{"id":"arxiv/2603.17934","url":"https://arxiv.org/abs/2603.17934","tags":["math.NA","math.OC"],"score":0,"title":"State-dependent temperature control in Langevin diffusions using numerical exploratory Hamiltonian-Jacobi-Bellman equations","domain":"math.NA","source":"arxiv","authors":["Taorui Wang","Xun Li","Gu Wang","Zhongqiang Zhang"],"reviews":0,"subject":"math.NA","summary":"Choosing how much noise to add in Langevin dynamics is essential for making these algorithms effective in challenging optimization problems. One promising approach is to determine this noise by solving Hamilton-Jacobi-Bellman (HJB) equations and their exploratory variants. Though these ideas have been demonstrated to work well in one dimension, extension to high-dimensional minimization has been limited by two unresolved numerical challenges: setting reliable control bounds and stably computing the second-order information (Hessians) required by the equations. These issues and the broader…","category":"research","comments":0,"disputed":false,"division":"math","mentions":0,"discussed":0,"published":"2026-03-18","reactions":{}}