Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Question Find matrices $A,B$: such that there exists matrices $X,Y$: $AX-BY=[A,B]$ and $XA-YB=[A,B]$, here $X,Y$ are also matrices, [A,B]=AB-BA - commutator.

"Find" means "say something worth" any help is welcome. In particular it is not clear for me:

0) Are there some nice families of solutions except $[A,B]=aA +bB$?

1) What is dimension of variety $A,B$?

2) Is it reduced? If not how many components this variety has?

... and further in future one might be asking algebraic-geometric questions about is the corresponding ideal radical ? normality, singularities...

Remarks. For fixed $A,B$ we have linear system of equations on $X,Y$. however this system is always degenerate. For generic matrices $A,B$ it would not have solutions.

See also these questions:

$\det(A \otimes B - B \otimes A) = 0$ why? Why $rk(M) = n^2-n$ ? Why x and -x in Spec(M) ?

share|cite|improve this question
up vote 4 down vote accepted

If $A$ and $B$ are invertible, the following condition is necessary. If $A$ and $B$ are invertible and $A B^{-1}$ has distinct eigenvalues, it is necessary and sufficient:

For all integers $n$, we have $$Tr{\Large (}[A,B] (AB^{-1})^k {\Large )} = Tr{\Large (}[A,B] (B^{-1} A)^k {\Large )}. \quad (\ast)$$

Proof of necessity: I'll write out the case of $k$ positive; $k$ negative is similar. The left hand side of $(\ast)$ is $$Tr((AX-BY) (A B^{-1})^k) = Tr (X (AB^{-1})^k A - Y (AB^{-1})^k B) = Tr (XAB^{-1} A B^{-1} \cdots A - Y A B^{-1} \cdots A)$$ where we have used the cyclic symmetry of trace. Similarly, the right hand side of $(\ast)$ is $$Tr((XA-YB) (B^{-1} A)^k) = Tr(X A (B^{-1} A)^k - Y B (B^{-1} A)^k) = Tr (XAB^{-1} A B^{-1} \cdots A - Y A B^{-1} \cdots A).$$ The two expressions are identical. $\square$

Proof of sufficiency: Let's study the more general question of, given matrices $P$ and $Q$, when there exist $X$ and $Y$ with $AX-BY=P$ and $XA-YB=Q$. Let $V \subseteq \mathrm{Mat}_n^2$ be the set of pairs $(P,Q)$ which can be written in this way.

Clearly, $V$ is a linear space, so it is defined by some linear relations. Every linear function on $\mathrm{Mat}_n^2$ is of the form $(P,Q) \mapsto Tr(PT-QU)$ for some matrices $T$ and $U$. Let's figure out for which $(T,U)$ the function $Tr(PT-QU)$ vanishes on $V$.

This occurs if and only if $$Tr((AX-BY) T -(XA-YB)U) = Tr (X (TA-AU)) - Tr (Y (TB-BU))=0.$$ for all $X$, $Y$. This occurs if and only if $TA=AU$ and $TB=BU$.

So we deduce the more general, but less explicit, condition:

Your equation is solvable if and only if we have $Tr([A,B] T) = Tr([A,B] U)$ for all $T$ and $U$ such that $TA=AU$ and $TB=BU$.

Now suppose $A$ and $B$ are invertible. Then $T A B^{-1} = A U B^{-1} = A B^{-1} T$. So $T$ commutes with $A B^{-1}$ and $U$ can be computed from $T$ as $U=A^{-1} T A$. With the additional hypothesis that $A B^{-1}$ has distinct eigenvalues, the space of matrices that commute with $A B^{-1}$ is spanned by the powers of $A B^{-1}$. $\square$.

Some comments on dimension: Note that the sequence $(A B^{-1})^k$ satisfies a linear recurrence of length $n$, by the Cayley-Hamilton theorem. So condition $(\ast)$ for any $n$ consecutive values implies $(\ast)$ for all $k$. Also, condition $(\ast)$ is trivial for $k=0$. So condition $(\ast)$ cuts out a space of codimension $\leq n-1$ in the open set of matrices for which $A$ and $B$ are invertible and $A B^{-1}$ has distinct eigenvalues. The space of solutions to your equation thus has dimension $\geq 2 n^2 - (n-1)$.

I didn't check carefully, but I think that the condition $[A,B] \in \mathrm{Span}(A,B)$ has dimension $n^2 + n+1$, so much smaller.

A general comment: The set of $(A,B)$ such that this equation is solvable will be constructable, by Chevalley's theorem. But I see no reason to expect that it is closed. Without a result saying that the image is closed, I don't see a way to equip the image with the structure of an algebraic variety. Without this, questions about singularities and more subtle invariants don't make sense.

share|cite|improve this answer
Thank you very much !!! That is great and unexpected, I did not thought this line of thought... Some replies 1) it seems to me dim of [A,B]=a A +bB is n^2+n the same as [A,B]=0 2) (A,B) - only contructible - seems yes - probably I was too hurrying up - what I actually keep in mind is variety (A,B,X,Y) with additional equation [X,Y]=0 "relaxed commuting variety". May be it is the same good as usual "commuting variety". – Alexander Chervov Jun 9 '12 at 20:01
See… for my arguments for n^2+n for [A,B]=aA+bB – Alexander Chervov Jun 9 '12 at 20:27
Note that my solution (like your question above) does not impose that $X$ and $Y$ commute. That extra nonlinear condition sounds hard to deal with. Also, you're right about $n^2+n$; I was working too fast. – David Speyer Jun 9 '12 at 22:00

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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