{"id":"arxiv/2608.28988","url":"https://arxiv.org/abs/2608.28988","tags":["math.OC","cs.DC"],"score":0,"title":"B$^3$-PWL: GPU-Batched Branch-and-Bound for Piecewise-Linear Optimization with SOS2 Constraints","domain":"math.OC","source":"arxiv","authors":["Yilin Guan","Shuqing Luo","Pingzhi Li","Tianlong Chen","Kaidi Xu"],"reviews":0,"subject":"math.OC","summary":"Piecewise-linear (PWL) optimization problems arise in many mixed-integer programming (MIP) optimization applications, including portfolio optimization, workforce scheduling, and resource allocation. But solving them to global optimality remains computationally expensive because branch-and-bound repeatedly solves LP relaxation subproblems. Existing solvers are largely CPU-centric, leaving the scalability of modern GPUs underutilized. Few prior GPU-accelerated branch-and-bound either targets neural network which is not suitable for general PWL optimization, or accelerates only auxiliary…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-08-29","reactions":{}}