[Feature Request] Filtering unreasonable structures before 02.fp in dpgen1

[liyibaipnu]

Summary

Add structure filtering (bad box / short distance) before 02.fp in dpgen1 to prevent unconverged FP tasks.

Detailed Description

Description:
Currently, dpgen1 examples seem to lack a mechanism for filtering out unreasonable structures before sending them to 02.fp. As a result, structures generated in 01.md are directly and randomly selected for 02.fp tasks. If abnormal cells or configurations are included, FP calculations may fail to converge, which in turn makes the whole 02.fp job unrecoverable (even when some structures do converge).

Suggestion:

1. Add a structure screening step between 01.md → 02.fp. Possible criteria include:

2. Cell deformation (exclude severely distorted boxes)

3. Minimum interatomic distance (exclude structures with atoms too close)

Background:
I am aware that dpgen2 has already implemented similar filtering, but due to the lack of comprehensive documentation, it is currently difficult for users to smoothly adopt dpgen2. Adding such functionality to dpgen1 would greatly improve the robustness and efficiency of the workflow.

Request:

Could the developers confirm whether such filtering already exists in dpgen1?

If not, would it be possible to add this feature in a future update?

Further Information, Files, and Links

No response

[wanghan-iapcm]

dpgen1 may filter out unreasonable cells, but the feature is not well documented. please also refer to #1819