A Unified Hard-Constraint Framework for Solving Geometrically Complex PDEs

Authors: Songming Liu, Hao Zhongkai, Chengyang Ying, Hang Su, Jun Zhu, Ze Cheng

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results on real-world geometrically complex PDEs showcase the effectiveness of our method compared with state-of-the-art baselines. We empirically demonstrate the effectiveness of our method through three parts of experiments.
Researcher Affiliation Collaboration 1Dept. of Comp. Sci. and Tech., Institute for AI, THBI Lab, BNRist Center, Tsinghua-Bosch Joint ML Center, Tsinghua University 2Peng Cheng Laboratory; Pazhou Laboratory (Huangpu), Guangzhou, China 3Bosch Center for Artificial Intelligence
Pseudocode No The paper describes its method using equations and textual descriptions but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] See the supplementary material.
Open Datasets Yes the 2D stationary incompressible Navier-Stokes equations, in the context of simulating the airflow around a real-world airfoil (w1015.dat) from the UIUC airfoil coordinates database (an open airfoil database) [29].
Dataset Splits No The paper specifies the number of collocation points for training and testing, but does not explicitly mention a separate validation set or split percentages.
Hardware Specification Yes The total amount of compute is around 50 GPU hours with NVIDIA V100 GPU.
Software Dependencies No The paper states that the method is 'implemented in PyTorch' but does not provide specific version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes We use Adam optimizer with a learning rate of 1e-3, and then use L-BFGS optimizer. The learning rate of Adam is decayed using cosine annealing schedule [29] (with a warm up of 1000 iterations). In each experiment, we sample Nf collocation points, Nb boundary points, and Ni initial points.