[BUG] Only the first frame is read using abacus/lcao/relax

[sies-delta]

Bug summary

When reading the abacus relaxation trajectory, dpdata's LabeledSystem only includes the first frame in the trajectory, namely the configuration without any relaxation. @dosu

dpdata Version

v1.0.1

Input Files, Running Commands, Error Log, etc.

python code as follows,

from dpdata import LabeledSystem
import os
sys = LabeledSystem("./test/chempot=0.02",fmt='abacus/lcao/relax')
print(sys)

the related directory of abacus results is linked below "Further Information, Files, and Links".
The output is

Data Summary
Labeled System
-------------------
Frame Numbers      : 1
Atom Numbers       : 92
Including Virials  : Yes
Element List       :
-------------------
Cu  Zn
91  1

but this trajectory should include 11 frames.

Steps to Reproduce

simply unzip the abacus directory and run the python code mentioned above

Further Information, Files, and Links

test.zip

[ZhouXY-PKU]

@dosu Please take a look at this issue.

[dosubot]

This is a known bug in dpdata v1.0.1 that was fixed in later versions. The critical fix came in PR #409 (merged January 2023), which refactored the frame reading logic to properly synchronize coordinate, energy, force, and stress data across multiple relaxation steps.

Solution: Upgrade dpdata to the latest version:

pip install --upgrade dpdata

The v1.0.1 release (August 2022) had issues with:

• Cell data not being properly reused when cells don't change between relaxation steps
• Energy/force/stress arrays not properly aligned with coordinate arrays
• Incomplete frames not being filtered correctly

After upgrading, your relaxation trajectory should correctly show all frames. If the issue persists after upgrading, please share the contents of your running_cell-relax.log or running_relax.log file, as there could be data completeness issues (e.g., unconverged SCF calculations being filtered out).

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}

  

[ZhouXY-PKU]

@dosu Are you sure？ v1.0.1 is actually the lastest version just released 2 weeks ago.

[pxlxingliang]

Hi @sies-delta , I have checked the outputs of your job, there is no coords outputted in the running _relax.log file, so dpdata can not collect the fully results. This is a bug of abacus, when relax_method is bfgs_trad (we will fix it in future version). Now I have supportted to read the coord/cell from STRU_ION*_D (need set "out_stru 1" in abacus INPUT to write these structure files) in dpdata, it can work in #957 .