I have two correlation matrices A and B. They are:

  • Real symmetric (with ones on the diagonal)
  • Positive semi-definite (eigenvalues are $\ge 0$)

I want to try to prove that the average of these two matrices $C={1\over2}A + {1\over2}B$ still has the same properties.

It was a few years since my linear algebra courses and would like some help along the way here. I am assuming it's possible as I have tested it numerically and for my 5000 randomly generated correlation matrices the property holds (at least within machine precision).

Grateful for any kind of help!


Assume $A$ and $B$ are positive semidefinite ($z^TAz\geq0$ for all $z\in\mathbb{R}^n$). So then $$ z^T\frac12(A+B)z=\frac12(z^TAz+z^TBz)\geq0, $$ and therefore$\frac12(A+B)$ is positive semidefinite. Symmetry is also preserved under addition $\frac12(A+B)^T=\frac12(A^T+B^T)=\frac12(A+B)$.

  • $\begingroup$ Ah of course, I had forgotten that $z^TAz \ge 0$ was a property and was only thinking of the eigenvalues. Thank you! $\endgroup$ – Fredrik Oct 4 '12 at 13:05

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.