{"id":"arxiv/2605.14846","url":"https://arxiv.org/abs/2605.14846","tags":["math.OC"],"score":0,"title":"Successive convex optimization for transformer encoder model predictive control","domain":"math.OC","source":"arxiv","authors":["Xingxiao Chen","Mark Cannon"],"reviews":0,"subject":"math.OC","summary":"We propose a data-driven Model Predictive Control (MPC) framework that employs a transformer encoder to generate multi-step predictions. To handle the nonconvex attention mechanism, we derive difference of convex (DC) representations of the transformer encoder components and embed them in a successive convex programming (SCP) iteration. Recursive feasibility and convergence of the SCP iterates are guaranteed, and each iterate yields a solution estimate satisfying the problem constraints. Under mild assumptions, the SCP iteration converges to a locally optimal solution of the MPC problem. The…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-05-14","reactions":{}}