GO-Diff: Data-free and amortized global structure optimization

Nikolaj Rønne, Tejs Vegge, Arghya Bhowmik

Keywords: Generative Diffusion Model, Global Structure Optimization, Amortized Optimization

TL;DR: Boltzmann-weighted diffusion model for global structure optimization

Abstract:

We introduce GO-Diff, a diffusion-based method for global structure
optimization that learns to directly sample low-energy atomic
configurations without requiring prior data or explicit
relaxation. GO-Diff is trained from scratch using a Boltzmann-weighted
score-matching loss, leveraging only the known energy function to
guide generation toward thermodynamically favorable regions. The
method operates in a two-stage loop of self-sampling and model
refinement, progressively improving its ability to target low-energy
structures. Compared to traditional optimization pipelines, GO-Diff
achieves competitive results with significantly fewer energy
evaluations. Moreover, by reusing pretrained models across related
systems, GO-Diff supports amortized optimization — enabling faster
convergence on new tasks without retraining from scratch.

Submission Track: Paper Track (Short Paper)

Submission Category: AI-Guided Design

Institution Location: Copenhagen, Denmark

AI4Mat Journal Track: Yes

AI4Mat RLSF: Yes

Submission Number: 2