Dear Christopher, Il giorno 05/mar/2011, alle ore 04.59, Christopher O'Brien ha scritto:
> Thank you for your suggestions regarding my problem. In my case, a further > decrease in the electronic convergence threshold, or increase in ecutrho, and > ecutwfc had little effect. I usually work with supercells or NEB calculations > with little or no symmetry. > It turns out that the parameter that affected the most significant reduction > in force was the "upscale" parameter. Your recommended value of 1.D3 was > usually sufficient, but really difficult cases can be handled with 1.D4. This > parameter seemed to be the 'magic bullet' for me when nothing else worked. > Why is the default value so small (10.0)? There's really no magic bullet, it is just a straightforward method to save computer time. You could have chosen conv_thr=1.d-9 and upscale=10, or conv_thr=1.d-6 and upscale=1.d4, you would end up in both cases with an electronic convergence threshold of 1.d-10 when the optimization procedure requires that (i.e. close to the convergence, when forces are small and energy changes very little, because you need very accurate forces to locate the minimum with a good precision). The point is that at the beginning of the relaxation your system is usually far from the optimum ionic configuration, hence it is not worth to spend a lot of cpu time in computing very accurate forces. Defaults are often good, but cannot encompass all possible needs of the users. Regards GS > =================================================================== > Christopher J. O'Brien > cjobrien at ncsu.edu > https://sites.google.com/a/ncsu.edu/cjobrien/ > > Ph.D. Candidate > Computational Materials Group > Department of Materials Science & Engineering > North Carolina State University > __________________________________________________________________ > Please send all documents in PDF. > For Word documents: Please use the 'Save as PDF' option before sending. > =================================================================== > > On Feb 24, 2011, at 4:08 AM, Giuseppe Mattioli wrote: > >> Dear Christopher >> You should try to reduce conv_thr in &ELECTRONS, in order to have a more >> accurate calculation of forces and stress. Try, for instance, with conv_thr >> = 1.0d-9 (and with a lower mixing_beta, 0.2~0.3 or so, if the electronic >> calculation does not converge). >> HTH >> >> Giuseppe >> >> Quoting Christopher O'Brien <cjobrien at ncsu.edu>: >> >>> Below this is a super-cell of an interface of Cu2O and Ti. The problem is >>> simply that the forc_conv_thr is not met within 100 iterations. >>> etot_conv_thr is almost always met first (even decreasing etot_conv_thr to >>> 1.0d-6). Typically, the force in this system decreases to ~0.01 Ry/B with >>> a delta F on the same order of magnitude but fluctuates about this value. >>> This means that I continually get the warning telling me to reduce the >>> convergence threshold. Ideally I would like obtain a maximum force of >>> ~1D-4. >>> Does anyone have any suggestions? I am at a loss. >>> >>> Thanks in advance, >>> Chris O'Brien >>> >>> >>> title = 'Cu2O-Ti-case2' , >>> calculation = 'vc-relax' , >>> restart_mode = 'from_scratch' , >>> wf_collect = .false. , >>> outdir = './', >>> wfcdir = './', >>> pseudo_dir = '/share2/cjobrien/pseudo/', >>> prefix = 'Cu2OTi', >>> disk_io = 'low', >>> verbosity = 'minimal', >>> tstress = .true., >>> tprnfor = .true., >>> forc_conv_thr = 1.0d-4, >>> etot_conv_thr = 1.0d-5, >>> / >>> &SYSTEM >>> ibrav = 0, >>> celldm(1) = 8.1878, >>> nat = 14, >>> ntyp = 3, >>> ecutwfc = 30.0 , >>> ecutrho = 300.0 , >>> occupations = 'smearing' , >>> smearing = 'mv' , >>> degauss = 0.01 , >>> / >>> &ELECTRONS >>> conv_thr = 1.0d-6 , >>> mixing_beta = 0.7 , >>> / >>> &IONS >>> ion_dynamics = 'bfgs', >>> / >>> &CELL >>> cell_dynamics = 'bfgs', >>> press = 0.0, >>> cell_dofree = 'xyz', >>> / >>> CELL_PARAMETERS >>> 0.929429709 0.0 0.0 >>> 0.0 0.929429709 0.0 >>> 0.0 0.0 2.924945006 >>> ATOMIC_SPECIES >>> Cu 58.70000 Cu.pw91-n-van_ak.UPF >>> Ti 47.88000 Ti.pw91-nsp-van.UPF >>> O 16.00000 O.pw91-van_ak.UPF >>> ATOMIC_POSITIONS alat >>> Cu 0 0 0 >>> Cu 0.504691013 0.504691013 0 >>> Cu 0.484702934 0.019988079 0.553031754 >>> Cu 0.019988079 0.484702934 0.553031754 >>> Cu 0.039976158 0.039976158 1.106063508 >>> Cu 0.464714854 0.464714854 1.106063508 >>> O 0.252345506 0.252345506 0.285738983 >>> O 0.024221288 0.024221288 1.544049139 >>> Ti 0.480469725 0.024221288 1.544049139 >>> Ti 0.024221288 0.480469725 1.544049139 >>> Ti 0.484702934 0.484702934 2.015504257 >>> Ti 0.019988079 0.019988079 2.015504257 >>> Ti 0.01575487 0.488936142 2.486959376 >>> Ti 0.488936142 0.01575487 2.486959376 >>> K_POINTS automatic >>> 8 8 4 0 0 0 >>> >>> =================================================================== >>> Christopher J. O'Brien >>> cjobrien at ncsu.edu >>> https://sites.google.com/a/ncsu.edu/cjobrien/ >>> >>> Ph.D. Candidate >>> Computational Materials Group >>> Department of Materials Science & Engineering >>> North Carolina State University >>> __________________________________________________________________ >>> Please send all documents in PDF. >>> 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 >>> >> >> Giuseppe Mattioli >> ISM-CNR, Italy >> >> >> >> > > _______________________________________________ > Pw_forum mailing list > Pw_forum at pwscf.org > http://www.democritos.it/mailman/listinfo/pw_forum ? Gabriele Sclauzero, EPFL SB ITP CSEA PH H2 462, Station 3, CH-1015 Lausanne -------------- next part -------------- An HTML attachment was scrubbed... URL: http://www.democritos.it/pipermail/pw_forum/attachments/20110305/380c84d6/attachment-0001.htm
