The maximum number of mutually commuting linearly independent complex matrices of order $n$ is equal to $\lfloor n^2/4\rfloor + 1$ ($\lfloor \cdot \rfloor$ denotes the integer part of $\cdot$). This result is attributed to Schur and a short proof can be found here. In an answer to this question, an explicit set $\mathcal{M}$ of linearly independent and mutually commuting matrices is written for $2n\times 2n$ complex matrices: $$ \begin{pmatrix}a\cdot\text{Id}_n & M_{n\times n} \\ 0_{n\times n} &a\cdot \text{Id}_n\end{pmatrix}, $$ where $a$ is a complex number, $\text{Id}_n$ is the order $n$ identity matrix and $M$ is an arbitrary complex matrix.

Now my question: can any nonzero $n\times n$ complex matrix be part of a subspace of ($\lfloor n^2/4\rfloor +1$) linearly independent mutually commuting matrices? If the answer is negative, what would be necessary?

To avoid misunderstanding, let me add some redundancy and rephrase my question as: given an arbitrary nonzero $n\times n$ complex matrix $A_0$, is it possible to find $\lfloor n^2/4\rfloor$ linearly independent $A_i$, $i=1,\ldots,\lfloor n^2/4\rfloor$, such that $A_iA_j-A_jA_i=0$ for $i,j=0,\ldots,\lfloor n^2/4\rfloor$?


The answer to my main question - if any nonzero matrix can be in the maximal subspace of mutually commuting matrices - is negative in general as Ewan Delanoy pointed out with the example of diagonalizable matrices with distinct eigenvalues.

The secondary question - what is necessary to be in the maximal commuting set - seems more complicated. A set of mutually commuting matrices can be simultaneously triangularized, then only upper (lower) triangular matrices need be analyzed. For $2n\times 2n$ matrices an explicit set $\mathcal{M}$ of solutions is provided above and for $(2n+1)\times(2n+1)$ matrices the following set, also called $\mathcal{M}$, is a maximal commuting subspace: $$ \begin{pmatrix}a\cdot\text{Id}_n & M_{n\times n+1} \\ 0_{n+1\times n} &a\cdot \text{Id}_{n+1}\end{pmatrix}. $$

This set of matrices is also suggested in the paper by M. Mirzakhani, "A simple proof of a theorem of Schur", already cited above.

Now I would like to change my secondary question for a more specific one: must any maximal commuting subspace have the form of $\mathcal{M}$?

I think the answer is affirmative and, if that is the case, it will answer this question in MO: "How many commuting nilpotent matrices are there?" because the set $\mathcal{M}$ would be the maximal abelian nilpotent subalgebra plus (a multiple of) the identity matrix.

  • 1
    $\begingroup$ A related question : is it necessary that $A_0$ be of the form $a{\mathsf{Id}}+N$ where $a$ is a scalar and $N$ is nilpotent ? $\endgroup$ Mar 12 '18 at 15:54
  • $\begingroup$ @EwanDelanoy, I think that all matrices in the maximal commuting subspace have the explicit form I give in the question. $\endgroup$
    – jobe
    Mar 13 '18 at 19:59

The answer is easily negative when $n\geq 4$ (so that $\lfloor \frac{n^2}{4}\rfloor+1 \gt n$), because then if you take $A_0$ to be diagonalizable with distinct eigenvalues, the commutant of $A_0$ (i.e. the space of matrices $B$ satisfying $A_0B=BA_0$) already has dimension $n$.

This also indicates a partial answer to your second question : a necessary condition is that the commutant of $A$ must have dimension at least $\lfloor \frac{n^2}{4}\rfloor+1$. This is probably not sufficient however, because there is no reason why the matrices in that commutant should commute with each other.

  • $\begingroup$ @user1551 Corrected, thanks. $\endgroup$ Mar 12 '18 at 16:17
  • $\begingroup$ Thank you for your answer. Defining the maximal commuting subspace for a matrix $M$ as the maximal subspace of mutually commuting matrices including $M$, it seems that the minimum dimension occurs for diagonalizable matrices with non-repeated eigenvalues. $\endgroup$
    – jobe
    Mar 13 '18 at 20:19

In case $A$ is nilpotent or unitriangular this is true, because there is an invertible matrix $S$ such that $SAS^{-1}$ is contained in the above set $\mathcal{M}$. Then $A\in S^{-1}\mathcal{M}S$, and hence is part of such a set of pairwise commuting matrices.

For matrices $A$ where its Jordan form cannot be chosen in $\mathcal{M}$, it need not be true.

Remark: The result has been generalized later by Jacobson for arbitrary fields:

N. Jacobson: Schur's theorem on commutative matrices.
Bull. Amer. Math. Soc. 50 (1944), 431-436.

See also our paper Minimal faithful representations of reductive Lie algebras, where it is Proposition $2.3$:

Proposition 2.3: Let $M$ be a commutative subalgebra of $M_d(K)$ over an arbitrary field $K$. Then $\dim M\le \lfloor d^2/4 \rfloor +1$, and the bound is attained.

  • $\begingroup$ Thank you for your answer. I already knew that Schur's theorem is valid for other fields, but I am really interested in the complex case, then I omitted this fact to avoid unnecessary complications. I didn't know the concept of reductive Lie Algebras, it seems interesting. $\endgroup$
    – jobe
    Mar 13 '18 at 20:21

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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