Predicting challenging phase transitions with Bayesian active learning

This repository contains all the data to reproduce the calculations of the article: L. Bastonero, G. Joalland, C. Cignarella, L. Monacelli, and N. Marzari, Predicting challenging phase transitions with Bayesian active learning (2026).

It contains:

• archive.aiida: AiiDA archive containing the data and provenance of the simulations performed using AiiDA (see also How to navigate)
• CsPbI3.tar.gz and Li2O.tar.gz: tarballs of the corresponding material containing the following folders:
  • analysis: contains jupyter notebooks to reproduce the main figures of the article.
  • datasets: contains .extxyz format files representing the datasets from reference SSCHA-DFT and active learning calculations
  • The following folders are found under alpha_black and delta_yellow subfolders for CsPbI3, and under no subfolder for Li2O:
    • vcrelax_dft: contains the script to run a geometry optimization with DFT (pw.x binary of Quantum ESPRESSO) using aiida-quantumespresso
    • phonons_dfpt: contains the script to run a harmonic phonon calculation with DFPT (ph.x binary of Quantum ESPRESSO) using aiida-quantumespresso
    • sscha_active_learning: contains the script to run and related results of the active learning SSCHA with FLARE and AiiDA
    • sscha_dft_reference (for alpha-CsPbI3 and Li2O only): contains the script to run and related results of the reference SSCHA-DFT calculations
    • sscha_ml (for CsPbI3 only): contains the script to run and related results of the production SSCHA calculations with the ML potential to calculate the phase diagram of CsPbI3
  • train (for CsPbI3 only): training of the production ML potential using the pruned dataset of alpha and delta phases of CsPbI3
  • pseudos (for CsPbI3 only): the pseudopotentials for Cs, Pb, and I
• utilities.tar.gz: other useful scripts used throughout the study

How to reproduce

The calculations can be performed and reproduced with the following AiiDA plugins and codes:

• aiida-core (version >= 2.3)
• aiida-quantumespresso (version >= 4.4.0)
• Quantum ESPRESSO, pw.x and ph.x binaries (version >= 7.2)
• python-sscha (version >=1.7.0)
• cellconstructor (version >=1.7.0)

How to navigate

The AiiDA database can be explored either directly on Materials Cloud Archive or by importing the archive.aiida archive in a local AiiDA installation (after having installed the dependencies; see How to reproduce).