[CP2K-user] [CP2K:14172] Re: SCF convergence issues

Ryan Rogers rr... at nyu.edu
Fri Nov 13 23:25:34 UTC 2020


Dear Marcella,

Thank you very much for your suggestion. I tested removing the MM region,
but unfortunately see similar behavior.
My configurations are generated from pure MM simulations with a custom
force field which could possibly be allowing certain atoms to get a little
too close. However, as the thermalized atom positions only vary (for the
most part) by a few tenths of an Angstrom, no problematic close contacts
can be found visually.

I've pasted below an example of the first SCF cycle from such a job,
showing that the "Total energy" starts off much lower than I'm expecting
(usually on the order of -2,000 +/- 500). During the SCF, the "Total
energy" drops very low. When a job like this is allowed to continue, it
won't crash on its own, but the "Total charge density on r-space grids"
will become too large after 2-3 SCF cycles.

Is there any advice about how to handle atoms that are potentially "close"
but not chemically "wrong"? These could potentially include
hydrogen-bonding pairs like O-H, N-H, etc.

Any help is greatly appreciated in advance!

Sincerely,
Ryan Rogers
rr... at nyu.edu <trr... at email.uark.edu>
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

################################################################################
 SCF WAVEFUNCTION OPTIMIZATION

  ----------------------------------- OT
---------------------------------------
  Minimizer      : DIIS                : direct inversion
                                         in the iterative subspace
                                         using   7 DIIS vectors
                                         safer DIIS on
  Preconditioner : FULL_ALL            : diagonalization, state selective
  Precond_solver : DEFAULT
  stepsize       :    0.15000000                  energy_gap     :
 0.00100000
  eps_taylor     :   0.10000E-15                  max_taylor     :
    4
  ----------------------------------- OT
---------------------------------------

  Step     Update method      Time    Convergence         Total energy
 Change

------------------------------------------------------------------------------
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     1 OT DIIS     0.15E+00   66.7     0.13322879     -8451.7456190410
-8.45E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     2 OT DIIS     0.15E+00   65.4     0.13364185    -10643.9042780022
-2.19E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     3 OT DIIS     0.15E+00   65.7     0.15497766    -13025.7969156521
-2.38E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     4 OT DIIS     0.15E+00   65.5     0.14868108    -14464.8777105315
-1.44E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     5 OT SD       0.15E+00   65.5     0.17759857    -14843.1753581151
-3.78E+02
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     6 OT DIIS     0.15E+00   65.8     0.32256964    -13383.1417204580
 1.46E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     7 OT DIIS     0.15E+00   65.6     0.19569799    -14813.0440620911
-1.43E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     8 OT DIIS     0.15E+00   65.3     0.24452346    -15865.0025958057
-1.05E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
     9 OT DIIS     0.15E+00   65.5     0.26678603    -14795.2757334147
 1.07E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    10 OT DIIS     0.15E+00   65.7     0.36307397    -17004.3658543152
-2.21E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    11 OT DIIS     0.15E+00   65.6     0.48782211    -17176.2139668702
-1.72E+02
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    12 OT DIIS     0.15E+00   65.5     0.45773260    -15669.9100288352
 1.51E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    13 OT DIIS     0.15E+00   65.5     0.66194382    -18842.9056350951
-3.17E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    14 OT DIIS     0.15E+00   65.6     0.63269527    -16132.4918799441
 2.71E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    15 OT DIIS     0.15E+00   65.7     0.73064928    -16834.2508612376
-7.02E+02
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    16 OT DIIS     0.15E+00   65.5     1.15829587    -27460.2661838299
-1.06E+04
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    17 OT DIIS     0.15E+00   65.7     0.70782891    -15281.5935133825
 1.22E+04
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    18 OT DIIS     0.15E+00   65.7     0.98480536    -20548.0516400897
-5.27E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    19 OT DIIS     0.15E+00   65.6     1.39954759    -33842.8143338371
-1.33E+04
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    20 OT DIIS     0.15E+00   65.6     1.33809404    -29354.7460547364
 4.49E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    21 OT DIIS     0.15E+00   65.4     1.62600866    -36147.2741396100
-6.79E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    22 OT DIIS     0.15E+00   65.6     1.21990609    -24956.2600892884
 1.12E+04
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    23 OT DIIS     0.15E+00   65.6     1.32400959    -27440.1826004675
-2.48E+03
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    24 OT DIIS     0.15E+00   65.5     1.81006425    -42812.1680574805
-1.54E+04
  Adding QM/MM electrostatic potential to the Kohn-Sham potential.
    25 OT DIIS     0.15E+00   65.7     0.89367434    -18701.9345340245
 2.41E+04

  Leaving inner SCF loop after reaching    25 steps.


  Electronic density on regular grids:      -1218.0000000005
-0.0000000005
  Core density on regular grids:             1217.9999999992
-0.0000000008
  Total charge density on r-space grids:       -0.0000000013
  Total charge density g-space grids:          -0.0000000013

  Overlap energy of the core charge distribution:
0.00009990728015
  Self energy of the core charge distribution:
 -4772.82460287547656
  Core Hamiltonian energy:
2367.11980010682919
  Hartree energy:
 -15786.54494725671611
  Exchange-correlation energy:
-509.06361063631039
  Dispersion energy:
-0.62127327012014
  QM/MM Electrostatic energy:
0.00000000000000

  Total energy:
 -18701.93453402451269

  outer SCF iter =    1 RMS gradient =   0.89E+00 energy =
-18701.9345340245
################################################################################



On Thu, Nov 5, 2020 at 10:20 AM Женя Елизарова <zhene... at gmail.com>
wrote:

> I see.
>  Actually, I would like to know is it possible to run a single point
> energy calculation of the QM subsystem without the removal of the MM part,
> not in relation to the previous posts (in general)?
>
>
> Best wishes,
> Evgenia
>
> четверг, 5 ноября 2020 г. в 19:08:44 UTC+3, Marcella Iannuzzi:
>
>>
>> Dear Evgenia Elizarova
>>
>> Is this question related to the previous posts in this conversation?
>> If yes, what I meant is to remove all the MM part and just carry out a
>> DFT calculation of the QM part.
>> Regards
>> Marcella
>>
>> On Thursday, November 5, 2020 at 3:29:42 PM UTC+1 zh... at gmail.com
>> wrote:
>>
>>>
>>> Dear  Marcella Iannuzzi
>>>
>>> I've just started to explore opportunities of the cp2k package. I've
>>> done some tutorials about single-point calculations of ethane molecule,
>>> QM/MM simulations, and some more. I am very interested in single point
>>> energy calculation for the QM part of the system. As I understood, I have
>>> to define the force_eval section (method - quickstep), and also I have to
>>> define subsections: dft, subsys, qmmm. Did I understand correctly? Also, i
>>> have some questions.
>>> Do I have to define the MM subsection? In subsys section, I should
>>> define the whole system?  How to define for which part of the system run a
>>> single point calculation?
>>>  Could you help me, please?
>>>
>>> Best wishes,
>>> Evgenia Elizarova
>>> понедельник, 2 ноября 2020 г. в 11:55:28 UTC+3, Marcella Iannuzzi:
>>>
>>>> Dear Ryan Rogers
>>>>
>>>> There is apparently a problem with the conservation of the charge on
>>>> the QM grid.
>>>>
>>>> Did you try to run a single point energy calculation for the QM part
>>>> alone?
>>>> Kind regards
>>>> Marcella
>>>>
>>>> On Thursday, October 29, 2020 at 7:11:42 PM UTC+1 r... at nyu.edu wrote:
>>>>
>>>>> I might add that I have not been able to identify any obvious problems
>>>>> with the configurations (e.g. overlapping or too close atoms, etc.) when I
>>>>> encounter these errors.
>>>>>
>>>>> On Monday, October 26, 2020 at 4:07:46 PM UTC-5 Ryan Rogers wrote:
>>>>>
>>>>>> Dear CP2K community,
>>>>>>
>>>>>> I am having issues in DFT QM/MM force calculations on molecular
>>>>>> crystals of paracetamol (acetaminophen). I am describing here the two
>>>>>> problems I most often experience. I am currently unable to identify the
>>>>>> cause or any pattern in the problems I encounter. The root of the problems
>>>>>> could be something other than what I identify below; I am pointing out the
>>>>>> problematic features in the output that are most obviously to me. All input
>>>>>> and output files are included.
>>>>>>
>>>>>> *1. Total energy falls into "hole" and never converges.
>>>>>> (CP2K_problemTotalE_conf_0636.tar.gz)*
>>>>>> Personal experience tells me to expect a "Total energy" for these
>>>>>> systems on the order of -2,000 (Hartree) and a "Hartree energy" on the
>>>>>> order of +2,000 (Hartree).
>>>>>> In these jobs, I find an initial "Hartree energy" on the order of
>>>>>> >-10,000 (Hartree), which appears to send the SCF wavefunction optimization
>>>>>> down a path of non-convergence, in which the "Total energy" can easily
>>>>>> become on the order of -100,000 (Hartree) before I kill the job.
>>>>>>
>>>>>> *2. Total charge density on grids grows too large.
>>>>>> (CP2K_problemEGrids_conf_0623.tar.gz)*
>>>>>> In these jobs, the Total energy looks reasonable, and the Convergence
>>>>>> looks promising in the few SCF cycles of steps.
>>>>>> However, the total Change never drops below my threshold, and
>>>>>> eventually the "Total charge density on r-space/g-space grids" becomes much
>>>>>> too large.
>>>>>>
>>>>>> My configurations are extracted from MD trajectories, so the atoms
>>>>>> have perturbations from their perfect crystal positions. One confusing
>>>>>> observation is that very similar QM/MM configurations selected from other
>>>>>> frames of the same trajectory often have no problems.
>>>>>> My configurations are constructed from a cluster of several molecules
>>>>>> in the QM region with usually another layer usually 1-2 molecules thick
>>>>>> making up the MM region. (In the attached sample images, the size of the
>>>>>> stick molecules alludes to a larger/smaller basis set used, while the MM
>>>>>> atoms are denoted as points.) Because I am not including integer numbers of
>>>>>> unit cells, I am not using PBC.
>>>>>>
>>>>>> Any advice about both/either problem will be greatly appreciated!
>>>>>>
>>>>>> Sincerely,
>>>>>> Ryan Rogers
>>>>>> r... at nyu.edu
>>>>>> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>>>>>>
>>>>> --
> 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 cp... at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/0c182814-c0e4-42f2-84e0-cae9bb11377an%40googlegroups.com
> <https://urldefense.proofpoint.com/v2/url?u=https-3A__groups.google.com_d_msgid_cp2k_0c182814-2Dc0e4-2D42f2-2D84e0-2Dcae9bb11377an-2540googlegroups.com-3Futm-5Fmedium-3Demail-26utm-5Fsource-3Dfooter&d=DwMFaQ&c=slrrB7dE8n7gBJbeO0g-IQ&r=qS7CCskWE2QhguKgOenq3w&m=wgaFjFy2qpPxzILbfwkIfmQ8sUL66bmnFebYZvRJvmI&s=Blmbnkyn5duoggvGHwBKMbJw2m5smcSZCQHB01QHAIc&e=>
> .
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20201113/cca0425b/attachment.htm>


More information about the CP2K-user mailing list