Creating a potential by training data on multiple sets of temperatures

[ramic-k]

Hello,

I was wondering is it possible to create one potential that will be able to be used in LAMMPS for different temperatures. For example I have performed VASP AIMD calculations at 50K increments from 0 to 1000K. If possible, how would one go about training a model like that? I tried the following, but I dont think it worked (I am interested in a phonon spectrum, and using the model trained using input script below, the phonon spectrum doesnt change with temperature. If I train a model at each temperature, and then use that for LAMMPS and calculate the phonon spectrum, I do see the difference):

{
    "_comment": " model parameters",
    "model": {
    "type_map": ["Y", "H"],
    "descriptor" :{
        "type":     "se_e3",
        "sel":      "auto",
        "rcut_smth":    0.50,
        "rcut":     5.80,
        "neuron":       [4, 8, 16],
        "resnet_dt":    false,
        "seed":     1,
        "_comment":     " that's all"
    },
    "fitting_net" : {
        "neuron":           [240, 240, 240],
        "resnet_dt":    true,
        "seed":     1,
        "_comment":     " that's all"
    },
    "_comment": " that's all"
    },

    "learning_rate" :{
    "type":     "exp",
    "decay_steps":  5000,
    "start_lr": 0.001,  
    "stop_lr":  3.51e-8,
    "_comment": "that's all"
    },

    "loss" :{
    "start_pref_e": 0.02,
    "limit_pref_e": 1,
    "start_pref_f": 1000,
    "limit_pref_f": 1,
    "start_pref_v": 0,
    "limit_pref_v": 0,
    "_comment": " that's all"
    },

    "_comment": " traing controls",
    "training" : {
    "training_data": {
        "systems":      ["./deepmd_data_5K/0/",
                                "./deepmd_data_5K/1/",
                                "./deepmd_data_20K/2/",
                                "./deepmd_data_20K/0/",
                                "./deepmd_data_20K/1/",
                                "./deepmd_data_20K/2/",
                                "./deepmd_data_50K/0/",
                                "./deepmd_data_50K/1/",
                                "./deepmd_data_50K/2/",
                                "./deepmd_data_100K/0/",
                                "./deepmd_data_100K/1/",
                                "./deepmd_data_100K/2/",
                                "./deepmd_data_150K/0/",
                                "./deepmd_data_150K/1/",
                                "./deepmd_data_150K/2/",
                                "./deepmd_data_200K/0/",
                                "./deepmd_data_200K/1/",
                                "./deepmd_data_200K/2/",
                                "./deepmd_data_250K/0/",
                                "./deepmd_data_250K/1/",
                                "./deepmd_data_250K/2/",
                                "./deepmd_data_296K/0/",
                                "./deepmd_data_296K/1/",
                                "./deepmd_data_296K/2/",
                                "./deepmd_data_300K/0/",
                                "./deepmd_data_300K/1/",
                                "./deepmd_data_300K/2/",
                                "./deepmd_data_350K/0/",
                                "./deepmd_data_350K/1/",
                                "./deepmd_data_350K/2/",
                                "./deepmd_data_400K/0/",
                                "./deepmd_data_400K/1/",
                                "./deepmd_data_400K/2/",
                                "./deepmd_data_450K/0/",
                                "./deepmd_data_450K/1/",
                                "./deepmd_data_450K/2/",
                                "./deepmd_data_500K/0/",
                                "./deepmd_data_500K/1/",
                                "./deepmd_data_500K/2/",
                                "./deepmd_data_550K/0/",
                                "./deepmd_data_550K/1/",
                                "./deepmd_data_550K/2/",
                                "./deepmd_data_600K/0/",
                                "./deepmd_data_600K/1/",
                                "./deepmd_data_600K/2/",
                                "./deepmd_data_700K/0/",
                                "./deepmd_data_700K/1/",
                                "./deepmd_data_700K/2/",
                                "./deepmd_data_750K/0/",
                                "./deepmd_data_750K/1/",
                                "./deepmd_data_750K/2/",
                                "./deepmd_data_800K/0/",
                                "./deepmd_data_800K/1/",
                                "./deepmd_data_800K/2/",
                                "./deepmd_data_850K/0/",
                                "./deepmd_data_850K/1/",
                                "./deepmd_data_850K/2/",
                                "./deepmd_data_900K/0/",
                                "./deepmd_data_900K/1/",
                                "./deepmd_data_900K/2/",
                                "./deepmd_data_950K/0/",
                                "./deepmd_data_950K/1/",
                                "./deepmd_data_950K/2/",
                                "./deepmd_data_1000K/0/",
                                "./deepmd_data_1000K/1/",
                                "./deepmd_data_1000K/2/"],
        "batch_size":   "auto",
        "_comment":     "that's all"
    },
    "validation_data":{
        "systems":      ["./deepmd_data_5K/3/",
                                "./deepmd_data_20K/3/",
                                "./deepmd_data_50K/3/",
                                "./deepmd_data_100K/3/",
                                "./deepmd_data_150K/3/",
                                "./deepmd_data_200K/3/",
                                "./deepmd_data_250K/3/",
                                "./deepmd_data_296K/3/",
                                "./deepmd_data_300K/3/",
                                "./deepmd_data_350K/3/",
                                "./deepmd_data_400K/3/",
                                "./deepmd_data_450K/3/",
                                "./deepmd_data_500K/3/",
                                "./deepmd_data_550K/3/",
                                "./deepmd_data_600K/3/",
                                "./deepmd_data_700K/3/",
                                "./deepmd_data_750K/3/",
                                "./deepmd_data_800K/3/",
                                "./deepmd_data_850K/3/",
                                "./deepmd_data_900K/3/",
                                "./deepmd_data_950K/3/",
                                "./deepmd_data_1000K/3/"],
        "batch_size":   "auto",
        "_comment":     "that's all"
    },
    "numb_steps":   1000000,
    "seed":     1,

    "_comment": " display and restart",
    "_comment": " frequencies counted in batch",
    "disp_file":    "lcurve.out",
    "disp_freq":    100,
    "numb_test":    4,
    "save_freq":    1000,
    "save_ckpt":    "model.ckpt",
    "disp_training":true,
    "time_training":true,
    "profiling":    false,
    "profiling_file":"timeline.json",
    "_comment": "that's all"
    },

    "_comment":     "that's all"
}

One thing I didn't understand compared to the examples is how to set batch_size and numb_btch in validation_data section, so I set the batch_size to auto and removed numb_btch completely.

Thank you for this wonderful tool!

[wanghan-iapcm]

You may want to try the descriptor se_e2_a, or the hybridization of se_e2_a and se_e3. Only using se_e3 dose not provide you a good model.

[McAlexDudu]

@wanghan-iapcm @ramic-k I have done similar test on dpgen. I use se_e2_a with 800K,1000K,1200K as training data produced through init_bulk, and in LAMMPS part, the mode_jobs set NVT with six temperature(50K,100K,300K,500K,600K,700K), totally run 9 iteration, the result as below.

But it seem the candidate still high and I guess many different temperature as training data may not be a good idea?