Learning Preconditioners for Conjugate Gradient PDE Solvers

Authors: Yichen Li, Peter Yichen Chen, Tao Du, Wojciech Matusik

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct extensive experiments to demonstrate the efficacy and generalizability of our proposed approach on solving various 2D and 3D linear second-order PDEs.
Researcher Affiliation Academia 1MIT CSAIL 2Tsinghua University 3Shanghai Qi Zhi Institute.
Pseudocode Yes Algorithm 1 PCG
Open Source Code No The paper provides a project webpage link (https://sites.google.com/view/neuralPCG) which is not a direct link to a source-code repository as specified in the guidelines.
Open Datasets No The dataset is generated by the authors through simulation: 'Our dataset is constructed by simulating trajectories of each PDE on a given mesh domain with various initial conditions and boundary conditions.' No concrete access information (link, DOI, specific citation) for a publicly available dataset was provided.
Dataset Splits No The paper mentions a 'training set' and 'test set' but does not explicitly describe a separate 'validation' set or its split.
Hardware Specification Yes All experiments are conducted using the same hardware setup equipped with 64-core AMD CPUs and an NVIDIA RTX-A8000 GPU.
Software Dependencies Yes All learning-based methods are written using the Pytorch (Paszke et al., 2019) Framework with the Pytorch-Geometric (Fey & Lenssen, 2019) Package and CUDA11.6.
Experiment Setup Yes We use Adam optimizer with the initial learning rate set to 1e-3. We use a batch size of 16 for all Heat-2d, Poisson-2d, and Wave-2d experimental environments. We use a batch size of 8 for the Poisson-3d environment. Table 6: GNN architecture hyper-parameter. Env Heat-2D Wave-2D Poisson-2D Poisson-3D l 1 1 2 2 h 16 16 16 16 nmp 5 5 5 3 lmp 1 1 2 2 hmp 16 16 16 16