I have the following (real) matrix which I need to be positive semi-definite,

$P = \begin{bmatrix} P_1 & -\frac{1}{2}(P_1+P_2)\\-\frac{1}{2}(P_1+P_2) & P_2\end{bmatrix} \succeq 0$.

I think this is only the case when, $P_1 = P_2 \succeq 0$, but I couldn't find a way to prove this. I was therefore wondering if this is even the case and if so how to prove this.


Indeed your claim is correct.

The characteristic polynomial of the matrix is:

$$\lambda^2-(P_1+P_2)\lambda-\frac{1}{4}(P_1-P_2)^2,$$ so that the eigenvalues are: $$ \lambda_\pm=\frac{P_1+P_2\pm\sqrt{2(P_1^2+P_2^2)}}{2}. $$ To ensure that both eigenvalues are non-negative it should hold: $$ P_1+P_2\ge0;\quad (P_1+P_2)^2\ge 2(P_1^2+P_2^2)\Rightarrow 0\ge (P_1-P_2)^2. $$ The last inequality can however hold only if $P_1=P_2$.

  • $\begingroup$ I forgot to mention that P1 and P2 are symmetric matrices of an arbitrary size, so not necessarily scalar. Will it still hold in that case? $\endgroup$ – Patbott Apr 13 '18 at 8:35
  • $\begingroup$ @Patbott 0f course the proof is valid only for numbers, not for matrices. You should reformulate the question. Probably the best way is to write a new one. $\endgroup$ – user Apr 13 '18 at 8:59

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.