DeePKS

Overview

DeePKS-kit supports the development of learned energy functionals for quantum-chemistry systems. The README distinguishes a perturbative scheme, DeePHF, from a self-consistent scheme, DeePKS. Rather than identifying a single pretrained predictive model, the linked sources describe a Python library and command-line toolkit for training models and using them in electronic-structure calculations.

Its deepks interface divides the workflow into five operations. train fits a neural-network-based post-HF energy functional, while test evaluates a post-HF model against supplied data and reports statistics. scf performs a self-consistent field calculation using a supplied energy model, and stats collects statistics from SCF results. iterate combines these operations into iterative training of a self-consistent model. The documented inputs therefore include model-training or testing data, an energy model for SCF calculations, and configuration parameters; outputs include trained models, calculation results, and reported statistics. The excerpts do not specify complete data schemas or output formats.

The README points to examples for a single water molecule and for training on water clusters, with separate configuration guidance for local execution rather than Slurm scheduling. It also links a methods paper for further background. PyTorch and PySCF must be installed separately because the package does not install them automatically. Although the project describes its goal in terms of accurate energy functionals, the source excerpts contain no benchmark results establishing accuracy, transferability, or computational cost for a particular system.

Key Features

• Supports both perturbative DeePHF and self-consistent DeePKS energy-functional workflows.
• Provides `deepks train` for fitting neural-network-based post-HF energy functional models.
• Provides `deepks test` for evaluating post-HF models on supplied data and reporting statistics.
• Runs self-consistent field calculations with a supplied energy model through `deepks scf`, with result statistics collected through `deepks stats`.
• Combines training, testing, SCF calculation, and statistics operations through `deepks iterate` for self-consistent model training.
• Documents single-water and water-cluster examples, including input-parameter guidance and a local-execution configuration alternative to Slurm.

Use Cases

• Intended evaluation: use the single-water example to examine the documented workflow before adapting it to another quantum-chemistry system.
• Intended evaluation: train and test a post-HF energy functional on suitable supplied data, using the reported statistics to assess its behavior.
• Intended evaluation: explore iterative self-consistent model training with the water-cluster example and compare results under a defined evaluation protocol.

How to Use

1. Read the project README to choose between the perturbative DeePHF and self-consistent DeePKS workflows. Consult the linked methods paper for scientific context.
2. Prepare a Python environment following the README. Install PyTorch and PySCF separately before installing DeePKS-kit; these requirements are not installed automatically.
3. Follow the README’s repository installation procedure, then inspect its linked single-water example as an initial evaluation case rather than assuming suitability for a new system.
4. For water-cluster training, inspect the README’s linked args.yaml parameter guidance. Review the linked shell.yaml configuration if you intend to execute locally instead of using Slurm.
5. Select the documented train, test, scf, stats, or iterate operation for your workflow. Assess the resulting statistics on your chosen data; the source excerpts do not establish performance outside the documented examples.