1: \begin{abstract}

2: 	We construct a convergent family of outer approximations for the problem of optimizing

3: 	polynomial functions over convex bodies 

4: 	subject to polynomial constraints.

5: 	This is achieved by generalizing the \emph{polarization hierarchy}, which has previously been introduced for the study of polynomial optimization problems over state spaces of $C^*$-algebras, to convex cones in finite dimensions.

6: 	If the convex bodies can be characterized by linear or semidefinite programs, then the same is true for our hierarchy.

7: 	Convergence is proven by relating the problem to a certain \emph{de Finetti theorem} for \emph{general probabilistic theories}, which are studied as possible generalizations of quantum mechanics.

8: 	We apply the method to the problem of nonnegative matrix factorization, and in particular to the \emph{nested rectangles problem}.

9: 	A numerical implementation of the third level of the hierarchy is shown to give rise to a very tight approximation for this problem.

10: \end{abstract}

11: