare
it with the equation for the transform y(s) of Y(t), where
m
Y(t) + L any,n)(t) = F(t),
n= 1
Y(O) = Y'(O)= o o o = y'm- 1)(0)= O,
to show that y(S) = SZ(S,S).Thus deduce the following formula for the solution of the
problem in Y in terms of the solution of the simpler problem in Z:
Y(t) = io
'
-Z(t - " .)d..
oot
25. Use the formula in Probo 24 to find the solution of the problem
Y"(t) - Y(t) = F(t), Y(O)= Y'(O)= O,
in the form Y(t) = J~F(r)sinh(t - Note that - F(-r)is a particular solution of
-r)d-ro
the differential equation in Z(t,-r)o
3
Elementary Applications
..x Fig.12
obey Hooke's law, SOthat the force exerted by the free end is proportional
to the displacement of that end. The factor k of proportionality is called
the spring constant. We also assume that the mass of the spring can be
neglected in comparison with the mass m and that no frictional forces or
external forces act on m.
Let X denote the displacement of m from the position of equilibrium;
that is, let the origin O denote the position of m when the spring is not
deformed. Then according to Newton's second law of motion,
d2X
(1) (k > O).
m dt2 = kX -
Let Xo denote the initial displacement and Vothe initial velocity, so that the
function X(t) satisfies the conditions
(2) X(O) = Xo, X'(O) = Vo.
xowo. WoXo
where Wo = A
m
-, tan ex= -, Vo smex = y,Wo 2Xo 2 + Vo2'
.
ELEMENTARY APPLICATIONS [SEC. 28 87
Fig.13
where the eoefficient of damping e is a po sitive constant. Let the mas s start
from the origin with initial velocity Vo:
X(O) = O, X'(O) = Vo.
The equation in the transform x(s) becomes
ms2x(s) - mvo = - kx(s) - esx(s)
or, if we write 2b = e/m, and again write illo2 = k/m,
Ir b2 < illo2, that is, if the coefficient of damping is smalI enough that
e2 < 4km,
then the formula for the displacement is
(6)
(7)
We can see from this formula that the mass m moves in the direction of Vo
until the time t = l/b, then reversesits direction and approaches O as t tends
to infinity.
When e2 > 4km, a similar motion of the mass takes place. The dis-
cussion of this case and the case of other initial conditions is left to the
problems.
The mathematical problem treated in this section can be interpreted
also as a problem in electric circuits. This welI-known analogy between
problems in vibrations of mechanical systems and electric-circuit theory will
be observed for other problems in this chapter. NaturalIy, the notation and
terminology differ in the two types of problems.
In the electric circuit shown in Fig. 14,let Q be the charge accumulated
in the capacitor e at time t, and 1 the current in the circuit, so that
(8) I(t) = Q'(t).
./
88 SECo 29] OPERATIONAL MATHEMATICS
Fig.14
This equation requires that I(t) be positive when Q(t) is increasing. Ir the
positive sense of ftow I(t) of positive charges is taken in the clockwise
direction in Fig. 14, then Q(t) measures the charge on the upper plate of the
capacitor.
The circuit has a resistance R and a coil of inductance L. Since the
sum of the three voltage drops in the circuit is zero in this case of no impressed
voltage, then
1
(9)
LI'(t) + CQ(t) + RI(t) = O.
The system of first-order differential equations (8) and (9) can be used
directly to determine I(t) and Q(t) in terms of initial values of those functions.
By eliminating I(t) from the two equations, however, we see that
1
(10) LQ"(t) + RQ'(t) + CQ(t) = O.
Except for the notation used, this equation is the same as Eq. (4). When the
resistance is negligible, R = O,the equation reduces to our Eq. (1).
The initial conditions in the electrical problem can be made the same
as those in the mechanical problem. For example, if the capacitor has an
initial charge Qo and if the initial current is 10, then
which are the same as the initial conditions (2) in our first mechanical
problem.
~
m
Ftt)
Fig.15 O~x
wo=A.
Then we can write
xos + Vo 1 1
(3) x(s) = s2 + Wo2 + -m f(s)---z- ..
s + Wo
Hence the displacement for any F(t) can be written, with the aid of the
convolution, as
~ sin wot + ~
(4) X(t) = Xocos wot +
Wo mwo o f
' sin wo(t - -r)F(r) d-r,
then
f(s) = F{~ - exp(;tos1
90 SECo 29] OPERATIONAL MATHEMATICS
1 - exp(-toS)
(6) x(s) = FO
[
m S(S2 + (02) S(S2 + (02) ].
Now
where l5(t)is the unit impulse symbol (Sec. 13). The constant M o represents
the increase in momentum of the mass m at the instant t = O.
X(t)
Fig.16
1
..
ELEMENTA.RY APPLICATIONS [SECo 30 91
Xt
lO, 2FO)
"
o
Fig.17
Mo.
(9) X (t ) = -SInWoto
mwo
satisfies the condition M( +0) = Mo. Ir the mass started from rest, then its
momentum jumped to the value M o at the instant t = 00
30. RESONANCE
and, if w ::f:wo,
l Fow .
(2) X(t) = Xocos wot + -Wo
[Vo +
m (w
2
- Wo
2
)J
SIn wot
- Fo
20
m(w - Wo2)SInwto
92 SECo 30] OPERATIONAL MATHEMATICS
That is, the motion is the superposition of two simple harmonic motions,
one with frequency Wo and known as the natural component of vibration,
and the other with frequency w which is called the forced component of the
vibration. Note that the natural vibrations are not present in case
Fow
Xo = O, Vo= 2 2 '
m(Wo - w )
(3)
The presence of the repeated quadratic factor in the denominator here shows
that X(t) will contain an unstable component (Sec. 26) having the form ofthe
product of t by a cosine function. In fact,
l Fo . Fo
(4) X(t) = Xo cos wot + ---"2vowo
+ _ SIRwot - - t cos wot.
Wo ( 2m) 2 mwo
In view of the last term here, the amplitude of the oscillations of m increases
indefinitely.
In this case the force F(t) is said to be in resonance with the system.
We note in particular that if Xo = O and Vo = - Fo/(2mwo) the resonance
type of motion reduces to
Fo
X(t) = .-,.2mwo
- t cos wot,
--
-- --
X(t)
"'"
"'".....-
.....---
Fig.18
ELEMENTARY APPLlCATIONS [SEC. 30 93
PROBLEMS
Note that the vibration of the mass is unstable under that periodic 2-in. oscillation of
the support, and that within only four cycles of the oscillations (t = 8n/wo < 2 sec),
displacements of the mass mount to approximately l ft.
94 SECo 30] OPERATIONAL MATHEMATICS
6. If the coefficient of damping for the oscillator shown in Fig. 13 is so large that
e2 > 4km and if X(O) = Oand X'(O) = vo,(a) show that
X(t) = :
V. et
smh (at) exp ( - 2m') wherea =
p-4km
2m .
(b) Find X'(t) and show that the mass m moves in the direction of vo until the
instant
11 e+2am
t = 2a og e- 2am'
when it turns and approaches the origino
7. Let the mass m in the damped oscillator shown in Fig. I3 be released from rest
with an initial displacement X(O) = xo, and write 2b = e/m. (a) If e2 < 4km and
WI = (W02 - b2)t, show that
. b 1t
wh eresm(X=-, O;;(X< "2;
Wo
(b) if e2 = 4km show that
X(t) = xoe-bt(l + -bt),
and hence that the mass never moves across the origino
8. When the force on the undamped oscillator shown in Fig. 15is the periodic function
F(t) = Fo sin (wt + 8),where 8 is a constant, show that resonance occurs when w = Wo.
9. When t = O. let the current in the circuit shown in Fig. 14 be zero and let the
capacitor have a positive charge Qo. If R > 2jL!C, derive the formula
Rt . (Xt t
2Qo 4L
l(t)= --exp
(XC (
--
2L)
smh-
2L (
2
(X= R --
C)[ ]
for the current. Show that l(t) ;;O and 1(00) = O.
10. The current 1, and the charge Q on the capacitor in the circuit shown in Fig. 19,
are functions of t that satisfy the conditions
L-dI + -Q + RI = E 1(0)= O,
dt C o' Q = f~ I(T)dT,
when Q and 1 are initial1y zero, where t is the time after cIosing the switch K, and the
electromotive force Eo is constant. If b = R(2L)-1 and Wl2 = (LC)-l - b2 > O,
derive the formula
1 = EOe-bl sin wlt.
WIL
Fig.19
ELEMENTARY APPLICATIONS [SEC. 31 95
When an external force F(t) acts parallel to the X axis on the mass in the
damped oscillator shown in Fig. 13,the equation of motion becomes
(1) mX"(t) = - kX(t) - eX'(t) + F(t).
Ir X(O)= X'(O) = O,the solution of the transformed equation is
1 1
(2) x(s) = -m f(s)
It follows from Theorem 14, Seco26, that X(t) consists of terms of the types
A cos (wt + e) and Be-bt cos (wI + el), where A, B, e, and el are constants.
ConsequentIy the component of oscillation with frequency WI is nearIy
damped out after a sufficiently long time, and the periodic forced component
of vibration
(4) Xp(t) = A cos (wt + e) (A > O)
remains. Its amplitude A is, according to Seco 26, 214>1(iw)1 where
4>1(S) = (s - iw)x(s). Thus
(5) A = IFol 1 .
m I(iw+ b)2 + wl21
The square of the absolute value in the denominator can be written as
IW02 - W2 + 2ibwl2,which is (W2- W02)2 + 4b2w2,and by expanding and
completing the square, we can write that polynomial in w as (W2 + 2b2 -
Wo 2)2 + 4b2(wo 2 - b2). Thus
IFol
(6) A= - ~ u~ ~-~ ~_..
(8) A = ~= IFol
, 2bmool eOOl'
We can seethat the expression(3) for X(5)cannot contain a repeated
factor of the type (52 + (2)2 in its denominator unless e = Oso that b = O.
Hence no value of the frequency w of our periodic force will induce a com-
ponent of X(t) of type t cos (wt + 9) as it did for the undamped oscillator.
The amplitude A of the forced component of vibration is still given by
formula (5) when the force Fosin wt is replaced by a force Fosin (wt + IX)
(Prob. 3, Seco32).
In case e2 ~ 2km, it tums out that the amplitude A of the forced
vibration tends toward an upper bound IFol/kas 00tends to zero. When w
is small and t is large enough that wt is near an odd multiple of n/2, the
magnitude of the force Fo sin wt is then near IFol. Under static conditions
the force IFol would displace the end of the spring a distance IFol/k. The
formula for X(t) when e2 = 2km, found in Probo 4, Sec. 32, shows that when
w is small the mass may oscillate slowlywith an amplitude near IFol/k.
F!tl
mI
m
A k kl
I
OL ..x
Fig.20
ELEMENTARY APPLlCATlONS [SECo 32 97
The transforms x(s) and XI(s) of X(t) and X l(t) therefore satisfy the simul-
taneous algebraic equations
Foro
(ms2 + cs + k + k1)x(s) - k1xi(S) = -y- 'l'
S +ro
unless kdml = ro2,in which case the numerator ofthe above fractioh cancels
with the.factor in the denominator, leaving
1
(4)
x(s) = FOromlp(s)'
The inverse transform X(t) of the function (4) represents a damped
oscillation of the mass m if each of the four roots, real or imaginary, of the
98 SECo 32] OPERATIONAL MATHEMATICS
equation p(S)= Ohas a negative real part (Secs.24 to 26). Now the poly-
nomial p(s) is the determinant of this system of homogeneous equations in
x and Xl:
(ms2+ es + k + kl)x - klxl = O,
(5)
-klx + (mls2 + kl)XI = O.
When the determinant vanishes, that system is satisfied by numbers x and
Xl' both differentfrom zero.
Let s be any one of the four roots of the equation p(s) = O and let
z,z1 denote a pair of nonvanishing roots of the system(5)that corresponds
to that value of s, where z and z1may be either real or imaginary numbers.
We now write X = z and Xl = Zl in Eqs. (5), then multiply the members of
those equations by the complex conjugates Zand Zl' respectively,and add
to obtain the equation
(6) (mzz + mlzlzl)s2 + ezzs + B = O,
where B = kzz + kl(zz - ZZI- ZIZ + ZIZ)
= klzI2 + kllz - zl12 > O.
Ifwe write A = mlzI2 + mllzl12,then Eq. (6)becomes
(7) As2 + elzl2s + B = O,
where the coefficients A, clzl2, and B are positive numbers that depend on
the value of s. However, the number s is given in terms of those positive
coefficients by the quadratic formula
1
(8) s = 2A (- clzl2 :f: .Je21z14 - 4AB).
Whether the radical here is real or imaginary, it follows from formula (8)
that the real part of s is negative. This completes the proof that the oscillation
X(t) is damped.
Thus the forced component of the vibration of the main mass m is
eliminated by the system mI' kl, if the natural frequency of that system
coincides with the frequency of the exciting force:
view of Eq. (9), the absorber is designed for a fixed frequency ro of the exciting
force Fo sin rot.
By solving the above equations for XI (s),we can see that the mass mi
has an undamped component of vibration of the type (3).
PROBlEMS
1. If the initialconditionsX(O)= X'(O) = Oused in Seco31 for the damped oscillator
with exciting force F(t) are replaced by the conditions X(O)= XO,X'(O)= vo, show
that the additional terms in the formula for X(t) represent damped oscillations.
2. When F(t) = FoD(t)and X(O)= X'(O) = Ofor the damped oscillator considered in
Seco31, where e2 < 4km, show formally that
Fo b.. e 2 k 2
X( t) = -e- s!nrolt = -,rol =- - b ,
mrol (b 2m m )
and note that the jump in the momentum of the mass m at the instant t = Ois Fo.
3. Show that formula (6), Seco31, for the amplitude A of the forced component of
vibration for the damped~scillator is valid (a) when the exciting force F(t) is Focos (J)f;
(b) when F(t) = Fo sin (rot + IX).
4. If e2 = 2km for the damped oscillator considered in Seco31, and if F(t) = Fosin wt
and X(O)= X'(O) = O,show that
Hence when ro is small and rot is large, show that X(t) is approximately (Fo/k)sin rol.
5. In Seco31, if e2 = 4km, F(t) = Fosinrot, and X(O)= X'(O)= O,show that X(t)
has the form A cos (rot + IX)+ (B + Ct)e-bt, and when ro is small that IAI is near its
upper bound IFol/k.
6. Let the oscillator describedin Probs. 1 and 2, Seco30,for whichk = Ib/ft and t
t
mg = lb, be subject to a damping force - eY'(t) where e = f4Ib/ft/sec. (a) If the
upper end of the spring is fixed so that Z(t) = O,and if Y(O)= Yoand Y'(O)= O,show
that after about nine cydes of oscillation of the mass the amplitude of the displacements
Y(t) is reduced to approximately o.llyo!. (b) When the upper end is oscillated so that
Z(t) = Zosin rot, show that resonance occurs if ro = ror where ror = )159.5 rad/sec,
a frequency slightly less than roo(cf. Seco31). (e) When ro = rorin part (b),show that the
amplitude A ofthe undamped forced compone~t ofvibration exceeds 121zol.
7. Let the two masses in the system treated in Seco32 have arbitrary initial displace-
ments and velocities. When condition (9) is satisfied by the eIements of the absorber,
show that the vibration of the main mass again contains no undamped component.
8. (a) If the exciting force Fosin rot in the system of Seco32 is replaced by the force
Fosin (rot + IX),where IXis any constant, show that when condition (9) is satisfied the
vibration of m is again entirely damped.
100 SECo 32] OPERATIONAL MATHEMATlCS
(b) If the exciting force is replaced by Al sin rolt + A2 sin ro2t,where the A's and
ro's are constants and rol f= ro2' show that the values of mi and kl cannot be adjusted
so that alI undamped vibrations of m are absorbed.
9. Let the force F(t) be removed from the mass m in Fig. 20 and let the end A of the
spring k be moved horizontalIy so that its distance from the walI is F(t)/k. Show that
the differential equations of motion of m and mi are then the same as in the original
problem.
10. A system of two masses and two springs in series, with no damping, is shown in
Fig. 21. The weights of the two masses are mlg = 8lb and m~ = 2lb (g = 32 ft/sec2).
The spring constants have the values kl = 24lb/ft and k2 = 8lb/ft. If a periodic force
F = Focos 2nct acts on m2' find the frequencies c (cycles per second) of the force for
which resonance occurs in the vibration of mi' Assume zero initial displacements and
velocities. Also show tltat resonance occurs in the vibration of m2for those same values
of e. Ans. e = 4n-I, 4 J3 n-l cycles/sec.
Fig.21
11. For the system shown in Fig. 21, if mi = m2 = ls, kl = ?s, k2 = ?s, and if both
masses start from rest in their equilibrium positions, find the formula for the displace-
ment X l(t) of mi when the force F(t) on m2 is arbitrary.
Ans. X l(t) = (6 sin t - J6 sin t J6) * F(t).
12. InitialIy the two masses shown in Fig. 22 are at rest and the spring has its natural
length. ThenaconstantforceF = Fo(t > O)actsonml' Ifthesystemisfreefromdamp-
ing, find the formula for the displacement of m2 and note that the displacement consists
of a simple harmonic motion superimposed upon a uniformly accelerated motion.
Also show that ifFo > O,the spring is always compressed by an amount proportional to
1 - cosrot,wherero2= k/ml + k/m2'
Ans. 2w2(ml + m2)X 2(t) = Fo(ro2t2 - 2 + 2 cos rot).
I Fig.22
13. If the force F = O in the system shown in Fig. 22 and if the masses are initialIy
released at rest from positions Xl = al and X 2 = a2' show that the frequency of
vibration of the masses has the value ro = (k/ml + k/m2)t. When a2 = -alml/m2'
I show that XI = al cos rot and X 2 = a2 cos rot and describe the vibration.
I 14. Let the force in the system shown in Fig. 22 be the periodic force F = Fosin rot,
where ro2 = k/m2' and let a viscous dampingforce -eX(t) act on mi' Find the steady-
state vibration of m2 and note that it does not depend on c. Ans. - Fok- 1sin rot.
! 15. In Fig. 23 the end E of the first spring has a periodic displacement Y = A sin rot.
i The two springs are identical. Find the value of ro for which this damped system is in
resonance, when e2 < 4km. Ans. 2m2ro2= 4km - e2.
I
1
[
ELEMENTARY APPLlCATlONS [SEC. 32 101
Fig.23
E
~
I
I
~y
k
I
~X
m
16: In Fig. 24 the two masses are unit masses (m=:=1)and the two springs are identical.
Initially, the springs have their naturallength, the first mass has velocity vo, and the
second mass is at rest. Find the undamped component of vibration of those masses.
Ans. !(vo/.jk) sin t .jk.
Fig.24
17. For the undamped system of two equal masses and three identical springs shown
in Fig. 25, the initial conditions are XI(O) = al' X2(O)= a2, X'I(O)= X(O)= O.
(a) If lall * la21,show that the components of vibration of each mass have frequencies
-fl0 and J3k/m. (b) If a2 = al' show that both masses vibrate in unison with
frequency -fl0. (e) If a2 = - al' showthat the massesvibrate in oppositedirections
with frequency J3k/m.
f I
x(s) =~ mls2 + es + k
S2 + W2(mls2 + es + k)(ms2 + es + k + k) - (es + k)2
for the transform of the displacement X(t) of m, then show that the amplitude A of the
forced component A cos (wt + O)of X(t) is given by the formula
A2 (mw2 - k)2 + e2w2
F? = [(mw2 - k)(mw2 - k) - mkw2]2 + e2w2[(m+ m)w2 - kp.
(A graphical examination of A2 as a function of W2 would indicate a value of e for
whichthe rangeof valuesof A2willbe as smallas possiblefor all w.)
See Den Hartog, op. cit., pp. 93 ff.,ror a detailed discussion.
102 SECo 33] OPERATIONAL MATHEMATICS
Fig.26
(3)
~l QI(t) = LII;(t).
Ll
Fig.27
.1
ELEMENTARY APPLlCATIONS [SECo 33 103
PROBLEMS
1. If 11(O)= 12(0)= Oin the network shown in Fig. 28, while the capacitors have the
same initial charge Qo, show that
wherero = (CL)-t.
e
11 R 12
Fig.28
2. Ifl1(0) = 12(0) = Oin thenetwork shown in Fig. 28 and ifQI(O) = alandQ2(0) = a2,
find the undamped component 1 (t) of the current 11(t), the steady-state current through
the first inductance coil. Ans. 1 = -!(al + a2)wsin rot,ro = (CL)-t.
3. Show that the network indicated in Fig. 28 represents an electrical analog of the
mechanical system shown in Fig. 24 where QI and Q2 correspond to the displacements
of the two masses.
4. In Fig. 29 the currents 11(t) and 12(t)and the charge Q(t) are initially zero.
(a) Show that the current 11(t) is unstable under the impressed voltage V =
Vosin (rot + ex)if ro2 = (L1q-l + (L2Q-I.
(b) Show that the system indicated in Fig. 22 is a mechanical analog to the
network here with X I(t) and X 2(t)corresponding to Ql(t) and Q2(t),where Q'l(t) = 11(t),
Q(t) = 12(t), and Q(t) = Ql(t) - Q2(t).
Fig.29
ELEMENTARY APPLlCATIONS [SECo 33 105
5. A prescribed current l(t) is supplied to the network shown in Fig. 30. The initial
values of 1I(t)and Q(t)are zero. Use the following notation: V(t) is the resulting potential
drop from the junction A to the junction B, m02 = (LC)-I, 2b = (RC)-I, and
ml2 = m02 - b2, given that b2 < m/.
(a) Show that the transfer function from input l(t) to output 1(t) is
y( s)
- m o2
- S2 + 2bs + mo2'
thus thatI (t) = l(t) * Y(t), where Y(t) = m/ml -Ie-br sin mi t.
(b) Find V(t) when l(t) = lo, a constant. Ans. V(t) = 10(Cm)-e-br sin mlt.
Fig.30
6. In a simple LC circuit with inductance coil and capacitor in series, let the impressed
voltage be a periodic function, with period T, of the type
N
V(t) = ao + L (a. cos nmt + b. sin nmt),
n=1
where m = 2n/T. Show that the current l(t) in the circuit is unstable ifthe frequency m
of V(t) has any one ofthe values
l 1
m- - (m = 1,2,..., N),
-myU
providedthat amand bmare not both zero.
7. In the network shown in Fig. 31 all currents and chargesare zero at the instant
t = Owhen the switch K is closed. If the impressed voltage is given by the equation
V = Vosin mt, where m2 = 2(LC)-I, find the steady-state value 12(t) of the current
12(t)and note that it is independent of the value of the resistance R.
Ans. 12(t) =- VoCmcos mt.
Fig.31
8. In the system shown in Fig. 25 apply an exciting force to the first mass and viscous
damping to the second mass. Show that the resulting system is a mechanical analog
ofthe network shown in Fig. 31.
106 SECo 34] OPERATIONAL MATHEMATICS
9. Initially the currents and the charge on the capacitor are zero in the network shown
in Fig. 32. The voltage drop from Al to B1 caused by the mutual inductance of the
two coils is MI;(t), and the drop from A2 to B2 caused by that mutual inductance is
Ml(t),whereM2 < L1L2. Theimpressedvoltageisgivenbytheequation V = Vocoscot.
When the values of C and L2 are adjusted so that L2C = co-2, show that the current
Il(t) in this idealized resistance-free circuit has a simple periodic variation at all times
with a frequency greater than co.
Fig.32
10. Show that the network in Fig. 33 represents an analog of the damped absorber
shownin Fig. 26 if initial currents and charges are zero, where charges Q and Q - Q
correspond to displacements X and Xl and the voltage V corresponds to the force
Fo sin cot.
Q
~c 1
Ql
R 11
Fig.33
(1)
ELEMENTARY APPLICATIONS [SECo 34 107
Here F(t) is the Fourier cosine transform (Chap. 10) of (X2 + a2)-1.
Note that F is bounded: F(t)1;;;; f~ (X2 + a2)-1 dx = n/12al; also,
-
F( t) = F(t).
Let us first proceed formalIy to transform with respect to t and
interchange the order of integration with respect to x and t:
eo dx eo s dx
(2) I(s) = fo L{costx}
x +a
2 2 = fo
eo
s 1 1
= - . f
o ( x2 + a2 - x2 + S2 )
d
x
n
= 2a s +
1
a
if a > Oand s > O. Thus the integral has the value
n
(3) F(t) = - e-al (a > O,t E:;;O).
2a
TOsee that the formal step (2)is sound, we first note that, whenever
T and s are positive constants,
T T
eo cos tx eo 1
(4)
o fe-sI
ox+a
2f 2 dx dt =
ox+a
2 2
o f
e-sI cos tx dt dx, f
because the absolute value of the entire integrand does not exceed
(X2 + a2) - 1, a function independent of t whose integral from x = O
to x = 00 exists. The integral F(t)e-st therefore converges uniformly
with respect to t when t E:;;O by the Weierstrass test and (Sec. 15) the
interchange of order of integration with respect to x and t made in
step (4)is valido Also, F is continuous (t E:;;O)and bounded so, according
to the theorem on uniqueness, if L{ F} is given by formula (2), then F
must be the exponential function (3).
The final integral in Eq. (4) can be evaluated by integration by
parts, and the equation can be written
(5)
fOTe-SIF(t)dt = Leog(x,T)dx,
oX+s
2 2 =_
2
;
s
hence F(t) = n/2. In view of Eq. (6), F(t) is an odd function: F( - t) =
- F(t), and F(O)= O. Therefore
n n
(8) F(O) = O.
F(t) = 2" (t > O); F(t) = -2" (t < O);
The Weierstrass test cannot be used here to justify step (7).
In fact, if our formal result (8) is correct, then F is discontinuous at the
origin t = O,and the integral (6) cannot converge uniforrnlyover an
interval that includes the origino
Now the function S(r) = sin rlr, r =1=
O, S(O) = 1, that is,
r2 r4
(9) S(r)= 1 - -3! + -5! - . . . (-oo<r<oo)
is continuous, and IS(r)1~ 1, for all real r. Note that the sum of the
alternating sees (9) is positive and not greater than unity when
Irl ~ 1, and if Irl > 1, then Isinrlrl < 1. To see that the integral (6)
exists, we note that its integrand tS(tx) is continuous in t and x and,
if to, xo, and XI are positive constants, an integration by parts shows
that
fo -
x
dx+--cos txo
txo f XI)
a)
tx
y- dx (t > O),
x tx
r dx
l 1 2
IR(t X )I ::;;-
::;;- < when X> X(. +-
, - tX tX - toX
Thus integral (6) converges uniformly with respect to t
(t ~ to > O),and wecanwrite, with the aid of anelementaryintegration,
T co 1
T
f lO
e-SIF(t) dt =
f f
o X
-
lo
e-sI sin tx dt dx
ox+s
-y- (5 > O),
The uniform convergence of the integral SO' h(T,x) dx and the fact that
h(T,x) -. Oas T -. 00, uniform1y with respect to x, foIlow from con-
ditions (11). Thus as T -. 00, that integral vanishes, and Eq. (10) leads
to step (7), so the bounded continuous function F has the transform
tn/s, and therefore F(t) = tn when t > O. This completes the proof.
. IncidentaIly, we have shown that Si(00) = tn, for when t > O,we
found that F(t) = !1t, and Eq. (6) can be written
co sin r . 1t
(1) E1(t) =
f I
=-dy =
y f 1
~dx
X
(t > O).
(2) =-
s) s f 1
- - - dx = -log - (s > O).
J,1 x(x + (X X + s) s x+s ] 1
The proof that expression (2) represents L{El(t)} is left to the problems.
The basic form of transform 100, Appendix A, Table A.2, is, therefore,
1
(3) L{E1(t)} = -log(s + 1) (s > O).
s
In Seco 22 we found that the transform of the sine-integral function,
Si t, is S-1 arccot s. Now let us wrte the transform of the cosine-integral
function
(4) Ci t =- oocos r dr = - (00cos tx dx (t > O).
f.I r J1 x
An integration by parts, first over a bounded interval (l,X), shows that
. sin t 1 oo sin tx
(5) Clt=---
t t f
1
-
X2
dx (t > O);
cost
,-dr
sinr
+ smt
. oocos r
-dr
1o r f.I r
oo sin r oo dr
= cost fo -drr, -
f (sinrcost - cosrsint)-
r
when t > O. By replacing Si ((0) here by tn (Sec. 34) and substituting tx
for r - t, we can write the last expression in the form
1t oo sin (r - t) 1t oo sin tx
(7) -cos t -
2 f,r dr = -cos t -
2 f -dx.
ox+l
ELEMENTARY APPLICATIONS [SEC. 35 111
(s > O).
~ I} = (~n - Sis) cos s + Ci s sins
.
(12) LL2
PROBLEMS
Establish the integration formulas in Probs. l to 3, where t ~ O.
1. l '" sin tx
o x,(x2:1- I)dx
1t
= 2(1 -
-/
e ),
the Fourier sine transform of X-1(X2 + 1)-1 (Chap. 13).
'" sin2 tX '" 1 - eos2tx
2. f -r-
-'" x
dx =
o xl,x d = 1tt.
- exp ( - tx2) d
3.
lo
'" l
x
2 X
-- y'1tt.
c:
5. (x ~ O).
1)fi, d,= 1te' erfe fi
f.o.'"(-r +e-X<
(ef. transformation 111, Appendix A, Table A.2.)
112 SECo 35] OPERATIONAL MATHEMATlCS
eo sin tx eo n 1-
6. fo fi dx f
= 2 o sin ty2 dy = ( 2t)
(t > O).
[Note that y4 + S2 = (y2 + S)2 - 2si = (y2 - y.j2s + s)(y2+ y.j2s + s).]
eo 1 /1r
7. fo exp(-tx2)cos2txdx =2 V-e-' (t > O).
then prove that the limit of that integral as to -> Oand T -> 00 is the integral (2),Seco35,
which represents L{E(t)}.
9. Use the transformation (3), Seco35, to show that, when a > O,
logs - loga .
L{e"'E(at)} = s-a (s> a; cf. transformatlOn 97, Appendix A, Table A.2).
Obtain formally the transforms given in Probs. 10 to 12, if a > O and b> O
(see Probo 9).
eo -sr
I
10.
fo t+a ~dt =L - = easE(as)
{ t+a }
(cf.transformation118, AppendixA,Table A.2.)
eo sin tx n . ..
14. fo x + adx = (2 - SI at) cos at + Ci at Sin ato.
eo cos tx n . . .
d S C
1.
5
fo x + a x = ( 2- 1at ) Sin at - at cos ato
r l
I Wo I
I I.
0.----------------------
I
.."
., x-e x-2e
y
Fig.34
114 SECo 36] OPERATIONAL MATHEMATICS
We have used a convenient extension, awoS.(x) when x > 2e, ofthe function
aW(x) in the right-hand member of Eq. (4) where O < x < 2e. The par-
ticular extension used is immaterial because we first seek a function Y(x)
that satisfiesEq. (4)when O < x < 2e and conditions Y(O)= Y'(O)= O,and
contains two arbitrary constants A = Y"(O)and B = Y"'(O).Since
A B 1 -es
y (S) = 3"
S
+ "4
S
+ awose
S
,
This function satisfies Eqs. (4) and (5) and the continuity conditions.
Since Y"'(O)= B = -awoc, it follows from Eq. (2) that the shear at
-
x = O has the value - woc,or F(O)= Wo as we should expect, since the
magnitude of the vertical force exerted by the support must be the same as
the totalload on the beam. Actually, the displacements Y(x) for this can-
tilever beam can be found by first noting that the shear - Y"'(x)/a at each
section is the totalload on the span to the right of that section.
The bending moment at the end x = Ois Y"(O)/aor Ala:
o %
Fig.35
(1)
T J2i = I (y - r)-tH'(r)dr.
This is an integral equation of convolution type in the unknown function
H'(y). We may write it in the form
T fii = y-t * H'(y).
Let h(s)be the Laplace transform of H(y) with respect to the variable y.
Since H(O) = O,it followsformallyfrom Eq. (1)that
the differential equation of the curve in terms of the variables x and y is,
according to Eq. (2),
dx
1+ 2 = 2gT2 =~,
( dy ) n2y y
Noting that x = Owhen y = O,we see that the parametric equations of the
tautochrone are therefore
a
(3) x = ~(O + sinO), y = 2(1 - cos O).
(4)
j2gF(Y) = I: (y - 17)-tH'(17)d17.
38. SERVOMECHANISMS
Simple integral equations as well as differential equations arise in the theory
of automatic control. As a special case we consider servomechanisms that
force the angle of tum 0o(t) of a rotating shaft to follow closely the angle of
tum 0(t) of a pointer or indicator, where t denotes time. The shaft and
1 See Bcher, M., "Integral Equations," p. 6, 1909.
118 SECo 38] OPERATIONAL MATHEMATICS
(5) <jJ(s) -- - s3
s38;(s) + CsE>;(O)
+ Cs2 + K2S + B3'
In the special case E>;(t)= At, K2 = BC, Eq. (5) becomes
As
(6) <jJ(s) =-
and since the )?olynomials in the numerator and denominator here are of
degrees one and three in s, respectively, it folIows from our earlier observa-
tions (see the problems, Seco27) that c1I(O)= O; that is, the initial value ofthe
angle of deviation is zero. When C > B, all values of s that make the poIy-
nomial in the denominator vanish ha ve negative real parts, and con-
sequently (Sec. 26) c1I(t)contains only damped components. But when
C < B, c1I(t)has an unstable oscillation.
t = 't' + d't'and the number of survivals at any future time t, out of these
replacements, is
R'('t')H(t - 't')d't'.
The total amount of equipment in service at time t is the sum of these
survivals from the replacements during every time interval d't' between
120 SECo 39] OPERATIONAL MATHEMATICS
(2)
F(t) = F(O)H(t) + { R'(T)H(t - T) dT.
We have assumed here that the equipment F(O)on hand at time t = Ois all
new; then R(O) = O.
Ir the amount F(t) that must be in service at each instant is known and
if the survival factor H(t) is known, then Eq. (2) is an integral equation of
convolution type in R'(t). Its solution gives the formula by which replace-
ments must be made.
The equation is an integral equation in the survival factor H(t) when
F(t) and R(t) are known.
In either case, the transformed equation is
(3) . f(s) = F(O)h(s) + s r(s)h(s).
Then
(4) ()
- f(s) - F(O)h(s)
rs - s h(s) ,
and R(t) is the inversetransform of that function.
Suppose the mortality is exponential in character so that
H(t) = e-kt
and that the amount of equipment on hand is to be a constant,
F(t) = b.
Then h(s)= l/(s + k) and f(s) = bIs,and it followsfrom Eq. (4)that
1
r(s) = bkz.
s
Therefore replacements must be made at such arate that the total equipment
replaced up to time t is
R(t) = bkt,
a result that is easily verified as the solution of Eq. (2). Thus replacements
must be made at the rate of bk pieces per unit time.
PROBLEMS
1. In the example solved in Seco 36 let the end x = 2c of the beam be pin-supported,
rather than free, with no other changes in conditions. Find the vertical force exerted
by the beam on the pin, and the vertical force and bending moment exerted on the
supportatx = O. Ans. *wo;HWo;ncWo'
ELEMENTARY APPLICATIONS [SECo 39 121
2. Both ends x = Oand x = 2e of a beam are pin-supported. Find the vertical force
on each support when a transverse load Wo is distributed uniformly over the span
e < x < 2e. Ans. Wo/4; 3Wo/4.
3. Solve Probo 2 if both ends are built in rather than pin-supported.
Ans. f,Wo;tiWo'
4. The end x = O of a beam is built in and the end x = 2e is pin-supported. The
load per unit length is bx on the span O < x < e, and b(2e - x) on the span e < x < 2e.
Find the vertical force on each support in terms of the total load Wo on the beain.
Ans. ~~Wo;gwo.
5. In addition to a distributed load W(x) a single concentrated load W acts between
the ends of a beam, at position x = b. Then the shear F(x) is continuous except for a
jump W at x = b. Apply formula (4), Seco4, to the function Y"'(x) to show that the
transform of Eq. (3),Seco36, is
S4y(S) - s3y(0) - s2y'(0) - sY"(O) - Y"'(O) - aWe-bs= aw(s).
.Compare this equation in y(s) with the one obtained formally by replacing W(x) in
Eq. (3), Seco36, by W(x) + Wt5(x - b) and transforming as if Y"'(x) were continuous,
where t5(x) is the unit impulse symbol.
6. In Seco37, let the time T of descent be proportional to.JY, T.j2g = 2BJY,
whereB > 1. Showthat the curveof descentis the linex = yJB2=l.
7. Ire = k = 21 in Eq. (2), Seco38, find the output angle 80(t) corresponding to the
constant input angle 8{t) = l(t > O), and compare them graphically. Assume that
80(0) = 80(0) = O. AIso, note that the value of the output lags behind that of the
input until t = 3n/4. Ans. 80(t) = 1 - J2e-t sin(t + n/4).
8. For the servomechanism corresponding to Eqs. (4) and (5), Seco38, consider this
special case: 8{t) = At, B3 = CK2. Ir a represents the argument of the complex
number K + Ci, derive the formula
for the angle of deviation, and note the undamped component of Cl>(t).
9. Let the servomechanism discussed in Seco 38 supply only a corrective torque
proportional to the deviation angle CI>(t)while the shaft itself is subject to a damping
torque proportional to its angular velocity. Ir 80(0) = 80(0) = O,show that
k 8Js),
80(s) = rs2 + es + 'o
Fo + I: W(~) d~ - F(x) = O,
(1)
Brief derivations of this equation for some elementary physical functions will
now be given. The derivations are helpful in writing modifications of the
equation and in setting up boundary conditions for specific problems.
123
124 SECo 40] OPERATlONAL MATHEMATICS
y
...-..-
::::-.I
-H I
I
I
Y I
I
I
" I
o
Fig.36
First, let Y(x,t) denote the displacement at time t away from the x axis
of a point (x, Y) of a string in the x Y plane under tension P (Fig. 36). The
string is assumed to be flexible enough that all bending moments transmitted
between its elements can be neglected; thus each element pulls tangentially
on an adjacent element with a force of magnitude P. Suppose further than
conditions are such that the magnitude H of the x component of the tension
remains essentially constant at all times for all points (x, Y); in particular all
displacements Y(x,t) are assumed small compared with the length of the
string. Finally let the slope angle IXremain small in order that each small
element of length of the string may be approximated by the length of its
projection Llx on the x axis. All these idealizing assumptions are satisfied,
for instance, by strings of musical instruments under ordinary conditions of
operation.
The vertical component of tension is the vertical force Vexerted by the
part ofthe string to the left ofpoint (x, Y)on the part to the right ofthe point.
It is proportional to the slope of the string, - VI H = tan IX(Fig. 36); that is,
O
I
"
j "
Fig.37
126 SECo 41] OPERATIONAL MATHEMATICS
o G)---%---t~ ~
'-y
Fig.38
(10) a2 =-.Es
p
Finally, it should be remarked that Eq. (1) is a special case of the
telegraph equation
(11) Yxx(x,t) = KLY,,(x,t) + (RK + SL)Y,(x,t) + RSY(x,t),
where Y(x,t) represents either the electric potential or the current at time t at
a point x units from one end of a transmission line or cable.2 Here the
elements of resistance, inductance, etc., are distributed along the cable or
through the circuit, in contrast to the lumped elements in elementary circuits
(Sec. 33) that lead to ordinary differential equations in the currents. The
cable has resistance R, electrostatic capacity K, leakage conductance S, and
self-inductance L, all per unit length. When R and S are so small that their
effect can be neglected, Eq. (11) reduces to Eq. (1), where a2 = (KL)-l.
that the end may be considered infinitely far from the origin, and having its
other end looped around the Y axis. The loop, initially at the origin, is
moved in some prescribed manner along the Yaxis (Fig. 39) so that Y = F(t)
when x = O and t ~ O, where F(t) is a prescribed continuous function and
F(O) = O. Ir the string is initiallyat rest on the x axis, let us find the formula
for Y(x,t).
The above conditions on Y(x,t) can be written
(1) Yrr(x,t)= a2 Yxx(x,t) (x > O,t > O),
(2) Y(x,O) = Yr(x,O)= O (x > O),
(3) Y(O,t)= F(t), lim Y(x,t) = O
x->oo
(t ~ O),
where a2 = H/p (Sec.40). The equation ofmotion (1) implies that no external
forces act along the string.
A problem composed of such conditions is called a boundary value
problem in partial differential equations. We shall use a formal procedure to
solve the problem and then show how our result can be verified as a solution.
Ir y(x,s) is the Laplace transform of Y(x,t), then, in view of the initial
conditions (2), L{ Yrr} = s2y. AIso,
Ji1t)
o
Fig.39
128 SECo 41] . OPERATlONAL MATHEMATICS
Here we have used the symbol for ordinary rather than partial differentiation
because s is only a parameter in the problem ; no differentiation with respect
to s is involved.
A convenient form of the general solution of Eq. (4) is
y(x,s) = Cle-sx/a + C2e'x/a,
where CI and C2 may be functions of s. This solution could of course be
obtained by transforming the members of Eq. (4) with respect to x. We
consider s as positive, since Laplace transforms generally exist for all s
greater than some fixed number. Then C2 = O,if y(x,s) is to approach zero
as x tends to infinity. The first of conditions (5) is also satisfied if C 1 = f(s),
and the solution of the transformed problem is
(6) y(x,s) = e-(x/a)s(s).
The translation property (Sec. 11) enables us to write the inverse trans-
form of y(x,s) at once:
=0 when t ~~.a
Since F(t) is continuous and F(O) = O,we can see that the function Y(x,t)
described by formula (7) is continuous when x ~ O and t ~ O, including
points on the line x = at in the xt planeo The function clearly satisfies the
boundary conditions (2) and (3).
Any function of t - x/a is easily seen to be a solution of the wave
equation (1) when its derivative of the second order exists. Let F(t) satisfy
these additional conditions: F'(t) and F"(t) are continuous when t ~ O
except possibly for finite jumps at t = ti (i = 1, 2, . . .). The function Y(x,t)
described by formula (7) then satisfies Eq. (1) except possibly at points (x,t)
on the lines t - x/a = ti and t - x/a = O in the quadrant x > O,t > O. In
this sense, then, the function (7) is verified as a solution of our boundary
value problem regardless of the validity of formal steps taken to obtain that
function. In case F"(t) is continuous whenever t ~ O, and F(O)= F'(O)=
F"(O)= Oso that Y(x,t) and its partial derivatives of first and second order
are also continuous on the line t = x/a, then the function (7)satisfiesEq. (1)
PROBLEMS IN PARTlAL DIFFERENTIAL EQUATIONS [SEC. 41 129
with no exceptions when x > O and t > O and represents a solution of the
boundary value problem in the ordinary sense. Note that conditions of
exponential order are not involved in the verification of our solution.
According to formula (7), a point of the string x units from the origin
remains at rest until the time t = x/a. Starting at that time, it executesthe
same motion as the loop at the left-hand end. The retarding time x/a is the
time taken by a disturbance to travel the distance x with velocity a. Since
a = (H/p)t, note that it has the physical dimensions ofvelocity. The vertical
force from left to right at a point (Sec. 40) is -HYix,t); thus the vertical
force exerted on the loop to make it move in the prescribed manner, as
found from formula (7), is
H
V(O,t)= -F'(t)
a = JPijF'(t).
(t>l)
"
Th"'b
O alt-l)-O
Fig.4O
130 SECo 42] OPERATIONAL MATHEMATICS
The function Y(X,t) above can also represent the longitudinal dis-
placements in a semi-infinite elastic bar (Sec. 40), initially at rest and un-
strained. The distant end of the bar is held fixed and the end x = O is
displaced in a prescribed manner, Y(O,t) = F(t). When F(t) is given by
formula (8), the solution (9) or Fig. 40 shows that at time t the section of the
bar fram x = a(t - 1) to x = at is strained while the remainder is unstrained
and at res.
(4)
~
,
I
I
I % I
O I I
I
{
lat.-'lgt2) I
Fig.41
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SEC. 42 131
where the primes denote derivatives with respect to x. Since the constant
- g/S3is a particular solution of Eq. (4), the general solution ofthat equation
is
PROBLEMS
Y(x,t) = G(t) - G( t - ~)
where G(t) = J~J~ F(-r) d-rdr if t ~ O, G(t) = O if t ;;;O.
8. The end x = Oof a semi-infinite stretched string is looped around the Yaxis which
exerts no vertical force on the loop, and the distant end is fixed on the x axis. The string
is displaced into the position Y = e - X(x ~ O) and released from rest in that position
at the instant t = O. If no external forces act on the string, set up the boundary value
problem for the displacements Y(x,t) and find its solution in the form Y = e-a, cosh x
if x ;;;ato Y = e-X cosh at ifx ~ ato Suggestion: Find A(s) so that A(s)e-X is a particular
solution of the nonhomogeneous equation a2y" - s2y = -se-X in y(x,s).
9. The force per unit area on the end x = O of a uniform semi-infinite elastic bar
(Fig. 42) is F(t): - EYx(O,t)= F(t). Ifthe infinite end is fixed and the initial displacement
and velocity of each cross section are zero, set up the boundary value problem for
longitudinal displacements Y(x,t) and derive the solution
Y(x,t) = iG( t - ~)
where G(r) = J~ F(-r) d-r if r ~ O, G(r) = O if r ;;;O; thus Y(x,t) = O when t ;;;x/a. Note
that the displacement of the end x = Ois Y(O,t)= (a/E)G(t).
w - f3
-
~,
-r+-r--t ~~
~
iIfi
~~:::~ O
Fig.42
1
10. When the pressure F(t) in Probo 9 is a finite impulse: F(t) = EFo when t < to.
F(t) = Owhen t > to, show that Y = Owhen t ;;;x/a, y = (at - x)Fo when x/a ;;;t ;;;
to + x/a, and Y = Foato when t ~ to + x/a. Study this function Y(x,t) graphically.
11. Under the instantaneous impulse ofpressure F(t) = 1<5(t)in Probo 9 show formally
that
al
if x < at, Y(x,t) = O if x > ato
Y(x,t) = E
Note that in this hypothetical case a part of the bar is displaced into a region already
occupied by another parto
12. Let the pressure F in Probo 9 be constant, F(t) = Fo.
(a) Show formally that Ey(x,s) = aFos-2 exp( -sx/a) and hence
(b) Draw graphs of Y(X,t) here as a function of x with t fixed, and as a function
of t with x fixed. Note the jumps in Yxand Y;where x = at and the removable discon-
tinuities in the functions Yxxand Y;,there.
13. For the function Y(x,t) and its transform y(x,s) found in Probo 12, show that
L{ Y;,}is zero, not S2y as assumed in the formal solution; also that L{ YxJ is zero, not
d2y/dx2. Note, however,that L{Y;, - a2yxx}= O= s2y- a2 d2y/dx2.
14. If a rigid mass m is attached to the end x = O of the semi-infinite bar in Probo 9
(Fig. 42), when the bar has unit cross-sectional area, and the force F(t) is a constant Fo,
show why the boundary condition at that end becomes
mY;,(O,t)= EYx(O,t)+ Fo.
Write b = E/am and derive the formula
x
Y(x,t) = -aFo xlI
E [ t - -a - -b + -e-
b
b(' Xla
- o t- - .
a)
Js (
15. An unstrained semi-infinite elastic bar is moving lengthwise with velocity - Vo
when, at the instant t = O, the end x = O is suddenly brought to rest, the other end
remaining free (Fig. 43). Set up and solve the boundary value problem for the longi-
tudinal displacements Y(x,t). AIso show that the force per unit area exerted by the
support at x = O upon the end of the bar is Evo/a.
Ans. Y = -vot when t ~ x/a, Y = -vox/a when t ~ x/a.
- Vo
.
,I
~ x
~
I
Ag.~
16. Let the end x = O of the bar in Probo 15 meet an elastic support at the instant
t = O,such that the pressureexertedby that support upon the end is proportional to
the displacement of that end: - EYAO,t) = - BY(O,t),or
:fx(O,t)= b Y(O,t) (b = B/E).
Show that the velocity ofthat end is -voe-ab' since abY(O,t) = -vo(1 - e-ab').
Let the ends of a semi-infinite string stretched along the positive x axis be
kept fixed, and let the string be given some prescribed displacement Y = 1I(x)
initially a~d released from that position with initial velocity zero. Here
11(0) = 11(00) = O. Then th~ boundary value problem in the transverse
displacements Y(x,t) is
(x > O,t > O),.
134 SECo 43] OPERATIONAL MATHEMATICS
u(z,s) = foooe-ZXy(x,s)dx.
and performing the inverse transformation with respect to z, with the aid of
the convolution, we ha ve
aC
2s
- ~
2ao I
x cI>(~)e-(s~/a) d~.
(5)
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SECo 44 135
Substituting this into Eq. (4), we can write the result in the form
+ J: W(~)e-[S(x-~)]/a d~.
In order to combine the last two integrals, let W(x) represent the odd extension
of the function W(x):
(7) tPl(X) = W(x) ifx ~ O,
= -W( -x) ifx ;;;;O;
thus Wl( -x) = -Wl(X) for all x. Then Eq. (6) can be written
2y(x,s) = L{W(x + at)} + L{W(x - at)}
and therefore
(8) Y(x,t) = t[W(x + at) + Wl(X - at)].
It is easily seen that a twice-differentiable function of the variable
x + at, or of the variable x - at, as well as each linear combination of such
functions, satisfies the linear homogeneous wave equation Y" = a2yxx. Qur
continuous function (8)therefore satisfiesthe wave equation for all x and t
where W"(x + at) and Wl(x - at) existo It satisfies the condition Y,(x,O)= O
when x > Owherever W'(x)exists, and it satisfies the remaining conditions of
problem (1) because W(x) = W(x) when x > O, W1(-at) = -W(at) and
W(oo,t) = O. .
Instantaneous positions of the string can be sketched by adding
ordinates as suggested by formula (8). When t is fixed, the graph of !<I>(x-
. at), for instance, is obtained by translating the graph of the function !w(x),
defined for all real x, to the right through the distance al.
--+
o % %
Fig.44
aFo 1 se
and, when x = e, y( e ,S) = - E s"2tanh -.a
In Seco23 we found that S-2 tanh (bs/2) is the transform of the triangular
wave function H(b,t) of period 2b (Fig. 10). Therefore the displacement of
the end x = e is
aFo 2e
(4) Y(e,t) = EH ( ~,t ) ;
that is, the end moves byjerks as indicated in Fig. 45.
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATlONS [SECo 44 137
~~
(0.2C;0)
Q (4cti ,01
~
Fig.45
since (1 + z)-1 =L
o
(- 1tzn when O < z < 1. Therefore
3c-x
ti
k ti
Fig.46
138 SECo 44] OPERATIONAL MATHEMATICS
AN IMPULSE OF PRESSURE
y E s cosh (se/a)
The displacements Y(x,t) can be represented by a seriesof the step functions
So{t - [(2n + l)e:!: xJ/a}. Those displacementsare not continuous.
To see the behavior of the end x = e, we note that
al 1 se
y(e,s)= -E -s tanh-.a
Thus the end jumps suddenly back and forth between two fixed posi-
tions. It is possible to demonstrate a close approximation to that behavior by
substituting for the bar a loosely wound coil spring. Ir, when the spring is
hanging from one end, the free lower end is given a sharp tap, the lower end
tends to move as indicated.
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SECo 44 139
PROBLEMS
1. A string is stretched between two fixed points (0,0)and (c,O)oIf it is displaced into
the curve Y = b sin (nxjc) and released from rest in that position at time t = O,set up
and solve the boundary value problem for the displacements Y(x,t). Verify the result
fully and describe the motion of the string. Ans. Y(x,t) = b cos (natjc) sin (nxjc).
2. If the initial displacement of the string in Probo 1 is changed to
Y(x,O) = b sin nnx (O~ x ~ e),
e
where n is any integer. derive the formula
nnat o nnx
Y(x,t) = bcos-sm-.
e e
Note that the sum of two or more of these functions with different values of n and b
is a solution of the equation of motion that satisfies aIl the boundary conditions if the
initial displacement Y(x,O) corresponding to such a superposition of solutions is a
certain linear combination of the functions sin (nnxjc).
3. As a special case of the initiaIly displaced semi-infinite string considered in Seco43,
where Y(O,t)= Y(co,t) = O, let the end segment O~ x ~ 1 be plucked so that the
initial displacement has this form:
Y(x,O) = sin nx if O~ x ~ 1, Y(x,O)= O if x ~ 1,
where again Y,(x,O)= O when x > O. Show the displacement Y(x,t) graphically as a
function of x when t is fixed and at > 1, and note that the displacement is a wave
formed by letting one cycle of the curve Y = t sin nx, -1 < x < 1, move to the right
with velocity a. (Boys have used such plucking of a kite string to dislodge parachutes
hooked onto the string and blown toward the kite.)
4. A long string stretched along the entire x axis is released at rest from a prescribed
initial displacement. Thus
( - co < x < 00, t > O),
Y(x,O) = <I>(x), Y,(x,O)= O (-00 < x < 00),
x-lim
- 00Y(x,t) = O, lim Y(x,t) = O
x-oo (t ~ O).
Assuming <I>"(x)
exists for all x, derive the formula
Y(x,t) = t[<I>(x+ at) + <I>(x at)] -
and verify that solution.
5. A semi-infinite string stretched along the positive x axis with ends fixed,
Y(O,t)= limx_ooY(x,t) = O,is given an initial velocity Y,(x,O)= g(x), where g is section-
ally continuous on each bounded interval, limx_oog(x) = O,and g is integrable from
zerotoinfinity. (a)Ifg istheoddextensionofg,sothatg(r) = g(r)andg(-r) = -g(r)
if r > O,show that the transverse displacements can be written
Y(x,t) = 2
a
_f
1 x+ar
x-al
_1
g(r)dr = 2 [G(x+ at) - G(x - at)],
a
140 SECo 44] OPERATIONAL MATHEMATICS
EYz(O,t)
n r--
I
I
2Fo II I
I
o 3c
a
Fig.47
7. Let the constant force Fo in Probo 6 be replaced by a finite impulse of duration 4e/a:
. 4e . 4e
F(t) = Fo IfO < t<-, a F(t) = O If t>-.
a
(a) Show that the force per unit area exerted by the bar on the support at the end
x = Ois that shown in Fig. 47 up to the time t = 3e/a after which time the force is zero.
(b) Show that the end x = e moves with constant velocity Vo = aFo/E during
the time intervalO < t < 2e/a, then with velocity - Voduring an equal time interval
and that Y(e,t) = Owhen t ~ 4e/a.
8. A constant longitudinal force Fo per unit area is applied at the end x = e of an
elastic bar (Fig. 48) whose end x = Ois free and which is initially at rest and unstrained.
Set up the boundary value problem for the longitudinal displacements Y(x,t) and show
that Ey(x,s) = aFos- 2 cosh (sx/a)/sinh (se/a).
(a) Derive the formula
Y;(O,t)= Vo
.;0
f So[t - (2n + 1)~J a where Vo
2aFo.
= E'
hence note that the free end moves with velocity Vo from time t = e/a until t = 3e/a,
then with velocity 2vo until t = Se/a,etc.
- -
(b) In terms ofthe function {T k} = (r k)So(T k), show that -
F. 00
Y(x,t) = ; .~o [{at + x - e(2n+ 1)}+ {at - x - e(2n+ 1)}].
o e
Fig.48
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SECo 44 141
nEY(e,t) = 2eFoIsinrotl.
10. The end x = O of a bar or heavy coil spring (Fig. 49) is free. The end x = e is
displaced in a prescribed manner, Y(c,t) = G(t). If the bar is initially unstrained and
at rest, find the transform y(x,s) of longitudinal displacements.
J 01
Fig.49
(a) When G(t) = Fot, show that the displacement Y(O,t) of the free end is repre-
sented by the function obtained by integrating the function EYx(O;r)in Fig. 47 from Oto
t and show the displacement graphically.
(b) If G(t) = bt when t ;;4eJa and G(t) = 4beJa when t ;;4eJa, show that the
free end moves with a uniform velocity 2b to a new position and remains there, as shown
in Fig. 50.
Y(O,t)
Fig.50
11. An unstrained elastic bar is moving lengthwise with velocity Vo when at the
instant t = O its end x = e meets and adheres to a rigid support (Fig. 51).
(a) If A is the area of the cross section of the bar and M denotes the square-wave
function (Fig. 9), show that the force on the support can be written
AEvo 2e
-AEy"(e,t) = ---;-M ( -;i',t) (a2= ~).
- Vo
[
O 1-.
Fig.51
142 SECo 44] OPERATIONAL MATHEMATICS
8(x s) = ~ 1
, S2 [- cosh (sx/a)
cosh (se/a)]
for the transform of the angular displacements 0(x,t) of the cross sections, where
a2 = EJ p (Sec. 40).
(a) Ir H denotes the triangular-wave function (Fig. 10), show that the displace-
ment of the middle cross section is
0(0,t) = Jt ift ~ ~
- a'
Je 2e e . e
0(0,t) = -a - JH-,t - -a ft f;-.a
( a )
- "
( -e
. r I,
O
+ ~-
e
Fig.52
14. Let M denote the square-wave function (Fig. 9). Show that the torque acting on
the support at the end x = e of the shaft in Probo 13 is a-1 E.IJM(2e/a, t).
15. The end x = Oof a cylindrical shaft is kept fixed. The end x = e is rotated through
an angle 80 and, when all parts have come to rest, that end is released; thus 0(x,0) =
8ox/e (Fig. 53). Show that the angular displacement of the free end at each instant can
be written
a 2e
0(e,t) = 80 - 80cH ( -;' t) ,
of-
Fig.53
~ e
45 EQUATIONS OF DIFFUSION
Let U(x,y,z,t) denote the temperature at a point (x,y,z) within a solid at time t.
Let S be aplane or curved surface passed through that point and let the
coordinate n represent directed distance along the line normal to S at the
point. Then the flux of heat, the quantity of heat per unit area per unit time
transferred across S at the point by conduction, is given by the formula
d
(1)
<II(x,y,z,t)= - K dn U(x,y,z,t),
When the temperature function has the simple form U(x,t), consider
a portion of the solid in the shape of a prism parallel to the x axis (Fig. 54).
Since there is no temperature variation with respect to distance normal to
the lateral surface of the prism, no heat is conducted across that surface.
According to formula (2), an element of the prism extending from position
Fig.54
144 SECo 45] OPERATlONAL MATHEMATICS
where k = K/e and k is called the thermal diffusivity of the material. Equa-
tion (5) is the simple form of the heat equation or the equation of diffusion.
When U(x,t) represents the concentration of a diffusing substance in
weight per unit volume, Eq. (3) follows as before if e = 1 and K is the coeffi-
cient of diffusion. Thus, when K is constant, U(x,t) satisfies Eq. (5) where
k= K.
The heat equation or equation of diffusion for U(x,y,z,t)
(6) U, = k(Uxx + Uyy + Uzz),
when coefficients are constant, can be derived in a similar manner by con-
sidering a three-dimensional element of volume ~x ~x ~Z.1
Modifications of the heat equation are easily written. For example,
let U(x,t) represent temperatures in a slender wire along the x axis. Let heat
be generated at a time rate R(x,t) per unit volume of the wire and let heat
loss from the lateral surface take place at arate proportional to the difference
between the temperature of the wire and the temperature Uo of the surround-
ings. The modification of Eq. (3) then shows that
1
(7) U,(x,t) = kUxix,t) + -R(x,t)
e - h[U(x,t) - Uo],
46 TEMPERATURES
IN A SEMI-INFINITE
SOLIO
Let us now derive a formula for the temperatures U(x,t) in a semi-infinite
solid x ~ O,initially at temperature zero, when a constant flux of heat is
maintained at the boundary x = O (Fig. 55). In this idealized case of a thick -
slab of material, we shall substitute for the thermal condition at the right-hand
boundary the condition that U tends to zero as x tends to infinity. The
boundary value problem is then
(1) Ulx,t) = kUxx(x,t) (x > O,t > O),
(2) U(x,+O) = O (x > O),
- Ku,,(O,s)= <Po,
s
lim u(x,s) = O.
x-oo
u(x,s) = <PO{e-M/S/k.
Ksys
Accordingto formula (5),Seco27, we can write
-'
4>0-!
-10 x
-1
Fig.55
146 SECo 47] OPERATIONAL MATHEMATICS
and in view of the factor l/s in our formula for u(x,s) it follows that
4>
, X2 d.
U(x,t) = ~
K ~ -
1t1o
exp ---
( 4h ) .Ji
4>ox
f oo
= Kv :1t x/(2./ktA2exp (- A )dA,
1 2
where the second integral is obtained from the first by the substitution
A = x/(2.)kT). Upon integrating the last integral by parts, we find that
2
4k t)
- 2x
f oo exp (- A(!)dA .
x/(2./kt ]
We can therefore write our formula in terms of the complementary error
function (Sec. 27) in the form
-~
x erfc ~ -
(4)
[
U(x,t) = 4>0 2 {kiexp
K V -; ( 4kt) ( 2Jkt)J
.
We can show that the function (4) satisfies all our conditions (1), (2), and (3).
We observe that
1
K 1t
U(O,t)= 24>1f fi.
Thus the temperature of the face of the solid must vary as Jt in order that
the fluxof heat through the faceshall be constant.
Let us first study the flux of heat through the face of the solid,
~(t) = -KU",(O,t).
The transform of this function, - Ku",(O,s),according to formula (1), is
K K 1
(2) cp(s)= y'k.fif(S) = y'ksf(s) .fi'
Since s-t = L{(nt)-t} and sf(s) = L{F'(t)} + F(+0), assuming that F is a
continuous function,then
(3) ~(t) =~
pjt [
F(+0) +
fo~
r F'(t - 't) d't
]
,
(4)
o f Fig.56
e-x./Sik = L exp -~ x ,
{ 2v1nkt3
( 4kt)}
U(x,t) = 2#
X
io
t F(t -
Ti
T) X2
exp ( - 4h ) dT.
Substitutinga new variable of integration, we have for the general tempera-
ture formula
oo 2
2
fi f
(7) U(x,t) = - F t x
txl../ki ( - 4k,F ) e-;.2 dA..
(9)
U(x,t) =.Fo erfc
I (2~).
Since this is a function of Fo and x/Jki only, it follows that the rapidity o
heating is proportional to k; for if k is increased and t decreased so that k
is unchanged, the temperature at any given distance x from the face is thc
same. It is also interesting to note that for a fixed k two points Xl and x2 \Vi
have equal temperatures at times ti and t2 provided x/-.; = X2/.jt;, thaJ
is, if
(10)
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SECo 47 149
This is sometimes called the law of times in the conduction of heat in semi-
infinite solids.
Ir Fo is positive, formula (9) assigns a positive value to U(x,t) for each
x and t beca use the complementary error function has only positive values.
At each interior point of the solid U(x,t) is small when t is sufficiently small,
but the temperature does not retain its initial value U(x,O)= O during an
interval of time after t = O. That instantaneous though highly attenuated
transfer of the thermal disturbance at the surface can be noted in other
cases. In Seco46, for example, the flux
PROBLEMS
.
1 The initial temperature of a semi-infinite solid x ~ Ois zero. The surface tempera-
ture is this step function:
U(O,t)= Fo ifO < t < to, U(O,t)= O if t > to.
Obtain the temperature distribution formula
x
U(x,t) = Foerfc" r,:;
~kt
XX
= Fo erf - erf-
[ 2Jk(t - to) 2.jki J
2. A thickslab ofiron withthermal diffusivityk = 0.15 cgs (centimeter-gram-second)
unit is initially at OC throughout. Its surface is suddenly heated to a temperature of
500C and maintained at that temperature for 5 min, after which the surface is kept
chilled to OC. (See Probo 1.) Find the temperature to the nearest degree at a depth of
10 cm below the surface (a) at the end of 5 min; (b) at the end of 10 mino
Ans. 146C; (b) 82C.
3. Solve Probo 2 if the slab is made of firebrick with diffusivity k = 0.007 cgs unit.
Ans. (a)OC; (b) Oc.
4. The surface of a thick slab of concrete for which k = 0.005 cgs unit, initially at OC,
undergoes temperature changes described in Probo 2. Show that at each instant the tem-
perature at any depth x in the concrete slab is the same as the temperature at the depth
X2 = J30x in the iron slab. Generalizethis result for materialswith diffusivitiesk
and k2 and any common time interval to of heating the surfaces of the slabs.
5. At time t = O,the brakes of an automobile are applied, bringing the automobile to a
stop at time to. Assuming that the rate of generating heat at the surface of the brake
150 SECo 47] OPERATIONAL MATHEMATlCS
Hence show that this teniperature is greatest at the instant t = tto and that ths maximum
temperatureis j2U(O,to). .
6. Theinitialtemperatureofa semi-infinitesolidx ~ Ois zero. The inwardfluxofheat
through the facex = Ois a prescribed function 4I(t)oftime and the distant face is kept at
temperaturezero. Derivethe temperatureformula
(x > O).
7. An impulse of heat of quantity Qo per unt area is introduced through the face
x = O of a semi-infinite solid x ~ Oat the instant t = Oand the face is nsulated when
t > O,so that -KUx(O,t) = Qo c5(t).(That instantaneous source ofheat may be realized
approximately by burning a layer of highly combustible fuel at the face.) Ir U(x,O)= O
when x > O,show formally that the subsequent temperature is
X2
U(x,t) = ~QK /ff
-1ftexp --4kt (t > O).
( )
(See Probo 8 for a verification.) Observe that U(x,t) > Owhenever t > O.
8. The function I/I(x,t)= t-t exp [-x2/(4kt)] and its derivative I/Ix are known as
fundamental svlutivns vf the heat equativn I/Ir= kl/lxx'
(a) Verify that 1/1satisfies that equation when t > O,that 1/1 iO,t) = Oand I/I(oo,t) =
O when t > O,and that I/I(x,+O)= O when x > O; then complete a verification of the
formal solution of Probo 7 by showing that the total heat content of the solid per unit
area offace, f~ cU(x,t) dx, relative to the initial heat content, is Qo.
(b) Show that the functions 1/1and I/Ixare unbounded in a neghborhood of the
point x = t = O when x > O and t > O. Note that functions AI/I(x,t), where A is an
. arbitrary constant, could be added to the solution (4),Seco46, to producea familyof
solutions of the problem in that section if unbounded solutions were permitted.
9. For a solid whose temperature is a function U(x,t) show that the total quantity
of heat transmitted across a unit area of a plane x = Xo can be written, formally,
PROBLEMS IN PARTlAL DIFFERENTIAL EOUATIONS [SECo 48 151
Q(xo) = -/CKs-o
limJif(s) (K = ck).
For the solid in Probo 7, show that Q(xo) = Qo.
10. The initial temperature distribution of a solid x ~ Ois prescribed:
U(x,O) = g(x) (x > O).
If U(O,t) = O when t > O and UAoo,t) = O, show formally that, when t > O,
1 OJ
(r - X)2 (r + X)2
48 TEMPERATURES IN A SLAB
The initial temperature of a slab of homogeneous material bounded by the
planes. x = Oand x = 1is Uo. Let us find a formula for the temperatures-in
this solid after the face x = O is insulated and the temperature of the face
x = 1is reduced to zero (Fig. 57).
The temperature function U(x,t) satisfies the following conditions
(2) u(1,s)= O.
U(x,O)-UQ u-o
o x
Fig.57
152 SECo 48] OPERATIONAL MATHEMATICS
The solution of the ordinary differential equation (1) that satisfies the
first of the conditions (2) is
u(x,s) = USo
+ e cosh xA,
(3) u(x,s) = uo
[! - ! coshx# ].
s s cosh1#
Let us write
q=f,
(ex ~ O).
[ 2ykt
r,::
]} .
We shall not take up the verification of this formula sincea complete
discussionwould be lengthy. However, it is not difficult to show with the
aid of the ratio test that the series converges uniformly with respect to x
and t and that the series can be differentiated term by termo Since the value of
the complementary error function here decreases rapidly as n increases, the
convergence of the seriesis rapid, especiallywhen t is small. Moreover the
error function is one that is tabulated so that the series is well adapted to
computation.
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SECo 50 153
Fig.58
u(x,s)
= f( s) sinhx-fi1: .
sinhv s
Proceeding as in the last section, and using the convolution, we find
that
'"
2 '" (m - X)2
(1) V(x,t) = V~L
7tn=O {f F t-
t(m-x)/./l [
4A2 J e-A2dA
'"
(m + X)2
- ft(m +x)/./l F[t - 4A2 J e-A2 dA} (m = 2n + 1).
When the temperature of the facex = 1 is constant,
(2) F(t) = Fo,
our formula can be written
'" 2n + 1 + X 2n + 1 - x
(3)
V(x,t) = Fo t [erf( 2jt ) - erf( 2.Ii )J .
Details are left to the problems.
A thin semi-infinite plate occupies the space x ~ O,O~ Z ~ Zo, - 00 < y < oo.
Heat transfer takes place at the faces Z = O and Z = Zo,into a medium at
temperature zero,accordingto the linear lawofsurfaceheat transfer(Sec.45);
154 SECo 50] OPERATIONAL MATHEMATICS
u-o
t t
S-::'::==::Y
+ + :c
U{x.O)-O
Fig.59
but the thickness Zoofthe plate is small enough that variations oftemperature
with z can be neglected. Ir the initial temperature is zero and the end x = O
is kept at temperature Fo (Fig. 59), the boundary value problem for the
temperature function becomes
Ut(x,t) = kU xAx,t) - hU(x,t) -t ~i. (x > O, t > O),
(1)
U(x,O)= O, U(O,t)= Fo, lim U(x,t) = O.
L -1 {exp
(-xPi)} = 2~ nkt3 exp ( - 4kt) .
It follows from formula (2) that
t
(3)
U(x,t) = ~
2~
F. x
""
o ie-ht exp ---
(
X2
4kt ) tt
dt
2F. hx2
~
= '\! n Jtx../ki exp ( -A2 - 4k12A ) dA,
(4)
~ 100 exp ( -AZ - ~~) dA = eZaerfe (r +~) + e-Za erfe (r - ~).
This formula can be verified by noting that its two members have the same
derivative with respeet to the parameter r and that both members vanish as
r tends to infinity. With the aid offormula (4) the solution (3) of our problem
becomes
(5)
U(x,t) = ~[ebXerfc
(2~ + Jht) + e-bx erfe (2~- Jht)1
whereb =.Jhik. Note that U(x,t) > Ofor eaeh x whenevert > O,if Fo > O.
In obtaining Eq. (5) we have found a desirable form of the inverse
transform of the funetion (2). The verifieation of our formal solution of
problem (1)is left to the problems.
U(%,O)-A
o a %
Kl.kl
Fig.60
156 SECo 51] OPERATIONAL MATHEMATICS
The solution of Eq. (7) that satisfies the first of conditions (9) is
fS A
u(x,s)= Cl coshx-vI; + -; (O~ x < a),
and the solution of (8) that satisfies the second of conditions (9) is
(x > a).
u(x,s) = C2exp (-xJf)
By applying the conditions (10) to these functions, the values of Cl and C2
are easily found. The formulas for u(x,s) can then be written
- A 1 - A e-ala-x) + e-ala+X
(O< X < a),
(11) u(x,s)- -; ( 1 -~. 1 - Ae-2aa )
A(l + A)e-aplx-a)- e-aI2a+px-pa)
(12) u(x,s) = (x > a),
where
r; (k;
u = vk;' Jl = -VI<;'
and therefore UJl= .;;k; and IAI< 1.
Equation (11) can be written in the form
u(x,s) =
A
!s - 1-
2
A f
"=0 s
A"(e-alma-x) + e-alma+x (O< X < a),
PROBLEMS IN PARTIAL DIFFERENTIAL EQUATIONS [SEC. 51 157
PROBLEMS
1. Completethe derivationofformula (1),Seco49.
2. Whenx = 1,showthat thegeneraltermofthe seriesinformula(3),Seco49,becomes
<I>(O,t)
= KFo(~e-ht + AerfJht).
5. In the problem (1) of Seco50, make the substitution
V(x,t) = ehIU(x,t)
and show that the resulting problem in V(x,t) is a special case of one solved in Seco47.
158 SECo 51] OPERATIONAl MATHEMATICS
7. In Prob. 6, let the face x = 1 be insulated when t > Owhile U = Oas before at the
face x = O. Find the flux of heat outward through the face x = O.
00 2n + l
[
Ans. K 1 - 2 "~o(-1)" erfe 2../i J.
8. Heat is generated in a long bar x $;;Oat arate R(t) units per unit volume [Seco45,
Eq. (7)],and the lateral surface of the bar is insulated. Ir the initial temperature is zero
and the end x = Ois kept at that temperature while the infinite end is insulated (Fig. 61),
derive the temperature formula
2 oo X2
U(x,t) = Q(t) - r= Q( t
V 1t
l -
4U2 ) exp(-''2)dl,
!:x;./fi
l t
where Q(t) = -
e o i
R(t) dt.
u-0-c~;;;;:::;;;;~~:::r:;:~!;::;:::)
O UIz,O-O -
-1/4-
--.,.7,
Fig.61
9. In Probo 8, let the rate of generation of heat be a eonstant B and show that
B
i
U(x,t) = e oerf 2../fT
t ~dt.
10. The lateral surface of a bar of unit length is insulated while its ends x = O and
x = 1are kept at temperaturezero. Heat is generatedthroughoutthe bar at a eonstant
rate of B units per unit volume. Ir the initial temperature is B(x - x2)/(2K), show that
the bar retains that initial temperaturedistributionwhent > O.
11. Units of time and distance are ehosen so that k = 1 and a given bar has unit
length. The lateral surface and the end x = O are insulated. The initial temperature of
the bar is zero, and the end x = 1 is kept at that temperature. Ir heat is generated
throughout the bar at arate of R(t) units per unit volume, derive the temperature
formula
12. Let the semi-infinite evaporation plate (Sec. 50, Fig. 59) have a uniform initial
concentration of moisture U(x,O) = uo, and let the end x = O be kept dry, U(O,t) = O,
while the infinite end is impervious to moisture. Derive, and verify, this formula for
the concentration :
-h, f x
U(x,t) = uoe er rv;.
2yKt
13. The entire surfaceof a long porous cylinder is impervious to moisture. The
cylinder consistsof two semi-infiniteparts, the first ( - 00 < x < O)of material with a
coeffieientof diffusion of moisture K and the second (O < x < 00) with a eoefficient K 2
(Fig. 62). Initially, the concentration of moisture is zero in the part x < O and Uo in
the part x > O.Let IXbe the ratio of equilibrium concentrations in the two parts so that,
as indicated in Seco51, the concentration U(x,t) satisfies the eonditions
U( +O,t) = IXU(-O,t),
at the interface x = O. Derive the formula
Uo -x
U(x,t) = - rr;;-; whenx < O,
IX + p erfe ( 2yKt )
X
= - Uo + p erf" rr;;-;) whenx > O,
IX + p(
IX
.yK2t
where p = .JK/K2' Show that the ratio ofthe limiting coneentrations, as t - 00, in
two parts is 0(.
L---G:r~~
~UI#IJ)-O K~::::::~
~UI#IJ)-"
O
Fig.62
1
14. Derive the integration formula (4), Seco50. Suggestion:Write
All the problems treated in this chapter involve partial differential equations
and boundary conditions that are linear,that is, offirst degree in the unknown
function and its derivatives. The limitation of our operational method to
the treatment of such linear boundary value problems is a natural one, since
we have presented no formula giving the Laplace transform of the product
/
160 SECo 52] OPERATIONAL MATHEMATlCS
We give now a synopsis of theorems and definitions from the basic theory of
functions of a complex variable, material needed to develop futher the
theory ofthe Laplace transformation. For complete treatments ofthe topics,
including proofs of statements and theorems, the reader may refer to books
on the subject.l Derivations of extentions of the Cauchy integral formulas
are given in the final section of this chapter.
53 COMPLEX NUMBERS
We shall not review the algebra of complex numbers except to remind the
reader of properties of absolute values and conjugates. We write
z = x + iy = r(cos 9 + i sin 9),
1 See, for instance, the author's "Complex Variables and Applications," 2d OO.,1960, and
references listed in the Bibliography in that book.
162
FUNCTIONS OF A COMPLEX VARIABLE [SECo 54 163
if Z2 "" O.
Neither the statement z 1 > z 2 nor z 1 < z2 has meaning unless z 1 and z 2 are
both real numbers.
Ir m and n are integersand z "" O,then
me . me
znfn = r"'fn ( ros --;-+ i SID --;-) , Izmfnl= Izlmfn.
AIso zz = IzZ..
A neighborhood of a point Zoin the complex plane is a two-dimensional
circular domain, or disk, consisting of all points z such that Iz - zol < (,
where ( is some positive number.
Here u and v are single-valued in the domain consisting of all points of the
complex plane except those for which y = O and x ~ O.
A function f has a limit Woat a point Zo,
if and only if for each positive number lO there is some number ~ such that
lim f(z)
%-%0
= %-%0
lim (u(x,y) + iv(x,y)]= Uo+ ivo
provided the limit exists, that is, provided that for each positive number lO
there is a number ~ such that
Fig.63 o
AIso, sin z and cos z are periodic with period 2n, while sinh z and cosh z are
periodic with period 2ni.
The forms of the trigonometric identities are the same as those for
functions with real arguments; for instance,
The same is true for the relations between the six hyperbolic functions.
Furthermore, the formulas for the derivatives of all those functions retain
the same form when the variable is real. Thus for all z in a region where w(z)
is analytic, the composite functions sin w(z) and cos w(z) are analytic functions
of z and
d
-sm
. w = cosw-
dw d
-cosw
. dw
= -sm w-.
dz dz' dz dz
The absolutevalues of sin z and cos z are not bounded as Iyl -. oo. It is left to
the problems to show that
(5) IsinZl2 = sin2 x + sinh2 y, Icos Zl2 = COS2x + sinh2 y,
From those formulas we can see that the zeros of sin z and cos z are just the
real zeros of those functions, while the zeros of sinh z and cosh z are pure
imaginary numbers.
PROBLEMS
1. (a) Interpret the triangle inequalities (Sec.53) in tenns oflengths of sidesoftriangles.
(b) Extend the first triangle inequality to apply to n complex numbers as follows:
IZ1+ Zz + ... + z.1 ;;; Iz11 + Izzl + ... + Iz.l.
2. Ir Izzl =FIz31,prove that
7. Show that (a) e-Z = lje"; (b) exp (Zl + Z2) = (exp zl)(exp Z2);(e) lexp (-z2)1 is not
bounded for all z.
8. In Seco55, prove (a) identities (5); (b) identities (6).
9. Prove that (a) sin z = O if and only if z = :tmr; (b) sinh z = O if and only if z =
:tmri, wheren = O,1,2,....
I
I 10. Prove that tanh z is analytic except at the points z = :t(n - !)ni, where n = 1,
2,... .
11. Show that (a) tan (z + n) = tan z; (b) tanh (z + ni) = tanh Z.
12. Establish the inequalities sinh Ixl ~ Isinh zl ~ cosh X.
13. Use the definition of f'(z) when dz = dx, then when dz = idy, to show that
f' = Ux+ ivx= Vy- iuywhen f'(z) exists. Hence deduce that the components u and v
necessarily satisfy the Cauchy-Riemann conditions at points wheref'(z) exists.
I 14. In polar coordinates, where z = reiO,r > O, verify that the Cauchy-Riemann
conditions take the form
I ou 1 ov,
or = -;: 00'
I
and that f'(z) = e-iO[u,(r,O)+ iv,(r,9).
56 CONTOUR INTEGRALS
Consider first a complex-valued function G of a real variable t, with com-
ponents V and V,
G(t) = V(t) + iV(t).
(1)
f G(t) dt = f V(t) dt + i f V(t) dt
(2)
If G(t)dtl ~ f'G(t)'dt.
The limit b ofthe integral (1)can be replaced by <X)ifthe corresponding
integrals of V and Vexist. The comparison test for convergence of integrals
(Prob. 14, Seco5) is easily extended to the improper integral of G. Thus if
FUNCTlONS OF A COMPLEX VARIABLE [SECo 56 169
(3)
Ir" G(t) dtl ~ LX) IG(t)1dt.
where q" Ijt, q,', and Ijt' are continuous and the derivatives do not vanish
simultaneously for any of the values of t, or a curve formed by joining a
finite number of such ares endwise. Then e is called a contour. If e is closed
and do es not intersect itself, it is called a closed contour; boundaries of
triangles, rectangles, and circles are examples.
Ir the parametric equations (4) represent a contour e over which
a ~ t ~ b and if j(z) is sectionally continuous over e, that is, if the function
j[q,(t) + iljt(t)], a < t < b, is sectionally continuous, then the contour
integral of j over e can be defined in terms of an integral of type (1), namely
(5)
fe j(z) dz = lb f[ q,(t) + iljt(t)][q,'(t) + iljt'(t)]dt.
(6)
lfe j(z) dzl ~ fe If(z)lldzl~ ML,
if If(z)1~ M for all z on e, L is the lengthof e, and Idzlis the elementof
are length of e :Idzl= Iq,'(t) + W(t)1 dt = ds.
Using the components u(x,y) and v(x,y) ofj(z), we can write the contour
integral (5) in terms of real-line integrals over e as follows:
(7)
fe j(z) dz = fe (udx - v dy) + i fe (u dy + v dx),
57 INTEGRALTHEOREMS
We now state basic theorems on contour integrals of analytic functions.
(1) fe j(z) dz = O.
A simply connected domain D is an open-connected region such that
every closed contour within it endoses only points of D. The interior of a
closed contour and the entire z plane are examples, but the exterior of a
closed contour is not simply connected, nor is the domain between two
concentric cirdes.
The Cauchy integral theorem permits C to be any closed contour
within a simply connected domain throughout which f is analytic. The
theorem is easily extended so as to permit C to be replaced by the boundary
B of certain multiply connected regions, as follows.
Let Co denote a closed contour, and CI, C2, ..., Cn a finite number
of closed contours within Co with no interior points in common (Fig. 64).
Let R be the closed multiply connected region consisting of all points on
and within Co except for points interior to Cj(j = 1,2,..., n). Let B denote
the boundary of R, consisting of Co and all Cj described in such a direction
that the regio n R lies on the left of the boundary. Then iff is analytic at all
points of R, it is true that
(2)
fB f(z) dz = O.
The value offat each point interior to a closed contour C is determined
by the values on C by the following theorem.
(3)
1
f(zo) = ---: -
if(z)
2m e z - Zo
dz,
where the integration is counterclockwise around C.
y
o Fig.64
FUNCTIONS OF A COMPLEX VARIABLE [SECo 58 171
(5)
f z
ZO
f(~) d~ = F(z),
fZO
f(z) dz = F(Zl) - F(zo)
58 POWER SERIES
Let f be analytic over the interior of a circle Iz - zol = ro. Then at each
interior point z it is true that
(1)
(2) L an(z -
n=O
zo)n,
where an are complex constants, is always a disk about the point Zo,
-
Iz zol < ro, unless the series converges only when z = Zoor for all z. Let
f(z) denote the sum of the series. The series can be differentiated term by
172 SEC, 58] OPERATIONAL MATHEMATICS
term to give a power series that converges over that disk to j'(z). Thus
series (2) represents an analytic function over that disk; in fact, it is Taylor's
series representation, about the point Zo, of its sum f(z). Thus
l
an = ,f(nJ(zo) (n = 1,2,. ..).
n.
The power series (2) converges absolutely and uniformly with respect to z
over each interior disk Iz - zol ~ rl, where rl < ro.
Since is an entire function, its Maclaurin series (zo = O),
ex> l
(3) e% = l + _zn (lzl < (0),
n= 1 n!
represents that function for all z. Consequently,the function g(z) = exp (Z2)
has the power series representation
A l
(5)
n = 2ni c(zJ -f(z)zot+1 d
Z
(n = O,:t 1, :t2,.. .).
Fig.65 o x
convergent series of form (4) is necessarily the Laurent series, and hence
the values of the integrals (5) can be seen from such a representation.
For example, from Maclaurin's series (3)for eZwe can write this Laurent
series representation:
(6) exp ~= 1+ ~Z + ~2! ~Z2 + ~3! ~Z3 + ... (lzl > O).
( Z)
Here A_1 = 1, A_2 = 1, and Zo= O. According to formula (5), if e is a
closed contour described counterclockwise around the origin, then
A-I A_2
(1) f( Z) = -Z - Zo
+
(z - Zo)
2 + . .. + Ao + Al (Z - Zo) +...
(O< Iz - zol < ro),
(3)
fe f(z) dz = 21ti(P1 + P2 + . . . + pm>,
where Pn denotes the residue off at Znand where the integration is in the
positive direction around C. Note that the singular points Znare necessarily
isolated because of their finite number.
In the representation (1) the series of negative powers of (z - zo) is
called the principal part of f(z) about the isolated singular point Zo. The
point Zois an essential singular point offifthe principal part has an infinite
number ofnonvanishingterms. It is a poleofordermirA - m =1= Oand A-n = O
when n > m. It is called a simple pole when m = 1; thus, if Zo is a simple pole,
a number ro exists such that
The function exp (I/z) has an essential singular point at the origin, in
view of its representation (6), Seco58. The function
cos z l 00 (- I)n l l Z2
-=-
z2
L:
Z2n=O (2n)!
-Z2n=___+-+...
Z2 2! 4!
(lzl > O)
From the Laurent expansion (1) offabout Zoit follows that Zois a removable
singular point of 4>,that 4>is analytic at Zoif we make the definition
(6)
Then <pis represented by Taylor's series when Iz - zol < ro, and it can be
seen from Eq. (5)that
(7)
This is a useful formula for the residue off at a pole. For a simple pole
(m = 1) it becomes
(8) A -1 = 4>(zo) = lim (z -
%~%o
zo)f(z).
-
We also note that If(z)1 OCJas z -
Zo whenever Zo is a pole.
Conversely, for a givenj, suppose that a positive integer m exists such
that the function 4>,defined by Eq. (5), is analytic at Zowhen 4>(zo)is suitably
defined and that this value 4>(zo):F O. Then Taylor's series for 4>shows that
f has a pole of order m at Zoand that the residue offthere is given by formula
(7), or by (8) if m = l. For example, if
e-Z e-Z - l
f(z) = Z2+ n2 = z + ni z - ni
176 SECo 59] OPERATIONAL MATHEMATICS
PROBLEMS
1 . When e is the boundary of a square with opposite vertices at the points z = Oand
z = 1 + i, evaluate the integral f e Z2dz directly in terms of realline integrals to show that
its value is zero.
2. When e is the circle Izl = 2 described counterclockwise, use Cauchy's integral
theorem or Cauchy's integral formula to prove that
(a) r~
Jcz + 9
dz = O;
(b) fe sec~dz = O;
(e)
i_Z2 + 4
1 ao
(b) ~=
1 z - I
.=0
Z2. (lzl < 1);
ao Z2.-
(e) sinhz = I
n=1
(lzl< (0);
Z ao 1
(lzl> 1).
(d) z - 1 = .~o?
FUNCTlONS OF A COMPLEX VARIABLE [SECo 60 177
_21 '
i eZl
-:--
1tI e Sinh z
dz = 1- 2 cos 1tt + 2 cos 21tt.
(3)
liz) = ~(cos ~ + i sin~) (n < O< 3n, r > O).
Here 12(Z)= -11 (z). The number of branches is unlimited. A branch with
the positive real axis as its branch cut is
(5) zl/n = ~(cos ~ + i sin~) (00 < O < 00 + 2n, r > O).
o x
Fig.66
FUNCTIONS OF A COMPlEX VARIABLE [SECo 61 179
fl(Z) I z'
= .=0 (lzl< 1).
Then f is analytic in the region Izl < 1,the region of convergence ofthe series;
but the function is undefined for all other values of z because the series
diverges whenever Izl ;;;;;1. The series is the Mac1aurin series representing
the function (1 - Z)-1 in the region; thus f(z) = (1 - Z)-1 when Izl < 1.
180 SECo 61] OPERATIONAl MATHEMATICS
off1 into the entire z plane, excluding the point z = 1,becausefis everywhere
analytic except at that point and f(z) = f1 (z) when Izl < 1.
As another example, consider the Laplace transform
We also note that ifthe Laplace integral were known to represent an analytic
function in the half plane and to have the value l/x when z = x there, then
the integral represents l/z there because the latter function is analytic and
has the values l/x on the x axis.
Finally, consider the branch (2), Seco 60, of the function zt. In the
half plane y > O it is given by the conditions
is ana1ytic except on the positive real axis, and its values coincide with those
off1 in the half plane y > O. Thus f is the analytic continuation of f1 down-
ward across the negative real axis.
The following principie of rejlection is easily established with the aid
of the Cauchy- Riemann conditions. Let a function w = f(z) be analytic in
. some region R that includes a segment of the x axis and is symmetricwith
respect to that axis. Hf(x) is real whenever x is a point on that segment, then
(1) f(z) = w
wheneverz is in R. Conversely,if conditionl!lis satisfied,thenf(x) is real.
Condition (1) can be written f(z) = f(z), or f(z) = f(z).
FUNCTIONS OF A COMPLEX VARIABLE [SECo 62 181
As examples, the entire functions Z2, eZ, and sin z are real when z is
real, and the complex conjugate of each function is the same as the function
of Z. But the entire functions iz and Z2 + i are not real when z = x; their
conjugates are different from iz and Z2 + i.
-y+i/3
R x
Fig.67 -y-i/3
182 SECo 62] OPERATIONAL MATHEMATICS
(2) f(zo) = ~ [f
2m CRz
f(z)
- Zo
dz -
j Y+ifJ f(z)
y-ifJ Z - Zo
dZ
],
where C R is described in the positive sense.
Whenpoint z is on CR,then Iz - zol~ R - IZol and, since If(z)1<
Mlzl- k when Izl = R and R is sufficiently large, it follows that
f(z) M 1
- <
I z - Zo 1
RkR - IZoI whenR>ro.
The integrands of the integrals in formula (2) are continuous. The length of
CR is less than 2nR; therefore (Sec. 56)
I fCRz
f(z)
- Zo z
d
I
2nRM
< Rk(R- IZol)- Rk(l - IZolR
-1)
- 2nM
and the last member clearly tends to zero as R -+ 00,since k > O. It also
tends to zero as p -+ 00, because R2 = p2 + y2. Sincef(zo) is independent
of p, we see from formula (2) that
1 . Y+ifJ f(z)
(3) f(zo) = ---; bm
j
2m fJ~oo y-ifJ Z - Zo
- dz;
1 . f(y + iy) d
+ Jp f(y
.- iy) d
].
fJ
= -- l 1m . y y
2n fJ~ 00 [J o Y + IY - Zo o Y - IY - Zo
The integrands of the last two integrals are of the order of 1/1+ 1 as
y -+ 00,and sincek + 1 > 1, the improper integrals, as p -+ 00, exist and
(4) f(zo) = -~
2n [i oo f(Y. + iy) dy +
o y + IY - Zo I o f(Y. + iy) dY
-00 y + IY - Zo ] ,
cP(z)= f(z)
z - Zo
along the line x = y is the residuef(zo) of cPat its singular point Zo. The proof
was based upon the analyticity of cP(z)(z - zo) throughout the half plane and
the fact that cPis (f}(l/zk+ 1) there. But Taylor's series.,representation of the
analytic functionf about the point Zo shows that Zo is a removable singular
point of cPif and only iff(zo) = O. Thus if cP is also analytic at zo, its integral
along the line x = y is zero, the value f(zo), where f(z) = (z - zo)cP(z), and
we have established the following extension of the Cauchy integral theorem.
FUNCTIONS OF A COMPLEX VARIABLE [SECo 62 183
Theorem 2 Let cjJbe analytic over a half plane x ~ y and (9(Z-k) as Izl -+ 00
there, where k > l. Then
Y+iOJ OJ
(5)
f y-iOJ cjJ(z)dz=i f -OJcjJ(Y+ iy)dy=O.
The proof of formula (5) applies as well when cjJis analytic and (9(Z-k)
in the left half plane x ~ y, if k > 1. Also, the residue theorem (Sec. 59) can
be extended in the same manner to the case in which cjJhas a finite number
of singular points interior to either half planeo For the left half plane that
extension of the residue theorem can be stated as follows.
(6)
f Y-IOJ
. cjJ(z)
dz = 2ni(pl + P2 + . . . + Pn).
l-iOJ
2 =
Z
"
I
f OJ
-OJ (1
dy
. 2 = O.
+ IY)
When t is real and t ~ Oand y = 1, the conditions in Theorem 3 are
satisfiedby the function eZl/(Z2+ 1)with simple poles :!:i; therefore
1 1 + OJ eZt eit e - it .
2ni f 1- OJZ2 + 1
(8) - -dz=---=smt (t ~ O).
2i 2i
As we shall see in the following chapter, the left-hand member of formula (8)
is an integral representation of the inverse Laplace transform of 1/(s2+ 1).
184 SECo 62] OPERATIONAL MATHEMATICS
PROBLEMS
1. Use Maclaurin's series for cos Z to prove that the function cos.j; is single-valued
and entirewhendefinedas l at z = O.
2. Point out why the function
f Y+lOO z"g2(Z)
y-loo
dz = O (n = 0,1,2,...;)/ > O).
is analytic over the halfplane x > Oand that its analytic continuation into the rest ofthe
z plane,excludingthe points :t i, is the functionfez) = 1/(Z2+ 1). .
f-100~-_?:!.
lOO
Z2- k2- k
(k > O).
.
+iP lP+iP
o 'Y
+-. c
{3-i{3
Fig.68
12. Use the residue theorem (Sec. 59) and give details of the proof of Theorem 3,
Seco62.
6
The Inversion Integral
We shall now extend our theory ofthe Laplace transformation by letting the
letter s in the transform f(s) represent a complex variable. As before F(t)
represents a real-valued function of the positive real variable t; but the
transformf(s) can assume complex values. We shall see that properties ofthe
transformation airead y obtained by assuming that s is real carry over to the
case in which s is complex.
63 ANALYTIC TRANSFORMS
When s is a complex variable,
s = x + iy,
the Laplace transform of a real-valuedfunction F(t),
(1)
186
THE INVERSION INTEGRAL [SECo 63 187
because the integral here converges uniformly with respect to x and y when
x ~ xo(xo > a); also Uxis continuous there. From the second of Eqs. (3)
we see that the integral (6) also represents v.(x,y); hence u and v satisfy the
first Cauchy-Riemann condition over the half plane x > a. In the same
manner we find that the second condition Uy = - Vxis satisfied there. Those
conditions along with the continuity of the first-order partial deriva tives of u
and v establish the analyticity of f over the half plane x > a (Sec. 54).
188 SECo 64] OPERATIONAL MATHEMATICS
I'(S) = - foO:>
te-xt(COS yt - i sin yt)F(t) dt = L{ - tF(t)}.
Since tF(t) is also sectionalIy continuous and of exponential order, our
result applies as welIto that function; thus
(x> rx),
(s = x + iy)
f(s) = {O:>e-stF(t) dt = L{F}
is an analytic function of s over the half plane x > rx. Its Laplace
integral is absolutely and unlformly convergent over each half plane
x ~ xo(xo > rx),and f and its derivatives are bounded there. Also,
(7) <nJ(s)= L{( -ttF(t)} (n = 1,2,...; x > rx),
(8) f(s) = f(s) (x> rx).
Formula (7) was derived in Seco19 when s is real.
The conditions in Theorem l can be relaxed. For instance F may
become infinite at t = to in such a way that (t - to)kF(t) remains bounded as
t -+ to, where k < 1. Then the conclusions in the theorem are still valid. As
.an example, the transform of t-t is analytic in the half plane x > O, and
formula (7) applies to it.
64 PERMANENCEOF FORMS
We have seen that the Laplace integral of F(t) represents a functionf(s) that
is analytic in a half plane x > ex,where s = x + iy. Ir the integration is
THE INVERSION INTEGRAL [SECo 65 189
LHF(t)} = 1(%)J(J.)dJ.,
and agree that s is real here.
The permanence of the forms of those properties is again a consequence
ofthe fact that the steps used in their derivations are independent ofthe real
or complex character of the parameter s. However, the derivations could be
rewritten when s = x + iy by taking the corresponding steps with the real
integrals (3), Seco63, that represent u(x,y) and v(x,y).
For points s =x + iy in the half plane x ;;;Xo we are to prove that for
each positive number E there is a number '. such that
(2)
/f(s)1 < E wheneverIsl> '. and x ;;;Xo.
The Laplace integral of F converges uniformly with respect to s in the
half plane x ;;;Xo (Theorem 1), so the remainder for that improper integral
can be made uniformly small in absolute value for all s there. That is, to the
given number E there corresponds a number T. independent of s such that
-s fo e-SIF(t)dt
TI (TI
= e-S'F(t)J o - Jo e-S'F'(t)dt.
The absolute value ofthe right-hand member does not exceed the function
(TI
iflsl > 2M
E '
that is, condition (2) is established by selecting r. as 2M/E.
Weaker-order properties onf(s), such as the boundedness of xf(x + iy),
can be derived by assuming only that F itself is sectionally continuous and
of exponential order (Probs. 7 and 8). Stronger ones will now be noted in
case F or its derivatives are continuous and of exponential order.
THE INVERSION INTEGRAL [SECo 65 191
Theorem 3 If F is continuous and F' and F" are sectionally continuous over
each intervalO ;;;t ;;;T while both F and F' are (9(e"'t),then
Theorem 4 If F and F' are continuous and F" and F'" are sectionally con-
tinuous over each intervalO ;;;t ;;;T while F, F', and F"are (9(e"'t),then
s2f(s) - sF(O)is bounded over any half plane x ~ Xo where Xo > oc;in
fact
(8) lim [s2f(s)
1.1 00
- sF(O)] = F'(O) (x ~ Xo > oc),
PROBlEMS
1 . State why none of the following functions can be Laplace transforms of real-valued
sectionally continuous functions of exponential order (Theorem 1).
1 1
(a) s; (b) ssins
+ l'. (ti) 7;
S
(e)~.
S+I
2. Use Theorem 2 to show that none of the following functions can be Laplace
transforms of a function F of exponential order such that F and F' are sectionally
continuous.
s
(a) 1; (b) s - 2;
3. Determine an order property of the transform f(s) from the character of the
function F in each of the following cases, and verify by writing f(s). (a) F(t) = sin t;
(b) F(t) = cosh t; (e) F(t) = t sin t; (ti) F(t) = (t - 1)80(t - 1).
l l 1
Ans. (a) (!J2'
s x ~ Xo > O; (b) (!J-,x
s ~ Xo > 1; (e) (!J3'X
s ~ Xo > O.
then represent f as a power series in s convergent for all s to verify that f is entire.
6. Ir a function F is sectionally continuous over some interval O < t < T but con-
tinuous when t ~ T, and ifF' is sectionally continuous over every such interval and both
F and F' are (!J(e"'),use the formulafor L{F'} to prove that the transformf(s) is (!J(I/s)
over the half plane Re s ~ Xo when Xo > IX.
7. Ir F is sectionally continuous over each interval O < t < T and (!J(e"'), prove that its
transform f(s) satisfies the condition that xf(x + iy) is bounded over the half plane
x ~ Xowhen Xo > IX;consequently
lim f(x + iy)
x-a>
= o.
8. According to the Riemann-Lebesgue theorem in the theory of convergence ofFourier
series,l if a function G is sectionally continuous over an interval a < t < b, then
lim
y co f G(t)cos yt dt = lim f
b
a y oo a
b G(t)sin yt dt = O,
a result that is plausible if graphs of the integrands for large values of y are considered.
, See the author's "Fourier Series and Boundary Value Problems," 2d OO.,p. 87, 1963.
THE INVERSION INTEGRAL [SECo 66 193
f
eyl (l.
(3) _
2 lim e'Y'f(y+ iy) dy,
1t (1 00 -(1
does existo This is illustrated in Probo 7, Seco70. Ir the improper integral (2)
does exist, its principal value exists and is equal to the improper integral.
The principal value (3) is called the complexinversionintegral for the
Laplace transformation. We use the symbol Li - 1 for that linear integral
transformation of f(s), a symbol that is intended to suggest an integration
as well as an inverse Laplace transformation; thus
1 Y+i(l
(4) Li - 1 {J(s)} = ---: lim
1 i(l e'Zf(z)dz.
2m (1 <X>Y-
194 SECo 66] OPERATlONAL MATHEMATlCS
y
-y+i(3
z=-y+iy
o x
-
-y i(3
Fig.69
When f(s) satisfies certain conditions, we shall see that the value of
the real integral (5) is independent of y for all values of y greater than a
prescribed number. Even for simple functions f, however, the integration
involved in the real form (5) is generally difficult. To evaluate the inversion
integral, we shalI use the complex form (4) in conjunction with the theory of
residues or processes of changing the path of integration.
THE INVERSION INTEGRAL [SECo 67 195
67 CONDITIONS ON I(s)
The following theorem gives conditions on a function J that are sufficient
to ensure that L- 1 {J(s)} exists and represents a function F whose transform
is J(s). Useful properties of F(t) also follow from the conditions on J(s),
properties that give the initial value F(O), the continuity and exponential
order of F.
Since k > 1, the improper integral s of Mlyl-k from Yo to ex)and from - ex)
to - Yo exist, so the inversion integral exists as an improper integral, and
it is equal to its principal value.
1 See books by Doetsch and by Widder listed in the Bibliography.
196 SECo 67] OPERATIONAL MATHEMATICS
AIso, since f(x) is real, then, according to the reflection principIe for
analytic functions,f(z) = f(z), and the inversion integral takes its real form
00
eyt
(5) L-l{J(S)} = - [u(y,y)cos ty - v(y,y) sin ty] dy.
1t 1o
Now the components u(y,y) and v(y,y) of f(y + iy), as well as f itself, satisfy
condition (4). Thus the absolute value of the integrand of integral (5) is less
than 2My-k when y > Yo, a function that is integrable from Yo to 00 and
independent of t. Consequently, integral (5) converges uniforrnly with
respect to t, and it follows that L-1{J} is a continuous real-valued function
F(t) for all rt.al t.
To show that F(t) is independent of y when y ~ el, we use a second
path ofintegration x = y',where y' > y. Sinceetzf(z)is analytic when x ~ y,
the integral of that function around the boundary of the rectangle ABCD
in Fig. 70 vanishes, according to Cauchy's integral theorem. On sides BC
and AD, Izl ~ fJand, since If(z)1< Mlzl-\
M N(t)
letz' (z)1 < etx- <-
J fJk = fJk
where N(t) is the greater ofthe two numbers Mety and Mety'. The length of
-
those sides is y' y; hence the absolute value of the integral of etzf(z) along
either side BC or AD does not exceed N(t)(y' - y)fJ-k,a quantity that vanishes
as fJ -+ oo.
The vanishing of the contour integral around the boundary can be
expressed in the form
fAB
etzf(z) dz +
iCD
etzf(z)dz =
f AD
etzf(z) dz r etzf(z) dz.
- JBC
o 'Y x
A DI'Y'-iP
I I Fig.70
THE INVERSION INTEGRAL [SECo 67 197
The left-hand member tends to zero as P-+ 00 because the member on the
right does so. But we have seen that the limit as P-+ 00 of each of the
integrals on the left exists when k > 1. It follows that
Y+iP Y'+IP
p
lim f
oo y-IP
e'Zf(z)dz = lim
p f
oo y'-IP
e'Zf(z)dz,
- 00
.
eUYf(1X+ iy) dy ~ -
I 2n - 00f If(1X+ iy)1dy.
The last integral exists because f is lP(y-k) as Iy/ -+ 00 where k > 1, and f
is continuous. Its value is independent of t. Therefore F(t) is lP(eCZ/)
as t -+ oo.
To evaluate F when t ~ O,it is convenient to select a positive number
for y; this can be done since y can be any number in the range y ~ IX.Then
over the half plane x ~ y, le'zl = e'" ~ l when t ~ Oand
le'Zf(z)1< Mlzl-k where k > 1.
It follows from our extension of Cauchy's integral theorem (Theorem 2,
Sec.62)that
Y+loo
f Y-JC()
. e'Zf(z) dz = O (t ~ O);
when z = IX+ iy. Since the convergence of the inversion integral is uniform
with respect to t, the order of integration can be interchanged:
oo T
l
(6) - L{F}=-lim
2nT
f(z)
00 f f
e-(S-z)ldtdy
-00
oo
o
= .2- lim
2n T 00 f- [ 00
f(z) - f(z) e-(S-Z)T
s - z s- z ]
dy.
I
f-",s-z
'" f(z) e-(s-Z)T dy
l ~ e-(~a)T
a (Xf'"-'" If(z)1 dy.
The member on the right vanishes as T -. 00 becausefis absolutely integrable
along the line x = (x. Thus
(8) L{ F(t)} = L{L-1[f]} = f(s) (Re s > (X).
The proof of Theorem 5 is now complete.
68 CONDITIONS ON F{t)
The foregoing conditions under which the inversion integral formula is
valid are severe. They are not satisfied, for example, by the function l/s,
which is @(l/sk) where k = 1, nor by transforms of functions F that are
discontinuous or for which F(O):f: O. By using a Fourier integral theorem
and qualifying the function F instead of f, we can relax the conditions so
that the inversion integral formula can be seen to apply in nearly all cases
of interest to uso In fact we shall see that our formula is only a modified
form of the Fourier integral formula.
Let G be a function defined for all real t, sectionally continuous over
each bounded interval ofthe taxis, and absolutely integrable from - 00 to oo.
We agree to define G(t) at each point to where G is discontinuous as the mean
value of its limits from the right and left at to:
(1) G(to) = ![G(to + O)+ G(to - O)].
Also let the derivative G' be sectionally continuous over each bounded
interval; this condition actually implies that G itself is sectionally continuous
(Prob. 10, Seco70).
Then G(t) is represented in terms of trigonometric functions of t by
the F ourier integral formula 1
1 "' '"
(2) G(t) = -
1t o f f
G('r)cos y(t - .) d. dy
-00
(-oo<t<oo)
1 p oo. .
= - lim [G(.)e,y(t-f)+ G(.)e-,y(t-f)]d. dy.
f. f -'"
21tP-+00o
Since the integral of IG(.)I from - 00 to 00 exists, the inner integralof each
term within the brackets converges uniformly with respect to y to a con-
1 See, for instance, the author's "Fourier Series and Boundary Value Problems," 2d ed., pp. 83,
115 Ir., 1963.
THE INVERSION INTEGRAL [SECo 68 199
p eiyt
f.o f co
-co
G(r)e-iy<d1:dy +
f o
-P
eiYI
f co
-co
G(1:)e-iYt d1: dy.
Note that the existence of the principal value of the improper integral with
respect to y from - 00 to 00 is ensured here, but not the convergence of the
improper integral itself.
Now let F denote a function defined when t ;;;O and (9(eat)as t -+ 00,
such that F', and therefore F itself, is sectionally continuous over each interval
O < t < T. If we write
The inner integral represents f(y + iy), where f(s) is the Laplace transform
of F(t). Ir z = y + iy, it follows that, for all real t,
I Y+iP
(6) eY1G(t) = : lim e'Zf(z) dz = Li -1{J(S)}.
2m P-co Jy-ip
When t < O, G(t) = O,and therefore Li -1{J(S)} = O.
When t > O,eY1G(t)= F(t), and Eq. (6) represents F(t) as the inversion
integral Li-1 {J(s)}. But sincethe function G representedby formula (5)has
the mean value (1) at eachpoint to of discontinuity, the inversion integral
converges to F(to) at a point to where F has a jump, provided we agree that
(7) F(to) = t[F(to + O)+ F(to - O)] (to > O).
When t = O,the value of the inversion integral is ![F(+ O)+ O].
The existence of the inversion integral (5)in the form of the principal
value of the improper integral along the line x = y was established here,
but the improper integral itself may not exist under the conditions stated.
The following theorem is now established.
intervalO < t < T, then the inversion integral of f along any line
Re s = y, where y > oc,exists and represents F(t):
(8) (t > O).
At a point to(to > O) where F is discontinuous,the inversion integral
has the mean value (7) of F(t); when t = O, it has the value tF( + O),
and whent < O,it has the valuezero.
The conditions here on F can be relaxed because the conditions on G
for the validity of the Fourier integral formula (2) can be modified in various
ways.l For instance, on each bounded interval let the function G have at
most a finite number of infinite discontinuities but still be absolutely inte-
grable over the entire taxis, where each interval can be subdivided into a
finite number of open intervals such that G is monotonic (nondecreasing or
nonincreasing) over each one. Then the Fourier integral expression (2) has
the value t[G(t + O)+ G(t - O)] at each point t where G(t + O)and G(t - O)
exist.
As a consequence of those m9difiedconditions, for instance,
because the function G(t) = e-ytct when t> O, G(t)= O when t < O, is
represented by its Fourier integral formula when t > O and y > O, and
because L{t-i} = (n/s)i when Re s> O.
Even though its conditions are on F, the function usually sought,
rather thanf, Theorem 6 serves as a useful supplement to Theorem 5. The
following example and corollary illustrate this.
Example Let,j'; denote the branch J;eiOI2, r > O, 101< n/2, where
s = reiO.Show that the inversionintegral formula applies,when y > O,
to the function
f(s) =
sys+l
f .
On a half plane Re s ~ oc> O, the branch ,j'; is analytic and
s,j'; + 1 never vanishes sin<::eRe (s + l/,j';) > O. Thus f is analytic
there but only (9(1/s), which fails to satisfy the order condition in
Theorem 5. By carrying out one step of a division, however, we can
write f as a sum of two components,
1 1
(9) f(s) = - - -
S S2,j'; +s
1 See, for instance, H. S. Carslaw, "Fourier's Series and Integrals," chap. 10, 1930, and E. C.
Titchmarsh, "Theory of Fourier Integrals," p. 13, 1937.
"
(t ~ O)
(2) F(t) = f~[G(r) - H(!)] d!
properties of the function f The two theorems in this section are useful for
that purpose. Their proofs foIlow from Theorem 5 and properties of uni-
formly convergent integrals (Sec.15).
When the inversion integral in Theorem 5 is differentiated with respect
to t under the integral sign, we obtain L -1 {sf(s)} :
_1 a I oo
f f
oo
(1) -;-[etzf(z)]dy = 2
_ etzzf(z) dy,
21/: - 00ut 1/: - 00
F1(t) = L-I
{ ;/+ }.
S2 s s
of transfonning derivatives,
p<n)(t)= L-1{s"f(s) - s"-1F(O) - ... - F(n-1)(0)}.
If F(n)(t)is to be the inverse transform of s"f(s), then it is necessary that
F(O),. .. , F(n-1)(0) all vanish.
Of course the function f in the above theorem may involve variables
independent of s provided the statements in the theorem are understood to
apply for fixed values of those parameters. Conceming differentiation and
continuity with respect to such a parameter p, the following theorem can be
stated from properties of unifonnly convergent integrals.
PROBLEMS
1. Apply Theorem5 to the functions(a)f(s) = (S2 - 1)-1 and (b)g(s)= s-2e-s to
show that their inverse transforms F(t) and G(t) are represented by inversion integrals,
also to find order and continuity properties of F and G and to show that F(O)= G(O)= O.
Verify those properties by writing F and G as known inverse transforms.
2. ApplyTheorem8 to the functionf(s) = S-4 to deduce that the inverse transform
F(t) and its derivatives offirst and second order at least are continuous when t ;; O,and
19(e"")if IX> O, and that F(O) = F'(O) = F"(O) = O. Verify those properties by recaIling
that F(t) = t3/6.
THE INVERSION INTEGRAL [SECo 70 205
3. Apply Theorem 6 to the transform of the function e-' and simplify the inversion
integral, when }' = O,to establish the integration formula
lo
eo cos ty + y sin ty d
2
Y + 1
y=ne
-, (t > O).
4. When f(s) in Theorem 5 is the transform (s + 1)-2 of te-', simplify the inversion
integral, when )' = O,to pro ve the integration formula
lo
eo O - y2)cos ty + 2y sin ty d = nte-t
(y2 + 1)2 Y
(t ;;;;O).
5. Let ~ denote the branch j;eiBI2, -n < IJ< n, r > O,where s = reiO,an analytic
function over each half plane Re s ;;;;IX> O. If f(s) = exp (-~), prove that If(s)1 <
exp(-.;i) overthe halfplane and thatf(s) is l!I(s-m)there for every positive integer m.
Show that Theorem 8 applies to establish f(s) as the transform of a function F(t) with
continuous derivatives of all order (t ;;;;O)such that
F(O) = O and F(n)(o)= O (n = 1,2,.. .).
[Compare the transformation (6), Seco27.]
6. Use power series to prove that the function g, where
is entire and hence bounded over a neighborhood of the origin in the z plane. Then
apply our corollary, Seco68, to the function
f(s) = !e-lls
s = !s - sO - S2e-l/')
to prove that L -1{J} is a continuous function F(t) when t > O,l!I(e.')if IX> O,and that
F(+O) = 1.
7. State why the inversion integral formula applies when}' = 1to the known transform
f(s) = s-le-s for all t; but show that when t = 1 the Cauchy principal value Li-l{J}
must be used because the improper integral itself fails to exist for that value of t. AIso
verify that, when t = 1, Li-l{J(S)} = t.
8. The entire functiong(s) = exp (S2)is not l!I(S-k)over a right half plane for any
positive k. But prove that Li-l{g}, integrated along the imaginary axis, is the function
eo
l
F(t) = -n f
o
exp (- y2) cos ty dy.
A known integration formula [Eq. (7), Seco74] shows that F(t) = tn-t exp (- t2/4), and
thus (transformation 109,Appendix A, Table A.2)
satisfies all conditions for its representation by the exponential form of the Fourier
integral formula. When t = O,however, show that the Cauchy principal value (3),Seco68,
must be used in that representation, for the improper integral itself,with respect to y, does
not existo
10. Prove that F(t) itself is sectionally continuous over an interval if its derivative
F'(t) is sectionally continuous there. Note that F' is continuous over each of a finite
number of subintervals, such as a ~ t ~ b, when properly defined at the end points, so
F is continuous when a < t < b and, when a < to < b, then
Throughout this and the following two sections, f(s) denotes a function
that is analytic for all finite values of the complex variable s except for a
set of isolated singular points
confined to some left half plane Re s < y. We assume also thatf(s) satisfies
conditions under which its inversion integral along the line x = y converges
to the inverse transform F(t), say the conditions in either Theorem 5, Seco67,
or Theorem 6, Seco68. Then F(t) can be represented formally by a series,
finite or infinite, depending on the number of singular points, and we shall
establish practical conditions under which the representation is valido
Since etz is an entire function of z, the singular points z = Snoff(z) are
the singular points of the integrand etzf(z) of the inversion integral. Let
Pn(t) denote the residue of the integrand, for any fixed t, at the isolated
singular point Sn:
Let the path be made up of the line segment joining the points y - iPN,
Y + iPN,and some contour CNbeginning at the second and ending at the
THE INVERSION INTEGRAL [SECo 71 207
52
O
Fig.71
first of these two points and Iying in the half plane x ~ y (Fig. 71). Then
N
l
(2) _2l .f
nI
Y+PN
y-PN
etzf(z)dz + _
2nI '
fCN
etzf(z) dz = L Pn(t).
n=1
As PN- 00, the value of the first integral here tends to L-1{J(S)},
sincethe inversionintegralis the limitofthe correspondingintegralinvolving
--
P,as P 00 in any manner. Let the numbers PN(N = 1,2,...) be selected
so that PN 00, and let the curves CNtogether with the line x = y endose
the points S1, S2, ... , SN, if the number of singular points is infinite. If the
number is finite, let all of them be endosed when N is greater than some
fixed number. Then if j(z) and CNsatisfy additional conditions under which
_1 . lim
(3)
f
2nI N-oo CNetzf(z) dz = O,
it follows, by letting N - 00 in Eq. (2), that 00
L-1{J(S)} = L Pn(t),
n= 1
or by a finite series in case the number of singular points is finite. The series
on the right is necessarilyconvergent because the limit, as N 00, of the
left-hand member of Eq. (2) exists.
-
Since the inversion integral represents F(t) by hypothesis, the inverse
transform off (s)is represented as the series of residues of etzf(z):
00
(4) F(t) L1 Pn(t).
= n=
It is not essential that the line x = y and the curve CNendose exactIy
N of the poles, of course. F or example, if two poles are induded in the ring
208 SECo 72] OPERATIONAL MATHEMATICS
between CN and CN+1, the residues at these two poles are simply grouped as
a single term in the series.
It is sometimes convenient to use the series representation (4) directly
and formally, without regard to conditions under which the inversion
integral represents F(t) and the integral over CN tends to zero. The function
F so obtained may be such that its transform can be shown to be f(s), or
such that it satisfies all conditions in a problem whose solution was sought
by the transformation method. As we shall see in the problems at the end
of this chapter, however, there are simple cases in which such a formal
procedure leads to incorrect results.
72 RESIDUES AT POLES
At singular points Sn' which are poles of the functionf, useful formulas for
the residues Pn(t) of e'Zf(z) can be found.
Let Sn be a pole of f of order m. Over a neighborhood of that pole
with the point Snitself excIuded (O< Iz - snl < cJ, f is represented by a
Laurent series
Taylor's series expansion ofthe entire function e'z about the point Snis
~-1
(3) [
e'z. = exp(snt) 1 + t(z - Sn) + ... + (m - 1)!(Z - Snr-1 + ... . ]
The coefficientof (z - sn)-l in the Laurent expansion of e'zf(z) over the
domain O< Iz - snl < cn, the residue Pn(t) of that function at Sn' is the
coefficientof (z - Snr- 1 in the product of the power series(2)and (3). Thus
we find that
~-1
~
Pn(t) [
= exp(snt) A-1 + tA_2 +... + (m - 1)!A-m . ]
Formula (4) has a useful feature. It presents the residue of e''i(z) at a
pole Snof order m as the inverse Laplace transform, with respect to z, of the
principal part of the Laurent expansion (1) of the function f
When sn is a simple pole (m = 1) of f
(5) Pn(t) = exp (snt)A":: 1 = exp (snt) lim [(z - sn)f(z)].
z s,.
THE INVERSION INTEGRAL [SECo 72 209
(8) F(t) = L - 1
P S)
{
qS )} = f P,
n= 1q Sn)
Sn)
exp (snt).
In caseP and q are polynomials, the number of poles offis finite, and formula
(8) becomes Heaviside's expansion (4), Seco24.
Assume now that f(z) = f(z) except at the singular points off, all of
which are isolated, a condition that is satisfied if f(z) is real over the real
axis except at singular points (Sec. 61). Let SI = a + ib, where b :i=O, be a
pole off of order m. Then over a domain O < Iz - sIl < c1,
(10)
when 0< Iz - sIl < Cl, that is, when 0< Iz - sIl < Cl. By writing S for
z here, we see that f(s) has a Laurent series expansion in the domain
O < Is - sIl < Cl with the A's as coefficients, and, since A-m :i=O, SI is also
a pole off of order m. According to formula (4) then, the sum of the residues
of etzf(z) at the poles S1 = a + ib and S1is P1(t) + P1(t), where
m-l
...
-
Pl(t) - e
(a+ib)t
[A_l + tA_2 + + (m -
t
I)!A-m ] .
Therefore the component of F(t) corresponding to the pair of poles a :t ib
(b :i=O) of f(s) is the real-valued function
Let us write the residue A - 1 off at a pole SI =a+ ib in its polar form as
(12) A_1 = r1 exp (iOl).
Ir s1is a simple pole, formula (11) then gives the component of F corresponding
to the poles a :t ib as
(13) Pl(t) + Pl(t) = 2e"trl cos (bt + (1) (b :F O,r > O),
a result that includes the component G(t) obtained in Seco26, wherefwas a
quotient of polynomials.
In case the singular points off consist only of the simple poles O, :t inro
(n = 1,2,.. .), wherefhas residue ro at z = Oand residues r" exp (i0,,)at inro,
formula (4), Seco71, becomes
00
(2) lim
f et%f(z)dz =O (t > O).
N-<e CN
y
'PN+iPN "(+i(3N
CN
-(3N "(
o %
(3N-i(3N "(-i(3N
Fig.72
212 SECo 73] OPERATIONAl MATHEMATICS
CIRCULAR ARCS
(4)
IfCN e'Zf(z)dz < ---
1 RN "/2 f exp (tRNeos O)RNdO
2M ,,/2
= --- exp (- tRN sin </J)RN
d</J.
RN 1o
The graph of sin </Jshows that sin </J> 2</J/nwhen O < </J< n/2. Therefore
the final member of eondition (4) is less than
2M "/2 A,. nM 1 - exp (-tRN)
f
</J
PARABOllC ARCS
(8)
I JCN
r eIZf(z)dz I
< 2~
aN f -aN e'xJX X ++ 2aN
y aN dx.
Let N be large enough that aN > 21yl. Then when t > O,
1 y + 2aN y JIx
aN f-taN f
(10) --- e d d
y IX! + 2aN x< aNkJaN/2 -taN e x
xx+aN
which tends to zero as N -+ 00 if t > O, so condition (2) is satisfied on the
parabolic ares. Again we find that the right-hand member of condition (8)
vanishes as N -+ 00 when t = Oin case k > l. Our results can be summarized
as follows.
The sum of the integral in formula (3) and the integral along the path
ACDD'C' A', consisting of the circular ares and line segments shown in
Fig. 73, is zero, since the integrand isanalytic except on the nonpositive real
axis. Thus if IAc denotes the integral of e'=j(z) over the arc AC, etc., we can
write
(4)
Let R and ro denote the radii of the large and small cicular arcs; thus
R2 = y2 +p2 so that P -+ 00 when R -+ oo. Along the arc AC, z = Rel,
dz = iReiOdO, and zt = JRei0/2. Hence the integrand of the integral is a
continuous function of (}whenever f. ~ 0, wheref. is the anglebetweenDC
or D'C' and the negative real axis. For any fixed R, the limit ofthe integrals
IAC and IA'c', as f.-+ 0, therefore exists. Likewise for any fixed positive ro
the limits of the integrals over the other parts of the path exist. Since formula
(4)is true for everypositive f. and the integral on the left is independent of f.,
it follows that we can let each of the integrals on the right have their limiting
valuesas f. -+ 0, and consider hereafterthe path in Fig. 74.
THE INVERSION INTEGRAL [SECo 74 215
y
'Y + ifJ
c
x
c'
'Y-ifJ
Fig.73 Fig.74
Write J AC = lim o JAC, and so on, for the integrals over the other
arcs and lines. We shall now let ro approach zero and R tend to infinity, so
that the left-hand member ofEq. (4),which is incidentally independent of ro,
becomes the inversion integral or F(t), except for a factor - 2ni.
When z is on the circle r = ro, Z = roei9,and the integral over that
circle can be written .
./R sin JL
= 4i f. exp ( - tJL2)- dJL.
o JL
Now on ares AB and A' B', Re (tz - zt) ~ ty; thus the integrand of the
integrals J AB and J B'A' satisfies the condition
and JC'B' satisfies the same condition. The value of the last integrand does
not exceed exp ( - 2tRCP/n),as pointed out under Circular Ares, Seco73, and
an elementary integration then shows that J BCand JC'B'tend to zero as
R -+ 00, if t > O.
The limit in formula (5) therefore vanishes, and
2 '" sinp.
(6) F(t) = l - - exp (-tp.2)-dp. (t > O).
f.
n o P.
'" l O(2
(7) exp (- tp.2)COSocp.dp. = - A- exp -- (t > O),
f.o 2 t ( 4t )
2 l/(2 jij
l
(8) 1 - erfc - .
F(t) = 1 - erf ( 2.fi } - ( 2-fi }
I See, for instance, R. V. Churchill, "Complex Variables and Applications," 2d ed., p. 171,
exercise 7, 1960. Formula (7) also follows from the result in Probo 7, Seco35.
THE INVERSION INTEGRAL [SECo 74 217
PROBLEMS
1 . Instead of simply using partial fractions, show that Theorems 5 and 10 apply to
1(s)when
2s + 1 1
(a) f(s) = S(S2 + 1)' (b) f(s) = S2(S+ 1)'
and find F(t) as a sum of residues. Ans. (a) F(t) = 1 - cos t + 2 sin t.
2. Use identities (6),Seco55, to prove that the function tanh z is bounded over each of
the half planes x ~ y and x ~ -y when y> O, and on the lines y = INn(N = 1,
2, . . .) uniformly with respect to N. Thus show that the function
f(s) = tanh
S2
s
satisfies all conditions in Theorems 5 and 10 such that F(t) is represented by a series of
residues as follows:
F(t) = 1 - .! f
n2 n =1 (2n
1
.. ,
co (2n
s
- I)nt (t ~ O).
Note that, according to Seco23, F is the triangular-wave function H(2,t) shown graphi-
cally in Fig. lO.
3. In Probo 4, Seco24, we found that the function
1 1
f(s) = S2 - s sinh s
is the transform of the periodic function F for which
F(t) =t when -1 < t < 1,F(t + 2) = F(t).
Show that s = Ois a removablesingular point of f, that f satisfiesall conditionsin
Theorems6 and 10,and thus that F has the sine seriesrepresentation
2 00 (-1)"+1
F(t) = 1t
- n= 1 n sin nnt.
4. Use' Maclaurin's series for coshz to show that the function g(s)= cosh ;,
s #- O,g(O)= 1, is an entire function of s for any branch of ;. Then show that the
function
1
f(s) =-
s cosh .jS
is analytic except for simple poles s = O,s = -(n - t)2n2and, with the aid of one of
identities (6), Seco5}. that f satisfies all conditions in Theorems 5 and 10 when CN are
ares of parabolas v
r sin 8/2 = Nn. Thus find L -1{J} in the form
= 1 + -4 '--
~ (-I)n (2n - 1)2n2
F(t) -
nn=12n-1
exp
[ 4 ]
t (t > O).
218 SECo 74] OPERATIONAL MATHEMATICS
to find F(t) directly. Here we can write y = Oand use the integration fonnula (8), Seco34.
Note that the residue procedure can represent F(t) only where F(t) = Obecause e'"f(z)
is entire.
9. From properties ofthe transfonn ofthe function erfc (tt-i) noted in Seco74, prove
that the function and each ofits derivatives tends to zero as t -+ +0.
10. When,JS = ~i912( -n < (J< n, r > O),use the procedure in Seco74 to prove
that
L;l -
1
{ ,JS }
=-
2
n0 i '"
o
exp(-12) d1 =-
1
jitt
(t > O)
11 . Generalize Theorem 10 for rectangular paths by letting the sides of CNlie on lines
x = -CPNand y = IPN and usingonlya oonditionof boundednessIf(z)1< B on the
sides x = -CPN when t > O,where e and B are positive oonstants independent of N.
12. Prove Theorem 10 when CN are semicircles Iz - yl = PN'x ~ y, and If(z)1<
Mlzl-k on CN. -
7
Problems in Heat Conduction
We shall now illustrate the use of the theory just developed in solving further
boundary value problems in the conduction of heat in solids. We present
examples of problems that cannot be fully treated with the more elementary
theory used in Chap. 4.1
The formal solution of the problem in the next section is followed by
a complete mathematical treatment of that problem. The purpose is to
illustrate a means of rigorously establishing the solutions of such problems.
Since the procedure is lengthy, the reader is advised to use it sparingly in
his workon the sets of problems that follow. A clear understanding of the
formal method of solution is of primary importance.
1AItemative methods for solving a few of the problems in this and the following chapter are
presented in chap. 7 of the author's "Fourier Series and Boundary Value Problems," 2d ed.,
1963.
219
220 SECo 75] OPERATlONAL MATHEMATlCS
where Fa is a constant.
The problem in the transform of U(x,t) is
su(x,s) = uxx(x,s) (O< x < 1),
where the symbol fi denotes some branch ofthe double-valued function sto
As long as fi
represents the same branch, regardless of which one, in both
,I
L
i
i
{:::::::::~:~::::}o
O x-l . Fig.75
PROBLEMS IN HEAT CONDUCTION [SECo 75 221
p(x,s) = Fosinh
s x-fi, q(s)= sinh-fi
Hence s = - n2n2 are simple poles, where the residues of e"u(x,s) are
22 .'
P (x,-n n ) _"2,,2, F.01 SIn nnx " ' 2 2
(5) ' e
2 2 = 2 2 ( - 1) 2nm exp ( - n n t) o
q ( -n n ) -n n
At least formally then (Sec. 71) the inverse transform of u(x,s) is the sum of
the residues (4) and (5) of e"u(x,s) at the poles (3); that is,
2 00 (-1)"
(6)
[
U(x,t) = Fo x + - L-
n "= 1 n ]
exp ( - n2n2t) sin nnx .
sinh x.jS
u(x,s) = F.o 1:
s sinhy s
is analytic with respect to s in any half plane Re s ~ y where y > O. To
examine its order in that half plane, let us write
but this does not assure us that U(x,t) -. Fo as x -. 1, which is the condition ~
the temperature function should satisfy.
By writing hyperbolic sines in terms of exponential functions and
carrying out one step of a division, we find that I
It is the first term on the right that has a weak order with respect to s when I
x = 1. Now we can write ~
where
(10)
g(x,s) = ~oexp [-(1 + x).jS]exp[1 -2(1
- exp-(-2.jS)
x).jS] - 1.
With the aid of the inversion integral evaluated in Seco 74, we find
! that
1Cn e'%u(x,z)dz,
taken along a curve Cn of a family of curves (n = 1,2,...) between the poles,
tends to zero as n tends to infinity (Sec. 71).
To select curves Cn so that lu(x,z)1has a suitable order property when-
ever z is on Cn, we first note that
then
(2)
Isinhj;12 = sinh2(fi cos~) + 1 ~ 1.
-a~
Fg.76
226 SECo 78] OPERATIONAL MATHEMATICS
For if is a smalI positive angle and z is on either part of any of the parabolic
paths Cn on which 181 < 1t - , it can be seen from the Eq. (4) that for each
fixed x (x < 1)a number M. independent ofn exists such that lz1'u(x,z)1< M..
for any constant k. This order property, together with the inequality (4)
satisfied on the remaining part of Cn, is sufficient to show that the representa-
tion (5) is valid when t = Oand x < 1. The proof, made by first selecting
smalI and then n large, is left to the problems. It establishes the Fourier
the representation (3), Seco 49, of U(x,t) as a series of error functions, we can
see that
l
(2) whenever t > Oand O < x < l.
F:o U(x,t) > O
(4)
and
(5) lim
,-o [
cI>(I,t) + F~ J
v 1I:t
= o.
&(e'tt)than to pro ve they are bounded for all t. Note that V xx satisfies the
eonditions imposed on VI because V, = V xx'
We have seen that the funetion
W(X,S)= i O e-S'W(x,t) dt + i
.
'0 e-S'W(x,t) dt
(to > O),
(1) ( )
- f( ) sinhx.fi
uX,s - s r:.
sinh v s
(3)
U(x,t) = {F(t - T)V,(x;r)dT.
The series obtained by differentiating this series term by term with respect
to t does not converge when t = O; but it was shown that the function
V,(x,t)is continuous when t ~ Oand O~ x < 1and that
V,(x,O) = O (O~ x < 1).
The differentiated series simply fails to represent V,(x,t)at t = O.
To arrive at another form of the temperature function U, we assume
F continuous, F' sectionallycontinuous, and F of exponential order. Then
L{ F'(t)} = sf(s) - F( + O),
and u(x,s) = F( +O)v(x,s) + L{F'(t)}v(x,s).
(4)
U(x,t) = F( +O)V(x,t) + { F'(t - T)V(X,T) dT.
The two formulas (3) and (4) give the temperature U(x,t) in terms of
the temperature V(x,t) corresponding to a fixed surface temperature. They
are two forms of Duhamel's formula (Sec. 85). The above series for V(x,t)
can be substituted into formula (4), and it can be shown that the temperature
232 SECo 81] OPERATIONAL MATHEMATICS
X X(X2 1) 00
u(x,s)=
. A ["2
s + 3'-.S L an(x)~J .
+ n=O
x(x2 - 1)
[
A xt + 3! J.
The residue of IU(X,Z)at the simplepole z = -n2n2 is
' sinh x'-; 2A( _1)n-l . 2 2
z z cosh
2A.-;.-; zJ z=-n2,,2= n3n3 Sin (mrx) exp (-n n t).
ConsequentIythe formula for the temperatures can be written
X3 - x 2 00 (_1)n-l
(2) U(x,t) = A -
[ 6
+ xt + "3
n
L
n=l n
3 exp ( - n2n2t) sinmrx .
J
PROBLEMS IN HEAT CONDUCTION [SECo 81 233
12 ex>(_I)n-l
x - X3 = "3
1t n=1 n
3 sin n1tx (O ;;; x ;;; 1).
PROBLEMS
Fig.77
W(O,t) = B, and W(X,O)= A. Ir V(X,t) is the function (5), Seco80, so that V, = V xx'
V(x,O) = V(O,t) = Oand V(I,t) = F(t), show by superposition that
5. When t ~ to > O and O ~ x ~ 1, prove that the series (6), Seco75, and the series
obtained by differentiating that series once or twice with respect to x or once with
respect to t are unifonnly convergent with respect to x and t. The series then represent
continuous functions of x and t, and tennwise differentiation is valid, since the terms
themselves are continuous functions. As a consequence, show that the function V(x,t)
defined by the series satisfies the heat equation V, = Vxx when O < x < l and t > O,
and end conditions V( +O,t) = O, V(I - O,t) = Fo, when t > O.
6. According to Abel's test for uniform convergence, a series 1::",. An T,,(t) converges
unifonnly with respect to t if the series 1::'=1 An converges, and if functions T,,(t) are
bounded unifonnly with respect to t and n and such that T,,+(t) ~ T,,(t). Given that
the Fourier series representation (6), Seco77, is valid, use Abel's test when O ~ t ~ to
to show that the function defined by Eq. (6), Seco75, is continuous with respect to t
when t ~ Oand that it satisfiesthe initial condition V(x,+O) = Ofor eachfixedx such
that O < x < l. This, together with the result in Probo 5, again establishes the solution
(6), Seco 75.
7. Give details of the proof outlined in Seco 77 of the Fourier sine series representation
(6) ofthe function x(O ~ x < 1). Suggestion: Show first that when z is on the parabolic
arc Cn and 181< 1t - (, then
ZU(X,Z)
~ cosh(x-fi cos8/2) < 2 exp[-(1 - x>-fi sin (/2].
I Fo I
- cosh(-fi cos8/2) l + exp(- 2-fi sin(/2)
8. As in Seco80 let V(x,t) denote temperatures in the slab with faces x = O and x = l
when V(x,O) = V(O,t) = O, V(l,t) = 1, and k = 1, and write V(x,t) = O when t < O.
Let V(x,t) be the temperatures under the same conditions except that the face x = l is
kept at temperature A from t = O ~o t = to and thereafter at temperature zero; thus
V(I,t) = A - ASo(t - to) (t > O).
(b) Note that the total quantity of heat conductedacrossa unit areaof a section
x = xo, Q = - K So Vx(xo,t)dt, is the value of L{ Vx(xo,t)} when s = O if that trans-
fonn exists when s = O. Thus for the above slab with temperatures V show fonnalIy
that
Q = -AKto.
9. Obtainthesolutionof the temperatureproblem
V,(x,t) = V xx(x,t) (O< x < 1,t > O),
V(x,O) = 1, V(O,t) = V(I,t) = O,
I For a proofsee the author's "Fourier Series and Boundary Value Problems," 2d ed., chap. 10,
1963, which also treats uniqueness of solutions.
~
~
PROBLEMS IN HEAT CONDUCT/ON [SECo 81 235
in the fonn
~
U(x,t ) = -4 L..- sin(2n - 1)1tx
1
2 2
exp [ - (2n - 1) 1t t] .
1t n= 1 2n -
- ~ ~ (_l)n-I
I -
[- (2n - 1)21t2t
U( ) (2n - 1)1tx
x,t - 1t /;;'1 2n - I cos 21 exp 412 J
for the temperatures in a wall with its face x = O insulated and its face x = 1 kept at
temperature U = 1, if the initial temperature is zero (Fig. 78) and if k = l.
UI%,OI-O U-l
o %
Fig.78
11. Establish the fonnula in Probo 10 as a solution of the boundary value problem.
12. Find U in Probo 10 as a series of error functions.
Ans. U = I - U o(x,t), where U o is the function (4), Seco48, when Uo = k = l.
13. Obtain the solution of the problem in Seco48 in the fonn
( )-
uX,s - I h sinhx-fi + -fi coshx-fi .
s h sinh -fi + JS cosh -fi
, ,
U(.x.O)-O
U(1,t)-l
0'-
-10 %
-
----- Fig.79
Obtain formally the solution
U(x,t) -- hx + I
h+ I
- 4 L-
~ 2sinan(1-. 2x) exp ( - an2t),
n=1 an - s10 an
PROBLEMS IN HEAT CONDUCTION [SECo 82 237
82 ARBITRARY INITIALTEMPERATURES
Let the initial temperature distribution in a bar or slab be a prescribed func-
tion g(x) of the distance from one face. When the lateral surface of the bar is
insulated and the ends are kept at temperature zero (Fig. 80), units can be
selected so that the boundary value problem for the temperature in the bar
. becomes
U,(x,t) = Uxx(x,t) (O< x < 1, t > O),
U(x,+O) = g(x) (O< x < 1),
U(+O,t) = U(l - O,t) = O (t > O).
Fg.80 o
l.
(6)
U(x,t) = 2 n~1 exp (- n2n2t) sin nnx f. o g(~) sin nn~ d~.
When t = O, the series here is the Fourier sine series for the function g(x),
on the intervalO < x < 1. In fact, the boundary value problem in U(x,t)
here is especially well adapted to the classical method of solution by using
separation of variables and Fourier series, a method we shall discuss in
Chap. 9.
VERIFICATION
Solution (6) can be established by the procedure used in Secs. 76 and 77.
Since some variations are needed, we outline the method here. Over any
PROBLEMS IN HEAT CONDUCTION [SECo 82 239
From Eqs.J.3) and (4) it can be shown that u(x,s) is O(r-t) when s is on
the parabolas V
r sin (J/2 = (n - t)n. Therefore the series (6) represents the
solution (9) when t > O.
In addition to this, we can see from Eq. (8) that the inversion integral
of the function u(x,s) - g(x)/s converges to zero when t = O and that it is
represented by its series of residues when t = O, provided O< x < 1. It
follows that
00 l
(lO)
f
2 n~l sin nnx o g(~)sin nn~d~ - g(x) =O
(O< x < 1),
83 TEMPERATURES IN A CYLlNDER
Let us derive formally the temperature function U(r,t) for a solid circular
cylinder of infinite length whose initial temperature is zero and whose
surface is kept at unit temperature (Fig. 81). Units of time and length can
be chosen so that the boundary value problem becomes
I
U,(r,t) = Urr(r,t)+ -Ur(r,t) (O~ r < 1, t > O),
r
I
I
I
I
U(r,OI-OI U-l
I
I
I
~~--T----
,," r I 1 .......
Fig.81
....
PROBLEMS IN HEAT CONDUCTION [SECo 83 241
V(Po,T) = A.
Our temperature formula then takes the form
(10) V(P,T) = A
[ - ,,=f
I 2 JO(IX"p/PO)exp
1 IX"J 1(IX,,) (
- IX/~T
Po }J
.
PROBLEMS
1 . The initial temperature of a slab is U(x,O)= Ax. Ir the faces x = O and x = I
are kept at temperature zero, derive the temperature formula
2AI 00 (-1)"-1 n21t2kt . n1tx
U(x,t)= - L
1t ,,=1 n
exp
(
- y-
I )
Sln-.
I
2. Derive the following formula for the temperature function in Probo 1:
4. The facesx = Oand x = Iof a slab are insulated (Fig. 82)and the initial temperature
distribution is prescribed: U(x,O)= g(x). Find the transform of the temperature
function U(x,t) in the form
'
l
i
u(x,s) = - k og@R(x,~,s) d~,
where
2
i'
an= 1 o g(x)cosdx
mr:x
(n = O, 1, 2, . . .).
U(x.O)-g(x)
o %
1-
Fig.82
5. In Probo4, if g(x) = A when O < x < 1/2and g(x) = Owhen 1/2 < x < 1,show that
-
U(x,t ) -- A 2A ~ (_l)n- (2n - 1)1tx (2n - 1)21t2kt
2 +-'- 2
1tn=n- 1 cos I exp [ - 12 J
.
6. The facex = Oof a slab is kept at temperature zero while the face x = 1 is insulated.
If k = 1 and the initial temperature is U(x,O)= x, derive the temperature formula
-~ -
U(x,t)
00 (-1)"+
- 1t2n~ (2n -
. (2n - l)1tx
1)2SIn 2 exp [- (2n 1)21t2t
4 J.
7. The initial temperature of a cylinder of infinite length is zero. If the surface r = 1
is kept at temperature A from t = Oto t = to and at temperature zero thereafter, and if
k = 1, derive the following formula for the temperatures in the cylinder:
W(r,t) = A[U(r,t) - U(r, t - to),
where U(r,t) is the function defined by formula (9), Seco83, when t 6; Oand U(r,t) = O
when t < O.
244 SECo 83] OPERATIONAL MATHEMATICS
8. The flux of heat into a cylinder of infinite length through its surface r = l is a
constant, so that U,(I,t) = Ao If k = l and the initial temperature is zero, derive the
temperature formula
where p, P2' oare the positive roots ofthe equation J (fJ) = O,given that the equation
o o
has only real roots and that J'(Pn)"# 00 Note that according to Bessel's equation
-J;;(x) = Jo(x) - J(x)/x, and since -J;;(x) = J~(x) it follows that J~(Pn)= JO(Pn)o
9. Use symbolic transforms, as in Seco13,to indicate formally that the solution of the
problem
R(O,~,s)= R(I,~,s) = O,
u(x,s)= - f R(x,~,s)gR)d~
84 EVAPORATION
FROM A THICK SLAB
We now illustrate a type of problem that is especiallywell adapted to treat-
ment by the Laplace transformation in contrast with other methods.
Let C(x;r)denote the concentration of moisture at time 't" in a porous
semi-infinite solid (x ~ O)with a uniform initial concentration Co (Fig. 83).
Evaporation of moisture takes place at the face x = O into a dry adjacent
medium according to the linear law tbat the outward flux is E[C(O,'t")- O],
where the positive constant E is a coefficient of evaporation. A boundedness
condition on C will take the place of a condition of diffusionat a right-hand
boundary. The condition that C(x,'t")is 19(eat)uniformly with respect to x,
whenever ex> O,will be sufficient. Then if K is the coefficient of diffusion
of moisture in the solid, C(x,'t")satisfies the conditions
for some constant N and each positive constant p. The problem in the
transform of V becomes,formally,
---
O
--10 C(X, O) = Co
--
x
Fig.83
246 SECo 84] OPERATIONAL MATHEMATICS
Ir.j';=jr ei912where -11: < (J < 11:and s = rei9,the solution of problem (4)
can be written as
I h
(5)
u(x,s) = -s - s(h + .j';s) exp(- x.j';).
SinceRe.j'; < Owhen e =F11:,
then h + .j'; =FOand u is an analytic
function of s everywhere except on the nonpositive real axis. The ray (J = 11:
is the branch cut of.j'; and of u. The inversion integral on the line Re s = y
applies to both terms on the right in Eq. (5) since the second term is (Il(S-3/2),
uniformly with respect to x. According to our theorems in Chap. 6, it
represents a function
Since Iy-l sin yl < I when y> O,we can see that the integrand in
formula (7)satisfiesthe condition .
Thus Ux(O,t)= hU(O,t) when t > O,so the function U with representations
(6) and (7) satisfies all conditions in boundary value problem (3).
That function has the transform (5); thus
l h l h
u(Os) =- - =- -
, s s(h + JS) s - h2 JS(s - h2)
and it follows from Probo 4, Seco17,or transform 40, Appendix A, Table A.2,
that
2 fco
(8) U(O,t)= exp (h2t) erfc (h.jt) = ~Jo exp(-y2 - 2h.jty)dy ~ l.
Using formulas (7) and (8) and the substitutions (2), we can write the
concentration C in the slab and at the face as
(9) C(X,-r=
)
2hCo
-
1t f
o
co h sin xy + y cos xy
2 2
y(y + h )
2
exp ( - K -ry ) dy,
85 DUHAMEL'S FORMULA
In Seco80 we obtained a formula for the temperatures in a bar with variable
end temperature, in terms ofthe temperature function when the end tempera-
ture is constant. We now obtain a more general formula that simplifies
heat conduction problems in the same way.
Let U(x,y,z,t) be the temperatures in any solid, filling a regio n R, that
is initialIy at temperature zero throughout. Let the temperature at every
point of some part S of the boundary be a prescribed function F(t) of time,
and let the remainder S' of the boundary be kept at temperature zero
(Fig. 84).
Ir A(U) represents the linear differential form
l a au a au a au
(1) A(U)=- - K- +- K- +- K- ,
e[ ax( ax ) ay( ay ) az( az )]
248 SECo 85] OPERATIONAL MATHEMATICS
S'
Fig.84
-1 onS.
s
Since s is a parameter in the linear homogeneous differential equation
(5), the product of sf(s) by the solution v is also a solution. But according to
conditions (8),
sf(s)v = O onS'
= f(s) onS;
PROBLEMS IN HEAT CONDUCTION [SECo 85 249
thus the function sf(s)v also satisfies all the conditions (5) and (6). Ir the
problem consisting of conditions (5) and (6) is to have but one solution, then
sf(s)v must be the same as the solution u:
(9) u(x,y,z,s) = sf(s)v(x,y,z,s).
Since sv is the transform of v" it follows from Eq. (9) with the aid of
the convolution that
PROBlEMS
1. From formula(5),Seco84, for u(x,s)and transform 86,AppendixA, Table A.2,
obtainthisformulafor the concentrationofmoisturein the slabx ;;; withevaporation
at the surface:
C(X,f)
= c{ erf(~) + exp (hx + h2 Kf) erfc (~ + h~) J.
2. Identify the two expressions in formula (8), Seco85, for the concentration U(O,t)
at the face of the semi-infinite solid and thus show that < U(O,t)~ 1 when t ;;;O.
3. A layer of finite thickness extending from the face of the slab in Seco84 is initially
dry. The initial concentration of the rest of the semi-infinite slab is uniformo Choose
units so that the relative concentration U(x,t) satisfies the conditions
U, = U""',X> O,t > 0, U(x,O)= So(x - 1), u ,.(O,t) = hU(O,t),
U(%.O)-O
A'-
-
Fig.85
8. The diffusivity k of the earth 's soil in a certain locality is 0.005 cgs unit. The tempera-
ture of the surface of the soil has an annual variation from -
8 to 22C. Assumingthe
variation is approximately sinusoidal (Prob. 6), show that the freezing temperature
will penetrate to a depth of approximately 170cm (considerably less, because of the
latent heat of freezing).
9. In Probo 8 find the approximate depth at which the variation of temperature with
time is 6 months out of phase with the variation of the surface temperature. Show that
the amplitude of the variation at that depth is less than 1C.
Ans. x = 705 cm = 23 n, approximately.
10. Let the functions V(x,t) and W(y,t) satisfy the heat equations V. = kJ.~x and
~ = kjy, respectively. Prove by direct substitution that the product of those tem-
perature functions,
U(x,y,t) = V(x,t)W(y,t),
u, = k(Uxx + Uyy).
Ir in addition V(O,t) = V(a,t) = O and W(O,t) = W(b,t) = O, and if V(x,O) = f(x) and
W(y,O) = g(y), then show that U(x,y,t) represents the temperatures in a rectangular
plate (Fig. 86) with insulated faces, if the edges are at temperature zero and the initial
temperature is
U(x,y,O) = f(x)g(y).
U-o
(a,b)
Fig.86 o U-o x.
11. Use the product of solutions (Prob. 10) to obtain the following formula for the
temperaturesin an infiniteprism with a squarecrosssection,if the initial temperature
is A and the surfacetemperatureis zero, if the unit of length is the side of the square,
and k = 1:
V(x,y,t) = A U(x,t)U(y,t),
where U is the temperature function found in Probo 9 of Seco81.
12. With the aid of Probo 10, derive the formula
4 x 00 sin(2n - I)lty
U(x,y,t) = -7t erf ( 2y r.t.) n=L 1 2n - l exp[ - (2n - l )2lt2t]
for the temperatures in the semi-infiniteslab x ;;;; O,O;;;; Y ;;;;1 with its boundary at
temperature zero, if U(x,y,O)= 1 (Fig. 87) and k = 1.
252 SECo85] OPERATIONALMATHEMATICS
YI U-O
U-O
"'''1
O
U(x,y,Oj
U-O
-1
\ " Fig.87
13. Generalize the method of Probo 10 to the case of three dimensions, and give an
illustration of its use in finding temperatures in a cube.
14. Use the generaJized convolution property found in Probo 20, Seco27, to make a
formal derivation ofthe following more general Duhamel formula. Let A(U) denote the
differential form (1), Seco85, P the point (x,y,z), n the distance normal to the boundary
surface S or S', and let a(P), b(P),H, F, and G represent prescribed functions. The tem-
perature function U(P,t) in the region R is to satisfy the conditions
U, = A(U) + H(P,t), U(P,O)= G(P) (P in R);
dU
(P on S or S').
a(P)U+ b(P)dn = F(P,t)
When the functions H(P,t) and F(P,t) are replaced by H(P,t') and F(P,t') respectively,
where t' is any fixed vaJue of t, V(P,t,t') denotes the solution of the problem. Write the
problem in the transform v(P,s,t')and transform its members with respect to t', using the
same parameter s, to obtain a problem in the iterated transform D(P,s,s)and thus derive
this generaJization of formula (10):
a f'
U(P,t) = ai Jo V(P, t - T, T) dT.
Note that if a(P) = F(P,t) = Owhen P is on S', then the surface S' is insulated. Thus the
generalized formula could be applied to Probo 14, Seco81.
t
A
8
Problems in Mechanical Vibrations
Fg.88
constant force Fo per unit area acts parallel to the bar on the other end
(Fig. 88). Let all parts of the bar be initially at rest and unstrained. The
displacements Y(x,t) then satisfy the conditions in the boundary value
problem
Y,,(x,t) = a2 Yxx(x,t) (O< x < c, t > O),
(1) Y(x,O) = Y,(x,O) = O,
a function that is analytic at the point s = O. In viewof Eq. (2) then the
origin is a simple pole of (f'y(x,s) when x =1= O, and the residue there is
(aFo/E)(x/a), or Fox/E.
The residue of y itself at the simple pole Sncan be written
(6)
- 8Foc (-l)n . (2n - l)nx (2n - l)nat
Pn + Pn = n2E (2n - 1)2 SIn 2c COS 2c .
(1) x = -- 8e ~
i.J ( - 1)n
sm (2n -
o l)nx
(O ~ x ~ e).
n2 n=1 (2n - 1)2 2e
Each term of series (1) is an odd function of x, as is x itself, so the series
represents the function x over the interval - e ~ x ~ eo When x is replaced
by x + 2e, each term merely changes sign so the sum of the series is an
antiperiodic function for all x, with period 2e; consequendy it is periodic
with period 4e. Thus for all real x the series represents the triangular-wave
function H shown in Fig. 89, described by the two conditions
(2) H(x) = x (-e ~ x ~ e), H(x + 2e) = -H(x) ( - co < x < co)o
'The function H also has these properties for all x:
(3) H(x) = -H( -x) = H(x + 4e) = H(2e - x).
(4) 8e
2i.J
~ (-I)n-1
2sm
. o (2n - l)nx
= H(x ) (- co < x < co).
.n n=1(2n - 1) 2e
The series is, in fact, the Fourier series representing the odd periodic function
H for all Xo
-
We write m = (2n l)nf(2e) and note that
2 sin mx cos mat = sinm(x + at) + sinm(x - at).
H[x)
Fig.89
PROBLEMS IN MECHANICAL VIBRATIONS [SECo 87 257
Let the end x = O of the bar again be fixed while a simple periodic force
F(t) = A sin rot (ro > O)
per unit area acts lengthwise at the end x = e. Ifthe bar is initially unstrained
and at rest, only the end conditionat x = e in problem (1), Seco86, need be
changed to read
(1) EYx(e,t) = A sin rot.
If iro is not equal to any one of the numbers Sn' that is, if
ro#--
(n - !)na
(n = 1,2,...),
e
the singular points :t iro and :t Snare all simple poles unlessthe value of x
is such that some are removable, in which case they serve as simple poles
with residue zero.
r:
According to formula (13), Seco72, the component of Y corresponding
to the pair of poles :t iro of y is a-term of the type b cos(rot + O); here the
real numbers b and Odepend on x. Similar components correspond to a
pair of simple poles :t Sn' Thus Yis represented formally by a series of
this type:
00
in the denominator of expression (2) for y(x,s) is such that q(iw,) = q'(icD,)= O
while q"(iw,) =FO. Consequently the points s = :t iw, are poles'of y f the
second order, and corresponding to that pair of poles, according to formula
(16), Seco72, Y(x,t) contains an unstable component of the type
(4) tb,(x)cos[w,t + I/I,(x)], where b,.(x) =Fo.
The remaining component of Y is periodic with frequency Wl>the common
frequency of all its periodic terms.
That unstable oscillation of sections of the bar is called resonanee.
The periodic external force is in resonance with the bar when its frequency
w coincides with any one of the resonance frequencies
The frequenciesw, depend on the physical properties of the bar and the
mannerin whichthe bar is supported. If the end x = O of the bar is free
or elastically supported, for instance, we find that the set of resonance
frequenciesis differentfrom the set (5).
To produce resonance, the external force need not be restricted to the
simple form A sinw,t. For any prescribed force F(t) at the end x = e, the
transform of Y(x,t) is
a sinh (sx/a)
(6)
y(x,s) = "Ef(s) s cosh (se/a)'
89 VERIFICATIONOF SOLUTIONS
When the formal solution of a boundary value problem can be written in
terms of a finite number of simple functions, that form is usually best for
verifying the solution, as we illustrated in Seco87. But to establish a solution
represented only by an infinite series or by an improper integral, the
procedure based on properties of its transform, used in Chap. 7, may be
useful. Let us illustrate a combination of the two methods for the problem
in the preceding section.
The following transform of displacements Y(x,t) in a bar with its end
x = Ofixed and with pressure A sin wt applied at the end x = e was found
260 SECo 89] OPERATIONAL MATHEMATICS
formally:
B sinh (sx/a)
(1)
y(X,S)= S2+ ro2s cosh (se/a)
The quotient of hyperbolic functions here was shown (Sec. 86) to be bounded
uniformly in x over half planes Re s ~ ')1and Re s ;;;- ')1 when ')1 > O;
cosh (sx/a)/cosh (se/a) is also bounded there. Hence y is l!J(S-3)and Yx is
l!J(s- 2) over those half planes, uniformly in x, and L- 1 {y} along a line
Re s = ')1represents a continuous function Y(x,t) such that Y(x,O) = Y(O,t) = O;
also Y has continuous derivatives of first order,
Y,(x,t)= L- 1{sy(x,s)}, y':(x,t)= L- 1 {yx(x,s)},
such that Y,(x,O)= Oand Ey':(e,t) = A sin rot. Hence Y satisfies all boundary
conditions in the problem, but the order of y is not adequate to show directly
that Y/t = a2yxx.
To prove that Y does satisfy the wave equation, we write
1 1 ro2
S2 + ro2 = S2 - S2(S2 + ro2)'
(2) u(x s) = -
B sinh (sx/a)
v(x s)
- Bro2sinh (sx/a)
= S3(S2 + ro2) cosh (se/a).
, S S2 cosh (se/a)' ,
Then y(x,s) = u(x,s) + v(x,s)
where v is l!J(s-S)and Vxxis (9(S-3). It follows readily that L-l{V} satisfies
the wave equation.
The function L- 1{u} can be written as a linear combination of
known solutions of the wave equation since (Sec. 87)
2au(x,s)= Bs-1 L{2x - H(x + at) - H(x - at)}
where H is the triangle-wavefunction shown in Fig. 89. Thus L-l{U} is
the function
B '
U(x,t) = 2a fo [2x - H(x + a-r)- H(x - a-r)]d-r
x+al
= _B 2xt - -l
2a [ af
H@ d~ .
x-al ]
In terms of the periodic function
an-I
d o
2n-1
( mn=~n ) o
PROBLEMS
1. The end x = O of an elastic bar is free while a constant longitudinal force Fo per .
unit area acts longitudinally at the end x = c (Fig. 48). The bar is initially at rest and
unstrained. One formula for the longitudinal displacements Y(x,t) in the bar was
found in Probo 8, Seco44. Derive this altemate formula:
Y = -Fo 2
x + 3a t - c -
22 2 12c2
r '-- ---y-cos-cos-
~ (_I)n mrx mrat
.
6Ec[3 n n= n c c ]
2. Reducethe solutionof Probo1 to the form
F.
Y(x,t) = 12~c[6x2+ 6a2t2- 2C2 + Q(x + at) + Q(x - at)]
where Q is this periodic function:
(-c<x<c), Q(x + 2c) = Q(x) (-00 < x < 00).
Use that form to verify the solution of the boundary value problem.
3. An unstrained elastic bar or heavy coiled spring is clamped along its length c
so as to prevent longitudinal displacements, then hung from its end x = O (Fig. 90).
At the instant t = O,the clamp is releasedand the bar or coil vibrateslongitudinally
because of its own weigh. Thus
Y,,(x,t) = a2Yxx(x,t)+ g (O< x < c, t > O),
where g is the acceleration of gravity. Complete the boundary value problem, write
m = n - 1. and derive the formula
g 4C2 <X>1 . mnx mnat
Y(x,t) = -
2a2 [
x(2c - x) - - "-sm-cos-
n3 nf'm3 c c ]
.
Fig.90
5. The ends x = :t e of an unstrained shaft spinning about its axis with angular
velocity w are brought to rest at the instant t = O (Fig. 52, Probo 13, Seco44). Derive
this formula for the angular displacements of its cross sections:
O( ) -- 8cw ~ (_1)n-1 (2n - 1)1tx.
SIn (2n - 1)1tat
.
'--
1t a "=1 (2n - l )
O' x,t 2 2 cos
2e 2e
6. Reducetheformulafore in Probo5 to the form
w
e (x,t) = 2a [H(x + at) - H(x - at)],
where H is the triangular-wave function shown in Fig. 89 and represented by series (4),
Seco87. Use this simple representation of e to verify the solution of the boundary
value problem in Probo 5.
7. The end x = O of an elastic bar is fixed while a force proportional to t2 acts
10ngitudinalIy at the end x = l so that Yx(l,t) = At2, where Y(x,t) denotes longitudinal
=
displacements. The unit of time is such that a l. If the bar is initially at rest and
unstrained, derive the formula
X3 l l
where
Y(x,t) = A [ X(t2 - 1) + 3" + '2R(x + t) + '2R(x- ]
t)
X3
R(x) = x - 3"
(-1 ~ x ~ 1), R(x + 2) = -R(x) (-00 < x < (0).
w,--- r1ta (r = 1, 2, . .. ).
e
Note that the lowest frequency W1 is twice as great as the lowest in the set (5), Seco88.
11 .
If the end x = O of the bar in Probo 10 is free, rather than fixed, show that the
resonance frequencies are the same as the set (5), Seco88.
264 SECo 90] OPERATIONAl MATHEMATICS
12. A simple periodic longitudinal force acts uniformly on all elements of an elastic
bar with its end x = O fixed and its end x = e free, so that the displacements Y(x,t)
satisfy the conditions
1';,(x,t) = a2Y"",(x,t) + B sin (Ot,
(0= (r - t)na
, e (r = 1,2,...).
13. A simple periodic transverse force acts uniformly on all elements of a stretched
string of length e with fixed ends, so that the transverse displacements Y(x,t) satisfy
the equation 1';, = a2y= + B sin (Ot. Let the string be initially at rest in its position
of equilibrium. Show that resonance occurs only when (O= (o,where
14. The supports of the ends of a stretched string vibrate in a transverse direction
so that
Y(O,t)= A sin at, Y(c,t) = B sin pt.
The string is initially at rest on the x axis. In general, show that resonance takes place
if either a or p has one of the values n1ta/c. But if p = IX,show that the resonance
frequencies are 2n1ta/c when B = -A, and (2n - l)na/c when B = A, where
n = 1,2, Hint: When p = a and B = -A, showthat
y( ) -
x,s -
~
S2 + a2
sinh [s(c - 2x)/(2a)]
sinh [sc/(2a)] .
A string stretched between the origin and a point (e,O)is given a prescribed
initial displacement Y(x,O) = g(x) and released from rest in that position.
To find its transverse displacements Y(x,t), we solve the problem
p(x,S)
y(x,s) = a sinh (sela
. xs c
" . (e - ~)sa"
+ SInh -
a f
x
( ) SInh
g."
a
.".
Ir we write
2 c . ml:~
(1) bn = - f. g(~)SIn-a~,
e o e
then p(x,sn)=tbne cos mI:sin (ml:xfe),and the residue ofy at Snis tbn sin (ml:xfe).
Thus the sum ofthe residues of esty(x,s) at the pair of poles :tsn is (Sec. 72)
.
bnSIn-COS-.
nnx nnat
e e
(2) "
en
Y(x ,t) -- f..J bn sln -cos-
. nnx nnat
n= 1 e e '
where the coefficients bn are given by formula (1).
Since Y(x,O)= g(x), formula (2)implies that
en nnx
L bn sin- e = g(x)
n= 1
(O ~ x ~ e).
Each term of this series is an odd periodic function of x with period 2e.
Consequently, for all x the series should represent the odd periodic extension
G of g defined by the conditions
G(x) = g(x) when ~ x ~ e,
(3)
G(-x) = -G(x), G(x + 2e) = G(x) for all x.
~ ,.
0-
L!BSintl1tFig.91
PROBlEMS IN MECHANICAl VIBRATIONS [SECo 91 267
Fig.92
268 SECo 92] OPERATIONAL MATHEMATICS
two points s = :t. iafJ.n/care poles of y of the second order and Y contains
a term of resonance type bn(x)t cos [wnt + 8n(x)] with the resonance frequency
11
lB sin c"t
I
1-
(e,O)
Fig.93
PROBLEMS IN MECHANICAL VIBRATlONS [SECo 92 269
(5) y (x,s )
= 2
Bw 2 sin qx sinh..qe + sinh qx sin qe .
s +w 2 sm qe smh qe
The Maclaurin series representations of the sine and hyperbolic sine
functions show that the final fraction in Eq. (5) is an analytic function of q2,
and therefore of s, except at points where the denominator vanishes. The
point s = Ois a removable singular point. The denominator vanishes when
qe = :i:n1t and qe = :i:in1t; that is, whenever q2e2 = :i:n21t2. In view of
Eqs. (3)and (5)then, the singular points of y(x,s) are
in21t2a
(6) s = :i:iw, s = :i:---y-
e
(n = 1,2,...).
(7) (n = 1,2,...),
the derivative of the entire denominator in the expression (5) for y(x,s)
does not vanish at any of the points (6),nor does the numerator exceptfor
particular values of x. Those points are therefore simple poles of y(x,s).
In this case the formula for the displacements has the form
co
The reader can show that the expression for y(x,s) then has the same
denominator as it does in Eq. (5). The resonance frequencies are again those
given by formula (7). Other cases are included in the problems at the end
of the chapter.
y .F(t
o (e,O) x
Fig.94
PROBLEMS IN MECHANICAL VIBRATIONS [SECo 93 271
(8)
I
Y(x,t) = ot o F(t - t)Z(x;r)dt,
or
(9)
Y(x,t)= E F(t - t)Zt(X,t) dt.
These are two forms ofDuhame1's formula for the resolution ofthe problem
in Y(x,t) with a variable end condition into one with a fixed end condition.
The derivation of these formulas can be extended easily to other
problems. In the case of transverse vibrations of bars, for instance, the
derivative of the fourth order with respect to x replaces Yxx(x,t),and addi-
tional boundary conditions are involved. In the case of transverse displace-
ments in a membrane, the Laplacian of Y replaces Yxx in our problem.
In these cases, the steps in the derivation of Duhamel's formula are the
same as in the case treated above.
PROBLEMS
1. An elastic bar with its end x = O kept fixed is initially stretched so that its longi-
tudinal displacements are Y(x,O) = Ax (O~ x ~ e), then released from rest in that
position at the instant t = Owith its end x = e kept free. Derive these two formulas
for its displacements:
Fig.95
When a longitudinal periodic force F(t) = A sin rot acts at the end x = e, show that
resonance takes place if ro has any one of the values aa../e (n = 1,2,...), where a.. is
a positive root of the equation tan a. = he/a.. Show those roots graphically (cf. Fig. 92).
3. A string stretched from the origin to a point on the x axis is initially straight with
a uniform velocity in the direction of the Y axis, as if a moving frame supporting the
end points is brought to rest at the instant t = O. There is a smalldampingforceper
unit length proportional to the velocity. Select units so that the transverse displace-
ments satisfy the conditions
Y,,(x,t)= Yxx(x,t)- 2bY,(x,t) (O< x < n, t > O,O< b < 1),
Y(O,t)= Y(n,t) = Y(x,O)= O, Y,(x,O)= Vo.
1[p is a branch of the function (S2 + 2bs)t, show that
Y(x,t) = 4voe-h'
n .=1
fsin (2n - l)x sin mt
(2n-l)m
4. A string is stretched from the origin to the point (e,O)and given an initial veiocity
Y,(x,O)= g(x) but no initial displacement. Derive this formula for its displacements:
2
Y(x,t) = -
00 I . nnx . nnat
L -sm-sm-e
na.=ln e i
o
e . nn~
g(~)sm-d~.
e
5. In Probo 4 let G(x) be the odd periodic extension, with period 2e, of the velocity
function g(x) and show formaIly that
2 Y,(x,t) = G(x + at) + G(x - at)
x+ar
I
and Y(x,t) = -2
f
a x-at
G(~)d~.
1 Fig.96
1
the telegraph equation. Note that equations (a) are better adapted to initial and end
conditions that involve values of both I and V, however, than is the telegraph equation
(see Probs. 7 to 9).
7. Supposethat R = S = O in the transmission line shown in Fig. 96, and that the
current and voltage are zero initiaUy. If one end is kept grounded, V(O,t)= O, and
the current at the end x = e is proportional to t, find the transform v(x,s) of V(x,t)
using the transmission line equations (a), Probo 6. Thus show that V(x,t) is an eIectrical
analog of the displacements Y(x,t) in the bar in Seco86, where a2corresponds to (KL)- 1.
8. In the transmission line shown in Fig. 96, R = S = O and I(x,O)= V(x,O)= O.
If V(O,t)= A sin rot and the end x = e is grounded, find the values of ro for which
the voltage and current become unstable, using the transmission line equations (a),
Prob.6. Ans. ro = mrc-l(KL)-t (n = 1,2,...).
9. Solve Prob. 8 when the condition that the end x = e is grounded is replaced by the
condition that the circuit is open at that end.
Ans. ro = (n - t)1tc-l(KL)-t (n = 1,2,...).
10. One end of a beam is built into a rigid support (Fig. 97); thus if Y(x,t) denotes
transverse displacements,
Y(O,t)= YJO,t) = o.
y;- .
o x
Fig.97
y
YlO,tl-B sin (lit Fig.98
12. Both ends of a beam of length e are built into rigid supports. The beam is initially
at rest with no displacements. A simple periodic force per unit length acts along the
entire span, in a direction perpendicular to the beam, so that the transverse displace-
ments Y(x,t) satisfy the equation
Show that resonance occurs when w = 4aa.//C2,where ex.is any positive root of the
equation tan ex= - tanh ex.
and that the boundary value problem in o (x,t) is an analog of the problem in Seco91.
where a2 now represents EJ p and p is mass of the material per unit volume. Show that
torsional resonance takes place when w = aA..I(4e),where A..is any positiveroot of
the equation tan A. = k/A.and where k = epl p/l.
PROBLEMS IN MECHANICAL VIBRATIONS [SECo 93 275
15. The propeller shaft in Probo 14 is 150 ft long with a diameter of 10in., and made
of steel. The moment of inertia 1 of the propeller equals that of a steel disk 4 in. thick
and 4 ft in diameter. The steel weighs 0.28 lb/in.3 (gp = 0.28),and Es = 12 X 106 lb/in. 2
Show that the lowest ftywheel speed at which torsional resonance takes place is approx-
imately 135 revolutions per minute.
9
Generalized Fourier Series
We shall use theory developed in Chap. 6, on the inversion integral for the
Laplace transforrnation, to assist us in establishing some basic theory of
eigenvalue problems in differential equations. The theory shows that the
solutions of certain types of homogeneous boundary value problems in
ordinary differential equations form orthogonal sets of functions, and that an
arbitrary function can be represented by a series of functions of such a set, a
generalized Fourier series.
A classical method of solving some types of linear boundary value
problems in partial differential equations begins with a separation ofvariables,
then requires a representation of a prescribed function by a generalized
Fourier series. The orthogonal functions in the series are the solutions of an
eigenvalue problem that arises from the separation of variables. The method
will be illustrated at the end of this chapter.
In the following chapter we shall see how such eigenvalue problems
are encountered in a process of designing an integral transformation to fit
276
...
GENERALIZED FOURIER SERIES [SECo 94 277
We begin with some concepts and an existence theorem from the theory
of linear ordinary differential equations.
The adjoint D!y of the linear differential form
(1)
of the second order is this related form :
(2) D!J = (IX2(X)y(X)]" - (IXl(X)y(X)]' + IXO(X)y(X).
(3)
vD2y - + IXIVY].
yD!v = d~(rJ.2VY'- (rJ.2V)'Y
(5)
V~2Y - y~2V = :x(Vy' - v'y)r].
When both members of the differential equation
(6) y"(x) + Pl(X)Y'(x) + Po(x)y(x) = f(x)
are multiplied by an integrating factor
(7)
r(x) = exp [f Pl(X) dx J.
the equation assumes its self-adjoint form
(8) (r(x)y'(x)]' - Q(x)y(x) = F(x),
where Q(x) = - r(x)Po(x) and F(x) = r(x)f(x).
Adjoints of differentialforms of order higher than 2 are defined in a
manner that corresponds to the definition (2) (Prob. 3, Seco97). But linear
differential equations of higher order do not always have self-adjoint forms.
..
278 SECo 95] OPERATIONAL MATHEMATICS
THEOREM
which clearly has a solution Y(X,A.)==O. But when the parameter A has any
one ofthe values n2n2 (n = 1,2,.. .),problem (10)has the additional solutions
(11) y = k sin nnx,
where the constant k is arbitrary. AIso, note that the problem
(12) y"(x) + n2y(x) = O, y(O)= O, y(l) = 1,
has no solution, since the function k sin nx that satisfies the first two condi-
tions fails to satisfy the condition y(1) = 1 for any value of the constant k.
95 GREEN'S FUNCTIONS
1 See, for instanee, E. L. Inee, "Ordinary Di!ferential Equations," pp. 72 !f., 1927, also, p. 123 for
more on adjoints, or E. C. Titehmarsh, "Eigenfunetion Expansions," p. 6, 1946.
GENERALlZED FOURIER SERIES [SECo 95 279
(3) (x # ~);
A function that satisfies conditions (2) to (4) and the stated continuity
conditions is called Green's function for problem (1). Specifically, and for a
problem with general linear homogeneous boundary conditions, we describe
the function as follows.
y-G(x,f
o (a,O) (b,O) x
Fig.99
J
I
DEFINITION
equations
(r(x)Gx(x,~)]x - Q(x)G(x,~) = 15(x - ~),
G(a,~) = G(b,~) = O,
~
(O,b)
10,a)
o (b,O) x Fig.100
GENERALIZED FOURIER SERIES [SECo 96 281
(9)
y(x) = f F(~)G(x,~)d~
the integrand of each integral and its first two partial derivatives with respect
to x are continuous functions of (x,~) in the triangular regions 1;. and T2,
respectively, in Fig. 100. Differentiation by the Leibnitz formula (Sec. 14) is
therefore valido We find that
(10)
(ry')' = f F(~)[r(x)Gx(x,mx d~ + F(x)r(x)J(x),
where J(x) = G..,(x,x - O)- Gx(x,x + O). Thus J(x) is the jump in the
value of Gx as the point (x,~) moves across the diagonal from triangle TI into
triangle ~ (Fig. 100). According to Eq. (2), J(x) = l/r(x). In view of Eq. (3),
Eq. (10) can now be written
has a unique solution u(x) of class C' on that interval. Let v(x) be the unique
solution, of that class, of the initial-value problem
= u(~)v(x) when x ~ ~.
u(b)r(b)
GENERALIZED FOURIER SERIES [SECo 97 283
(8) (rw')' - Qw = O,
w(a)= w(b)= O.
Since v satisfies the same differential equation and v(b) = O,it followsthat
v(x) = ku(x) when u(b) = O,where ku'(b) = 1. The following theorem is now
established.
(1)
(J,g) = f f(x)g(x) dx = (g,f).
The norm of a function J, the generalized concept of the length of a vector, is
(fJ)t; that is,
(2)
IIfII = {f If(xW dx y.
Two functions f and g are orthogonal on the interval if (J,g) = O. This
signifies nothing about actual perpendicularity, but instead that the product
of the functions changes sign in such a way that
(3)
f f(x)g(x) dx = O.
Corresponding to units of length that differ from one coordinate axis
to another in vector analysis, we introduce a nonnegative weight function p
and define the inner product with respect to p, of two functions, as
(5) (a ~ x ~ b, n = 1,2,...)
Ir so, and if it is true that (cPmJ)= ::= 1 cn(cPm,cPJ,we can determine the
constants cn. F or (cPm,cPn)
= Owhen n "" m, and 11 cPm11= 1. Thus (J,cPm)= cm;
that is
Series (6) with coefficients (7) is the generalized Fourier series for f
corresponding to the orthonormal set of functions </1n.The coefficients cn
are called the Fourier constants for the function f In case the orthogonal sets
are generated by certain eigenvalue problems known as Sturm-Liouville
problems, we shall establish useful conditions on f under which its genera-
lized Fourier series converges to f(x).
The set of functions Yn(x) = sin nx is a well-known example of ortho-
gonal sets. The set is orthogonal with unit weight function on the interval
(O,n)and IIYnl12 = n/2 for each n; thus the orthonormal set consists of the
functions
(8)
</1n(x) = A sin nx (O< x < n, n = 1,2,.. .).
which converges to f(x) when O < x < n at each point where f is continuous,
if f' is sectionally continuous on the interval.l
PROBLEMS
1. Verify Lagrange's identities (3) and (5), Seco94.
2. Use Lagrange's identity to show that v(x) is an integrating factor for the differential
equation
if v is not identically zero and satisfies the homogeneous equation D~v = 0, where
D! is the adjoint of D2. Thus show that D2y = f reduces to this equation of first order:
1 See the author's "Fourier Series and Boundary Value Problems," 2d ed., 1963; also, ortho-
gonal sets are treated somewhat more fully in Chap. 3 of that book.
286 SECo 97] OPERATIONAL MATHEMATICS
= -r(b)[u(b)cosP + u'(b)sinP],
GENERALlZED FOURIER SERIES [SECo 97 287
in the form
G(x,~) = u(x)v(~)
e
when x ;; ~
= u(~)v(x) when ~ ;; x.
e
y(x) = f F(~)G(x,~)
d~
of the boundary value problem in y for which Green's function G was constructed in
Probo 8.
10. When the constant k is neither zero nor an integer, use Green's function (Probs. 8
and 9) to solve this problem for y:
y"(x) + ey(x) = F(x), y'(0)= O, y'(7t)= o.
Ans. y(x)k sin k7t = I: F(~) cos k~ d~ cos k(7t - x) + f F@ COS k(7t - ~)d~ cos kx.
12. Interpret the symmetry property G(x,xo) = G(xo,x) of Green's function in terms
of displacementsofthe stringconsideredin Seco95 and concentratedloads,at pointsx
and Xo.
13. Supposea function w of c1assC' is not identicalIyzero and satisfiesthe homo-
geneous problem
[r(x)w'(x)]' - Q(x)w(x) = O, w(a) = w(b) = O.
f F(x)w(x) dx = O.
IIIustrate that result using the problem
y"(x) + y(x) = - 3 sin 2x, y(O) = y(7t) = O,
is orthonormal on the interval (O,e)with unit weight function, and that the generalized
Fourier series for a functionJ, with respect to the set, is the Fourier eosine series
l 2 '" mrx mr.;
-
e o [ f(.;)d.; + -e cos- e [o f(.;)cos-d.;e
n= 1
(O < x < e).
98 EIGENVALUE PROBLEMS
The homogeneoustwo-point boundary value problem
[r(x)Y'(X,A)]'- [q(x) + Ap(X)]Y(X,A)= O (a < x < b)
(1)
A1y(a,A) + A2Y'(a,A) = O, B1y(b,A) + B2Y'(b,A)= O,
involving a parameter A in the manner indicated, is called a Sturm-Liouville
problem or system.l The constants Al, A 2, B l' and B2 are real and independ-
ent of A. The functions p, q, r, and r' are real-valued and continuous, and
p(x) > Oand r(x) > O,over the entire interval a ~ x ~ b.
Problem (1) has the trivial solution y = O,for everyvalue of A. But for
certain values Anof A,called eigenvalues or characteristic numbers,the problem
has a non trivial solution
Y(X,An) = Yn(x),
and similarly for the second boundary condition. Thus the functions y and z
must be such that
and from these and Lagrange's identity (5), Seco94, it follows that
(6)
(J1.- A) f pyz dx = W(b)r(b) ...,.W(a)r(a) = O.
(7) f p(x)y(x)z(x) dx = O
(A =f. J1.).
which also satisfies the homogeneous equation f!)2X = A.x. The reader can
X(a) = Oand X'(a) = O because
verify that it satisfies the initial conditions
W(a) = O. Therefore X(x) == O, the unique solution of that equation under
those conditions. Moreover, the coefficient of y(x) in formula (9) cannot
vanish, in view of the second of boundary conditions (2) and because z(a)
and z'(a) cannot both vanish, for the determinant of the pair of equations
A2z(a) - Alz'(a) = O,' A1z(a)+ A2z'(a)= O
in z(a) and z'(a) has the value A22 + Al 2. That value is positive for our real
numbers Al and A2.
Likewise, the coefficient of z(x) in formula (9) is a nonzero constant.
SinceX(x) == O,it followsthat there is a constant k such that
(10) z(x) = ky(x) (k*O,a~x~b).
Finally, we show that all eigenvalues are real and have corresponding
real-valued eigenfunctions. For if y(x) is an eigenfunction corresponding to
an eigenvalue ..1.,then by writing complex conjugates of all terms of the
Sturm-Liouville problem (1), we can find that
f!)2Y= y, Aly(a) + A2Y'(a) = Bji(b) + B2y'(b) = O.
Thus y is an eigenfunction corresponding to and, according to Eq. (6), it
follows that
f
( - ..1.) p(x)y(x)y(x) dx = O.
But y is not identically zero, and yy = IYl2~ O,while p(x) > O,so the integral
here has a positive value; therefore, - A.= O. Thus = ..1.,
so A.is real.
Now since y and y are eigenfunctionscorresponding to the same real
eigenvalue ..1.,then y and y are linearly dependent; that is, for some constant
k, y = ky. In case k = -1, then y = - y and iy = iy, so iy is a real-valued
eigenfunction. If k * -1, then (1 + k)y is the eigenfunction y + y, or 2 Re y,
which is real-valued. That is, each eigenfunction can be made real-valued by
multiplying it by an appropriate complex constant.
We have now established the following properties of characteristic
numbers and functions.
f p(x)Ym(x)Yn(x)
dx = O (A.m* ..1.n)'
GENERALlZED FOURIER SERIES [SECo 99 291
99 A REPRESENTATIONTHEOREM
In the sections to follow, the existence of an infinite sequerice o eigenvalues
Anofthe Sturm-Liouville system will be established. We then 'show that the
generalized Fourier series of orthonormal eigenfunctions, for each function f
of a broad class, does converge to f(x) on the interval (a,b).
Let Yndenote real-valued eigenfunctions of the system
(1) [r(x)y'(x)]' - [q(x) + Ap(x)]y(x) = O (a < x < b),
A1y(a)+ A2J'(a)= O, B1y(b)+ B2J'(b)= O,
in case eigenfunctionsexist. Then the normalized eigenfunctions
(2)
f
00
(4) L en~n(x)
n= 1 where en = p(x)f(x)~n(x) dxo
Our theorems on representing the inversion integral by a series of
residues are useful in establishing the following representation theorem for
the series (4)0
1 The theorem is true without the condition onf", but in our proofit is convenient to assumef"
sectionaIly continuous.
292 SECo 100] OPERATIONAL MATHEMATICS
prove the general case. The proof exhibits interesting devices for finding
properties of solutions of differential equations when the equations are too
general to solve.
To shorten the analysis, we now consider the case where a2 = b2 = Oin the
boundary conditions of our reduced Sturm-Liouville problem. Except
for changes in notation, that problem then reads
(1) X"(x) - [A + q(x)]X(x) = O, X(O) = X(l) = O.
We introduce, to serve as a guide to the analysis, a boundary value
problem in partial differential equations whose solution is associated with
problem (l). Let Y(x,t) denote transverse displacements in a string stretched
between points (0,0)and (1,0)when the string is attached along its span to an
elastic medium that may be nonuniform. If the string starts from r:est with a
prescribed initial displacement Y = f(x), and if the unit of time is properly
chosen, then
Y,r(x,t)= y':x(x,t) - q(x) Y(x,t) (O < x < 1, t > O),
(2)
Y(O,t)= Y(l,t) = Y,(x,O)= O, Y(x,O) = f(x).
The differential equation and both two-point boundary conditions art
homogeneous. Hence the method of separation of variables may apply to
problem (2). First, to find all functions not identically zero of type X(x)T(t)
that satisfy al! homogeneous conditions, we can write
XT" = X"T - qXT, X(O)= X(l) = T(O) = O,
T"(t) X"(x ) q(x )X (x )
Consequently - = =A
T(t) X(x) ,
where the parameter A is independent of x and t, because T"jTis independent
of x and equal to a function that is independent of t. The function X therefore
satisfies all conditions in problem (1). Thus A must be an eigenvalue Anof
that Sturm-Liouville problem and, except for an arbitrary constant factor,
X(x) must be the corresponding orthonormal eigenfunction 4>n(x).The
function T(t) then satisfies the system T" = AnT, T(O) = O. The functions
X.T can therefore be written
(3) (n = 1, 2, . . .),
satisfies all conditions in the problem, including the condition Y(x,O) = f(x),
294 SECo 102] OPERATIONAL MATHEMATICS
where
when t = O.
L-l{y(X,s) - f~X)} = O
Then wecan anticipate that the sum of residuesofy may bejust the series(5),
and that carefulanalysiswill show that the seriesconvergesto f(x).
where
(6)
h(x,s) = v(x,s) f: f(~)u(~,s) d~ + u(x,s) f j(~)v(~,s) d~.
For each fixed x, h is an entire function of s. This follows from the Cauchy-
Riemann conditions, because the products uv are entire functions of s.
Consequently y is an analytic function of s except at the zeros of the
entire function u(l,s), which are necessarily isolated. But when u(1,s) = O,
we see from problem (2) that u(x,s) is a solution of our Sturm-Liouville
problem (1), Seco101, in which A.= S2. Ir A.nis the corresponding eigenvalue
and A denotes one of the two square roots of that real number, then
s = :t Sn where Sn =..::,. Thus if ljJnis the normalized eigenfunction,
u(x, :t sn) = kljJn(x)for some nonzero constant k. Since u'(O,sn)= 1, it follows
that ljJ~(O)=1:O and
(7)
(10)
ljJ~(1)h(x,:tsn) = :~~~; f j(~)ljJn(~) d~
ljJn(x)
= Cn ljJ~(O) ,
where Cnis the Fourier constant for our functionj Also, note that
(12)
(S2 - s/) f u(x,s)<Pn(x)dx = - <p~(I)u(l,s).
o u(x,s)<Pn(x)dx n
and the sum of those equal residues is cn<Pn(x).Similarly, if Sn= O,we find
that the residue ofy at snis cn<Pn(x).Hence the sum ofresidues ofthe function
y at all its poles s = ISn is formally the generalized Fourier series forf:
00
L cn<Pn(x).
n=1
when s =1= o, as is easily verified or derived (cf. Probo 17, Seco 104). The
solution (3) is unique because the difference w - z of two solutions w and z
of problem (2) must satisfy a completely homogeneous initial-value problem,
-
so that w z ==O. But problem (1) shows that z = u(x,s)is the solution of
GENERALlZED FOURIER SERIES [SECo 103 297
(4)
su(x,s) = sinh sx + f: q(~)u(~,s) sinh s(x - ~) d~,
where s is any complex number.
Likewise, for all s the solution v ofthe initial value problem (3), Seco102,
satisfies the integral equation
(5)
sv(x,s) = sinh s(x - 1) + f q(~)u(~,s) sinh s(~ - x) d~.
when O~ x ~ 1;
where R 1 and R2 are continuous functions whose absolute values are less
than 2qo when p > 2qo. In particular,
1
(10) su(l,s)e-a = e-a sinh S + -R1(l,s),
S
where if
We found that u(l,s) cannot vanish unless S has real or pure imaginary
values :t.JI,;. If S is real and not zero, S = :tu, then e-a sinh S approaches
:ti as o' -+ 00, and if follows from Eq. (10) that a number')l exists such that
u(l,s) =FO when o' ~ ')l. All real zeros of the entire function u(1,s),therefore,
le on a bounded interval of the real axis; hence they are finite in number.
When S is apure imaginary number, o' = Oand s = ir,so
(12)
(13)
With the aid of the last two equations we find that the norm 11rnu(x,sn) 11
Since u is of class C' in x, then 4>nis also of that class. Qur principal results
are stated as follows.
GENERALlZED FOURIER SERIES [SECo 104 299
than afinite
infinite number
set (ll) ofthem are A.n.
of eigenvalues positive. The value of
All eigenvalues - A.nand
arjeal, is arbitrarily
no more
close to an integral multiple of 1t when n is sufficiently large. The corre-
sponding orthonormal eigenfunctions <Pnare of class C', and <Pn(x}and
<p~(x}satisfy the boundednessconditions (14).
To prove that the series of residues of y converges to f(x}, we first find order
properties of Green's function G and our function y.
Againwewriteu = IResi and p = Isl. When x ~ ~,
) - u(x,s}vR,s} - [se-"Xu(x,s}][se-"(l-~)v(~,s)]
G(x,~,s- u(l,s) - s2e-"u(1,s)e"(~-X) .
In view of conditions (7) and Eq. (10), Seco103, then
4
(1)
.h
IG(x,~,s)1 < pe1 -" sms + s -l R l'S
(1 )1
The right-hand member here is &(s- 2) on the half plane Re s ~ 'Yand on the
paths CN, in view of condition (2). The left-hand member of the equation
therefore satisfies the same conditions, and so its inversion integral along
the line Re s = 'Y has the value zero when t = O; also (Sec. 73) its integral
over CN vanishes as N -. oo. That is, if CN denotes the closed rectangle
with the side along Re s = 'Yadded to CN, then
-. oo.
(4) ICN [ y(x,s) + ~ I: f(~)G~~(x,~,s)
d~ ds -. O ] as N
Now let us assume 1", and therefore f' and J, sectionally continuous.
It will suffice to let f have just one jump of an amount jo at a point XO
(O< Xo < 1). Let x be a fixed point of the interval at whichf is continuous,
say O < x < Xo. The jump in G~ at the point ~ = x is 1, as we can see from
the expression for G in terms of u and v. Then
l x xo l
if we write f(O) for f( + O) and f(l) for f(1 - O). Note that G~x,O,s)is
u'(O,s)v(x,s)/u(l,s)and u'(O,s)= 1, etc. Thus in terms of the function
(6)
GENERALIZED FOURIER SERIES [SECo 104 301
The final member of the sum (6) is (!J(s-Z)on CNas we can see by
integrating the sectionally continuous function l' G~by parts and recalling
that G itself and therefore such terms as jG(x,x,s) and gf"G d~ are (!J(S-l).
Thus the integral around CN of that member tends to zero as N -+ oo.
We find, corresponding to Eqs. (8) and (9), Seco103, that
(7) sv(x,s)= -sinh s(l - x) + S-lea(1-X)R3(x,s),
(8) v'(x,s) = coshs(l - x) + s-lea(1-X)R4(x,s),
where R3 and R4 are bounded when Isl > 2qo and O ~ x ~ 1.
We write the coefficient of jo in the sum (5) in the forms
1
~ 1 [sinh sx + s-leaXR1(x,s)][cosh s(l
s u( ,s) - xo) + S-lea(1-xo)R4(xo,s)]
sinh sx cosh
. s(l - xo) + R1(1,s)sinh
;--. sx cosh s(l - xo) + g 1(X,Xo,S)
s smh s s smh s se-a smh S + R1(1,s )
tf(e,s)= si~h es
s smh s where - 1 < e < 1.
Actually, tf is the transform of a square-wave function which vanishes over
an interval containing the origin t = O(Prob. 9, Seco104).
An examination ofthe integral of 1tf1 around CN shows that JCNtfds -+ O
as N -+ 00 (Prob. 11 below). Similarly, the integrals around CN of the
coefficients of f(O) andf(l) in the sum (5) vanish as N -+ 00; the reduction
in terms of the function tfis simplerfor those coefficients.Thus JCNgds -+ O
as N -+ oo.
Consequently, condition (4) reduces to the equation
(9)
.
11m
i[
N oo CN
-
f(X)
y(x,s) - -
S]
ds = O;
302 SECo 104] OPERATIONAL MATHEMATICS
that is, the sum of the residues of the integrand at the poles inside CN con-
verges to zero as N -+ 000 But the sum of residues of y were found in Seco102
to be 'I:.cnl/Jn(x);therefore
f
00
(lO) L cnl/Jn{x) -
n= 1
f{x) = O,
where Cn = f{x)l/Jn{x)dx.
There are no essential changes in the proof of formula (lO) when
Xo < x < 1. Thus :rheorem 4 is established for the reduced form of the
Sturm-Liouville problem with boundary conditions X{O) = X(1) = O. A
similar procedure, but more involved, can be followed to prove the theorem
when the more general boundary conditions of type (5), Seco 100, are usedo
Theorem 4 as stated in Seco 99 then follows by transforming the
reduced problem back to the original Sturm-Liouville problem with the
substitutions (2), Seco 100.
PROBLEMS
1 . Prove that the eigenvalues A..of the Sturm-Liouville problem
y"(x) - (A.+ h)y(x) = O, O< x < e; y(0) = y(c) = O,
where h is real and constant, consist of the numbers -h- n2n2/el (n = 1,2,0 o0)0 Note
that not more than a finite number of them are positive, and that - h is not an eigenvalueo
Show that the normalized eigenfunctions 4J. are .j2jC sin (nnx/c) and that according
to Theorem 4, iff" is sectionally continuous (O< x < e), then f(x) is represented at
each point wherefis continuous by its Fourier sine series on the interval (O,e):
2 QO nnx nn~
f(x) =- Lsin -
e .= 1 e l c
o
f(~) sin - e d~ (O< x < e).
in .=f 12n
~sin
- 1
(2n -e l)xx = 1 (O< x < c)o
o
f(~) d~ +
QO
-e.=L1cos - e l
c
o
nn~
f(~) cos -
e
d~ (O< x < e),
Then use them to derive the following expressions for the eigenvalues and eigenfunctions
of the problem in y:
Al = 0, A.= -4a.2 (n = 2,3,.. o)
prove that two eigenfunctions y and z and their corresponding eigenvalues A and .L
must satisfy the condition
J: y(x)z(x) dx
(A - .L) = 0,
and hence that ifA '# .L,y is orthogonal to z on the interval (O,e);also that all eigenvalues
must be real.
7. In Seco 100, complete the derivation of form (1) of the Sturm-Liouville equation
and of the substitutions (2) that produce the reduced form (4).
8. Derive formulas (7) and (8), Seco104, for sv(x,s) and v'(x,s).
9. Let'l' be this periodic square-wave function:
'I'(c,t)= if < t < 1 - e or if 1 + c < t < 2,
'I'(c,t)= 1 if 1 - e < t < 1 + c,
'I'(c,t + 2)= 'I'(t) whent > 0,
-
where ;;;e < 1; also '1'( c,t)= - 'I'(e,t). Show '1' graphicallyas a function of t.
Use Theorem 10, Seco23, to prove that its transform is the function t/J(e,s)used in
Seco104:
sinh es
L{'I'(e,t)} = t/J(c,s)= s sinh s (-1 < e < 1, Re s > O).
11 . If we write s =v+ i", the sides of the cIosed rectangle CN used in Seco104 lie
on the lines v = y(y > O), v = - PN, and" = IPN, where PN = (N - t)7t, N = 1,2,....
For the function
sinh es
(-1 < e < 1)
tf(e,s)= s sinh s
establish this property, used in Seco104:
lim
N-C() JCN
f tf(c,s)ds = O,
with theaid ofthe result found in Prob.lO.
12. If the function f is continuous (O~ x ~ 1) and satisfies the boundary conditions
in the reduced Sturm-Liouville problem (1), Seco 101, so that f(O) = f(l) = O,and if!"
is sectionally continuous, prove that condition (9), Seco104, is satisfied uniformly in x;
hence that
<X>
f(x) L
= n=l c.4>.(x) (O~ x ~ 1)
X.(x) = A. fX.(~)G(x,~,O)d~
where A.is the eigenvalue corresponding to X. and G is Green's function (4), Seco102.
Notation used in Probs. 14 to 18: Let N.[X] and N[X]and their derivatives
with respect to the real-valued parameters exand Pdenote these boundary values on the
interval O~ x ~ 1:
14. Use Green 's function (Probs. 8 and 9, Seco97) to write the solution of the boundary
value problem (c) above in the form
,-
,-O'
-
o U-}1x1 (1,0) Fig.101
X(x) = e coshx..ji.
It satisfiesthe third condition if
(6) h cosh ..ji + ..ji sinh ..ji = O.
According to Theorem 3, all eigenvalues of problem (4) are real. Ir A > O,
GENERALIZED FOURIER SERIES [SECo 105 307
an equation that has an infinite set of roots :!:CXn(n = 1,2,. . .) that can be
approximated graphically. In fact the number CXn is only slightly greater than
mI:when n is large.
The complete set of eigenvalues of problem (4) is therefore
(n = 1,2,. ..),
where the numbers cxnare the positive roots of Eq. (7),and the corresponding
eigenfunctions are
Xn(X) = COS CXnX.
CXn
CXn
(10) (1 1 (1
Cn= Jo F(x)<fJix)dx = Pn Jo F(x) cos OCnx
dx.
Note that the series used to represent F(x) here,
G() e
(11) F(x) = cos
n=1Pn
...!!. (l(nX (O< x < 1),
is a generalized Fourier series, not the Fourier cosine series for the interval
O<x<1.
The formal solution of our problem is therefore
G() e
(12) U(x,y) = exp ( - IXnY) COS IXnX,
n= 1 Pn
...!!.
where the constants Pnand en are given by formulas (8) and (10).
We order the positive numbers (l(nso that (l(n+1 > (l(n'Then the functions
exp (-(I(nY) in our formal solution (12), Seco 105, are bounded, and non-
increasing with respect to n:
0< exp(-(I(nY) ~ 1, exp(-IXn+1Y)~ exp(-(I(nY) (y ~ O).
According to Abel's test (Prob. 6, Seco81), therefore, the series
G() e
(2) (y ~ O,O < x < 1)
n~1 P: cos (l(nXexp (-(I(nY)
converges uniformly with respect to y when y ~ O. Since its terms are con-
tinuous functions of y, its sum U represents a continuous function of y when
y ~ O. Qur formal solution therefore satisfies the condition
U(x, + O) = U(x,O) = F(x) (O< x < 1).
GENERALlZED FOURIER SERIES [SECo 107 309
From Eq. (8), Seco105, we seethat Pn > l/fi. Also, a fixed number M
exists such that ICnl/Pn< M. Whenevery ~ Yowhere Yo> O,the absolute
values of the terms in series (2) are less than the constants M exp ( - anyo);
also the absolute values of the deriva ti ves of those terms, of first and second
order, with respect to x and y, are less than Man exp (-anYO) and
M an2 exp ( - anyo), respectively. The series of those constant terms converges,
according to the ratio test, since an - me O as n-oo. -
Hence series (2)
and the series of derivativesof its terms convergeuniformlywith respect to
x and Y over each strip O ~ x ~ 1, Y ~ Yo > O. The terms are continuous
functions, so the series converge to a continuous function U(x,y) and its
continuous partial derivatives.
In particular, U and Ux are continuous functions of x on the interval
O ~ x ~ 1,when Y > O. They satisfy the homogeneous boundary conditions
Ux(1,y) + hU(l,y) = O
because the terms in series (2) satisfy those conditions. The terms also satisfy
Laplace's equation (1), Seco 105; hence the sum U satisfies that equation
when Y > Oand O < x < 1.
FinaIly, we note that the sum U(x,y) satisfies the remaining condition
-
U O and y -
00 because each term satisfies that condition and the re-
mainder after N terms is arbitrarily smaIl in absolute value, uniformly with
respect to x and y(y ~ Yo),when N is sufficientlylarge.
Thus formula (12),Seco105, is established as a solution ofthe boundary
value problem in the temperature function U.
But if the interval does not contain the origin X = O as an end point or as
an interior point, these eigenvalue problems in Bessel's equation, with a
homogeneous boundary condition at each end of the interval, are non-
singular special cases of the Sturm-Liouville problem.
LEGENDRE'S EQUATION
(5)
[(1 - X2)X'(X)]' - AX(X) = O (-1 < x < 1),
X and X' continuous on the closed interval - 1 ;;;; x ;;;; 1,
is another example of a singular problem. Legendre's differential equation
here has singular points at both end points x = :!: 1 of the interval. Its solu-
tion satisfies the continuity condition only if A = - n(n+ 1),wheren =
O,1,2, .... Corresponding to those eigenvalues, the eigenfunctions are the
Legendre polynomials
(6) )- 1
P( x--L.- ~ (-1}i (2n - 2j)!
x
n-2j (O!= 1)
n 2nj=O j! (n - 2j)!(n - j)!
wherem = tn ifn is even and m = t(n - 1)ifn is odd. They form an ortho-
gonal set with weight function p(x) = 1 on the interval (-1,1). The repre-
sentation of a functionj on that interval, such thatf' issectionallycontinuous,
as a generalized Fourier series of the eigenfunctions, can be established
rather easily by using properties of Pn. It can be written
FOURIER INTEGRALS
PROBLEMS
Apply the method of separation of variables to the boundary value problems in partial
differential equations that follow.
1. The steady-state temperature problem (cf. Seco105)
u ",,(x,y) + Uyy(x,y) = O, IU(x,y)1< M, (O< x < 1,Y > O)
U x(O,y) = U x~,y) = O, U(x,O)= F(x) (O< x < 1),
where M is some constant. Show that A.= Ois an eigenvalue of the associated Sturm-
Liouville problem and obtain the formal solution in the form
1 "" 1
-2 00
7. Let U(x,y) be the steady-state temperatures in a thin plate in the shape of a semi-
infinite strip, from which heat is transferred at the faces into a medium at temperature
zero, so that
where An = (1 - cos (Xn)/(h+ cos2 (Xn)and (XI' (X2,. .. are the positive roots of the
equation tan (X= -(X/h.
8. Let U(x,y)be the steady'state temperaturesin an infiniteprism bounded by the
planesx = O,Y = O,x = 1,andy = 1. If U = 1onthefacey = 1andifUx = - hU at
x = 1,whereh > O,and U = Oon the other two faces(Fig.102),derivethe formula
Y/ U=1 0,1)
Fig.102
U=O
0
O U=O--X
-
=
-O.
.-
U(x,t)= 2 f
..!..(t+ Wm2 sin mx.
n=1m
Verify that this function satisfies the condition U(x,+O) = 1(0 < x < 1).
314 SECo 107] OPERATIONAL MATHEMATICS
Fig.103
10. The end x = 1 of a stretched string is elastically supported (Fig. 103)so that the
transverse displacement Y(x, t) satisfies the condition y"(l,t) = -hY(1,t), where h > O.
Also, let Y satisfy the conditions
Y(O,t)= O, Y(x,O) = bx, y'(x,O) = O.
where 1X1,1X2,. . . are the positive roots of the equation tan IX= -IX/h.
11. The longitudinal displacements Y(x,t) in a certain nonuniform bar satisfy the
conditions
V(r,z) = L A.
11=1
exp (-IX.z)t/!(rx.,r),
are the roots ofthe equation tJ(a,b) = O,all reaP am > O,and An are given by theformula
13. The temperature V(r,t) in a solid cylinder r ~ 1 of infinite length satisfies the
conditions
also V and V. are continuous when O ~ r ~ 1 and t > O. Show that the method of
separation of variables involves a singular eigenvalue problem of type (1), Seco 107.
With the aid of the integration formula
[XJo(X)dX = cJ(c),
solve for V in the form
V(r,t) = 2 L.. - J
~ J o(a.r)
exp(- a. t), 2
.= a. (a.)
where the a;s are the positive zeros of J o(a).
14. The temperature U(x,t) in a semi-infinite solid x ~ O, whose initial temperature
function is a step function, satisfies the conditions
and U is bounded. Show that the solution by separation of variables involves the
singular eigenvalue problem (8), Seco107, and derive the formula
2 "" sin a
U(x,t) = - -cosaxexp (-a2kt) da.
7t1o a
15. Prove that the eigenvalues and eigenfunctions of the singular problem
X" - J.X = O,O < x < co; X(O) = O; IX(x)1~ l,x > O;
are J. = -a2(a > O) and X = sin ax. If l' is sectionally continuous and absolutely
integrable over the half line x > O,note that, according to the Fourier integral formula
(Sec. 68) for odd functions, f is represented in terms of those eigenfunctions by the
F ourier sine integral formula
2 "" ""
f(x) = - 1 sinax 1 f(~) sin a~d~ (x > O).
7t o o
16. Use the results found in Probo 15 to solve the problem in Seco43 for the displace-
ments Y(x, t) of an initially displaced semi-infinite stretched string, namely,
I See H. S. Carslaw and J. C. Jaeger, "Conduction of Heat in Solids," 2d ed., p. 206, 1959, and
their tables, p. 493.
316 SECo 107] OPERATIONAL MATHEMATICS
in the form
y = ~ (00[sin ex(x+ at) + sin ex(x- at)] (00 cI>(~) sin ex~ d~ dex.
nJo Jo
Show how this reduces to the form (8), Seco 43.
17. Prove that the eigenfunctions of the singular problem
(1)
T{F} = f F(x)K(x)) dx = f(A.).
The resulting function f(A.) is the transform of F under that transformation
with kernel K.
Of course the class of functions from which F and K are chosen must
be restricted and a set A of values of the parameter A.must be specified so
as to ensure not only the existence ofthe integral (1) but also the uniqueness
of the inverse transformo That is, not more than one function F should
correspond to a given transform f(A.).
317
318 SECo 108] OPERATIONAL MATHEMATICS
Ir inverse transforms T-1 {J} are unique, the linearity condition (2) can be
written
(3)
For the Laplace transformation the kemel K(x) = e-.I.x has the simple
product property
(1) K(x,).)K(xO,A) = K(x + Xo, A).
Then if f(A) = L{F} and F(x) = O when x < O,it followedthat L has the
translation property (Sec. 11)
(2) f(A)K(xO,A) = L{ F(x - xo)} (xo > O).
This formula represents the product of the transform of F by the kemel as
the transform of a function related to F. We call it a kernel-product property
of the Laplace transformation. That property led us to the convolution
property (Sec. 16)
(4)
T{F} = f F(x)K(x,A) dx = f(A),
suppose we have a modified or weighted kemel-product property such that
for some specifiedfunction W(A)and for each function F there is a function
PF(x,xo) for which
(5) (a < xo < b).
(6)
f(A)g(A)W(A) = f g(A)w(A)K(y,A)F(y) dy
= f T{PG(x,y)}F(y)dy
f K(X,A) f F(y)PG(x,y) dy d~
320 SECo 110] OPERATIONAL MATHEMATICS
(7)
XFG(X) = f F(y)PG(x,y)dy.
We call the function X FG the convolutionof F and G for the trans-
formation T, with respect to the weight function w. It is the inverse
transform of the product fgw. That is,
(8) f(A)g(A)W(A) = T{XFG(x)} = T{XGF(x)},
110 EXAMPLE
We illustrate the results in the preceding section by deriving a convolution
formula for the Fourier sine transformation on the half line x > O,
(3)
H(x) = foo Ft(~)d~ = foo Ft(~)d~ + I: Ft(~)d~
(-00 < x < (0)
and note that it is even, H( - x) = H(x) or H(lxl) = H(x),and that it vanishes
as x --+ :t 00; also, H'(x) = Ft(x).
To find a icernel-product property of SJ..,we write
oo
2 1
lJ.<A) =l f -00 Ft(x) sin AXdx = f oo
-00
sin AX
x-H'(x)dx
00 oo
sin AX
[
= -H(x)
A ] f
-00-00
- H(x)cos AXdx.
x+y 1
also, PG(x,y) =
f.Ix-yl e-a~ d~ = -[e-alx-yl
o
- e-a(x+y)].
1 1
--
o o f oo
e-aYe-a(x+y)dy = -xe-ax.
o
Thus we have this useful Fourier sine transformation
PROBLEMS
1. Use the convolution property (7), Seco110,when
F(x) = e-ax, G(x) = e-bx (a > 0, b > 0, a '# b)
to obtain this Fourier sine transformation:
(This result can be found also by using partial fractions.) Evaluate the integral trans-
formation on the right to verify the result.
2. Interchange the order of integration in formula (6), Seco 110, to show that the
convolution X FGfor the Fourier sine transformation SA is symmetric in F and G:
XFG(x) = XGP(x).
GENERAL INTEGRAL TRANSFORMS 323
[SECo 110.
).2
1
+ 1-
- E {1 -lxl }
A '[e .
Verifythe result by evaluatingthe integral EA{e-lxl}.
5. When K(x,).)= cos ).x, a = O, and b = 00, our transformation T becomes the
Fourier cosine transformation eA on the half line (Chap. 13),
asinH = C{XFG(x)}.
8. Generalized convolution For a function H(x,y) of two variables, let h(A.,y)denote
the transform T{H} with respect to x,
. h(A.)= f h(A.,y)K(y,A.)dy,
if H and T are such that this iterated transform exists.
In case T has a kernel-product property
h(A.)w(A.) = f [f QH(X,y)dY]K(X,A.)dX.
Note that this reduces to property (8), Seco 109, in case H(x,y) = F(x)G(Y); then
QH = PF(x,y)G(y). Examples are given in Probs. 9 and 10.
9. When the transformation in Probo 8 is the exponential Fourier transformatin E
describedin Probo3 and w(A.)= 1, show that the generalizedconvolution property
becomes
GENERAL INTEGRAL TRANSFORMS [SECo 111 325
Ji(l) = (1 + iA)e-iA
l - 1
and verify that this agrees with the convolution property in Probo 9.
and let the interval (a,b) be bounded. For the sake of simplicity we assume
that aIl functions of x here, together with their derivatives of {he first and
second order, are continuous over the cIosed interval a ~ x ~ b, although
those conditions can be relaxed. Also, we assume that A(x) > O over that
cIosed interval. In applications to problems in partial differential equations,
F will be a function of x and the remaining independent variables, while
A, B, and e will be functions of x alone.
The boundary values with respect to x, to be prescribed as functions
of the remaining variables, are
that transforms D2F into an algebraic form in f(A), NaF, and NpF, it is
convenient to write D2F in terms of a self-adjoint differential form
(4) !!)2F = [r(x)F'(x)]' - q(x)F(x).
where the weight function p is rlA, and the kernel el>and the range of the
parameter A are yet to be determined. Then the transform of D2F becomes
f~eI>!!)2F
dx and, either by integration by parts or by using Lagrange's
identity (5), Seco94, according to which
we can write
(8)
T{D2F} = f F(X)!!)2e1>(X,A)dx + [(eI>F' - eI>'F)rJ:.
(9)
(11)
T{F} = f F(x)p(x)$n(x) dx = f(An)
(n = 1,2,...).
a F(~)p(~)
$n(~)
11$n 11d~ =
ex> $n(X)
n~l II$nlld(An),
Y(x) = f F(~)p(~)G(x,~)d~
(a ~ x ~ b).
GENERAL INTEGRAL TRANSFORMS [SECo 113 329
- r(b)Np[DzF]Np<l>".
+ Na[DzF]N~<I>"
330 SECo 113] OPERATIONAL MATHEMATICS
T{D2F} = fK(X,A)D2[F(X)] dx
is to be written in terms Ofj(A) and the two boundary values NaF and NpF.
In terms of the adjoint D! of D2, Lagrange's identity (3), Seco94, shows that
d
KD2F - FD!K = d)AKF' - (AK)'F + BKF]
where r(x) = exp f: [B(~)/AR)] d~, and r' = rB/A. Therefore when F and K
are of class C",
(2)
(3)
(4) D!K - AK = O,
GENERAL INTEGRAL TRANSFORMS [SECo 113 331
Note that
(6) (n = 1,2,...).
Formulas (3) to (6) follow also from those in Seco 111 when the latter
are written in terms of the kernel Kn and the coefficients A, B, and e in D2,
where Kn = pCl>n.
The basic operational formula for T{D2F} must be modified in case
F or F' has jumps. Suppose, for instance, that both functions are eontinuous
over the interval a ;;;; x ;;;;b exeept at one interior point x = e where they
have jumps jc and j~:
where T{F} = f~FpCl>ndx and Cl>nand An are the characteristic functions and
numbers of the Sturm-Liouville problem (10), Seco111.
We cite one more property of Sturm-Liouville transforms f(An).
When F is sectionally continuous, its Fourier constants
en = IICI>nll-l f F(x)p(x)CI>n(x)dx
(n = 1, 2, . .. )
(8)
Suppose Tis such that no eigenvalue Anis zero. Then Green's function
G(x,c)for problem (4), Seco112,exists. It is continuous, but its derivative
Gx has thejumpj~ = [r(c)r 1at the point x = c (a < c < x); also,
(2) (x"' c),
According to the modified operational property (7), Seco113, then
0= AnT{G(x,c)} - r(c)[r(c)r1$n(c).
Thus the Sturm-Liouville transform of G(x,c) is
1
(3) (a < c < b).
T{G(x,c)} = :f$n(c)
n
The continuity of G(x,c)and the sectional continuity of its first deriva-
tive are sufficient (Theorem 4, Seco 99) for the validity of our inversion
formula (1), Seco 112. Therefore we have the representation
(7)
h(An) = ;n {r(b)Np[<I>n]NpH - N~[<I>n]N",H}
If constant values, not both zero, are assigned to N",Hand NpH here, then
H is the solution of the differential equation D2H = O that satisfies those
boundary conditions. We note that in case IX= P = O,formula (7)becomes
(8) T{H} = An-l[r(b)<I>~(b)H(b) - <I>~(a)H(a)].
If F(x) == 1 (a ~ x ~ b), then D2F = C(x) and our formula for T{D2F}
reduces to
(9) - r(b)Np[<I>n]cosP.
T{C(x)} = AnT{l} + N~[<I>n]COSIX
Thus when <l>n,An, and either of the two transforms T{C(x)} or T{I} are
known, the other transform can be found. If C(x) = Co, a constant other
than one of the eigenvalues,then T{1}can be found,since
(10)
(4)
Sm{F}= f F(x) sin (n - t)xdx =fs(m) (n = 1,2,...).
334 SECo 115] OPERATIONAl MATHEMATlCS
!.
<X>
(6) F(x) = -
n n= 1
f.(m) sin mx (m = n - O < x < n).
(7) (m = n - !).
For other particular transforms see Probs. 5 to 8, Seco 116, and Table 2,
Seco 126.
We can derive kernel-product and convolution properties for Sm,
whose kernel is an odd function that is antiperiodic with the period 2n.
Let F3 be the extension of F which is odd and antiperiodic with period 2n;
~~ .
~ ~w=[~m~=[~m~-[~m~
which is even and antiperiodic with period 2n; also F(x) = - F3(x) and
Fo(n) = Fo(-n) = O. Then
beca use the integrand is periodic with period 2n. AIso, the integrand is an
even function of ~, SOthe last equation is the kernel-product property
(10) (m = n - t).
(11)
if x + Xo~ n
if x + Xo~ n.
(12)
In terms of the notation used in Seco 109, w(A.)= 2/m and PF(x,xo) is
the function inside the braces in property (10). Hence the convolution of
two functions H(x) and F(x) corresponding to our kernel-product property is
XHF(X)
which can be written
= f H(y)[Fo(x - y) - Fo(x + y)]dy,
Example 1 Let the interval be unbounded, O < x < 00, and let the differ-
ential form and the prescribed boundary value and order condition
be these:
(1) D2F = F"(x), x> O; N"F = F(O), F(oo) = F'(oo) = O.
GENERAL INTEGRAL TRANSFORMS [SECo 116 337
(2)
T{F} = fOF(x)eI>(x,..1.)dx = f(..1.)
" (11)
T{F} = I: F(X)XJo(J1.~)dx = fo(J1.},
GENERAL INTEGRAL TRANSFORMS [SECo 116 339
PROBLEMS
1. When D2F = F"(x), O< x < b, N.F = F(O), and NpF = F(b), show that the
Stunn-Liouville transformation becomes the .finite Fourier sine transformation on the
interval (O,b),
b mrx
T{F} = Sn{F} = f
o F(x) sin Tdx = fs(n)
(n = 1,2,. .. );
+ nb1t[F"(O)- (-l)nF"(b).
(b) If F(4)(X) = O, F(O) = F(b) = F"(O) = O, and F"(b) = 6b, show that F(x) =
x(X2 - b2) and fs(n) = 6b4( -lf/(n1t)3.
3. When D2F= F"(x), O< x < b, N.F = F'(O),and NpF = F'(b), show that the
Sturm-Liouville transformation becomes the .finite Fourier cosine transformation on
the interval (O,b),
b n1tX
(n = O,1,2, . . . )
Cn{F"}= -(n;ffc(n) - F'(O)+ (-l)nF'(b)
and that Cn-l{fc} is given by the Fourier cosine series
1 2 00 n1tX
(O< x < b).
F(x) = bfc(O)+ b n~lfc(n)cosT
340 SECo 116] OPERATIONAL MATHEMATICS
4. Find the Sturm-LiouvilIe transformation on the interval (0,1) that resolves the
differential form F"(x) in terms of the transform of F and the boundary values F(O)
and hF(I) + F'(l), where h is a positive constant, and show that it has the operational
property
T{F"} = -/l/T{F} - F(O) + [hF(l)+ F'(l)]cOS/l.,
wherep. (n = 1,2,... ) are the positive roots of the equation tan /l = h//l (cf. Seco 105).
5. Show that Green's function for the problem Y"(x) = 0, Y(O)= Y'(1t)= O is the
function
G(x,e) = - x (O~ x ~ e), G(x,e) = - e (e ~ x ~ 1t);
then apply formulas (3) and (6), Seco114, when T is the modified finite Fourier trans-
formation Smdefined in Seco115, where m = n - t. to obtain the inverse transforms
1 sin me l cosme
Sm- {--;;r-} = -G(x,e), Sm- { ;;;- } = So(x- e),
whereO< e < 1tand Sois the unit step function.
6. Showthat the transformation(Sec.115)
(m = n - t)
Sm{F} = s: F(x) sin mx dx = f.(m)
has the property
S -1 ~ = 1tX - X2
m { m3 } 2'
F(x,y) = ~ I
1t.=1
e-my sin mx =~
1t
1m
.=1
I (exp iz).-t,
where z = x + iy and lexp izl < 1. Sum the series here to show that
8. For the transformation S", in Seco115, where m = n - t, show that if y> O and
independent of x and z = x + iy, then
- e-my
S", 1
{- }
m
2
= - 1m
7t
L (expizr
00
n= 1 m
(.t~ O)
C/l{F} = {') F(x) cos J.lXdx = fc(J.I)
(Prob. 5, Seco110 and Chap 13). Show that
where the real numbers J-ljare the positive roots of the equation Jo(.t)= O. Since
J(.t) = - J 1 (.t), show that
(4)
T{ U} = f U(x,y) sin J.lnXdx = un(Y)
(n = 1,2,...).
Now Na.U = P(y), IX= O,and N~$n = $~(O)= J.ln;also NpU = O,and
An= - J.l/. Therefore
T { U xx} = - J.ln2un(y) + J.lnP(y)
(7)
1 - cosJ.ln 1 - cosJ.ln
un(y) = Qo 3 - Qo 3 exp (- J.lnY).
J.ln J.ln
GENERAL INTEGRAL TRANSFORMS [SECo 118 343
l - COSJl.n
Z(x) = -~ 1 + 2h +
21+h
!X2 = -T-l
2 { Jl./ }
+4 ~
- Qo 1 + 2h - 2
(8) U( X,Y) -
[
2 1 + hx x L...
n= 1
1 - cos Jl.n exp (- Jl.nY).
1 + h cos 2 Jl.n 3
Jl.n ].
SIn Jl.nx
Form (8) can be verified fully as the solution of problem (1) in this
special caseP = O,Q = Qo. Note that if only a finite number of terms of
the infinite series is used in formula (8), the resulting function satisfies all
conditions in problem (1) except the condition at the base y = O. Also, we
see that U has the property
. Qo l + 2hx - x 2 .
11m U(x,y) = -2 -
y-oo ( 1+ h )
The differential form Uyy and boundary conditions in problem (1) are
such that the Fourier sine transformation with respect to y, on the half line
y > O, applies to the problem provided that the transforms of both the
functions P and Q exist; but that is not the case when Q(x,y) = Qo. In case
the second method does apply, we have an advantage of obtaining a different
representation of the function U. Note that problem (1) is not adapted to a
Laplace transformation. Since the partial differential equation is not
homogeneous, the method of separation of variables does not apply.
For further properties of transformation (4) see Seco 128.
"'
118 OTHER BOUNDARYVAlUE PROBlEMS
Other special and singular cases of Sturm-Liouville transformations will be
applied to boundary value problems in the chapters to follow. Operational
properties of some of them are adequate to represent solutions of problems
in more than one form, sometimes in quite simple forms. But in each
application of our process an eigenvalueproblem must be solved to deter-
mine the kernel of the transformation; then the transformed problem must
be solved to find the transform of the unknown function. Failure of the
transformedproblem to have a unique solution suggeststhat conditions are
missing or incorrect in the boundary value problem.
344 SECo 118] OPERATlONAL MATHEMATICS
Note that the coefficientsA, B, and e are functions of x alone; rxand /3are
real and constant. Then as before
where the e's are constants. When Tis applied to Eq. (1) and (3) with inter-
changes of operations with respect to x and y, then, in view of conditions (2),
the resulting differential equation is
GENERAL INTEGRAL TRANSFORMS [SECo 118 345
PROBLEMS
(m= n - t)
Sm{U(x,y)} = J: U(x,y) sin mx dx = u.(y)
applies to this problem in steady temperatures:
U",,(x,y) + Uyy(x,y)= O, U bounded, (O<x<x,y>O),
U(O,y)= U",(x,y)= O, U(x,O)= 1 (O< x < x).
Show that u.(y) = m-le-my and thus (Prob. 8, Seco116)
2. Use the transformation Sm(Sec. 115) with respect to x to find the solution of the
problem
uxx(x,y) + Uyy(x,y)= O, U bounded, (O< x < 1t,Y > O),
U(O,y)= Ux(1t,y)= O, Uix,O) = -1,
in the form U(x,y) = Sm-1{m-2e-my}; thus show that
x
U(x,y)= -log
1t
x cosh y/2 + cos x/2
i
cos h y /2 - cos x/2 + o tF(t,y)dt
where (Probs. 6 and 7, Seco116)
U(x,t) = 1 - x - 2 ~
L..
(t + l)-N (2n - l)1tX
cos-
n=1 N 2
4. Determine the integral transformation (a) with respect to y, (b) with respect to x,
that applies to the problem
U xx(x,y) + Uyix,y) + 2Ux(x,y) = O (O< x < 1, O < y < 1),
U(O,y)= G(y), U(l,y) = O, Uy(x,O)= O, Uy(x,l) = H(x),
and write the transformed problem in each case.
Ans. (a) uC<x,n)= f~ U(x, y) cos n1ty dy (n = O, 1,2, . . . );
u; + 2u~ - n21t2uc = (_l)n+1 H(x), uC<O,n)= gc(n), uc(l,n) = O.
(b) u(ln,y) = f~ U(x,y)~ sinn1txdx (ln = -1 - n21[2,
n = 1,2,...);
u" - (1 + n21t2)u= -n1tG(y), u'(ln'O)= O,u'(ln,l) = h(ln)'
5. The line y = O is a singularity of the equation
Vxx(x,y)+ y2Vyix,y) + y~(x,y) + F(x) = O.
If V is to be bounded over the rectangle O < x < 1t, O < Y < b, solve that partial
differential equation under the conditions
VAO,y)= V(1t,y)= V(x,b) = O.
n m
Ans. V(x,y) = fx io
'
F(r)dr dt - -
2 <X>
f(l
L -+ -b
1t n= 1 m
) y
( }
cos mx,
6. Note why the Fourier sine transform (Sec. 110) on the half line x > O applies to
this problem on transverse displacements of a stretched string with prescribed initial
velocity (Prob. 5, Seco44):
Y,,(x,t) = a2 Y",,(x,t) (x > O,t > O)
Y(x,O) = O, Y,(x,O) = G(x), Y(O,t)= O.
Use property (4), Seco110, to show that
1 "+a'
i
Y(x,t) = 2a ',,-a" G(r)dr
and verify that solution of the problem.
Without solving the following problems for V(x,y), determine procedures that
are adapted to their solution.
7. V",,(x,y)+ Vyy(x,y)- V(x,y)= O, (x> ,O < Y < 1),
V(O,y)= 1, Vy(x,O)= O, Vy(x,l)= - V(x,l) + e-".
Ans. Fourier sine transformation with respect to x (Sec. 110); the Sturm-
Liouvilletransformationin Probo4, Seco116,with respectto y, where h = 1.
8. V",,(x,y)+ X-l V,,(x,y)- Vyy(x,y)= Q(x,y) (O< x < 1,y > O),
V(l,y) = O, V(x,O)= O, Vy(x,O)
= O.
Ans. The finiteHankel transformationin Probo12,Seco116;the Laplacetrans-
formation with respect to y.
11
Finite Fourier Transforms
sine transformation
(1) (n = 1, 2, .. .).
Sn{F} = J: F(x) sin nx dx = f.(n)
The basicoperationalproperty(12),Seco111,becomes
(2) Sn{F"(x)} = -n2f.(n) + n[F(O) - (-l)nF(n)]
when F is of class C" (O~ x ~ n), or even if F and F' are continuous while F"
is sectionally continuous over the interval. lt follows that
(3) Sn{F(4)(X)}= n4f.(n) - n3[F(0)- (-l)nF(n)]
+n[F"(O) - (-l)nF"(n)]
if F is of class C(4). Formulas for transforms ofhigher-ordered iterates ofthe
form F" can be written easily.
Transforms of many elementary functions can be found by evaluating
the integral Sn{F}. Some can be found readily from property (2); as examples,
when F(x) ==1, or when F is x, x(n - x), or x, that formula gives these
transforms :
(4)
(5)
Sn -1 {f~~)} = J: ; (n - r)F(r) dr - I: (x - r)F(r) dr.
For Sn the series representation (1), Seco 112, of the inverse Sturm-
Liouville transformation becomes the Fourier sine series representation of
F(x):
2 DO
Ij
~
350 SECo 120] OPERATIONAL MATHEMATICS
because the last two integrands are periodic functions of r with period 2n.
The sum of those integrands is an even function, so
(3)
(5)
fs(n) = f F(x) sin nx dx = - n fa" Fo(x) cos nx dx.
If Xo is a constant, then
~f.(n)
n
sin nxo =_
2
1
IIt
-It
Fo(x)[sin n(x - xo) - sin n(x + xo)]dx,
and this kernel-product property folIows easily:
2 x+xO
- f.(n) sin nxo = S" .
(6)
n {f x-~
Fl(r) dr
}
The corresponding convolution of two sectionalIy continuous functions
F and H is (Sec. 109)
lt X+Y
(7)
XFH(X) = i
o H(y) fx-y F1(r)drdy,
I
2f.(n) = -It Fl(x)sinnxdx I
= -;; -It F;(x)cosnxdx.
Thus a kernel-product property with weightfunction w = 2n can be obtained,
namely,
(10)
352 SECo 120] OPERATIONAL MATHEMATICS
PROBLEMS
1. Use properties of S. to obtain these transforms:
6. Derivethe inversetransform
- ifO ~ x ~ e < n
.
S -1 nsinne
{ n2 }
= (n e)x
.{ (n - x)e if e ~ x ~ n,
(a) with the aid of formula (3), Seco114, when T is S.;
(b) with the aid ofthe kernel-product property (6),Seco120,whenf.(n) = n-l.
7. If the real numbers al' a2, . . . are such that the series al + a2 + ... is absolutely
convergent, prove that
(n = 1,2,...).
FINITE FOURIER TRANSFORMS [SECo 120 353
= 1m [Iog (1 + eiZ)],
and thus derive the inverse transform
2 sin x
S -1 -e
( -1)n+ 1 -ny = - arctan if y> O.
n { n } n ( eY+cosx )
showthat
n
Sn{Y(x)} = n4 + 4c"
12. If Y1is the odd periodic extension of the function Y in Probo 11, with period 2n,
find the solutionof the problem .
in the form
" x+'
Z(x) = i i
o F(t) X.-/ Y(r) dr dt.
13. Express the convolution integral (7), Seco120, in terms of the functions H and F
by the formula
I i
+ " H(y) 2"-X-Y F(r)drdy -
"-X o i i
x
o
H(Y) X-Y
o
F(r)drdy
15. Establish the following relation between the Fourier sine transformation on the
interval (O,b)introduced in Probo 1, Seco116, and OUTtransformation Snon the interval
(O,n):
b . nnx b b
Jo F(x) SIDTdx = Sn{ ( )} . ,/;rx
I
FINITE FOURIER TRANSFORMS [SECo 121 355
cxr 1
Y(x) = Jo J, F(r)drdt + 2xfc(O)(x- X)2 + C.
The constant C is determined by the value assigned to y.(O). We can simplify
the iterated integral to write
fc(n) X 1
(7) 7 = c" {Ix (x - r)F(r)dr + 2xJ.(O)(x- X)2 + A}
L.
356 SECo 122] OPERATIONAL MATHEMATICS
(12)
Cn -1{2fc(n)hc(n)} = s: [F2(x - y) + F2(x + y)]H(y) dy,
where F and H are sectionally continuous. The integral here will be written
in terms of H and the function F in Seco 123.
(3)
F(x,O) = -~log [2(1 - cos x)] = -~ log (2 sin~),
and our results suggest that
ifn =O
(4)
if n = 1, 2, . . . .
The improper integral Cn{F(x,O)}can be evaluated with the aid ofthe theory
of residues1 to prove that formula (4) is correct.
(3)
Sn{H(x)} = -nCn{J: H(r)dr},
(4)
Cn{H(X) - ~hc(O)} = nSn{f: H(r)dr- ;hc(O)},
when H is sectionally continuous.
Let F1 and F2 be the odd and even periodic extensions, with period 2n,
of a function F that is sectionally continuous on the interval (O,n), and let
k be real and constant. Then by the method used to derive property (1),
Seco120, we find that
(5) 2f.(n) sin nk = Cn{Fl(X + k) - Fl(X - k)},
(6) 2.fc(n)sin nk = Sn{F2(x - k) - F2(x + k)}.
I Cf. seco 73 of the author's "Complex Variables and Applications," 2d ed., 1960.
358 SECo 123] OPERATIONAl MATHEMATICS
(7)
P * Q(x) = f" P(x - r)Q(r) dr
is periodic with period 2n and continuous. Let us call it the faltung integral
of P and Q on the interval (-n,n). We find that
(9)
F2 * H 2(X) = f o [F2(x - r) + F2(x + r)]H(r) dr = ~l l(x),
when O< x < n, where 1are theseintegrals involving F and H:
(lO)
11 = s: F(r)H(x - r) dr, 12 = f F(r)H(r - x) dr,
(12)
Fl * H1(x) = I: [Fl(X - r) - Fl(X + r)]H(r)dr
= 11(X) - 12(x) - 13(x) + 14(x),
(13)
Fl * H2(x) = I: [F1(x.- r) + Fl(X + r)]H(r)dr
Then froID the product properties (5) and (6) we find these joint convolution
properties:
(14) -2f.(n)hs(n) = Cn{F * H(x)},
(15) 2f.(n)hc(n) = Sn{F * H2(x)}.
PROBLEMS
e sinh en
(d) Cn{coshe(n - x)} = n2 + e2
if e#: O.
then -1 n3 n 2 1 3 1 4
Cn
{yc(n)} = 45 - 6'X + 6X - 24nX .
3. If a #: Oand b #: O,derive the transforms
(a2 - b2)(_1)n cosh bx cosh ax
a =C -
( ) (n2+ a2)(n2+ b2) n { b sinhbn a sinhan}
3' h2 C -I
(-1)"
(b)
2a SIn an n { (n2 + a2)2 }
= (an cosh an + sinh an) cosh ax - ax sinh an sinh ax.
4. We noted in Probo 9, Seco120, that when y> O,
n sin x
-e-ny = S 2arctan (n = 1,2,.. .).
n n{ eY-cosx }
Use formula (2), Seco103, and its special case Co{F'(x)} = F(n:)- F(O)to prove that,
when n = O,1,2, . . . , and y > O,
V=o h V=A
4r
Let the planes x = Oand y = Obe kept at potential zero and the plane x =n
at another fixed potential V = A (Fig. 104). Then
(2) V(O,y)= O, V(n,y)= A (y > O),
V(x, O) = O (O< x < n),
(3)
IV(x,y)1 < M
(4) V.(n,y )
= hSn{1} - 2An( -1)" (1 - e -ny)
n
(5) V(x,y) = -
nn=1
L vs(n,y)sin nx.
l - (-l)n - ~( - (-l)n+l ~
S
n
-1
{ n3 } - 2 n x
), Sn-l
{ n } =
n'
e-ny 2 sin x
S-
n
1
{
( - l)n+ 1 -
n } = - arctan
n eY + cos X ,
and
Sn-l{l - ~_l)n e-ny} = a(x,y),
where
2 sin x
(6) a(x,y) = - arctan -:-- (y ~ O).
n SInh y
Sn -1
{ l - (-1)"
n3 e
-ny
} = U(
x,y ),
where
" x
f f
X
(7) U(x,y) = - (n - r)a(r,y) dr - (x - r)a(r,y) dr.
n o o
X A sin x
(8) V(x,y) = h -(n
[
2 - x) - U(x,y) ] + -n ( x - 2 arctan ey + cos x ) ,
where U(x,y) is given by Eq. (7) in terms of the function (6). Both inverse
tangent functions in volved here are known solutions of Laplace's equation;
consequently it is not difficult to verify that the function (8) satisfies all
conditions in our boundary value problem [(1) to (3)].
heat. Units are chosen so that the boundary value problem in the temperature
function V(x,y,t) becomes
(O< x < n, y > O,t > O),
(1) V(x,y,O) = O, V x(O,y,t)= -1,
V x(n,y,t) = O, V(x,O,t) = O;
also IVI < Mt throughout the slab, for some constant M.
The presence of the differential form V xx together with prescribed
values of V x at x = O and x = n indicates that the finite Fourier cosine
transformation, with respect to x, can be used here. The form VI and a
prescribed value of V at t = O indicate the Laplace transformation with
respect to t. Let us begin with the cosine transformation.
The transform with respect to x,
(2) W(n,y,t) = Cn{V(x,y,t)} (n = O, 1,2, . . .),
PROBLEMS
1 . A steady-state temperature function V satisfies the conditions
Vxx(x,y)+ y,y(x,y) = O (O < x < n, y > O),
V(O,y)= O, V(n,y)= A (y > O),
V(x,O)= B (O< x < n);
3. When F(x) = A sin x in Probo 2, note the values of u.(n,y) to show that
U(x,y)= Ae-Y sinx. Verifythat solutiono
4. Solvethis problemfor V whenO~ x ~ 1land y > O:
Vxx(x,y)+ Y,y(x,y) = G(x) (O< x < n, y > O),
V(O,y)= V(n,y) = V(x,+O) = O,
FINITE FOURIER TRANSFORMS [SECo 125 365
o
(x - r)G(r) dr
X
i
- -11: o (n - r)G(r) dr.
YI v=o (1T',):)
=0 V=lo
Fig.105
VD
O - "
U
21t o
(b) When F(x) = A, note the valuesof un,y)and deducethat U(x,y)= A.
(e) When F(x) = A cos x, note the values of un,y); thus show and then verify that
U(x,y) = Ae-' cos x.
7. The facesx = Oand x = n of a semi-infinite slab O ;;;;x ;;;;n, y ;;; O, are insulated
and the base y = O is kept at temperature V = O. A steady source of heat KQ(x) with
mean value zero, J~Q(x)dx = O, is distributed throughout the slab; thus Vxx+ Yy, +
Q = O,where V(x,y) is the steady-state temperature. If Vis bounded, derive the formulas
2 00 1
V(x,y) = F(x) L ,qn)e-n, cosnx
- -n n=1n
where F(x) = f;(X - r)Q(r)dr and U is the temperature function in Prob. 6 corres-
ponding to this function F.
8. Ir V is a bounded harmonic function in the semicircular region r < a, O < () < n,
whose values are prescribed as F() on the semicircIe and as zero on the bounding
diameter (Fig. 106),then
r2v,.,(r,() + rf;(r,() + V8(r,()= O, lV(r,()1 < M (r < a, O < () < n),
V(r,O)= V(r,n) = O, V(a - O, () = F(().
x Fig.106
10. The transverse displacement Y(x,t) in a stretched string with a steady transverse
force distributed along it satisfies the conditions
Y,,(x,t) = a2y':x(x,t) + F(x) whenO< x < n and t > O,
Y(O,t)= Y(n,t) = Y(x,O)= Y,(x,O)= O.
Find the solution, and verify it, in the form
1
Y(x,t) = 2a2[2G(x) - G1(x + at) - G1(x - at)]
where G1 is the odd periodic extension, with period 2n, of the function
x " x
G(x) = -
n o I
(n - r)F(r)dr -
o
(x - r)F(r)dr. I
I Cf. seco 101 of the author's "Complex Variables and Applications," 2d ed., 1960.
,
11. The ends of a stretched string are fixed at the origin and the point (n,O). The
string is initially at rest along the horizontal x axis, then it drops under its own weight.
Thus the vertical displacements Y(x,t) satisfy the equation Yr,= a2Yxx + g, where g is
the acceleration of gravity. If Q(x) denotes the odd periodic extension of the function
tx(n - x), where O < x < n, with period 2n, derive the formula
12. If a steady transverse load is distributed along a beam, the transverse displace-
ments Y(x,t) satisfy an equation
82y 84Y
af = -a2 8x4 + F(x).
If the ends x = Oand x = n are hinged so that Yand Yxxvanish there, and if the initial
displacement and velocity are zero, derive the formula
1 2 > f,(n) .
Y(x,t) = zG(x) - ~
a
L ~ cos n2at SInnx,
na 0=1 n
where G(4)(X)= F(x) and G(x) = G"(x) = Oat x = Oand at x = n.
13. Note that for the elementary problem
(O < x < n, t > O),
where
F(t,t) = f f(r) dr.
14. Obtain the solution of the diffusion problem
U,(x,t) = Uxx(x,t) - h(t)U(x,t) + A (O< x < n, t > ( .,
U(O,t)= U(n,t) = U(x,O)= O,
368 SECo 126] OPERATIONAL MATHEMATICS
u, = UJCJC
+ Uyy + A (x > O, O < Y < 11:,t > O),
i r
(5) -
m
f.(m)hs(m) = Sm
{fo H(y)
fx-y F3(r)dr dy} .
Table 2 summarizes properties of Sm and lists transforms of some
functions.
Table 2 Modified sine transforms' !,(n - !)
J.(m) (m = n - t,n = 1,2,...) F(x) (O < x < n)
F(x)
s: F(x) sin mx dx = Sm{F}
2 00
6 I
1
2m f.(m)
2
r O
"
'F(r)dr+ x
x+y
rx F(r)dr
7 -mf.(m)h.{m) f O
H(y)
fx-y
F3(r) dr dy
8 -1
m
cosme O ir O<x<e
9 I --m (O<c<n)
{1 ir c<x<n
1
10 I _(_1)"+1 x
m2
sin me X ir O~x~e
11 I (O < e < n)
m { ir e~x~n
e
2
12 I
m3
6
13 I -( -1)"
m4
14 I
sinhex
(-lr+1
m2 + e2 coshen
15 m - ee""(-I)"
I
m2 + e2
k(-lr+1 sin kx
16 (k,p :!:t.:!:f,...)
m2 -p cos kn
1 For details see Secs. 115 and 126 and the problerns rollowing Seco 116.
369
370 SECo 127] OPERATIONAL MATHEMATlCS
(6) sin qn
iy
o
F(x) sin qnx dx =h fl-yl e-hx
fl-yx ehrF(l - r) dr cos qnx dx
FINITE FOURIER TRANSFORMS [SECo 127 371
R(z) = f e-hrF(r) dr
if z < 1, R(z) = O if z > 1.
(8)
in terms of this convolution of F and G:
(9)
XFG(x) = f G(y)[F2(x + y) + F2(x - y)] dy
- 2hG(1 - x) * F(l - x) * e-hx.
Here the asterisk denotes the convolution integral used with Laplace trans-
forms (Sec. 16); thus
(11)
fc(qn) cos qn = Ch{ F(l - x) - he-hx s: ehrF(l - r) dr} .
From property (3) we find in the usual way that
(12)
J;(q ) (1 1 r
~J = Ch{ Jo ( 1 + h - r) F(r)dr + Jo (r - x)F(r)dr} .
These transforms can be found in various ways:
(13)
(14)
1Another derivation of the property can be found in the author's paper, Generalized Finite
Fourier Cosine Transforms, Micho Matho Jour., vol. 3, pp. 85-94,1955-56.
372 SECo 128] OPERATIONAL MATHEMATICS
A RELATED TRANSFORMATION
Since the above kernel cos qnx is an even function of X, the transformation
(15)
Ch{F(x)} = fl F(x)cosqnxdx =!c(qn)
of functions defined on the interval (-1,1), where qn (n = 1,2,...) are again
the positive roots of Eq. (2), is expressed in terms of Ch by the equation
(16)
It follows from property (3) that, when F is of class C" on the interval
-1 ;;;;x ;;;; 1, Ch resolves F"(x) in terms of the boundary values hF( -1) -
F'( -1) and hF(l) + F'(l):
(7)
Sk{F(l - x) - ke-kxI: rF(l - r)dr} = sinPn f F(x)cos p,.xdx.
Let FI be the odd periodic extension of F with period 2:
(12) XFG(x) =
f o
G(y)
f
x-y
Fl(r) dr dy + 2G(1 - x) * F(l - x) * e-kx,
where the final term, the iterated convolution used with Laplace transforms,
is given by formula (10), Seco 127.
374 SECo 128] OPERATIONAL MATHEMATICS
(13)
(14) Pn .
Sk{sin rx} = r SInPn
Sk{e-kx} = k2 + p/' r2 - Pn - .
PROBLEMS
1. Use the transformations Sm(Sec. 126) to solve for Y(x) when
d4y
dX4 - y = 2x, Y(O)= Y"(O)= Y'(n) = Y"'(n) = O.
2 ex>(-1)'
L1 my- e-my sin mx
V(x,y)= x + -n ,= (m = n - !)
2 x
=x - -
I
n n-x
sin r/2 dr.
arctan --;--
( SInh Y/2)
3. A harmonic function V(r,O)in the semicircular region r < 1, O < O< n, satisfies
these conditions :
V(r,O)= 1 - ~
n arctan 20 1-r'
sin0/2
4. Ir V(r,O)satisfies the conditions stated in Probo 3 in the part of the upper half plane
exterior to the semicircIe, the unbounded region r > 1, O < O< n, show that
V(r,O)= 1 - ~
n arctan 20 r-1
sin(0/2) (r > 1, O < O < n).
I
I
J
FINITE FOURIER TRANSFORMS [SECo 128 375
1 + -l - e if O~ x ~ e,
(a ) Ch- 1 --
cos qnc - h
{ q/ } 1 .
1 f
{ +-x I c~x~l;
if O < x < c,
ifc<x<1.
Y(x,t) = B 1 +
[ !h - x - 2h f
n=1
cos qnat) C~S}qnX) '
qn (h + sm qn) J
where ql' q2'... are the positive roots of the equation tan qn = h/qn' Also,use trans-
form (a) in Probo 5 to show that, when at ~ 1,
Y(x,t) = B(at - x) if x ~ at, Y(x,t) = O if x ~ ato
7. Use the transformation Ch in Seco127 to solve the following problem for steady-
state temperatures in a semi-infinite wall with heat transfer at one face into a medium
at uniform temperature A :
vxx(x,y) + Y,y(x,y) = O (O< x < 1, Y > O),
Vx(O,y)
= O, - Vx(l,y)= h[V(I,y)- A] (y > O),
V(x,O)= O, Y,(x,oo)= O (O< x < 1),
whereh > O. Obtain the solution in the form
V(x,y)= A 1 - 2h2
[
f cosqn~o~qnX exp(-;qnY>l,J
n= 1 h + sm qn qn
where tan qn = h/qn and qn > O.
8. Let the temperatures U(x,t) in a slab with surface heat transfer at one face, and
with heat generated within the slab, satisfy the conditions
U,(x,t) = Uxx(x,t) + F(x) (O< x < 1, t > O),
U(O,t) = O, - Ux(I,t) = bU(I,t), U(x,O) = O,
where b > O. Derive the formula
U(x,t) =
(
1 -~
1 + b) o f l rF(r)dr +
Ix
I (x - r)F(r)dr
where p, P2'... are the positive roots of the equation tan P. = - pJb and
T{F'(x)} = Fc1>]':."
- f" dx = ..lf(..l)+
Fc1>' c1>(n,..l)[F(n) - F(-n)]
if I>and ..l satisfy the homogeneous conditions
(1) I>'(x) = - ..lc1>(x), I>(n,..l)= I>(-n).
This last equation is a periodic boundary condition on I>.The non trivial
solutions of problem (1) are I>(..l,x) = e-inx and ..l= in where n = O,Il,
I 2, . . .. Thus T is the finite Fourier exponential transformation
That set is the basis for the exponential form of the Fourier series on that
intervaJ.l In terms offe(n) that series representation of F(x) can be written as
1 n=m
(6)
F(x) = 2n l~ n];.m fe(n)einx
( - n < x < n), .
valid if F and F' are sectionally continuous over the interval and F is defined
as its mean value from the right and left at each point of discontinuity. Note
that the sum in Eq. (4) ineludes the term corresponding to n = O. The
equation is a formula for En-l{fe(n)}.
In case the infinite series of terms lfe(n)l, where n runs from - 00 to 00,
converges, then
(7) (-n<x<n);
that is, the sum of the series here exists, and so it is the same as the principal
value used in formula (6). In fact, whenever the series of terms lfe(n)1con-
verges, then series (7) converges uniformly to a continuous sum F(x) whose
transform is fe(n).
In terms of Fourier sine and cosine transforms
+ s: [F(-x) - F(x)]sin nx dx
where n = O,:t 1, :t 2,. . .. In particular, when F is an evensectionallycon-
tinuous function on the interval (-n,n), we have a useful formula for
obtaining exponential transforms from cosine tran~orms:
(8) if F( -x) = F(x) (n = O,:t 1, :t2,.. .).
(12) if n =FO.
(5) [2"
(-1)nfe(n) = En{F,,(x- 1t)} = Jo F(x - 1t)e-inxdx.
Note that F,,(x - 1t)= F(x + 1t)if -1t < x < O,and F,,(x - 1t)= F(x - 1t)
if O < x < 1t. The corresponding convolution property is
(7)
XFG(x) = f" G(y)F,,(x- y)dy.
In terms of G and F directly, that convolution has the form
"+X "
(8)
XFG(x) = f-" G(y)F(x - y) dy + I"h G(y)F(x - y + 21t)dy
when - 1t < X < O; but when O < x < 1t,
r -n F(x)e-inXdx = En{F(x)}
F(x)
1 m
2 f.(n) (n = O, i: 1, i:2,...) -1'
2n..~ n=L-m f.(n)einx
3 inf.(n)+ (-I)n[F(7f)- F(-n)] F'(x)
4 J~(n - m) (m = i: 1, i:2,...) eimxF(x)
5 e-in'f.(n) F.(x -
y) (Sec. 130)
6 f.(n)g.(n)
f. G(y)F.(x- y)dy
7 o ir n,p O,f.(O)= 2n 1
8 o ir n,p i: l,f.(i: 1)= n cos x
9 o ir n,p i:1,f.(1) = -ni,f.(-I) = ni sin x
(-I)n
10 x
21ti - ir n,p 0,f.(0) = o
X+ n ir -n<x<O
11 2ni
n ir n,p 0,f.(0) = o { x-n ir o < x < n
(-1). 2 eCX
12 :- (e ,p O, - 1, - 2,.. .)
e - In 2 sinh en
e<.ec;" ir - n < x < O
13 2 sinhen (e2,p O,-1, - 2, . . .)
e - in { e-".e'" ir O< x < n
14
(-Ir
4n~n ir n,p 0,f.(0)= ~n3
(-I)n cosh ex
15 (e2 ,p O, -1, -2,...)
n2 + e2 2e sinh en
(_I)n+1 cos ex
16 (e,p O, i: 1, i:2,...)
n2 - e2 2e sin en
(_l)n+1 1 .
17 -(2x SIOx + cosx)
~ ir n,p i: I,f.(i: 1) = O 4n
18
1 1 - b2
blnl (-1 < b < 1, b ,p O)
2n 1 + b2 - 2bcos x
380 SECo 130] OPERATIONAL MATHEMATICS
PROBLEMS
1. Ir F(x) is sectionally continuous over the interval (-n,n) andJ.(O) = O,prove that,
when n = :t 1, :t 2, . . . ,
K
f.(n) X X
-::r =
n
En
{f-n (r - x)F(r) dr - _21tf-n rF(r) dr} .
2. Use En with respect to t to find the periodic solution Y(t), with period 2n, of the
equation
Y'(t) - eY(t) + P(t) = O,
where P(t + 2n) = P(t) for all t, and e2 =f.O,-1, - 2, . . . .
'+2K
3. Ir the external force acting on the mass m shown in Fig. 15 (Sec. 29) is periodic,
the unit of time can be chosen so that the displacement X(t) of m satisfies an equation
mX"(t)= -kX(t) + P(t) where P(t + 2n) = P(t) and k > O.
Write Jo= jki;. When Wo =f.1,2,..., and P is sectionally continuous over its
period, obtain these formulas for the periodic displacements with period 2n:
X( t ) -- ~ ~ p.(n)
'-- 2 2e
in' -- f:K P(t + n - . t) COSJot dt
2mnn=_""WO-n 2mwosmwOn
'+2n
4. Let P(x) be a prescribed periodic function with period 2n such that P" is everywhere
continuous. Use En to find the periodic function Y(x) of period 2n that satisfiesthe
integral equation
5. If P(t) and Y(x,t) are both periodic in t with period 211:, derive the solution of the
problem
in the form
6. In Probo 6, Seco85, let the unit of time be chosen so that the temperature V(x,t)
in the semi-infinite solid x ~ Osatisfies the conditions
V(O,t)= A sin t, V(oo,t) = O;
A
= -rlogr ifn = j:l.
2
Thus derive the formula
A r
[ - r9 sin 9 - (2- r log r) cos 9
U(r,9) = 211:1 -
co (-1)"
2 "~2 n2 - 1r"cos n9 ].
Fig.107
I
I'
382 SECo 130] OPERATIONAL MATHEMATICS
when F is sectionally continuous. Thus derive Poisson's integralformula for the har-
monic function in the disk (cf. Probs. 8 and 9, Seco125),
V(r,O)= ~
x (a2- r2)F(IX) L. (r < a, -1t ~ O~ 1t).
21tI -x . . -. ua
12
Exponential Fourier Transforms
L
384 SECo 131] OPERATIONAL MATHEMATICS
functions is
and the right-hand member here has a limit as a -. - 00 and b -. 00, so the
integral on the left has the same limit, and it followsthat
Since the inner integral here is.fe(a),the formula represents the inverse ofthe
exponential Fourier transformation:
1
f
(J o
-00
.fe(a)eiu:da = - E-A.fe(a)}
2 1t
(-00 < x < (0).
386 SECo 132] OPERATIONAL MATHEMATICS
When x = O, the integral E-Afe(a)} does not exist because (a2 + 1)-1 is
integrable from - 00 to 00 while a(a2 + 1)-1 is not. But
- i fJ a - i fJ a
- f~da
21t - fJa + 1
=O and lim -
fJ-+00 21t f
- fJ a
--y--
+1
da = O.
'
therefore
ti
lim _21 f fJ
fe(a)da = 2 _1 f
oo
~
da
1
1
=_
2
= F(O).
fJ-+oo 1t -fJ n -ooa +
No two functions F and G that satisfy the conditions on F in Theorem 2
can have identical transforms. For if F and G are absolutely integrable from
-
- 00 to 00, then F - G is also, since IF GI ;;;IFI + IGI. Thus we see that
the function F - G satisfies all those conditions. But Ea{F - G} = O if F
and G ha ve identical transforms and, according to the inversion formula
(2), F(x) - G(x) = Ofor all X. In that cIass of functions, therefore, there is a
one-to-one correspondence between functions and their transforms.
Theorem 2 gives conditions on F that are sufficient for the validity of
inversion formula (2). We now establish sufficient conditions on the func-
tion fe(a) instead, conditions that also establish fe as a transform and as an
inverse transform; but one of our conditions is apt to be difficult to apply.
-<X>
f(a)eixda = -E_xU(a)}
1
27t
But the right-hand member here representsf(a) for all real a, according to the
Fourier integral formula (1), sincefsatisfies the conditions ofvalidity ofthat
formula. Thus
f(a) = f~oo<1>(
-x)e-ix dx = E{<I>(
-x)}.
Equation (5) shows that 4>(-x) is represented by the inversion formula (7).
To illustrate Theorem 3, let us find the inverse transform of
(11)
E~{exp(_X2)} = ~ exp ( - :).
Then from properties (3) and (10) we find these transforms:
(12)
E~{exp ( - C2X2)} = ; exp ( - ::2) (e real, e =1=
O),
PROBLEMS
1. If F(x) = 1 when -b < x < b and F(x) = Oelsewhere, show that
~sin ba
1.(11.)= 11. if 11."# O, and 1.(0) = lb.
2. In tenns of our unit step function So, where So(x)= Oif x < Oand So(x)= 1 if
x > O,we found in Seco132that
1 l-ia.
E{e-XSo(x)}= 1 + ia = 1 + 11.2
390 SECo 133] OPERATIONAL MATHEMATICS
ife > O;
1 .
(b) E.{xe-XSo(x)} = (1 + io,? ,
1
(e) E.{ e"So(- x)} = 1 - ia'
3. Use transform(10),Seco132,of exp(-Ixl) to prove that
2e
(a) E.{exp(-elxl)} =--r-2
e +a ife > O;
2 e-ic. if e is real;
(b) E.{exp (-Ix - el)} = 1 + a2
-4ia
(e) E.{xexp(-Ixl)} = (1 + (2)2'
4. From transform(9),Seco132,of (1 + X2)-1 deducethat
5. Apply Theorem 3 to the function f(a) = exp (-(2) to establish the transform
E.{xF'(x)} = - ;a[a/.(a)].
(b) Apply E. formally to find a solution Y, for which E.{ Y} exists, ofthe ditTerential
equation
show that F"(x) = 2a2[1 - 2F(x)] when alxl < 1t/2and hence that
showthat
2
E{F(x)} = -;-(1
IIX - cos CIX).
(1)
XFG(x) = t: F(t)G(x - t) dt = F * G(x)
(-00 < x < (0).
Note that the eonvolution or faltung operation here, F * G(x), is not the one
used with Laplaee transforms sinee the integral has limits - 00 and 00 rather
than O and X. Both eonvolutions are the same in case F(x) = G(x) = O
whenever x < O.
Before establishing eonditions on F and G under whieh Ea{XFG} =
Je(rx)ge(rx),
we need some properties of the operation F * G and information
on the nature of the function X FG(X).
We assume in this seetion that our functions F(x) and G(x) are sectionally
continuous on each bounded interval of the x axis, and that they are bounded
and absolutelyintegrableJrom - 00 to oo.
Sineea eonstant M exists such that lG(x - t)1< M, then IF(t)G(x - t)1<
MIF(t)l. But MIF(t)1is independent of x and integrable from - 00 to oo.
"
Hence the convolution integral (1) converges absolutely and uniforrnlyfor all
real x. Moreover,
(2)
XFG(x) = f: G(y)F(x - y) dy = XGF(x).
(4)
le(x) = fe F(t)G(x - t) dt
can be written as a sum of integrals from -e to tI' tI to t2, etc., whose in-
tegrands are continuous in x and t. ConsequentIy le is continuous when
- e ~ x ~ e.
Next let G be the unit step function So(x - k) where - 2e < k < 2e.
Then G(x - t) = So(x - t - k) = 1 ift < x - k, and it vanishesift > x-k.
If - e ~ t ~ e, then t ~ x - k when e ~ x - k, and so
if x ~ k + e.
le(x) = fe F(t)So(X- t - k)dt = fe F(t)dt
Since So(X - t - k) = Owhen t > - e> x - k, then
if x ~ k-e.
Thus IAx) has constant values on each of the half lines x ~ k + e and
x ~ k-e. On the remaining interval of the x axis, k - e < x < k + e, we
note that both t and x - k lie on the interval (- e,e),and thus
e "-k
f
le(x) = -e F(t)So(x- k - t)dt = -e F(t)dt f
(k - e ~ x ~ k + e).
This continuous function of x assumes the above constant valuesat the end
-
points k e and k + e of the interval. Therefore le is continuous for all x
when G(x) = So(x - k).
EXPONENTIAL FOURIER TRANSFORMS [SECo 135 393
(5)
X FG(X) = fe F(t)G(x - t) dt + Re(x)
and the absolute value of the remainder RAx) can be made arbitrarily small,
uniformly for all x, by making e sufficiently large. Let x be fixed and its
increments .1x be such that l.1xl < b, where b is some positive constant, and
select e such that e > Ixl + b. Then
(-oo<x<oo)
F * G(x) = J:CX}F(t)G(x - t) dt
converges absolutely and uniformly to afunetion XFG(x) that is bounded
and eontinuous for all real X. Also, F * G(x) has the eommutative and
distributive properties (2) and (3).
(3)
Q(x) = t: lF(t)IIG(x - t)1dt
( - 00 < x < (0),
and the integral here converges uniformly with respect to x. With an in-
version of order of integration (Sec. 15),we can write
rb Q(x) dx =
Jo f-00
oo IF(t)1 rb IG(x
Jo
- t)1dx dt.
or
a~i~ooLO Q(x) dx, f: 00 Q(x) dx,
exists, and hence the improper integral
(4)
f~ 00 Q(x) dx = f~ f~ 00 00 IF(t)IIG(x - t)1dt dx
converges. Since le-iuXFG(x)1 ~ Q(x) for all a, it follows that the Fourier
integral
(5)
EXPONENTIAL FOURIER TRANSFORMS [SECo 135 395
(7)
fe f:", F(t)G(x - t)e-i"X dx dt.
Here the integral with respect to x converges uniformly in t when Itl < e
because F is bounded, IF(t)1 < M, and the remainderfrom x = b to 00
satisfies the conditions
~ M L~e IG(Y)Idy.
f:", q(c,x)dx
where q is the function
(9)
fe(tl)ge(tl)= !~~ f~", q(c,x) dx.
We note that q(c,x)-+ e-i"xXFG(x) as e -+ 00, and that limit is uniform with
respect to x because the improper integral XFGconverges uniformly for all x.
396 SECo 136] OPERATIONAL MATHEMATICS
Since
lim
c-+oo foo
-00
q(c,x)dx=
f oo e-iaxXFG(x)dx,
-00
(10)
f [e-iaxXFG(X) - q(c,x)]dx + foo e-iaxXFG(x)dx
-
fa
-00
q(c,x)dx+ oo e-iaxXFG(x)dx - ooq(c,x)dx.
Jb Jb
The improper integrals here converge absolutely; those involving q con-
verge uniformly with respect to c. Thus their absolute values can be made
small, independendy of e, by taking a large and negative and b large and
positive. Then by taking e large and positive, the absolute value of the first
integral can be made small beca use the integrand tends to zero uniformly
in x as e ~ oo. This completesthe proof of Theorem 5 since it followsthat
Loo:(X - t, t)dt.
Example 1 Let F(x) be such that F' is continuous and F itself is absolutely
integrable, from - 00 to 00, and such that F(:t 00) = O. To find a
solution of the boundary value problem
Wx(x,y) + Jv,,(x,y) + 2yW(x,y) = F(x) (- 00 < x < 00,y> O),
(1)
W(x,O) = O, W(:t oo,y) = O,
398 SECo 137] OPERATIONAL MATHEMATICS
(4)
W(x y) = exp ( - y2) I: F(x - y + t) exp (t2) dt.
It is not difficultto verifythat the function (4)satisfiesall condi-
tions in problem (1).
Example 2 To obtain a formula for the bounded harmonic function
V(x,y) in the halfplane y > Othat assumes values F(x) on the boundary
y = O,we solve this problem:
Vxix,y) + Vyy(x,y)= O (- 00 < x < 00, y > O),
(5)
V(x, +0) = F(x), lV(x, y)1 < M,
where M is some constant and, over the entire x axis, the function F is
bounded and continuous except possibly for a finite number of finite
jumps.
To solve the problem using the transformation Ea with respect to
x, we shall assume some auxiliar y conditions on F, conditions associated
with our method of solution but not necessarily required in order that
the resulting formula for V(x,y) will be valido We assume that F(x) is
such that its transform f..(ex)exists and that the problem will have a
solution V(x,y) whose transform vJex,y)with respect to x exists and is
bounded when y > O for each fixed ex. Further conditions on V are
needed to write Ea{Vxx} = -rx2ve and to justify steps in transforming
problem (5) into the problem
2 d2
(y > O),
(6) - + dy2ve(rx,y)= O
ex vAex,y)
ve(rx,O)= fe(rx), Ive(rx,y)1< N(rx),
where N is independent of y.
EXPONENTIAL FOURIER TRANSFORMS [SECo 137 399
(9) V(x,y) =
F.
: t - x
arctany
Fo 1t 1t
]-'" = -; 2 + 2 = Fo,
( )
i I
(10)
1 '"
i y dt 1 1t
2
V(x,y) = -1t o (t-x )2 +y 2 = -1t( _
X
+ arctan -
Y)
.
PROBLEMS
1. Let F(x) be sectionally continuous over an interval -c < x < c, while F(x) = O
when x < -c and when x > c. Use the exponential Fourier transformation to find
a solution of the differential equation
Y"(x) - Y(x) + 2F(x) = O (-00 < x < (0)
such that Y' is everywhere continuous and Y(i: (0) = Y'( i: (0) = O,in the form
I
q
400 SECo 137] OPERATIONAL MATHEMATlCS
3. Let h(y) be continuous when y ;; O,and let F(x) have a continuous derivative F'
for all x and vanish as x --+i: oo. Derive the solution of the problem
also F(x) --+O and Y(x,t) --+O as x --+i: oo. Transform with respect to x to derive the
formula
Ans. U(x,t)= ~
2.jidZi f-
<X> F(~) exp
<X> [- (x - ~)2 d~
4kt J
O>
1
=
f
j1r - <X>F(x + 2r.jki) exp(- r2)dr.
13
Fourier Transforms on the
Half Line
Fourier sine and cosine transforms of functions defined over the half line
x > O were introduced in Chap. 10 as examples of general integral trans-
forms. They transform the differential form F" in terms of the transform of
F itself and the initial values F(O)and F'(O),respectively. Operational prop-
erties of the two transformations are noted here in greater detail and sum-
marized in Appendix D; some applications are given. AIso, we present a
modified transformation that resolves F" in terms of a boundary value
F'(O) - hF(O).
(ex> O),
(1) Sa{F(x)}= La:> F(x) sin exxdx = f.(ex)
401
402 SECo 138] OPERATIONAL MATHEMATICS
(5)
s: F"(x) sin ax dx = -a2 s: F(x) sin ax dx
+ [F'(x)sinax - aF(x)cosax].
As e -+ 00, the bracketed term tends to (1.F(O);thus the integral on the left
has the limit Ser{F"}when the integral on the right has a limit.
By replacing F by F" in Eq. (5) and integrating by parts, we find that
(6)
under these conditions : F and its derivatives up to F'" are continuous when
x ~ Oand vanish as x -+ 00, F(4)is sectionalIy continuous on each bounded
interval, and J.(ex)exists; similarly for transforms of further iterates of the
operator d2/dx2.
Sufficient conditions for the inversion formula (3), the Fourier sine
integral formula, are that F be absolutely integrable from O to 00 and that
F' be sectionalIy continuous on each bounded interval. Then the formula
represents F(x) where F is continuous, and the mean value of F(x + O)and
F(x - O)where F has a jump.
FOURIER TRANSFORMS ON THE HALF LlNE [SECo 138 403
(7)
(9) if k > O.
(6) if k > O,
fc(ak) = ~Ca{F(I)}
(7) f;(a) = Ca{ -x2F(x)},
(8) lim fc(a) = O.
(x ~ O)
F(x) = {") G(r)dr
satisfies our conditions since F'(x) = - G(x) and F( (0) = O. Thus property
(1) can be written
(3)
The existence of the trbnsforms!s and h for all positive a, and the
natural agreement that fs(-;a) = - fs(a) and fc(-a) = fc(a), enables us to
write the relations
fs(a + k) + fs(a - k) = Scz{2F(x)cos kx},
(7)
fs(a + k) - fs(a - k) = Ccz{2F(x)sin kx},
fc(a + k) + fc(a - k) = Ccz{2F(x)cos kx},
(8)
fc(a - k) - fc(a + k) = Scz{2F(x) sin kx}.
where F2(x) = F(lxl) and G2(x) = G(lxl) when - 00 < x < oo. According to
Theorem 5, Seco 135, the product
(1)
2fc(a)gc(a) = Ccz{L'"F(r)[G(x + r) + G(lx - rl)]dr}.
Under the same conditions on F and G, and in the same manner, we
can establish these joint properties of sine and cosine transforms:
(2)
2fs(a)g.(a) = Ccz{L'" F(r)[G(x + r) - Gt(x - r)]dr} ,
FOURIER TRANSFORMS ON THE HALF LINE [SECo 142 407
(3)
2.fs(IX)gc(lX)
= S..{foa>F(r)[G(lx - rl) - G(X + r)] dr}
(x ~ O)
H(x) = la> G(t) dt
2f.(IX)gs(lX)= 21Xf.(IX)hc(lX)
(2)
C..L ~ A = (~ - Si IXC)sin IXC- Ci IXCcos IXC,
--
408 SECo 143] OPERATIONAL MATHEMATICS
(3)
{' H(x) dx = o.
Let us assume also that whenever x ~ O, the integral
(4)
P(x) = {>OI H(t) dt dr
exists and that P has a cosine transformo Since P"(x) = - H(x) and P'(O) = O,
the cosine transforms of P and H satisfy the relation
(5) pc(a) = h.(a).
y
...-;
--
Hb:)
O V=F(x) x Fig.108
FOURIER TRANSFORMS ON THE HALF LINE [SECo 143 409
e-ay = C ~ y
a { 7t X2 + y2 }
and, according to the convolution property (1), Seco 141, it follows fram
Eq. (7) that
ex>
1 1
(8) V(x,y) = P(x) + -
y
7t Of [F(r)- P(r)J
[ (x+r )
2
+y
2+
(x-r
2
) +y ]
2 dr.
(9)
V(x,y) = P(x) = {ex>L H(t) dt dr,
a result that is easily verified as a solution.
In case no source is present, then H(x) = O,and
I
1 Y Y
(10) fex>
[
V(x,y)= ; Jo F(r) (x + rf + y2 + (x - r)2+ y2 dr ] 11
when x ~ Oand y > O. That is an integralformula for the harmonic function
V in the quadrant such that Vx(O,y)= O and V(x,+O) = F(x). In this case
where H(x) = O, the problem can be solved also by using the sine trans-
formation with respect to y.
When H(x) = Oand F(x) = 1,wefindfromformula(10)that V(x,y) = 1,
and this is cIearIy a solution of the problem even thoughfc(x) does not exist.
410 SECo 144] OPERATIONAL MATHEMATICS
ln cas.:: H(x) = D, and .F(x) ~ 1 W'h.::n O <' x <' e W'hil.:: F(x) - O ovhen
(5)
PROBLEMS
7. (a) Transform Laplace's equation Yx" + Y,p= O with respect to x to derive the
integral formula
y y
1 f'"
[
V(x,y) = -;)0 F(r) (r - X)2 + y2 - (r + X)2 + y2
dr
]
for the harmonic function in the quadrant x ~ O, Y > O, such that V(O,y)= O and
V(x,+O) = F(x).
(b) In case F(x) = 1 whenever x> O,then, even thoughf.(cx)does not exist, show
that the formula in part (a) gives the valid solution
V(x,y) = ~ y = 1- ~
1tarctan ~ 1t1m [Log (x + iy).
FOURIER TRANSFORMS ON THE HALF LlNE [SECo 144 413
(e) Verify the solution given by the integral formula in part (a) when
F(x) = 1 if O < x < 1, and F(x) = O if x > 1.
where
U(x,t) = -
i
1t o
u.(a,t)sin (XXda
also Z and its derivatives are to vanish as x -> 00 and y -> oo. Derive the formulas
I
b (">
Z(x,y) = '2Jo (2 + ay) exp [- (1 + y)et)sin ax da }
.
- bx + bx 1+ y .
- X2 + (1 + y)2 y [X2+ (1 + yf)2 I
The first form is helpful in verifying the solution. I
I
I
414 SECo 145] OPERATIONAL MATHEMATlCS
12. In Probo 11 let the conditions on the edge y = O be replaced by the conditions
- b xy
- - x2+ (1+ y)2'
= i oo"
o F(x)<I> (x,A) dx - /
F'(O) -
<1>'(0,..1.)
hF(O)
- h<l>(O)
F(O)
<1>(0,..1.)
I .
(6)
O(a) = arctan h where 0< O(a)< 2'
then 41= Jh2 + a2 sin (ax + O),and 1;,takes the form
(7)
1;,{F(x)} = .Jh2 + 11.2100F(x) sin [ax + 0(11.)]
dx.
In Seco 140 we noted conditions on F under which
aC{F(x)} = -S{F'(x)}.
Under those conditions form (5) for 1;,can be written
(8) 1;,{F(x)} = fT(a) = S{hF(x) - F'(x)}.
Formally then we can invert the sine transformation to write
oo
2
(9) hF(x) - F'(x) = -
f
11: o
fT(a)sin ax da (x > O).
We invert the order of integration here and use the known Laplace trans-
I
form of sin a(x + t) with respect to t, with parameter h, to write this formula
for the inversetransformation T,.-l{JT(a)} :
f
(12) F(x) =~
nJo flX) fT(a) a cos ~~ + a.
h~sinax da.
(x > O).
(2)
Q(x) = F(x) + h {IX)F(r) dr
and, since S,,{-Q'} = aC,,{Q}, we have this alternate representation of
T,.{F} :
(3)
~fT(a)~ik)
a = 2aqia) cos ak + 2hqJa) sin ak
1 The proof, with the aid of Fourier sine and cosine integral formulas, is not difficult even for
a somewhat more general integral formula. See R. V. Churchill, Generalized Fourier Integral
Formulas, Mich. Math. Jour., vol. 2, pp. 133-139, 1953-54.
j
FOURIER TRANSFORMS ON THE HALF UNE [SECo 146 417
(7)
X FG(X) = 100 G(r)PF(x,r) dr
(8)
SimilarIy,
(9)
Since
2 t
e-at - -C
- 1t a{ t2 + X2 } '
(10)
eky
U(x,y) = -1ty f f e-kt o P(r)[ t+x-r
2 (
t
)2 2 (
t+x+r
t
)2J dr dt.
Formula (8) then gives a representation of V(x,y).
The solution represented by Eqs. (8) and (lO) can be simplified and
verified in case F(x) = 1,in whichcase P(x) = h, eventhough the transforms
fT(ex)and p.(ex)do not exist.
14
Hankel Transforms
148 INTRODUCTION
(1)
420
HANKEL TRANSFORMS [SECo 149 421
where n is zero or a positive integer. Tht form arises when the Laplacian
operator V2 is written in terms of cylindrical coordinates; in that case our
variable x here is the radial coordinate usually called r or p (cf. Seco83).
In terms of a self-adjointform we write
1f n2
(2) D2F = x1[xF'(x)' - x-F(x)} (O< x < 1,n = 0,1,2,...),
so in the notation used in Seco 111, p(x) = r(x) = x and q(x) = -n2jx.
1 As noted in Seco 83, basic properties and representations of Jn are developed in the author's
"Fourier Series and Boundary Value Problems." For extensive treatments ofBessel functions
see G. N. Watson, "A Treatise on the Theory of Bessel Functions," 2d ed., 1944, or A. Gray,
G. B. Mathews, and T. M. MacRobert, "A Treatise on Bessel Functions and Their Applications
to Physics," 2d ed., 1952. For a shorter account see F. Bowman, "Introduction to Bessel
Functions," 1958.
422 SECo 149] OPERATIONAL MATHEMATICS
(3)
T{F} = f F(x)X<l>(X,A)dx = f(A),
with weight function p(x) = x, that transforms D2Fin terms of A,j(A), and
the boundary value F(l). According to Eq. (8), Seco111, we can write
Thus if <1>
T {D2F} = f
and A are such that
F(X{ (x<l>')'- : <1>]dx + [x(<I>F' - <1>'
Fm.
n2
(4) - -<1> and <1>(1)) = O,
(x<l>')'
x = AX<l>(X))
then
(5) T{ D2F} = Af(A) - <I>'(l,A)F(l).
when F is of cIass C" (O~ x ~ 1) and /l1, /l2,'" are the positive zeros of
J n(J1.).Numerical values of the first four zeros of J o(J1.)
were given in Seco83.
-
When n = O, D2[1]= O and D2[X2 1] = 4. Thus these particular
transforms foIlow from formula (10):
1
(11) HOj{l} = -J1(J1.j),
/lj
(i = 1,2,...)
(1)
ifj = i,
where /lb /l2,' .. are the positive zeros of Jn(/l).
For a function F on the interval the generalized Fourier series in terms
of the functions of that set is the Fourier-Bessel series
424 SECo 151] OPERATIONAL MATHEMATICS
That series converges to F(x) on the intervalO < x < 1 if F' is sectionally
continuous there and if F is defined as the mean value of its limits from the
right and left at each point where F is discontinuous. Then we can write
00
(2) F(x) =2
j= t
[Jn+t(.uj)r2fH(.uj)Jn(.ujx) (O< x < 1),
(4)
where J n(.uj)= O,J.lj > O,n = O, 1,2,. . . , a result that is valid if F is sectionally
continuous on the interval (0,1).
i,
f
_J
1
(2)
(4) (i = 1, 2, .. .),
where Pi are the nonnegative zeros of the function hJiP) + pJ~(P). The
transformfH(P)is a sequenceof numbers, a sequencethat depends on the
valuesgivento n and h. The basicoperationalproperty(2)now takes the form
(5)
Hnh{F"(X) + ~F'(X) - ::F(X)} = -P/fH<Pi)
+ JiP)[hF(l) + F'(l)],
valid if F is of class C" on the interval O ~ x ~ 1.
Again, for each fixed pair of constants n and h, where n = O,1,2, ...
and h ~ O,the set of functions Jn(PIX),Jn(P2X),,,,,is orthogonal on the
interval (0,1),with weight function x. Using known formulas for norms of
Jn(PiX),we can write
O if i ~ j, where j = 1,2,. . . ,
~ I
(6) = P/ + h2 -
Hnh{Jn(PjX)} n2 2
if i = j,
{ 2P/ [Jn(Pj)]
except that in case n = h = O,then P1 = O and J 0(0) = 1 so
(8)
but if n
F(x) - 2 ;t, Po' +~: - n' [~:gJ],J.(P,X)
= h = O,then P1 = O,and the first term in the summation is the
constant 18(0) = f~ tF(t) dt.
r < x < 1);
(%,y,z)
z
y
------------------
% Fig.109
HANKEL TRANSFORMS [SECo 152 427
then
(6)
,1,.
)
4 ~ ~ usH{Jlj,n,z)
) ',1,.
L.J L.J [Jn+l (Jlj WJ n{Jljr SIO n'l'
U(r,'I"z = -1tn=lj=l
where UsHis given by formula (5) and Jlj are the positive zeros of J n{Jl),which
depend on the value of n.
In case the condition U(l,cjJ,z)= 1 is replaced by
Ur(1,cjJ,z)= -hU(1,cjJ,z) + P(cjJ,z),
PROBLEMS
1. Derive the finite Hankel transfonns (13) and (14), Seco 149, of the functions X'
and Jn(cx).
2. Obtain the finite Hankel transfonns
J.ll-4
(a) HoAx2} = -r-J(J1.j),
J.lj
I
d
428 SECo 152] OPERATIONAL MATHEMATICS
- eJ-l/l(JJ.}.c)[F(e
+ O)- F(e - O)] - eJo(JJ.}.c)[F'(e
+ O)- F'(e - O)]
(j = 1, 2, . . .).
fH(JJ.j) = HOj{F} = s: F(x)xJo(JJ.r)dx
4. Let G(x,e) be Green's function (Sec. 95) for the differential form D2 when n = O,
such that
l h+ n
fo x.+ lJ.<PX)dx = H.h{X.} = yJ.<P) (n = O, 1, 2, . . .).
6. When n = O and h > O and p are the positive zeros of the function hJo<P)+
- pJ 1 (P),and if
pJ'o<P),or hJ o(P)
(i = 1,2,. ..),
fH<P)= HOh{F}= s: F(x)xJo<Px)dx
derive the formula
l l
1 y t 1
p/fH<P) = HOh {Jx f.o yF(t)dtdy + h f.o tF(t)dt } .
HANKEL TRANSFORMS [SECo 152 429
ifi = 2,3,...,
HOO{F"(X)+ ~F'(X)} = -P/Hoo{F} + F'(l)Jo(/J)
= F'(l) ifi = 1, since p = O,
where P, P2, . . . are the nonnegative zeros of J (/J). Show that
Hoo{1} =O if i = 2, 3, . . . , if i = 1;
ifi = 2,3...,
if i = 1.
8. If e> O and Jo(e) t= O, show that the function Y of class C" (O~ x ~ 1) that
satisfies the conditions
1
Y"(x) + -x Y'(x) + e2 Y(x) = x2 - 1 Y(l) = O,
has the transform
9. Solvethe problem
1
Y"(x) + -x Y'(x) - e2Y(x) + F(x) = O, O<x<l; Y(l) = O,
in the form
(j = 1,2, . . .).
fHJ.j)= I: F(t)tJ oJ.t) dt
L
.
430 SECo 152] OPERATIONAL MATHEMATICS ~
/
10. Use the Hankel transformation HOj with respect to r to derive, in only a few steps,
the formula (9), Seco83, for temperatures U(r,t) in a circular cylinder of infinite length,
when
1
U,(r,t) = Urr(r,t) + -Ur(r,t)
r (O ;;i; r < 1, t > O),
U(r,O)= O, U(I,t) = 1.
11. Use results found in Probo 7 to show that the appropriate Hankel transform of
U(r,t), where
1
U,(r,t) = Urr(r,t) + -:Ur(r,t) (O;;i;r < 1, t > O),
U(r,O)= O, Ur(l,t) = A,
has the value
ifi=2,3,o..,
1[P~zJo(fJ) - zJo(fJ)exp(-p/t)]
if i = 1,
4[~ +M4t +~)]
and thus obtain the formula for U(r,t) found in Prob:S, Seco 830
12. In Probo 8, Seco144, an integral representation was found for the temperature
distribution W(r,x) in a semi-infinite circular cylinder whose axis is the positive x axis,
when ~
also, W is to be continuous and bounded when O ;;i;r ;;i; 1 and x > 00 Use a Hankel
transformation to derive this series representation of W:
.., 1 J o(p.jr)
W(r,x) = 2 L --exp(-JljX)
j= Jlj J(p.j)
when O ;;i; r < 1 and x > O,where Jo(Jlj) = O and Jlj > 00
13. Let transverse displacements Z(r,t) in a membrane stretched over a circular frame
satisfy the conditions
1
Ztt = Zrr + -Zr - 2kZ, (O;;i;r < 1, t > O),
r
Ifthe coefficient of damping 2k is such that O < k < Jl where Jl, Jlz," oare the positive
zeros of J o(p.)in order of ncreasing magnitude, derive the formula
~fH(J1.j)
Jl)
f ro !.P 2F(t)dtdP } '
= Hlj { , ~
P { }
and since H1j{r} = Jlj-IJ2(J1.j), according to transform (13), Seco 149, show that
1 r- r3
-J2(J1. j ) = Hl" { - .
Jlj3 ) 8 }
15. If Y = r sin ,p and V2V + Ay = O iriside a half cylinder 0< r < 1, O < ,p < n,
z > O,and if V = Oon the boundaries,then V(r,4>,z)
satisfiesthe conditions
1 1
+ V.Z+ ArslD'I' = O
"~
v" + -v, + -V4>4> (O< r < 1, O < ,p < n, z > O),
r r2
The solution found there can be obtained by using the nonsingular Hankel
transformation
(5)
where
H o{F(r)} = f F(r)rtf(ai,r)dr
(i = 1,2,. ..),
(6)
and ai are the positive zeros of the function tf(a,b).But the transformation
H o applies also to problems in which the differential equation and all
boundary conditions in problem (4) are made nonhomogeneous.
I
i
~ J
HANKEL TRANSFORMS [SECo 154 433
on the half line x > O by A.f(A.). That operational property is the principal
one for applications of such transforms to problems in differential equations.
To construct T so that it has that property, we write
(1)
T{D2F} = LX) [(XF')' - : FJ~dX
(4)
(6) (x >,0).
Y"(x) + !.
X
Y'(x) - x2
n2 Y(x) = - F(x) (x > O)
modified form
1 n2
(8)
(9)
X FG(X)= {ex>F(y)PG(x,y)dy.
Formulas (4) to (9) are presented here to display their complexity. In
case n = O,the results are included in Table 4, but even in that case the
kernel-product and convolution properties are difficult to use. The integer
n can be replaced by v, where v ~ O,in those formulas.2
1 Watson, G. N., op. cit., p. 367.
2 Derivations and conditi~ns of validity of those kemel-product and convolution properties
are inc1uded in J. S. Klein's doctoral dissertation, "Some Results in the Theory of Hankel
Transforms," University of Michigan, 1958. Earlier, J. Delsarte, J. Math. Pures Appl. (9), vol.
17, pp. 213-231,1936, and Acta Math., vol. 69, pp. 259-317,1938, gave convolution properties
associated with Hankel transforms.
-
436 SECo 156]
OPERATlON\l MATHEMATICS
F(x)
f""
o F(x)xJo(ax)dx = Ho.{F}
2
{"" fH(a)aJO(ax)da = Hox{fH}
I
3
F"(x) + 'XF'(x)
4 fH(ka) (k > O)
"" ""t
5 -
f f x
-F(t)dtdy
y Y
6 1tJo(ak)fH(a) (k> O)
r o
F[(X2 + k2
"-.
- 2kx cos O)t]dO
7
fo""yG(y) s: F[(X2 + y2 - 2xy cos O)t]dOdy
1 I
8
a x
I I -ex
9 (e> O) -e
J a2 + e2 x
10
II (e> O)
12 (e> O)
.jX2 + e2
13 1 .j X2 + e2 - e
(e> O) -Iog
2 .j X2 + e2 + e
I ir O< x < e
14 (e> O)
{ O ir x > e
15 (e> O)
~Jo(~)
HANKEL TRANSFORMS [SECo 157 437
{X>G(x)~Jv(ax)dx = g(a;v).
/
G(x) . g(a;v).
and
F(x) = .fi fH(a) = .fi
Those tables list several properties of transforms g(a; v)in which different
values of the index v occur in the same property. Such properties follow
from relations such as Eq. (5), Seco 154, between J~, Jn, and Jn+l' or the
recurrence relation
Let vH(a,z)denote the Hankel ttansform of V with respect to'r, for which
n = O:
PROBLEMS
1. UsetheknownLaplace transformation /
L{ J O(lXt)} = (S2 + 1X2) -1/2 (s > O, IX> O)
to establish the transform H o. {x -1 e -ex} listed in Table 4. Also, differentiate with respect
to the parametere to showthat, if e > Oand m = 1,2,...,
2. Note that our conditions of validity of the inversion formula (6), Seco 154, are
satisfied by the function F(x) = e-ex whene> O.Thus
when x > O.
where G(x)= O
e2 - x2 + 2c210g~ if O< x ~ e,
{ e if x ~ e.
4. Let V(r,z) be bounded in the half space z > O, r ~ O, O ~ f/>< 21t, where r,z,
and f/>are cylindrical coordinates, and satisfy the conditions
1
V2V(r,z)= v" + -v,
r + V.z= O (r > O, z > O),
1
V(r,+O) = 'r2 + e2,1I2'
v" + !v,
r + r12V",,,,+ V.Z+ F(r)sin f/>= O
J
/
in the region r > O,O < cP < n, Z > O,and if V = Oon the boundaries cP = O,cP = n,
and Z = O,derive formally the formula
a:>
a:>1 - e-U
V(r,cP,z)= sincP
f o ex
J1(exz)
f o
F(P)J1(exp)dpda..
V(r,z) -c
-
f
a:>
--
sinh
o cosh z
exz J 1(exC)
ex
( ) dex.
J otXr
15
Legendre and Other Integral
Transforms
i
I
x Fig.110
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SEC. 158 443
then requiring that ~2~ = A~ and that ~ and F be of class C" on the
interval - 1 ~ x ~ 1. Thus
(3) [(1 - X2)<l>'(X,A)]'
= A~(X,A) when - 1< x< 1
and ~ is to be of class C" on the closed interval - 1 ~ x ~ 1.
That singular eigenvalue problem in Legendre's differential equation (3)
has as its set of eigenvalues
(4) A = -n(n + 1) (n = O, 1,2, . . . ).
"
The corresponding eigenfunctions are the Legendre polynomials Pn(x) of
degree n:
(5) - - 1 m (-lY
~(X,A) - Pn(X)- 2
(2n - 2j)! -2]
n ]=0 -:--
J. L (n - 2")1
'.] .(n - J.")Ix"
I
(6)
T,,{F(x)} = f 1 F(x)Pn(x) dx = fL(n)
(n = O, 1, 2, . . . )
(O < e < n)
f1 F(x)dx = O.
Then by solving the differential equation
[(1 - X2)Y'(X)]' = -F(x)
for Yand requiring Yand Y' to be continuous on the interval -1 ~ x ~ 1,
we obtain the property
(3) A(n) =
n(n + 1)
T,.
{fo +-
X
s - 1 5-1
F(t) dt ds +
s C
}
(n = 1, 2, .. . ).
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SECo 159 445
The value of the transform on the right when n = O depends on F and the
choice of the constant C.
Legendre polynomials have the product propertyl
n
(4)
where
Pn(cos f3)Picos O) =
1
-
1t oi Pn(COSv) da
(5) ~v=~f3~O+~f3~O~~
The following convolution property of 1'"has been established with the aid
of that property.2
Let F(x) and G(x) be sectionally continuous functions with Legendre
transforms fL(n) and gL(n). Then
(6)
A(n)gL(n) = T,,{XFG(x)} = 50" XFG(cos O)Pn(cos O) sin O dO
cos v being given by formula (5). The convolution here is not a simple one
to use.
Some useful transforms of particular functions can be found from the
representation
00
of a generating function for Pn. When t is fixed, the series converges uniformly
with respect to x, and it follows that
2tn
(9) T.{(1 - 2tx + t2)-t} = -
n 2n+1
wheren = 0, 1,2, . . .. Let us use the symbol1;n+ 1 {F} to denote the integral
transformation of F that appears here; that is,
(1)
T2n+l{F(x)} = f F(x)P2n+ l(X) dx = JL(2n + 1)
(n = 0, 1, 2, .. . ).
Since T2n{F.} = 0, the Legendre series representation (10), Seco 158, of the
function F1 becomes
00
F1(x) = L
n=O
(2n + 1 + !)T2n+1{F.}P2n+1(X) (-1 < x < 1).
because F1(x) = F(x) when < x < 1. The formula is valid if F' is s~c-
tionally continuous over the interval (0,1).
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SEC. 160 447
(6) 1 {[(1
'1'211+ - x2)F'(x)]') = -(2n - 1)(2n + 2)JL(2n + 1)
+ (2n + 1)P2n(0)F(0).
(7) (n = O, 1, 2, . . . ),
and T211+
1{F2} = O. It follows that '1'211
has the inversion formula
00
(9) F(x) = L
n=O
(4n + 1)JL(2n)P211(x) (O < x < 1),
that resolves the differential form ~2F on the interval (0,1) in terms of
1;n{F} and the initial value F'(O) of F'. AIso, the set of functions P2n is
orthogonal on the interval (O,1).
[
~
V2V = r12 (r2v,)r+ sin () (Sin ()J.-9)o
J
=O (r < 1, O < () < n),
and V -> F(cos ()) as r -> 1; also V and its partial derivatives of first and
second order are to be continuous interior to the sphere. We write x = cos ().
Then
(1) [r2 V,(r,x)]r + [(1 - x2)Vx(r,x)]x= O (r < 1),
(2) V(1 - O,x) = F(x) (-1 < x < 1).
Ir vL(r,n) is the Legendre transform of V with respect to x on the
interval (-1,1),
This form of the solution is also obtained when the problem in Vis solved
by the method of separation of variables.
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SEC. 161 449
The productfL(n)r" in Eq. (5) is the transform of the convolution (7), Seco159,
of the functions F and T"-1{r"}; that is,
1 - r2 " "
(7) V(r,cosO)= -
2n I o
F(cosP)sin P
I o
(1 + r2 - 2r cos v)- 3/2da dp
PROBLEMS I
1. Using orthogonality properties of Pn(x)on the interval (-1,1), display the ortho-
gonality and norms for the set of polynomials P2n+I(X),n = 0,1,2,..., on the interval I
(0,1) by showing that for each m (m = O,1,2, . . .)
- O if n # m
T2n+1{P2m+1(x)} = { (4m+ 3)-1 ifn = m
where(Sec.160) T2n+I{F} = f~F(x)P2n+l(X)dx.
2. Display the orthogonality and normsoftheset ofpolynomialsP2n(x),n = O,1,2,...,
on the interval (0,1) by showing that for each m(m = 0,1,2,.. .)
- O ifn#m
T2n{P2m(x)}
= { (4m + 1)-1 ifn = m
Also if U is interpreted as steady temperatures and Ur(l,x) as flux of heat, note why
condition (a) is to be expected in order that temperatures remain steady.
Under condition (a) derive the formulas
ifn = 1,2,...,uL(r,0) = C,
1 1""2n+l
U(r,cosO)= -2 C + 2 L - _n=1 n
fL(n)r"Pn(cosO)
uL(r,n)=
i r1
-vL(p,n) dp
op
if n = 1,2,..., uL(r,O)
= C,
=i t
1 1 1
(r < 1),
O V(rt,cos O)dt + '2C
where C is an arbitrary constant. The relation also follows from the series representa-
tions of U and V. It can be verified even if F, U, and V are not independent of C/>,
provided that
I: F(c/>,cos O) sin O dO = O.
"I
i. Let U(r,cos O) be harmonic interior to the sphere r = 1 and satisfy this condition
It the surface:
V(c,x) = V(r,O)= 0,
where A is a constant. With the aid of the transform 1'2n+1 {1}found in Probo 3, derive
the formula
'"
V(r,cos O) = I (4n
n=O
+ 3)iL(r,2n + I)P2n+ I(COSO)
Thus Ln(x) are eigenfunctions and A.= -n are eigenvaluesof the singular
eigenvalue problem consisting of the Sturm-Liouville equation
[xe-Xy'(x)]' - A.e-Xy(x)= O (x > O)
and the requirement that y be of class C" on the half line x ~ O.
We find that the Laguerre integral transformation
oo ~
(3)
Gn{F(x)} = fo F(x)e-xL.(x) dx = fG(n) (n = 0,1,2,...)
has the operational property
(4) Gn{(xe-XF')'} = Gn{xF"(x) + (1 - x)F'(x)} = -nfG(n),
valid if F is of class C" when x ~ Oand if F is (!)(e<l1as x -. 00, where IX< 1.
I McCully, Joseph, The Laguerre Transform, SIAM Review, vol. 2, pp. 185-191, 1960. The
author includes a convolution property for the transformation.
LEGENDRE AND OTHER INTEGRAL TRANSFORMS [SECo 163 453
(1)
Ms{ F(x)} = 1"" F(x)xS-1 dx = fM(S)
with the operational property
(2) Ms{xF'(x)} = -sMs{F} = -SfM(S).
For further iterations of the operator x d/dx, property (2) leads to the
formula
(5) (n = 1,2,...).
454 SECo 163] OPERATlONAL MATHEMATICS
and thus Ix'l = xa when x > O. The integrand of the Mellin integral (1) is
then 19(xa- 1) as x -+ O and we shall assume a > O to ensure the convergence
of the integral at its lower limit. Suppose that F(x) is 19(X-k) as x -+ 00,
where k> O. Then the integrand is 19(I/xk-a+l) as x -+ 00, so the Mellin
transform J.(a + ia) exists when k - a > O and a > O, that is, when
O < a < k, provided that F(x) is 19(x-k) as x -+ 00, where k > O.
When F is continuous (x ~ O) and F' is sectionally continuous, an
integration by parts shows that M.{xF'} exists and that property (2) is valid
if O < (T< k.
The Mellin transformation can be described in terms of the exponential
Fourier transformation Ea. Writing s = (T+ ia and substituting x = e-t,
we find that
that is,
(6)
f
lIIIII!IMI"'f'I'I!'!'!!'ftlH!I"!!It!H"
if F(x) is (9(X-k) as x --+00 and ifO < u < k, where x = e-l. Thus by writing
e-I = x, we have an inversionformula for Ms:
1 p .
(7) F(x) = - lim
f
2n p oo -{I
fM(U
U+iP
+ iex)x-u-,,, dex
i
1
= _ . lim fM(S)X-Sds.
2nI p '" u-iP
(8) (y > O)
BIBLIOGRAPHY 457
Tablas
...
Appendix A
Tables of Laplace Transforms
F(t)
Y+IP
~
I
2 f(s) -lim
2ni P-., I,-1' et%f(z)dz
3 sf(s) - F(O) F'(t)
4 5"f(s)- r I F(O)
- 5"-2F'(O)- .. . - F(n-I)(O) F(n)(t)
I
5 I -f(s)
s f~ F(T)dT
F(t - b) if t > b
6 I e-bY(s) (b > O)
{O ift<b
7 I f(s)g(s)
fo F(T)G(t - T)dT
8 I /,(s) - tF(t)
9 I pn)(s) (-1)"tnF(t)
10 l. r f(x) dx
I
-F(t)
t
~~
II
12 I f(es)-(ea)> O)
f(s
f" e-"F(t)dt
13 I o
I -e -as
>O) F(t) if F(t + a) = F(t)
APPENDIX A 459
f
14 "o e-si F(t)
1 + e -as
dt (a > O) F(t) if F(t + a) = - F(t)
00 a
15 2:
.~o s"
~ (k > O)
1
3 (n = 1.2 )
s" (n - 1)!
1
4
fi J:;
5
1
s,fi 2jf
6 (n = 1.2 )
1 x 3 x 5...(2n - l)fi
r(k)
7 (k > O) rk-I
T
8 e"'
s-a
1
9 te"'
(s - a)2
1 1
10 (n = 1.2 )
(s - ar (n - l)(-Ie"I \
11 r(k)
(k> O)
(s - at
1 1
-(e"'-eI")
(s - a)(s - b) a-b
s 1
-(ae"' - beb')
(s - a)(s - b) a-b
I For more extensive tables of Laplace transforms see the books by A. Erdlyi et al.. vol. l.
G. Doetsch et al.. or V. A. Ditkin and A. P. Prudnikov. listed under Tables in the Bibliography.
2 Here a. b. and (in entry 14) e represent distinct constants.
460 OPERATIONAL MATHEMATICS
14
1 (b - e)e'" + (e - a)e" + (a - b)e"
(s - a)(s - b)(s - e) (a - b)(b - e)(e - a)
1 1 .
15 - sm al
S2+ a2 a
S
16 cosat
S2 + a2
1 .
17 - smhal
a
S
18 cosh at
S2 - a2
1 1
19 2"(1 - cos at)
s(s2+ a2) a
1 1 .
20 "3(at - smat)
S2(S2+ a2) a
1 1 .
21 3(sm al - al cosal)
(S2+ a2)2 2a
S 1 .
22 -smat
(S2 + a2)2 2a
S2 1
23 -(sin at + at cosat)
(S2 + a2)2
2a
S2 - a2 t cos at
24
(S2 + a2j2
S cos at - cos bt
25
(S2 + a2)(s2 + b2) b2 - a2
1 1
26 -e"'sinbt
(s - a)2 + b2 b
s-a
27 e'"cosbt
(s - a)2 + b2
s 1
32 1 s----
- a 22a (eosh al - eos at)
8a3s2
33 1
(S2 + a2)3
e' dn
34
'1 s t-T
--
s Ln(/) = - -(tne-')
n! dtn
s 1
35 I
(s-a) )mea'(1 + 2al)
1
36 1 -
2fit3(eh' - e"')
1 1
371-
,fi+a )m - ae""erfe(ajt)
38 1 ,fi
s - a2
39 I ,fi
s + a2
40 1 1
JS<s - a2)
1 2 fajl
41 1
JS<s + a2) -e-a"
a..fi J,o eA'dA
b2 - a2
42 1 e""[b - a erf(ajt)] - beh" erfe (bJi)
(s - a2)(b + ,fi)
1
43 1 e"" erfe (aJi)
JS<,fi + a)
1
441
(s + a).;+b
b2 J
45 1 - a2
JS<s - a2)(';; + b)
n!
462 1 (1 - sr
s"+t
(2n)!)mH 2n(Ji)
47 I -(1 - s) n!
, H ~
sn+l,fi 1t(2n + 1)! 2n+ 1('" t)
Js + 2a - 1
-fi
1
49
~fi+b
50 r(k)
(k >0)
(s + 4(s + b)k fiL ~ br\-tt4+b1. Ik_t(a ~ bt)
1
51
~.fi+b<s + b) te-tI4+bIt[Io(a~ bt) + Il(a ~ bt)]
52 ~-.fi
~+.fi
53
(a- bY' (k >0)
(~ + fi+b)2k
54 (~ + .fi)-2Y (v> -1)
.fi~
1
55
J S2 + a2
56 (~-st
(v> -1) aYJ y(at)
JS2 + a2
1
57 (k > O)
(S2 + a2t'
58 (k> O)
65
1 I + coth !ks 1 + [t/k] = n when (n - I)k < t < nk
s(1 - e-Os)= ~ (n = 1,2,...) (Fig.5)
O when O < t < k
1 1 + a + a2 + ... + a"-I
66
s(eh - a)
{ when nk < t < (n + I)k (n -- 1,2,...)
1 M(2k,t) = (-1)"- I
67 - tanh ks
s
when 2k(n - 1) < t < 2kn
(n = 1,2,... )(Fig. 9)
1 1 1 - (-1)"
-M (kt ) + - =
68 2 '2 2
s(1 + e-k')
when(n - I)k < t < nk
1
69 "2 H(2k,t) (Fig. 10)
s tanh ks
70 F(t) = 2(n - 1)
s sinh ks when(2n - 3)k < t < (2n - I)k (t > O)
k 1tS
73 -coth- Isin ktl
S2+p 2k
1
74 Hsin t + Isin ti)
(52 + 1)(1 - e-ns)
1
75 -e-Ik/')
5
I
1
76 1"
.Jii cos 2ft
I
1
77
.Jii cosh2ft
1
78
"j;k sin2ft
1
79
J;k sinh2ft
1 t I.-I)/2
80 (/l > O)
~e-Ik/') ( k} J.-I(2ft)
464 OPERATlONAL MATHEMATICS
81 (JI> O)
82 (k> O)
83 (k ~ O)
erfe (2~)
1 P
84 (k ~ O)
jmexp ( -41 )
85 (k ~ O) 2 f!...exp-~ -kerfe ~
V; ( 4t) ( 2Jr )
86 (k ~ O)
s(a + ./S) -e"ke""erfe(a.j + 2~) + erfe(2~)
e -k./S
87 (k ~ O)
fl.a + ./S) e"ke""erfe (aJr" + 2~)
e- k,/$+.i O when O < t < k
88
Js(s + a) { e-t"loHaJt2=k2) when t > k
when O < t < k
89
\ ~ { ~o(aJt2=k2) when t > k
e-k(~-S)
91 (k ~ O)
~
92
1 k- 1 r'(k) log t
96
l< log s
(k> O)
t {[r{kJF - f(k) }
APPENDIX A 465 .
1 J>t r
101 logd -(e- - e")
s-b I
102
2
104 (a > O) -[al log a + sin al - al Ci (al)]
a
2
\ 105 -(1 - eos ar)
I
2
106 -(1 - eoshae)
e
k 1 .
107 aretan -s -Sin
e
kI
1 k
108 -aretan - Si (ke)
s s
1 (2
109 e"'" erfe (ks) (k> O) -exp
k-fi ( - 4k2)
111 (k> O)
-fi
e'" erfe .Jks
nJt<e + k)
1 \ O when O< e < k
112
fierfe (.Jks)
{(no-t whene > k
1 1
113 (k> O)
Jse'" erfe (.Jks) Jn(t + k)
I The exponential-integral funetion El(t) is defined in See. 35. For tables of this funetion and
other integral funetions, see, for instance, Jahnke and Emde, "Tables of Funetions."
2The eosine-integral funetion is defined in Seco35. Si t is defined in Seco22.
I
/ L..
466 OPERATIONAL MATHEMATICS
1
114 I erf (~) ;t sin (2kfi)
115 I -
1, /Serfc - k -e1 -2e"".'t
Js (
Js ) Fe
[t(2k - t)]-t when O < t < 2k
116 I ne-kslo(ks) {O when t > 2k
k - t
117 e-ksll(ks)
{ ~kJt(2k - t) when t > 2k
when 0< t < 2k
1
118 e'''EI(as)
t + a (a > O)
1
119 I -1 - se"EI(as)
,
a (t + a)2 (a > O)
1
120 I
( - Sis) coss + Ci s sin s t2 + 1
t
121 I
(i - Sis) sin s - Ciscoss t2 + 1
/
Appendix B
Tables of Finite Fourier
Transforms 1
ro F(x)sinnxdx = S"{F}
F(x)
2 '"
2 I f.(n) - I f.(n) sin nx
1t n= 1
4 I 2f.(n)
n -
n ~ f
(n - r)F(r) dr - fo
(x - r)F(r) dr
8 I n-x
n
1 x
9 I -(-lr+1
n n
1
10 I -[1
n - (-1)"]
-x ifx<c
11 I
n
cosnc (O< e < n)
{n-xifx>c
- c)x ifx ~ c
(n
12 I ..;
n sin nc
(O< c < n)
{(n - x)c if x ~ e
1 See Chap. 11 for details and operations not tabulated here.
467
468 OPERAT/ONAL MATHEMATICS
1I
Tabla B.1 ,I
(continuad)
6n
13 1 - x(n - x)(2n - x)
n3
2
14 I -[1 - (-1)"] x(n - x)
n3 .
15 I n2 +I 2
-(-1)" --[1-(-1)"]
n n3
16 I 1t(-I)" - --
( n3
6 n
n2)
17 1 [I - (-O"eCO]
n2 + el
n sinh e(n - x)
181 (e #<O)
n + c sinh en I
n
19 I [1 - (-1)" cosh cn] cosh ex
n + c
201
n
(k #< O, + 1, + 2,.. .)
sin k(n - x)
n - sin kn
n
21 I Oirn #<m;f.(m) =- 2 sinmx (m = 1, 2,. ..)
n
22 1 [1 - (-l)"coskn] cos kx (k ,, :t 1, :t 2, .. .)
n-k
1 - (-O"'+'. cos mx
23 I n n-m
2 2 Irn #<m;f.(m) = O (m = 1, 2, .. .)
j e-ny (y> O)
n cosh y
sin x
- cos x
bn 2 bsinx
32 I - (-1 < b < 1) -arctan -
n n 1 - bcosx
1 2 sinx
33 I -e-ny (y > O) - arctan
n n e'-cosx
2 2bsinx
34 I 1-(-1)n bn (-I<b<l) - arctan-
n n 1 - b2
---- 2 sinx
35 I 1 - (-I)n e -n, (y> O) - arctan-
n n sinhy
F(x)
I: F(x)cosnxdx = Cn{F}
2 I .fc(n) .fc(0) +
1t
~ ~
1t..=1
.fc(n) cos nx
. /.,(0)
4 I
n
/.,(n)ifn = 1,2.... f. (x - r)F(r)dr+ -(x~. - n)2+ A
5 (-Ir/.,(n) F(n - x)
6 2.fc(n)cos ne Fix + e) + F2(x- e)
7 28n {F} sin ne F(x + e) - F(x - e)
1
8 -Sn{F} irn = 1,2,...
" - s: F(r)dr+ A
9 2/.,(n)ht(n)
I:. F2(x- r)H2(r)dr
10 /.,(n) ir n .. 0;/.,(0) + An ir n = O F(x) + A
11 I O irn = 1,2'00' ;(O) = n 1
I
12 I - ir n = 1,2.... ;.fc(0) = O
n -~ log (2 sin~)
2 l ifO < x < e
131 -n sin ncir n ..O;/.,(0)= 2c - n { - 1 ire < x < n
I
J
470 OPERATIONAL MATHEMATICS
nI
14 x
(-1)"n2 - 1 ifn 'fi 0;.fc(0) = 2"
21t 1t3
15 2( - 1)"if n 'fi O;h(O) = -
n - 3
16
1. (n
--- - X)2 n
2" If n 'fi O; .fc(0) = O
n 2n 6
241t
17 --:-(-I)n fn 'fi 0;.fc(0) = O
n
18 cosh e(n - x)
~'fi~
e sinh en
(-I)"e'n-I 1
19 ~'fi~ -e'X
n2 + e2 e
1 cos k(1t - x)
20 (k 'fi O. :t 1, :t 2, . . .)
n2 - k2 k sin k1t
f 24
Oifn'fim;.fc(m)=z
1t
cos mx (m = 1,2,...)
25
2e(_I)n+! a cosh ex
~'fi~
(n2 + e2)2 OC( e sinh e1t )
26 ~
1 1 - b2
(-1 < b < 1, b #<O)
; 1 + b2 - 2b cos x
27 sinhy
~>~
1tcoshy - cos x
bn 1
28 -n ifn 'fi O;h(O) = 0(-1 < b < 1) --Iog(l + b2 - 2bcosx)
1t
30
1a sinh y
~>~
1t iy( cos x - cosh y )
n bn
31 -2 -n! if n -# O.' j,(0)
e = 1t exp (b cos x) cos (b sin x)
j
Appendix e
Table of Exponential Fourier
Transforms 1
F(x)
f"" F(x)e-1OXdx= E.{F}
2 I J.(rx)
(m = 1,2,...)
lim -1 -p
p"" 27t
P"')(x)
r J.(rx)e'OXdrx
.
43 I (irx)'"I.(rx)
if(rx) xF(x)
5 I eic"fe(rx) (e real) F(x + e)
6 I J.(rx - e) (e real) e'uF(x)
10 I J.(rx)ge(rx)
1
r-"" F(t)G(x - t) dt
112 ---
e + irx (e> O) e-e",so(x)
1
12 - (e > O) I "'So(-x)
e - irx
131
(e + irx)2
(e> O) I xe-e",So(x)
1 See Chap. 12 for details and Sec. 136 for references to other tables.
2S0(X) = O if x < O, So(x)= 1 if x> O.
471
472 OPERATlONAL MATHEMATICS
1
14 (e> O) - xe""So(- x)
(e - ia)2
2e
15 (e> O) exp (-elxl)
4ica
17 (e> O) -x exp (-elxl)
(e2 + a2)2
n!
18 (Rep > O) (n = 1,2,...)
(p + ia)n+1
e
19 7t exp (-elal) (e> O)
e2 + X2
20 (e> O)
2ex
21 i7ta exp ( - elal) (e> O)
- (e2 + x2)2
22 (e> O)
2 .
23 -sIDea (e> O) So(x + e) - So(x - e)
a
-1 if -e < x < O,
24 21-cosea
ia (e> O)
{1 ifO < x < e, Oflxl > e
25 (e> O)
COS2ex iflxl < ~
{O iflxl >-
2e
a1
i
i
i
J
Appendix D
Table. of Fourier Sine and
Cosine Transforms 1
F(x)
(' F(x)sinax dx = S.{F(x)}
2 '" 2
2 f.(a) -Itof f.(a) sin ax da = -SAf.(a)}
It
3 F"(x)
i
4 -E.{ (E. in Appendix C) F(x)
2 F.(xJ}
5 f.(ak) (k> O)
6 !;la) -x2F(x)
2
7 - I.<a)g.(a)
a f'"o F(r)r+r
I%-rlG(r)dr dr
8 2f.(a) cos ak F.(x + k) + F.(x - k)
9 f.(a + k) + f.(a - k) 2F(x) cos kx
10 f').a) (1. in Table D.2) -xF(x)
11 a/.{a) -F'(x)
12 ~.(a) sin :rk F2(x- k) - F2(x+ k)
13 f.(a - k) - fcta + k) 2F(x)sin kx
. For details and propertics not listed here see Chap. 13. Rererences to other tables are given
in Seco 142
473
474 OPERATlONAL MATHEMATlCS
14 2f.(a)g,(a)
{O F(r)[GUx - rl) - G(x + r)]dr
2ea
18 (e> O)
(e2 + )2
8c3a
19 (e> O) x(l + ex)e-CX
(e2 + (2)3
1 fil
20
.fi ,r~ fi
2 x
21 (e> O)
;-e2 + X2
4e x
22 (e> O)
-; (e2 + x2j2
4x 3e2 - X2
23 (e> O)
1t (e2 + X2)3
24 (e> O) --2 e2
1tX e2 + x2
2 e
25 (e> O) - aretan -
a 1t X
26 (e> O)
x
27
a (e> O) erfe =
2Jc
1 - eosea 1 ifO < x < e
28
a (e> O) {Oifx > e /
4 x
29
sin ea I x +e
30 (e> O) -Iog-
a 1t Ix - el
APPENDIX D 475
fO F(x)
o F(x)cosrxxdx = C.{F(x)}
2 O> 2
2 fc(rx) -non
s fc(rx)cosrxxdrx = -Cx{fc(rx)}
5 fc(rxk) (k> O)
6 - x2 F(x)
7 2fc(rx)gc(rx)
{O> F(r)[G(x + r) + G(lx - rl)] dr
8 2fc(rx)cos rxk (k > O) F(lx - kl) + F(x + k)
9 fc(rx+ k) + fc(rx- k) 2F(x) cos kx
10 f~(rx) (fs in Table D.I) xF(x)
1
11 -f.(rx)
rx fO>
x F(r) dr
12 2f.(rx)sin rxk (k > O) F(x + k) - F1(x - k)
13 f.(rx + k) - f.(rx - k) 2F(x) sin kx
14 2f.(rx)g,(rx)
so> o F(r)[G(x+ r) - G1(x- r)]dr.
e
15 (e> O)
e2 + rx2
2
16 (l + x)e-X
(1 + rx2)2
1
17
1 + rx4
exp ( - fi) sin ( + fi)
rx2
18
1 + rx4
exp ( - fi) cos ( + fi)
1
19 [21
.fi. V~vx
20 2 e
(e> O)
~'e2 + X2
21 2 1 - X2
n (l + X2)2
/
e-a. - e-ea.
22 I (e> O)
a
23 I -1 -e -c
(e> O)
1
-Iog 1 + -X2
e2
a 1t ( )
1 X2
24 I exp (-ea2) (e> O) c: exp --
...1te ( 4c)
1 . 1 2
25 I -e-' SIna - arctan -
a 1t X2
477
478 INDEX
j
I
INDEX 481