The Cooper Union

Department of Electrical Engineering

ECE114 Digital Signal Processing
Lecture Notes: Complex Analysis
September 6, 2011

Complex Numbers
z = x + jy = rej where ej = cos + j sin . Note ej2n = 1 i¤ n is an integer.
conjugate: z = z = x jy = re j
jzj2 = zz , =argument of z = arg (z)
The unit circle is jzj = 1.
log z = ln jzj + j arg (z) = ln r + j
Actually, log z is multiple valued: log z = ln r + j + j2n , n= integer.
Note the natural logarithm is written ln for real parameters and log for complex. In
MATLAB, log, abs and angle return the (complex or real) natural log, magnitude and
argument, respectively.
If z, w are complex: z w = ew log z . Thus in general z w is multiple valued, unless w = k =
integer: z k is single valued.
Example 1 j j = ej(j =2+j2n ) = e =2 e 2n , so j j has 1 many values, all positive real, with
values arbitrarily small (n ! +1) and arbitrarily large (n ! 1).
z 1=N = e(ln r+j +j2n )=N = r1=N ej =N ej2 n=N , which yields N distinct values for 0 n
j2 n=N th th
N 1. The N values e are the distinct N roots of unity ( is an N root of unity if
N
= 1), and are equally spaced around the unit circle. The twiddle factor is WN = e j2 =N
is de…ned in the context of DFT/FFT. It is a primitive N th root of unity, so-called because
its powers WNk , 0 k N 1, are all the distinct N th roots of unity.
=4 1=3
Example 2 8ej = 2ej =12
1; ej2 =3
; e j2 =3

Analytic Functions and Singularities

A complex-valued function of a complex variable f (z) can be expressed as:
f (z) = u (x; y) + jv (x; y)
where u; v are each real-valued functions of two real variables, x; y. The derivative f 0 of f is
de…ned as:
f (z + z) f (z)
f 0 (z) = lim
z!0 z
0 0
If f exists at zo , then f is analytic at zo . If f exists at all points within some region, then f
is said to be analytic in the region. If f is analytic for all z, f is called entire. For example,
polynomials and ez are entire.
The existence of the complex derivative f 0 is a much stronger condition than in the case
of real functions. Speci…cally, as z ! 0 in any way, the limit must always yield the same
answer.

1
Example 3 z , jzj, and most general expressions involving conjugate or magnitude are not
analytic for any z. Let f (z) = z = x jy. First, if z = x purely real:

f (z + x) f (z)
= x= x ! +1
x
But if z = j y is purely imaginary then:

f (z + j y) f (z)
= j y=j y ! 1
j y

Polynomials, ratios of polynomials, exponential and trigonometric functions are analytic

except at points where the denominator goes to zero. For example, tan z = sin z= cos z is
not analytic at z = =2 + n , n= integer.
If f is analytic in a region, it may be possible to extend f beyond this region so it
stays analytic. This is called analytic continuation. The analytic continuation is uniquely
determined by the values on the boundary.
P
Example 4 f (z) = 1 n 1
n=0 z is analytic for jzj < 1, but the series does not converge for
jzj 1. However, we can use the formula 1= (1 z) to perform analytic continuation to all
z 6= 1. That is, f^ (z) = 1= (1 z) has the property that f^ = f , jzj < 1, P
and f^ is de…ned and
analytic for jzj 1, z 6= 1 as well as for jzj < 1. For example, f (2) = 1 n
n=0 2 is nonsense
but the analytic continuation f^ (2) = 1 1 2 = 1 is perfectly valid.

f (z) = u (x; y) + jv (x; y) is analytic at z i¤ u; v satisfy the Cauchy-Riemann (C-R)

equations:
@u @v @u @v
= , =
@x @y @y @x

De…nition 5 singularity: If f is analytic in a region except at an isolated point zo , then

f has a singularity at z = zo .

Theorem 6 types of singularities: There are only three types of singularities, which can
be classi…ed according to the behavior of limz!zo jf (z)j:

1. removable singularity: limz!zo f (z) exists. If f (zo ) is set equal to this limit, f becomes
analytic there. For example, f (z) = sin z=z is analytic at z = 0 if we de…ne f (0) = 1,
so z = 0 is a removable singularity.

1. pole: limz!1 jf (z)j = 1. Then f must blow up with polynomial growth: f (z) =
g (z) = (z zo )m where g is analytic at zo , g (zo ) being nonzero …nite. Equivalently:
8
< 0, k > m
lim (z zo )k f (z) = …nite, not 0, k = m
z!zo :
blows up, k < m

The multiplicity of the pole is m.

1
Here, “convergence” of a series or integral ALWAYS refers to absolute convergence.

2
 2. essential singularity: limz!zo jf (z)j does not exist: it is neither a …nite constant nor
1. For example, e1=z ! 0 if z = x, x < 0, x ! 0, and e1=z ! 1 if z = x, x > 0,
x ! 0, but e1=z = 1 if z = jy. Thus, limz!0 jf (z)j does not exist.

Theorem 7 Behavior at an essential singularity (Picard’s theorem): If zo is an

essential singularity, then in any arbitrarily small neighborhood of zo , f takes on all complex
values in…nitely often, except for exactly one value.

Example 8 f (z) = e1=z has an essential singularity at z = 0. For all > 0 and all w 6= 0,
we can …nd jzj < such that f (z) = w. Setting e1=z = w, w =6 0, we get:
1
z = 1= log w =
ln jwj + j + j2n

By picking jnj large enough we can …nd a z with arbitrarily small magnitude. Note e1=z 6= 0
for any z.

To repeat, these are the only possibilities! If f is analytic in a region except for poles,
it is called homomorphic. If f (z) = (z zo )m g (z) where g is analytic at zo and g (zo ) 6= 0,
then f has a zero at zo of multiplicity m. That is:
8
f (z) < 0, k < m
lim = …nite, not 0, k = m
z!zo (z zo )k : blows up, k > m

Similar to the concept of multiplicity of poles is the multiplicity of zeros.

Problem 9 Theorem 10 zero of an analytic function: If f is analytic at z0 and

f (z0 ) = 0, then either f (z) is identically zero everywhere, or f (z) = (z z0 )m g (z) where
g is analytic at z0 and g (z0 ) 6= 0; m is called the multiplicity of the zero.

In complex analysis, 1 is considered as a single point. The expression z ! 1 means

that jzj ! 1, the behavior of the arg z can be arbitrary. We say f is analytic at 1, has
poles or zeros at 1, etc., according to its behavior as jzj ! 1. Speci…cally:

Theorem 11 behavior at 1: We say f (z) is analytic at 1 if f (z) is analytic for all

…nite jzj > R, for some …nite radius R, and limz!1 f (z) exists (in this case we denote the
limit as f (1)). The function f has a zero at 1 with multiplicity m if it is analytic there
and f (z) = g (z) =z m where g is analytic at 1 and g (1) 6= 0; if f (1) = 0, then either
there is …nite multiplicity or f is identically zero. The function f has a pole at 1 if it is
analytic for all …nite jzj > R, for some …nite radius R, and limz!1 jf (z)j = 1; in this
case, f (z) = z m g (z) where g is analytic at in…nity and g (1) 6= 0, and m is called the
multiplicity. If f is analytic for all jzj > R for some …nite R but limz!1 jf (z)j does not
exist and is not 1, then f has an essential singularity at 1, and it satis…es the behavior
described in Picard’s theorem (i.e., it takes on all …nite complex values except exactly one in
every region jzj > R).

3
Example 12 Consider:
z3 + 1 z3 + 1 z3 + 1
f1 (z) = ez , f2 (z) = , f3 (z) = 3 , f4 (z) = 5
2z + 3 2z + 3 2z + 3
Then f1 has an essential singularity at 1 (f1 is still called entire); f2 has a double pole at
1 (it blows up like z 2 ; f2 (z) =z 2 ! 1=2 as z ! 1); f3 and f4 are both analytic at 1, with
f4 having a double zero there (it dies out like 1=z 2 ; z 2 f4 (z) ! 1=2 as z ! 1).
A ratio of polynomials is a rational function and is uniquely determined by its poles,
zeros and a constant gain factor:
an z n + + a1 z + a0 (z z1 ) (z zn )
f (z) = =K
bm z m + + b1 z + b0 (z p1 ) (z pm )
If we count poles or zeros at 1 (corresponding to the cases n > m or n < m, respectively),
and count multiplicities, then the total number of poles = the total number of zeros =
max (n; m).

Power Series and Laurent Series

A power series:
X
1
f (z) = cn (z zo )n
n=0
P1
is absolutely convergent in a region if n=0 jcn j jz zo jn < 1 at all points z in the region.
f is analytic in the region of absolute convergence, and conversely if f is analytic at zo then
it has such a power series representation.2
n
In particular, if f is analytic at zo (f 0 exists) then all derivatives f (n) = ddznf exist at zo !!
In fact, cn = n!1 f (n) (zo ).
A real-valued function f (x) is called analytic if it is equal to such a power series. Note
2
that even if f (n) exists for all n it may not be analytic! For example, f (x) = e 1=x , x 6= 0,
2
f (0) = 0. Then f (n) = (polynomial in x 1 ) e 1=x ! 0 as x ! 0, so f (n) (0) = 0 for all n.
This suggests a power series with all terms 0, but f (x) 6= 0!!! Hence, f is not analytic at 0
even though it has in…nitely many derivatives.
If f (x) is a real analytic function then it can be extended to f (z) as an analytic function;
2
that is, the power series in x will work with complex z. Note that e 1=z in fact has an
2 2
essential singularity at z = 0 (if z = jy ! 0, e 1=z ! 1, but if z = x ! 0, e 1=z ! 0).
More generally, a Laurent series is a power series with both positive and negative powers:
X
1
f (z) = cn (z zo )n
n= 1
P
It is analytic in the region of convergence (ROC), which is z for which 1
n= 1 jcn j jz zo jn <
1. The ROC, if it exists, has the form of an annular ring: R1 < jzj < R2 .
2
The term holomorphic is often used interchangeably with analytic. In fact, holomorphic means specif-
ically a function of a complex variable that is di¤erentiable, and analytic is more generally any function
that can be represented by an absolutely convergent power series (e.g., real valued or matrix valued analytic
functions). A basic theorem in complex analysis is that a complex function is holomorhpic i¤ it is analytic.

4
 2
Example 13 cn = en : the series does not converge for any z
P P 1
Example 14 f (z) = 1 n 1
n=0 z = 1 z for jzj < 1; f (z) =
n 1
n= 1 z = 1 z for jzj > 1.
To see the second case:
X 1 X1
z 1 1
n 1
z = z z m= 1
= , for z 1 < 1
n= 1 m=0
1 z 1 z

Hence, the function 1 1 z has two di¤erent Laurent series corresponding to two distinct ROCs.
Note that the ROCs are bounded by the circle jzj = 1 which passes through the pole at z = 1.
In general if f (z) has various poles, it has several di¤erent Laurent series representations
with ROCs bounded by circles passing through the poles. Two Laurent series for the same
function cannot have partially overlapping ROCs: either the ROCs do not overlap at all,
or the two series have the exact same ROC and in fact are the same series (have identical
coe¢ cients).

Contour Integration
Let be an arc or contour, not necessarily closed, with f analytic on . Then the contour
integral of f on is:
Z XN
f (z) dz = lim f (zi ) zi
j zi j!0
N !1 i=1
where zi lie Ron and zi = zi zi 1 . This contrasts with the line integral against an arclength
parameter, f (x; y) ds in that the weight function is ds = jdzj, i.e. the di¤erential arclength
s = j zj.
If = C is a simple closed curve, we take the direction of integration to be counterclock-
wise (ccw) by default, and denote it as:
I
f (z) dz
C
Clockwise integration can be replaced by minus the counterclockwise integral.
H
Theorem 15 f is analytic in a region (without holes) i¤ C f (z) dz = 0 for all simple closed
curves C in the region.
In fact, the Cauchy-Riemann equations, on the one hand, and this integral theorem, on
the other hand, can be shown to be equivalent via Green’s Theorem. H
If f is analytic on a simple closed curve C, but not necessarily inside, then C f (z) dz is
not necessarily zero. The residue of f about C is:
I
1
resf = f (z) dz
C 2 j C
If f has a singularity at z0 , then the residue of f at z0 is:
I
1
resf = f (z) dz
z0 2 j C
where C is a circle centered about z0 of su¢ ciently small radius such that f is analytic in
and on C except for the singularity at z0 .

5
Theorem 16 If CH can be continuously
H deformed to C 0 without passing through any singu-
larities of f , then C f dz = C 0 f dz. If f is analytic on C and inside except for singularities
at z1 , z2 , , zN , then:
I XN
1
f (z) dz = resf
2 j C i=1
zi

Lemma 17 If C is a simple closed curve containing zo :

I
1 1 1, n=0
n+1 dz = [n] = 0, otherwise
2 j C (z zo )

Proof Without loss of generality, let C be a circle of radius R about zo : z = zo + R ej ,

: 0 ! 2 , dz = jR ej d . Therefore:
I Z 2 Z
1 dz 1 jR ej d R n 2
= = e jn d = (n)
2 j C (z zo )n+1 2 j =0 Rn+1 ej(n+1) 2 =0

P
Theorem 18 (Cauchy Integral Formula) If f (z) = 1 n= 1 cn (z zo )n and C is a sim-
ple closed curve encircling zo within the ROC of the series then:
I
1 f (z)
cn = dz
2 j C (z zo )n+1

As a special case, if f is analytic in and on C and zo is inside C then:

I
1 f (z)
f (zo ) = dz
2 j C z zo
and for n = 1; 2; 3; : I
(n) n! f (z)
f (zo ) = dz
2 j C (z zo )n+1
g(z)
Corollary 19 Let f (z) has an mth order pole at zo , that is f (z) = (z zo )m
where g is
analytic at zo . If zo is a simple pole (m = 1) then:

resf = g (zo )
zo

and if m > 1 then:

1
resf = g (m 1)
(zo )
zo (m 1)!
Thus, we can evaluate integrals by computing residues.

6
Maximum Principle and Other Properties
There are restrictions on the locations of extrema of analytic functions. The basic result is:

Theorem 20 (Maximum Principle) Let f be analytic in and on C. Then maxz in C jf (z)j

occurs on C. If the maximum value is also achieved at a point inside C, then f is constant.

The following result can be derived from the maximum principle.

Corollary 21 (Rouche’s Theorem) If f and g are analytic inside and on C, and jf j > jgj
on C (strict!), then f and f +g have the same number of zeros (counting multiplicities) inside
C.

Example 22 z 6 + 3z 2 + 1: let f = 3z 2 , g = z 6 + 1. For jzj = 1, jf j = 3, jgj 1 + 1 = 2, so

jf j > jgj. Thus, f and f + g have the same number of zeros in jzj < 1. That is, z 6 + 3z 2 + 1
has exactly two zeros in jzj < 1, and therefore 4 zeros in jzj > 1. For z 6 + 2z 2 + 1, this test
fails: j2z 2 j jz 6 + 1j for jzj = 1, namely at z = 1 these two terms are equal.

Let C be a simple closed curve, f analytic on C with no poles or zeros on C. As z winds

around C once (in the ccw direction), let w = f (z) trace out a curve in the w-plane. The
winding number N of is the number of times encircles the origin; N > 0 if ccw, N < 0
if cw (clockwise).

Theorem 23 (Cauchy’s Principle of the Argument) If f is analytic on C with no ze-

ros or poles on C, having winding number N , Z zeros in C and P poles in C, then:

N =Z P

Proof. Let w = f (z) trace out a curve in the w-plane as z winds around C once in the
ccw direction. We …rst show:
I I I
1 1 1 dw
N= d arg w = d log w =
2 2 j 2 j w

Note log w = ln jwj + P j arg (w) and closes on itself; as winds around the origin, ln jwj
stays single valued so ln jwj = 0, whereas the argument increments by 2 for each ccw
encirclement
P of the origin, and by 2 if cw. That is, as w winds around the origin N times,
arg w = 2 N . Thus:
I I 0
1 1 f (z)
N= d log f (z) = dz
2 j C 2 j C f (z)

Now write f (z) as: Y Y

f (z) = g (z) (z zi )mi = (z pi )ni
where g is analytic with neither poles nor zeros inside or on C, and zi , pi are the zeros and
poles, respectively, of f in C. Then:
X X
log f = log g + mi log (z zi ) ni log (z pi )

7
Now: I I
1 1 g 0 (z)
d log g (z) = dz = 0
2 j C 2 j C g (z)
0
since g=g has no poles in C (its poles would be poles of g or zeros of g). Also, if zo is inside
C: I I
1 1
d log (z zo ) = d arg (z zo ) = 1
2 j C 2 C
Hence: I X X
1
N= d log f (z) = 0 + mi ni = Z P
2 j C

The Cauchy Principle of the Argument is the basis of the Nyquist diagram in control
theory.

Conformal Mapping Properties

If we view a complex function w = f (z) as a mapping of points, curves and regions in the
z-plane to points, curves and regions in the w-plane, it is called a conformal mapping. A
conformal mapping preserves oriented angles (if two curves intersect at some angle in the
z-plane, the the corresponding curves in the w-plane intersect at the same angle). It turns
out a mapping is conformal i¤ it is representable by an analytic function.
Here, we are concerned with two issues in conformal mapping: the Schwarz Re‡ection
Principle, and the special class of maps called bilinear transforms.
We …rst de…ne the notion of a symmetric point. The general de…nition is beyond the
scope of these notes. The symmetric point of the complex number z with respect to a line
is the complex point that is the re‡ection of z in the line. The symmetric point is on the
other side of, and equidistant from, the line. The symmetric point of the complex number z
with respect to a circle is the complex point along the same ray emanating from the center
of the circle, and such that the geometric mean of the distances from the center is the radius
of the circle. The notion of a symmetric point can be generalized to a wide variety of curves.
If z lies on the curve C, then its symmetric point with respect to C is itself. We shall only
require the following three cases:

De…nition 24 The following are three special cases of the de…nition of a symmetric point:

1. The symmetric point of s with respect to the real axis is s .

2. The symmetric point of s with respect to the imaginary axis is s.

3. The symmetric point of z with respect to the unit circle is 1=z . In particular, 0 and
the point at 1 form a symmetric pair.

The basic result from complex analysis that we shall rely on is:

Theorem 25 (Schwarz Re‡ection Principle) If an analytic function maps the curve C

onto the curve , then it maps symmetric points of C to symmetric points of .

8
Example 26 If H(s) is a rational function that is real for real s, then H(s) = H (s )
for all complex s. Now, H (s ) is a ratio of polynomials in s whose coe¢ cients are the
conjugates of the coe¢ cients of H (s). Hence, all coe¢ cients of H(s) must be real, and the
poles and zeros occur in complex conjugate pairs. Note that here we have generalized the
relation H( ) = H ( ) for real to the case of complex . Viewed another way, if we de…ne
R( ) = Re (H( )) = (H( ) + H ( )) =2, for real s = , then the analytic continuation of
R to the complex plane is R(s) = (H(s) + H (s )) =2. Indeed, in general, H (s) is not
an analytic function (the conjugate function f (s) = s is not analytic). But the double
conjugation in H (s ) has the e¤ect of conjugating the coe¢ cients of H while leaving s
unconjugated, so that H (s ) is in fact analytic.

A mapping of the form:

az + b
w= (1)
cz + d
is called a linear fractional transform, a bilinear transform or a Möbius transform. If ad bc =
0, the mapping is degenerate, reducing to w = b=d constant. Therefore, we specify that
ad bc 6= 0. When cz + d = 0, we de…ne w = 1; and when z = 1 we de…ne w = a=c. Note
that ad bc = 0 prevents az + b = 0 and cz + d = 0 at the same point z.

Theorem 27 A bilinear transform w = f (z) given by (1) satis…es the following properties:

1. If w = f (z) is de…ned via fa; b; c; dg and v = g(w) is de…ned via fa0 ; b0 ; c0 ; d0 g, then
v = g(f (z)) is a bilinear transform speci…ed by fa00 ; b00 ; c00 ; d00 g where:

a00 b00 a0 b 0 a b
= (2)
c00 d00 c0 d 0 c d

2. It is a one-to-one mapping of the closed complex plane onto itself (the closed complex
plane includes the point at in…nity); thus, for each z, w = az+b
cz+d
exists, and for each w,
a0 w+b0 0 0 0 0
there is a unique point z = c0 w+d0 where the coe¢ cients fa ; b ; c ; d g are found as the
elements of the inverse of the matrix formed by fa; b; c; dg; note ad bc 6= 0 ensures
this matrix is invertible.

3. A circle or a line in the z-plane maps to a circle or a line in the w-plane. Here, a line
can be thought of as a circle which contains the point at 1, so a line or circle is called
a generalized circle.

4. Consider a curve C that cuts the z-plane into disjoint regions 1 , 2 ; for example, a
line has a left side and a right side, and a simple closed curve has an inside and an
outside; then a bilinear transform w = f (z) maps C to a curve in the w-plane which
cuts the plane into disjoint regions 01 and 02 , and moreover either the entire region
0 0 0 0
1 maps to the entire region 1 and 2 to 2 , or 1 maps to 2 and 2 to 1.

5. The region mapping discussed in 4 above obeys the following rule. If we trace out C in
a certain direction and then one region, say 1 , is to the left of the curve. As z follows
C then w follows in a certain direction, and hence one region, say 01 is to the left
of . Then 1 maps to 01 . For example, if C is a circle, then its inside is on the left

9
 as C is traced out in the counterclockwise direction. Suppose is a circle, too. Then
the inside maps to the inside or to the outside depending on whether is traced out
counterclockwise or clockwise, respectively.

6. The bilinear transform is uniquely determined by selecting any three points in the z-
plane and specifying where they map to in the w-plane.

Proposition 28 The most general analytic one-to-one mapping of the unit circle onto itself
is:
z k
w = ej (3)
k z+1
where 2 R and jkj 6= 1. Moreover, if jkj < 1, this maps fjzj 1g onto fjwj 1g, and if
jkj > 1 this maps fjzj 1g onto fjwj 1g. The inverse map is:

e j w+k
z= (4)
k e j w+1
If k = 0, then w = ej z is a special case, and if k = 1, then w = ej z1 is another special
case. The most general analytic one-to-one mapping of the unit circle onto the imaginary
axis is:
z+1
w= (5)
z
where 2 Rn f0g and Re ( ) 6= 0. Moreover, if > 0, then if Re ( ) > 0 this maps
fjzj 1g onto the LHP fRe (w) 0g and if Re ( ) < 0 this maps fjzj 1g onto the RHP
fRe (w) 0g. If < 0, these conditions are reversed. The inverse map is:
w+
z= (6)
w
This represents the most general analytic one-to-one mapping of the imaginary axis onto the
unit circle.

Corollary 29 The most general rational functions that map the unit circle onto itself, unit
circle to imaginary axis, or imaginary axis to unit circle are products of the respective …rst-
order functions in the above proposition.

We recognize mappings of the unit circle onto itself as digital all-pass transfer functions,
and mapping of the imaginary axis to the unit circle as analog all-pass transfer functions.
The Schwarz Re‡ection Principle can similarly be used to characterize real (or zero-phase)
analog and digital transfer functions (i.e., transfers functions that are real valued in the
frequency domain).

10