Let $A, B$ be two positive semidefinite matrices. Prove that if both $A-B$ and $B-A$ are positive semidefinite, then $A = B$.

I can show that their diagonal elements are the same but for others I have no idea. Any help will be appreciated.


One route is to observe that if $\lambda,v$ is an eigenvalue/eigenvector pair of $M$ with , then $(-M)v=-\lambda v$, meaning that $-\lambda$ is an eigenvalue of $(-M)$. Since $M:=A-B$ is positive semidefinite, conclude that any eigenvalue of $M$ satisfies $0\leq \lambda\leq 0$, so $\lambda=0$, so that $M=0$.

  • $\begingroup$ Thank you very much! Thats a neat proof $\endgroup$ – honzaik Mar 15 at 18:15

By assumption $A-B=-(B-A)$ is both positive semidefinite and negative semidefinite. This can only happen if $A-B=0$.

  • $\begingroup$ its this equivalent to saying that $x^TAx=0 \forall x \implies A = 0$ which is wrong (math.stackexchange.com/questions/358281/…) ? $\endgroup$ – honzaik Mar 15 at 17:57
  • $\begingroup$ I assumed we are working over $\mathbb C$ and so we are using $x^*Ax$. So the link is not relevant. $\endgroup$ – chhro Mar 15 at 18:01
  • $\begingroup$ actually, the argument still works since the real positive semidefinite matrices=real symmetric matrices with nonnegative eigenvalues. I don't see the point of mentioning the link. I didn't use that in the argument. $\endgroup$ – chhro Mar 15 at 18:05
  • $\begingroup$ well i assumed you meant $0 \geq -x^T (B-A) x = x^T (A-B) x \geq 0$. I guess when I hear positive semidef etc I assume this definition with vectors. $\endgroup$ – honzaik Mar 15 at 18:10

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.