This might be elementary, but I'm struggling to write the product of a sums of squares polynomial and an odd power of a variable using its Gram matrix.

Following Pablo Parrilo's notation, let $p(x,y):=[\mathbb{x}]_d^t G [\mathbb{x}]_d$ be a polynomial in $\mathbb{R}[x,y]$ which is a sum of squares of degree $2d$, where $[\mathbb{x}]_d^t:=[1,x,y,x^2,xy,y^2,x^3,x^2y,xy^2,y^3,\dots,y^d]$ is a vector of monomials and $G$ is the Gram matrix of $p(x,y)$.

How can I compute the Gram matrix of $x.p(x,y)$, for example?

I know that the matrix of the product will have dimmension $\binom{n+d+1}{d+1}$ and there will be a block of zeros somewhere, empirically.

I also know how to solve this problem for things like $x^{2s}.p(x,y)=x^s[\mathbb{x}]_d^tG[\mathbb{x}]_dx^s=[\mathbb{x}]_{d+s}^t\overline{G}[\mathbb{x}]_{d+s}$ with even degree, because then $\overline{G}$ will be sort of an extension of $G$ by adding null columns and rows conveniently.

Can I represent the matrix of $x^{2s-1}.p(x,y)$ in a similar way?

Parrilo has shown that a polynomial is SOS iff there exists a positive semidefinite Gram matrix to it, but at least does this matrix representation always exists, without assuming non-negativity?

Thanks in advance.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.