/ML-quantum-vibrational-spectroscopy

Dipole and polarizability models developed using the approach Kapil, V., Kovács, D. P., Csányi, G., & Michaelides, A. (2023). First-principles spectroscopy of aqueous interfaces using machine-learned electronic and quantum nuclear effects. Faraday Discussions, 10.1039.D3FD00113J. https://doi.org/10.1039/D3FD00113J

MIT LicenseMIT

MACE dielectric response Model for bulk water

A MACE model developed in Ref. 1 predicts the dipole moment, polarization, and their real space derivatives of bulk water. The dataset is taken from Ref. 2.

The following repository and commit were used to generate the model:

  • Repository: git@github.com:venkatkapil24/mace.git
  • Commit: ea9178fc7d6cdb54eb850623e09000c5ae27243e

The following CLI command was used:

mace_run_train \
    --name="MACE_dipole_pol" \
    --train_file="_INSERT_" \
    --test_file="_INSERT_" \
    --model="AtomicDipolesMACE" \
    --E0s="average" \
    --num_channels=32 \
    --max_L=2 \
    --r_max=5.0 \
    --loss="dipole" \
    --dipole_weight=1.0 \
    --polarizability_weight=1.0 \
    --dipole_key="REF_dipole"  \
    --polarizability_key="REF_polarizability" \
    --weight_decay=5e-10 \
    --clip_grad=1.0 \
    --batch_size=64 \
    --valid_batch_size=64 \
    --max_num_epochs=100 \
    --scheduler_patience=20 \
    --patience=50 \
    --eval_interval=1 \
    --ema \
    --error_table='DipoleRMSE' \
    --default_dtype="float64"\
    --device=cuda \
    --seed=123 \
    --restart_latest \
    --save_cpu \
    --compute_polarizability

Evaluating the model

Use the following CLI command to evaluate the dipole moment and the polarization:

mace_eval_mu_alpha \
  --configs="_INSERT_ \
  --model="MACE_dipole_pol.model" \
  --output="_INSERT_" \
  --device=cuda \
  --batch_size=10 \

Use the additional flag if you also want to estimate the spatial derivatives of the dipole moment and the polarization:

mace_eval_mu_alpha \
  --configs="_INSERT_ \
  --model="MACE_dipole_pol.model" \
  --output="_INSERT_" \
  --device=cuda \
  --batch_size=10 \
  --compute_dielectric_derivatives \

References

  1. Kapil, V., Kovács, D. P., Csányi, G., & Michaelides, A. (2023). First-principles spectroscopy of aqueous interfaces using machine-learned electronic and quantum nuclear effects. Faraday Discussions, 10.1039.D3FD00113J. https://doi.org/10.1039/D3FD00113J
  2. Grisafi, A., Wilkins, D. M., Csányi, G., & Ceriotti, M. (2018). Symmetry-adapted machine learning for tensorial properties of atomistic systems. Physical Review Letters, 120(3), 036002. https://doi.org/10.1103/PhysRevLett.120.036002