notesum.ai

Published at November 25

VICON: Vision In-Context Operator Networks for Multi-Physics Fluid Dynamics Prediction

math.NA

physics.flu-dyn

Released Date: November 25, 2024

Authors: Yadi Cao, Yuxuan Liu, Liu Yang, Rose Yu, Hayden Schaeffer, Stanley Osher

Arxiv: http://arxiv.org/abs/2411.16063v1

Refer to caption

Rollout Rescaled $L^{2}$ Error	Case	Ours (single step)	Ours (multi step)	MPP-B (McCabe et al., 2023)
Step 1 [1e-2]	PDEArena-Incomp	11.10	11.10	6.17
	PDEBench-Comp-LowVis	15.61	15.61	16.37
	PDEBench-Comp-HighVis	5.97	5.97	20.90
Step 5 [1e-2]	PDEArena-Incomp	23.00	25.79	14.81
	PDEBench-Comp-LowVis	24.56	39.54	45.45
	PDEBench-Comp-HighVis	19.73	28.24	42.93
Step 10 [1e-2]	PDEArena-Incomp	36.18	35.02	27.89
	PDEBench-Comp-LowVis	37.47	51.50	64.11
	PDEBench-Comp-HighVis	57.88	79.52	144.46
Last Step [1e-2]	PDEArena-Incomp	77.81	70.83	93.52
	PDEBench-Comp-LowVis	39.03	48.82	65.32
	PDEBench-Comp-HighVis	71.17	96.23	185.29
All average [1e-2]	PDEArena-Incomp	56.27	51.32	55.95
	PDEBench-Comp-LowVis	27.08	36.29	46.68
	PDEBench-Comp-HighVis	30.06	40.06	72.37