[CP2K-user] diverging SCF wavefunction optimization for several nanocrystal-ligand systems (after they have been successfully relaxed)

Travis polla... at gmail.com
Thu Oct 25 17:56:27 CEST 2018


Direct mixing is inefficient and the number of steps required for 
convergence grows rapidly with system size. Switch from DIRECT_P_MIXING to 
BROYDEN_MIXING or PULAY_MIXING to reduce the number of steps for 
convergence. 
See https://www.cp2k.org/_media/events:2015_cecam_tutorial:tong_qsscf.pdf


On Wednesday, October 24, 2018 at 10:34:54 AM UTC-4, kat.... at gmail.com 
wrote:

> Hi, 
> I have an issue with diverging SCF wavefunction optimization for  5 
> different systems of silicon nanocrystals with various ligands. So far this 
> has not occurred (I have successfully simulated 20 other silicon 
> nanocrystals with quite complicated ligands), so it does not seem to be 
> consistent issue... 
>
> For silicon nanocrystal with 235 atoms in the core and surface covered 
> with Sulfur I use these parameters (extract from cp2k.out for energy_force):
>
> SCF PARAMETERS         
>                         Density guess:                                   
> RESTART
>                         
> --------------------------------------------------------
>                         max_scf:                                           
>   200
>                         max_scf_history:                                 0
>                         max_diis:                                         
>      4
>                         
> --------------------------------------------------------
>                         eps_scf:                                       
>  5.00E-04
>                         eps_scf_history:                           0.00E+00
>                         eps_diis:                                       
> 1.00E-01
>                         eps_eigval:                                   
> 1.00E-05
>                         
> --------------------------------------------------------
>                         level_shift [a.u.]:                               
>  0.00
>                         added MOs                                       
> 100    0
>                         
> --------------------------------------------------------
>                         Mixing method:                           
> DIRECT_P_MIXING
>                         
> --------------------------------------------------------
>                         No outer SCF
>
> Number of electrons:                                                       
> 1780
>  Number of occupied orbitals:                                             
>    890
>  Number of molecular orbitals:                                             
>  4875
>
>  Number of orbital functions:                                             
>   4875
>  Number of independent orbital functions:                                 
>   4875
>
>  Extrapolation method: initial_guess
>
> and I get this divergence in the optimization already after 2nd step:
>
>  SCF WAVEFUNCTION OPTIMIZATION
>
>   Step     Update method      Time    Convergence         Total energy   
>  Change
>   
> ------------------------------------------------------------------------------
>      1 P_Mix/Diag. 0.40E+00    4.2     0.10908716     -2354.4136559975 
> -2.35E+03
>      2 P_Mix/Diag. 0.40E+00    5.0     0.44462605     -2354.4126056799 
>  1.05E-03
>      3 P_Mix/Diag. 0.40E+00    5.1     0.53885704     -2354.3562265414 
>  5.64E-02
>      4 P_Mix/Diag. 0.40E+00    5.0     3.07899832     -2352.5424379413 
>  1.81E+00
>      5 P_Mix/Diag. 0.40E+00    5.0    33.29887563     -1891.1308608487 
>  4.61E+02
>      6 P_Mix/Diag. 0.40E+00    5.3    30.17765973      4133.1886655512 
>  6.02E+03
>      7 P_Mix/Diag. 0.40E+00    5.1    32.02311479      3797.3039280923 
> -3.36E+02
>      8 P_Mix/Diag. 0.40E+00    5.1    31.83587407      1215.6949983041 
> -2.58E+03
>      9 P_Mix/Diag. 0.40E+00    5.1    33.45037356      4635.2206528830 
>  3.42E+03
>     10 P_Mix/Diag. 0.40E+00    5.1    30.96157640      2861.7052141337 
> -1.77E+03
>     11 P_Mix/Diag. 0.40E+00    5.1    29.36705839      2709.3853227498 
> -1.52E+02
>     12 P_Mix/Diag. 0.40E+00    5.5    31.13837241      3593.9217491598 
>  8.85E+02
>     13 P_Mix/Diag. 0.40E+00    5.1    34.63292667      4412.8333731638 
>  8.19E+02
>     14 P_Mix/Diag. 0.40E+00    5.1    30.85152040      4458.1077941490 
>  4.53E+01
> .......
>  Leaving inner SCF loop after reaching   200 steps.
>
>
>   Electronic density on regular grids:      -1779.9999999908       
>  0.0000000092
>   Core density on regular grids:             1779.9999997505       
> -0.0000002495
>   Total charge density on r-space grids:       -0.0000002402
>   Total charge density g-space grids:          -0.0000002402
>
>   Overlap energy of the core charge distribution:               
> 0.00000002843702
>   Self energy of the core charge distribution:             
>  -5795.76572189967919
>   Core Hamiltonian energy:                                   
> 2938.96179612088508
>   Hartree energy:                                           
>  7673.11151856082142
>   Exchange-correlation energy:                               
> -640.40137237794443
>
>   Total energy:                                             
>  4175.90622043252006
>
>  *** WARNING in qs_scf.F:479 :: SCF run NOT converged ***
>
>
>  ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):           
>  55520.780056994452025
>
> But the system is fully relaxed, so the SCF run converged before (for the 
> GEO_OPT) - see below the relaxation cp2k.out extract:
>
> --------  Informations at step =    25 ------------
>   Optimization Method        =                   CG
>   Total Energy               =     -2354.4137112872
>   Real energy change         =        -0.0000012787
>   Decrease in energy         =                  YES
>   Used time                  =              475.723
>
>   Convergence check :
>   Max. step size             =         0.0023939758
>   Conv. limit for step size  =         0.0030000000
>   Convergence in step size   =                  YES
>   RMS step size              =         0.0003373842
>   Conv. limit for RMS step   =         0.0015000000
>   Convergence in RMS step    =                  YES
>   Max. gradient              =         0.0004156999
>   Conv. limit for gradients  =         0.0004500000
>   Conv. in gradients         =                  YES
>   RMS gradient               =         0.0000585848
>   Conv. limit for RMS grad.  =         0.0003000000
>   Conv. in RMS gradients     =                  YES
>  ---------------------------------------------------
>
>
>  *******************************************************************************
>  ***                    GEOMETRY OPTIMIZATION COMPLETED                   
>    ***
>
>  *******************************************************************************
>
>                     Reevaluating energy at the minimum
>
>  Number of electrons:                                                     
>   1780
>  Number of occupied orbitals:                                             
>    890
>  Number of molecular orbitals:                                             
>   890
>
>  Number of orbital functions:                                             
>   4875
>  Number of independent orbital functions:                                 
>   4875
>
>  Extrapolation method: ASPC
>
>
>  SCF WAVEFUNCTION OPTIMIZATION
>
>   ----------------------------------- OT 
> ---------------------------------------
>   Minimizer      : DIIS                : direct inversion
>                                          in the iterative subspace
>                                          using   7 DIIS vectors
>                                          safer DIIS on
>   Preconditioner : FULL_KINETIC        : inversion of T + eS
>   Precond_solver : DEFAULT
>   stepsize       :    0.15000000                  energy_gap     :   
>  0.20000000
>   eps_taylor     :   0.10000E-15                  max_taylor     :         
>     4
>   ----------------------------------- OT 
> ---------------------------------------
>
>   Step     Update method      Time    Convergence         Total energy   
>  Change
>   
> ------------------------------------------------------------------------------
>      1 OT DIIS     0.15E+00    3.5     0.00000227     -2354.4136673140 
> -2.35E+03
>      2 OT DIIS     0.15E+00    4.2     0.00000210     -2354.4136705162 
> -3.20E-06
>      3 OT DIIS     0.15E+00    4.1     0.00000166     -2354.4136799605 
> -9.44E-06
>      4 OT DIIS     0.15E+00    4.1     0.00000127     -2354.4136892114 
> -9.25E-06
>      5 OT DIIS     0.15E+00    4.1     0.00000111     -2354.4136936860 
> -4.47E-06
>      6 OT DIIS     0.15E+00    4.1     0.00000103     -2354.4136965337 
> -2.85E-06
>      7 OT DIIS     0.15E+00    4.1     0.00000092     -2354.4137006183 
> -4.08E-06
>
>   *** SCF run converged in     7 steps ***
>
>
>   Electronic density on regular grids:      -1779.9999999998       
>  0.0000000002
>   Core density on regular grids:             1779.9999997509       
> -0.0000002491
>   Total charge density on r-space grids:       -0.0000002490
>   Total charge density g-space grids:          -0.0000002490
>
>   Overlap energy of the core charge distribution:               
> 0.00000002843702
>   Self energy of the core charge distribution:             
>  -5795.76572189967919
>   Core Hamiltonian energy:                                   
> 1266.52936221037999
>   Hartree energy:                                           
>  2771.64162064259563
>   Exchange-correlation energy:                               
> -596.81896159999383
>
>   Total energy:                                             
> -2354.41370061826001
>
>  ENERGY| Total FORCE_EVAL ( QS ) energy (a.u.):           
>  -2354.413700618260009
>
> Please, anyone has an idea what am I doing wrong? 
>
> Thank you in advance for any advice... 
>
> best regards
>
> Katerina (Uni Amsterdam, NL)
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20181025/ad3e5602/attachment-0001.html>


More information about the CP2K-user mailing list