MathOverflow is a question and answer site for professional mathematicians. 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

(I tried asking that on, but did not get a satisfying answer. I am trying here as well, in case someone here will have more insight. The question was eventually abandoned there. I personally think it is a challenging question.)

Let $C(y)$ be the following function:

$C(y) = A \operatorname{diag}(A^t y) A^{-1}$

where $A$ is an invertible $m \times m$ matrix and $y$ is an $m \times 1$ vector, everything is over the reals.

Let's assume I know the values of $C(y)$ for any given $y$ (I can basically calculate $C(y)$ for any $y$). Is there a way to identify the matrix $A$ (or the set of solutions satisfying the equality above)?

Few insights that I gained through over time and through

There should be some connection between $A$ and eigendecomposition. Basically, $C(y)$ can be diagonalized for any $y$. It might mean that $A$ consists of scaled eigenvectors of $C(y)$ in certain cases.

Any hints would be greatly appreciated.

Here is the version of the question:

share|cite|improve this question
Two questions: (1) does $A^t$ mean the transpose of $A$, or the $t$'th power of $A$? (2) does 'diag($B$)' refer to the vector whose elements are the diagonal elements of $B$? – Daniel Mansfield Sep 28 '11 at 23:04
I should have clarified this: $A^t$ is indeed the transpose, and $\mathrm{diag}(x)$ is an $m \times m$ matrix with the vector $x$ on the diagonal and 0 everywhere else. – dotproduct Sep 28 '11 at 23:07
thanks Douglas, good idea. I added the link. – dotproduct Sep 28 '11 at 23:11

Maybe I do not understand the question, but since the vector of eigenvalues of $C(y)$ is $A^t y$, computing the spectral decomposition for $y = e_1, \dotsc, e_m$ (the basis vectors) gives you $A.$ The problem is overdetermined, since you are also postulating that $A$ is the conjugating matrix.

share|cite|improve this answer
thanks! this makes sense to me, but if I understand correctly, doesn't your method imply there is a unique solution? I don't think there is a unique solution for $m=1$. I will try to think how to settle this issue. – dotproduct Sep 28 '11 at 23:05
@Igor > since the vector of eigenvalues of C(y) is Aty Isn't this true only if A is an orthogonal matrix? The question only assumes that it is invertible. I'm posting this as an answer because I'm not allowed to comment. – psd Sep 28 '11 at 23:09
I guess I am wrong and that A only needs to be invertible for this to be a vector of eigenvalues. – psd Sep 28 '11 at 23:14
oh, whoops, for $m=1$, $C(y) = ay$ not $a^2 y$. My mistake. – dotproduct Sep 28 '11 at 23:17
I guess I'm missing something: the spectral decomposition is not unique, you can permute the eigenvalues around. So without saying something more, you only get the rows of $A$ up to permutation. – euklid345 Sep 29 '11 at 3:10

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.