15
$\begingroup$

$\DeclareMathOperator\GL{GL}$Let $G$ be a subgroup of $\GL_n(\mathbb{C})$ such that for every $g \in G$ there exists $c \in \GL_n(\mathbb{C})$ for which $cgc^{-1}$ is unitary (or, which is the same, $g$ is diagonalizable with unimodular eigenvalues). Does there exist $c \in \GL_n(\mathbb{C})$ such that $cGc^{-1}$ consists of unitary matrices?

A similar question is discussed in Venkataramana's answer to Groups of matrices in which all elements have all eigenvalues equal in modulus, and this way I can obtain that $G$ is isomorphic to a subgroup of the unitary group.

Let $\mathbb{R}^n = V$, and $0 = V_0 \subset V_1 \subset V_2 \subset \dotsb \subset V_k = V$ be subspaces such that each $V_i$ is invariant under $G$ and the induced action of $G$ on $V_{i+1}/V_i$ is irreducible. Then I get a homomorphism: $G \to \GL(V_1 \oplus (V_2/V_1) \oplus \dotsb \oplus (V_{k}/V_{k-1}))$, and (see the link above) I can assume that its image consists of unitary matrices. Its kernel is trivial, since this homomorphism preserves eigenvalues and the kernel consists of diagonalizable matrices with all eigenvalues equal to $1$.

Now I see two ways:

  1. This homomorphism (denoted as $S$ and seen as a bijection: $G \to \operatorname{Im}S$) is continuous. If $S^{-1}$ is continuous, I think I'm able to prove that $G$ is bounded. But I'm not sure that it's continuous.

  2. $\operatorname{tr}(ab)$ was a non-degenerate form on $\operatorname{Mat}_{n \times n}(\mathbb{C})$. Will it be non-degenerate on a subspace generated by $G$?

  3. Something else….

$\endgroup$
3
  • $\begingroup$ If all operators in $G$ are uniformly bounded, then $G$ preserves the norm $\|x\|:=\sup_{g\in G} |gx|$ (where $|x|$ is some fixed norm). Thus it preserves the John ellipsoid of the unit ball of $\|\cdot\|$, that yields affirmative answer. $\endgroup$ May 25, 2023 at 20:12
  • $\begingroup$ Fedor Petrov, thank you for noting this! The hard part is to prove that the operators are bounded. $\endgroup$ May 25, 2023 at 20:18
  • 1
    $\begingroup$ The conclusion is negative as [Dave Benson's answer] shows. However, I think it can be shown that one can block-triangulate so that diagonal blocks are unitary. In particular, the conclusion is positive in the irreducible case. $\endgroup$
    – YCor
    May 25, 2023 at 21:25

3 Answers 3

16
$\begingroup$

An example in $GL(3, {\mathbb C})$ was first given by Bass (answering a question by Kaplansky) in Example 1.10 of

Bass, Hyman, Groups of integral representation type, Pac. J. Math. 86, 15-51 (1980). ZBL0444.20006.

In section 1 of his paper Bass also discusses the general structure of subgroups of $GL(n, {\mathbb C})$ where every element is unitarizable.

Here is the example. Start with a free subgroup $F=\langle s, t\rangle$ of rank 2 in $SU(2)$ acting linearly on ${\mathbb C}^2$ with generators acting as matrices $A, B$. Now, deform $F$ to a free group of affine transformations by $$ \rho(s)=A, \rho(t)x= Bx+ v, $$ where $v$ is any nonzero vector in ${\mathbb C}^2$. Then each element of $G=\rho(F_2)$ is affine-conjugate to its linear part, an element of $SU(2)$. At the same time, $G$ has no fixed points in the complex-affine space ${\mathbb C}^2$. It follows that $G$ is unbounded. Lastly, use the fact that the group of complex-affine transformations of ${\mathbb C}^2$ embeds in $GL(3, {\mathbb C})$ sending an affine transformation $$ x\mapsto Mx + v $$ to the matrix $$ \left[\begin{array}{cc} M&v\\ 0&1 \end{array}\right]. $$ Under this map unbounded subsets map to unbounded subsets. This gives an example of a non-unitarizable subgroup of $SL(3, {\mathbb C})$ such that each individual element is unitarizable.

$\endgroup$
14
$\begingroup$

Take the group of matrices of the form $\left(\begin{smallmatrix}A&B\\0&A^{-1}\end{smallmatrix}\right)$ with $A\in U(2)$. Inside this, choose two generators for a free group as the values of $A$, in such a way that no two words or their inverses give the same eigenvalues, and with two independent scalar matrices as values of $B$, and see what they generate; in the group generated by these two matrices, all elements are diagonalisable since they don't have a repeated eigenvalue (apart from the identity), but this group is not bounded so it can't be conjugated into a unitary group.

Edit (26 May 2023) in response to comments:

Presumably the example of Bass (I don't seem to be able to find it in his writings) is similar, but using the group $G$ of matrices of the form $\left(\begin{smallmatrix}A&b\\0&1\end{smallmatrix}\right)$ with $A\in U(2)$ and $b\in\mathbb{C}^2$. This is similar but easier, so let me discuss this. Choose a pair of elements of $U(2)$ close to the identity, generating a dense free subgroup consisting of semisimple elements. Since the Lie algebra of $G$ is generated by two elements, we may attach vectors to these two generators so that the subgroup of $G$ generated is dense, hence unbounded.

$\endgroup$
6
  • $\begingroup$ Sorry about the multiple edits, it took me a while to get the example working properly. I think it should be okay now. $\endgroup$ May 25, 2023 at 20:58
  • 3
    $\begingroup$ You can even lower the dimension to 3 (the example is due to Bass). $\endgroup$ May 25, 2023 at 20:58
  • $\begingroup$ I was originally trying dimension 2, and - surprise, surprise - it didn't work properly. $\endgroup$ May 25, 2023 at 21:11
  • $\begingroup$ What is the 3-dimensional example? $\endgroup$ May 25, 2023 at 21:32
  • 1
    $\begingroup$ Ah! I didn't see @MoisheKohan's response above. Yes, this is essentially the example of Bass. $\endgroup$ May 26, 2023 at 10:19
7
$\begingroup$

It does not seem to have been mentioned in your question, or in the answer and comments that a periodic subgroup $G$ of ${\rm GL}(n,\mathbb{C})$ satisfies your condition. There were several theorems of Jordan, Burnside and Schur which culminated in the theorem of Schur that a periodic subgroup of ${\rm GL}(n,\mathbb{C})$ is similar to a group of unitary matrices. See Section 36 of chapter V of Curtis and Reiner (1962).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.