[CP2K:7031] Re: Molecular crystal in a triclinic cell
Alex
nedo... at gmail.com
Thu Oct 8 01:58:26 CEST 2015
Thanks, that is really useful, but still not much of a gain in terms of
speed.
I was also looking at trying D3, but the examples from slide #14 are
definitely missing from my CP2K installation. I suppose I will stick with
PBE + NON_LOCAL VDW for the moment, just to get some idea, but then consult
with the cited paper.
Alex
On Wednesday, October 7, 2015 at 5:42:20 PM UTC-6, S Ling wrote:
>
> Hi
>
> Please replace your &SCF section with the following settings, which should
> significantly accelerate your SCF calculations:
>
> &SCF
> MAX_SCF 20
> EPS_SCF 1.000E-06
> SCF_GUESS RESTART
> &OT T
> MINIMIZER DIIS
> PRECONDITIONER FULL_ALL
> ENERGY_GAP 0.001
> &END OT
> &OUTER_SCF T
> EPS_SCF 1.000E-06
> MAX_SCF 20
> &END OUTER_SCF
> &END SCF
>
> You can use PBE pseudopotential for BEEF functional. It takes time to
> generate and test the performance of new basis sets and pseudopotentials.
> The PBE pseudopotential is perhaps the "closest" choice you have for BEEF
> functional, and it should be good enough for BEEF functional, but you need
> to test it.
>
> In terms of choice of XC functional, please note BEEF functional contains
> parameters which were optimised for several data sets chosen by the
> original authors. There is no guarantee that it is good for your system. So
> you need to test the functional on your specific system. You may want to
> try PBE+D3 which I mentioned in the slides. Generally speaking, D2/D3
> should be cheaper than non-local vdW based methods, but they may have their
> own advantages and disadvantages.
>
> With respect to the supercell sizes, please keep in mind CP2K uses
> Gamma-point approximation. I am not sure whether you will be able to get
> correct electronic structure of your system if you use a 1x1x1 cell with
> cell dimensions smaller than 5 A. If the electronic structure is wrong,
> then the forces, stress tensor and optimised lattice parameters will also
> be wrong. You will be more likely to get better electronic structure if you
> use bigger supercells.
>
> SL
>
>
> On 7 October 2015 at 19:34, Alex <ned... at gmail.com <javascript:>> wrote:
>
>> Thanks Matt.
>>
>> I enforced the angles and changed my SCF section to, hope this looks
>> reasonable:
>>
>> &SCF
>> &OT ON
>> MINIMIZER DIIS
>> PRECONDITIONER FULL_ALL
>> ENERGY_GAP 0.001
>> &END OT
>> SCF_GUESS RESTART
>> EPS_SCF 5.0E-6
>> MAX_SCF 300
>> &END SCF
>>
>> Given the above, I do not see much of an improvement speed-wise, to be
>> honest. The memory footprint is about half of what it was, I can see that
>> much.
>>
>> Your comment regarding XC is definitely well taken, we here don't believe
>> PBE+VDW out of the box should be accurate. I am tempted to try the setup
>> from p. 21 of the pdf you suggested, but does this BEEF work in stock CP2K
>> and does one need to change the potentials (I am using GTH potentials). I
>> suppose this boils down to whether there is an input file from the authors
>> for others to try...
>>
>> Alex
>>
>>
>> On Wednesday, October 7, 2015 at 2:37:42 AM UTC-6, Matt W wrote:
>>>
>>> Hi Alex,
>>>
>>> one concern would be the XC functional. There are some slides from a
>>> recent CECAM meeting at
>>> http://www.cp2k.org/_media/events:2015_cecam_tutorial:ling_vdw.pdf with
>>> some partial inputs of setting up DFT-vdw functionals. I'm not sure PBE+vdw
>>> is great.
>>>
>>> Otherwise, there are some things that might be a bit more efficient:
>>>
>>> Changing to the OT method instead of using diagonalization would give
>>> you a very large speed up, I would think (order of magnitude probably here).
>>>
>>> It might be a good idea to fix the angles of the cell and do an
>>> optimization (CELL_OPT/KEEP_ANGLES) then optimize again without the
>>> constraint.
>>>
>>> Matt
>>>
>>> On Tuesday, October 6, 2015 at 7:35:18 PM UTC+1, Alex wrote:
>>>>
>>>> Hi all,
>>>>
>>>> We're trying to simulate an n-octane crystal (triclinic, coordinates
>>>> and cell vectors from literature) -- just a static cell + geometry
>>>> optimization. Here's the issue: when simulating a single cell (1 1 1), the
>>>> crystal cohesive energy is far above the experiment -- an order of
>>>> magnitude larger in absolute value. As a vacuum ref we're simulating an
>>>> isolated octane molecule. In all cases, a NON_LOCAL vdw energy correction
>>>> is used.
>>>>
>>>> When simulating a total of eight cells (2 2 2), and the simulation is
>>>> still running, so the results are preliminary, things appear to be much
>>>> more reasonable. The input file is below (I can provide the coordinates as
>>>> well, if needed). Can you just by simple inspection tell me if anything is
>>>> obviously wrong?
>>>>
>>>> Thanks a lot,
>>>>
>>>> Alex
>>>>
>>>> **********************************
>>>> &GLOBAL
>>>> PROJECT opt
>>>> PRINT_LEVEL MEDIUM
>>>> RUN_TYPE CELL_OPT
>>>> &END GLOBAL
>>>> &MOTION
>>>> &GEO_OPT
>>>> OPTIMIZER BFGS
>>>> &END
>>>> &CELL_OPT
>>>> EXTERNAL_PRESSURE [bar] 1.0
>>>> OPTIMIZER BFGS
>>>> &END
>>>> &END MOTION
>>>>
>>>> &FORCE_EVAL
>>>> METHOD QS
>>>> STRESS_TENSOR ANALYTICAL
>>>> &DFT
>>>> BASIS_SET_FILE_NAME BASIS_SET2
>>>> POTENTIAL_FILE_NAME GTH_POTENTIALS2
>>>> &MGRID
>>>> NGRIDS 5
>>>> CUTOFF 500
>>>> &END MGRID
>>>> &QS
>>>> METHOD GPW
>>>> &END QS
>>>> &SCF
>>>> SCF_GUESS RESTART
>>>> EPS_SCF 5.0E-6
>>>> MAX_SCF 300
>>>> ADDED_MOS 100
>>>> &MIXING
>>>> METHOD BROYDEN_MIXING
>>>> ALPHA 0.1
>>>> NBROYDEN 8
>>>> &END MIXING
>>>> &END SCF
>>>> &XC
>>>> &XC_FUNCTIONAL PBE
>>>> &END XC_FUNCTIONAL
>>>> &VDW_POTENTIAL NON_LOCAL
>>>> &END VDW_POTENTIAL
>>>> &END XC
>>>> &END DFT
>>>> &SUBSYS
>>>> &CELL
>>>> ABC 4.160 4.750 11.000
>>>> ANGLES 94.80 84.50 105.10
>>>> PERIODIC XYZ
>>>> SYMMETRY TRICLINIC
>>>> MULTIPLE_UNIT_CELL 2 2 2
>>>> &END
>>>> &TOPOLOGY
>>>> COORD_FILE_NAME noct_2mol.pdb
>>>> COORD_FILE_FORMAT PDB
>>>> MULTIPLE_UNIT_CELL 2 2 2
>>>> &END
>>>> &KIND C
>>>> BASIS_SET DZVP-MOLOPT-SR-GTH
>>>> POTENTIAL GTH-PBE-q4
>>>> &END KIND
>>>> &KIND H
>>>> BASIS_SET DZVP-MOLOPT-SR-GTH
>>>> POTENTIAL GTH-PBE-q1
>>>> &END KIND
>>>> &END SUBSYS
>>>> &END
>>>>
>>>> --
>> 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 <javascript:>.
>> To post to this group, send email to cp... at googlegroups.com <javascript:>
>> .
>> Visit this group at http://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/20151007/c9a68d66/attachment.html>
More information about the CP2K-user
mailing list