I'm having difficulty solving a linear algebra problem:
Let $A,B,C,D$ be real $n \times n$ matrices. Show that there is a non-zero $n \times n$ matrix $X$ such that $AXB$ and $CXD$ are both symmetric.

There is an accompanying hint:
Show that the set of all matrices $X$ for which $AXB$ is symmetric is a vector space, and compute its dimension.

I introduced some notation: let $S(A,B)$ denote the set of all $n \times n$ real matrices such that $AXB$ is symmetric. It's easy to prove that for any $A,B$ real $n \times n$ matrices $S(A,B)$ is a subspace of $\cal {M}_{n \times n} (\mathbb{R})$-it's closed under addition, scalar multiplication, and contains the zero matrix. Now I'm not entirely sure where to go. I'd like to be able to show that the space $S(A,B) \cap S (C,D)$ has dimension greater than $0$, but I haven't made any progress thus far.

Any help would be greatly appreciated!


Hint: The product $AXB$ has entries that can be expressed as linear combinations of the entries of $X$, where the coefficients come from the entries of $A$ and $B$. The symmetry condition imposes equations on the entries, so we have a linear system of equations with unknowns the off-diagonal entries of $X$. How many equations are there? You should be able to use this to compute the dimension, and it will be big enough to get what you want.


Let $T$ map every matrix $X$ to $AXB$.

  1. Show that $T$ is a linear transformation.

  2. What is the preimage of the space of symmetric matrices under $T$?

  3. What is the dimension of the space of symmetric matrices?

  4. What can you deduce from 1 and 2 about the dimension of $S(A,B)$?

  5. Use 4 to prove the intersection is nontrivial.

  • $\begingroup$ @MattSamuel: Who said anything about nonzero? $S(A,B)$ is the space of matrices $X$ such that $AXB$ is symmetric, not such that $AXB$ is nonzero symmetric. I think it works out in the end - every addition to $T$'s kernel adds a vector that maps to 0, and takes away at most a vector that maps to a nonzero symmetric matrix. I think I have a proof but I'm not 100% sure about it, can you provide a counterexample? $\endgroup$ – Meni Rosenfeld Jun 17 '15 at 19:28
  • $\begingroup$ I believe I see the point you are driving at, but am still having difficulty: let $S$ denote the space of all symmetric $n \times n$ real matrices. If we can prove that $T |_{S(A,B)}$ gives an isomorphism from $S(A,B)$ to $S$ then we'll be set-the latter space has dimension $\frac{n^2 + n}{2}$, then a counting argument shows the intersection is non-trivial. I can't see how to prove that $T|_{S(A,B)}$ is surjective, and I believe that for the argument to work this is necessary.. $\endgroup$ – aherring Jun 18 '15 at 20:27
  • $\begingroup$ @aherring: $T$ can have a nontrivial kernel, which $S(A,B)$ contains, so $T|_{S(A,B)}$ needn't be injective, and our plan isn't to show that it is an isomorphism. Rather, we want to prove $\dim T^{-1}(S) \ge \dim S$ (it can be strictly larger, obvious example is $A=B=0$. Decompose the dimension of the domain of $T|_{S(A,B)}$, and use theorem about dimension of intersection to bound $\dim(S)\cap Im T$. It should all work out unless I made a mistake somewhere. PS I originally thought that in general $\dim T^{-1}(S)\ge\dim S$ is an obvious well-known theorem, but even if not, we can prove it. $\endgroup$ – Meni Rosenfeld Jun 18 '15 at 20:57

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.