You are on page 1of 27

Complex Analysis

by Dr. Gbor Csnyi Michelmas 2008: M 121pm LR3, F 910am LR3B

Lecture 1. Introduction to complex variables. CauchyRiemann equations. Taylor expansion. Lecture 2. Classication of singularities. Laurent expansion. Cauchys theorem and contour integration. Lecture 3. The residue theorem. Evaluation of the residue at a simple pole. Further examples of contour integration. Lecture 4. Jordans lemma. Fourier transforms. Lecture 5. Multi-valued functions: branch points and branch cuts. Contour integration for multivalued functions. Lecture 6. Cauchy principal value. Summation of series. Lecture 7. 2D uid ow. Conformal mapping. Joukowski transformation.

Lecture 1: Complex functions of complex variables


A complex function f (z) is a mapping from the complex plane to the (same) complex plane. Consider for example f (z) = z 2: f (z) maps z into the point u + iv, so from the (x, y) plane into the (u, v) plane. f (z) = z 2 = u = x2 y 2 v = 2xy (x + i y)2 = x2 y 2 + i2xy

f (z) f (z) = u + iv

z = x + iy

Other examples of complex functions: sin z, log z, ez. For real arguments z = x, each of these functions takes the obvious value, sin x, log x, ex, respectively. For complex values, the functions are dened via their series expansions, so z2 z3 + + 2! 3! 3 5 z z sin z = z + + 3! 5! ez = 1 + z + Series expansions will be central to complex analysis. Note that this explains the link between the two forms of a complex number z = x + i y = rei. Substitute z = i into the above, ei = 1 + i + (i)2 (i)3 (i)4 (i)5 + + + + 3! 4! 5! 2! 2 4 3 5 = 1 + + +i + + 2! 4! 3! 5! = cos + isin

ei

cos

Dierentiation
Just as for a real function, the derivative is dened as the following limit, f (z) =

df f (z + z) f (z) = lim dz z z0

Note however, that the value of the derivative cannot depend on the path through which z approaches zero. Examples: f (z) = z 2 f (z) = 2z f (z) = z n f (z) = nz n1 f (z) = log z f (z) = 1/z The formulas are the same as for real variables. In general, if and only if f (z) exists at a point z, then f (z) is said to be analytic (or regular or holomorphic) at that point. However, there may be points for which f (z), does not exist, and/or f (z) diverges (blows up). When these points are isolated, they are called singularities. For example, f (z) = 1/z has a singularity at z = 0, everywhere else it is analytic. There may be more than one, or an innite number of singularities, e.g. f (z) = tan z has singu3 3 larities at z = , 2 , 2 , 2 , 2 ,

CauchyRiemann equations
The above condition for the existence of the derivativethat its value is independent of the path of zis a rather strong condition. Write f (z) = u(x, y) + iv(x, y), and let us consider two dierent paths: y

z z

x 1. z = x f (z) = dz = x = x + i x 2. z = iy f (z) = dz =

df

df

1 f i y

= i y + y

If f (z) is to be analytic, then f (z) must be independent of the path, so equating real and imaginary parts, we get, v u = y x v u = x y 3

In fact, the relationship holds both ways, f (z) is an analytic function of z Cauchy Riemann equations are true

Complex analysis is basically the study of complex analytic functions. The condition of a complex function being analytic is much stronger than the analogous condition for a two dimensional real function (f : R2 R2) to be smooth or dierentiable. The CauchyRiemann equations relate the derivatives in the two directions to each other. Examples 1. f (z) = z 2 = (x + i y)2 = (x2 y 2) + i2x y
u x u y

= 2x = 2y

v y

= 2x
v x

= 2y

2. f (z) = z = x i y
u x u y

=1

v y v x

=1

=0

=0

not analytic 3. f (z) = ez = e(x+i y) = excos y + iex sin y (check at home!)

Singularities
In addition to analytic functions (for which the CauchyRiemann equations hold at every point), it is very useful to think about functions, which are analytic almost everywhere, except at special points called singular points. Consider the function f (z), which may have isolated singularities at particular values of z, everywhere else it is analytic. For example, f (z) = 1/(z 2) is analytic everywhere, except at z = 2. All other points are called ordinary points, to distinguish them from the singular points. The function can be represented by a Taylor series expansion about any ordinary point. Take an ordinary point z0, then

f (z) =
n=0

an (z z0)n

an =

1 (n) f (z0) n!

Taylor expansion about an ordinary point

R=1 z =2+i R=2 z=0 z =2

Expand f (z) = 1/(z 2) about z0, using f (z) = 1 z 2


1

f (z) =

1 (z 2)2

f (z) =

2 (z 2)3

etc.

i. Expand around z0 = 0 f (z) = f (0) + zf (0) + 2 z 2 f (0) + f (z) = 2 4


1 z z2 8

converges for |z | < 2


1

ii. Expand around z0 = 2 + i f (z) = f (z0) + (z z0) f (z0) + 2 (z z0)2 f (z0) + f (z) = i + (z z0) i(z z0)2 + converges for |z z0| < 1

The Taylor series expansion converges over a circular disk, centered at z0, and extending out to the nearest singularity.

Lecture 2: Classication of singularities, Cauchys theorem


Expansion about a pole We can expand f (z) about a singularity using negative powers of (z z0). If we achieve this with a nite number of negative powers, the largest of which is N , then the singularity is called a pole of order N . The expansion of f (z) at z = z0 looks like

f (z) =
n=N

an(z z0)n =

a1N aN + + (z z0)N (z z0)N 1

+ a0 + a1(z z0) + a2(z z0)2 +

This series is called the Laurent series expansion for f (z). The function we considered above, f (z) = 1/(z 2) has a simple pole (a pole of order one) at z = 2, and the expansion around it just has one term: 1/(z 2). Consider a less trivial case, f (z) = z 2/(z 1)2. Here, f (z) has a pole of order 2 at the point z0 = 1. Expand f (z) about this point by writing the Taylor series for the numerator around z0 = 1, z 2 = 1 + 2(z 1) + (z 1)2 Thus the Laurent series expansion about the pole is f (z) = Examples 1. f (z) = 1/(z 2)2 has a pole of order 2 at z = 2 has two simple poles (poles of order 1) at z = 2 and z = 1 1 + 2(z 1) + (z 1)2 1 2 = + +1 (z 1)2 (z 1)2 z 1

2. f (z) = 1/(z 2)(z 1)

3. f (z) = 1/sin3z has poles of order 3. To see this, note that near z = 0, sin z z, so f (z) 1/z 3. Near z = n, putting w = z n, sin z = sin(w + n) = ( 1)n sin w ( 1)n w f (z) ( 1)n/w 3 = ( 1)n /(z n)3 Essential singularities Now consider the case where f (z) has a pole around z0 such that the Laurent series extends to , so that

f (z) =

an(z z0)n

In this case, f (z) has an essential singularity at z = z0. For example, e1/z = 1 + 1 1 1 1 1 + + + z 2! z 2 3! z 3

has an essential singularity at z = 0. The function diverges so strongly, that lim z n e1/z = (in other words, the limit does not exist)

z0

for any nite n. 6

Branch points There are points which are special not because the function under consideration diverges there, but because no expansions exists around it. Such singular points can arise because of the multivalued nature of the function. In the analysis of real functions, normally only one value is associated with a given argument. In the case of complex functions, it is very useful to allow more than one (or perhaps even an innite number of) function values to be associated with a particular function argument. A simple example is f (z) = z 1/2, which has two values for each value of z. To see this, use the polar form of a complex number, f (z) = z 1/2 = r 1/2ei/2. Consider the possible values of the function as the argument z moves around the unit circle. At z = 1, both f (z) = 1 and f (z) = 1 are valid function values, corresponding to = 0 and = 2, respectively. If we start from the f (z) = 1 value, and try to make the function remain continuous as z traverses the unit circle, we get f (z) = 1 when we arrive back to z = 1. If we continue and sweep the unit circle a second time, we get back to the f (z) = 1 solution. The problem is to do with going around the branch point z = 0, where the two solutions come together, because f (0) = 0 is the only solution there. Thus it is not surprising that no Taylor or Laurent series exists around z = 0, since its value would have to be ambiguous. y z r Branch cut x

Branch point

One way to visualize such surfaces is to imagine that the complex plane is cut near the branch cut (hence the name), and the sections are glued together to form a continuous (but necessarily selfintersecting) object, as shown below. Note that the function does not diverge at the singular point z = 0, its value is unambiguously zero there. The branch cut is not unique. In the image below, dont confuse the branch cut (e.g. positive real axis) with the half line where the surface intersects itself (e.g. negative real axis). The image is a projection where the height of the surface is the real part of the function, the colour is the imaginary part.

An example of a function with an innite number of values at every point is f (z) = log z. Part of the corresponding surface (reminiscent a staircase) is below (the real surface extends in all directions to innity).

Cauchys Theorem
The following theorem holds, which might be very surprising at rst sight: If f (z) is analytic inside a closed curve C , then
C

f (z)dz = 0

C A No singularities inside Proof: Recall Greens theorem:


C

F ds =

( F ) da h g dxd y x y

gdx + hd y =
C

In the above case, let f (z) = u + iv, so f (z)dz =


C C

(u + iv)(dx + id y) udx vdy + i vdx + udy


C

=
C

= = 0

u v v u + dxdy + i dxdy x y x y by the CauchyRiemann equations. 8

Corollary: Now pick two points A and B along the path C, and it follows that the integral of f between the two points along the two segments of C are equal. But this is true for any closed curve C that contains the two points, so the integral of f between the points is independent of the path taken! The power of complex analysis rests on these two results.

C1

C2 C = C1 C2

From Cauchys theorem, the integral over the entire curve is zero, but it is also equal to the sum of the integrals over the separate pieces, 0 =
C

f (z)dz f (z)dz f (z)dz


C2

=
C1

f (z)dz
C2

f (z)dz =
C1

The negative sign comes in because one of the paths is traversed backwards in completing the loop.

Lecture 3: Residue Theorem


Residue Theorem
Now suppose we want to evaluate the integral of a function I = C f (z)dz around a closed contour that does contain isolated singularities inside, at z = z1, z2, z3.

z1 z2 z3 C

Using the fact that the contour can be distorted without aecting the value of the integral (as long as we do not cross any singularities during the distortion), the new contours shown below will do just as well:

C1 C2 C3

The last contour has innitesimal circles C1, C2 and C3 around each singularity inside C, connected by straight line segments that are innitesimally separated. Because the lines segments are traversed in opposite directions, their contribution to the integral vanishes. So we are left with I =
C1

f (z)dz +
C2

f (z)dz +

I1 + I2 + The task now is to calculate the integral around each singularity over a (very small) circle. To do this, expand f (z) using the Laurent series about each pole zi, in turn. For example, near z1,

f (z) =
n=N

an(z z1)n

Now, because we will want to take z around in a small circle, let z z1 = ei,

f (z) =
n=N

an n ei n along the circular path C1 10

dz = ei id

dz C1 z1 So
2

I1 =
C1

f (z)dz =
0 2

d iei
n=N

an nein

= i
n=N

n+1 an
0

ei (n+1) d
2

For n 1, the integral vanishes, for n = 1, it is 0 ei(1+1)d = 2, so I1 = 2i a1. The coecicent a1 in the Laurent expansion of f (z) about z1 is called the residue of f (z) at the pole z1. Repeating the above process for all the poles inside the original contour, we get I =
C

f (z)dz = I1 + I2 +

= 2i (sum of residues of poles inside C)

This amazing result is the Residue Theorem. It says that to evaluate an integral on a closed path C, no matter how complicated in shape, all you need to do is get the a1 coecient of the Laurent expansion at each pole inside C.

Examples of countour integration


We can use contour integration to evaluate dicult real denite integrals. Example 1. I =
0

dx/(1 + x2)

Consider the function over the entire complex plane, so f (z) = 1/(1 + z 2). The function has poles at z = i. Let us evaluate the integral over the closed curve C = C1 + C2, a very large semicircle with radius R :

C2

i C1

The integral over C is the sum over the parts, i.e. C1, the integral along the real axis (which is twice what we want, 2I), and the integral over the curved part, C2. Along C2, z = Rei, so dz = 1 + z2 iReid 0 1 + R2 ei2 11 as R

C2

C2

So, using the residue theorem, we have 2I =


C

dz = 2i 1 + z2

(residues)
1 2i 1 z i

To get the residues, note that 1/(1 + z 2) = 1/(z i)(z + i) = 1 2i z = i is 1/2i. Therefore the integral is I = 2 2i = /2.

1 z+i

, so the residue at

Compare this with the solution from the formula book: dx/(1 + x2) = tan1x, so the denite 1 1 integral is I = tan tan 0 = /2. Notice that the value of the rst term is ambiguous. No such ambiguity arises in the contour integration. Example 2. I =
2 0

cos2 d

Consider the unit circle in the complex plane: z = rei, r = 1. Then dz = idei. The function we want to integrate needs to be written in terms of a complex variable, cos = d =
2 0

cos2 d = = =

1 i 1 1 e + ei = z+ 2 z 2 dz/iei = dz/iz 2 1 dz 1 z+ z iz 2 i (z + 2/z + 1/z 3)dz 4 i 2i2 = 4 d/(2 + cos)

which has a pole at z = 0 with residue of 2

Example 3. I =

2 0

As before we work on the unit circle, so let z = ei, d = dz/iz and cos = complex integral is then
2

1 2

z + 1/z . The

I =
0

= i

d = 1 + cos 2dz z 2 + 4z + 1

dz iz(2 + z/2 + 1/2z)

Now consider the integrand for poles: z 2 + 4z + 1 = (z + 2)2 3 = 0 when z = 2 3 . The ve root (say ) is outside the unit circle, the +ve root, , is inside. To get the residue at , we break up the fraction, 1 z 2 + 4z + 1 = 1 ( )1 ( ) 1 = + (z )(z ) z z

The nal expression is like the Laurent series around z = (we were to further expand the rst term in a Taylor series around z = ), so the residue is ( )1 = (2 3 )1. Hence the inegral is I = i2(2i)(2 3 )1 = 2/ 3

Summary of Residues
Cauchys theorem is that along a closed contour C, C f (z) dz = 2 i (residues), where the sum goes over all the residues of poles inside the contour. The residue is the a1 coecient of the Laurent series expansion of the function around each pole. Given the Laurent series f (z) = a1 + a0 + a1(z z0) + a2(z z0)2 + z z0 12

The residue is given by a1 = limzz0 [(z z0)f (z)] . Example: f (z) = z2 1 with residues at z = 1. What is the residue at z = 1 ? Residue = limz1 (z 1)
z+2 (z 1)(z + 1) z+2

=2
q(z) , p(z)

Now suppose the function is given in the form f (z) = simple zero at z0. Then, for z near z0,

where q(z) is analytic and p(z) has a

q(z) q(z0) + (z z0)q (z0) + 0 + (z z0)p (z0) + p(z) q(z0) + O(z z0) (z z0)p (z0) so the residue at z0 is a1 = q(z0)/p (z0) . If f (z) has a pole of order N at z = z0, then aN a1N a + + + 1 + a0 + a1(z z0) + a2(z z z0 (z z0)N (z z0)N 1 z0)2 + (z z0)Nf (z) aN + a1N (z z0) + + a1(z z0)N 1 + a0(z z0)N + a1(z z0)N +1 + f (z) d dz
N 1

(z z0)Nf (z)

(N 1)!a1 +

N! (N + 1)! a0(z z0) + a1(z z0)2 + 1! 2!

So taking the limit z z0, we get the residue, a1 = 1 lim (N 1)! zz0 d dz
N 1

(z z0)Nf (z)

13

Lecture 4: Jordans Lemma and Fourier Transforms

Consider the integral I = eiaz g(z)dz around a semicircle of radius R in the upper half plane (UHP). Jordans lemma states that in the limit R , I = 0 provided that: i. g(z) 0 as z in the UHP, and ii. a is real and a > 0 Proof Write z = Rei and dz = Rei id, so I = magnitudes of the factors in the integrand,
0

ieiaRcos aRsin g(Rei)Rei d. Now consider the

|i| = 1, eia R cos = 1, ei = 1 and since g(z) tends to zero for large R, let g(Rei) < G(R) for some G(R) (independent of ). Then
/2

|I | <

ea R sinRG(R)d = 2
0 0

ea R sinRG(R)d

Now 2/ sin for 0 /2, so ea R sin e2aR/ for the same range. So
/2

|I | < 2 =

RG(R)e2aR/ d
0

G(R) 1 eaR a

Now take the limit R , and since a > 0 and G(R) 0, we have |I | 0. A simple corollary is that if a < 0, then the integral over a semicircle in the lower half plane vanishes.

Example: electrical response

14

For an example use of Jordans Lemma, let us consider the response of an electronic circuit to a voltage spike, i.e. a delta function signal, V (t) = A(t). We will derive the response to this pulse from the solution to harmonic case (sinusoidal voltage input). Recall that the Fourier Transform of V is written as

V () =

eit A(t)dt = A

Which means that the magnitude of the voltage in the frequency domain is constant, similar to white noise (bonus quesion: how is this dierent from actual white noise?). The inverse transform is V (t) = 1 2

1 V ()eit d = 2

Aeit d

so we can think of the voltage V (t) = A (t) as made up of an innite sum of harmonic signals eit, each of magnitude A d/2. In our specic case, we know that the current response to a sinusoidal input eit is I(t) = So the total response is I(t) = A 2

eit R + iL eit d R + iL

which we now need to evaluate. The integrand can be rewritten as eit eit = iR R + iL iL L which has a simple pole at = i R/L. We want to integrate along the real line, so we need to close the contour integral either in the upper or lower half plane. If t < 0, we close the contour in the lower half plane, so by Jordans Lemma, the integral along the semicircle vanishes. plane iR/L

There are no poles in the lower half plane, so the integral along the real axis must vanish also, i.e. I(t) = 0 for t < 0. This is good, because it would have been strange to get a response before the signal! plane iR/L

15

For t > 0, we close the contour in the upper half plane, and again the integral along the semicircle vanishes due to Jordans Lemma. But now there is a pole inside the contour, so I(t) = eit A d 2 R + iL A = 2i (residue at pole = iR/L) 2 eR t/L A Rt/L = Ai = e iL L

giving an exponential decay as the response to a spike voltage.

16

Lecture 5: Multi-valued functions


Now we look at multi-valued functions in more detail. Remember that complex functions (just like real functions) are mappings, they map points from the complex plane to other points in the complex plane. For functions that are analytic everywhere, such a mapping maps each point in the complex plane to a single point (for real functions, this property is among the requirements for calling a mapping a function). We saw previously that in order to call some complex mappings functions, we have to relax this requirement, and this would be nice to do, because these mappings are in fact analytic almost everywhere. Our rst example is f (z) = z 1/2, which maps every point to two points. Let us look at what the function does on the unit circle, where z = ei, so f (z) = ei/2. In order to see that the multi-valuedness cannot be avoided, we consider how the function varies continuously as z traverse the unit circle counterclockwise starting from z = 1. c b g c z f j a e e f d b f (z) h /2 a j

i d

g h

Notice how for example points a and f are adjacent in the zplane, but a and f are not adjacent in the f (z)plane. When we make a complete revolution with z, f (z) only traverses half the circle, so even though we are back with z where we started from, now we have a dierent value for f (z). A second revolution with z completes the single revolution in f (z). Of course one option would be to restrict the argument of z to 0 arg(z) < 2 (note the strict inequality), and then we would have a one-to-one mapping, but it would have a gaping discontinuity on the positive real axis, or wherever we chose the lower limit of arg(z). Another option is to view the values of f (z) not on a single complex plane, but on a Riemann surface, which in this case consists of two complex planes (one for each revolution of z), joined together smoothly at a halfline (where the discontinuity appeared before). The joining is called the branch cut (indicated with a zigzagged line), and in fact it does not have to be a straight half-line, it can be any continuous non-self-interesecting curve which starts at the origin and goes o towards innity. The resulting surface is independent of how we choose the branch cut.

Once we have picked a branch cut, and we consider the point z moving smoothly around in the complex plane, we have to make sure that it either does not cross the branch cut, or if it does, we have to remember that the function values are only continous if the path of z crosses the branch cut a second time. Now let us consider a more complicated example, f (z) = z2 1 = (z 1)(z + 1) = 1 +2 i 2 r1r2 e , where the radii and angles are measured with respect to the two poles z = 1. Each factor in f (z) gives rise to a branch point at their respective poles, so for a unique deni17

tion of the function value we have to insert two branch cuts, each one attached to a branch point. There are two convenient choices: z r2 e f 1 2 c d 1 r1 1 a b

Here, the ranges of the angles are 1, 2 , and the function is discontinuous across the segment between the branch points. To see this, consider the function value at the labelled points. At a and b, 1 = 2 = 0, f (z) = r1r2 ei0 = r1r2 . At c, 1 = , 2 = 0, f (z) = r1r2 ei/2 = i r1r2 , however at d, 1 = , 2 = 0, f (z) = r1r2 ei/2 = i r1r2 , so the function is discon tinuous here. Over on the other side at e, 1 = , 2 = , f (z) = r1r2 ei = r1r2 and at f , i = r1r2 , so again the function is continuous. 1 = , 2 = , f (z) = r1r2 e However, we can also choose the branch cuts dierently: z r1 |w| |w| 1 2 i|w| i|w| 1 1 |w| |w| w = f (z)

r2

The ranges are now 0 1 < 2 and 2 . The function is again discontinuous upon crossing these new branch cuts. Summary In order to work with a multi-valued function, we have to i. Find the branch points ii. Choose branch cuts iii. Choose a value for the function at a single point Example 1. Suppose we want to evaluate the real integral

I=
0

x 1/3 dx 1+x 18

In order to do so, let us consider the complex version J=


C

z 1/3 dz 1+z

along a suitable contour C. Since z 1/3 has a branch point at z = 0, we choose the positive real axis as the branch cut, and break up the contour as shown. C2 R 1 C4 C3 C1

The complex integrand has a simple pole at z = z0 = 1, where the residue is given by limzz0 [(z z0)f (z)], (z + 1)z 1/3 = ( 1)1/3 = (ei)1/3 = ei/3 1+z z1 lim The integral around the whole contour is the sum of the integrals around the segments, J = J1 + J2 + J3 + J4 Ji =
Ci

f (z)dz

Taking each in turn, for the integral around the branch point let z = ei, dz = izd |J4| = 1/3 ei/3 iei d = 2/3 1 + ei ei/3 i ie d 1 + ei

C4

C4

So |J4| 0 as 0. For the outer arc, let z = Rei, dz = izd, |J2| = R 1/3 ei/3 iRei d = R2/3 1 + Rei ei/3 iei d = O(R1/3) 0 as R 1 + Rei

C2

C2

We are left with the two straight segments. Now J1 = I, the integral we actually want, and for J3, let z = xei2, dz = dx, z 1/3 = x1/3ei2/3, so
0

J3 =

x1/3 ei2/3 dx = ei2/3I 1+x

So adding up all the integrals, we have J = I(1 ei2/3). But using the Residue Theorem, we also know that J = 2i (residues) = 2iei/3, so 2iei/3 = (1 ei2/3)I 2 2 2i cos i sin = 1 cos + i sin 3 3 3 3 2 I = 3 19

Example 2.

I =
0

dx 1 + x3

The integrand is not even, so we cannot extend the integral to . The denominator has a suciently large power of x such that the complex integral will vanish on a large outer loop. The trick is to consider a slightly modied complex analogue, J =
C

ln z dz 1 + z3

Note that this is a dierent integral tour be the same as for Example 1. arc, J4 ln as 0, so J4 straight segments, above the branch So J = J1 + J3 =
0

on the real axis, ln z 1. Let the branch cut and the conThe integrand now has three poles inside C. For the inner 0. For the outer arc, similarly J2 0 as R . On the cut ln z = ln x, and below the branch cut ln z = ln x + 2i. ln x dx 1 + x3
0

ln x + 2i dx = 2iI 1 + x3

Now we compute the residues inside C. The poles are at z0 = ei/3, ei , ei5/3, so the sum of the residues gives J = 2i = 2i lim (z ei/3) ln z ln z ln z + lim (z ei) + lim (z ei5/3) 1 + z3 1 + z 3 zei5/3 1 + z 3 zei

zei/3

i/3 i i5/3 + + (ei/3 ei)(ei/3 ei5/3) (ei ei/3)(ei ei5/3) (ei5/3 ei)(ei5/3 ei/3) = 2 2i4/9 3 /2 = i 24 3 /9 = 2iI I = 2 3 /9 Alternative method We can evaluate the original complex integral J =
C

dz/(1 + z 3) over a dierent contour:

R C3 2/3 C1

C2

Along C2, the integral vanishes. Along C1, the integral is I. Along C3, let z = re2i/3, therefore dz = dre2i/3 and z 3 = r 3. So dz = 1 + z3
0

C3

dr 2i/3 e = e2i/3 I 1 + r3

The residue at the pole inside the contour is 1/3e2i/3, so we have J = 1 e2i/3 I = 2i/3e2i/3 20

But ei/3(ei/3 ei/3) = ei/3( 2i)sin /3 = iei/3 3 , so I = 2/ 3 as before. Incidentally, the integral is not that hard even without the use of complex numbers (hint: using the known real root of 1 + z 3 = 0, break up the fraction into a sum of three fractions, which can all be integrated using the Formula Book after some variable substitutions).

21

Lecture 6: Cauchy Principal Value


We have been talking about integrals such as

f (x)dx

without worrying too much about whether the integral is well dened (i.e. exists) or not. Suppose that f (x) = x, then clearly
x1 x1 x2

lim

lim

xdx =
x2

x2/2

x1 x2

does not exist. However,


R R

lim

xdx =
R

lim

x2/2

R R

=0

does exist. The key point is that the two limits are taken towards innity in a controlled way, namely at the same rate. Suppose we want to integrate a function on the real line, which has a pole on the path of inte4 gration, e.g. I = 1 dx/(x 2). Clearly, the integrand diverges at x = 2, so with our denitions thus far, the integral does not exist. But is there a consistent way of extending the denition of an integral to allow the above to exist? The integrand is an odd function around the pole, so one gets the feeling that the positive and negative divergences should cancel once we integrate. y

1/(x 2) 1 2 4 x

The way to formalize this feeling is to break up the integral into two parts:
4

I =
1

dx lim x2 0

2 1

= lim

0 0

= lim (ln + ln 2 ln ) = ln 2

4 1 [ln(2 x)]2 + [ln(x 2)]2+

dx + x2

4 2+

dx x2

Notice that we reversed the direction of integration in the rst integral in order to be able to properly evaluate it. The crucial point is that the limit process again was the same for both integrals. The notation for the principal value is
4

P.V.
1

dx = ln 2 x2

Suppose we wanted to evaluate the integral using contour integration, using the residue theorem. Again, we break up the integration path into parts, and connect the two segments by a small semicircle: 22

C1 a

z0 C3

C2 b

I = P.V.
a

f (z)dz =
C1

f (z)dz +
C2

f (z)dz

=
C1 +C2 +C3

f (z)dz

f (z)dz
C3

We need to evaluate the integral over the small semicircle, so put z = z0 + ei and expand f (z) about the point z0 in a Laurent series,

f (z) =
n=1

an(z z0)n

Then
0

f (z)dz =
C3

id
n=1

an n ei nei = ia1

as 0

Note that if the small semicircle is taken in the upper half plane, the integral over it is ia1. To calculate the integral, we close the contour by a large semicircle in the upper half plane, C4, so the total integral around the contour will be equal to 2i (residue at z0). f (z)dz = 2a1
C

I = P.V.
C1 +C2

f (z)dz =
C

f (z)dz

C3

f (z)dz
C4

f (z)dz
C4

= 2ia1 ia1 = ia1


C4

f (z)dz

f (z)dz

Note that if we took the small semicircle in the upper half plane, then the total contour would have contained no poles, so the integral around it would be zero, but the contribution around C3 is ia1 and thus the result is the same.

Summation of series
Let f be a function that is analytic at z = n for integer n. The function cot(z) has simple poles at z = n, and it is easy to check that the residue of f (z)cot(z) at z = n is f (n)/. Let us consider a contour which is a square of side 2N + 1 centered on the origin:

N N +1

CN

23

First we show that cot(z) is bounded on the contour, with the bound independent of N . On the horizontal sides, |cot(z)| = = = eiz + eiz eiz eiz |eiz | + |eiz | |eiz | |eiz | ey + ey ey ey coth|y | coth(/2) since |y | > 1/2
1

On the vertical sides, we have z = N + 2 + it, |cot(z)| = |tan(i t)| due to phase shift by (N + 1/2) 1 e2t = 1 1 + e2t Hence there is a constant M such that |cot(z)| M for any z on any CN . Now suppose that f (z) is such that for large enough |z |, |f (z)| A/z 2, and consider the integral of f (z)cot(z) on CN . The magnitude of the integral is less than the maximum of the integrand multiplied by the length of the contour (this can be shown rigorously), f (z) cot(z)dz
CN

A M 4(2N + 1) N2 0 as N

But the integral is also the 2i the sum of the residues of the enclosed poles. Taking f (z) as 1/z 2, the residues are n 0 are 1/n2 and at the origin the residue is /3. So

0 = (sum of residues) = /3 + Hence

1 n2

(n

0)

n=1

2 1 = 2 6 n

where the extra factor of two arises because the sum over residues ranges from to , whereas the series is only over positive n.

24

Lecture 7: Conformal mapping and uid ow


Let us recall a few facts from the theory of uid dynamics. We are concerned with a uid of density (x), moving with a velocity v(x). The continuity equation (conservation of mass) is expressed by + (v) = 0 t Let us specialize to uids that are incompressible, so we get + v = v = 0 t Furthermore, at high Reynolds number viscosity can be neglected, resulting in irrotational ow , i.e. v = 0. In this case, we can dene a potential , , such that v = , simply by integrating v,
x

(x) =

v ds

and the denition is independent of the integration path (starting at a suitable origin) due to the vanishing curl (via Greens Theorem). Substituting into the continuity equation gives () = 2 = 0 i.e. Laplaces equation. Therefore, nding the velocity eld of a uid ow in a given situation is equivalent to solving Laplaces equation subject to appropriate boundary conditions. The typical no slip condition on a solid boundary is expressed by v = = 0. However, for inviscid ow, this is inappropriate (why?), and the no penetration condition is used instead, i.e. that the velocity component perpendicular to the surface is zero at the boundary of the solid. How can complex analysis help here? Remember the CauchyRiemann equations, if f (z) = (x, y) + i(x, y) is an analytic function, then = x y Dierentiating further, we get 2 2 2 2 = = = 2 2 x xy yx y 2 2 2 + = =0 x2 y 2 and = y x

Similarly, 2 = 0, i.e. both the real and imaginary parts of an analytic function are harmonic, they are solutions of Laplaces equation. So solving Laplaces equation in two dimensions is equivalent to nding an analytic function that satises the appropriate boundary conditions. (And we leave the issues of uniqueness in all this up to the mathematicians). Now let us identify the real part of f with the potential of the inviscid, incompressible ow, vx = vy x = y 25

Using the fact that the derivative of an analytic function is independent of the direction, we have f (z) = +i = i x x y x = vx iv y

so the derivative of f gives us the velocity eld. The = constant curves are equipotentials, and are perpendicular to the velocity eld. But again using the CauchyRiemann equations, = = vy x y = = vx x y So is perpendicular to , therefore the = constant curves are perpendicular to the equipotentials and are along the velocity eld. These are streamlines, is the stream function. Test particles released into the ow travel along the streamlines.

Conformal mapping
Suppose we have a solution to Laplaces equation 2 = 0 and a corresponding analytic function f (z) = + i. Let us introduce an analytic mapping z = F (w), with an inverse w = G(z), where w = u + iv. Lemma 4. 2 (x,y) = 0
2 (u,v) = 0

Proof. Write f as a function of w, f (z) = f (F (w)). Then df d f dz f (z) = = dw dz dw G (z) Now f (z) and G(z) are analytic, so f (w) is an analytic function everywhere except where G (z) = 0 or . Therefore the real (and imaginary) parts of f (w), that is (w) and (w) are harmonic. Hence f (w) also represents a valid solution to Laplaces equation. The above equation relates the velocity elds in (x, y) space to that in (u, v) space, vu ivv = 1 v iv df = f (z) = x y G (z) G (z) dw

Note that f (z) = f (F (w)) implies that the potential value at point z is the same as the potential value at the mapped point w by construction, but the corresponding velocity eld is dierent. A very useful class of transformations are called Mbius mappings, given by w= az + b , cz + d a, b, c, d C, ad bc 0

which maps circles and lines into circles and lines (but can map circles into lines and vice versa). For special values of a, b, c, d, this transformation can realise i. Translation: w=z+b 26

ii. Rotation:

w = eiz

iii. Magnication: w = rz iv. Inversion: w = 1/z

Example: Joukowski aerofoil Amazingly, the following transformation maps a circle (with its origin slightly shifted) to a pretty good approximation of an aircraft wing cross section: w2 = w+2 or equivalently, w =z+ 1 z z 1 z +1
2

So having the solution of Laplaces equation with boundary conditions on a circle (see the example sheet) immediately gives the solution corresponding to the aerofoil. Before the advent of modern computers, this is how aircraft wings were designed! By adjusting the transformation slightly or moving the origin of the circle, one can get dierent aerofoil shapes, with dierent drag and lift, and therefore it is possible to optimise the shape.
1.5

0.5

0.5

1.5 2.5

1.5

0.5

0.5

1.5

2.5

Check out http://www.av8n.com/irro/java/conformi1_1.html for an interactive demo!

27

You might also like