{"id":"arxiv/2610.32299","url":"https://arxiv.org/abs/2610.32299","tags":["physics.comp-ph","cs.LG","math.NA","physics.flu-dyn"],"score":0,"title":"CoPiT: A conditional Position-induced Transformer for aerodynamic field prediction on unstructured meshes","domain":"physics.comp-ph","source":"arxiv","authors":["Junfeng Chen","Yongchen Fan","Zhonghua Qiao","Kailiang Wu"],"reviews":0,"subject":"physics.comp-ph","summary":"High-fidelity Reynolds-averaged Navier--Stokes (RANS) simulations remain a standard tool for aerodynamic analysis, yet each solve on an unstructured mesh is computationally expensive, making many-query tasks such as design exploration, optimization, and multi-condition evaluation costly. Neural operators provide mesh-flexible surrogates for mapping geometry and operating conditions to flow fields; however, in many existing architectures, global parameters such as the freestream state and design variables are concatenated with pointwise inputs, leaving their influence on intermediate features…","category":"research","comments":0,"disputed":false,"division":"physics","mentions":0,"discussed":0,"published":"2026-10-23","reactions":{}}