Abstract:In 1888, Hilbert proved that the cone $\mathcal{P}_{n+1,2d}$ of positive semidefinite forms in $n+1$ variables of degree $2d$ coincides with its subcone $\Sigma_{n+1,2d}$ of those forms that are representable as finite sums of squares if and only if $(n+1,2d) = (2,2d)_{d\geq1}$ or $(n+1,2)_{n\geq1}$ or $(3,4)$. These are the Hilbert cases. In [GHK23, GHK24], we applied the Gram matrix method to construct cones between $\Sigma_{n+1,2d}$ and $\mathcal{P}_{n+1,2d}$, defined by projective varieties containing the Veronese variety. In particular, we introduced and examined a specific cone filtration $$\Sigma_{n+1,2d} = C_0 \subseteq \ldots \subseteq C_n \subseteq C_{n+1} \subseteq \ldots \subseteq C_{k(n,d)-n} = \mathcal{P}_{n+1,2d}$$ and determined each strict inclusion in non-Hilbert cases. This gave us a refinement of Hilbert's 1888 theorem. Here, $k(n,d)+1$ is the dimension of the vector space of forms in $n+1$ variables of degree $d$. In this paper, we show that the intermediate cones $C_i$'s are closed and describe their interiors and boundaries. We discuss the membership problem for the $C_i$'s, present open problems concerning their dual cones and generalizations to cones defined by toric varieties.

Maximum information divergence from linear and toric models

Logarithmic Voronoi polytopes for discrete linear models

Logarithmic Voronoi cells

Maximum Likelihood Estimation of Toric Fano Varieties

Toric invariant theory for maximum likelihood estimation in log-linear models

Degrees of the Wasserstein Distance to Small Toric Models

Maximum likelihood degree of the two-dimensional linear Gaussian covariance model

Toric geometry of entropic regularization

Tropical toric maximum likelihood estimation

Toric Fiber Products in Geometric Modeling

Likelihood Geometry of Reflexive Polytopes

Efficiently deciding if an ideal is toric after a linear coordinate change

Linear Convergence in Hilbert's Projective Metric for Computing Augustin Information and a Rényi Information Measure

Maximum Likelihood Estimation for Nets of Conics

Maximal Information Divergence from Statistical Models defined by Neural Networks

Separating Cones defined by Toric Varieties: Some Properties and Open Problems

Geometry of rational quasi-independence models as toric fiber products

Toric supervarieties with one odd dimension

Toric Multivariate Gaussian Models from Symmetries in a Tree

The minimal fibering degree of a toric variety equals the lattice width of its polytope

Maximizing Multivariate Information With Error-Correcting Codes