Every two disks are conformally equivalent, after translation and scaling. An annulus has an additional geometric parameter: the ratio of its outer and inner radii. Can a clever conformal map change that ratio? It cannot. The logarithm of the ratio, called the modulus up to a conventional factor, is a conformal invariant.
Proving this requires care at the boundary. A biholomorphism is initially defined only inside the annulus, so we cannot assume it extends continuously to either boundary circle. We first prove the weaker boundary information that is actually needed. Then the harmonic function and a winding-number argument force the map's form.
The theorem
Definition 3.7.1
(Round annulus and its modulus)
For , write
The constant is a convention; equality of moduli is equivalent to equality of . Any annulus with is carried onto by , so it suffices to study the normalized annuli .
Concept lensGeometric
An annulus as a flat cylinder
The exponential map sends the vertical strip onto , and exactly when . So is the strip with and identified: a cylinder of height and circumference . The modulus is height divided by circumference, which explains the factor . In this picture the rotations become vertical translations , and the exchanging maps become the half-turns , which swap the two edges.
Theorem 3.7.2
(Round-annulus rigidity)
Let . If is biholomorphic, then . Every automorphism of has one of the forms
The first type preserves the two boundary components; the second exchanges them. Conversely, every displayed map is an automorphism. Thus round annuli are biholomorphic exactly when their moduli agree.
A holomorphic map with nonzero derivative need not be globally injective. In particular, maps an annulus onto another annulus and is locally conformal there, but for it is -to-one. The classification relies on global biholomorphicity.
What happens near each boundary circle?
The first step is to extract from the boundary information that the later argument uses.
Lemma 3.7.3
(Boundary circles correspond)
Let be biholomorphic. Then assigns to each boundary circle of one boundary circle of , the two assignments are different, and as tends to a source radius, tends uniformly to the radius of the assigned target circle.
Proof
(A boundary component cannot split between two target circles)
Consider a sequence approaching either source boundary circle. If a subsequence of converged to a point , continuity of the inverse would give , a contradiction. Hence all cluster values lie on or .
For the outer circle, choose a sequence with . The sets
are nonempty compact connected subsets of the closed target annulus and decrease with . Their intersection is nonempty and connected by the nested compact connected-set theorem. It is exactly the cluster set of as : one direction follows from the definition; for the other, choose in the th collar whose image approaches a prescribed point of the intersection. This cluster set lies in the union of two disjoint circles, so connectedness places it entirely in one of them. The same proof applies to the inner source circle.
These two assignments are different. Indeed, approach either target boundary circle by points and put . A subsequence of converges in the compact closed source annulus. An interior limit would, by continuity of , make tend to an interior point, a contradiction. Thus it approaches a source boundary circle whose cluster set contains the chosen target boundary point; that cluster set lies in one target circle, which is its assigned component. Both target circles must therefore appear among the two assignments, making the assignments a bijection.
Finally, the modulus tends uniformly to the assigned boundary radius as tends to each source radius. Otherwise a sequence violating uniformity would have a target cluster value on the wrong circle. We have not asserted pointwise extension of itself; the limiting modulus is enough for the next step.
Proof X-ray
Why boundary behavior needs a proof
Compactness and inverse continuity force cluster values onto the target boundary. Connected collars prevent one source boundary circle from mapping partly toward each of two disjoint target circles. Surjectivity forces both target components to occur. These steps give uniform limiting moduli without assuming an extension theorem for the map.
Common mistake
(Assuming that the map extends to the boundary)
A biholomorphism is defined only on the open annulus. Writing , or applying the maximum principle to on the closed annulus, silently assumes a continuous extension that has not been proved. The argument here needs, and proves, only the uniform limits of the modulus ; these are enough to give the harmonic function continuous boundary values.
Harmonic boundary values determine the modulus
Proof
(A harmonic function on the annulus)
Since is holomorphic and never zero, is harmonic: locally a holomorphic logarithm exists and its real part is . The preceding uniform modulus convergence makes converge uniformly to a constant on each boundary circle, either or . Defining those boundary values gives a function continuous on the entire closed annulus.
If preserves boundary components, has values on and on . The harmonic function with these values is
If it exchanges components, the corresponding formula is
To justify uniqueness, subtract the proposed formula from . The difference is harmonic in , continuous on its compact closure and zero on both boundary circles. The harmonic maximum principle applied to it and its negative makes it zero throughout. Thus in either case
Proof
(Winding and injectivity force the exponent)
Locally choose holomorphic logarithms of and . The Cauchy–Riemann equations applied to their real parts give
Integrating around any positively oriented circle in yields
The left side is the integer winding number of the closed curve around zero. Hence is a nonzero integer. The derivative of vanishes throughout the connected annulus, so for a nonzero constant .
If , two distinct points on the same source circle differing by a suitable th root of unity have the same image. Injectivity therefore gives or . In the preserving case ; in the exchanging case . Both imply . The boundary moduli then give in the first case and in the second. The displayed maps are evidently holomorphic bijections with holomorphic inverses, completing the proof.
The quotient is globally holomorphic because never vanishes, so the integral is well-defined even though the logarithms were local. The integer in this argument is a winding number, not a choice of a globally defined logarithm on the annulus. Indeed, has no single-valued holomorphic logarithm on . The local logarithms and their derivatives suffice.
Examples and counterexamples
Worked example 3.7.4
(Distinguishing two annuli)
The annuli and normalize by scaling to and . Their moduli are respectively and , so no biholomorphism exists between them. Equal area or similar appearance would not settle this question; the radius ratio does.
Worked example 3.7.5
(An automorphism exchanging boundaries)
On the map sends toward and toward . It satisfies , so it is its own inverse. Its derivative never vanishes in the annulus. Despite exchanging the boundary circles, it is orientation-preserving as a real map because its complex derivative is nonzero.
Worked example 3.7.6
(Composing exchanging automorphisms and their fixed points)
Compose two exchanging automorphisms of :
So the composite is a rotation, consistent with the fact that exchanging the boundary circles twice preserves them. For fixed points, means , giving the two points . Both lie on the circle , which is inside because . A rotation with has no fixed point in , since there. Hence every automorphism other than the identity fixes at most two points, and those points lie on the middle circle of radius .
The power maps show what goes wrong when injectivity is dropped.
Counterexample mode
Local conformality is weaker than a biholomorphism
The map sends onto and has derivative throughout its domain. But and have the same image, so it is not injective. There is no contradiction with the invariant: , and the rigidity theorem requires a biholomorphism.
The classification is for round annuli with positive inner radii. A punctured disk has a different end: the inner boundary has collapsed to a point, so it is not covered by the hypothesis . Nor have we classified arbitrary doubly connected domains here.
Exercises
Quick check
(Decide whether and are biholomorphic, and give a map if they are.)
Compare the ratios of inner to outer radius.
Solution · Answer
Both radius ratios are , so the moduli agree. The scaling maps the first annulus bijectively onto the second.
Quick check
(List all automorphisms of sending to itself.)
Test both families of automorphisms at the point .
Solution · Answer
For a rotation to fix , . For to fix , again . The two maps are and . The radius is the geometric mean of the boundary radii, so the exchanging map fixes that circle setwise.
Quick check
(Show that maps onto , compute its winding number on a source circle, and explain why it is not a biholomorphism.)
Use the argument-principle integral of .
Solution · Answer
Moduli of images satisfy , and every target point has a cube root with modulus in . The winding integral is ; distinct cube-root rotations share an image, so the map is not injective.
Quick check
(Is complex conjugation an annulus automorphism in the holomorphic sense? Compare it with .)
Check complex differentiability, not only bijectivity.
Solution · Answer
Conjugation maps the annulus bijectively onto itself but is not holomorphic: its difference quotient depends on the direction of approach. In contrast is holomorphic with nonzero derivative on and is its own inverse. Exchanging boundary circles does not mean reversing complex orientation.