GS-DFT rewrites DFT with Gaussian Splatting, simulating 2,742 atoms on four H200s

AllThingsApx · x · 2026-09-28

GS-DFT: bringing 3D Gaussian Splatting ideas into quantum chemistry

Researchers from Mila, Université de Montréal, Princeton and others propose GS-DFT (Gaussian Splatting for Density Functional Theory). Instead of fixed atom-centered basis sets, molecular orbitals are represented as a cloud of Gaussians whose positions, shapes and mixing coefficients are jointly optimized by gradient descent to minimize the energy — no training data needed. The authors describe it as "3D Gaussian splatting with the renderer replaced by quantum mechanics."

Two key solver components:

Adaptive density fitting with screening for efficient two-electron integral evaluation;
Regularized differentiable orthogonalization of molecular orbitals.

Results:

The optimized basis matches the accuracy of the largest conventional basis sets with a fraction of the parameters, converging systematically in energy, density and nuclear forces;
At equal parameter count it captures stretched-bond and anion physics that fixed bases only recover with specialized augmentation;
Peak memory scales quadratically with cloud size, enabling simulations of systems up to 2,742 atoms on four H200 GPUs.

The authors note the next frontier is reliable forces at lower compute cost.

Original post →

GS-DFT rewrites DFT with Gaussian Splatting, simulating 2,742 atoms on four H200s

More from Research