{"id":"arxiv/2607.13701","url":"https://arxiv.org/abs/2607.13701","tags":["math.OC","eess.SY"],"score":0,"title":"Lifting-Free Quadratic Sum-Of-Squares Programming","domain":"math.OC","source":"arxiv","authors":["Gabriel F. Machado","Ross Drummond","Morgan Jones"],"reviews":0,"subject":"math.OC","summary":"Quadratic Sum-Of-Squares (QSOS) optimization problems appear in system identification and machine learning, but standard Schur-complement and second-order cone liftings enlarge conic dimensions and create computational bottlenecks for interior-point methods. This paper introduces a lifting-free regularization that preserves the original conic structure by adding a norm penalty to SOS variables, yielding closed-form primal updates and an unconstrained, concave dual with Lipschitz-continuous gradient. Accelerated first-order methods efficiently maximize this dual, and convergence analysis…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-07-15","reactions":{}}