Chapter 3

Elementary Functions

3.1 Polynomials and Rational Functions

As we indicated in Sec. 2.2, polynomial functions of z are functions of the form
pn (z) = a0 + a1 z + a2 z 2 + · · · + an z n , (1)
and rational functions are ratios of polynomials:
a0 + a1 z + a2 z 2 + · · · + am z m
Rm,n (z) = . (2)
b0 + b1 z + b2 z 2 + · · · + bn z n
The degree of the polynomial (1) is n if the complex constant an is nonzero. The
rational function (2) has numerator degree m and denominator degree n, if am  = 0
and bn  = 0. The analyticity of these functions is quite transparent: polynomials are
entire, and rational functions are analytic everywhere except for the zeros of their
denominators. These families are ideal for launching our survey of the elementary
analytic functions.
To begin, we take a specific polynomial of degree three as a prototype for our
analysis:
p3 (z) = 12 + 10z − 4z 2 − 2z 3 ≡ −2z 3 − 4z 2 + 10z + 12. (3)
We are going to study how p3 (z) is characterized by its zeros and by the values of its
derivatives at a single point.
It is easy to verify (by substitution) that the zeros of p3 (z) are 2, −1, and −3. The
reader is no doubt acquainted with the fact that this implies p3 (z) can be expressed in
factored form as
p3 (z) = −2(z − 2)(z + 1)(z + 3). (4)
A familiar line of reasoning that leads to the factored form goes roughly as follows.
One can always divide a “dividend” polynomial by a “divisor” polynomial to obtain a
The identically zero polynomial is assigned a degree of −∞. This has little significance other
than to avoid “exceptions” (see Prob. 22).

From Chapter 3 of Fundamentals of Complex Analysis with Applications to Engineering, Science, and Mathematics,
Third Edition. Edward B. Saff, Arthur David Snider. Copyright © 2003 by Pearson Education, Inc. All rights reserved.
99
 100 Elementary Functions

“quotient” polynomial and a “remainder” polynomial whose degree is less than that of
the divisor:
dividend = divisor × quotient + remainder . (5)
If z 1 is any arbitrary complex number, then division of pn (z) in (1) by the degree-
one polynomial z − z 1 must result in a remainder of lower degree: in other words, a
constant,
pn (z) = (z − z 1 ) pn−1 (z) + constant , (6)
where the quotient polynomial pn−1 (z) has degree n − 1. But suppose z 1 happens
to be a zero of pn (z). Then by setting z = z 1 in (6) we deduce that the (constant)
remainder is zero. Thus (6) displays how (z − z 1 ) has been factored out from pn (z);
we say pn (z) has been “deflated.” For our prototype p3 (z), factoring out the first zero
z 1 = 2 results in

−2z 3 − 4z 2 + 10z + 12 = (z − 2)(−2z 2 − 8z − 6).

Now if z 2 is a zero of the quotient pn−1 (z) (and consequently of pn (z)) we can
deflate further by factoring out (z − z 2 ), and so on, until we run out of zeros, leaving
us with the factorization

pn (z) = (z − z 1 )(z − z 2 ) · · · (z − z k ) pn−k (z). (7)

Since we knew 3 zeros for p3 (z), Eq. (4) displays its “complete” deflation down
to factors of degree one (and the degree-zero factor −2).

Example 1
Carry out the deflation of the polynomial z 3 + (2 − i)z 2 − 2i z .
Solution. Obviously z 1 = 0 is a zero, and the first deflation is trivial:

z 3 + (2 − i)z 2 − 2i z = z (z 2 + (2 − i)z − 2i).

The quadratic formula (Example 3, Sec. 1.5) provides 2 zeros of the degree-two poly-
nomial:

−(2 − i) ± (2 − i)2 − 4(1)(−2i)
z2, z3 = = −2, i
2
(see Prob. 21, Exercises 1.5, for the square root) and the factored form is

z 3 + (2 − i)z 2 − 2z = z(z + 2)(z − i).

If we contemplate the deflation of the degree-six polynomial p6 (z) = z 6 + z 4 −

4i z 2 − 4z + 4 − 3i, we are confronted with the question as to how we find a zero of
The proper name for this division procedure is the Division Algorithm.
The reader may be aware of the fact that, in theory, deflation can always be continued until
pn−k (z) has been reduced to a constant. This will be discussed shortly.

100
3.1 Polynomials and Rational Functions 101

p6 (z). A logically prior question is: how do we know p6 (z) has any zeros? Recall that
in Sec. 1.1 we observed that x 2 + 1 had no zeros among the real numbers, and that we
had to extend our number system to the complex numbers C to create the zeros ±i.
This raises the chilling possibility that there is a horde of polynomials with coefficients
in C, which don’t have zeros in C, and that we will have to keep extending the number
system.
The hero of the saga is Gauss. In his doctoral dissertation of 1799 he proved
the Fundamental Theorem of Algebra, demonstrating that no further extensions are
necessary:

Theorem 1. Every nonconstant polynomial with complex coefficients has at

least one zero in C.

We immediately conclude that a polynomial of degree n has, in fact, n zeros, since

we can continue to factor out zeros in the deflation process until we reach the final,
constant, quotient. Repeated zeros are counted according to their multiplicities; for
example, z 4 + 2z 2 + 1 = (z − i)2 (z + i)2 has zeros in two points, each of multiplicity
2, and we count them as 4 zeros.
Gauss constructed four proofs of the Fundamental Theorem in his lifetime. The
easiest one is based on complex integration, which we study in Chapter 4; the details
of the proof are postponed to Sec. 4.6.
Returning to the deflation process, with the issue of existence of zeros for the
quotients settled we have a complete factorization of any polynomial

pn (z) = an (z − z 1 )(z − z 2 ) · · · (z − z n ). (8)

The display (8) conveys a lot of information. It demonstrates that a polynomial pn (z)
of degree n has n zeros: no less (if we count multiplicities) and no more (why?); and
that pn (z) is completely determined by its zeros, up to a constant multiple (an ). If
two polynomials of degree n have the same (n) zeros, then they are simply constant
multiples of each other. Furthermore, the factorization shows that z 0 is a zero of p(z)
of multiplicity precisely k if and only if

pn (z) = (z − z 0 )k q(z),

where q(z) is a polynomial with q(z 0 )  = 0 .

Example 2
Show that if the polynomial p(z) has real coefficients, it can be expressed as a product
of linear and quadratic factors, each having real coefficients.
Solution. The nonreal zeros of a polynomial with real coefficients occur in com-
plex conjugate pairs (Prob. 17, Exercises 1.2). So if, say, z 2 is the conjugate of z 1 , we

101
 102 Elementary Functions

can combine (z − z 1 )(z − z 2 ) in (8) to obtain

pn (z) = an (z − z 1 )(z − z 2 ) · · · (z − z n )
 
= an z 2 − (z 1 + z 1 )z + z 1 z 1 · · · (z − z n )
 
= an z 2 − 2(Re z 1 )z + |z 1 |2 · · · (z − z n ).
The pair of complex factors of degree one has been replaced by a real factor of degree
two. Combining the remaining complex factors similarly (and noting that the real
zeros, if any, generate real degree-one factors) we achieve the required display. For
example, recalling the fifth roots of unity, the reader can verify that
z 5 − 1 = (z − 1)(z 2 − 2 cos(2π/5)z + 1)(z 2 − 2 cos(4π/5)z + 1).

Of course the Fundamental Theorem only tells us that there are zeros; it doesn’t
tell us how to find them. We do know how to find all zeros of polynomials of degree
one (trivial!) and two (the quadratic formula). The references at the end of this chapter
describe algorithms for finding zeros of polynomials of degrees three and four. Re-
markably, Abel and Galois ended a centuries-long search by proving that no similar
such algorithm exists for polynomials of degrees five or higher!
Therefore to implement the factored form of a polynomial in general we have to
rely on numerical approximations for the zeros. Newton’s method (Prob. 10, Exer-
cises 2.7) is an excellent zero-finder for any analytic function if a good initial estimate
is available. The references describe other algorithms that are specifically designed for
polynomials.
As motivation for the next topic, consider the following example.
Example 3
Express the polynomial p3 (z) of Eq. (3) in terms of powers of (z − 1) (instead of
powers of z).
Solution. The task is to find coefficients d0 , d1 , d2 , and d3 so that
p3 (z) = 12 + 10z − 4z 2 − 2z 3 (9)
= d0 + d1 (z − 1) + d2 (z − 1)2 + d3 (z − 1)3 . (10)
A brute-force solution would be to expand (10) in powers of z, match the coefficients
with (9), and solve the resulting 4 equations in 4 unknowns. Somewhat craftier would
be to note that z = (z − 1) + 1; call this ζ + 1, and expand p3 (z) = p3 (ζ + 1) in
powers of ζ , replacing ζ by (z − 1) at the end:
p3 (ζ + 1) = 12 + 10(ζ + 1) − 4(ζ + 1)2 − 2(ζ + 1)3
= 12 + 10ζ + 10 − 4ζ 2 − 8ζ − 4 − 2ζ 3 − 6ζ 2 − 6ζ − 2
= 16 − 4ζ − 10ζ 2 − 2ζ 3 ;
p3 (z) = 16 − 4(z − 1) − 10(z − 1)2 − 2(z − 1)3 .
(11)
Abel, Niels Henrik (1802–1829); Galois, Evariste (1811–1832).

102
3.1 Polynomials and Rational Functions 103

From this example we readily generalize to conclude that any polynomial can be
rewritten in powers of (z − z 0 ), for arbitrary z 0 . But there’s an easier way of accom-
plishing this “re-expansion.” First note that the coefficients of p3 (z), as displayed in
(10), are directly expressible in terms of the values of p3 and its derivatives at z = 1.
For example, if we differentiate (10) twice, we eliminate d0 and d1 ; and if we set z = 1
in what’s left, we eliminate d3 ; only d2 (times 2) survives. So from (10) we get via
successive differentiation
p3 (1) = d0 ,
p3 (1) = 1d1 ,
p3 (1) = 2 · 1d2 ,
p3 (1) = 3 · 2 · 1d3 ,
p3(4) (1) = p3(5) (1) = · · · = 0.
The pattern is clear. The coefficient of (z − z 0 )k , in the expansion of a polynomial
pn (z) in powers of (z − z 0 ), is given by its kth derivative, evaluated at z 0 , and divided
by k factorial:
(n)
pn (z 0 ) p (z 0 ) p (z 0 ) pn (z 0 )
pn (z) = + n (z − z 0 )1 + n (z − z 0 )2 + · · · + (z − z 0 )n
0! 1! 2! n!
 pn(k) (z 0 )
n
= (z − z 0 )k . (12)
k!
k=0

This is known as the Taylor form of the polynomial pn (z), centered at z 0 . Of course,
we can compute the Taylor form of p3 (z) centered at 1 directly from (9):
p3 (1) = 12 + 10 · 11 − 4 · 12 − 2 · 13 = 16 = d0 ,
p3 (1) = 10 − 4 · 2 · 11 − 2 · 3 · 12 = −4 = d1 ,
etc. (compare (11)). The “standard” form (3) of the polynomial p3 (z) is, then, its
Taylor form centered at z 0 = 0. Usually one uses the nomenclature Maclaurin form
for the Taylor form centered at 0; the standard form (9) is thus its Maclaurin form.
If the Taylor form of a polynomial starts with a term of degree one
pn (z 0 ) p (z 0 ) pn(n) (z 0 )
pn (z) = (z − z 0 ) + n (z − z 0 )2 + · · · + (z − z 0 )n ,
1! 2! n!
it clearly signals that z 0 is a zero of the polynomial. For instance, the Taylor form of
p3 (z) centered at its zero z 0 = 2 is
30 32 12
p3 (z) = (0) − (z − 2) − (z − 2)2 − (z − 2)3 . (13)
1! 2! 3!
Remember that 0! = 1.
Brook Taylor (1685–1731) published his discovery in 1715, but it had already been anticipated
by James Gregory (1638–1675) 40 years earlier.
Colin Maclaurin, 1698–1746.

103
 104 Elementary Functions

In fact, not only does the absence of the constant term p3 (2) highlight the fact that
p3 (z) is zero when z = 2, but (13) also provides an alternative demonstration of how
(z − 2) can be factored out. When more of the leading terms in the Taylor form for the
general polynomial pn (z) are missing, it looks like
(k) (k+1) (n)
pn (z 0 ) pn (z 0 ) pn (z 0 )
pn (z) = (z − z 0 )k + (z − z 0 )k+1 + · · · + (z − z 0 )n
k! (k + 1)! n!
 
(k) (k+1) (n)
pn (z 0 ) pn (z 0 ) pn (z 0 )
= (z − z 0 )k + (z − z 0 ) + · · · + (z − z 0 )n−k .
k! (k + 1)! n!
(14)
(k)
Thus we see that if pn (z) has a zero of multiplicity precisely k at z 0 , then pn (z 0 )  = 0,
( j)
while pn (z 0 ) = 0 for 0 ≤ j < k.
Now we direct our attention to rational functions. Since they are ratios of poly-
nomials, all of the previous considerations can be applied to their numerators and
denominators separately. Probably the most enlightening display comes from the fac-
tored form:
am (z − z 1 )(z − z 2 ) · · · (z − z m )
Rm,n (z) = , (15)
bn (z − ζ1 )(z − ζ2 ) · · · · · · (z − ζn )
where {z k } designates the zeros of the numerator and {ζk } designates those of the
denominator. We assume that common zeros have been cancelled. The zeros of the
numerator are, of course, zeros of Rm,n (z); zeros of the denominator are called poles
of Rm,n (z). (Zeros and poles can, of course, be multiple.) Clearly the magnitude of
Rm,n (z) grows without bound as z approaches a pole.

Example 4
Find all the poles and their multiplicities for

(3z + 3i)(z 2 − 4)
R(z) = .
(z − 2)(z 2 + 1)2
Solution. The zeros of the denominator are 2, i, and −i, which are candidates
for the poles. To determine whether they truly are, we factor both numerator and
denominator and cancel common terms:
(3z + 3i)(z 2 − 4) 3(z + i)(z − 2)(z + 2)
=
(z − 2)(z + 1)
2 2 (z − 2)(z − i)2 (z + i)2
3(z + 2)
= .
(z − i)2 (z + i)
Thus we see that the only poles of R(z) are at z = i of multiplicity 2 and z = −i of
multiplicity 1.

A rational function maps neighborhoods of its poles into neighborhoods of infinity, in the sense
of Sec. 1.7.

104
3.1 Polynomials and Rational Functions 105

Knowledge of the poles of Rm,n (z) enables its expression in terms of partial frac-
tions. Some examples of the partial fraction expansion are
3z 2 + 4z − 5 1 1 1
= + + (16)
(z − 2)(z + 1)(z + 3) z−2 z+1 z+3
and, for multiple poles,
4z + 4 −1 8 −7 6
= + + + . (17)
z(z − 1)(z − 2)2 z z − 1 z − 2 (z − 2)2
Note that the partial fraction decomposition focuses on the poles of the rational func-
tion, conveying little information about its zeros.

Theorem 2. If
a0 + a1 z + a2 z 2 + · · · + am z m
Rm,n (z) = (18)
bn (z − ζ1 )d1 (z − ζ2 )d2 · · · (z − ζr )dr
is a rational function whose denominator degree n = d1 + d2 + · · · + dr exceeds
its numerator degree m, then Rm,n (z) has a partial fraction decomposition of
the form
(1)
A0 A(1) A(1)
d1 −1
Rm,n (z) = + 1
+ ··· +
(z − ζ1 )d1 (z − ζ1 )d1 −1 (z − ζ1 )
(2)
A(2)
0
Ad2 −1
+ + ··· +
(z − ζ2 ) 2
d (z − ζ2 )
(r ) (r )
A0 Adr −1
+ ··· + + · · · + , (19)
(z − ζr )dr (z − ζr )
( j)
where the {As } are constants. (The ζk ’s are assumed distinct.)

The reader may have had some experience with partial fractions in the evaluation
of integrals. Advanced applications will be discussed in later chapters.
Before proceeding with the proof of Theorem 2, assume for the moment that every
suitable rational function has a partial fraction decomposition and let us consider how
to find it.
The brute-force procedure consists in rearranging the proposed form (19) over a
common denominator and comparing the resulting numerator, term by term, with the
original numerator of Rm,n (z). This results in a system of linear equations for the
( j)
unknown coefficients {As }. A quicker, more sophisticated method for evaluating the
( j)
{As } is illustrated in the following example.
In most applications, when the numerator degree of a rational function equals or exceeds its
denominator degree, one performs long division to extract the “polynomial” part of the function and
applies the partial fraction decomposition to the residual fraction.

105
 106 Elementary Functions

Example 5
Reproduce the partial fraction decomposition of the rational function in (17).
Solution. The desired form is
(1) (2)
(3) (3)
4z + 4 A0 A0 A0 A1
R(z) = = + + + . (20)
z(z − 1)(z − 2)2 z z − 1 (z − 2)2 z−2
Note that if the proposed display (20) is multiplied by z, canceling the z in the first
(1)
denominator, and then the result is evaluated at z = 0, only the number A0 would
survive; in other words, A(1)
0 = limz→0 z R(z). But this limit is readily evaluated from
the original formula for R(z):
4·0+4
A(1)
0 = lim z R(z) = = −1.
z→0 (0 − 1)(0 − 2)2
Similarly, we read off
4·1+4
A(2)
0 = lim (z − 1)R(z) = =8
z→1 1(1 − 2)2
(3) 4·2+4
A0 = lim (z − 2)2 R(z) = = 6.
z→2 2(2 − 1)

To get A(3) (3)

1 , we multiply (20) by (z − 2) , then we differentiate to kill off the term A0 :
2

 (1) 
d   d A0 (z − 2)2 A(2)
0 (z − 2)
2
(3) (3)
(z − 2) R(z) =
2
+ + A1 (z − 2) + A0
dz dz z z−1
 (1) 
d A0 (z − 2)2 A(2)
0 (z − 2)
2
(3)
= + + A1 .
dz z z−1
(3)
After the differentiation, everything but A1 will still have at least one factor of (z −2),
so
d   d 4z + 4
A(3) = lim (z − 2) 2
R(z) = lim
1 z→2 dz z→2 dz z2 − z
(22 − 2)4 − (4 · 2 + 4)(2 · 2 − 1)
= = −7.

(22 − 2)2

From the method employed in the above example, it can be seen that if Rm,n (z)
can be written in the form (19), then a general expression for the coefficients is

( j) 1 ds
As = lim [(z − ζ j )d j Rm,n (z)]. (21)
z→ζ j s! dz s

This formula will enable us to give the

106
3.1 Polynomials and Rational Functions 107

( j)
Proof of Theorem 2. Let the constants As be defined by formula (21) and
consider, at first, the sum of those partial fractions in (19) that are associated with the
pole ζ1 (the so-called singular part of Rm,n (z) around ζ1 ). We claim that the difference

1 −1 1 −1
A(1)
d d
1
A(1)
s
Rm,n (z) − = Rm,n (z) − s (z − ζ1 )
s
(22)
(z − ζ1 )d1 −s (z − ζ1 )d1
s=0 s=0

has no pole at ζ1 . Writing

1 −1
d
T (z) := A(1)
s (z − ζ1 ) ,
s
(23)
s=0
P(z) := a0 + a1 z + a2 z 2 + · · · + am z m ,
Q(z) := bn (z − ζ2 )d2 (z − ζ3 )d3 · · · (z − ζr )dr ,

the expression (22) becomes

P(z) T (z) P(z) − Q(z)T (z)

− = .
(z − ζ1 ) Q(z) (z − ζ1 )
d 1 d 1 (z − ζ1 )d1 Q(z)
Thus our claim will be proved if we show that the polynomial P − QT has a zero at
ζ1 of order at least d1 , that is

(P − QT )(s) (ζ1 ) = 0, for s = 0, 1, . . . , d1 − 1. (24)

For this purpose observe that (23) is the Taylor form of the polynomial T centered at
ζ1 . Hence
T (s) (ζ1 )/s! = A(1)
s , for s = 0, 1, . . . , d1 − 1.

From the defining formula (21) we also know that

1 (s) P(z)
A(1)
s = g (ζ1 ), where g(z) := (z − ζ1 )d1 Rm,n (z) = .
s! Q(z)
Thus
T (s) (ζ1 ) = g (s) (ζ1 ), for s = 0, 1, . . . , d1 − 1.

But these last equations, when written in the equivalent form (g −T )(s) (ζ1 ) = 0, imply
(24) since

(P − QT ) = Q(g − T )
(P − QT ) = Q(g − T ) + Q (g − T )
(P − QT ) = Q(g − T ) + 2Q (g − T ) + Q (g − T ),

etc. This verifies the claim that the difference (22) has no pole at ζ1 .

107
 108 Elementary Functions

Similarly, we deduce that the difference between Rm,n (z) and its singular part
around ζ2 has no pole at ζ2 , and so on. From this it follows that
1 −1 r −1
A(1) A(r )
d d
s s
Rm,n − − · · · − (25)
(z − ζ1 )d1 −s (z − ζr )dr −s
s=0 s=0

is a rational function with no poles, i.e., a polynomial. But every term in (25) ap-
proaches 0 as |z| becomes unbounded, so this must be the zero polynomial, which
yields equality (19).

EXERCISES 3.1
1. A polynomial p(z) of degree 4 has zeros at the points −1, 3i, and −3i of respective
multiplicities 2, 1, and 1. If p(1) = 80, find p(z).
2. Show that if the polynomial p(z) = an z n +an−1 z n−1 +· · ·+a0 is written in factored
form as p(z) = an (z − z 1 )d1 (z − z 2 )d2 · · · (z − zr )dr , then
(a) n = d1 + d2 + · · · + dr ,
(b) an−1 = −an (d1 z 1 + d2 z 2 + · · · + dr zr ),
(c) a0 = an (−1)n z 1d1 z 2d2 · · · zrdr .
3. Write the following polynomials in factored form:

(a) z 5 + (2 + 2i)z 4 + 2i z 3 (b) z 4 − 16

(c) 1 + z + z 2 + z 3 + z 4 + z 5 + z 6

4. Show that if p(z) = z n + an−1 z n−1 + · · · + a0 is a polynomial of degree n ≥ 1 and

|a0 | > 1, then p(z) has at least one zero outside the unit circle. [HINT: Notice that
the leading coefficient an = 1 and consider the factored form of p.]
5. Write the following polynomials in the Taylor form, centered at z = 2:

(a) z 5 + 3z + 4 (b) z 10 (c) (z − 1)(z − 2)3

6. If p(z) = an z n + an−1 z n−1 + · · · + a0 (an = 0), then its reverse polynomial p ∗ (z)
is given by
p ∗ (z) = an + an−1 z + · · · + a0 z n .
(a) Show that p ∗ (z) = z n p(1/z).
(b) Show that if p(z) has a zero at z 0 (= 0), then p ∗ (z) has a zero at 1/z 0 .
(c) Show that for |z| = 1, we have | p(z)| = | p ∗ (z)|.
7. Prove that if the polynomial p(z) has a zero of order m at z 0 , then p (z) has a zero
of order m − 1 at z 0 .

108
3.1 Polynomials and Rational Functions 109

8. Show that if the polynomials p(z) and q(z) have, respectively, zeros of order m and
k at z 0 , then the product polynomial p(z)q(z) has a zero of order m + k at z 0 .
9. Show that if z 0 is a zero of pn (z) of order d, then for all z sufficiently near z 0 , there
are positive constants c1 and c2 such that c1 |z − z 0 |d ≤ | pn (z)| ≤ c2 |z − z 0 |d .
10. Show that if pn (z) has degree n, then for all z with |z| sufficiently large, there are
positive constants c1 and c2 such that c1 |z|n < | pn (z)| < c2 |z|n .
11. For each of the following rational functions find all its poles and their multiplicities.

3z 2 + 1 z2 + 4
(a) (b)
z 3 (z 2 + 2i z + 1) (z − 2)(z − 3)2
3
2z + 3 2z 2
(c) (d) +
z + 4z + 4
2 z2 + 3z + 2 z + 1
12. Let Rm,n (z) = P(z)/Q(z) and rm,n = p(z)/q(z) be two rational functions each
with numerator degree m and denominator degree n. Show that if Rm,n and rm,n
agree at m + n + 1 distinct points, then Rm,n = rm,n for all z.
13. Use formula (21) to find the partial fraction decompositions of each of the following
rational functions:
3+i 2z + i z
(a) (b) 3 (c) 2
z(z + 1)(z + 2) z +z (z + z + 1)2
5z 4 + 3z 2 + 1
(d) [HINT: First apply the division algorithm to
2z 2 + 3z + 1
obtain a rational function of suitable form.]
14. Show that if the rational function R(z) has a pole of order m at z 0 , then its derivative
R (z) has a pole of order m + 1 at z 0 .
15. (Residue) Let R = P/Q be a rational function with deg P < deg Q. If ζ is a pole
of R, then the coefficient of 1/(z − ζ ) in the partial fraction expansion of R is called
the residue of R(z) at ζ and is denoted by Res(ζ ). Using formula (21), compute
each of the following residues:
2z + 3
(a) Res(i) for R(z) =
(z − i)(z 2 + 1)
z 3 + 4z + 9
(b) Res(−1) for R(z) =
(2z + 2)(z − 3)5
2z 2 + 3
(c) Res(0) for R(z) =
z 2 (z 2 + 2z + i)
z2 − 9
(d) Res(3i) for R(z) =
(z 2 + 9)2
2z 3 + 3
(e) Res(0) for R(z) =
z 3 (z + 1)

109
 110 Elementary Functions

16. Show that if Rm,n (z) is a rational function with numerator degree m and denomina-
tor degree n, then for all |z| sufficiently large, there are positive constants c1 and c2
such that c1 |z|m−n < |Rm,n (z)| < c2 |z|m−n .
17. Show that if p(z) = an (z − z 1 )d1 (z − z 2 )d2 · · · (z − zr )dr , then the partial fraction
expansion of the logarithmic derivative p / p is given by

p (z) d1 d2 dr
= + + ··· + .
p(z) z − z1 z − z2 z − zr

[HINT: Generalize from the formula ( f gh) = f gh + f g h + f gh .]

18. Show that if
d1 d2 dr
R(z) = + + ··· + ,
z − z1 z − z2 z − zr
where each di is real and positive and each z k lies in the upper half-plane Im z > 0,
then R(z) has no zeros in the lower half-plane Im z < 0. [HINT:Write R(z) =
d1 (z−z 1 ) r (z−zr )
|z−z 1 |2
+ · · · + d|z−z 2 . Then sketch the vectors (z − z k ) for Im z k > 0 and
r|
Im z < 0. Argue from the sketch that any linear combination of these vectors with
real, positive coefficients (dk /|z − z k |2 ) must have a negative (and hence nonzero)
imaginary part. Alternatively, show directly that Im R(z) > 0 for Im z < 0.]
19. Show that if all the zeros of a polynomial p(z) lie in the upper half-plane, then the
same is true for the zeros of p (z). [HINT: See Probs. 17 and 18.]
20. Generalize the geometric argument in Prob. 19 to show that if all the zeros of a
polynomial p(z) lie on one side of any line, then the same is true for the zeros of
p (z).
21. When a set of points lies entirely on one side of a straight line, let us say that the
line “shelters” the point set. If z 1 , z 2 , . . . , zr is any finite set of points in the plane,
their convex hull is the common “territory” that is sheltered by every straight line
that shelters the {z j }. Argue that Prob. 20 proves the Gauss-Lucas Theorem: if p(z)
is any polynomial, all the zeros of p (z) lie in the convex hull of the zeros of p(z).
22. Show how exceptions to the statement “the degree of the product of two polynomials
equals the sum of the degrees of each factor” are finessed by the choice (−∞) as
the degree of the identically zero polynomial.

3.2 The Exponential, Trigonometric, and

Hyperbolic Functions
The complex exponential function e z plays a prominent role in analytic function the-
ory, not only because of its own important properties but because it is used to define the

Francois Edouard Anatole Lucas (1842–1891) invented the Tower of Hanoi puzzle.

110
3.2 The Exponential, Trigonometric, and Hyperbolic Functions 111

complex trigonometric and hyperbolic functions. Recall the definition from Chapter
1: If z = x + i y,
e z = e x (cos y + i sin y). (1)
As a consequence of Example 2 in Sec. 2.4, we know that e z is an entire function
and that
d z
e = ez .
dz
The polar components of e z are readily derived from (1):
|e z | = e x , (2)
arg e z = y + 2kπ (k = 0, ±1, ±2, . . .). (3)
It follows that e z is never zero. However, e z does assume every other complex value
(see Prob. 4).
Recall that a function f is one-to-one on a set S if the equation f (z 1 ) = f (z 2 ),
where z 1 and z 2 are in S, implies that z 1 = z 2 . As is shown in calculus, the exponential
function is one-to-one on the real axis. However it is not one-to-one on the complex
plane. In fact, we have the following.

Theorem 3.

(i) The equation e z = 1 holds if, and only if, z = 2kπi, where k is an integer.

(ii) The equation e z 1 = e z 2 holds if, and only if, z 1 = z 2 + 2kπi, where k is
an integer.

Proof of (i). First suppose that e z = 1, with z = x + i y. Then we must have

|e z | = |e x+i y | = e x = 1,
and so x = 0. This implies that
e z = ei y = cos y + i sin y = 1,
or, equivalently,
cos y = 1, sin y = 0.
These two simultaneous equations are clearly satisfied only when y = 2kπ for some
integer k; i.e., z = 2kπi.
Conversely, if z = 2kπi, where k is an integer, then
e z = e2kπi = e0 (cos 2kπ + i sin 2kπ) = 1.
Proof of (ii). It follows from the division rule that
ez1 = ez2 if, and only if, e z 1 −z 2 = 1.

111
 112 Elementary Functions

But, by part (i), the last equation holds precisely when z 1 − z 2 = 2kπi, where k is an
integer.

One important consequence of Theorem 3 is the fact that e z is periodic. In general,
a function f is said to be periodic in a domain D if there exists a nonzero constant λ
such that the equation f (z + λ) = f (z) holds for every z in D. Any constant λ with
this property is called a period of f . Since, for all z,
e z+2πi = e z ,
we see that e z is periodic with complex period 2πi. Consequently, if we divide up the
z-plane into the infinite horizontal strips
Sn := {x + i y| − ∞ < x < ∞, (2n − 1)π < y ≤ (2n + 1)π} (n = 0, ±1, ±2, . . .),
as shown in Fig. 3.1, then e z will behave in the same manner on each strip. Further-
more, from part (ii) of Theorem 3, it follows that e z is one-to-one on each strip Sn . For
these reasons any one of these strips is called a fundamental region for e z .

Figure 3.1 Fundamental regions for e z .

From the identity

ei y = cos y + i sin y,
and its obvious consequence
e−i y = cos y − i sin y,
we deduce, by subtracting and adding these equations, that
ei y − e−i y ei y + e−i y
sin y = , cos y = .
2i 2
These real variable formulas suggest the following extensions of the trigonometric
functions to complex “angles.”

112
3.2 The Exponential, Trigonometric, and Hyperbolic Functions 113

Definition 1. Given any complex number z, we define

ei z − e−i z ei z + e−i z
sin z := , cos z := .
2i 2

Since ei z and e−i z are entire functions, so are sin z and cos z. In fact,

d d ei z − e−i z 1 iz
sin z = = (ie − (−i)e−i z ) = cos z, (4)
dz dz 2i 2i
and, similarly,
d
cos z = − sin z. (5)
dz
We recognize that Eqs. (4) and (5) agree with the familiar formulas derived in calculus.
Some further identities that remain valid in the complex case are listed below:

sin(z + 2π) = sin z, cos(z + 2π) = cos z. (6)

sin(−z) = − sin z, cos(−z) = cos z. (7)
sin2 z + cos2 z = 1. (8)
sin(z 1 ± z 2 ) = sin z 1 cos z 2 ± sin z 2 cos z 1 . (9)
cos(z 1 ± z 2 ) = cos z 1 cos z 2 ∓ sin z 1 sin z 2 . (10)
sin 2z = 2 sin z cos z, cos 2z = cos2 z − sin2 z. (11)

The proofs of these identities follow directly from the properties of the exponential
function and are left to the exercises. Notice that Eqs. (6) imply that sin z and cos z are
both periodic with period 2π.

Example 1
Prove that sin z = 0 if, and only if, z = kπ, where k is an integer.
Solution. If z = kπ, then clearly sin z = 0. Now suppose, conversely, that
sin z = 0. Then we have
ei z − e−i z
= 0,
2i
or, equivalently,
ei z = e−i z .
By Theorem 3(ii) it follows that

i z = −i z + 2kπi,

which implies that z = kπ for some integer k.

113
 114 Elementary Functions

Thus the only zeros of sin z are its real zeros. The same is true of the function
cos z; i.e.,
π
cos z = 0 if and only if z = + kπ,
2
where k is an integer.
The other four complex trigonometric functions are defined by
sin z cos z 1 1
tan z := , cot z := , sec z := , csc z := .
cos z sin z cos z sin z
Notice that the functions cot z and csc z are analytic except at the zeros of sin z, i.e., the
points z = kπ, whereas the functions tan z and sec z are analytic except at the points
z = π/2 + kπ, where k is any integer. Furthermore, the usual rules for differentiation
remain valid for these functions:
d d
tan z = sec2 z, sec z = sec z tan z,
dz dz
d d
cot z = − csc2 z, csc z = − csc z cot z.
dz dz
The preceding discussion has emphasized the similarity between the real trigono-
metric functions and their complex extensions. However, this analogy should not be
carried too far. For example, the real cosine function is bounded by 1, i.e.,
| cos x| ≤ 1, for all real x,
but
e−y + e y
| cos(i y)| = = cosh y,
2
which is unbounded and, in fact, is never less than 1!
The complex hyperbolic functions are defined by a natural extension of their defi-
nitions in the real case:

Definition 2. For any complex number z we define

e z − e−z e z + e−z
sinh z := , cosh z := . (12)
2 2

Notice that the functions (12) are entire and satisfy

d d
sinh z = cosh z, cosh z = sinh z. (13)
dz dz
One nice feature of the complex variable perspective is that it reveals the intimate con-
nection between hyperbolic functions and their trigonometric analogues. The reader
can readily verify that
sin i z = i sinh z, sinh i z = i sin z, cos i z = cosh z, cosh i z = cos z, (14)

114
3.2 The Exponential, Trigonometric, and Hyperbolic Functions 115

and using these formulas one can transform the trigonometric identities (6)–(11) into
their hyperbolic versions. For example, replacing z by i z in the identity (8) yields the
familiar hyperbolic identity cosh2 z − sinh2 z = 1.
The four remaining complex hyperbolic functions are given by
sinh z cosh z 1 1
tanh z := , coth z := , sech z := , csch z := .
cosh z sinh z cosh z sinh z

EXERCISES 3.2
√
1. Show that e z = (1 + i)/ 2 if and only if z = (π/4 + 2kπ )i, k = 0, ±1, ±2, . . ..
2. Let f (z) = e z − (1 + z + z 2 /2 + z 3 /6). Show that f (3) (0) = 0.
100
3. Find the sum k=0 ekz .
4. Let ω (= 0) have the polar representation ω = r eiθ . Show that
exp(log r + iθ ) = ω (base e logarithm).
5. Write each of the following numbers in the form a + bi.
exp(1 + i3π )
(a) exp(2 + iπ/4) (b)
exp(−1 + iπ/2)

(c) sin(2i) (d) cos(1 − i)

(e) sinh(1 + πi) (f) cosh(iπ/2)

6. Establish the trigonometric identities (8) and (9).

7. Show that the formula ei z = cos z + i sin z holds for all complex numbers z.
8. Verify the differentiation formulas (13).
9. Find dw/dz for each of the following.
1
(a) w = exp(π z 2 ) (b) w = cos(2z) + i sin
z
(c) w = exp[sin(2z)] (d) w = tan3 z

(e) w = [sinh z + 1]2 (f) w = tanh z

10. Explain why the function f (z) = sin(z 2 ) + e−z + i z is entire.

 cos z 
11. Explain why the function Re is harmonic in the whole plane.
ez
12. Establish the following hyperbolic identities by using the relations (14) and the cor-
responding trigonometric identities.
(a) cosh2 z − sinh2 z = 1
(b) sinh(z 1 + z 2 ) = sinh z 1 cosh z 2 + cosh z 1 sinh z 2

115
 116 Elementary Functions

(c) cosh(z 1 + z 2 ) = cosh z 1 cosh z 2 + sinh z 1 sinh z 2

13. Show the following.
(a) sin(x + i y) = sin x cosh y + i cos x sinh y
(b) cos(x + i y) = cos x cosh y − i sin x sinh y
14. Prove the following.
(a) ei z is periodic with period 2π .
(b) tan z is periodic with period π .
(c) sinh z and cosh z are both periodic with period 2πi.
(d) tanh z is periodic with period πi.
15. Prove that cos z = 0 if and only if z = π/2 + kπ , where k is an integer.
16. Verify the identity
z2 + z1 z2 − z1
sin z 2 − sin z 1 = 2 cos sin ,
2 2
and use it to show that sin z 1 = sin z 2 if and only if z 2 = z 1 + 2kπ or z 2 =
−z 1 + (2k + 1)π , where k is an integer.
17. Find all numbers z (if any) such that
(a) e4z = 1
(b) ei z = 3
(c) cos z = i sin z
18. Prove the following.
sin z cos z − 1
(a) lim =1 (b) lim =0
z→0 z z→0 z
[HINT: Use the fact that f (0) = lim [ f (z) − f (0)]/z.]
z→0

19. Prove that the function ez is one-to-one on any open disk of radius π .
20. Show that the function w = e z maps the shaded rectangle in Fig. 3.2(a) one-to-one
onto the semi-annulus in Fig. 3.2(b).
21. (a) Show that the mapping w = sin z is one-to-one in the semi-infinite strip

S1 = {x + i y | − π < x < π, y > 0}

and find the image of this strip. [HINT: See Prob. 16]
(b) For w = sin z, what is the image of the smaller semi-infinite strip

S2 = {x + i y | − π/2 < x < π/2, y > 0}?

116
3.2 The Exponential, Trigonometric, and Hyperbolic Functions 117

Figure 3.2 Mapping of a rectangle under w = e z .

22. Prove that for any m distinct complex numbers λ1 , λ2 , . . . , λm (λi = λ j for i = j),
the functions eλ1 z , eλ2 z , . . . , eλm z are linearly independent on C. In other words,
show that if c1 eλ1 z + c2 eλ2 z + · · · + cm eλm z = 0 for all z, then c1 = c2 = · · · =
cm = 0. [HINT: Proceed by induction on m. In the inductive step, divide by one of
the exponentials and then take the derivative.]

23. Below is an outline of an alternative proof that sin2 z + cos2 z = 1 for all z. Justify
each step in the proof.

(a) The function f (z) = sin2 z + cos2 z is entire.

(b) f (z) = 0 for all z.

(c) f (z) is a constant function.

(d) f (0) = 1.

(e) f (z) = 1 for all z.

24. Using only real arithmetic operations, write a computer program that, when given
as input the real and imaginary parts (x, y) of z = x + i y, produces as output the
real and imaginary parts of (a) e z , (b) sin z, and (c) cosh z.

25. The behavior of the function e1/z around z = 0 is extremely erratic; in Sec. 5.6
this point is classified as an “essential singularity.” Find values of z, all located in
the tiny disk |z| ≤ 0.001, where e1/z takes the value (a) i, (b) −1, (c) 6.02 × 1023
(Avogadro’s number), (d) 1.6 × 10−19 (the electronic charge in Coulombs).

117
 118 Elementary Functions

3.3 The Logarithmic Function

In discussing a correspondence w = f (z) we have used the word “function” to mean
that f assigns a single value w to each permissible value of z. Sometimes this fact is
emphasized by saying “ f is a single-valued function.” Of course there are equations
that do not define single-valued functions; for example, w = arg z and w = z 1/2 .
Indeed, for each nonzero z there are an infinite number of distinct values of arg z
and two distinct values of z 1/2 . In general, if for some values of z there corresponds
more than one value of w = f (z), then we say that w = f (z) is a multiple-valued
function. We commonly obtain multiple-valued functions by taking the inverses of
single-valued functions that are not one-to-one. This is, in fact, how we obtain the
complex logarithmic function log z.
So we want to define log z as the inverse of the exponential function; i.e.,

w = log z if z = ew . (1)

Since ew is never zero, we presume that z  = 0. To find log z explicitly, let us write
z in polar form as z = r eiθ and write w in standard form as w = u + iv. Then the
equation z = ew becomes
r eiθ = eu+iv = eu eiv . (2)

Taking magnitudes of both sides of (2) we deduce that r = eu , or that u is the ordinary
(real) logarithm of r :
u = Log r.

(We capitalize Log here to distinguish the natural logarithmic function of real vari-
ables.) The equality of the remaining factors in Eq. (2), namely, eiθ = eiv , identifies v
as the (multiple-valued) polar angle θ = arg z:

v = arg z = θ.

Thus w = log z is also a multiple-valued function. The explicit definition is as follows.

Definition 3. If z  = 0, then we define log z to be the set of infinitely many values

log z : = Log |z| + i arg z

(3)
= Log |z| + i Arg z + i2kπ (k = 0, ±1, ±2, . . .).

All logarithms in this text are taken to the base e and are hereafter abbreviated log or Log. The
notations ln and Ln are not used in this book. Leonhard Euler (1707–1783) conceived the idea that
the logarithm could be extended to negative and complex numbers.

118
3.3 The Logarithmic Function 119

The multiple-valuedness of log z simply reflects the fact that the imaginary part of
the logarithm is the polar angle θ; the real part is single-valued. As examples, consider

log 3 = Log 3 + i arg 3 = (1.908 . . .) + i2kπ,

log(−1) = Log 1 + i arg(−1) = i(2k + 1)π,
log(1 + i) = Log |1 + i| + i arg(1 + i)
1 π 
= Log 2 + i + 2kπ ,
2 4
where k = 0, ±1, ±2, . . ..
The familiar properties of the real logarithmic function extend to the complex case,
but the precise statements of these extensions are complicated by the fact that log z is
multiple-valued. For example, if z  = 0, then we have

z = elog z ,

but
log e z = z + 2kπi (k = 0, ±1, ±2, . . .).
Furthermore, using the representation (3) and the equations

arg z 1 z 2 = arg z 1 + arg z 2 , (4)

z1
arg = arg z 1 − arg z 2 , (5)
z2
one can readily verify that

log z 1 z 2 = log z 1 + log z 2 , (6)

and that
z1
log = log z 1 − log z 2 . (7)
z2
As with Eqs. (5), we must interpret Eqs. (6) and (7) to mean that if particular values
are assigned to any two of their terms, then one can find a value of the third term so
that the equation is satisfied. For example, if z 1 = z 2 = −1 and we select πi to be the
value of log z 1 and log z 2 , then Eq. (6) will be satisfied if we use the particular value
2πi for log(z 1 z 2 ) = log 1.
Recall that in Sec. 1.3 we used the notion of a branch cut to resolve the ambiguity
in the designation of the polar angle θ = arg z. We took Arg z to be the principal value
of arg z, the value in the interval (−π, π], which jumps by 2π as z crosses the branch
cut along the negative real axis. Other branches argτ z resulted from restricting the
values of arg z to (τ, τ + 2π] and shifting the 2π-discontinuities to the ray θ = τ .
Clearly the same artifice will generate single-valued branches of log z. The prin-
cipal value of the logarithm Log z is the value inherited from the principal value of the
argument:
Log z := Log |z| + i Arg z. (8)

119
 120 Elementary Functions

Figure 3.3 Analyticity domain for Log z.

[Notice that we can use the same convention (Log with a capital L) for the principal
value as for the usual (real) value, since Arg z = 0 if z is positive real.] Log z also
inherits, from Arg z, the discontinuities along the branch cut; it jumps by 2πi as z
crosses the negative real axis. However, at all points off the nonpositive real axis,
Log z is continuous, and this fact enables us to prove the next theorem.

Theorem 4. The function Log z is analytic in the domain D ∗ consisting of all

points of the complex plane except those lying on the nonpositive real axis (see
Fig. 3.3). Furthermore,
d 1
Log z = , for z in D ∗ . (9)
dz z

Proof. Let us set w = Log z. Our goal is to prove that, for z 0 in D ∗ and w0 =
Log z 0 , the limit of the difference quotient
w − w0
lim (10)
z→z 0 z − z0
exists and equals 1/z 0 . We are guided in this endeavor by the knowledge that z = ew
and that the exponential function is analytic so that
z − z0 dz
lim = = ew0 = z 0 .
w→w0 w − w0 dw w=w0

Observe that we will have accomplished our goal if we can show that
w − w0 1
lim = lim , (11)
z→z 0 z − z0 w→w0 z−z 0
w−w0

It is convenient here to use form (10) instead of the usual form


lim Log (z 0 + z) − Log z 0 /z.
z→0

The equivalence of these limits can be seen by putting z = z − z 0 .

120
3.3 The Logarithmic Function 121

because the limit on the right exists and equals 1/z 0 . But (11) will follow from the
trivial identity
w − w0 1
= z−z 0 , (12)
z − z0 w−w 0

provided we show that

(a) As z approaches z 0 , w must approach w0 , and

(b) For z  = z 0 , w will not coincide with w0 [so that the terms in Eq. (11) are
meaningful].

Condition (a) follows from the continuity, in D ∗ , of w = Log z. Condition (b)

is even more immediate; if w coincided with w0 , z would have to equal z 0 , since
z = ew . Thus w = Log z is differentiable at every point in D ∗ and hence is analytic
there.

Corollary 1. The function Arg z is harmonic in the domain D ∗ of Theorem 4.

Corollary 2. The real function Log |z| is harmonic in the entire plane with the
exception of the origin. (See Prob. 8.)

Example 1
Determine the domain of analyticity for the function f (z) := Log(3z − i). Compute
f (z).
Solution. Since f is the composition of Log with the function g(z) = 3z − i,
the chain rule asserts that f will be differentiable at each point z for which 3z − i lies
in the domain D ∗ of Theorem 4. Thus points where 3z − i is negative or zero are
disallowed; a little thought shows that these points lie on the horizontal ray x ≤ 0,
y = 13 (see Fig. 3.4). In this slit plane, then, from Eq. (9):

d 1 d 3
f (z) = Log(3z − i) = (3z − i) = .

dz 3z − i dz 3z − i

Other branches of log z can be employed if the location of the discontinuities on

the negative axis is inconvenient. Clearly the specification

Lτ (z) := Log |z| + i argτ z (13)

Advanced readers will observe that the same proof could be applied to any function f (z) that is
analytic and one-to-one around z 0 and for which f (z 0 )  = 0, to conclude that the inverse function is
analytic and has derivative 1/ f (z 0 ).

121
 122 Elementary Functions

Figure 3.4 Analyticity domain for Log(3z − i).

Figure 3.5 Domain of L−π/4 (z).

results in a single-valued function whose imaginary part lies in the interval (τ, τ +2π].
Moreover, the same reasoning used in the proof of Theorem 4 shows that this function
is analytic in the complex plane excluding the ray θ = τ and the origin, and in this
domain,
d 1
Lτ (z) = .
dz z
Figure 3.5 depicts the domain of analyticity for L−π/4 (z). Of course, no branch of
log z is analytic at the origin, which is called a branch point for log z.
Thus far we have used the phrase “branch of log z” in a somewhat informal manner
to denote specific values for this multiple-valued function. To make matters more
precise, we give the following definition.

Definition 4. F(z) is said to be a branch of a multiple-valued function f (z) in

a domain D if F(z) is single-valued and continuous in D and has the property
that, for each z in D, the value F(z) is one of the values of f (z).

For example, Arg z is a branch of arg z and Log z is a branch of log z in the plane
slit along the negative real axis. In this same domain, the function defined by e(1/2) Log z
gives a branch of z 1/2 whose values all lie in the right half-plane.

122
3.3 The Logarithmic Function 123

Example 2
Determine a branch of f (z) = log(z 3 − 2) that is analytic at z = 0, and find f (0) and
f (0).
Solution. The multiple-valued function f (z) is the composition of the logarithm
with the analytic function g(z) = z 3 − 2. Thus, by the chain rule, it suffices to choose
any branch of the logarithm that is analytic at g(0) = −2. In particular, F(z) =
L−π/4 (g(z)) solves the problem. For this choice,

F(0) = L−π/4 (03 − 2) = Log 2 + iπ,

g (0)
F (0) = L−π/4 (g(0))g (0) = = 0.

g(0)

We conclude this section with a word of warning. When complex arithmetic is

incorporated into computer packages, all functions must of necessity be programmed
as single-valued. The complex logarithm, for instance, is usually programmed as our
“principal value,” Log z. This invalidates some identitites, such as Eq. (6), since it is
not true in general that Log z 1 z 2 = Log z 1 + Log z 2 . (See Prob. 3.)

EXERCISES 3.3
1. Evaluate each of the following.

(a) log i (b) log(1√− i)

(c) Log(−i) (d) Log( 3 + i)

2. Verify formulas (6) and (7).

3. Show that if z 1 = i and z 2 = i − 1, then y

Log z 1 z 2 = Log z 1 + Log z 2 .

4. Prove that Log e z = z if and only if −π < Im z ≤ π .

5. Solve the following equations.

(a) e z = 2i (b) Log(z 2 − 1) = iπ

2 (c) e2z + e z + 1 = 0

6. Find the error in the following “proof” that z = −z: Since z 2 = (−z)2 , it follows
that 2 Log z = 2 Log(−z), and hence Log z = Log(−z), which implies that z =
eLog z = eLog(−z) = −z.
7. Use the polar form of the Cauchy-Riemann equations (Prob. 6 in Exercises 2.4) to
give another proof of Theorem 4.
8. Without directly verifying Laplace’s equation, explain why the function Log |z| is
harmonic in every domain that does not contain the origin.

123
 124 Elementary Functions

9. Determine the domain of analyticity for f (z) = Log(4 + i − z). Compute f (z).
10. Show that the function Log(−z) + iπ is a branch of log z analytic in the domain D0
consisting of all points in the plane except those on the nonnegative real axis.
11. Determine a branch of log(z 2 + 2z + 3) that is analytic at z = −1, and find its
derivative there.
12. Find a branch of log(z 2 + 1) that is analytic at z = 0 and takes the value 2πi there.
13. Find a branch of log(2z − 1) that is analytic at all points in the plane except those
on the following rays.
(a) {x + i y | x ≤ 12 , y = 0}
(b) {x + i y | x ≥ 12 , y = 0}
(c) {x + i y | x = 12 , y ≥ 0}

14. Prove that there exists no function F(z) analytic in the annulus D : 1 < |z| < 2
such that F (z) = 1/z for all z in D. [HINT: Assume that F exists and show that for
z in D, z not a negative real number, F(z) = Log z + c, where c is a constant.]
15. Find a one-to-one analytic mapping of the upper half-plane Im z > 0 onto the infi-
nite horizontal strip

H := {u + iv| − ∞ < u < ∞, 0 < v < 1}.

[HINT: Start by considering w = Log z.]

16. Sketch the level curves for the real and imaginary parts of Log z and verify the
orthogonality property discussed in Sec. 2.5.
17. Prove that any branch of log z (cf. Definition 4) is analytic in its domain and has
derivative 1/z.
18. Prove that if F is a branch of log z analytic in a domain D, then the totality of
branches of log z analytic in D are the functions F + 2kπi, k = 0, ±1, ±2, . . ..
[HINT: Use the result of Prob. 17.]
19. How would you construct a branch of log z that is analytic in the domain D consist-
√ points in the plane except those lying on the half-parabola {x + i y : x ≥
ing of all
0, y = x}?
20. Using only real arithmetic operations, write a computer program whose input (x, y)
is the real and imaginary parts of z = x + i y and whose output is the real and
imaginary parts of

(a) Log z (b) L−π/4 (z) (c) L0 (z) (d) L4π (z)

21. Find a counterexample to the rule log (z 1 z 2 ) = log z 1 + log z 2 for the software
system your computer uses.

124
3.4 Washers, Wedges, and Walls 125

3.4 Washers, Wedges, and Walls

Now that we have established that the function log z, when suitably restricted, is ana-
lytic, the discussion in Sec. 2.5 implies that we have a new pair of harmonic functions
(the real and imaginary parts) that we can use to solve steady-state temperature prob-
lems.
The real part, Log |z| (or Log r in polar-coordinate jargon), is constant on circles
centered at the origin (obviously!). So steady-state temperatures, or electrostatic volt-
ages, or any quantity φ(x, y) that is governed by Laplace’s equation φx x +φ yy = 0 and
is constant on concentric circles must vary logarithmically between the circles. Exam-
ple 1 can be interpreted as describing a thermally conducting “heat pipe” separating
two fluids at different temperatures.

Example 1
Find a function φ(x, y) that is harmonic in the washer-shaped region between the
circles |z| = 1 and |z| = 2 and takes the values φ = 20 on the inner circle and φ = 30
on the outer circle. See Fig. 3.6.

Solution. As in Example 2, Sec. 2.5, we gain some flexibility by noting that

for any value of the constants A and B the function φ(x, y) = A Log |z| + B is
harmonic, and we only have to adjust A and B so as to achieve the prescribed boundary
conditions. We require that

A Log 1 + B = 20 and A Log 2 + B = 30 ,

Figure 3.6 Boundary value problem for a heat pipe.

125
 126 Elementary Functions

so B = 20, A = 10/ Log 2, and the solution is

Log |z|
φ(z) = 10 + 20 .

Log 2
If the circles were centered at z 0 , we would simply use the solution form
A Log |z − z 0 | + B.

Note that we did not worry about the branch cuts that encumber the Log z function
for this application, since we only used the real part, Log |z|, whose values do not
depend on the choice of branch. We need to be a little more fastidious when we use
the imaginary part arg z, which is constant along rays emanating from the origin. The
following example can be interpreted as describing the temperature distribution in a
wedge whose sides are maintained at fixed temperatures.

Example 2
Find a function φ(x, y) that is harmonic in the wedge-shaped region depicted in Fig.
3.7 and takes the values φ = 20 on the upper side and φ = 30 on the lower side.
Solution. Strictly speaking, we cannot employ the form A Arg z + B because the
nonpositive axis, which is the branch cut for Arg z, lies in the region of interest. But
the remedy is obvious: we simply use another branch of arg z, such as arg0 z, whose
discontinuities lie on the positive axis. In other words, we measure the polar angle
θ = arg z from 0 to 2π, instead of from −π to +π. The boundary conditions are
3π 5π
A + B = 20 and A + B = 30 ,
4 4
so A = 20/π, B = 5, and the solution is
20
φ(z) = arg0 z + 5 .

π

Figure 3.7 Boundary value problem for a wedge.

126
3.4 Washers, Wedges, and Walls 127

Figure 3.8 Boundary value problem for upper half-plane.

The more sophisticated boundary value problem depicted in Fig. 3.8 can be re-
garded as an alignment of several 1800 wedges. If we create the function template

φ(x, y) = A1 Arg(z − x1 ) + A2 Arg(z − x2 ) + · · · + An Arg(z − xn ) + B,

which is harmonic in the upper half-plane, we can adjust the constants so as to satisfy
the boundary conditions. An example will make the process clear.

Example 3
Find a function φ(x, y) that is harmonic in the upper half-plane and takes the following
values on the x-axis: φ(x, 0) = 0 for x > 1, φ(x, 0) = 1 for −1 < x < 1, and
φ(x, 0) = 0 for x < −1.
Solution. We enforce the given boundary conditions on the form

φ(x, 0) = A1 Arg(z + 1) + A2 Arg(z − 1) + B.

For z = x > 1, Arg(z − 1) = Arg(z + 1) = 0, so

A1 · 0 + A2 · 0 + B = 0 .

For z = x, −1 < x < 1, Arg(z − 1) = π , Arg(z + 1) = 0, so

A1 · 0 + A2 · π + B = 1 .

For z = x < −1, Arg(z − 1) = Arg(z + 1) = π, so

A1 · π + A2 · π + B = 0 .

Therefore B = 0, A1 = −1/π, A2 = 1/π, and the solution is

1 1
φ(z) = − Arg(z + 1) + Arg(z − 1) .

π π

In closing we note that the simple function φ(x, y) = x (the real part of f (z) = z)
is harmonic and is constant along vertical lines. Thus the format Ax + B can be used

127
 128 Elementary Functions

Figure 3.9 Boundary value problem for a slab.

Figure 3.10 Isotherms.

to find steady state temperatures in a conducting wall, or slab, separating two temper-
ature baths; the solution to the problem depicted in Fig. 3.9 is φ(x, y) = 10x + 10.
The level curves or isotherms for φ for these four geometries are depicted in Fig. 3.10.
Note that the boundary values are discontinuous for the wedge examples; tiny in-
sulators would have to be embedded in the wedge vertices to maintain the temperature
drops. Nonetheless, the solutions in the interiors are smooth functions of position.
This a characteristic of Laplace’s equation; its solutions are, in fact, infinitely differ-
entiable except at the boundaries. It is interesting to note how the isotherms near the
corners form “fan” patterns to mollify the discontinuities.
In Chapter 7 we will see that by using the techniques of complex analysis, one can
often reduce the problem of solving Laplace’s equation in domains with complicated
geometries to solving it in a washer, wedge, or wall. So the tools developed in this
section will be extremely significant.

128
3.4 Washers, Wedges, and Walls 129

EXERCISES 3.4
1. Find a solution to the boundary value problem depicted in Fig. 3.11 and evaluate it
at (0,0).

Figure 3.11 Boundary value problem for Prob. 1.

2. Find a solution to the boundary value problem depicted in Fig. 3.12 and evaluate it
at (0,0).

Figure 3.12 Boundary value problem for Prob. 2.

129
 130 Elementary Functions

3. Find a solution to the boundary value problem depicted in Fig. 3.13 and evaluate it
at (1,1).

Figure 3.13 Boundary value problem for Prob. 3.

4. Find a solution to the boundary value problem depicted in Fig. 3.14 and evaluate it
at (2,3).

Figure 3.14 Boundary value problem for Prob. 4.

5. Find a general formula for a solution φ(x, y) to the boundary value problem inside
the washer depicted in Fig. 3.15.

6. Find a general formula for a solution φ(x, y) to the boundary value problem inside
the disk depicted in Fig. 3.16. (If you’re not sure, apply the formula derived in the
previous exercise and let the inner radius r1 go to zero. Now don’t you feel foolish?)

7. Prove that the function (arg z)(Log |z|) is harmonic. [HINT: It occurs as the imagi-
nary part of what analytic function?]

130
3.5 Complex Powers and Inverse Trigonometric Functions 131

Figure 3.15 Boundary value problem for Prob. 5.

Figure 3.16 Boundary value problem for Prob. 6.

3.5 Complex Powers and Inverse

Trigonometric Functions
One important theoretical use of the logarithmic function is to define complex powers
of z. The definition is motivated by the identity
z n = (elog z )n = en log z ,
which holds for any integer n.

131
 132 Elementary Functions

Definition 5. If α is a complex constant and z  = 0, then we define z α by

z α := eα log z .

This means that each value of log z leads to a particular value of z α .

Example 1
Find all the values of (−2)i .
Solution. Since log(−2) = Log 2 + (π + 2kπ)i, we have

(−2)i = ei log(−2) = ei Log 2 e−π−2kπ (k = 0, ±1, ±2, . . .).

Thus (−2)i has infinitely many different values.

Using the representations of Sec. 3.3, we can write

z α = eα(Log |z|+i Arg z+2kπi) = eα(Log |z|+i Arg z) eα2kπi , (1)

where k = 0, ±1, ±2, . . .. The values of z α obtained by taking k = k1 and k = k2 ( =

k1 ) in Eq. (1) will therefore be the same when

eα2k1 πi = eα2k2 πi .

But by Theorem 3 of Section 3.2 this occurs only if

α2k1 πi = α2k2 πi + 2mπi,

where m is an integer. Solving the equation we get α = m/(k1 − k2 ); i.e., formula

(1) yields some identical values of z α only when α is a real rational number. Conse-
quently, if α is not a real rational number, we obtain infinitely many different values
for z α , one for each choice of the integer k in Eq. (1). On the other hand, if α = m/n,
where m and n > 0 are integers having no common factor, then one can verify that
there are exactly n distinct values of z m/n , namely,
m   m 
z m/n = exp Log |z| exp i (Arg z + 2kπ) (k = 0, 1, . . . , n − 1). (2)
n n
This is entirely consistent with the theory of roots discussed in Sec. 1.5. In summary,

z α is single-valued when α is a real integer;

z α takes finitely many values when α is a real rational number;
z α takes infinitely many values in all other cases.

It is clear from Definitions 4 and 5 that each branch of log z yields a branch of
z α . For example, using the principal branch of log z we obtain the principal branch
of z α , namely, eα Log z . Since e z is entire and Log z is analytic in the slit domain

132
3.5 Complex Powers and Inverse Trigonometric Functions 133

D ∗ = C\(−∞, 0] of Theorem 4, the chain rule implies that the principal branch of
z α is also analytic in D ∗ . Furthermore, for z in D ∗ , we have

d α Log z d α
(e ) = eα Log z (α Log z) = eα Log z .
dz dz z

Other branches of z α can be constructed by using other branches of log z, and since
each branch of the latter has derivative 1/z, the formula

d α 1
(z ) = αz α (3)
dz z
is valid for each corresponding branch of z α (provided the same branch is used on
both sides of the equation). Observe that if z 0 is any given nonzero point, then by
selecting a branch cut for log z that avoids z 0 , we get a branch of z α that is analytic in
a neighborhood of z 0 .
“Branch chasing” with complicated functions is often a tedious task; fortunately,
it is seldom necessary for elementary applications. Some of the subtleties are demon-
strated in the following example.

Example 2
Define a branch of (z 2 − 1)1/2 that is analytic in the exterior of the unit circle, |z| > 1.

Solution. Our task, restated, is to find a function w = f (z) that is analytic

outside the unit circle and satisfies

w2 = z 2 − 1. (4)

Note the principal branch of (z 2 − 1)1/2 , namely,

e(1/2) Log(z
2 −1)
,

will not work; it has branch cuts wherever z 2 −1 is negative real, and this constitutes the
whole y-axis as well as a portion of the x-axis (see Fig. 3.17). But if we experiment
with some alternative expressions for w, we are led to consider z(1 − 1/z 2 )1/2 as a
solution to (4). The principal branch of (1 − 1/z 2 )1/2 , i.e.,

e(1/2) Log(1−1/z ) ,
2

has cuts where 1 − 1/z 2 is negative real, and this occurs only when 1/z 2 is real and
greater than one—i.e., the cut is the segment [−1, 1], as shown in Fig. 3.18. Thus

w = f (z) = ze(1/2) Log(1−1/z

2)

satisfies the required condition of analyticity outside the unit circle.

133
 134 Elementary Functions

Figure 3.17 Branch cut for e(1/2) Log(z

2 −1)
.

Figure 3.18 Branch cut for e(1/2) Log(1−1/z ) .

2

Now that we have exponentials expressed in terms of trig functions (Sec. 1.4),
trig functions expressed as exponentials (Sec. 3.2), and logs interpreted as inverses of
exponentials, the following example should come as no surprise. It demonstrates that
the arcsine is, in fact, a logarithm.

Example 3
The inverse sine function w = sin−1 z is defined by the equation z = sin w. Show that
sin−1 z is a multiple-valued function given by

sin−1 z = −i log[i z + (1 − z 2 )1/2 ]. (5)

Solution. From the equation

eiw − e−iw
z = sin w = ,
2i
we deduce that
e2iw − 2i zeiw − 1 = 0. (6)
We remark that Eq. (3) can be written in the more familiar form

d α
(z ) = αz α−1
dz

with the proviso that the branch of the logarithm used in defining z α is the same as the branch of the
logarithm used in defining z α−1 .

134
3.5 Complex Powers and Inverse Trigonometric Functions 135

Using the quadratic formula we can solve Eq. (6) for eiw :

eiw = i z + (1 − z 2 )1/2 ,

where, of course, the square root is two-valued. Formula (5) now follows by taking
logarithms.

We can obtain a branch of the multiple-valued function sin−1 z by first choosing a
branch of the square root and then selecting a suitable branch of the logarithm. Using
the chain rule and formula (5) one can show that any such branch of sin−1 z satisfies
d 1
(sin−1 z) = (z  = ±1), (7)
dz (1 − z 2 )1/2
where the choice of the square root on the right must be the same as that used in the
branch of sin−1 z.

Example 4
Suppose z is real and lies in the interval (−1, 1). If principal values are used in formula
(5), what is the range of sin−1 z?
Solution. With principal values, Eq. (5) is realized as

Sin−1 z = −i Log[i z + e(1/2) Log(1−z ) ].

2
(8)

For |z| = |x| < 1, clearly 1 − z 2 lies in the interval (0, 1], and its Log is real; hence
the exponential in (8), which represents (1 − z 2 )1/2 , is positive real. The term i z, on
the other hand, is pure imaginary. Consequently, the bracketed expression in (8) lies
in the right half-plane. As a matter of fact, it also lies on the unit circle, since
 1/2   
i z + 1 − z2 = x2 + 1 − x2 = 1

(see Fig. 3.19). Taking the Log, then, results in values iθ, where −π/2 < θ < π/2,
and the leading factor (−i) in (8) produces
π π
− < Sin−1 x <
2 2
(in keeping with the usual interpretation).

For the inverse cosine and inverse tangent functions, calculations similar to those
in Example 3 lead to the expressions

cos−1 z = −i log[z + (z 2 − 1)1/2 ], (9)

i i +z i 1 − iz
tan−1 z = log = log (z  = ±i) (10)
2 i −z 2 1 + iz
To ensure that our method has not introduced extraneous solutions, one should verify that every
value w given by Eq. (5) satisfies z = sin w.

135
 136 Elementary Functions

Figure 3.19 Points i x + (1 − x 2 )1/2 for |x| < 1.

and the formulas

d −1
(cos−1 z) = (z  = ±1), (11)
dz (1 − z 2 )1/2
d 1
(tan−1 z) = (z  = ±i). (12)
dz 1 + z2
Notice that the derivative in Eq. (12) is independent of the branch chosen for tan−1 z,
whereas the derivative in Eq. (11) depends on the choice of the square root used in the
branch of cos−1 z.
The same methods can be applied to the inverse hyperbolic functions. The results
are
 
sinh−1 z = log z + (z 2 + 1)1/2 , (13)
 
cosh−1 z = log z + (z 2 − 1)1/2 , (14)
1 1+z
tanh−1 z = log (z  = ±1). (15)
2 1−z

EXERCISES 3.5

1. Find all the values of the following.

(a) i i (b) (−1)2/3 (c) 2πi

(d) (1 + i)1−i (e) (1 + i)3

2. Show from Definition 5 that if z = 0, then z 0 = 1.

3. Find the principal value (i.e., the value given by the principal branch) of each of the
following.

(a) 41/2 (b) i 2i (c) (1 + i)1+i

136
3.5 Complex Powers and Inverse Trigonometric Functions 137

4. Is 1 raised to any power always equal to 1?

5. Give an example to show that the principal value of (z 1 z 2 )α need not be equal to the
product of principal values z 1α z 2α .
6. Let α and β be complex constants and let z = 0. Show that the following identities
hold when each power function is given by its principal branch.
zα
(a) z −α = 1/z α (b) z α z β = z α+β (c) = z α−β
zβ
7. Find the derivative of the principal branch of z 1+i at z = i.
8. Show
√ that all solutions of the equation sin z = 2 are given by π/2+2kπ ±i Log(2+
3), where k = 0, ±1, ±2, . . .. √ [REMARK: This solution set can also be repre-
sented as π/2 + 2kπ − i Log(2 ± 3, k = 0, ±1, ±2, . . ..]
9. Derive formulas (9) and (11) concerning cos−1 z.
10. Show √ √ cos z = 2i are given by π/2 + 2kπ −
that all the solutions of the equation
i Log( 5 + 2), −π/2 + 2kπ + i Log( 5 + 2), for k = 0, ±1, ±2, √ . . .. [REMARK:
set can also be represented as π/2 + 2kπ − i Log( 5 + 2), −π/2 +
This solution √
2kπ − i Log( 5 − 2), for k = 0, ±1, ±2, . . .. ]
11. Find all solutions of the equation sin z = cos z.
12. Derive formulas (10) and (12) concerning tan−1 z.
13. Derive formulas (13) and (14) for sinh−1 z and cosh−1 z.
14. Derive the formula d(sinh−1 z)/dz = 1/(1+z 2 )1/2 and explain the conditions under
which it is valid.
15. Find a branch of each of the following multiple-valued functions that is analytic in
the given domain:
(a) (z 2 − 1)1/2 in the unit disk, |z| < 1. [HINT: z 2 − 1 = (z − 1)(z + 1).]
(b) (4 + z 2 )1/2 in the complex plane slit along the imaginary axis from −2i to 2i.
(c) (z 4 − 1)1/2 in the exterior of the unit circle, |z| > 1.
(d) (z 3 − 1)1/3 in the exterior of the unit circle, |z| > 1.
16. According to Definition 5 the multiple-valued function c z , where c is a nonzero
constant, is given by c z = e z log c . Show that by selecting a particular value of log c
we obtain a branch of c z that is entire. Find the derivative of such a branch.
17. Derive the identity
 
1/2
−1 1 1
sec z = −i log + −1 .
z z2

Using principal values determine the range of Sec−1 x when x > 1, and when
x < −1. Compare this with the ranges listed in standard mathematical handbooks:
0 ≤ Sec−1 x < π/2, for x ≥ 1; −π ≤ Sec−1 x < −π/2, for x ≤ −1.

137
 138 Elementary Functions

18. Using only real arithmetic operations, write a computer program whose input (x, y)
is the real and imaginary parts of z = x + i y and whose output is the real and
imaginary parts of
(a) the principal branch of z α
(b) Sin−1 z
(c) Sec−1 z
(d) the branch of (z 2 − 1)1/2 discussed in Example 3.5
19. Determine the inverse of the function

w = q(z) := 2e z + e2z

explicitly in terms of the complex logarithm. Use your formula to find all values of
z for which q(z) = 3.

3.6 *Application to Oscillating Systems

Many engineering problems are ultimately based upon the response of a system to
a sinusoidal input. Naturally, all the parameters in such a situation are real, and the
models can be analyzed using the techniques of real variables. However, the utilization
of complex variables can greatly simplify the computations and lend some insight
into the roles played by the various parameters. In this section we shall illustrate the
technique for the analysis of a simple RLC (resistor-inductor-capacitor) circuit.
The electric circuit is shown in Fig. 3.20. We suppose that the power supply is
driving the system with a sinusoidal voltage Vs oscillating at a frequency of ν =
ω/(2π) cycles per second. To be precise, let us say that at time t
Vs = cos ωt. (1)
Our goal is to find the current Is drawn out of the power supply.

Figure 3.20 R LC circuit.

138
3.6 *Application to Oscillating Systems 139

Across each element of the circuit there is a voltage drop V which is related to the
current I flowing through the element. The relationships between these two quantities
are as follows: for the resistor we have Ohm’s law
Vr = Ir R, (2)
where R is a constant known as the resistance; for the capacitor
d Vc
C = Ic , (3)
dt
where C is the capacitance; and for the inductor
d Il
Vl = L , (4)
dt
where L is the inductance.
One can incorporate formulas (1)–(4), together with Kirchhoff’s laws which ex-
press conservation of charge and energy, to produce a system of differential equations
determining all the currents and voltages. The solution of this system is, however, a
laborious process, and a simpler technique is desirable. The complex exponential func-
tion will provide this simplification. Before we demonstrate the utilization of complex
variables in this problem, we shall make some observations based upon physical con-
siderations.

Figure 3.21 Resistor circuit.

First, notice that if the capacitor and inductor were replaced by resistors, as we
illustrate in Fig. 3.21, the solution would be an elementary exercise in high school
physics. The pair of resistances R and Rc are wired in parallel, so they can be replaced
by an equivalent resistance R|| given by the familiar law
1 1 1
= + ,
R|| R Rc
George Simon Ohm, a professor of mathematics, published this law in 1827.
Gustav Robert Kirchhoff (1824–1887.)

139
 140 Elementary Functions

Figure 3.22 Equivalent resistor circuit.

or
R Rc
R|| = . (5)
R + Rc
This resistance then appears in series with Rl , yielding an effective total resistance
Re f f given by
Re f f = R|| + Rl , (6)
and (from the point of view of the power supply) the circuit is equivalently represented
by Fig. 3.22. Equations (1) and (2) yield the current output Is :
cos ωt cos ωt
Is = = R Rc
.
Re f f
R+Rc + Rl
Since this model is so easy to solve, it would clearly be advantageous to replace
the capacitor and inductor by equivalent resistors.
The second point we wish to make involves the nature of the solution Is for the
original circuit. If the power supply is turned on at time t = 0, there will be a fairly
complicated initial current response; this so-called “transient,” however, eventually
dies out, and the system enters a steady state in which all the currents and voltages
oscillate sinuosoidally at the same frequency as the driving voltage. This “in-synch”
behavior is common to all damped linear systems, and it will be familiar to anyone who
has ridden a bicycle over railroad ties or mastered the art of dribbling a basketball. For
many applications it is this steady-state response, which is independent of the initial
state of the system, that is of interest.
The complex variable technique that we shall describe allows us to replace ca-
pacitors and inductors by resistors, and is ideally suited for finding the steady state
response.
The technique is based upon exploiting Euler’s identity eiωt = cos ωt + i sin ωt,
which enables us to express the power supply voltage cos ωt in terms of the exponen-
tial,
cos ωt = Re eiωt . (7)
In fact, the more general sinusoidal forms a cos ωt + b sin ωt and A cos(ωt + γ ) can
also be expressed in an exponential form Re αeiωt :
a cos ωt + b sin ωt = Re(aeiωt − ibeiωt ) = Re αeiωt , (8)
The widespread adoption of this technique in electrical engineering is traditionally attributed to
Charles Proteus Steinmetz (1865–1923).

140
3.6 *Application to Oscillating Systems 141

where α = a − ib; or
A cos(ωt + γ ) = Re[αeiωt ], (9)
where α = Aeiγ . The complex factor α = a − ib = Aeiγ is known as a phasor in this
context. (See Problem 2 for a generalization.)
The advantage of the exponential representation for these functions lies in its com-
pactness, and in the simplicity of the expression for the derivative; differentiating eiωt
(with respect to t) merely amounts to multiplying by iω. Therefore, the derivative of
the general sinusoid in Eq. (8) can be computed by
d d d
(a cos ωt + b sin ωt) = Re αeiωt = Re (αeiωt )
dt dt dt
= Re iωαeiωt
(= −aω sin ωt + bω cos ωt).
To exploit these advantages, we must make one more observation about the circuit
in Fig. 3.20. Each of the elements is a linear device in the sense that the superposition
principle holds; that is, if an element responds to the excitation voltages V1 (t) and
V2 (t) by producing the currents I1 (t) and I2 (t), respectively, then the response to the
voltage V1 (t) + βV2 (t) will be the current I1 (t) + β I2 (t), for any constant β. This is
a mathematical consequence of the linearity of Eqs. (2)–(4); it will still hold true, in
the mathematical sense, even if the functions take complex values. Furthermore, since
we know that the circuit responds to real voltages V1 (t) and V2 (t) with real currents,
it follows that both I1 (t) and I2 (t) are real, so I1 (t) = Re[I1 (t) + i I2 (t)]. In short, if
the mathematical response to the complex voltage V (t) is I (t), then the response to
the (real) voltage Re V (t) will be Re I (t).
With these tools at hand, let us return to the solution of the problem depicted in
Fig. 3.20. The supply voltage, given in Eq. (1), can be represented by

Vs (t) = Re eiωt . (10)

Our strategy will be to find the steady-state response of the circuit to the complex
voltage eiωt , and to take the real part of the answer as our solution.
From the earlier observations about the nature of the steady-state response and
Eqs. (8, 9), we are led to propose that the sinusoidal current or voltage in any part
of the circuit can be written as a (possibly complex) constant times eiωt ; furthermore,
differentiation of such a function is equivalent to multiplication by the factor iω. Thus,
in the situation at hand, Eq. (3) for the behavior of the capacitor becomes

iωC Vc = Ic , (11)

and Eq. (4) for the inductor becomes

Vl = iωL Il . (12)

But the voltage-current relationships (11) and (12) now have the same form as Eq. (2)
for a resistor; in other words, operated at the frequency ν = ω/2π, a capacitor behaves

141
 142 Elementary Functions

mathematically like a resistor with resistance

1
Rc = , (13)
iωC
and an inductor behaves like a resistor with resistance

Rl = iωL . (14)

These pure imaginary parameters are artifacts of our postulating a complex power-
supply voltage. Engineers have adopted the term “impedance” to describe complex
resistances.
Having replaced, formally, the capacitor and inductor by resistors, we are back in
the situation of the simple circuit of Fig. 3.21. The effective impedance of the series-
parallel arrangement is displayed in Eqs. (5) and (6), and substitution of the relations
(13) and (14) yields the expression
R
iωC
Re f f = + iωL .
R+ 1
iωC

The current output of the power supply is therefore

eiωt
Is = Re ,
Re f f
which after some manipulation can be written as

R cos ωt − [R 2 ωC(1 − ω2 LC) − ωL] sin ωt

Is = . (15)
R 2 (1 − ω2 LC)2 + ω2 L 2
Again complex variables can be used to provide a more meaningful interpretation
of the answer. If we define φ0 and R0 by

φ0 := Arg Re f f , R0 := |Re f f |,

then Is can be expressed as

eiωt 1 i(ωt−φ0 ) 1
Is = Re = Re e = cos (ωt − φ0 ). (16)
R0 eiφ0 R0 R0
This displays some easily visualized properties of the output current in relation to the
input voltage (1). The current is, as we indicated earlier, a sinusoid with the same
frequency, but its amplitude differs from the voltage amplitude by the factor 1/R0 .
Furthermore, the two sinusoids are out of phase, with φ0 measuring the phase dif-
ference in radians. The numbers R0 and φ0 can be computed in terms of the circuit
parameters and the frequency, and the circuit problem is solved.
If the power supply voltage had been a more complicated sinusoid, like a cos ωt +
b sin ωt or A cos(ωt − γ ), we would express it as an exponential Re αeiωt using the

142
3.6 *Application to Oscillating Systems 143

prescription (8) or (9). The complex current response would simply then be multiplied
by the constant (“phasor”) α, and

αeiωt
Is = Re .
Re f f

In conclusion we reiterate the two advantages of using complex notation in ana-

lyzing linear sinusoidal systems. First, the representation of general sinusoids (8) is
compact and leads naturally to a reinterpretation of the sinusoid in terms of amplitudes
and phases, as is evidenced by comparing Eqs. (15) and (16). And second, the process
of differentiation is replaced by simple multiplication. It is important to keep in mind,
however, that only the real part of the solution corresponds to physical reality and that
the condition of linearity is of the utmost importance.

EXERCISES 3.6
1. What is the steady-state current response of the circuit in Fig. 3.20 if the power
supply voltage is Vs = sin ωt?
2. Find a formula for the complex constant (“phasor”) α so that the general sinusoid at
frequency ν = ω/(2π ) ,

a1 cos(ωt + γ1 ) + a2 cos(ωt + γ2 ) + · · · + am cos(ωt + γm )

+ b1 sin(ωt + δ1 ) + b2 sin(ωt + δ2 ) + · · · bn sin(ωt + δn )

is expressed as Re αeiωt .
3. Verify the expression in Eq. (15).
4. Using the techniques of this section, find the steady-state current output Is of the
circuits in Fig. 3.23.
5. In the limit of very low frequencies, a capacitor behaves like an open circuit (infinite
resistance), while an inductor behaves like a short circuit (zero resistance). Draw
and analyze the low-frequency limit of the circuit in Fig. 3.20, and verify expression
(15) in this limit.
6. Repeat Prob. 5 for the high-frequency limit. (What are the behaviors of the capacitor
and inductor in this case?)
7. The operation of synchronous and induction motors requires that a rotating magnetic
flux be established inside the motor; that is, if we regard a cross section plane of
the motor to be the complex z-plane, the motor needs a magnetic flux vector which
rotates like the complex number eiωt . Now a solenoid whose coils are wound around
an axis directed along the complex number α produces a magnetic flux in the interior
of the motor given by I (t)α, where I (t) is the real current in its windings. See
Fig. 3.24.

143
 144 Elementary Functions

Figure 3.23 Electrical circuits for Prob. 4.

(a) Show that three inductors, wound around axes directed along the complex
numbers 1, ei2π/3 , ei4π/3 , and carrying the “3-phase” currents I0 cos ωt,
I0 cos(ωt − 2π/3), I0 cos(ωt − 4π/3) respectively, will produce the desired
rotating flux (by superposition of their individual fields). [HINT:You will find
the algebra easier if you remember

ei(ωt−2π/3) + e−i(ωt−2π/3)
cos(ωt − 2π/3) = .]
2

(b) Would two symmetrically placed 2-phase inductors work? Four? Five? Can
you figure out the general rule?

8. In radio transmission, a carrier radio wave takes the form of a pure sinusoid,
A cos ωt. The carrier wave can transmit an information signal f (t) by using modu-
lation. Two forms of modulation are

144
3.6 *Application to Oscillating Systems 145

Figure 3.24 Rotating magnetic flux. Figure 3.25 Modulation schemes.

Amplitude Modulation: the amplitude A is perturbed by a small multiple (β) of the

information signal, so that the transmitted signal equals [A + β f (t)] cos ωt;
(Narrow Band) Frequency Modulation: the angular frequency ω is perturbed by a
small multiple (β) of the information signal, so that the transmitted signal equals
A cos [ω + β f (t)]dt. (Here the constant of integration is not important, and we
are identifying the “instantaneous angular velocity” associated with the sinusoid
cos g(t) as g (t).)

(a) Show that if f (t) is also a sinusoid cos ωm t, and if β is sufficiently small, the
FM signal approximately equals the real part of eiωt [A + i(Aβ/ωm ) sin ωm t],
while the AM signal (exactly) equals the real part of eiωt [A + β cos ωm t].
[HINT: sin α ≈ α and cos α ≈ 1 for small |α|.]
(b) Which of the diagrams in Fig. 3.25 depicts AM, and which depicts FM, for a
sinusoidal information signal?
(c) To get a feeling for AM and FM, use software to plot the transmitted AM and
FM signals for 0 ≤ t ≤ 1 with increments t = 0.0005, for A = 1, ω = 300,
f (t) = cos 18t, and β = 18. (This value of β is chosen excessively large to
highlight the modulations.)

SUMMARY
Polynomials and rational functions of z are analytic except at the poles of the latter.
Every polynomial can be completely factored into degree-one polynomial factors with
complex coefficients; the factorization displays the multiplicity of each zero. The

145
 146 Elementary Functions

Taylor form of the polynomial displays all of the derivatives of the polynomial at the
expansion “center”, as well as the multiplicity of the zero there (if any). A rational
function P/Q, where P and Q are polynomials having no common factors, is said to
have poles at the points where the denominator vanishes. Whenever deg P < deg Q,
the rational can be written as the sum of constants times simple fractions of the form
1/(z − z k ) j , where {z k } are the poles of the rational function and the integer j ranges
from 1 to the multiplicity of the pole.
The complex sine  and cosine  functions are  defined in terms of the exponential
functions: sin z = ei z − e−i z /2i, cos z = ei z + e−i z /2. When z is real these
definitions agree with those given in calculus. Further, the usual differentiation formu-
las and trigonometric identities remain valid for these functions. Complex hyperbolic
functions are also defined in terms of the exponential function, and, again, they retain
many of their familiar properties.
The complex logarithmic function log z is the inverse of the exponential function
and is given by log z = Log |z| + i arg z, where the capitalization signifies the natural
logarithmic function of real variables. Since for each z  = 0, arg z has infinitely many
values, the same is true of log z. Thus log z is an example of a multiple-valued function.
The real and imaginary parts of log z, namely Log r and θ in terms of the polar
coordinates, are harmonic functions taking constant values on circles and wedges re-
spectively. They are useful in solving boundary value problems for Laplace’s equation.
The concept of analyticity for a multiple-valued function f is discussed in terms of
branches of f . A single-valued function F is said to be a branch of f if it is continuous
in some domain, at each point of which F(z) coincides with one of the values of f (z).
Branches of log z can be obtained by restricting arg z so that it is single-valued and
continuous. For example, the function Log z = Log |z| + i Arg z, which is the princi-
pal branch of log z, is analytic in the domain D ∗ consisting of all points in the plane
except those on the nonpositive real axis. The formula d(log z)/dz = 1/z is valid in
the sense that it holds for every branch of log z.
Complex powers of z are defined by means of logarithms. Specifically, z α =
e α log z for z  = 0. Unless α is an integer, w = z α is a multiple-valued function whose

branches can be obtained by selecting branches of log z. The inverse trigonometric

and hyperbolic functions can also be expressed in terms of logarithms, and they too
are multiple-valued.
The analysis of sinusoidally oscillating systems can be greatly simplified with the
use of complex variables. In particular, the differentiation operation, applied to terms
containing the factor eiωt , reduces to simple multiplication (by iω).

Suggested Reading
The algorithms for extracting the zeros of degree-three and degree-four polynomials
are presented in the following tables:

[1] Standard Mathematical Tables. The Chemical Rubber Company, Cleveland

(continuing editions).

146
Chapter 3 – Suggested Reading 147

[2] Handbook of Mathematical Tables and Formulas. R.S. Burington. Handbook

Publishers, Sandusky, Ohio (continuing editions).

The following text may be helpful for further study of boundary value problems for
Laplace’s equation (Sec. 3.4):

[3] Snider, A.D. Partial Differential Equations: Sources and Solutions, Prentice-
Hall, Upper Saddle River, NJ, 1999.

The following texts may be helpful for further study of the concepts used in Sec. 3.6:

Electrical Circuits

[4] Scott, D.E. An Introduction to Circuit Analysis: A Systems Approach, McGraw-

Hill Book Company, New York, 1987.

[5] Hayt, W.H., Kemmerly, J.E., and Durbin, S.M. Engineering Circuit Analysis,
6th ed. McGraw-Hill, New York, 2001.

Sinusoidal Analysis

[6] Lathi, B.P. Signal Processing and Linear Systems, Oxford University Press,
New York, 2000.

[7] Guillemin, E.A. Theory of Linear Physical Systems. John Wiley & Sons, Inc.,
New York, 1963.

147