Notes on Complex Analysis
Export PDFGitHub

§11.3 Topological Equivalence and Biholomorphic Equivalence

11.3 Topological Equivalence and Biholomorphic Equivalence

Lemma 11.3.1.

If 𝑷 is a permutation matrix, and 𝑫 is a diagonal matrix, then there exists a diagonal matrix 𝑫′ such that 𝑷𝑫=𝑫′𝑷. Similarly, there exists a diagonal matrix 𝑫″ such that 𝑫𝑷=𝑷𝑫″.
Proof.

Let 𝑫=diag(𝑑1,…,𝑑𝑛) and let 𝜎 be the permutation corresponding to 𝑷, 𝑷𝒆𝑖=𝒆𝜎(𝑖) for each standard basis vector 𝒆𝑖. Define

𝑫′=diag(𝑑𝜎−1(1),…,𝑑𝜎−1(𝑛)).

Then for every 𝑖,

𝑷𝑫𝒆𝑖=𝑷(𝑑𝑖𝒆𝑖)=𝑑𝑖𝒆𝜎(𝑖),

while

𝑫′𝑷𝒆𝑖=𝑫′𝒆𝜎(𝑖)=𝑑𝑖𝒆𝜎(𝑖).

Hence

𝑷𝑫𝒆𝑖=𝑫′𝑷𝒆𝑖

for all 𝑖, so

𝑷𝑫=𝑫′𝑷.
Now apply this result to 𝑷⊤, and thus, (𝑷⊤𝑫)⊤=(𝑫″𝑷⊤)⊤⟺𝑫⊤𝑷=𝑷𝑫″⊤⟺𝑫𝑷=𝑷𝑫″ since diagonal matrices are invariant under transposition. ⁠□

Theorem 11.3.2 (Poincaré).

For any 𝑛≥2, the 𝑛-dimensional unit ball 𝐵𝑛 and the 𝑛-dimensional polydisk 𝔻𝑛 are not biholomorphically equivalent.
Proof.

Suppose, for the sake of contradiction, that there exists a biholomorphism 𝝋:𝔻𝑛→𝐵𝑛. Let 𝜶=𝝋(𝟎)∈𝐵𝑛, and define 𝚽=𝝋‖𝜶‖∘𝑼∘𝝋, where 𝑼 is a unitary matrix such that 𝑼𝜶=(‖𝜶‖,0,…,0) and 𝝋‖𝜶‖ is as in Proposition 11.2.16Proposition 11.2.16Proposition 11.2.16.

The definition of 𝚽 ensures that 𝚽:𝔻𝑛→𝐵𝑛 and 𝚽(𝟎)=𝟎. Then 𝚽−1∘Aut(𝐵𝑛)∘𝚽 consists of functions mapping 𝔻𝑛 to 𝔻𝑛, or that

𝚽−1∘Aut(𝐵𝑛)∘𝚽⊆Aut(𝔻𝑛)⟹Aut(𝐵𝑛)⊆𝚽∘Aut(𝔻𝑛)∘𝚽−1.

Similarly, 𝚽∘Aut(𝔻𝑛)∘𝚽−1⊆Aut(𝐵𝑛). Therefore, Aut(𝐵𝑛)=𝚽∘Aut(𝔻𝑛)∘𝚽−1, and

𝝍↦𝚽∘𝝍∘𝚽−1(11.3.1)

defines a group isomorphism between Aut(𝔻𝑛) and Aut(𝐵𝑛). Let Aut′(𝔻𝑛)<Aut(𝔻𝑛) and Aut′(𝐵𝑛)<Aut(𝐵𝑛) be subgroups fixing 𝟎. Therefore, (11.3.1)(11.3.1)(11.3.1) induces a group isomorphism between Aut′(𝔻𝑛) and Aut′(𝐵𝑛) as well.

By Theorem 11.2.11Theorem 11.2.11Theorem 11.2.11, every element of Aut′(𝔻𝑛) may be uniquely identified with a matrix in the form of

𝑷diag(ei𝜃1,…,ei𝜃𝑛),

where 𝑷 is a permutation matrix and (𝜃1,…,𝜃𝑛)∈[0,2π)𝑛. Hence Aut′(𝔻𝑛) is isomorphic to the group of unitary monomial matrices. The structure of Aut′(𝐵𝑛) is given by Proposition 11.2.17Proposition 11.2.17Proposition 11.2.17, and each element corresponds uniquely to a unitary matrix. Thus there is a natural isomorphism Aut′(𝐵𝑛)≅U(𝑛), the 𝑛×𝑛 unitary group.

For 𝑼∈U(𝑛), the spectral theorem allows it to be expressed in the form of 𝑽diag(ei𝜃1,…,ei𝜃𝑛)𝑽†. Hence, for any positive integer 𝑚, we have

𝑽diag(ei𝜃1𝑚,…,ei𝜃𝑛𝑚)𝑽†∈U(𝑛)

and

(𝑽diag(ei𝜃1𝑚,…,ei𝜃𝑛𝑚)𝑽†)𝑚=𝑽diag(ei𝜃1𝑚,…,ei𝜃𝑛𝑚)𝑽†⋯𝑽diag(ei𝜃1𝑚,…,ei𝜃𝑛𝑚)𝑽†.

The adjacent products of 𝑽†𝑽 simplify to the identity and the entire expression then simplifies to 𝑼. Hence the unitary group is divisible.

Consider the unitary monomial matrix

𝑷𝜏=(0100…01000…00010…00001…0⋮⋮⋮⋮⋱⋮0000…1).

inducing the permutation 𝜏, swapping the first and second entries. Assume that there exists some unitary monomial matrix 𝑸=𝑷𝜎𝑫 (where 𝑫 is diagonal and 𝑷𝜎 is a permutation matrix corresponding to the permutation 𝜎) such that 𝑸2=𝑷𝜏. This is equivalent to

𝑷𝜎𝑫𝑷𝜎𝑫=𝑷𝜏⟹𝑷𝜎𝑷𝜎𝑫′𝑫=𝑷𝜏⟹𝑷𝜎2𝑫″=𝑷𝜏,

where 𝑫′ and 𝑫″ are diagonal matrices, where the former existence is given by Lemma 11.3.1Lemma 11.3.1Lemma 11.3.1. Thus, 𝑷𝜎2=𝑷𝜏 (and 𝑫″=𝑰) since their permutation parts must match. This is an impossibility since 𝑷𝜎2 corresponds to an even permutation, while 𝑷𝜏 corresponds to an odd permutation. Thus, the unitary monomial group is not divisible.

By Proposition 11.2.19Proposition 11.2.19Proposition 11.2.19, the two groups cannot be isomorphic to each other.

This contradicts the existence of (11.3.1)(11.3.1)(11.3.1), and therefore, no such biholomorphism 𝝋 exists. ⁠□

Remark.

A more succinct proof of the nonexistence of an isomorphism in the proof of Theorem 11.3.2Theorem 11.3.2Theorem 11.3.2 can be briefly described by means of topology:

Let 𝑀𝑛 denote the subgroup of all monomial matrices in U(𝑛), or the subgroup of unitary matrices with exactly one nonzero entry in each row and each column, and those nonzero entries lying in U(1).

For each permutation 𝜎∈𝑆𝑛 (the symmetric group of permutations) let 𝑷𝜎 be the corresponding permutation matrix and define

𝑇𝜎={𝑫𝑷𝜎:𝑫=diag(ei𝜃1,…,ei𝜃𝑛)∈U(1)𝑛}.

Each 𝑇𝜎 is homeomorphic to the torus U(1)𝑛, and every element of 𝑀𝑛 lies in exactly one 𝑇𝜎. Hence

𝑀𝑛=⋃𝜎∈𝑆𝑛𝑇𝜎,

a disjoint union of |𝑆𝑛|=𝑛! copies of U(1)𝑛.

Each 𝑇𝜎 is clopen in 𝑀𝑛 by their pairwise disjointness, the topology of the torus, and the fact that their union is 𝑀𝑛. Therefore each 𝑇𝜎 is a connected component of 𝑀𝑛. Because each 𝑇𝜎 is connected, 𝑀𝑛 has 𝑛! connected components.

The elements of U(𝑛) may be unitarily diagonalized (by the spectral theorem) into 𝑽𝑫𝑽†, where

𝑫=diag(ei𝜃1,…,ei𝜃𝑛)

is a diagonal unitary matrix and 𝑽∈U(𝑛). Then there exists a connected path connecting (𝜃1,…,𝜃𝑛) to (0,…,0), which corresponds to the matrix 𝑽𝑰𝑽†=𝑰. Because every matrix is path-connected to the identity, U(𝑛) is connected.

Consequently for 𝑛≥2, the subgroup 𝑀𝑛 (which has more than one connected component) cannot be isomorphic to U(𝑛) as a topological group.

Although we do not justify these topological claims in detail here, it is worth noting, heuristically, why such topological considerations naturally arise.

A biholomorphism between two domains induces a homeomorphism with respect to their natural topologies (the compact-open topology) by 𝝍↦𝚽∘𝝍∘𝚽−1. Hence any induced topological invariant of an automorphism group, such as connectivity, is to be preserved under this equivalence. Of course, we are yet to verify the rigor and intuition used within the topology, but the intuitive picture already hints to the validity of the connectivity argument.