Energy of atomic oxgyen and smearing
909 views
Skip to first unread message
Natalie Austin
Jul 7, 2016, 3:29:41 PM7/7/16
to cp2k
Hello,
I wanted to calculate the atomic energy of oxygen. I used the CP2K tutorial for GEO_OPT of H2O as an example to generate the input file. What I noticed is that the energy calculation for atomic did not converge after 500 SCF steps if I did not use smearing. Would anyone be able to explain why this is? Because I thought for such a small atom smearing would not be necessary. The energy for atomic oxygen also converged when I used the OT method but I thought it best to use diagonalization to keep consistent with the input I use for calculations on my metal systems.
This is what I believe is the relevant part of the input form the calculation that converged. I indicated which lines were omitted from the calculation that did not coverge:
&FORCE_EVAL
METHOD QS
&DFT
BASIS_SET_FILE_NAME BASIS_MOLOPT
POTENTIAL_FILE_NAME GTH_POTENTIALS
&MGRID
CUTOFF 400
REL_CUTOFF 60
NGRIDS 4
&END MGRID
&QS
EPS_DEFAULT 1.0E-14
MAP_CONSISTENT
&END QS
&SCF
SCF_GUESS ATOMIC
EPS_SCF 1.0E-7
MAX_SCF 500
ADDED_MOS 200 Added MOS was omitted in the calculation that did not converge.
CHOLESKY INVERSE
&SMEAR ON
METHOD FERMI_DIRAC
ELECTRONIC_TEMPERATURE [K] 700 This smear section was omitted in the calculation that did not converge.
&END SMEAR
&DIAGONALIZATION
ALGORITHM STANDARD
&END DIAGONALIZATION
&MIXING
METHOD BROYDEN_MIXING
ALPHA 0.1
BETA 1.5
NBROYDEN 8
&END MIXING
&END SCF
&XC
&XC_FUNCTIONAL PBE
&PBE
PARAMETRIZATION REVPBE
&END PBE
&END XC_FUNCTIONAL
&VDW_POTENTIAL
DISPERSION_FUNCTIONAL PAIR_POTENTIAL
&PAIR_POTENTIAL
TYPE DFTD3
REFERENCE_FUNCTIONAL PBE
CALCULATE_C9_TERM .TRUE.
PARAMETER_FILE_NAME dftd3.dat
R_CUTOFF 15.0
VERBOSE_OUTPUT .TRUE.
&END PAIR_POTENTIAL
&END vdw_POTENTIAL
&END XC
UKS .TRUE.
MULTIPLICITY 3
Any help with this would be appreciated.
Thanks
I wanted to calculate the atomic energy of oxygen. I used the CP2K tutorial for GEO_OPT of H2O as an example to generate the input file. What I noticed is that the energy calculation for atomic did not converge after 500 SCF steps if I did not use smearing. Would anyone be able to explain why this is? Because I thought for such a small atom smearing would not be necessary. The energy for atomic oxygen also converged when I used the OT method but I thought it best to use diagonalization to keep consistent with the input I use for calculations on my metal systems.
This is what I believe is the relevant part of the input form the calculation that converged. I indicated which lines were omitted from the calculation that did not coverge:
&FORCE_EVAL
METHOD QS
&DFT
BASIS_SET_FILE_NAME BASIS_MOLOPT
POTENTIAL_FILE_NAME GTH_POTENTIALS
&MGRID
CUTOFF 400
REL_CUTOFF 60
NGRIDS 4
&END MGRID
&QS
EPS_DEFAULT 1.0E-14
MAP_CONSISTENT
&END QS
&SCF
SCF_GUESS ATOMIC
EPS_SCF 1.0E-7
MAX_SCF 500
ADDED_MOS 200 Added MOS was omitted in the calculation that did not converge.
CHOLESKY INVERSE
&SMEAR ON
METHOD FERMI_DIRAC
ELECTRONIC_TEMPERATURE [K] 700 This smear section was omitted in the calculation that did not converge.
&END SMEAR
&DIAGONALIZATION
ALGORITHM STANDARD
&END DIAGONALIZATION
&MIXING
METHOD BROYDEN_MIXING
ALPHA 0.1
BETA 1.5
NBROYDEN 8
&END MIXING
&END SCF
&XC
&XC_FUNCTIONAL PBE
&PBE
PARAMETRIZATION REVPBE
&END PBE
&END XC_FUNCTIONAL
&VDW_POTENTIAL
DISPERSION_FUNCTIONAL PAIR_POTENTIAL
&PAIR_POTENTIAL
TYPE DFTD3
REFERENCE_FUNCTIONAL PBE
CALCULATE_C9_TERM .TRUE.
PARAMETER_FILE_NAME dftd3.dat
R_CUTOFF 15.0
VERBOSE_OUTPUT .TRUE.
&END PAIR_POTENTIAL
&END vdw_POTENTIAL
&END XC
UKS .TRUE.
MULTIPLICITY 3
Any help with this would be appreciated.
Thanks
Matthias Krack
Jul 8, 2016, 4:11:45 AM7/8/16
to cp2k
Dear Natalie
you may try "direct_p_mixing" instead of "broyden_mixing". In addition, you may increase ALPHA to 0.4 for a faster initial convergence and thus earlier onset of DIIS/Diag. This mixing approach works often well for molecular as well as non-metallic systems when using diagonalisation.
Best regards
Matthias
you may try "direct_p_mixing" instead of "broyden_mixing". In addition, you may increase ALPHA to 0.4 for a faster initial convergence and thus earlier onset of DIIS/Diag. This mixing approach works often well for molecular as well as non-metallic systems when using diagonalisation.
Best regards
Matthias
Natalie Austin
Jul 11, 2016, 4:08:47 PM7/11/16
to cp...@googlegroups.com
Thank you Matthias, I tried as you suggested and unfortunately it did not work. I haven't been able to get convergence without smearing for O still.
Matthias Krack
Jul 11, 2016, 4:56:01 PM7/11/16
to cp2k
In that case, I suggest that you post the full input file.
M.
M.
Natalie Austin
Jul 13, 2016, 10:05:20 PM7/13/16
to cp2k
Hello Matthias,
The input file is below:
The input file is below:
&FORCE_EVAL
METHOD QS
&DFT
BASIS_SET_FILE_NAME BASIS_MOLOPT
POTENTIAL_FILE_NAME GTH_POTENTIALS
&MGRID
CUTOFF 400
REL_CUTOFF 60
NGRIDS 4
&END MGRID
&QS
EPS_DEFAULT 1.0E-14
MAP_CONSISTENT
&END QS
&SCF
SCF_GUESS ATOMIC
EPS_SCF 1.0E-7
MAX_SCF 500
CHOLESKY INVERSE
&DIAGONALIZATION
ALGORITHM STANDARD
&END DIAGONALIZATION
&MIXING
METHOD DIRECT_P_MIXING
ALPHA 0.1 # I also ran a case where I increased the value to 0.4 as you suggested.
ALPHA 0.1 # I also ran a case where I increased the value to 0.4 as you suggested.
BETA 1.5
NBROYDEN 8
&END MIXING
&END SCF
&XC
&XC_FUNCTIONAL PBE
&PBE
PARAMETRIZATION REVPBE
&END PBE
&END XC_FUNCTIONAL
&VDW_POTENTIAL
DISPERSION_FUNCTIONAL PAIR_POTENTIAL
&PAIR_POTENTIAL
TYPE DFTD3
REFERENCE_FUNCTIONAL PBE
CALCULATE_C9_TERM .TRUE.
PARAMETER_FILE_NAME dftd3.dat
R_CUTOFF 15.0
VERBOSE_OUTPUT .TRUE.
&END PAIR_POTENTIAL
&END vdw_POTENTIAL
&END XC
NBROYDEN 8
&END MIXING
&END SCF
&XC
&XC_FUNCTIONAL PBE
&PBE
PARAMETRIZATION REVPBE
&END PBE
&END XC_FUNCTIONAL
&VDW_POTENTIAL
DISPERSION_FUNCTIONAL PAIR_POTENTIAL
&PAIR_POTENTIAL
TYPE DFTD3
REFERENCE_FUNCTIONAL PBE
CALCULATE_C9_TERM .TRUE.
PARAMETER_FILE_NAME dftd3.dat
R_CUTOFF 15.0
VERBOSE_OUTPUT .TRUE.
&END PAIR_POTENTIAL
&END vdw_POTENTIAL
&END XC
MULTIPLICITY 3
UKS .TRUE.
&END DFT
&SUBSYS
&CELL
ABC 20.00000 20.00000 20.00000
ALPHA_BETA_GAMMA 90.00 90.00 90.00
PERIODIC NONE
&END CELL
&TOPOLOGY
COORD_FILE_NAME o.xyz
COORDINATE XYZ
&END TOPOLOGY
&KIND O
ELEMENT O
BASIS_SET DZVP-MOLOPT-SR-GTH-q6
POTENTIAL GTH-PBE-q6
&END KIND
&END SUBSYS
&END FORCE_EVAL
&GLOBAL
PROJECT o
RUN_TYPE ENERGY
PRINT_LEVEL MEDIUM
&END GLOBAL
&MOTION
&GEO_OPT
MAX_FORCE 0.0004
MAX_ITER 2000
OPTIMIZER BFGS
TYPE MINIMIZATION
&END GEO_OPT
&END MOTION
and the coordinate file:
1
o
O 7.46880 7.39185 7.47705
Natalie
UKS .TRUE.
&END DFT
&SUBSYS
&CELL
ABC 20.00000 20.00000 20.00000
ALPHA_BETA_GAMMA 90.00 90.00 90.00
PERIODIC NONE
&END CELL
&TOPOLOGY
COORD_FILE_NAME o.xyz
COORDINATE XYZ
&END TOPOLOGY
&KIND O
ELEMENT O
BASIS_SET DZVP-MOLOPT-SR-GTH-q6
POTENTIAL GTH-PBE-q6
&END KIND
&END SUBSYS
&END FORCE_EVAL
&GLOBAL
PROJECT o
RUN_TYPE ENERGY
PRINT_LEVEL MEDIUM
&END GLOBAL
&MOTION
&GEO_OPT
MAX_FORCE 0.0004
MAX_ITER 2000
OPTIMIZER BFGS
TYPE MINIMIZATION
&END GEO_OPT
&END MOTION
and the coordinate file:
1
o
O 7.46880 7.39185 7.47705
Natalie
Matthias Krack
Jul 14, 2016, 4:39:59 AM7/14/16
to cp2k
Dear Natalie
for a non-periodic calculation, you have to specify "PERIODIC none" in the &CELL and in the &DFT section using for instance MT as the Poisson solver for the electrostatic decoupling:
&POISSON
PERIODIC none
POISSON_SOLVER MT
&END POISSON
which is needed for the calculation of isolated systems.
That won't solve, however, your convergence problem for the O atom. The DIIS method employed with diagonalisation is a very efficient for SCF convergence acceleration, but it is also very sensible to numerical noise. That issue can be fixed by increasing the CUTOFF, e.g. from 400 to 600 Ry. I have attached the output of such a run.
You should also keep in mind that the convergence criterion applied for diagonalisation is different to the OT criterion. OT checks the energy convergence whereas the largest change in a density matrix element between two consecutive SCF steps is checked when using diagonalisation. Thus EPS_SCF can be 10-100 times larger with diagonalisation compared to OT. An EPS_SCF value of 1.0E-7 results in an energy convergence almost within machine precision as you can see in the attached output.
Best regards
Matthias
for a non-periodic calculation, you have to specify "PERIODIC none" in the &CELL and in the &DFT section using for instance MT as the Poisson solver for the electrostatic decoupling:
&POISSON
PERIODIC none
POISSON_SOLVER MT
&END POISSON
which is needed for the calculation of isolated systems.
That won't solve, however, your convergence problem for the O atom. The DIIS method employed with diagonalisation is a very efficient for SCF convergence acceleration, but it is also very sensible to numerical noise. That issue can be fixed by increasing the CUTOFF, e.g. from 400 to 600 Ry. I have attached the output of such a run.
You should also keep in mind that the convergence criterion applied for diagonalisation is different to the OT criterion. OT checks the energy convergence whereas the largest change in a density matrix element between two consecutive SCF steps is checked when using diagonalisation. Thus EPS_SCF can be 10-100 times larger with diagonalisation compared to OT. An EPS_SCF value of 1.0E-7 results in an energy convergence almost within machine precision as you can see in the attached output.
Best regards
Matthias
Natalie Austin
Jul 14, 2016, 4:40:37 PM7/14/16
to cp2k
Hello Matthias,
Thank you for your help.
I noticed in my previous calculations containing Cu based nanoparticles that I didn't see an energetic difference between using periodic none and periodic xyz (the default) in the POISSON section, but regardless moving on I will be sure to specify NONE for my isolated systems.
I had a feeling that increasing the CUTOFF would have helped. I will keep your advice in mind for future calculations. Thank you again.
Natalie
Thank you for your help.
I noticed in my previous calculations containing Cu based nanoparticles that I didn't see an energetic difference between using periodic none and periodic xyz (the default) in the POISSON section, but regardless moving on I will be sure to specify NONE for my isolated systems.
I had a feeling that increasing the CUTOFF would have helped. I will keep your advice in mind for future calculations. Thank you again.
Natalie
Reply all
Reply to author
Forward
0 new messages