SCF convergence issue with GAPW and all electron basis

[Hengyuan Shen]

Dear CP2K community,

I tried to calculate the XPS binding energy of some O atoms on a Pt surface. I used a big all electron basis for the interested O atoms, a smaller all electron basis for other O atoms, and GTH-MOLOPT basis for Pt with pseudopotential. However, the SCF calculation could not converge. I then tried the ground state calculation with a similar GPAW method and GPW method. For GPAW method, the SCF convergence also failed, and gave me an absurd energy. The GPW method converged after about 100 steps. Strangely, both calculations showed warnings about Fermi-Dirac smearing, although I used METHFESSEL_PAXTON smearing and 1000 ADDED_MO. Any idea on this issue? Any suggestions would be appreciated. Thank you.

Best regards,

Hengyuan

[Marcella Iannuzzi]

Dear Hengyuan,

The smearing method that you are using is obviously not working properly, also in the case of GPW.

It is not normal that the occupation is smeared down to the first MO.  You can add as many added MOs as you want, this does not change.

 *** WARNING in qs_mo_occupation.F:121 :: Fermi-Dirac smearing includes ***

 *** the first MO

Indications of a wrong behaviour are also the oscillating convergence  and the non-zero total charge density 

  Total charge density on r-space grids:       -0.0006221769

  Total charge density g-space grids:          -0.0006221769

My suggestion is to run the GPW calculation with the Fermi-Dirac smearing to confirm that one problem is the choice of the smearing method. 

Regards

Marcella

[Hengyuan Shen]

Dear Marcella,

Thank you very much for your suggestions. I switched to Fermi Dirac smearing, and the GPW calculation finishes normally now, with the correct total charge density. However, once I switch to GAPW calculation, the SCF still failed to converge, for either aug-cc-pVTZ or def2-SVP basis. I also tried to use all electron basis on only one O atom, but the convergence issue still persist. Do you have any suggestions on  converging SCF for the GAPW calculations? Thank you.

Best regards,

Hengyuan

[Nan Dong]

Hi Hengyuan, 

I ran the GAPW calculations with def2-SVP for O, and it converged (within 300 SCF step) 

Based on your def2-SVP.in, I changed something in SCF section, and attached input and output here. 

Hope this will help. 

Best,

Nanchen 

[Hengyuan Shen]

Dear Nanchen,

Thank you very much for your sample input and result. I used a similar setting as yours to run GAPW for 1 O atom with the def2-SVP basis and 4 O atoms with the aug-cc-pVTZ basis. The def2-SVP calculation converged, but it took almost 500 SCF steps, and there's a weird oscillation of Convergence between 1e-4 and 1e-3. The aug-cc-pVTZ calculation converges more slowly and can't converge within 500 steps. Do you have any idea what causes the slow convergence and how to improve it? Thank you! Below is my CP2K version:

 CP2K version 2026.1 (Development Version)
 Source code revision 7b94146043
 cp2kflags: omp libxc elpa parallel scalapack
 compiler: GCC version 14.3.0
 compiler options:
   -I /global/homes/h/hengyuan/hys_project/cp2k/build/src/start -I /glo
   bal/homes/h/hengyuan/hys_project/cp2k/src -I /global/homes/h/hengyua
   n/hys_project/cp2k/build/src -I /global/homes/h/hengyuan/hys_project
   /cp2k/src/base -I /global/homes/h/hengyuan/hys_project/cp2k/src/comm
   on -I /global/homes/h/hengyuan/hys_project/cp2k/src/motion -I /globa
   l/homes/h/hengyuan/hys_project/cp2k/src/dbm -I /global/homes/h/hengy
   uan/hys_project/cp2k/build/src/mod_files -I /global/homes/h/hengyuan
   /hys_project/cp2k/spack/spack/opt/spack/view/include -I /global/home
   s/h/hengyuan/hys_project/cp2k/spack/spack/opt/spack/__spack_path_pla
   ceholder__/__spack_path_placeholder__/__spack_pa/linux-zen3/elpa-202
   6.02.001-s5vz6kbd3dbethregansxz4e5tlqsafv/include/elpa_openmp-2026.0
   2.001 -I /global/homes/h/hengyuan/hys_project/cp2k/spack/spack/opt/s
   pack/__spack_path_placeholder__/__spack_path_placeholder__/__spack_p
   a/linux-zen3/elpa-2026.02.001-s5vz6kbd3dbethregansxz4e5tlqsafv/inclu
   de/elpa_openmp-2026.02.001/modules -I /global/cfs/cdirs/mp54/hys/cp2
   k/spack/spack/opt/spack/__spack_path_placeholder__/__spack_path_plac
   eholder__/__spack_pa/linux-zen3/mpich-4.3.2-6zwkkcll2y5mdmgz3at7bbns
   liyxgfbz/include -march=znver3 -mmmx -mpopcnt -msse -msse2 -msse3 -m
   ssse3 -msse4.1 -msse4.2 -mavx -mavx2 -msse4a -mno-fma4 -mno-xop -mfm
   a -mno-avx512f -mbmi -mbmi2 -maes -mpclmul -mno-avx512vl -mno-avx512
   bw -mno-avx512dq -mno-avx512cd -mno-avx512vbmi -mno-avx512ifma -mno-
   avx512vpopcntdq -mno-avx512vbmi2 -mno-gfni -mvpclmulqdq -mno-avx512v
   nni -mno-avx512bitalg -mno-avx512bf16 -mno-avx512vp2intersect -mno-3
   dnow -madx -mabm -mno-cldemote -mclflushopt -mclwb -mclzero -mcx16 -
   mno-enqcmd -mf16c -mfsgsbase -mfxsr -mno-hle -msahf -mno-lwp -mlzcnt
    -mmovbe -mno-movdir64b -mno-movdiri -mmwaitx -mno-pconfig -mpku -mp
   rfchw -mno-ptwrite -mrdpid -mrdrnd -mrdseed -mno-rtm -mno-serialize
   -mno-sgx -msha -mshstk -mno-tbm -mno-tsxldtrk -mvaes -mno-waitpkg -m
   wbnoinvd -mxsave -mxsavec -mxsaveopt -mxsaves -mno-amx-tile -mno-amx
   -int8 -mno-amx-bf16 -mno-uintr -mno-hreset -mno-kl -mno-widekl -mno-
   avxvnni -mno-avx512fp16 -mno-avxifma -mno-avxvnniint8 -mno-avxneconv
   ert -mno-cmpccxadd -mno-amx-fp16 -mno-prefetchi -mno-raoint -mno-amx
   -complex -mno-avxvnniint16 -mno-sm3 -mno-sha512 -mno-sm4 -mno-apxf -
   mno-usermsr --param=l1-cache-size=32 --param=l1-cache-line-size=64 -
   -param=l2-cache-size=512 -mtune=znver3 -g -O3 -O3 -Wno-deprecated-de
   clarations -Wno-maybe-uninitialized -Wuninitialized -Wuse-without-on
   ly -std=f2008 -fPIE -ffree-form -fimplicit-none -fno-omit-frame-poin
   ter -fbacktrace -fallow-argument-mismatch -funroll-loops -fopenmp -f
   preprocessed -J src -fpre-include=/usr/include/finclude/math-vector-
   fortran.h

--
You received this message because you are subscribed to a topic in the Google Groups "cp2k" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/cp2k/6--LsBoEEZw/unsubscribe.
To unsubscribe from this group and all its topics, send an email to cp2k+uns...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/84d452dd-bb63-4a23-b9b6-24bf2dcdede7n%40googlegroups.com.