Towards Symmetry-Aware Generation of Periodic Materials

Authors: Youzhi Luo, Chengkai Liu, Shuiwang Ji

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

Reproducibility Variable Result LLM Response
Research Type Experimental In this section, we evaluate our proposed Sy Mat method in two periodic material generation tasks, including random generation and property optimization.
Researcher Affiliation Academia Youzhi Luo, Chengkai Liu, Shuiwang Ji Department of Computer Science & Engineering Texas A&M University College Station, TX 77843 {yzluo,liuchengkai,sji}@tamu.edu
Pseudocode Yes Algorithm 1: Langevin Dynamics Sampling Algorithm for Coordinate Generation
Open Source Code Yes Our code is publicly available as part of the AIRS library (https://github.com/divelab/AIRS).
Open Datasets Yes We evaluate Sy Mat on three benchmark datasets curated by Xie et al. [52], including Perov5 [3, 2], Carbon-24 [35], and MP-20 [17].
Dataset Splits Yes For all three datasets, we split them with a ratio of 3:1:1 as training, validation, and test sets in our experiments.
Hardware Specification Yes The total training needs around 3, 5, and 10 hours on a GTX 2080 GPU for Perov-5, Carbon-24, and MP-20 datasets, separately.
Software Dependencies No The paper mentions software components like Sphere Net, VAE, and MLPs, but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes In our Sy Mat model, the Sphere Net model is composed of 4 message passing layers and the hidden size is set to 128. Also, all MLP models in the VAE decoder is composed of 2 linear layers with a Re LU function between them, and the hidden size is set to 256. During training, we set the learning rate to 0.001, the batch size to 128, and the epoch number to 1,000.