GO-Diff

GO-Diff: Data-free and amortized global structure optimization a generative diffusion framework for atomistic structure search. It couples a diffusion model with an energy/force calculator and iteratively refines the model through Boltzmann-weighted training, adaptive temperature annealing, and an adaptive replay buffer.

Installation

Install from pypi

pip install go-diff

(Optional) Install of MLIPs for running GO-Diff

E.g.

pip install mace-torch

Quick Start

import numpy as np

from ase.build import fcc111, surface
from mace.calculators import mace_mp

from agedi import AtomsGraph, create_diffusion
from agedi.diffusion import ForcefieldGuidanceConfig

from go_diff import GODiff, MinEnergyFilter
from go_diff.controllers import SampleController, BufferController, TemperatureSchedule, MomentumConsensusStop
from go_diff.noisers import WeightedConfinedCellPositions

##### SYSTEM #####
formula = "Pt"

template = surface('Pt', (1,2,2), 5, vacuum=8.0)
template.positions[:, 2] -= template.positions[:, 2].min()

confinement_above_zmax = np.array([0.0, 4.0])
confinement = confinement_above_zmax + template.positions[:, 2].max()


##### CALCULATOR #####
from mace.calculators import mace_mp
calc = mace_mp(model="medium", dispersion=False, default_dtype="float32", device='cuda')


#### DIFFUSION MODEL #####
diffusion = create_diffusion(noisers=(WeightedConfinedCellPositions(),), force_field=True)


#### GO-DIFF #####
godiff = GODiff(
    calculator=calc,
    diffusion=diffusion,
    temperature_schedule=TemperatureSchedule(fast=0.5, slow=0.9),
    sample_controller=SampleController(initial_N=16, target_ess=8),
    buffer_controller=BufferController(initial_buffer_size=16, max_buffer_size=96, adaption_rate=0.2),
    training_controller=MomentumConsensusStop(min_steps=100, patience=250, drop_factor=0.9),
    sample_config={
        "template": template,
        "formula": formula,
        "confinement": confinement,
        "ff_guidance": ForcefieldGuidanceConfig(guidance=1.0,)
    },
    dataset_config={
        "mask": "MaskFixed",
        "confinement": confinement,
        "regressor_data": "all_data", # use all data for training the regressor, not just the data in the buffer
    },
    trainer_config={
        "name": name
    },
    min_E=min_E,
    valid_structure_filters=[MinEnergyFilter(-200)],    
)

#### RUN GO-Diff #####
godiff.run(max_iterations=20)

See scripts/*.py for the full scripts.

Key components

Running the tests

<!-- CLONE package -->
git clone https://github.com/nronne/go-diff.git 
cd go-diff

<!-- INSTALL with pip including test extras-->
pip install ".[test]"

<!-- RUN pytest -->
pytest

Citation

If you use GO-Diff please cite:

Ronne, Nikolaj, Tejs Vegge and Arghya Bhowmik. "GO-Diff: Data-free and amortized global structure optimization" arXiv:2510.13448. Preprint, arXiv, October 15, 2025. https://arxiv.org/abs/2510.13448

Ronne, Nikolaj, and Bjork Hammer. "Atomistic Generative Diffusion for Materials Modeling." arXiv:2507.18314. Preprint, arXiv, July 24, 2025. https://arxiv.org/abs/2507.18314

If studying surface-supported systems please also cite:

Ronne, Nikolaj, Alan Aspuru-Guzik, and Bjork Hammer. "Generative Diffusion Model for Surface Structure Discovery." Physical Review B 110, no. 23 (2024): 235427. https://doi.org/10.1103/PhysRevB.110.235427

Optionally, if using any AGOX functionality please cite:

Christiansen, Mads-Peter V., Nikolaj Ronne, and Bjork Hammer. "Atomistic Global Optimization X: A Python Package for Optimization of Atomistic Structures." The Journal of Chemical Physics 157, no. 5 (2022): 054701. https://doi.org/10.1063/5.0094165