High negative pressure and overlapping atoms from trained potential

[mayankaditya]

Hello Admin,

I am using the deepmd code to train the potential for one of my systems which is a disorder. I created the pair potential (using ~500000 batch, force error ~10meV/A). The generated lammps potential gives very high negative value of pressure and after few steps the diffusive atoms are overlapping/merge with each other. I train the model with the following parameters.

**{
"_comment": " model parameters",
"use_smooth": true,
"sel_a": [150,200,200],
"rcut_smth": 2.0,
"rcut": 10.00,
"filter_neuron": [25, 50, 100],
"fitting_neuron": [240,240,240],
"filter_resnet_dt": false,
"axis_neuron": 32,
"fitting_resnet_dt": true,
"_comment": "that's all",
"_comment": " training controls",
"systems": ["./"],
"set_prefix": "set",
"stop_batch": 1000000,
"batch_size": 1,
"start_lr": 0.001,
"decay_steps": 5000,
"decay_rate": 0.95,
"start_pref_e": 0.02,
"limit_pref_e": 8,
"start_pref_f": 1000,
"limit_pref_f": 1,
"start_pref_v": 0,
"limit_pref_v": 0,
"seed": 1,
"_comment": " display and restart",
"_comment": " frequencies counted in batch",
"disp_file": "/dscrhome/mg422/lcurve.out",
"disp_freq": 100,
"numb_test": 100,
"save_freq": 100,
"save_ckpt": "model.ckpt",
"load_ckpt": "model.ckpt",
"disp_training": true,
"time_training": true,
"_comment": "that's all"
}

Kindly advise, how to deal with this issue.

Thanks,
Mayank

[amcadmus]

Hi Mayank,

Do you have the virial label in your data? If so could you please check the prediction accuracy on virial? And check if DFT gives you a high negative pressure.

Best, Han