In SDP Relaxations, Inaccurate Solvers Do Robust Optimization
From MaRDI portal
Abstract: We interpret some wrong results (due to numerical inaccuracies) already observed when solving SDP-relaxations for polynomial optimization on a double precision floating point SDP solver. It turns out that this behavior can be explained and justified satisfactorily by a relatively simple paradigm. In such a situation, the SDP solver (and not the user) performs some `robust optimization' without being told to do so. Instead of solving the original optimization problem with nominal criterion , it uses a new criterion which belongs to a ball of small radius , centered at the nominal criterion in the parameter space. In other words the resulting procedure can be viewed as a `' robust optimization problem with two players (the solver which maximizes on and the user who minimizes over the original decision variables). A mathematical rationale behind this `autonomous' behavior is described.
Recommendations
- A recursive algorithm of exactness verification of relaxations for robust SDPs
- Robust SOS-convex polynomial optimization problems: exact SDP relaxations
- A relaxation algorithm with a probabilistic guarantee for robust deviation optimization
- On approximate solutions for robust convex semidefinite optimization problems
- A convergent hierarchy of SDP relaxations for a class of hard robust global polynomial optimization problems
- On approximate solutions and saddle point theorems for robust convex optimization
- Robust optimizers for nonlinear programming in approximate dynamic programming
- Exact relaxations for parametric robust linear optimization problems
- A study on robust infeasibility of semidefinite programming
- Robustness in nonsmooth nonconvex optimization problems
Cites work
- A paradox in bosonic energy computations via semidefinite programming relaxations
- A Sum of Squares Approximation of Nonnegative Polynomials
- Bad semidefinite programs: they all look the same
- Certifying convergence of Lasserre's hierarchy via flat truncation
- Detecting Global Optimality and Extracting Solutions in GloptiPoly
- Exact algorithms for linear matrix inequalities
- Global optimization with polynomials and the problem of moments
- How to generate weakly infeasible semidefinite programs via Lasserre's relaxations for polynomial optimization
- scientific article; zbMATH DE number 1489799 (Why is no real title available?)
- On exact Polya and Putinar's representations
- On general minimax theorems
- Optimality conditions and finite convergence of Lasserre's hierarchy
- Optimization of polynomials in non-commuting variables
- Robust optimization
- Robust Solutions to Uncertain Semidefinite Programs
- Strange behaviors of interior-point methods for solving semidefinite programming problems in polynomial optimization
- Strong duality in lasserre's hierarchy for polynomial optimization
- Using SeDuMi 1.02, A Matlab toolbox for optimization over symmetric cones
- {\textsc{RealCertify}}: a Maple package for certifying non-negativity
Cited in
(6)- Conic programming: infeasibility certificates and projective geometry
- Positivity certificates and polynomial optimization on non-compact semialgebraic sets
- Sieve-SDP: a simple facial reduction algorithm to preprocess semidefinite programs
- Computing the Hausdorff boundary measure of semialgebraic sets
- Validating numerical semidefinite programming solvers for polynomial invariants
- Validating numerical semidefinite programming solvers for polynomial invariants
This page was built for publication: In SDP Relaxations, Inaccurate Solvers Do Robust Optimization
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q5233101)
