Skala

**Skala** is a deep learning exchange-correlation functional developed by Microsoft that achieves chemical accuracy for atomization energies. It uses message-passing for non-local correlation and integrates with PySCF for DFT calculations.

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

Overview

**Skala** is a deep learning exchange-correlation functional developed by Microsoft that achieves chemical accuracy for atomization energies. It uses message-passing for non-local correlation and integrates with PySCF for DFT calculations.

Reference Papers

Reference papers are not yet linked for this code.

Full Documentation

Official Resources

  • Source Repository: https://github.com/microsoft/skala
  • License: Open source (MIT)

Overview

Skala is a deep learning exchange-correlation functional developed by Microsoft that achieves chemical accuracy for atomization energies. It uses message-passing for non-local correlation and integrates with PySCF for DFT calculations.

Scientific domain: Deep learning XC functional with chemical accuracy
Target user community: Researchers needing ML-enhanced DFT with chemical accuracy

Theoretical Methods

  • Deep learning XC functional
  • Message-passing for non-local correlation
  • Chemical accuracy (<1 kcal/mol)
  • PySCF integration
  • GPU and CPU implementations

Capabilities (CRITICAL)

  • Chemical accuracy atomization energies
  • Non-local correlation via message passing
  • PySCF integration
  • GPU (Skala-GPU) and CPU (Skala-CPU)
  • Broad thermochemistry benchmarks

Sources: GitHub repository, arXiv:2506.14665

Key Strengths

Accuracy:

  • Chemical accuracy (<1 kcal/mol)
  • Better than hybrid functionals
  • Non-local correlation
  • Broad benchmark performance

Integration:

  • PySCF self-consistent
  • GPU acceleration
  • CPU fallback
  • Standard DFT workflow

Inputs & Outputs

  • Input formats: Molecular/periodic structures
  • Output data types: XC energy, potential, total energy

Interfaces & Ecosystem

  • PySCF: DFT engine
  • JAX/PyTorch: Backend
  • Python: Core

Performance Characteristics

  • Speed: Similar to hybrid DFT
  • Accuracy: Chemical accuracy
  • System size: Molecular
  • Automation: Full

Computational Cost

  • SCF: Similar to hybrid DFT
  • Training: Days on GPU

Limitations & Known Constraints

  • Molecular focus: Primarily molecules
  • PySCF only: Limited DFT code integration
  • New project: Still maturing
  • GPU preferred: For speed

Comparison with Other Codes

  • vs DM21: Skala has message-passing, DM21 is piecewise linear
  • vs DeePKS: Skala is standalone, DeePKS is perturbative
  • vs NeuralXC: Skala is more accurate, NeuralXC is simpler
  • Unique strength: Deep learning XC functional achieving chemical accuracy with message-passing non-local correlation

Application Areas

Accurate DFT:

  • Thermochemistry
  • Kinetics
  • Reaction energies
  • Benchmark studies

ML-DFT Research:

  • XC functional development
  • DFT accuracy improvement
  • Method comparison

Best Practices

  • Use PySCF for SCF calculations
  • Compare with standard functionals
  • Validate on target chemistry
  • Use GPU for speed

Community and Support

  • Open source (MIT)
  • Microsoft maintained
  • Azure AI integration

Verification & Sources

Primary sources:

  1. GitHub: https://github.com/microsoft/skala

Confidence: VERIFIED

Verification status: ✅ VERIFIED

  • Source code: ACCESSIBLE (GitHub)
  • Specialized strength: Deep learning XC functional achieving chemical accuracy with message-passing non-local correlation

Related Tools in 10.4 ML XC Functionals