SCFbench: Transferable Acceleration Data for Density Functional Theory

SCFbench provides data for accelerating Density Functional Theory (DFT) calculations using E(3)-equivariant neural networks, developed by the ByteDance Seed team in 2025. The dataset facilitates the prediction of electronic structure expansion coefficients to bypass or speed up traditional self-consistent field iterations. It is specifically designed to support the development of universally transferable acceleration methods in quantum chemistry.

Use Cases

• Predicting expansion coefficients for electronic structures to accelerate DFT
• Benchmarking the transferability of E(3)-equivariant neural networks
• Optimizing self-consistent field (SCF) iterations in molecular simulations

Strengths

• Linked to peer-reviewed methodology in Arxiv 2509.25724
• Supports E(3)-equivariant neural network training for physical symmetry
• Released under the permissive Apache 2.0 license

Limitations

• Highly specialized domain requiring expertise in quantum chemistry and DFT
• Specific column schemas and record counts are not detailed in the primary metadata

Provenance

Source

ByteDance Seed team, Arxiv 2509.25724

Freshness

Last updated March 2026.