deepmodeling / abacus-develop

An electronic structure package based on either plane wave basis or numerical atomic orbitals.
http://abacus.ustc.edu.cn
GNU Lesser General Public License v3.0
174 stars 136 forks source link

Inf in BAND_*.dat and Missing 1 to 36 in `BAND_2.dat` #3843

Closed Chentao168 closed 7 months ago

Chentao168 commented 7 months ago

Describe the bug

Fe, 2 atoms, nspin = 2, inf in BAND_1.dat and BAND_2.dat. Missing 1 to 36 in BAND_2.dat. image image

Expected behavior

No response

To Reproduce

20240329-band-inf.zip

Environment

ABACUS v3.5.4

Additional Context

No response

Task list for Issue attackers (only for developers)

kirk0830 commented 7 months ago

I did not reproduce your issue with input files you provided, I think I made minor changes and should not affect the behavior of as-written codes (if you are interested, see pull request draft #3855 ): BANDS_1.dat: image BANDS_2.dat: image

Chentao168 commented 7 months ago

I also tested in ABACUS version 3.6.0, where the aforementioned issue persisted, specifically the appearance of 'inf' in BAND_1.dat and BAND_2.dat, as well as missing entries from 1 to 36 in BAND_2.dat. The pull request #3855 has not yet been merged into the develop branch. Is it feasible to conduct tests using version 3.6.0?

kirk0830 commented 7 months ago

I also tested in ABACUS version 3.6.0, where the aforementioned issue persisted, specifically the appearance of 'inf' in BAND_1.dat and BAND_2.dat, as well as missing entries from 1 to 36 in BAND_2.dat. The pull request #3855 has not yet been merged into the develop branch. Is it feasible to conduct tests using version 3.6.0?

@Chentao168 Could you provide your environment setting information? This seems to relate to some undefined behavior

Qianruipku commented 7 months ago

ABACUS has bugs when out_band = 1 and kpar > 1。 Currently, when kpar > 1, there will be a memory overflow in the band calculation, resulting in the appearance of 'inf' in the results. The program has issues and does not support k-point parallelism, which needs to be addressed.