LP/SDP Hierarchies and Sum of Squares Proofs 2
summary
Real infeasibility certificates appear as an f + g = -1 certificate with f in the ideal of equations and g in the cone generated by inequalities; bounding the syntactic degree turns the search into a degree-bounded SDP that is solvable in practice. Pathologies include rational p that require irrational q or huge bit-complexity dependence, so one works with epsilon-approximations. Over the hypercube a delta-function SOS gives a trivial certificate; practical reductions use symmetry reduction and sparsity or quotient rings to keep the SDP tractable.
AI-written summary · about
chapters (13)
- 0:00Intro
- 11:05Inequality
- 14:50Possible constraints
- 18:20The real notional sense
- 21:40Infeasible equations
- 27:25polynomial equations
- 32:50non polynomial equations
- 38:00algebraically compact
- 41:28why is this even true
- 45:05why do I do this
- 50:00what matters
- 51:25semidefinite problems
- 52:50Explicit equations
related talks
44mCombinatorial Characterizations in Semidefinite Programming Duality: ...
01h 04mFrom Proofs to Algorithms for Machine Learning Problems
39mOptimization over the Hypercube via Sums of Nonnegative Circuit Polynomials
59mCertificates of Polynomial Non-Negativity via Hyperbolic Optimization
42mTight Semidefinite Programming Relaxations for Polynomial Optimization
32mLocal Statistics, Semidefinite Programming, and Community Detection
46mFaster SDP Hierarchy Solvers for Local Rounding Algorithms
31mA Positivity-First Approach to Sum-of-Squares Over the Hypercube
