Take the 2-minute tour ×
MathOverflow is a question and answer site for professional mathematicians. It's 100% free, no registration required.

I am looking for a procedure to find solution(s) for a square matrix equation

$H^T H = S$

where $H = H^\dagger$ is a hermitian ($n\times n$) matrix and $S$ is a given symmetric complex matrix. Due to hermiticity of $H$, $S$ should satisfy $n$ conditions $(\operatorname{im}(\operatorname{tr}(S^i))=0,\quad i = 1,...,n)$.

I am interested in simple solutions for small $n=3$ matrices. For $n=2$, this can be solved by an explicit parametrization of $H$ which leads to a quadratic equation giving four solutions. Since the problem is similar to taking a square root of a matrix, presumably there are $2^n$ solutions for this problem, too.

note: This question is perhaps similar to this one. Here, the equation is simpler but applies to complex, not real matrices.

share|improve this question
add comment

2 Answers

This is a very partial answer. Assume that $S$ is invertible ; then $S$ and $I$ are congruent and there is $K$ s.t. $K^TK=S$. Here a solution $H$ must be a hermitian matrix ; then $HH^T=\overline{S}$. Since $H$ is invertible, $HH^T$ and $H^TH$ are orthogonally similar. Consequently $S$ and $\overline{S}$ are orthogonally similar. In particular, $S$ must have the spectrum of a real matrix.

For the construction of $K$, have a look at page 130 of: Turnbull, Aitken. An introduction to the theory of canonical matrices.

share|improve this answer
add comment

Any Hermitian matrix is normal, i.e., we may write $H = UDU^*$ where $D$ is a real diagonal matrix, and $U$ is unitary. Thus, $S = (U^* D U)(U^*DU) = U^* D^2 U$ so we may diagonalize $S$ with the unitary matrix $U$. Hence, all eigenvalues of $S$ must by necessity be non-negative and real. Hence, $S$ must be positive semi-definite for it to be a solution. Furthermore, $S$ can be diagonalized with the unitary matrix $U$, so $S$ is normal.

Hence, we can find solutions iff $S^*S=SS^*$ and all eigenvalues are non-negative and real.

From here, we can easily see ways to find all solutions with the following method:

1) Diagonalize $S$, $S = U^*D'U$, where $D'$ have non-negative entries.

2) We can now choose $H = U D U^*$ where $D$ is any root of $D'$. There are be $2^n$ choices here.

(Note, $D^2=D'$).

EDIT: Ah, I misread it as $H^*H=S$. Maybe the above method can be used in some special case?

share|improve this answer
add comment

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.