Let $D$ be a real diagonal matrix $D=\operatorname{diag}(a_1,a_2,\ldots,a_n)$ with $a_1\le a_2\le\ldots\le a_n$. Assume that at least one of the $a_i$ is positive. Let $P$ be an irreducible, real, row-stochastic matrix (all entries between 0 and 1, the sum of the elements of every row is 1), with all its eigenvalues real. Let $k>0$ be a real number. I wish to prove that $D+kP$ has a positive eigenvalue. Do you think this is true?


Edit: Yes, it's true. Actually, the assumptions that $P$ is irreducible and row-stochastic can be removed. All we need is only that $P$ is (entrywise) nonnegative.

Since $P$ is nonnegative, so is $D-a_1I+kP$. By Perron-Frobenius theorem, the spectral radius $\rho(D-a_1I+kP)$ is an eigenvalue of $D-a_1I+kP$. Thus $\lambda=a_1+\rho(D-a_1I+kP)$ is a real eigenvalue of $D+kP$. It remains to show that $\lambda>0$.

As $D-a_1I+kP$ is entrywise bounded below by the nonnegative matrix $D-a_1I$, we see that $(D-a_1I+kP)^m$ is entrywise bounded below by $(D-a_1I)^m$ for every positive integer $m$. By using Gelfand's formula with Frobenius norm, it follows that $\rho(D-a_1I+kP)\ge\rho(D-a_1I)=a_n-a_1$. Therefore $\lambda=a_1+\rho(D+kP-a_1I)\ge a_n>0$.

  • $\begingroup$ A clever and very elegant answer. Thank you! $\endgroup$ – user67490 Mar 20 '13 at 11:03
  • $\begingroup$ Very beautiful and elaborated answer. $\endgroup$ – user67692 Mar 20 '13 at 17:28
  • $\begingroup$ @user67490 You can also thank user1551 by voting up his answer. $\endgroup$ – user940 Apr 18 '13 at 20:35
  • $\begingroup$ @user67692 Beautiful enough to warrant an upvote, don't you think? $\endgroup$ – user940 Apr 18 '13 at 20:36

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.