[Pw_forum] XDM stress dependent on k-point parallelization?
Alberto Otero de la Roza
aoterodelaroza at gmail.com
Tue Aug 1 17:22:46 CEST 2017
Yes, sorry, there is a bug in the XDM code related to the
parallelization. All XDM quantities are correct with the default
number of pools (i.e. mpirun -n x pw.x), so I recommend not using this
for now. I'll try to get it fixed soon and let the forum know when
Dr. Alberto Otero de la Roza
Department of Chemistry,
University of British Columbia, Okanagan
Kelowna, British Columbia, Canada V1V 1V7
* Malte Sachs <malte.sachs at chemie.uni-marburg.de> [2017-08-01 15:08:58 +0200]:
> Dear all,
> I have tried to perform vc-relax calculations with Quantum Espresso v6.0
> using the XDM Van-der-Waals correction. I get some strange results, because
> it seems that the calculated XDM stress depends on the k-point
> parallelization. As a test I performed three scf calculations using the same
> input (see attached) just variing the number of k-point pools:
> 2 pools with 3 k-points: XDM stress (-33.37 -33.37 -40.79)
> 3 pools with 2 k-points: XDM stress (-58.75 -58.75 -61.19)
> 6 pools with 1 k-point: XDM stress (-117.51 -117.51 -122.37)
> The stress ratio 1: 1/2 : 1/3 is the inverse of the k-point ratio per
> pool... Has somebody seen something like this before?
> Best regards,
> Malte Sachs
> Anorganische Chemie, Fluorchemie
> Philipps-Universität Marburg
> Hans-Meerwein-Straße 4
> 35032 Marburg (Paketpost: 35043 Marburg)
> Tel.: +49 (0)6421 28 - 25 39 5
> calculation = 'scf',
> title = 'UCl6_P-3m1'
> verbosity = 'high'
> prefix = 'UCl6_P3'
> outdir = './outdir/'
> pseudo_dir = '/home/Sachsm/Promotion/Pseudos/'
> tstress = .true.,
> wf_collect = .true.
> etot_conv_thr = 1.0D-8,
> forc_conv_thr = 1.0D-6
> ibrav = 4,
> celldm(1) = 20.55532424,
> celldm(3) = 0.461663392,
> nat = 21,
> ntyp = 2,
> nbnd = 90,
> occupations = 'fixed',
> ecutwfc = 100,
> ecutrho = 800,
> vdw_corr = 'xdm',
> xdm_a1 = 0.6512,
> xdm_a2 = 1.4633,
> mixing_beta = 0.4,
> mixing_mode = 'plain',
> conv_thr = 1.D-10,
> K_POINTS (automatic)
> 2 2 4 0 0 0
> Pw_forum mailing list
> Pw_forum at pwscf.org
More information about the users