Convergence rates of moment--SOS hierarchies with correlative sparsity via gluing and projection
This paper studies convergence rates of moment--SOS hierarchies with correlative sparsity. Following the method proposed in~\cite{tran}, we bound the Hausdorff distance between the set of sparse truncated moment sequences and sparse truncated pseudo-moment sequences. The central step is a quantitative gluing lemma on a clique tree: local moment errors and separator discrepancies in total variation control the error of a joint approximating measure. Over hypercube, polynomial kernels give the convergence rates $O((\log_2r)^3/r^2)$ for the correlative sparse quadratic module and $O(r^{-2})$ for the correlative sparse preordering, where the certificate degree is $2r$. On compact basic semialgebraic sets, slack variable lifting and global atom projection yield $O((\log_2r)^{3\Lo/2}/r^{\Lo})$ and $O(r^{-\Lo})$ with a Łojasiewicz exponent $0<\Lo \le1$. These bounds hold for fixed-degree Hausdorff distances and polynomial objective gaps. We also extend to result to the correlative sparse quadratic-module rate for scalar objectives with clique-local polynomial matrix constraints.
Authors
- Kim-Chuan Toh (ORCID: https://orcid.org/0000-0001-7204-8933)
- Hoang Tran (ORCID: https://orcid.org/0000-0001-7068-5621)
Publication Details
- Journal
- arXiv (Cornell University)
- Published
- 2026-11-16
- DOI
- https://doi.org/10.48550/arxiv.2611.20933
- Primary Topic
- Advanced Optimization Algorithms Research
- Type
- preprint