Electronic Convergence Issue in Geometry Optimization of Molecule Adsorption

Rishikanta Mayengbam

unread,

Jan 30, 2025, 11:24:09 AM1/30/25

to cp2k

Dear CP2K Users,

I am running a geometry optimization for a molecule adsorption system with vacuum along the y-axis, but I am struggling to achieve electronic SCF convergence even after 1000 iterations. I would appreciate any suggestions on improving the convergence.

Below is my input script:

&GLOBAL
PRINT_LEVEL MEDIUM
PROJECT_NAME mol_absorp_OPT
RUN_TYPE GEO_OPT
EXTENDED_FFT_LENGTHS T
&END GLOBAL

&FORCE_EVAL
METHOD QS
&DFT
UKS .FALSE.
BASIS_SET_FILE_NAME BASIS_MOLOPT
POTENTIAL_FILE_NAME POTENTIAL
&MGRID
NGRIDS 4
CUTOFF 600
REL_CUTOFF 60
&END MGRID
&QS
METHOD GPW
EPS_DEFAULT 1.000E-12
&END QS
&POISSON
PERIODIC XZ
PSOLVER MT
&END POISSON
&SCF
&DIAGONALIZATION ON
&END DIAGONALIZATION
&MIXING
ALPHA 0.5
METHOD PULAY_MIXING
NBROYDEN 8
&END MIXING
EPS_SCF 1.0E-5
SCF_GUESS ATOMIC
MAX_SCF 5000
ADDED_MOS 100
&END SCF
&XC
&XC_FUNCTIONAL
&PBE
&END PBE
&END XC_FUNCTIONAL
&VDW_POTENTIAL
POTENTIAL_TYPE PAIR_POTENTIAL
&PAIR_POTENTIAL
PARAMETER_FILE_NAME dftd3.dat
TYPE DFTD3
REFERENCE_FUNCTIONAL PBE
R_CUTOFF [angstrom] 16
&END
&END VDW_POTENTIAL
&END XC
&PRINT
&MULLIKEN OFF
&END
&HIRSHFELD OFF
&END
&END PRINT
&END DFT
&SUBSYS
&CELL
A 24.8975601196000014 0.0000000000000000 0.0000000000000000
B 0.0000000000000000 55.6124382018999981 0.0000000000000000
C 0.0000000000000000 0.0000000000000000 24.9615383147999985
PERIODIC XZ
&END CELL
&TOPOLOGY
COORD_FILE_NAME ABC.xyz
COORD_FILE_FORMAT XYZ
&END
&KIND Cs
BASIS_SET DZVP-MOLOPT-SR-GTH
POTENTIAL GTH-PBE-q9
&END KIND
&KIND Pb
BASIS_SET DZVP-MOLOPT-SR-GTH
POTENTIAL GTH-PBE-q4
&END KIND
&KIND Br
BASIS_SET DZVP-MOLOPT-SR-GTH
POTENTIAL GTH-PBE-q7
&END KIND
&KIND C
BASIS_SET DZVP-MOLOPT-GTH
POTENTIAL GTH-PBE-q4
&END KIND
&KIND N
BASIS_SET DZVP-MOLOPT-GTH
POTENTIAL GTH-PBE-q5
&END KIND
&KIND H
BASIS_SET DZVP-MOLOPT-GTH
POTENTIAL GTH-PBE-q1
&END KIND
&END SUBSYS
&END FORCE_EVAL

&MOTION
&GEO_OPT
OPTIMIZER BFGS
MAX_ITER 500
MAX_DR [bohr] 0.01
MAX_FORCE [hartree*bohr^-1] 0.01
RMS_DR [bohr] 0.01
RMS_FORCE [hartree*bohr^-1] 0.01
&END GEO_OPT

&CONSTRAINT
&FIXED_ATOMS
LIST 1 2 3 4 5 6 10 11 12 13 14 15 .........................
&END FIXED_ATOMS
&END CONSTRAINT

&PRINT
&TRAJECTORY
&EACH
GEO_OPT 5
&END EACH
FORMAT XYZ
&END TRAJECTORY
&VELOCITIES OFF
&END VELOCITIES
&FORCES ON
&END FORCES
&RESTART
BACKUP_COPIES 1
&EACH
GEO_OPT 10
&END EACH
&END RESTART
&END PRINT
&END MOTION

Marcella Iannuzzi

unread,

Jan 31, 2025, 7:31:44 AM1/31/25

to cp2k

Hi ...

How bad is the convergence? You might need smearing of the occupation number and a smaller alpha parameter in the density mixing.

Regards

marcella

Rishikanta Mayengbam

unread,

Feb 1, 2025, 1:41:35 AM2/1/25

to cp2k

Thanks for the response.

It was not able to converge even within 4000 steps. So, I tried many things, loosening the convergence criteria, changing cut-off and relative cut offs, etc. I haven't tried with the smearing, but have tried reducing the mixing alpha parameter.

Finally, I used CG Minimiser as follows and it's converging. Is this approach correct?

&GLOBAL

PRINT_LEVEL MEDIUM
PROJECT_NAME CsPb_Br3_TmPyPB_OPT

RUN_TYPE GEO_OPT
EXTENDED_FFT_LENGTHS T
&END GLOBAL

&FORCE_EVAL
METHOD QS
&DFT
UKS .FALSE.
BASIS_SET_FILE_NAME BASIS_MOLOPT
POTENTIAL_FILE_NAME POTENTIAL
&MGRID
NGRIDS 4

CUTOFF 500
REL_CUTOFF 50

&END MGRID
&QS
METHOD GPW
EPS_DEFAULT 1.000E-12
&END QS
&POISSON
PERIODIC XZ
PSOLVER MT
&END POISSON
&SCF

EPS_SCF 1.0E-6
SCF_GUESS RESTART
MAX_SCF 5000
&OT
MINIMIZER CG
PRECONDITIONER FULL_SINGLE_INVERSE
&END OT
&OUTER_SCF
MAX_SCF 500
EPS_SCF 1.0E-5
&END OUTER_SCF

COORD_FILE_NAME CsBr-TmPyPB.xyz

OPTIMIZER LBFGS
MAX_ITER 1000
MAX_DR [bohr] 0.002
MAX_FORCE [hartree*bohr^-1] 3.000E-4
RMS_DR [bohr] 0.002
RMS_FORCE [hartree*bohr^-1] 3.000E-4
&END GEO_OPT

&CONSTRAINT
&FIXED_ATOMS
LIST 1 2 3 4 5 6 ...........

&END FIXED_ATOMS
&END CONSTRAINT

&PRINT
&TRAJECTORY
&EACH
GEO_OPT 5
&END EACH
FORMAT XYZ
&END TRAJECTORY
&VELOCITIES OFF
&END VELOCITIES
&FORCES ON
&END FORCES
&RESTART
BACKUP_COPIES 1
&EACH
GEO_OPT 10
&END EACH
&END RESTART
&END PRINT
&END MOTION

Rishikanta Mayengbam

unread,

Feb 6, 2025, 3:34:05 AM2/6/25

to cp...@googlegroups.com

Dear Prof. Marcella,

Thank you for your suggestion! I tried applying smearing to the occupation numbers and reducing the alpha parameter in the density mixing, and it worked wonderfully. The convergence has improved significantly, and the calculations are now much more stable.

The OT method resulted in a weird optimized structure.

--
You received this message because you are subscribed to the Google Groups "cp2k" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cp2k+uns...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/cp2k/24aa87ce-f708-43fb-a7c7-59d7e068f636n%40googlegroups.com.

Rishikanta Mayengbam

unread,

Feb 11, 2025, 2:19:25 AM2/11/25

to cp2k

Hi users,

After the geometry optimization with OT method, the surface layers of the perovskite slab become so weird, some atoms of the molecule break and move far upwards.

So, I tried with the standard diagonalization. But after several steps, I stopped the geometry optimization in the mid-way because the structure changed in the similar fashion as above.

I have no idea what is going wrong here.

Can someone suggest something?

OT method_geo_opt.mp4

geo_OT.inp

geo_dia_method.inp

geo_dia_method.mp4

Tyler Sterling

unread,

Feb 11, 2025, 11:53:30 AM2/11/25

to cp2k

It seems like your structure is metallic since you are using smearing and added mos. In which case, you shouldn't be using OT or other minimization methods. Only use diagonalization. Also, if the structure is blowing up, make sure it is charge neutral.

Rishikanta Mayengbam

unread,

Feb 13, 2025, 3:02:30 AM2/13/25

to cp2k

Thanks Tyler.

My system is a perovskite slab over which an organic molecule is placed, both are semiconductors.

Now, I use diagonalization and removed smearing. The main issue was with the POISSON_PSOLVER, which was set to MT. With ANALYTIC, it converges quite smoothly and fast (in 4 hours, 128 cores).

The optimized structure quite resembles the one obtained from VASP package.

I would like to know: Is dipole corrections very essential for subsequent MD simulations for the optimized structure?

Reply all

Reply to author

Forward