Simplicial Regularizability of the Pseudo-Moment Cone and Carathéodory-Type Atomic Decomposition of Moment Matrices

Shucheng Kang, Heng Yang · 2026

Read Simplicial Regularizability of the Pseudo-Moment Cone and Carathéodory-Type Atomic Decomposition of Moment Matrices by Shucheng Kang, Heng Yang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-06854.txt.

Introduction

Shucheng Kang and Heng Yang study the geometry of the pseudo-moment cone, a structure underlying polynomial optimization, and how moment matrices break into atomic pieces. A moment matrix built from many weighted point evaluations should decompose into its rank-one atoms, but identifying the right face of the cone is the obstacle. The authors fix the degree and vary the dimension and number of atoms.

For generically chosen atoms in a stated regime, they prove the minimal face is simplicial and generated exactly by the planted atoms. That geometry yields a Carathéodory-type decomposition algorithm for spectrahedral cones, specialized into an efficient atomic decomposition for moment matrices. A stabilized numerical version recovers atoms strongly and may sample high-rank extreme rays beyond the proven regime.

Catalog record

CREATORShucheng Kang, Heng Yang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE279 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.06854

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Simplicial Regularizability of the Pseudo-Moment Cone and Carathéodory-Type Atomic Decomposition of Moment Matrices Shucheng Kang, Heng Yang Abstract We study the facial geometry of the (homogeneous) pseudo-moment cone Σ n , 2 ​ d ∗ \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} and its implications for atomic decomposition of moment matrices. For fixed d ≥ 2 \mathchar 29028\mathchar 12821\mathchar 28722 , we show that if a moment matrix is formed by O ⁡ ( n d ) \mathchar 29007\delimiter 67273472\mathchar 29038^{\mathchar 29028}\delimiter 84054785 generically chosen weighted atoms, then its minimal face in the matrix realization of the pseudo-moment cone is simplicial and generated by the planted rank-one atoms. Based on this geometric result, we develop a Carathéodory-type extreme-ray decomposition algorithm for spectrahedral cones and show that, when specialized to the pseudo-moment cone, it yields an efficient atomic decomposition method for generically generated moment matrices in the same regime. A stabilized numerical implementation demonstrates strong recovery performance, and suggests that outside the guaranteed regime the algorithm may serve as a practical sampler of high-rank extreme rays. 1 Introduction The (homogeneous) pseudo-moment cone is a fundamental geometric object in polynomial optimization, moment theory, and convex algebraic geometry. For instance, moment relaxations of polynomial optimization problems may be formulated as linear programs over this cone [ 22 , 28 ] , while atomic decomposition of a moment matrix may be interpreted as a conic decomposition into rank-one point-evaluation rays [ 3 , 17 ] . Following [ 3 ] , we denote this cone by \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} , where \mathchar 29038 is the number of variables and \mathchar 28722\mathchar 29028 is the degree. Despite its central role, our understanding of the facial structure of \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} remains limited. A major obstacle is the existence of higher-rank extreme rays—rays that do not correspond to point evaluations [ 4 ] . Existing results mainly focus on deciding the ranks of such rays and their construction [ 3 , 4 , 7 , 17 ] . In particular, there exist extreme rays of \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} of rank \mathchar 28723\mathchar 29028\mathchar 8704\mathchar 28722 for \mathchar 29028\mathchar 12821\mathchar 28723 and \mathchar 29038\mathchar 12821\mathchar 28723 , and of rank \mathchar 28726 for \mathchar 29028\mathchar 12349\mathchar 28722 and \mathchar 29038\mathchar 12821\mathchar 28724 [ 4 ] . The ternary case has also been studied in greater detail [ 7 ] . While these works reveal the complexity of the global facial structure of \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} , it is natural to ask whether simpler local geometric structures exist that can be leveraged for algorithm design. Contributions. We give an affirmative answer to the above question. Our contributions are threefold. • Simplicial regularizability of the pseudo-moment cone. Our main contribution is a new facial-geometric characterization of the pseudo-moment cone. For fixed \mathchar 29028\mathchar 12821\mathchar 28722 , we show that if a moment matrix is formed by \mathchar 29007\delimiter 67273472\mathchar 29038^{\mathchar 29028}\delimiter 84054785 generically chosen weighted atoms, then its minimal face in the matrix realization of the pseudo-moment cone is simplicial and generated by the planted rank-one atoms ( cf. Theorem 1 ). We call this property simplicial regularizability . Our proof is motivated by a geometric result on computing linear sections of varieties [ 20 , Theorem 15] . • Carathéodory-type atomic decomposition of moment matrices. Based on this facial-geometric result, we develop a Carathéodory-type extreme-ray decomposition procedure for spectrahedral cones; see Algorithm 1 . When specialized to {\cal\mathchar 29005}_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathsf{\mathchar 29040\mathchar 29043}} , the matrix realization of \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} , it yields an efficient atomic decomposition method for generically generated moment matrices with up to \mathchar 28674\delimiter 67273472\mathchar 29038^{\mathchar 29028}\delimiter 84054785 atoms. This leads to an asymptotic improvement over methods based on flatness extension, whose recovery upper bound is \mathchar 28674\delimiter 67273472\mathchar 29038^{\mathchar 29028\mathchar 8704\mathchar 28721}\delimiter 84054785 [ 18 , 21 ] , provided that one does not solve a hierarchy of semidefinite relaxations to generate high-order moments [ 25 ] . Conceptually, our algorithm can be viewed as a practical variant of the procedure underlying the Carathéodory-number bound in [ 19 , Theorem 4] . • Stabilized numerical implementation. We develop a stabilized numerical implementation of Algorithm 1 to improve its practical robustness. Experiments demonstrate strong recovery performance, suggest that the proven recovery regime is conservative, and indicate that outside the guaranteed regime the algorithm may also serve as a practical sampler of high-rank extreme rays of \mathchar 28678_{\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028}^{\mathchar 8707} . Algorithm 1 Carathéodory-Type Extreme Ray Decomposition 1: nonempty spectrahedral cone {\cal\mathchar 28995}\mathchar 12818\mathbb{\mathchar 29011}^{\mathchar 29006} , initial point \mathchar 29016\mathrel{{{{\mathchar 566\mathchar 61}}}}\mathchar 28720\mathchar 12850{\cal\mathchar 28995} 2: \{\mathchar 29044_{\mathchar 29035}\mathchar 24891\mathchar 29005_{\mathchar 29035}\}_{\mathchar 29035\mathchar 12349\mathchar 28721}^{\mathchar 29042} , with \mathchar 29016\mathchar 12349\mathchar 4944\displaylimits_{\mathchar 29035\mathchar 12349\mathchar 28721}^{\mathchar 29042}\mathchar 29044_{\mathchar 29035}\mathchar 29005_{\mathchar 29035} , \mathchar 29044_{\mathchar 29035}\mathchar 12606\mathchar 28720 and {{\mathbb{\mathchar 29010}}}_{\mathchar 8235}\mathchar 29005_{\mathchar 29035} an extreme ray of {\cal\mathchar 28995} for all \mathchar 28721\mathchar 12820\mathchar 29035\mathchar 12820\mathchar 29042 3: procedure RayDecomp ( {\cal\mathchar 28995}\mathchar 24891\mathchar 29016 ) 4: \mathchar 29035\mathchar 12346\mathchar 12349\mathchar 28721 , \mathchar 29016_{\mathchar 28721}\mathchar 12346\mathchar 12349\mathchar 29016 5: while \mathchar 29016_{\mathchar 29035}\mathrel{{{{\mathchar 566\mathchar 61}}}}\mathchar 28720 do 6: {\cal\mathchar 28998}_{\mathchar 29035}\mathchar 12346\mathchar 12349\mathrm{\mathchar 29037\mathchar 29033\mathchar 29038\mathchar 29030\mathchar 29025\mathchar 29027\mathchar 29029}\delimiter 67273472\mathchar 29016_{\mathchar 29035}\mathchar 24891{\cal\mathchar 28995}\delimiter 84054785 7: \mathchar 28994_{\mathchar 29035}\mathchar 12346\mathchar 12349 random matrix from any absolutely continuous distribution on \mathbb{\mathchar 29011}^{\mathchar 29006} 8: Generate \mathchar 29005_{\mathchar 29035}\mathchar 12850\mathbb{\mathchar 29011}^{\mathchar 29006} from the following linear SDP: \displaystyle\mathchar 29005_{\mathchar 29035}\mathchar 12850\argmin_{\mathchar 29017\mathchar 12850{\cal\mathchar 28998}_{\mathchar 29035}\mathchar 24891\mathrm{\mathchar 29044\mathchar 29042}\left\delimiter 67273472\mathchar 29017\right\delimiter 84054785\mathchar 12349\mathchar 28721}\left\delimiter 0\mathchar 28994_{\mathchar 29035}\mathchar 24891\mathchar 29017\right\delimiter 0 (1) 9: Find \displaystyle\mathchar 29044_{\mathchar 29035}\mathchar 12346\mathchar 12349\sup\{\mathchar 29044\mathchar 12821\mathchar 28720\mathchar 12906\mathchar 29016_{\mathchar 29035}\mathchar 8704\mathchar 29044\mathchar 29005_{\mathchar 29035}\mathchar 12850{\cal\mathchar 28998}_{\mathchar 29035}\} (2) 10: \mathchar 29016_{\mathchar 29035\mathchar 8235\mathchar 28721}\mathchar 12346\mathchar 12349\mathchar 29016_{\mathchar 29035}\mathchar 8704\mathchar 29044_{\mathchar 29035}\mathchar 29005_{\mathchar 29035} 11: \mathchar 29035\mathchar 12346\mathchar 12349\mathchar 29035\mathchar 8235\mathchar 28721 12: end while 13: return \mathchar 29042\mathchar 12346\mathchar 12349\mathchar 29035\mathchar 8704\mathchar 28721 , \{\mathchar 29044_{\mathchar 29035}\mathchar 24891\mathchar 29005_{\mathchar 29035}\}_{\mathchar 29035\mathchar 12349\mathchar 28721}^{\mathchar 29042} 14: end procedure Outline. The rest of the paper is organized as follows. § 2 reviews related work, and § 3 introduces the necessary notation and preliminaries. § 4 establishes the simplicial regularizability of the pseudo-moment cone. § 5 presents Algorithm 1 together with its stabilized implementation. Numerical experiments are given in § 6 , and § 7 concludes with a discussion of future directions. 2 Related Work Facial structures of the moment cone and the pseudo-moment cone. Prior work on pseudo-moment cones has focused mainly on extreme rays and related algebraic boundaries: the first nontrivial cases \delimiter 67273472\mathchar 29038\mathchar 24891\mathchar 28722\mathchar 29028\delimiter 84054785\mathchar 12349\delimiter 67273472\mathchar 28723\mathchar 24891\mathchar 28726\delimiter 84054785 and \delimiter 67273472\mathchar 28724\mathchar 24891\mathchar 28724\delimiter 84054785 were characterized via Cayley–Bacharach relations [ 5 , 3 , 6 ] ; positive Gorenstein ideals later provided a general framework and sharp rank thresholds for non-evaluation extreme rays [ 4 ] ; and, for ternary forms, Hankel spectrahedra and their algebraic boundaries were studied systematically [ 7 ] . Exact-arithmetic constructions of pseudo-moment certificates and extreme rays were proposed more recently in [ 17 ] . On the moment-cone side, truncated moment cones were

Read the complete text