[Pw_forum] vc-relax with diverging SCF error
Christopher O'Brien
cjobrien at ncsu.edu
Mon Aug 30 04:15:06 CEST 2010
Stefano, thanks for your response.
--- I have played around with the mixing_beta parameter in regards to this problem before. However, I have found that reducing it slightly (to 0.65, 0.5, etc...) causes either incredible changes in pressure or causes the initial SCF cycle to not converge. Keeping the mixing_beta at the default and letting it run, it can take 48+ hours to finish 3 BFGS steps using 16 procs, 4 pools, and 24 k-points.
--- Typically, the initial SCF cycle smoothly converges and calculates a reasonable pressure. However, after the first BFGS step the pressure skyrockets (1000's of kbar). I have tried to restrict the distance the ions can more and the degree to which the cell contracts but the problem still persists or, if the distance is set too low, the computational cost becomes prohibitive.
--- Going much beyond ecutwfc=30 Ry and ecutrho=300 Ry becomes prohibitively expensive. Even at ecutwfc=35Ry, the same problem occurs.
--- I have also been starting the vc-relax calculation from a 'relax' calculation. This also does not seem to have any effect.
Am I approaching this problem correctly? Should I abandon VC-RELAX completely?
Thanks,
Chris O'Brien
===================================================================
Christopher J. O'Brien
cjobrien at ncsu.edu
http://www4.ncsu.edu/~cjobrien/index.html
https://sites.google.com/a/ncsu.edu/cjobrien/
See what I'm citing at http://www.citeulike.org/cjo123
Ph.D. Candidate
Computational Materials Group
Department of Materials Science & Engineering
North Carolina State University
__________________________________________________________________
Please send all documents in PDF, HTML, RTF, DVI, PS or plain text.
For Word documents: Please use the 'Save as PDF' option before sending.
===================================================================
On Aug 29, 2010, at 3:06 PM, Stefano de Gironcoli wrote:
> I think that 22 Ry may be too small for Cu, even with USPP, especially
> if stress is calculated.
> try to reduce the mixing_beta.
>
> stefano
>
>
> Christopher O'Brien wrote:
>> PWSCF users:
>> I have been struggling for quite some time to perform a vc-relax calculation on a simple system of 96 Cu atoms. I am confident that the input is correct as an ion relaxation run converges in one scf cycle to a pressure of -11.54 kbar with the same parameters as the vc-relax calculation (shown below).
>>
>> As seen in the output file, the problem with the vc-relax calculation is that the SCF error will increase with each iteration. I have tried to remedy this issue by restricting the maximum movement distance of both the cell and ions to no avail. Unfortunately, I cannot turn on symmetry as I am using this as a test case for further calculations of Al-Cu solid solutions.
>>
>> Is there anything terribly wrong with my input? Should I impose symmetry despite the addition of Al to the Cu lattice?
>>
>> Thanks for your help.
>>
>>
>> ------------------------------------------------------------------------
>>
>>
>>
>> &CONTROL
>> calculation = 'vc-relax' ,
>> restart_mode = ‘from_scratch’ ,
>> wf_collect = .false. ,
>> outdir = './' ,
>> pseudo_dir = './' ,
>> prefix = 'Cu' ,
>> forc_conv_thr = 2.0d-2,
>> etot_conv_thr = 1.0d-3,
>> tstress = .true. ,
>> verbosity = 'minimal' ,
>> disk_io = 'none' ,
>> tprnfor = .true. ,
>> /
>> &SYSTEM
>> ibrav = 8,
>> celldm(1) = 9.505756,
>> celldm(2) = 1.735408,
>> celldm(3) = 4.921944,
>> nat = 96,
>> ntyp = 1,
>> ecutwfc = 22 ,
>> ecutrho = 220 ,
>> nosym = .true. ,
>> occupations = 'smearing' ,
>> degauss = 0.022 ,
>> smearing = 'methfessel-paxton' ,
>> /
>> &ELECTRONS
>> conv_thr = 1.0d-5 ,
>> mixing_beta = 0.7 ,
>> /
>> &IONS
>> ion_dynamics = 'bfgs' ,
>> /
>> &CELL
>> press = 0.0 ,
>> press_conv_thr = 5.0d0,
>> cell_dofree = 'xyz' ,
>> cell_dynamics = 'bfgs' ,
>> /
>> ATOMIC_SPECIES
>> Cu 63.54600 Cu.pw91-n-van_ak.UPF
>> ATOMIC_POSITIONS (crystal)
>> Cu 0.000000000 -0.000082163 -0.000018624
>> ...
>> K_POINTS automatic
>> 6 4 2 1 1 1 1
>>
>> ===================================================================
>> Christopher J. O'Brien
>> cjobrien at ncsu.edu
>> https://sites.google.com/a/ncsu.edu/cjobrien/
>> See what I'm citing at http://www.citeulike.org/cjo123
>>
>> Ph.D. Candidate
>> Computational Materials Group
>> Department of Materials Science & Engineering
>> North Carolina State University
>> __________________________________________________________________
>> Please send all documents in PDF, HTML, RTF, DVI, PS or plain text.
>> For Word documents: Please use the 'Save as PDF' option before sending.
>> ===================================================================
>>
>>
>> ------------------------------------------------------------------------
>>
>> _______________________________________________
>> Pw_forum mailing list
>> Pw_forum at pwscf.org
>> http://www.democritos.it/mailman/listinfo/pw_forum
>>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum
More information about the users
mailing list