Question about min_devi_f = 0 in model deviation stage

[chenggoj]

Description：
During the model deviation stage of DP-GEN (iter01), I noticed that min_devi_f (minimum force deviation) is exactly 0 in many steps of the LAMMPS NN-MD simulation. This occurs frequently across different timesteps while max_devi_f and avg_devi_f show reasonable values.

Relevant parameters from my param.json:

• model_devi_dt: 0.0005
• model_devi_f_trust_lo: 0.1
• model_devi_f_trust_hi: 3.0
• Ensemble: NVT
• Temperature: 873K
• Pressure: 1.0
• Trajectory frequency: 1000
• Total steps: 20000

Sample Data
Here's a snippet from my model_devi.out showing this behavior:
step max_devi_v min_devi_v avg_devi_v max_devi_f min_devi_f avg_devi_f
0 2.768609e-02 6.873493e-03 1.552153e-02 6.995181e-01 4.756727e-02 2.233931e-01
1000 8.919899e-02 1.897480e-02 5.112303e-02 1.636830e+01 0.000000e+00 3.319308e-01
2000 1.112451e-01 1.439360e-02 4.925055e-02 1.777609e+01 0.000000e+00 3.864138e-01
3000 8.376999e-01 7.933874e-02 3.993526e-01 3.052900e+02 0.000000e+00 3.898055e+00
4000 2.256255e+00 1.833313e-01 1.127322e+00 1.236023e+02 0.000000e+00 5.874100e+00
5000 1.256377e+00 1.153748e-01 6.713597e-01 1.745688e+02 0.000000e+00 6.884596e+00
6000 1.102776e+00 4.864065e-02 5.888506e-01 1.176719e+02 0.000000e+00 5.836812e+00
7000 9.601822e-01 1.028447e-01 5.532940e-01 1.423400e+02 0.000000e+00 6.794675e+00
8000 1.025749e+00 6.784166e-02 6.165037e-01 1.551683e+02 0.000000e+00 6.406366e+00
9000 1.098593e+00 1.017746e-01 6.343627e-01 1.686362e+02 0.000000e+00 5.894105e+00
10000 1.327279e+00 1.758605e-01 7.034756e-01 1.214956e+02 1.378852e-03 6.116430e+00
11000 1.313640e+00 5.598873e-02 7.357472e-01 1.704027e+02 0.000000e+00 6.394900e+00
12000 1.176754e+00 1.098788e-01 6.220313e-01 1.635384e+02 3.194611e-03 5.959711e+00
13000 1.138455e+00 3.650621e-02 6.511989e-01 1.378689e+02 0.000000e+00 6.116118e+00
14000 1.065351e+00 4.357848e-02 6.188082e-01 1.784189e+02 6.622296e-02 6.336310e+00
15000 1.005644e+00 1.336446e-01 6.560449e-01 1.601262e+02 0.000000e+00 6.560284e+00
16000 1.077902e+00 1.959097e-01 6.466314e-01 1.576194e+02 0.000000e+00 6.524390e+00
17000 1.041869e+00 1.562755e-01 6.219213e-01 1.865414e+02 4.931797e-02 6.313256e+00
18000 1.037294e+00 1.118517e-01 6.026657e-01 1.854644e+02 0.000000e+00 6.767444e+00

Questions

1. Is this behavior normal in the model deviation stage?
2. Could this indicate model convergence, or might it suggest potential issues with the model ensemble?
3. Should any parameters be adjusted to address this?

Any insights or guidance would be appreciated.