NARA: Nature-inspired Algorithm for Robust Atomic structure search

NARA is an atomic structure exploration framework based on the Firefly algorithm. The name NARA originates from the Korean expression nal-a (날-아), which represents the flying motion of a firefly. It has been developed as a companion package to LLUMYS, a machine learning interatomic potential that provides the "light" of the firefly.

Prerequisites

• llumys
• Python ≥ 3.10
• ASE ≥ 3.26.0

Earlier versions are compatible, but 3.26.0 or later is recommended because optimizer.irun now explicitly includes gradient evaluation/log

• scikit-learn
• scipy
• spglib
• e3nn
• pytorch ≥ 2.4

Please follow the installation guide on the official PyTorch website

• (optional) openequivariance, GCC≥9

For additional GPU acceleration. Requires installation of the matching CUDA toolkit for your PyTorch and CUDA driver versions.