[CP2K:9590] SCF_GUESS

[Xiaoming Wang]

Hi Matt,

You're right. After looking at the spin components, the two calculations 
did converge to quite different solutions. So which one should I trust? The 
one with lowest energy? But if I set  SCF_GUESS as ATOMIC, I cannot manage 
to find the that solution with RESTART. How could that be?

Best,
Xiaoming

On Friday, January 26, 2018 at 4:47:52 AM UTC-5, Matt W wrote:
>
> Assuming that the calculation converged, they might converge to different 
> electronic states.
>
> You could try and look at Mulliken spin densities or some other indicator 
> to see if the solutions seem different.
>
> Matt
>
> On Friday, January 26, 2018 at 4:09:47 AM UTC, Xiaoming Wang wrote:
>>
>> Dear Prof. Hutter,
>>
>> I encountered the same problem again. I am doing simple ENERGY 
>> calculations. I have reduced the EPS_SCF to 1.0E-8, but still the total 
>> energy I obtained are different for the two cases: one is using SCF_GUESS 
>> of ATOMIC, and the other is RESTART, where the restart wave function is 
>> from a converged one of a similar structure with several atoms slightly 
>> distorted.  Other features of my calculation are ROKS and SIC included.  I 
>> am wondering why those two setups give such different results. The energy 
>> difference can be as large as 0.01 Ha. Following is my input:
>>
>> &GLOBAL
>>
>>   PROJECT_NAME SS
>>
>>   RUN_TYPE ENERGY
>>
>>   PRINT_LEVEL MEDIUM
>>
>> &END GLOBAL
>>
>> &FORCE_EVAL
>>
>>   METHOD QS
>>
>>   &DFT
>>
>>     BASIS_SET_FILE_NAME BASIS_MOLOPT
>>
>>     POTENTIAL_FILE_NAME GTH_POTENTIALS
>>
>>     WFN_RESTART_FILE_NAME SS-RESTART.wfn
>>
>>     MULTIPLICITY 3
>>
>>     ROKS
>>
>>     &SIC
>>
>>      SIC_METHOD EXPLICIT_ORBITALS
>>
>>      SIC_SCALING_A 0.30
>>
>>      SIC_SCALING_B 0.00
>>
>>     &END SIC
>>
>>     &MGRID
>>
>>       CUTOFF 400
>>
>>       REL_CUTOFF 40
>>
>>     &END MGRID
>>
>>     &XC
>>
>>       &XC_FUNCTIONAL PBE 
>>
>>       &END XC_FUNCTIONAL
>>
>>     &END XC
>>
>>     &SCF
>>
>>       MAX_SCF 20
>>
>>       EPS_SCF 1.0e-8
>>
>>       CHOLESKY INVERSE
>>
>>       SCF_GUESS RESTART
>>
>>       &OT
>>
>>        ROTATION
>>
>>        PRECONDITIONER FULL_KINETIC
>>
>>        ENERGY_GAP 0.01
>>
>>        LINESEARCH 3PNT
>>
>>       &END OT
>>
>>       &OUTER_SCF
>>
>>        EPS_SCF 1.0e-8
>>
>>        MAX_SCF 50
>>
>>       &END OUTER_SCF
>>
>>     &END SCF
>>   &END DFT 
>>   &SUBSYS
>>    ....
>>    &END SUBSYS
>>
>> &END FORCE_EVAL
>>
>>
>>
>> Best,
>> Xiaoming
>>
>>
>> On Friday, October 27, 2017 at 4:32:51 AM UTC-4, jgh wrote:
>>>
>>> Hi 
>>>
>>> SCF_GUESS (RESTART/ATOMIC) is just for the very first guess, 
>>> meaning the first SCF iteration in the first step of a MD. 
>>>
>>> For initial guesses within one calculation (one MD step to the 
>>> next) you have to use 
>>>
>>> FORCE_EVAL / DFT / QS / EXTRAPOLATION  xxxxx 
>>>
>>> In any case, if the final result depends on any of these 
>>> choices, you have another problem, e.g. your convergence criteria 
>>> is not strict enough, or one calculation did not converge within the 
>>> given number of steps etc. 
>>>
>>> regards 
>>>
>>> Juerg Hutter 
>>> -------------------------------------------------------------- 
>>> Juerg Hutter                         Phone : ++41 44 635 4491 
>>> Institut für Chemie C                FAX   : ++41 44 635 6838 
>>> Universität Zürich                   E-mail: hut... at chem.uzh.ch 
>>> Winterthurerstrasse 190 
>>> CH-8057 Zürich, Switzerland 
>>> --------------------------------------------------------------- 
>>>
>>> -----cp... at googlegroups.com wrote: -----To: cp2k <cp... at googlegroups.com> 
>>>
>>> From: Xiaoming Wang 
>>> Sent by: cp... at googlegroups.com 
>>> Date: 10/26/2017 05:28AM 
>>> Subject: [CP2K:9590] SCF_GUESS 
>>>
>>> Dear all, 
>>> For BAND, OPT or MD calculations, one usually sets SCF_GUESS to RESTART, 
>>> which will use the WFN of the previous step as a initial guess. This will 
>>> usually accelerate the convergence of the SCF loops. However, The result is 
>>> different from that with SCF_GUESS sed to ATOMIC. I am confused why the 
>>> final results depend on the initial guess of the wfn. 
>>> Best,Xiaoming 
>>>
>>>
>>>
>>> -- 
>>>
>>> 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+... at googlegroups.com. 
>>>
>>> To post to this group, send email to cp... at googlegroups.com. 
>>>
>>> Visit this group at https://groups.google.com/group/cp2k. 
>>>
>>> For more options, visit https://groups.google.com/d/optout. 
>>>
>>>
>>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20180126/df8564b5/attachment.htm>