A new way to calculate energy bounds for electronic systems
Researchers developed a mathematical model to solve complex energy problems using a new optimization method.
A new computational model calculates spin-free sum-of-squares (SOS) energy bounds for electronic Hamiltonians. This model uses an exact reformulation of an established spin-free SOS model.
The researchers turned the problem into a convex quadratic problem. This problem uses a single symmetric matrix constrained between zero and the identity. They used a dimension-reduced simplicial method (DRSM) to solve it.
The dimension-reduced simplicial method (DRSM) is a cached implementation of fully corrective Frank--Wolfe. The Frank--Wolfe gap is the separation between recovered primal and dual objectives. Coulomb contractions are implemented through GPU-accelerated actions.
The method was tested on 65 FCIDUMP models, molecular geometries, and regular hydrogen rings. On eight H100 GPUs, a calculation for with 10,080 spatial orbitals reached a Frank--Wolfe gap of hartree in 5.02 hours.
Full-space hydrogen-ring calculations reached 10,000 orbitals. These results show that exact reformulation and reuse of Coulomb actions make optimizing the selected large-scale SOS relaxation practical.
Key facts
- The model reaches a Frank--Wolfe gap of hartree for in 5.02 hours.
- The dimension-reduced simplicial method (DRSM) uses cached operator products.
- Calculations reached 10,000 orbitals for full-space hydrogen rings.