Problems with Maximum Overlap Model (MOM)
Matt W
mattwa... at gmail.com
Mon May 15 10:49:34 CEST 2017
Yes, unfortunately that has been our experience too.
Possibly the algorithm depends quite strongly on the convergence
accelerator used and something better than linear mixing, but less
aggressive than DIIS is needed. Nothing that is directly available in CP2K
seems optimal anyway.
Matt
On Monday, May 15, 2017 at 9:36:54 AM UTC+1, Vladimir Rybkin wrote:
>
> Dear Ivan,
>
> according to my own experience MOM is not the most reliable approach. For
> non-trivial systems, I was never able to achieve convergence and always
> collapsed to the ground states, independently of my settings. I would not
> high hopes if I were you.
>
> Yours,
>
> Vladimir
>
> воскресенье, 14 мая 2017 г., 0:40:47 UTC+2 пользователь Ivan Infante
> написал:
>>
>> Hi,
>>
>> I am trying to use MOM and I have some problems.
>> I start from a ground state closed-shell converged wavefunction, and I
>> use the restart file for the MOM calculation. I just de-occupy the beta
>> HOMO to occupy the LUMO beta.
>> The HOMO-LUMO gap of this system is around 2.0 eV, so there shouldnt be
>> any problem to converge the excited state.
>> However, after a few iterations, when the DIIS kicks in, it immediately
>> swaps to the ground state energy (E = -5725.1289....) .
>>
>> Step Update method Time Convergence Total energy
>> Change
>>
>>
>> ------------------------------------------------------------------------------
>>
>> 1 P_Mix/Diag. 0.40E+00 13.8 0.32547826 -5724.9981579203
>> -5.72E+03
>>
>> 2 P_Mix/Diag. 0.40E+00 16.1 0.18145011 -5725.0429211687
>> -4.48E-02
>>
>> 3 P_Mix/Diag. 0.40E+00 16.0 0.11353690 -5725.0753406595
>> -3.24E-02
>>
>> 4 P_Mix/Diag. 0.40E+00 16.1 0.06733805 -5725.0961299910
>> -2.08E-02
>>
>> 5 DIIS/Diag. 0.20E-02 16.1 0.04123847 -5725.1090261343
>> -1.29E-02
>>
>> 6 DIIS/Diag. 0.11E-02 16.1 0.00192988 -5725.1289018949
>> -1.99E-02
>>
>> 7 DIIS/Diag. 0.31E-03 16.1 0.00128342 -5725.1289258147
>> -2.39E-05
>>
>> 8 DIIS/Diag. 0.28E-03 16.1 0.00032442 -5725.1289261285
>> -3.14E-07
>>
>> 9 DIIS/Diag. 0.96E-04 16.1 0.00018951 -5725.1289310034
>> -4.87E-06
>>
>> 10 DIIS/Diag. 0.45E-04 16.1 0.00004841 -5725.1289312796
>> -2.76E-07
>>
>> 11 DIIS/Diag. 0.17E-04 16.1 0.00003119 -5725.1289314498
>> -1.70E-07
>>
>> 12 DIIS/Diag. 0.11E-04 16.1 0.00001487 -5725.1289314674
>> -1.76E-08
>>
>> 13 DIIS/Diag. 0.28E-05 16.1 0.00000367 -5725.1289314718
>> -4.34E-09
>>
>>
>> *** SCF run converged in 13 steps ***
>>
>> I tried to play with the alpha mixing parameter, with a EPS_DIIS tight
>> (to avoid that the DIIS interfere) but no success:
>>
>> 1 P_Mix/Diag. 0.10E-03 13.8 0.32547826 -5724.9981579203
>> -5.72E+03
>>
>> 2 P_Mix/Diag. 0.10E-03 16.1 0.32544318 -5724.9981819920
>> -2.41E-05
>>
>> 3 P_Mix/Diag. 0.10E-03 16.1 0.32540810 -5724.9982060548
>> -2.41E-05
>>
>> 4 P_Mix/Diag. 0.10E-03 16.1 0.32537302 -5724.9982301087
>> -2.41E-05
>>
>> 5 P_Mix/Diag. 0.10E-03 16.1 0.32533795 -5724.9982541538
>> -2.40E-05
>>
>> 6 P_Mix/Diag. 0.10E-03 16.1 0.32530289 -5724.9982781901
>> -2.40E-05
>>
>> 7 P_Mix/Diag. 0.10E-03 16.1 0.32526782 -5724.9983022175
>> -2.40E-05
>>
>> 8 P_Mix/Diag. 0.10E-03 16.1 0.32523276 -5724.9983262361
>> -2.40E-05
>>
>> 9 P_Mix/Diag. 0.10E-03 16.1 0.32519771 -5724.9983502459
>> -2.40E-05
>>
>> 10 P_Mix/Diag. 0.10E-03 16.1 0.32516265 -5724.9983742469
>> -2.40E-05
>>
>> 11 P_Mix/Diag. 0.10E-03 16.1 0.32512760 -5724.9983982391
>> -2.40E-05
>>
>> 12 P_Mix/Diag. 0.10E-03 16.1 0.32509256 -5724.9984222225
>> -2.40E-05
>>
>> 13 P_Mix/Diag. 0.10E-03 16.1 0.32505752 -5724.9984461970
>> -2.40E-05
>>
>> 14 P_Mix/Diag. 0.10E-03 16.1 0.32502248 -5724.9984701628
>> -2.40E-05
>>
>> 15 P_Mix/Diag. 0.10E-03 16.1 0.32498745 -5724.9984941198
>> -2.40E-05
>>
>> 16 P_Mix/Diag. 0.10E-03 16.1 0.32495242 -5724.9985180680
>> -2.39E-05
>>
>> Here it seems that is converging, but the EPS_SCF remains very high.
>> When I use intermediete values of ALPHA, like 0.1 or 0.05, the
>> calculation goes slow, but tends still to converge to the ground state.
>>
>> Is there something that I can use to converge my calculation to the
>> excited state ?
>>
>> Thanks,
>> Ivan
>>
>>
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20170515/942d687f/attachment.html>
More information about the CP2K-user
mailing list