MA 205 Complex Analysis: Lecture 5

U. K. Anandavardhanan
IIT Bombay

August 19, 2019

[Logarithm]

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Announcements

We’ll have the first quiz on

Tuesday, August 27, 2019

in the Tutorial Hour.

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Introduction

So last lecture, almost all the time we spent discussing the

exponential function. We defined it by a power series which
converges everywhere. We characterized the exponential function
in two distinct ways. It’s the only function which is invariant under
differentiation if we normalize it such that the function takes the
value 1 at the point 0. It’s essentially the only function with the
property of converting addition in C to multiplication in C× . We
checked that, for a real variable, the exponential function matches
e x . e x is monotonic increasing, hence one-to-one, hence invertible.
d
This inverse is the logarithm. We checked that dx log x = x1 . It’s
also an easy check that log(xy ) = log x + log y , for x, y ∈ R.
Today we’ll discuss the complex logarithm.

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm

The real logarithm in MA 105 was a nice differentiable function,

with derivative x1 , and in MA 205, we want a nice holomorphic
function, with derivative z1 . Where will we hunt for this?
Remember: log x was the inverse of exp(x) = e x ; i.e.,

exp(log x) = x & log(exp(x)) = x.

If we blindly imitate, complex log “should be” the “inverse” of the

complex exponential. But exp(z) is not one-to-one:

exp(z + 2nπı) = exp(z) exp(2nπı) = exp(z),

since exp(2nπı) = 1 for any n ∈ Z. By the way, it’s easy to see

that this is the only way in which one-one-ness can be lost; i.e.,
exp(z) = exp(w ) implies z − w is an integral multiple of 2πı.

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm

Wait a minute. Why do we need one-one-ness to define the

inverse? If f is not one-to-one, f can take two distinct points to
the same point, say f (x) = f (y ) = a with x 6= y , and then f −1 will
take the image point a to both x and y . But a function doesn’t do
that; every point has a unique image under a function. In other
words, a function, by the very definition, is single-valued.
Okay, what if we weaken the notion a little bit? What if we are
willing to consider multi-valued functions? Then, we don’t need
one-one-ness to talk about inverses. The inverse will be
multi-valued, that’s all.

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm
Thus, exponential not being one-to-one was just a minor
head-ache. We have effective and cheap painkillers available in the
market: multi-valued functions! Okay, so let’s look at these
multi-valued inverses of the exponential. If z = re ıθ is given, we
need to get something, which when we exponentiate, we should
get z back. Since there’s an e ıθ in z, we immediately see that ıθ
has to be there in this something. Then there should be something
else, which gives r on exponentiation. Since r is a positive real, the
answer is clear: this is nothing but log r . Thus, what we hunt for is
something like
log r + ıθ,
whose exponential is indeed z. It’s multi-valued since θ can be
replaced with θ + 2nπ for any n in the above. The θ in z is called
the argument of z, denoted by arg(z). We can prescribe its range
to be any interval (α, α + 2π] so that the argument is single-valued.
When α = −π, we call it Arg(z), the principal argument.
U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5
Complex Logarithm
Have we then hunted down the logarithm? No, not yet. The
candidate
re ıθ 7→ log r + ıθ,
looks good, it is defined throughout C× , which is fine since we
don’t expect 0 to have a logarithm, because exponential is never
zero. But it has a very major drawback. Remember we had agreed
that the complex log should be holomorphic. This map is far from
being holomorphic on C× . It’s not even continuous. The real part
p
(x, y ) 7→ log x 2 + y 2
is fine; it’s differentiable. The imaginary part
z 7→ arg(z)
is the problem. This is not continuous. If arg ranges over
(α, α + 2π], take a point zα on the half-ray re ıα . As z → zα from
one direction, the limit of arg(z) is α, whereas from the opposite
direction, the limit is α + 2π. Thus, lim arg(z) does not exist.
z→zα
U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5
Complex Logarithm

The argument not being continuous on C× causes a major

head-ache. It makes sure that you cannot define a holomorphic
logarithm throughout C× . There’s no way-out. No cheap painkiller
will do. The only way to proceed in order to have holomorphic
logarithms defined is to do a surgery! We’ll cut and throw away
parts of C× in such a way that in the remaining part, maps of the
kind
log r + ıθ
with appropriate θ, turn out to be holomorphic. In that case, it
does give a logarithm, on the new domain, since exp of this is z as
we have seen already. Note the use: “a” logarithm. There are
many logarithms, as we can do surgery in many different ways. Of
course, a logarithm `1 on Ω1 will agree with a logarithm `2 on Ω2
over Ω1 ∩ Ω2 , as can be easily checked. (Agree = agree modulo
the 2nπ’s).

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm

Let’s now write down a logarithm. We first define it only on

Ω = {z | Re(z) > 0}, the right half plane. ` : Ω → C defined by
p y 
`(z) = log x 2 + y 2 + ı tan−1 .
x
In the first and fourth quadrants (which constitute the right half
plane), tan−1 yx is nothing but Arg(z). Thus,

(exp ◦`)(z) = z.

Also,
(` ◦ exp)(z) = z,
if exp(z) ∈ Ω. Both u(x, y ) and v (x, y ) are continuously
differentiable, and CR equations can be checked. So `(z) is
holomorphic. Check that `0 (z) is indeed z1 . It’s a logarithm!

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm
Can we enlarge the domain? Why not Re(z) ≥ 0 (minus the
origin)? i.e., include the y -axis. Only problem is the yx in the
formula for `. Easy to fix: define
x 2 + y 2 + ı tan−1 yx
 p


 log if Re(z) > 0
`(z) = log y + ı 2 π
if x = 0 and y > 0
 π
log(−y ) − ı 2 if x = 0 and y < 0.


Once again check differentiability and `0 (z) = z1 .

We would like to enlarge the domain further. But keep in mind
that it can’t be the whole of C× . First note that
p y 
log x 2 + y 2 + ı tan−1
x
is not a logarithm on the left half plane. If we exponentiate this,
we get −z and not z. This is because
y 
Arg(z) = tan−1 ±π
x
respectively in the second and third quadrants.
U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5
Complex Logarithm
Define ` on C minus the negative real axis as follows:
log x 2 + y 2 + ı tan−1 yx
 p


 if Re(z) > 0

π
log y + ı 2 if x = 0 and y > 0



`(z) = log(−y ) − ı π2 if x = 0 and y < 0
log x 2 + y 2 + ı tan−1 yx + ıπ
 p

if x < 0, y > 0




log x 2 + y 2 + ı tan−1 yx − ıπ
 p

if x < 0, y < 0.


Check that ` is differentiable throughout the domain and `0 (z) = 1

z
everywhere. This is a branch of the logarithm. Since `(z) is
nothing but
log |z| + ıArg(z),
we call this branch, the principal branch of the logarithm. This
domain cannot be further enlarged. Via a minimal surgery, we
removed the negative real axis in order to get holomorphy. Thus,
the negative real axis is a minimal branch cut. This corresponds to
the principal branch.
U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5
Complex Logarithm

You should convince yourself that there is nothing sacrosanct

about the negative real axis. You could define a branch of the
logarithm with branch cut being any half-ray, for example. If you
visualize C ∪ {∞}, note that the logarithm cannot be continuously
defined in any neighbourhood of 0 or ∞. Around any other point
it can be continuously (and holomorphically) defined in a small
enough neighbourhood. Is this clear to all of you? We say 0 and
∞ are the branch points for the logarithm.

Remark: Branch, branch cut, and branch point can be defined for
any multi-valued function. Picking one branch makes the
multi-valued function single-valued. Branch cut is a cut you
perform to make it holomorphic in the remaining part. This cut we
will do minimally usually but not always. Branch cuts are usually
constructed by connecting branch points.

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm

Now that we have defined log, we can define several other

functions. For instance, we define

z w = exp(w log z).

If log in the above formula is the principal branch, then note that
π π π
ıı = exp(ı log ı) = exp(ı · ı ) = exp(− ) = e − 2 ,
2 2
which is a real number!
ı am imaginary, but if ı apply my full power to myself, (ıı ), then
it’s real!
Keep in mind:

If you don’t apply yourself, good grades will remain imaginary!

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Aside

So imaginary power imaginary can be real. Can irrational power

irrational be rational?

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm

Similarly we define inverse trigonometric functions etc. For

instance, let w = sin−1 z. Then,

e ıw − e −ıw
z = sin w = .
2ı
Solve the quadratic to get:
p
e ıw = ız + 1 − z 2.

Thus,
1 h p i
sin−1 z = log ız + 1 − z 2 .
ı

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5

Complex Logarithm

Exercise: Analyse the functions

   
1+z 1 + ız
(i) log (ii) log .
1−z 1−z

See the natural maximal domains on which these are holomorphic.

Write down the real and imaginary parts u and v explicitly in these
domains. What are the branch points?

Exercise: Describe branch points and branch cuts for

p
(z − z1 )(z − z2 ) . . . (z − zn )

when (i) n is even, and (ii) when n is odd.

U. K. Anandavardhanan IIT Bombay MA 205 Complex Analysis: Lecture 5