D4FT: A Deep Learning Approach to Kohn-Sham Density Functional Theory
Authors: Tianbo Li, Min Lin, Zheyuan Hu, Kunhao Zheng, Giovanni Vignale, Kenji Kawaguchi, A.H. Castro Neto, Kostya S. Novoselov, Shuicheng YAN
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments are carried out to demonstrate the advantage of our approach in terms of efficiency and stability. In addition, we show that our approach enables us to explore more complex neural-based wave functions. 6 EXPERIMENTS In this section, we demonstrate the accuracy and scalability of D4FT via numerical experiments on molecules. |
| Researcher Affiliation | Collaboration | Tianbo Li* a, e, Min Lin* a, f, Zheyuan Hua, b, Kunhao Zhenga, Giovanni Vignalec, Kenji Kawaguchib, A. H. Castro Netoc, d, Kostya S. Novoselovc, d, Shuicheng Yana a SEA AI Lab, b School of Computing, National University of Singapore, c Institute for Functional Intelligent Materials, National University of Singapore, d Centre for Advanced 2D Materials, National University of Singapore; {elitb, flinmin}@sea.com |
| Pseudocode | Yes | Algorithm 1 Self-consistent Field Optimization for Kohn-Sham Equation |
| Open Source Code | No | Our code will be available on https://github.com/sail-sg/d4ft. |
| Open Datasets | No | The paper uses standard chemical basis sets (e.g., "6-31g basis set", "STO-3g basis set") and molecules (e.g., "Hydrogen", "Methane", "Water", "Oxygen", "Ethanol", "Benzene", "carbon Fullerene molecules"). While these are standard in computational chemistry, the paper does not provide a direct URL, DOI, specific repository name, or formal citation for a dataset instance in the typical machine learning sense. |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits with specific percentages, counts, or references to predefined splits applicable to its computational chemistry experiments. |
| Hardware Specification | Yes | All the experiments with JAX implementation (D4FT, JAX-SCF) are conducted on an NVIDIA A100 GPU with 40GB memory. As a reference, we also test with Py SCF on a 64-core Intel Xeon CPU@2.10GHz with 128GB memory. |
| Software Dependencies | No | The paper mentions key software like "JAX (Bradbury et al., 2018)" and "Py SCF (Sun et al., 2018)", and also "GPAW" and "Psi4" in Appendix E.3. However, it does not provide specific version numbers for these software dependencies, which are necessary for full reproducibility. |
| Experiment Setup | Yes | For D4FT, we use the Adam optimizer (Kingma & Ba, 2014) with a piecewise constant learning rate decay. The initial learning rate is 0.1, it is reduced to 0.01 to 0.0001 after 60 epochs. In total, we run 200 epochs for each molecule. A 3-layer MLP with tanh activation is adopted for gθ. The hidden dimension at each layer is 9. |