DeePKS-kit

**DeePKS-kit** is a package for developing machine-learned XC functionals for quantum chemistry. It implements both perturbative (DeePHF) and self-consistent (DeePKS) schemes, integrating with PySCF for accurate energy functionals.

10. NICHE & ML 10.4 ML XC Functionals VERIFIED
Back to Directory Official Website

Overview

**DeePKS-kit** is a package for developing machine-learned XC functionals for quantum chemistry. It implements both perturbative (DeePHF) and self-consistent (DeePKS) schemes, integrating with PySCF for accurate energy functionals.

Reference Papers

Reference papers are not yet linked for this code.

Full Documentation

Official Resources

  • Source Repository: https://github.com/deepmodeling/deepks-kit
  • License: Open source (LGPL-3.0)

Overview

DeePKS-kit is a package for developing machine-learned XC functionals for quantum chemistry. It implements both perturbative (DeePHF) and self-consistent (DeePKS) schemes, integrating with PySCF for accurate energy functionals.

Scientific domain: ML XC functional development (perturbative and self-consistent)
Target user community: Researchers developing and applying ML XC functionals

Theoretical Methods

  • DeePKS (self-consistent scheme)
  • DeePHF (perturbative scheme)
  • Neural network XC functional
  • PySCF integration
  • ABACUS integration

Capabilities (CRITICAL)

  • Self-consistent DeePKS
  • Perturbative DeePHF
  • PySCF and ABACUS integration
  • Neural network XC training
  • Transferable functionals

Sources: GitHub repository

Key Strengths

Dual Scheme:

  • Self-consistent (DeePKS)
  • Perturbative (DeePHF)
  • Both approaches available
  • Accuracy vs cost tradeoff

Integration:

  • PySCF (molecular)
  • ABACUS (periodic)
  • DP-GEN workflow
  • DeepModeling ecosystem

Inputs & Outputs

  • Input formats: Training data (QM calculations)
  • Output data types: Trained XC models, corrected energies

Interfaces & Ecosystem

  • PySCF: Molecular DFT
  • ABACUS: Periodic DFT
  • DP-GEN: Active learning
  • Python: Core

Performance Characteristics

  • Speed: Similar to hybrid DFT (SC)
  • Accuracy: Near coupled-cluster
  • System size: Molecular and periodic
  • Automation: Full

Computational Cost

  • SC calculation: Similar to hybrid
  • Training: Hours on GPU

Limitations & Known Constraints

  • Training required: Need reference data
  • PySCF/ABACUS only: Limited code support
  • Complex setup: Multi-step workflow
  • LGPL license: Copyleft

Comparison with Other Codes

  • vs Skala: DeePKS is perturbative+SC, Skala is message-passing
  • vs DM21: DeePKS is trainable, DM21 is fixed
  • vs NeuralXC: DeePKS is more sophisticated
  • Unique strength: Dual perturbative/self-consistent ML XC with PySCF and ABACUS integration

Application Areas

Accurate DFT:

  • Beyond-DFT accuracy at DFT cost
  • Molecular property prediction
  • Periodic system calculations
  • Multi-level calculations

XC Development:

  • Custom XC functional training
  • Active learning for XC
  • Transfer learning

Best Practices

  • Start with perturbative scheme
  • Validate against high-level QM
  • Use DP-GEN for training data
  • Fine-tune for target chemistry

Community and Support

  • Open source (LGPL-3.0)
  • DeepModeling community
  • GitHub repository

Verification & Sources

Primary sources:

  1. GitHub: https://github.com/deepmodeling/deepks-kit

Confidence: VERIFIED

Verification status: ✅ VERIFIED

  • Source code: ACCESSIBLE (GitHub)
  • Specialized strength: Dual perturbative/self-consistent ML XC with PySCF and ABACUS integration

Related Tools in 10.4 ML XC Functionals