4
$\begingroup$

For given $n$, the following $n\times n$ complex matrix $M=M^{\dagger}$ is called positive, if

$x^{\dagger}M x\geq 0$

holds for all complex vector $x=(z,z^2,\cdots,z^n)^T$ with arbitrary complex $z$, where $M^{\dagger}$ stands for the conjugate transpose of $M$.

Please note that this definition of positivity is not the same as positive definiteness since $x$ is nor ranging over all vectors here.

Of course, the set of such positive matrices forms a convex cone, I would like to have a closed form of the cone, for a given $N$, how to check whether $N$ is positive?

$\endgroup$
  • 1
    $\begingroup$ a matrix fulfilling your notion of positivity is called "positive definite" in linear algebra - that is, in each linear algebra textbook. you can look for answers there. $\endgroup$ – Delio Mugnolo Sep 13 '14 at 19:50
  • 4
    $\begingroup$ @Delio Please see the constraint on the vector $x$ in our definition. $\endgroup$ – gondolf Sep 13 '14 at 21:11
  • $\begingroup$ When you say $M^+$ stands for the complex conjugate of $M$, do you mean the conjugate transpose (often written as $M^\dagger$)? $\endgroup$ – S. Carnahan Sep 13 '14 at 23:32
  • 3
    $\begingroup$ "yes of course" - if you mean the conjugate transpose, say the conjugate transpose $\endgroup$ – Yemon Choi Sep 14 '14 at 12:00
  • 1
    $\begingroup$ Moreover, in its present form I find this question frustratingly unmotivated, with no indication of what effort the OP has put into attacking this problem $\endgroup$ – Yemon Choi Sep 14 '14 at 12:02
1
$\begingroup$

If your question were posed in the real case there is an easy answer. Note that any polynomial $p(\cdot)$ of even degree $2d$ can be written as $p(z)=x^TMx$, where $M$ is a Hermitian matrix and $x=(1,z,z^2,\ldots,z^d)$. In this language your question is how to characterize the cone of nonnegative polynomials (in terms of the matrix $M$): by a classical Theorem of Hilbert, real univariate polynomials are nonnegative if and only if they are sums of squares. That means that $M$ is psd.

Unfortunately, such nice characterization only holds in very few cases for multivariate polynomials (see http://www.msri.org/attachments/workshops/327/553_Lecture-notes_week1_Blekherman.pdf, Theorem 1.2). Moreover, in general, deciding nonnegativity of real polynomials is an NP-hard problem, and therefore one should not expect a closed form description of the cone of such matrices $M$.

Finally, even though I can't figure out a way to describe the complex case within the real one, I suspect there is no nice characterization of this cone.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Thank you. Notice here is onyl two variables case when we consider the complex version, so is it possible to find some way to do this? $\endgroup$ – gondolf Sep 14 '14 at 22:50
  • $\begingroup$ I thought about this, but one cannot simply reduce the complex case to the real one. This is because $i^2=-1$, which does not hold when we write complex numbers as two dimensional vectors. And even if you can do that, in the reference I put above you can see that for bivariate polynomials there are very few degrees where one has nonnegativity iff sum of squares. $\endgroup$ – Cristóbal Guzmán Sep 14 '14 at 23:03

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.