If I have an $n \times n$ real-valued non-symmetric matrix $\mathbf{M}$, which has determinant $|\mathbf{M}| > 0$, what can I say about the determinant of the matrix $\mathbf{Q}^T \mathbf{M}^{-1} \mathbf{Q}$, where $\mathbf{Q}$ is a real-valued $n \times m$ matrix?

In particular, can I say that $|\mathbf{Q}^T \mathbf{M}^{-1} \mathbf{Q}| > 0$?

For background, the matrix $\mathbf{Q}^T \mathbf{M}^{-1} \mathbf{Q}$ is the Schur complement of the block matrix in the following linear system:

$$ \left( \begin{array}{cc} \mathbf{M} & \mathbf{Q} \\ \mathbf{Q}^T & \mathbf{0} \\ \end{array} \right) \left( \begin{array}{c} \beta \\ \gamma \\ \end{array} \right) = \left( \begin{array}{c} \mathbf{f} \\ \mathbf{0} \\ \end{array} \right). $$

In the case that $\mathbf{M}$ is symmetric positive definite (i.e. symmetric, and $|\mathbf{M}| > 0$,), I believe I'm right in saying that $|\mathbf{Q}^T \mathbf{M}^{-1} \mathbf{Q}| > 0$.


Obviously, no. For example $Q = 0$. In fact, whenever $m > n$, you'll get a zero determinant.

We observe the case when $m \le n$. Let $Q = U \Sigma V^T$ be an SVD of $Q$. Then $$Q^T M^{-1} Q = V \Sigma^T U^T M^{-1} U \Sigma V^T.$$ So, $$\mathop{\rm det}(Q^T M^{-1} Q) = \mathop{\rm det}(V) \mathop{\rm det}(\Sigma^T U^T M^{-1} U \Sigma) \mathop{\rm det}(V^T) = \mathop{\rm det}(\Sigma^T U^T M^{-1} U \Sigma).$$

Denoting $\Sigma = \mathop{\rm diag}(\sigma_1,\dots,\sigma_m) \in \mathbb{R}^{n \times m}$, we see that $\Sigma^T U^T M^{-1} U \Sigma$ is $U^T M^{-1} U$ with $i$-th row multiplied by $\sigma_i$ and $j$-th row multiplied by $\sigma_j$, so $$\mathop{\rm det}(\Sigma^T U^T M^{-1} U \Sigma) = \left( \prod_{k=1}^m \sigma_k \right)^2 \mathop{\rm det}(U^T M^{-1} U) = \left( \prod_{k=1}^m \sigma_k \right)^2 \mathop{\rm det} M^{-1} = \frac{\left( \prod_{k=1}^m \sigma_k \right)^2}{\mathop{\rm det} M}.$$ So, $\mathop{\rm det}(Q^T M^{-1} Q) > 0$ if and only if $\mathop{\rm rank} Q = m$; otherwise $\mathop{\rm det}(Q^T M^{-1} Q) = 0$.

  • $\begingroup$ Thanks very much for your detailed reply. I should have said that one of the conditions of the linear system in my application is that $rank(Q) = m <= n$. From the last statement of your answer I guess that does confirm it then, $det(Q^T M^{-1} Q) > 0$ . Brilliant, thanks again! $\endgroup$ – Tom Jul 14 '13 at 3:01

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.