Broadening access to Skala creates a faster path to predictive DFT

Microsoft Research has announced the release of Skala 1.1, an updated deep-learning exchange-correlation functional designed to enhance the accuracy of density functional theory (DFT) simulations. Trained on 2.5 times more data than its predecessor, Skala 1.1 outperforms many existing functionals, including some of the most computationally expensive hybrid functionals, across a broad range of chemical challenges. It has achieved top rankings in the GMTKN55 benchmark, which assesses thermochemistry, reaction kinetics, and non-covalent interactions.

The new version of Skala not only improves accuracy but also emphasizes accessibility. It is now integrated into popular electronic-structure software packages such as CP2K, Psi4, FHI-aims, ORCA, and VASP, enabling a wider community of scientists to leverage its capabilities. Microsoft Research has also introduced a living benchmark that continuously tracks the computational performance of Skala across different software implementations and hardware platforms, facilitating ongoing optimization and community-driven progress.

Skala’s development follows a philosophy of continuous improvement, with each release designed to supersede previous versions while maintaining computational efficiency. The functional provides highly accurate energies, electron densities, dipole moments, and molecular geometries, supported by a large collection of high-accuracy quantum chemistry reference data.

The integration efforts include rigorous validation to ensure consistent accuracy across different software packages, exemplified by the successful implementation and testing within CP2K. Performance benchmarking indicates that Skala delivers computational costs comparable to semi-local functionals on both CPUs and GPUs, with ongoing improvements expected as new optimizations are implemented.

Overall, these developments mark a significant step toward making predictive, high-accuracy DFT simulations more accessible and integrated into scientific and industrial workflows, supporting advances in chemistry, materials science, catalysis, energy, and drug discovery.