MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have a $5 \times 5$ parametric nonnegative matrix and want to show that it's stable (in the sense that all eigenvalues are positive). It is not symmetric, but I do know in advance that it has 5 real eigenvalues.

I have also shown that it's a $P$-matrix (i.e. all principal minors are positive). So now I can try to check if it's sign-symmetric and if yes, I'll have stability by Carlson's theorem. However, the sign-symmetry involves checking lots of cases and I wonder if there is an alternative approach that I'm missing.

So, to put it in a general way:

Under which additional assumptions is a non-symmetric nonnegative P-matrix stable?

share|cite|improve this question
up vote 2 down vote accepted

Just a remark. Following your assumptions, your matrix is hyperbolic in the sense that you know that all eigenvalues are real-valued. I understand that it depends on some (real) parameters, then from a theorem due to Bronshtein, the characteristic roots are Lipschitz-continuous functions of the parameters.

Of course, if the roots are all positive at some value of the parameter (assume that there is one real parameter $t$ near the distinguished value $0$), then they will stay positive for $t$ near $0$. So the only case to look at is a root $\lambda(t)$ such that $\lambda(0)=0$. If it is a simple root, it is smooth and you just have to check for instance that $$ \dot \lambda (0)=0,\quad\ddot\lambda(0)>0 \tag 1 $$ which will ensure that $\lambda$ will stay positive near $t=0$. The characteristic polynomial $$ p(t,X)=\prod_{0\le j\le 4}(X-\lambda_j(t)) $$ and you may assume that you know $\frac{\partial p}{\partial X}(0,0)\not=0:$ then $$ p(t,X)=e(t,X)(X-\lambda_0(t)),\qquad p(t,\lambda_0(t))\equiv 0 $$ and $ \partial_t p +\partial_X p\dot \lambda_0=0,\quad \partial_t^2 p+2\partial_t \partial_X p \dot \lambda_0 + \partial_X^2 p\dot \lambda_0+\partial_X p \ddot \lambda_0 =0. $ As a result to ensure (1), it is enough to check $$ \frac{\partial p}{\partial t}(0,0)=0,\quad (\partial_t^2 p)(0,0) (\partial_X p)(0,0)<0. $$ Bazin.

share|cite|improve this answer
Bazin, thanks for the interesting answer. – Felix Goldberg Apr 30 '12 at 9:44

After some hard work showing sign-symmetry, I realized that the answer is much simpler:

Being a $P$-matrix is equivalent to having the property that the real eigenvalues of all principal submatrices are positive.

Since I had known that my matrix had real eigenvalues to begin with, I should have just QEDed it after showing $P$-matricity.

share|cite|improve this answer
However, I'm glad that Brazin gave his answer - it contains a useful idea that might help in other situations. – Felix Goldberg Apr 30 '12 at 15:17

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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