{"id":"arxiv/2608.19847","url":"https://arxiv.org/abs/2608.19847","tags":["math.OC"],"score":0,"title":"A Fixed-Penalty Linearized Augmented Lagrangian Method with Classical Multiplier Updates","domain":"math.OC","source":"arxiv","authors":["Benqi Liu","Kangkang Deng","Zichen Wang","Zaiwen Wen"],"reviews":0,"subject":"math.OC","summary":"Augmented Lagrangian methods are effective for nonlinear equality-constrained optimization, but solving their nonlinear primal subproblems can be expensive. For smooth nonconvex problems with deterministic or stochastic objectives, we propose a nonlinear-residual linearized augmented Lagrangian method (NR-LALM) that replaces this subproblem by a regularized Gauss-Newton-type step while retaining the classical multiplier update based on the nonlinear constraint residual. The resulting step is computed from one symmetric positive-definite linear system, but the mismatch between the linearized…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-20","reactions":{}}