Suppose $A$ is a $n \times m$ matrix and $B$ is a $m \times n$ matrix. Then it is known that $det(I_{n}+AB)=det(I_{m}+BA)$.

Is there an analogous identity of the form $det(P_{1}+AB)=det(P_{2}+BA)$, where $P_{1},P_{2}$ are positive definite? Or something like it?

  • $\begingroup$ Do you have a reference for "it is known that"? $\endgroup$ – Igor Rivin Jul 5 '12 at 13:21
  • $\begingroup$ This clearly can't be true as stated, if you take $A = B = I_n$ and $P_1 = \lambda I_n$ and $P_2 = \mu I_n$ for two distinct positive numbers $\lambda, \mu$. I don't know if there's something similar. I assume you are working over the field $\mathbb R$. Perhaps your best bet is to diagonalize $P_1$ and $P_2$ and try the explicit formula for the determinant in terms of permutations...? $\endgroup$ – Spiro Karigiannis Jul 5 '12 at 13:23
  • 3
    $\begingroup$ @Igor. See Exercise 14, Chapter 3 of my book *Matrices, Springer-Verlag GTM216. More generally, one has $$X^m\det(XI_n-AB)=X^n\det(XI_m-BA).$$ $\endgroup$ – Denis Serre Jul 5 '12 at 13:30

Given $P$, $\det(P+AB)$ does not depend only on $BA$. For example, take $n=m=2$, $P = \pmatrix{2 & 0\cr 0 & 1\cr}$, $A = \pmatrix{1 & t\cr 0 & 1\cr}$, $B = \pmatrix{0 & 1\cr 1 & -t\cr}$. Then $\det(P + AB) = 1 - t$ depends on $t$, but $BA = \pmatrix{0 & 1\cr 1 & 0\cr}$ doesn't depend on $t$. So any $P_2$ such that $\det(P+AB) = \det(P_2+BA)$ must depend on $t$.


Is this what you are looking for? $$ \det(P+AB)=\det(P)\det(I+P^{-1}AB)=\det P \det(I+BP^{-1}A)$$

Edit: the chain of equalities once ended with $=\det(P+PBP^{-1}A)$, but this only works when all matrices are square.

  • $\begingroup$ Well, not quite.. but it's a nice one. $\endgroup$ – Felix Goldberg Jul 5 '12 at 14:22
  • 1
    $\begingroup$ $B$ is not a square matrix, so you can't conjugate it by anything. $\endgroup$ – Will Sawin Jul 5 '12 at 15:36
  • $\begingroup$ @WillSawin: you are right, the last equality does not hold unless $A$ and $B$ are square. But maybe the second-to-last equality (which is basically the Schur complement formula) is a nice enough expression. $\endgroup$ – Federico Poloni Jul 5 '12 at 17:41

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.