Different eigenvalues from Hybrids + ADMM calculations in CP2K 2025.1 compared to 2024.1

[Max Pierre]

Hello CP2K community,

I have recently switched from cp2k-2024.1 to cp2k-2025.1 and initially got some strange output. Running some test calculations using the same input with both versions, I found that the energy eigenvalues are shifted, and the HOMO - LUMO gap is much larger in 2025.1.

- 2024.1:

*

      -0.15118136     -0.15058927     -0.15030345     -0.14919531
      -0.14827092     -0.14449236     -0.14178774     -0.14087384
      -0.14005256
 Fermi Energy [eV] :   -3.811024
 
  Lowest Eigenvalues of the unoccupied subspace spin            1
 -----------------------------------------------------
 OT| Eigensolver reached convergence in 763 iterations
       0.19181627
 
 HOMO - LUMO gap [eV] :    9.030610

 ENERGY| Total FORCE_EVAL ( QS ) energy [a.u.]:            -1322.383368758011557

*

- 2025.1:

*

      -0.25036349     -0.25011266     -0.24923456     -0.24855820
      -0.24640841     -0.24154336     -0.23787275     -0.23769990
      -0.23629481
 Fermi Energy [eV] :   -6.429909
 
  Lowest Eigenvalues of the unoccupied subspace spin            1
 -----------------------------------------------------
 OT| Eigensolver reached convergence in 887 iterations
       0.25231706
 
 HOMO - LUMO gap [eV] :   13.295805

 ENERGY| Total FORCE_EVAL ( QS ) energy [hartree]          -1322.383368757196422

*

Both calculations use PBE0 with ADMM pFIT3 and are started from a RESTART file from a PBE calculations, the system is amorphous Polyethylene. The 2024.1 HOMO-LUMO gap was pretty reasonable, the newer one is very high compared with experimental values and previously reported hybrid functional results. 

I didn't find anyone else experiencing this issue, does anyone have any idea what might be going on? 

Thank you,

Max

[fabia...@gmail.com]

Hi Max,

I don't know where the difference comes from. But to get the band gap you need to switch off purification, so neither of those values are reliable. Change  ADMM_PURIFICATION_METHOD MO_DIAG to ADMM_PURIFICATION_METHOD NONE.
See, e.g. here: https://groups.google.com/g/cp2k/c/WAdLCCMk47A/m/fP0AeCaRBAAJ

Cheers,

Fabian

[Max Pierre]

That was indeed the issue! For some reason in cp2k-2024.1 the ADMM band gaps were very close to the normal PBE0 ones even with purification switched on, so I hadn't noticed that anything was off before switching to 2025.1. I guess I was just lucky before that using the wrong purification setting wasn't causing any big issues for me in 2024.1.

Thank you for the help!

Max