Opening Scaling Density Functional Theory with Gaussian Splatting (GS-DFT)
Scaling Density Functional Theory with Gaussian Splatting (GS-DFT)
- Length
- 55 min
- Level
- Research
- Recorded
- Microsoft Research2026 · Public Lecture
Chapter 1 of 9: Before the lecture
Optimizes molecular orbitals via gradient descent over 3D Gaussian clouds to scale density functional theory.
Scaling density functional theory with Gaussian Splatting optimizes molecular orbitals via gradient descent over 3D Gaussian clouds without training data. The method represents molecular orbitals as a cloud of Gaussians whose positions, shapes, and mixing coefficients are optimized jointly to minimize the energy functional. Adaptive density fitting with screening evaluates two-electron integrals efficiently, while regularized differentiable orthogonalization maintains the required constraints. Empirical results demonstrate that the optimized basis matches large conventional basis sets with fewer parameters and better memory scaling.
Scaling density functional theory with Gaussian Splatting optimizes molecular orbitals via gradient descent over 3D Gaussian clouds without training data. The method represents molecular orbitals as a cloud of Gaussians whose positions, shapes, and mixing coefficients are optimized jointly to minimize the energy functional. Adaptive density fitting with screening evaluates two-electron integrals efficiently, while regularized differentiable orthogonalization maintains the required constraints. Empirical results demonstrate that the optimized basis matches large conventional basis sets with fewer parameters and better memory scaling.
OMol25 dataset
gs-dft paper
3D Gaussian Splatting for Vision
Paper on arXiv

This lecture
Choose a node to see how it connects to this lecture. What comes before is on the leftat the top. The ring is what it teaches. People and places are above. Other lectures and books are below.
Press a chosen node again to open it.
Your comment is kept while you sign up.Sign in