3
$\begingroup$

For certain classes of sets $S \subseteq \mathbb{R}^n$, there exist algebraic characterizations of real valued polynomials $p: \mathbb{R}^n \rightarrow \mathbb{R}$ that are positive on $S$. Several of these algebraic characterizations, known as positivstellensatz, are listed here in Wikipedia.

Some of these positivstellensatze can be generalized for matrix polynomials, i.e. a function $P$ from $\mathbb{R}^n$ to the set of all $r \times r$ symmetric matrices whose entries are polynomials. In this case, $P$ is understood to be positive in $S$ if $P(x)$ is positive definite for all $x \in S$.

For instance, in [1], Pólya's positivstellensatz for homogeneous polynomials is generalized to homogeneous matrix polynomials and in [2], Putinar's positivstellensatz is also generalized to matrix polynomials. Stengle's and Schweighofer's positivstellensatze are also generalized to matrix polynomials in [3].

Does Handelman's positivstellensatz have also a matrix polynomial version?

[1]: Scherer, C. W. "Relaxations for robust linear matrix inequality problems with verifications for exactness." SIAM Journal on Matrix Analysis and Applications 27(2) (2005): 365-395.

[2]: Scherer, C. W. and Hol, C. W. J. "Matrix sum-of-squares relaxations for robust semi-definite programs." Mathematical programming 107(1-2) (2006): 189-211.

[3]: Công-Trình, Lê. "Some Positivstellensätze for polynomial matrices." arXiv preprint arXiv:1403.3783 (2014)

$\endgroup$
2
$\begingroup$

The obvious generalization is false: Consider $A=\begin{pmatrix} 1+x &y \\y & 1-x \end{pmatrix}$ which is a linear matrix polynomial. It defines a compact set in the plane, namely the unit disc. The algebra $\mathbb{R}[A]$ of all polynomial expressions in $A$ is contained in $\mathbb{R}(x,y)[A]$ which is a two dimensional $\mathbb{R}(x,y)$-vector space since $A^2=2A+x^2+y^2-1$. Therefore $\mathbb{R}[A]$ can not contain all of these three $\mathbb{R}(x,y)$-linear independent matrices: $\begin{pmatrix}1 & 0 \\ 0 & 2 \end{pmatrix}$, $\begin{pmatrix}2 & 0 \\ 0 & 1 \end{pmatrix}$, $\begin{pmatrix}2 & 1 \\ 1 & 2 \end{pmatrix}$ which are all positive definite on the unit disc.

Edit: So Handelman's statement is the following: If the linear forms $l_1, \ldots, l_r$ define a compact set and if the polynomial $f$ is strictly positive on this set, then $f$ can be expressed as a sum of products of $l_1, \ldots, l_r$ with positive coefficients.

Now replace "(linear) polynomials" by "(linear) matrix polynomials" and you get a generalization to matrix polynomials which is false. The matrix $A$ plays hereby the role of $l_1, \ldots, l_r$.

$\endgroup$
  • $\begingroup$ thanks a lot for the answer :) Could you detail more in the answer what do you mean by the obvious generalization? $\endgroup$ – Shamisen Oct 4 '14 at 22:47
  • 1
    $\begingroup$ see my edit for the generalization I have in mind. $\endgroup$ – Hans Oct 5 '14 at 16:19

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.