Skip to content
Tutorials Center
My learning

Topics

Lesson 27 of 36 · Series and Residues

Taylor Series of Analytic Functions

Every analytic function is a power series on its largest disc. Learn Taylor's theorem in the complex plane, the radius of convergence, and how to find coefficients.

In real calculus, having infinitely many derivatives does not guarantee that a function equals its Taylor series. In complex analysis the story is much better: an analytic function is always equal to its Taylor series, on the largest disc that fits inside its domain. And the question "how far does the series converge?" has a clean geometric answer: up to the nearest singularity.

Taylor's theorem

Theorem. Let f be analytic on the open disc |z - a| < R. Then for every z in that disc

Text
f(z) = Σ_(n=0)^∞ c_n (z - a)^n,   c_n = f^(n)(a) / n! = (1/2πi) ∮ f(w) / (w - a)^(n+1) dw

where the integral is over any circle |w - a| = ρ with 0 < ρ < R. The series converges absolutely, and uniformly on smaller closed discs.

Where it comes from. Start from Cauchy's formula f(z) = (1/2πi) ∮ f(w)/(w - z) dw on a circle |w - a| = ρ with |z - a| < ρ. Write

Text
1/(w - z) = 1/((w - a) - (z - a)) = (1/(w - a)) · 1/(1 - q),   q = (z - a)/(w - a),  |q| < 1

and expand the geometric series 1/(1 - q) = Σ q^n. It converges uniformly on the circle, so you can integrate term by term, and each term gives c_n (z - a)^n with the coefficient formula above. That coefficient formula is exactly the derivative formula from Cauchy's derivatives lesson, so c_n = f^(n)(a)/n!.

The radius of convergence

The radius of convergence R of a power series is the largest radius for which it converges on |z - a| < R. For the Taylor series of f at a:

R is the distance from a to the nearest point where f fails to be analytic.

You do not need to analyze the coefficients. You just find the nearest singularity.

Example. f(z) = 1/(1 + z²) is a perfectly smooth real function on the whole real line, but its Taylor series at 0,

Text
1/(1 + z²) = 1 - z² + z⁴ - z⁶ + ...   (geometric series with ratio -z²)

converges only for |z| < 1. Why? Because 1 + z² = 0 at z = ±i, two singularities at distance 1 from the origin. Complex numbers explain a mystery of real analysis: the real series fails at x = 1.1 even though nothing seems wrong at x = 1.

Standard series

All of these are centered at 0:

Function Series Radius
e^z Σ z^n / n! ∞
sin z Σ (-1)^n z^(2n+1) / (2n+1)! ∞
cos z Σ (-1)^n z^(2n) / (2n)! ∞
1/(1 - z) Σ z^n 1
Log(1 + z) Σ (-1)^(n+1) z^n / n for n ≥ 1 1
(1 + z)^α Σ binom(α, n) z^n 1 (for non-integer α)

The function Log(1 + z) has a branch point at z = -1, at distance 1, which explains the radius. Entire functions (no singularities in the finite plane) have infinite radius.

Working with series

  • Add, subtract, multiply: series can be added termwise; products use the Cauchy product (convolution of coefficients).
  • Substitute: replace z by an expression: e^(z²) = Σ z^(2n)/n!.
  • Differentiate and integrate termwise inside the disc of convergence.
  • Divide by a nonvanishing series using the geometric series or long division.
  • Identity theorem. If two analytic functions have Taylor series that agree at a, they agree on every connected domain where both are analytic. A non-constant analytic function has isolated zeros: near a zero at a we have f(z) = (z - a)^m g(z) with g(a) ≠ 0, where m ≥ 1 is the order of the zero.

To find the order of a zero of f at a, compute the first non-vanishing derivative: if f(a) = ... = f^(m-1)(a) = 0 and f^(m)(a) ≠ 0, the order is m. Example: 1 - cos z has f(0) = f'(0) = 0 and f''(0) = 1, a zero of order 2.

Try it: convergence inside and outside the radius

We sum N terms of 1/(1 + z²) = Σ (-1)^k z^(2k) at four points: two inside the unit circle (0.5 and 0.9i), one just outside (1.1), and one outside along the imaginary axis (1.1i). The exact value is 1/(1 + z²). Inside the radius the error shrinks; outside, it explodes.

Python Partial sums of 1/(1+z^2)

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Try it: Taylor coefficients from a circle

By the coefficient formula, c_n = (1/2πi) ∮ f(w)/(w - a)^(n+1) dw. Parametrize the circle w = a + r e^(iθ) with N equally spaced angles: then c_n r^n is the n-th discrete Fourier coefficient of f(a + r e^(iθ)), which NumPy's FFT computes in one line. Below we recover the coefficients of e^z (should be 1/n!) and of 1/(1 + z²) (should be 1, 0, -1, 0, 1, ...) using radius r = 0.5, inside the convergence disc.

Python Coefficients by FFT

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

If you used a radius larger than 1 for 1/(1 + z²), the circle would enclose the poles and the formula would produce Laurent coefficients instead, which is the topic of the next lesson.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. What is the radius of convergence of the Taylor series of 1/(z² + 4) at 0?

    The singularities are z = ±2i, at distance 2 from the center.
  2. 2. The Taylor series of an entire function at 0 converges

    An entire function has no singularities, so the nearest-singularity distance is infinite.
  3. 3. The Taylor coefficient c_n of f at a is

    c_n = f^(n)(a)/n!, which equals (1/2πi)∮ f(w)/(w - a)^(n+1) dw.
  4. 4. f(0) = f'(0) = 0 and f''(0) ≠ 0. The zero of f at 0 has order

    The first non-vanishing derivative is the second, so the order is 2.

Summary

  • An analytic function equals its Taylor series Σ f^(n)(a)/n! (z - a)^n on every disc inside its domain.
  • The radius of convergence is the distance to the nearest singularity.
  • 1/(1 + z²) has radius 1 because of the poles at ±i, which explains a real-variable puzzle.
  • Coefficients can be computed from derivatives, from known series, or from a contour integral.
  • Zeros of a non-constant analytic function are isolated and have a finite order.

Next: Laurent series