{"id":"arxiv/2610.23788","url":"https://arxiv.org/abs/2610.23788","tags":["stat.ML","cs.LG","eess.SY","math.OC"],"score":0,"title":"Parametric Resonance in Primal-Dual Learning for Slowly Non-Stationary Constrained MDPs","domain":"stat.ML","source":"arxiv","authors":["Maxime Baelde"],"reviews":0,"subject":"stat.ML","summary":"Primal-dual methods for constrained Markov decision processes are analyzed under a two-timescale assumption, with non-stationarity entering through two channels. In the additive channel, where the constraint level oscillates, the dual oscillation saturates at a floor set by the moving optimum and is amplified linearly by the timescale ratio. In the parametric channel, where the constraint sensitivity oscillates, a coherent modulation carries no resonant forcing at any depth: the dual update's stiffness modulation cancels the primal update's damping modulation. Instability instead comes from…","category":"research","comments":0,"disputed":false,"division":"stat","mentions":0,"discussed":0,"published":"2026-10-16","reactions":{}}