DPGEN encountered the following issue while exploring the potential function of the PtNiMoO system.

[wangsl613]

Summary

DPGEN cannot perform the exploration step effectively.

DeePMD-kit Version

2.2.10

Backend and its version

tensorflow 2.16.1；dpdata 0.2.7；dpgen 0.10.6；

Python Version, CUDA Version, GCC Version, LAMMPS Version, etc

python 3.8.0；lammps 2023.08.02；gcc (GCC) 4.8.5 20150623 (Red Hat 4.8.5-39)；CentOS Linux release 7.8.2003 (Core)

Details

Dear DeePMD developers,

I used the same initial structure to run model_devi, only increasing the temperature by 100K. However, all the structures from the model_devi runs failed. Could it be that the temperature increase is too high, or is there an issue with my init_data (300K, ran over 800 steps of AIMD)? All structures are from the PtNiMoO system. Additionally, the 152 candidates from the first iteration's fp calculations resulted in similar structures.

Best regards.

dpgen.tar.gz

init.tar.gz：链接: https://pan.baidu.com/s/14jdGMoeSeYAd_7rdsgn5lA?pwd=w6ni 提取码: w6ni

[njzjz]

Do you get a different result when performing simulations under 100K? The current result might be expected, as structures under 200K are very different from that under 100K.

[wangsl613]

Thanks for the reply but my dataset is from AIMD at 300K temperature conditions.

[njzjz-bot]

The reported result is a scientific exploration outcome rather than a reproducible DP-GEN software defect. Model-deviation candidates can all fail when the explored configurations leave the domain covered by the initial data; raising the temperature by 100 K does not by itself guarantee usable configurations. The follow-up also confirms that the initial data were sampled at 300 K, so the result needs dataset/model-deviation analysis rather than a DP-GEN code change.

This issue is already labeled wontfix; I suggest closing it. Further discussion about exploration settings can continue in the user forum with the complete input and model-deviation statistics.

Coding agent: Codex
Codex version: codex-cli 0.149.0
Model: gpt-5.6-sol
Reasoning effort: xhigh