Notes on Complex Analysis
Export PDFGitHub

§2.5 Elementary Functions

2.5 Elementary Functions

Functions of one complex variable that are formed by compositions, sums, products, and powers of finitely many functions of the following form are known as elementary functions:

  1. Power functions including polynomials, rational functions, and their inverses.
  2. Trigonometric functions, hyperbolic functions, and their inverses.
  3. Exponential functions and their inverses.

Power functions are easily extendable to the complex plane by simply changing the real variable to a complex variable. The other two classes of functions have to be redefined and reinterpreted for the complex plane. It is well known that the exponential function can be expanded as

e𝑥=𝑥00!+𝑥11!+𝑥22!+𝑥33!+…=𝑥00!i0+i𝑥11!i1−𝑥22!i2−i𝑥33!i3+…=cos(𝑥i)+isin(𝑥i).

This is better written as

ei𝑥=cos(𝑥)+isin(𝑥),(2.5.1)

which is the famous Euler formula. Then for any complex number 𝑧=𝑥+i𝑦,

e𝑧=e𝑥+i𝑦=e𝑥(cos(𝑦)+isin(𝑦)).

Then trigonometric functions and exponential functions can be written in terms of each other:

sin(𝑧)=ei𝑧−e−i𝑧2i,cos(𝑧)=ei𝑧+e−i𝑧2,tan(𝑧)=ei𝑧−e−i𝑧i(ei𝑧+e−i𝑧)sinh(𝑧)=e𝑧−e−𝑧2,cosh(𝑧)=e𝑧+e−𝑧2,tanh(𝑧)=e𝑧−e−𝑧e𝑧+e−𝑧.

Hence, the following relationships are derived:

sin(𝑧)=−isinh(i𝑧),cos(𝑧)=cosh(i𝑧),tan(𝑧)=−itanh(i𝑧).

The complex logarithm, denoted 𝑤=log(𝑧), is the solution to 𝑧=e𝑤. We can then define the inverse trigonometric and hyperbolic functions.

We can also define the power function for non-integer powers with 𝑤=𝑧𝛼=e𝛼log(𝑧). Then power functions can all be written in terms of exponential functions and logarithms. Letting 𝑥=π in (2.5.1)(2.5.1)(2.5.1) yields eiπ=−1. Furthermore, we can see that exponentiation with an imaginary number is a rotation:

Theorem 2.5.1 (De Moivre).

∀𝑥∈ℝ, ∀𝑛∈ℕ,

(cos(𝑥)+isin(𝑥))𝑛=cos(𝑛𝑥)+isin(𝑛𝑥).

Since all elementary functions can be written in terms of exponential functions and complex logarithms, we will first study the exponential function.

  1. The exponential function e𝑧 never vanishes as |e𝑧|=e𝑥>0.
  2. Since e2πi=1, it is periodic over 2πi.
  3. It is also an entire function with (e𝑧)′=e𝑧.

    Write e𝑧=e𝑥+i𝑦=e𝑥(cos(𝑦)+isin(𝑦)) where 𝑥,𝑦∈ℝ. Let 𝑢(𝑥,𝑦)=ℜ𝔢(e𝑧)=e𝑥cos(𝑦) and 𝑣(𝑥,𝑦)=ℑ𝔪(e𝑧)=e𝑥sin(𝑦). The first order derivatives are respectively

    𝜕𝑢𝜕𝑥=e𝑥cos(𝑦),𝜕𝑢𝜕𝑦=−e𝑥sin(𝑦),

    and

    𝜕𝑣𝜕𝑥=e𝑥sin(𝑦),𝜕𝑣𝜕𝑦=e𝑥cos(𝑦),

    and indeed, the condition described by Theorem 2.2.3Theorem 2.2.3Theorem 2.2.3 is satisfied.

  4. For any two complex numbers 𝑧1 and 𝑧2, e𝑧1e𝑧2=e𝑧1+𝑧2.

    In fact, most real exponentiation rules are identical to those in the complex number field. Previously we claimed the periodic properties of e𝑧. For 𝑈⊆ℂ, a holomorphic function 𝑓:𝑈→ℂ is univalent over 𝑈 if it is injective over 𝑈. This means that the solutions 𝑧1 and 𝑧2 satisfying 𝑓(𝑧1)=𝑓(𝑧2) will also always satisfy 𝑧1=𝑧2.

  5. The function e𝑧 is univalent over any horizontal strip of height 2π.

    Let 𝑧1=𝑥1+i𝑦1 and 𝑧2=𝑥2+i𝑦2, with 𝑥1,𝑦1,𝑥2,𝑦2∈ℝ, and assume e𝑧1=e𝑧2. Then

    e𝑥1ei𝑦1=e𝑥2ei𝑦2.

    The moduli are equal, and therefore 𝑥1=𝑥2. By the periodic nature of exponentiation of imaginary numbers, 𝑦1−𝑦2=2π𝑘, where 𝑘∈ℤ. To satisfy univalence over a region 𝑈, we must exclude distinct points whose imaginary parts differ by an integer multiple of 2π. Thus, we may select 𝑈 to be any horizontal strip

    2π𝑘≤ℑ𝔪(𝑧)<2π(𝑘+1)

    or

    2π𝑘<ℑ𝔪(𝑧)≤2π(𝑘+1).

    Similar to the exponential function, any belt region with thickness 2π is a region over which log is univalent.

Next we examine the complex logarithm.

  1. From the periodicity of 𝑧=e𝑤, log is a multi-valued function.
  2. Let 𝑧=𝑟ei𝜃 and 𝑤=𝑢+i𝑣, where 𝑟,𝜃,𝑢,𝑣∈ℝ. Then

    𝑟ei𝜃=e𝑢+i𝑣,

    and e𝑢=𝑟, meaning that 𝑢=log(𝑟) and 𝑣=𝜃+2π𝑘, where 𝑘∈ℤ. Then

    𝑤=log(𝑟)+i(𝜃+2π𝑘),

    and using modulus-argument notation,

    log(𝑧)=log|𝑧|+iarg(𝑧),

    where arg(𝑧) is the multi-valued argument function. We denote the principal branch of the argument function by

    Arg:ℂ∖{0}→(−π,π].
    The principal branch of log(𝑧), or Log(𝑧), can be defined such that ℑ𝔪(Log(𝑧))∈(−π,π].

The functions sin and cos, through their exponential form, still satisfy the familiar properties such as their derivatives, periodicity of 2π, parity, sum and difference formulas, and the fundamental identities

sin2(𝑧)+cos2(𝑧)=1,sin(𝑧)=cos(π2−𝑧).

However, due to the extension, some properties do not hold. For instance, sin(𝑧) and cos(𝑧) are unbounded, as along the imaginary axis they resemble their hyperbolic counterparts, which are unbounded along the real line.

We now examine the regions over which they are univalent. Consider

cos(𝑧)=ei𝑧+e−i𝑧2.

Define the auxiliary functions

𝜉(𝑧)=i𝑧,𝜁(𝜉)=e𝜉,𝑤(𝜁)=𝜁+1𝜁2.

Then

cos(𝑧)=(𝑤∘𝜁∘𝜉)(𝑧).

𝜉 is clearly univalent on ℂ, as it is a linear map, specifically, a rotation by π2 radians followed by scaling. The function 𝜁 is univalent on any domain 𝑈⊆ℂ such that for all 𝜉1,𝜉2∈𝑈, 𝜉1−𝜉2≠2πi𝑘 for any 𝑘∈ℤ. If 𝜉1=i𝑧1 and 𝜉2=i𝑧2, then this translates to 𝑧1−𝑧2≠2π𝑘 for 𝑘∈ℤ. The function

𝑤(𝜁)=𝜁+1𝜁2

is univalent on regions excluding pairs (𝜁1,𝜁2) such that 𝜁1=1𝜁2. In terms of 𝑧, this condition becomes ei𝑧1ei𝑧2≠1, or equivalently, 𝑧1+𝑧2≠2π𝑘 for any 𝑘∈ℤ.

Combining these constraints, we conclude that cos(𝑧) is univalent on any vertical strip in the complex plane of width π, such as a region of the form

{𝑧∈ℂ:𝑘π<ℜ𝔢(𝑧)<(𝑘+1)π,𝑘∈ℤ}.

Let us now consider the specific region

{𝑧∈ℂ:0<ℜ𝔢(𝑧)<π},

and analyze how it is mapped under cos(𝑧).

  1. 𝜉(𝑧)=i𝑧 maps the region {𝑧∈ℂ:0<ℜ𝔢(𝑧)<π} to {𝜉∈ℂ:0<ℑ𝔪(𝜉)<π}.
  2. 𝜁(𝜉)=e𝜉 maps this region to the upper half-plane ℑ𝔪(𝜁)>0 since 0<Arg(𝜁)<π and 0<|𝜁|.
  3. 𝑤(𝜁)=𝜁+1𝜁2 maps ℑ𝔪(𝜁)>0 to ℂ∖((−∞,−1]∪[1,∞)).

Thus, the composition cos(𝑧)=𝑤∘𝜁∘𝜉 is univalent on the strip

{𝑧∈ℂ:0<ℜ𝔢(𝑧)<π},

and the image of this strip under cos is

ℂ∖((−∞,−1]∪[1,∞)).

We will now analyze the inverse cosine function, denoted arccos(𝑧). Consider

𝑧=ei𝑤+e−i𝑤2.

Then

(ei𝑤)2+1=2𝑧ei𝑤ei𝑤=2𝑧±4𝑧2−42𝑤=−ilog(𝑧±𝑧2−1).

Then arccos is also a multi-valued function. We can also define

arcsin(𝑧)=π2−arccos(𝑧).

Lastly, we will examine the power function. Let 𝛼=𝑢+i𝑣 where 𝑢,𝑣∈ℝ. Then

𝑧𝛼=exp(𝛼log(𝑧))=exp((𝑢+i𝑣)(log|𝑧|+iarg(𝑧))),

and in polar form,

𝑧𝛼=exp(𝑢log|𝑧|−𝑣arg(𝑧))exp(i(𝑣log|𝑧|+𝑢arg(𝑧))).

Let

𝑟𝑘=exp(𝑢log|𝑧|−𝑣arg(𝑧))

and

𝜃𝑘=𝑣log|𝑧|+𝑢arg(𝑧).

Then 𝑧𝛼=𝑟𝑘ei𝜃𝑘, where 𝑘∈ℤ. Analyzing the coefficient of 𝑣 in the exponent of 𝑟𝑘, 𝑧𝛼 is multi-valued if 𝑣≠0.

Then assuming 𝑣=0, we have

𝑧𝛼=|𝑧|𝑢exp(i𝑢arg(𝑧)).

Doing casework on 𝛼,

  1. If 𝛼=𝑢∈ℤ, then 𝑢 can be absorbed into 𝑘, and 𝑧𝛼 is single-valued.
  2. If 𝛼=𝑢∈ℚ with reduced fractional form 𝑝𝑞, where 𝑝,𝑞∈ℤ, 𝑞>0, and gcd(𝑝,𝑞)=1, then the multi-valued function 𝑧𝛼 is given by

    𝑧𝛼=|𝑧|𝑝𝑞exp(i𝑝𝑞(Arg(𝑧)+2π𝑘))=|𝑧|𝑝𝑞exp(i𝑝𝑞Arg(𝑧))exp(2πi𝑝𝑞𝑘),

    for 𝑘∈ℤ. These values are periodic with period 𝑞, since

    exp(2πi𝑝𝑞(𝑘+𝑞))=exp(2πi𝑝𝑞𝑘),

    as exp(2πi𝑝)=1 for integer 𝑝. To prove there are exactly 𝑞 distinct values, consider 𝑘=0,1,2,…,𝑞−1. The exponential factors are exp(2πi(𝑝𝑞)𝑘). These are distinct if, for 0≤𝑗<𝑘≤𝑞−1,

    exp(2πi𝑝𝑞𝑗)≠exp(2πi𝑝𝑞𝑘),
    which holds unless (𝑝𝑞)(𝑘−𝑗)∈ℤ, or equivalently unless 𝑞 divides 𝑝(𝑘−𝑗). Since gcd(𝑝,𝑞)=1, 𝑞 must divide 𝑘−𝑗, but |𝑘−𝑗|<𝑞 and 𝑘−𝑗≠0, a contradiction. Thus, 𝑧𝛼 has exactly 𝑞 distinct values.
  3. If 𝛼=𝑢∈ℝ∖ℚ, then 𝑧𝛼 is infinite-valued.

Lastly, there exist series representations of functions using power functions, namely Taylor series, and trigonometric functions, namely Fourier series. There does not exist another distinct representation using exponential functions, as trigonometric functions can be written in terms of them.