Chapter 5 Contour Integration and Transform Theory

5.1 Path Integrals

b

For an integral a f (x) dx on the real line, there is only one way of getting from a to b. For an integral f (z ) dz between two complex points a and b we need to specify which path or contour C we will use. As an example, consider I1 =
C1

dz z

and

I2 =
C2

dz z

where in both cases we integrate from z = 1 to z = +1 round a unit semicircle: C1 above, C2 below the real axis. Substitute z = ei , dz = iei d: 0 iei d I1 = = i ei but
2

I2 =

i d = +i.

The result of a contour interaction may depend on the contour. To formally dene the integral, divide C into small intervals, separated at points zk (k = 0, . . . , N ) on C , where z0 = a and zN = b. Let zk = zk+1 zk and let = max |zk |. Then we
k=0,...,N 1

dene f (z ) dz = lim
C 0

N 1

f (zk ) zk
n=0

where, as 0, N . Note that if C lies along the real axis then this denition is exactly the normal denition of a real integral. 77

R. E. Hunt, 2002

Elementary properties
If C1 is a contour from w1 to w2 in C, and C2 a contour from w2 to w3 , and C is the combined contour from w1 to w3 following rst C1 then C2 , we have that C f (z ) dz = C1 f (z ) dz + C2 f (z ) dz . (Obvious from denition; compare with the equivalent result on the c b c real line, a f (x) dx = a f (x) dx + b f (x) dx.) If C + is a contour from w1 to w2 , and C is exactly the same contour traversed backwards, then clearly C + f (z ) dz = C f (z ) dz . b a (Cf. a f (x) dx = b f (x) dx.) Integration by substitution and by parts work in C also. If C has length L, then f (z ) dz L max |f (z )|
C C

because at each point on C , |f (z )| max |f (z )|.

C

Closed contours
If C is a closed curve, then it doesnt matter where we start from on C : C f (z ) dz means the same thing in any case. (The notation denotes an integral round a closed curve.) Note that if we traverse C in a negative sense (clockwise) we get negative the result we would have obtained had we traversed it in a positive sense (anticlockwise).

5.2

Cauchys Theorem

A simply-connected domain is a region R of the complex plane without any holes; formally, it is a region in which any closed curve encircles only points which are also in R. By a simple closed curve we mean one which is continuous, of nite length and does not intersect itself. Cauchys Theorem states simply that if f (z ) is analytic in a simply-connected domain R, then for any simple closed curve C in R, f (z ) dz = 0.
C

78

R. E. Hunt, 2002

The proof is simple and follows from the CauchyRiemann equations and the Divergence Theorem in 2D: f (z ) dz =
C C

(u + iv )(dx + i dy ) (u dx v dy ) + i
C C

= =
S

(v dx + u dy ) u v x y
S

v u x y

dxdy + i

dxdy,

by applying the Divergence Theorem, where S is the region enclosed by C . But the CauchyRiemann equations show that both brackets vanish, since f is analytic throughout S . The result follows. This result is, of course, not true if C encircles a singularity (we could not then use the CauchyRiemann equations throughout S ).

Changing the Contour

Suppose that C1 and C2 are two contours from a to b and that there are no singularities of f on or between the contours. Let C be the contour consisting of C1 followed by the reverse of C2 . C is a simple closed contour, so f (z ) dz = 0
C

(no singularities are enclosed). Hence f (z ) dz

C1 C2

f (z ) dz = 0,

i.e. f (z ) dz =
C1 C2

f (z ) dz.

So if we have one contour, we can move it around so long as we dont cross any singularities as we move it. If f has no singularities anywhere, then chosen.
b a

f (z ) dz does not depend at all on the path

The same idea of moving the contour applies to closed contours; if C1 and C2 are closed contours as shown, then f (z ) dz =
C1 C2

f (z ) dz 79

R. E. Hunt, 2002

so long as there are no singularities between C1 and C2 . We prove this by considering the closed contour C shown: clearly 0=
C

f (z ) dz =
C1

f (z ) dz
C2

f (z ) dz

(the two integrals along the joins shown cancel).

5.3

The Integral of f (z )
b

For a real function f (x), a f (x) dx = f (b) f (a). This result extends immediately to complex functions, so long as both f and f are analytic in some simply-connected region R and the integration contour C lies entirely in R. Then
b

f (z ) dz = f (b) f (a)
a

for any complex points a, b in R. Note that the specied conditions ensure that the integral on the LHS is independent of exactly which path in R is used from a to b, using the results of 5.2. Examples: (i)
1 2 z dz = 2 (i 02 ) = 1 . (f and f are analytic in the whole of C, so the LHS is 2 path-independent.) i 0

(ii)

ez dz , where C is the semicircular contour joining 1 to +1 along |z | = 1 above the real axis, is equal to e e1 .
C

(iii)

z 1 dz via a straight contour. Note that z 1 is not analytic everywhere, so we do need to specify the contour; but we can dene a simply-connected region R, given 1 by Im z > 2 say, in which it is analytic, and C lies entirely in R. Let f (z ) = log z with the standard branch cut, so that f (z ) is also analytic in R; then
1+i

1+i 1+i

z 1 dz = log(1 + i) log(1 + i)
1+i

= log

2+ 3 i (log 4

2+ 1 i) 4

=1 i. 2

80

R. E. Hunt, 2002

(iv) Now consider 1+i z 1 dz via the contour shown. Dene R as in the diagram; we cannot now choose the standard branch cut for log z (since C would cross it), so we choose a cut along the positive imaginary axis, and dene log rei = log r + i where 32 < 2 . Then z 1 dz = log(1 + i) log(1 + i)
C

1+i

= log

2 + ( 5 4 )i (log

1 2+ 4 i)

= 3 2 i.

5.4

The Calculus of Residues

The Contour Integral of a Laurent Expansion

Consider a single term an (z z0 )n of an expansion, integrated round a closed curve C which encircles z0 in a positive sense (i.e., anticlockwise) once. For n 0, we can use Cauchys Theorem to obtain immediately an (z z0 )n dz = 0.
C

For n < 0, rst change the contour C to C , a circle of radius about z0 , using the ideas of 5.2. On C , z = z0 + ei and so
2

an (z z0 )n dz =
C 0

an n ein iei d
2

= ian n+1
0

ei(n+1) d
2

i(n+1) ia n+1 e n i(n + 1) = ia n+1 (2 ) n 0 n = 1 = 2ia1 n = 1

n = 1
0

n = 1

We deduce that for a function f (z ) with a singularity at z0 , and a contour C encircling the singularity in a positive sense,

f (z ) dz =
C n= C

an (z z0 )n dz = 2ia1 = 2i res f (z ).
z = z0

81

R. E. Hunt, 2002

We can also obtain the result as follows, using the method of 5.3: an (z z0 )n+1 C n = 1 n n + 1 an (z z0 ) dz = C n = 1 an log(z z0 ) C = 0 2ia1 n = 1 (because (z z0 )n+1 is single-valued) n = 1 (because changes by 2 )

The Residue Theorem

Suppose that f (z ) is analytic in a simply-connected region R except for a nite number of poles at z1 , z2 , . . . , zn ; and that a simple closed curve C encircles the poles anticlockwise. Then
n

f (z ) dz = 2i
C k=1

z = zk

res f (z ).

(We have just proved this in the case of a single pole.) Proof: Consider the curve C shown. C encircles no poles, so f (z ) dz = 0
b C

by Cauchys Theorem. But we can also work out the integral round C by adding together several contributions: the large outer curve (which is the same as C ), the small circles around each pole, and the contributions from the lines joining the outer curve to the inner circles. For each k , the contribution from the small circle round zk is 2i resz=zk f (z ) because the small circle goes clockwise round zk . Also, the contribution from the line joining the outer curve to the small circle cancels exactly with the contribution from the line going back. Hence n 0=
b C

f (z ) dz =
C

f (z ) dz +
k=1

2i res f (z )
z = zk

from which the result follows.

82

R. E. Hunt, 2002

5.5

Cauchys Formula for f (z )

Suppose that f (z ) is analytic in a region R and that z0 lies in R. Then Cauchys formula states that 1 f (z ) dz f (z0 ) = 2i C z z0 where C is any closed contour in R encircling z0 once anticlockwise. Proof: f (z )/(z z0 ) is analytic except for a simple pole at z0 , where it has residue f (z0 ). Using the Residue Theorem, f (z ) dz = 2if (z0 ) z z0

as required. Note: Cauchys formula says that if we know f on C then we know it at all points within C . We can see that this must be so by the uniqueness theorem of Chapter 2: u and v , the real and imaginary parts of f , are harmonic, so if they are specied on C (Dirichlet boundary conditions), then there is a unique solution for u and v inside C . Exercise: show that if instead f is analytic except for a singularity at z0 , and has a m Laurent expansion m= am (z z0 ) , then the coecients of the expansion are given by 1 f (z ) dz. an = 2i C (z z0 )n+1 If we dierentiate Cauchys formula with respect to z0 (dierentiating under the sign on the RHS), we see that f (z0 ) = 1 2i f (z ) dz. (z z0 )2

So f (z0 ) is known for all z0 inside C . Continuing this process, f (n) (z0 ) = n! 2i f (z ) dz, (z z0 )n+1

and f (n) (z0 ) is known. So at any point where f is analytic, i.e. dierentiable once, all its derivatives exist; hence it is dierentiable innitely many times.

5.6

Applications of the Residue Calculus

Suppose we wish to evaluate I=

0

dx 1 + x2

83

R. E. Hunt, 2002

(which we can already do using trigonometric substitutions).

Consider
C

dz 1 + z2

where C is the contour shown: from R to R along the real axis (C0 ) then returning to R via a semicircle of radius R in the upper half-plane (CR ). Now (1 + z 2 )1 = (z + i)1 (z i)1 , so the only singularity enclosed by C is a simple pole at z = i, where the residue is limzi (z + i)1 = 1/2i. Hence dz + 1 + z2 dz = 1 + z2
R R

C0

CR

dz 1 = 2i = . 2 1+z 2i

Now
C0

dz = 1 + z2

dx 2I 1 + x2

as R .

Consider CR dz/(1 + z 2 ): the integrand (1 + z 2 )1 is of order R2 on the semicircle, but the length of the contour is R. Hence dz R O(R2 ) = O(R1 ) 0 as R . 1 + z2

CR

Combining all these results and taking the limit as R , 2I + 0 = , i.e. I = /2. This example is not in itself impressive. But the power of the method is clear when we see how easily it adapts to other such integrals (for which it would not be easy, or would be impossible, to use substitutions). Examples: (i) We wish to calculate I=
0

(x2

dx + a2 )2

84

R. E. Hunt, 2002

where a > 0 is a real constant. We consider C dz/(z 2 + a2 )2 ; most of the above analysis is unchanged. The poles now occur at z = ia, and they both have order 2; only the pole at +ia is enclosed by C . The residue there is d 1 2 2 ia3 . lim = lim = = 1 4 3 z ia dz (z + ia)2 z ia (z + ia)3 8ia The integral round the semicircle still vanishes as R , since now
CR

(z 2

dz R O(R4 ) = O(R3 ). + a2 )2
1 3 ia ) = /2a3 , 2I = 2i( 4

Therefore i.e., I = /4a3 . (ii) For I =

0

dx/(1 + x4 ), the (simple) poles are at ei/4 , e3i/4 , ei/4 and e3i/4 .

Only the rst two poles are enclosed. The residue at ei/4 is lim
z ei/4

z ei/4 1 = lim =1 e3i/4 = 1 ei/4 4 4 4 3 1+z z ei/4 4z

1 i/4 e . Hence using LH opitals Rule, and at e3i/4 it is (similarly) 4 1 i/4 1 i/4 2I = 2i( 4 e +4 e ) = 2i( 1 )(2i sin ) = sin , 4 4 4 i.e., I = /2 2.

ei/4 and (iii) For I = 0 x2 dx/(1 + x4 ), the poles are as in (ii) but with residues + 1 4 1 ei/4 respectively (check for yourself). So the value of the integral is unchanged. 4 (iv) For I = 0 dx/(1 + x4 ) again, an alternative to the method used in example (ii) above (and similarly in example (iii) above) is to use a contour which is just a quarter-circle, as shown.

85

R. E. Hunt, 2002

Let C consist of the real axis from 0 to R (C0 ); the arc of circle from R to iR (C1 ); and the imaginary axis from iR to 0 (C2 ). Now C0 dz/(1 + z 4 ) I as R ; and, along C2 , we substitute z = iy to obtain dz = 1 + z4
0 R

C2

i dy = i 1 + (iy )4

R 0

dy iI 1 + y4

as R .

The integral along C1 vanishes as R , using the same argument as for CR above, but this time we only enclose one pole, which makes the calculation easier. Hence 1 1 i/4 e ) = 2 e3i/4 = I = /2 2 I iI = 2i( 4 as before.

Jordans Lemma
For many applications (in particular, ones involving Fourier transforms) we need to show that f (z )eiz dz 0
CR

as R , where > 0 is some real constant and f is an analytic function (except possibly for a nite number of poles). Jordans Lemma states that this is true so long as f (z ) 0 as |z | . For < 0, the same conclusion holds for the semicircular contour CR in the lower half-plane. Note that this result is obvious if f (z ) = O(|z |2 ) as |z | i.e., if f (z ) = O(R2 ) on CR by the following argument. First note that eiz = ei(x+iy) = ey eix , and y 0 on CR , so |eiz | = ey 1 on CR . Hence f (z )eiz dz R max |f (z )|
CR CR

= R O(R2 ) 0 as R . Jordans Lemma simply extends the result from functions satisfying f (z ) = O(|z |2 ) to any function satisfying f (z ) 0 as |z | . Examples: e2iz dz 0 as R ; z eiz dz 0 as R . z2

CR

CR

86

R. E. Hunt, 2002

The proof of Jordans Lemma stems from the fact that for 0 /2, sin 2/ . Now

f (z )eiz dz max |f (z )|
CR CR 0

|eiz | |Rei | d

= R max |f (z )|
0

eR sin d

[using y = R sin ]
/2

= 2R max |f (z )|
0
/2

eR sin d

2R max |f (z )|
0

e2R/ d

= (1 eR ) max |f (z )| 0 as R .

A similar proof holds on CR for < 0.

5.7

Laplace Transforms

The Fourier transform is a powerful technique for solving dierential equations and for investigating many physical problems, but not all functions have a Fourier transform: the integral dening the transform does not converge unless the function tends to zero at innity. To get around this restriction, we can use another kind of transform known as the Laplace transform. The price we pay is a dierent restriction: it is only dened for functions which are zero for t < 0 (by convention). From now on, we shall make this t assumption, so that if we refer to the function f (t) = e for instance, we really mean the 0 t < 0, function f (t) = et t 0. The Laplace transform of a function f (t) is dened by

(p) = f
0

f (t)ept dt

(p); and the where p may be complex. The notation [f ] or [f (t)] is also used for f symbol s is often used instead of p. Many functions for instance, t and et which do not have Fourier transforms do have Laplace transforms; however, there are still exceptions

87

R. E. Hunt, 2002

(e.g., et ). Laplace transforms are particularly useful in initial value problems, where we are given the state of a system at t = 0 and desire to nd its state for t > 0. Examples: (i) (ii) (iii) (iv)

L [1] = L [t] = L [e
t

0 0

1 ept dt = . p 1 tept dt = tept p

+
0

1 p

ept dt =
0

1 . p2

]=
0

e(p)t dt =

1 . p 1 1 pi p+i = p2 1 . +1

L [sin t] = L

1 1 it (e eit ) = 2i 2i

Note that, strictly speaking, in example (iii), the integral only converges for Re p > Re (otherwise the integrand, e(p)t , diverges as t ). However, once we have calculated the integral for Re p > Re we (p) to exist everywhere in the complex p-plane (except for singularities such as at p = can consider f in this example). This process of extending a complex function which is initially only dened in some part of the complex plane to the whole of the plane is known as analytic continuation.

It is useful to have a library of Laplace transforms to hand; some common ones are listed below. f (t) 1 et sin t sinh t et sin t (t) (p) f 1 p 1 p 2 p + 2 2 p 2 (p )2 + 2 1 f (t) tn tn et cos t cosh t et cos t (t t0 ) (p) f n! pn+1 n! (p )n+1 p 2 p + 2 p 2 p 2 p (p )2 + 2 ept0

Elementary Properties of the Laplace Transform

(i) Linearity:

L [f (t) + g(t)] = f(p) + g (p).

88

R. E. Hunt, 2002

(ii) Change of scale: using the substitution t = t,

L [f (t)] =
(iii) Shifting theorem:

f (t)ept dt =
0 t

f (t )e(p/)t dt =
0

1 p f .

L [e

(p ). (Easy to check.) f (t)] = f

(iv) Derivative of a Laplace transform:

L [tf (t)] = ddp f(p).

Proof:

(p) = f
0

f (t)ept dt

d f (p) = dp

tf (t)ept dt.
0

By repeating this trick n times, we see that the Laplace transform of tn f (t) is (n) (p). (1)n f Examples:

L [t sin t] = ddp p

1 2p = 2 ; 2+1 (p + 1)2

L [t ] = (1)
n

n! dn 1 = n+1 . n dp p p

(v) Laplace transform of a derivative:

L
Proof:
0

df dt

(p) f (0). = pf

df pt e dt = [f (t)ept ] 0 +p dt

(p) f (0). f (t)ept dt = pf

0

We can deduce that

L
and so on.

d2 f dt2

=p

df (p) pf (0) f(0) f(0) = p2 f dt

(p) f (0) as p , and pf (p) limt f (t) as p 0. (vi) Asymptotic limits: pf Proofs: from (v) above,

(p) = f (0) + pf
0

df pt e dt, dt

(p) f (0). Similarly, as so as p (and therefore ept 0 for all t > 0), pf p 0, ept 1 so that

(p) f (0) + pf
0

df dt = f (0) + [f (t)] 0 = lim f (t). t dt 89

R. E. Hunt, 2002

Solving Dierential Equations using Laplace Transforms

The Laplace transform is particularly suited to the solution of initial value problems. Example: solve y + 5y + 6y = 0 for y (t) subject to y (0) = 1, y (0) = 4. Taking Laplace transforms, and using the results for the Laplace transform of a derivative, we see that (p2 y (p) p + 4) + 5(py (p) 1) + 6 y (p) = 0, which we may solve for y (p): y (p) = p2 p+1 2 1 p+1 = = + 5p + 6 (p + 2)(p + 3) p+3 p+2

using partial fractions. We now need to invert y (p) to nd y (t); in general we must use the inversion formula described below, but in many cases (such as this one) it is possible to spot the answer using the library of transforms given above (and taking advantage of the fact that inverse Laplace transforms are unique). Here, we know that [et ] = 1/(p ); hence y (t) = 2e3t e2t .

The Convolution Theorem for Laplace Transforms

The convolution of two functions f (t) and g (t) is

(f g )(t) =

f (t t )g (t ) dt .

We are dealing here with functions which vanish for t < 0, so this reduces to
t

(f g )(t) =
0

f (t t )g (t ) dt

since g (t ) = 0 for t < 0 and f (t t ) = 0 for t > t. The convolution theorem for Laplace transforms then states that (p) [f g ] = f g (p).

Proof:

L [f g ] =
=

0 0 0 0

f (t t )g (t ) dt
t

ept dt

f (t t )g (t )ept dt

dt.

90

R. E. Hunt, 2002

From the diagram, we see that we can change the order of integration in the (t, t )plane, giving

L [f g] =
=

0 0 0 t

f (t t )g (t )ept dt dt

f (t )ept ept dt [substituting t = t t ]

g (t ) dt

=
0

(p)ept }g (t ) dt {f

(p) =f
0

g (t )ept dt

(p) =f g (p) as required.

The Inverse Laplace Transform

Inverting Laplace transforms is more dicult than inverting Fourier transforms because (p), we can calculate f (t) it is always necessary to perform a contour integration. Given f using the Bromwich inversion formula f (t) = 1 2i
+i

(p)ept dp. f
i

Here is a real constant, and the Bromwich inversion contour runs from i to + i along a straight line. (p). must lie to the right of all the singularities of f

91

R. E. Hunt, 2002

Note that it is possible to derive the Bromwich inversion formula from the inverse Fourier transform by (p) = f (ip) where f (k ) is the Fourier transform of f (t). The only substituting p = ik and noting that f dierence is in the detail of the inversion contour.

(p) has only poles, and no other singuSuppose that f larities; all these poles lie to the left of . When t < 0, consider the integral round the contour C shown consisting of C0 followed by CR . C encloses no poles, so (p)ept dp = 0. f
C

Now on CR , Re p , so Re(pt) t (since t < 0) and hence (p) = O(|p|2 ) as |p| i.e., if |ept | et . Therefore if f (p) = O(R2 ) on C then f R (p)ept dp Ret O(R2 ) 0 f
CR

(p) 0 as |p| , by a as R . In fact the same is true even if we only have f slight modication of Jordans Lemma. So in either case, (p)ept dp = lim f

(p)ept dp f
C0

= lim

(p)ept dp f
C CR

(p)ept dp f

= 0 0 = 0, and therefore for t < 0 the inversion formula gives f (t) = 1 2i (p)ept dp = 0 f

(as it must do, since f (t) = 0 for t < 0 by our initial assumption). When t > 0, we close the contour to the left instead, and once again we can show that (p)ept dp 0 f
CR

(p) 0 as |p| . Hence in the as R , so long as f limit R we obtain

n

(p)ept dp = 2i f
k=1

p=pk

(p)ept res f 92

R. E. Hunt, 2002

(p). We deduce that by the Residue Theorem, where p1 , . . . , pn are the poles of f
n

f (t) =
k=1

p=pk

(p)ept res f

(p) 0 as |p| . for t > 0, so long as f Examples: (p) = 1/(p 1). This has a pole at p = 1, so we must use > 1. We have f (p) 0 (i) f as |p| , so Jordans Lemma applies as above. For t < 0, therefore, f (t) = 0, and for t > 0, ept p1 = et .

f (t) = res

p=1

This agrees with our earlier result for the Laplace transform of et when = 1. (p) = pn . Here we need > 0, because there is a pole of order n at p = 0. For (ii) f t < 0, f (t) = 0 as usual. For t > 0, f (t) = res ept pn = lim = lim = 1 dn1 pt e (n 1)! dpn1 1 (tn1 ept ) (n 1)!

p=0

p0

p0

tn1 . (n 1)!

(p) 0 as |p| ? Consider the example (iii) What if f

p (p) = e ; f p

(p) . We need to calculate here, as p on the real axis, f f (t) = 1 2i ep pt e dp, p

but Jordans Lemma does not immediately apply. Note, however, that ep ept = ep(t1) = ept where t = t 1; so f (t) = 1 2i ept dp. p

93

R. E. Hunt, 2002

Now we can use Jordans Lemma: when t < 0, close to the right, and when t > 0, close to the left, picking up the residue from the pole at p = 0. Hence 0 t < 0, f (t) = 1 t > 0 0 t < 1, = 1 t > 1.

(p) = p1/2 ? We need to nd What function f (t) has Laplace transform f f (t) = 1 2i
+i

p /2 ept dp.
1

For t > 0 we can close the contour to the right as usual and obtain f (t) = 0. For t < 0, however, the branch cut gets in the way.

Use a contour as shown, with a small circle of radius round the origin and two large quarter-circles of radius R. Substituting p = ei on the small circle gives a contribution of

/2 ei/2 ee

iei d = O( /2 ) 0

as 0.

Similarly, the integrals round the two large quarter-circles vanish as R , using the method used to prove Jordans Lemma. Hence the required integral is equal to the sum of the integrals on either side of the branch cut: i.e., for t > 0, f (t) = 1 2i
0

r /2 ei/2 ert (dr)

1

r /2 ei/2 ert (dr)

1

[substituting p = rei and p = rei respectively] = 1 2i 2

2i
0
2

r /2 ert dr
1

es t ds
0

[substituting r = s2 ]

94

R. E. Hunt, 2002

1 = . t 1 1 So [t /2 ] = p /2 . This is a generalisation of the result that +1 1)/p where the Gamma function is dened by

L [tn] = n!/pn+1 to L [t] = ( +

() =
0

x1 ex dx

and can easily be shown to be equal to ( 1)! when is a positive integer.

95

R. E. Hunt, 2002