Investigating anharmonicities in polarization-orientation Raman spectra of acene crystals with machine learning

Authors: Paolo Lazzaroni, Shubham Sharma, Mariana Rossi

DOI: https://doi.org/10.1103/t4s2-45t8

Note: This repository contains input files, configuration files, and representative outputs. Full datasets and trajectory files are omitted due to size constraints, can be obtained upon request.

Repository Structure

alpha-ml/

Machine learning models and datasets for polarizabilties on anthracene.

• mace/ - MACE model files

  • anthracene.model - Trained MACE model for anthracene
  • train.xyz, test.xyz - Training and test datasets for model evaluation
  • train_mace_pol.sbatch - HPC batch submission script for model training
  • dfpt/control.in - FHI-aims input for DFPT calculations

• sagpr/ - SAGPR training files

  • train.xyz - Training dataset for SAGPR model
  • sagpr.sbatch - SAGPR sbatch training script with TENSOAP (https://github.com/alanmlewis/TENSOAP)

md-ml/

Machine learning models and datasets for MLIPs of anthracene and naphthalene.

• control.in-fhi-aims - FHI-aims input for DFT reference calculations

• MACE_input.sh - Shell script for MACE potential training

• n2p2_input.in - Configuration for N2P2 (Behler-Parrinello) neural network potential training

• anthra/ - Anthracene mace model and dataset

  • anthracene_trainset.extxyz - Extended XYZ format training set
  • anthra_float32_swa.model, anthra_float64_swa.model - Stochastic weight averaging models at different precision

• naph/ - Naphthalene mace model and dataset

  • naphthalene_trainset.extxyz - Training set for naphthalene
  • naphthalene_mace_tight.model - Tightly fitted MACE model

pigs/

Temperature-elevated path-integral coarse-graining (PIGS) delta potential model and generation. Fully based on https://github.com/venkatkapil24/Te-PIGS-spectroscopy-tutorial

• dataset.xyz - Dataset with centroid and physical forces

• train.sh - MACE training script

• deltaPMF_anthracene.model - Delta PMF model file

• dataset-generation/ - PIGS dataset generation

  • input.xml - input configuration for i-PI
  • nvt.centroid_force.extxyz, nvt.physical_force.extxyz - NVT ensemble trajectories (centroid and physical forces)

• production/ - Production PIGS runs

  • input.xml - i-PI input for PIGS production simulation (example usage)

ramanTensors/

Phonons and Raman tensor calculations for harmonic or RGDOS spectra.

• phonons/ - Phonon calculations with i-PI

  • minimize/ - Geometry optimization with MACE model

    • input.xml - i-PI input file
    • initial.xyz, optimized.xyz - Initial and optimized structures

  • phonons/

    • input.xml - Phonon calculation input with i-PI

• displacements-Rtensor.py - Generate +/- displaced structures along normal modes from referene geometries

• modes - Normal modes file from i-PI phonon calculation

• minus-displacement-example/ - Example calculation for negative displacements with i-PI replay mode

  • control.in, geometry.in - FHI-aims DFPT setup
  • input.xml - i-PI replay input file
  • init.xyz, minus.xyz - Initial and displaced geometries "trajectory"
  • pol-minus.pol_0 - Polarizability tensor for minus displaced structures

production-md/

Production molecular dynamics trajectories input files.

• input-nve.xml, input-nvt.xml - i-PI NVE and NVT input files

• rgdos-example/ - Example workflow for obtaining RGDOS correlation functions

  • 100K/

    • optimized.xyz - Optimized geometry

    • modes - Vibrational mode file printed by i-PI phonon calculation

    • pol-minus, pol-plus - Polarizability for +/- displaced geometries

    • 1x1x1/

      • nve/run1/ - NVE MD runs with i-PI example (1x1x1 cell)
        • nvt.md - MD output file
        • nvt.chk - NVT checkpoint file for NVE restart
        • nve.pos_0.xyz - MD trajectory
        • RESTART - i-PI input file
        • anthra100K.cif - Structure initialization for i-PI
        • ipi.out - i-PI simulation output
        • run_ase.py - Python script for ASE-based MACE model driver

---

Last updated: January 2026