Ab initio simulation of excited-state potential energy surfaces with transferable deep quantum Monte Carlo.

A geometrically transferable neural-network wavefunction method enables efficient, high-accuracy prediction of ground- and excited-state potential energy surfaces for ethylene, the carbon dimer, methylenimmonium, and a 96-electron rubredoxin active-site model.

• Why it matters: Computing full excited-state potential energy surfaces is difficult because accurate approaches are computationally demanding, forcing tradeoffs when modeling light excitation coupled to structural changes.
• What they did: The method uses weight sharing and a dynamical ordering of electronic states to optimize transferable neural-network wavefunctions, validated on four excited-state PESs.
• The result: It achieves up to two orders of magnitude cost reduction versus single-point calculations while reproducing ground- and excited-state PESs and their intersections at the highest accuracy.

AI summary of the abstract · may contain errors · not medical advice · read the paper.