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

Let $A=[A_{0}\ E;E^{T} \ B]$ be a real positive definite matrix and let $B$ be a principal submatrix. I am interested in tightly bounding $\frac{|B|}{|A|}$ from below in some "explicit" way that will involve the entries of $A$. ($A$ is actually a strictly diagonally dominant $M$-matrix, if that helps).

An obvious first approach is to try the Fischer-Hadamard inequality which gives:

$$ \frac{|B|}{|A|} \geq \frac{1}{|A_{0}|}. $$

This is not bad at all, but insufficient for my purposes (it gets the order of magnitude right in my examples but I need more than that).

Another approach is to let $C=A/B$ be the Schur complement and to observe that $|A|=|B||C|$ and so $\frac{|B|}{|A|} \geq \frac{1}{|C|}$; however, I can't find appropriate bounds on $|C|$ to use as the next step.

Do you know of such bounds? Or another way to tackle this problem?

share|cite|improve this question
Of course, the first approach I described is a special case of the second. – Felix Goldberg Dec 6 '12 at 21:38

From the standard block-matrix determinant formula, we get $$\det(A) = \det(B) \det(A_0 - E B^{-1} E^T)$$ so $$\frac{\det(B)}{\det(A)} = \frac{1}{\det(A_0 - E B^{-1} E^T)} = \frac{1}{\det(A_0) \det(I - A_0^{-1/2} E B^{-1} E^T A_0^{-1/2})}$$

Since $$ u^T (A_0 - E B^{-1} E^T) u = (u^T, \ (B^{-1} E^T u)^T) A \pmatrix{u\cr B^{-1} E^T u\cr} > 0 \ \text{for $u \ne 0$} $$ we get that $A_0 - E B^{-1} E^T$ is positive definite; together with the fact that $A_0^{-1/2}E B^{-1} E^T A_0^{1/2}$ is positive semidefinite this says $I \ge I - A_0^{-1/2} E B^{-1} E^T A_0^{-1/2} > 0$ and gives your lower bound $1/\det(A_0)$. Of course that lower bound is exact if $E = 0$. You could improve the lower bound if you could find some nonzero lower bounds on eigenvalues of $ A_0^{-1/2} E B^{-1} E^T A_0^{-1/2}$.

share|cite|improve this answer

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.