The default G-space implementation of exact-exchange terms with USPP is very inefficient. For the same cutoff, a hybrid functional run is much slower with USPP than with NCPP. The real-space implementation (tqr=.true.) should be much more efficient
Paolo On Wed, 2014-10-22 at 10:43 +0200, Florian Lazarevic wrote: > Dear all, > > recently I have installed the latest version of QE (Quantum ESPRESSO 5.10). > I tried to set a hybridfunctional calculation (pbe0). I have used a mixture > of NC-PP and US-PP. At the beginning the calculation seems to run properly. > However, after/during the first few EXX-steps the run seems to stuck. But it > did not crash (no error messages). It just keeps on running without writing > output anymore. You can find the input-file below. > > Is this problem related to the PP-Mixing? Without using US-PP > (Ta.pbe-nsp-van.upf ) I dont observe such issues. > In adition each done EXX-steps take very long time compaired to the none > mixing case (> 400000s). > > Best regards > Florian > > PS: Unfortunatly one of my favorite features - restart from hard crashes - is > no longer supported. Are there plans to change this feature back. > > ##input-file## > &control > calculation = 'scf' > restart_mode='from_scratch', > tstress = .true. > tprnfor = .true. > pseudo_dir = './pseudo', > outdir='./scratch' > prefix='HfO2_R', > / > &system > celldm(1) = 19.5260000 > ibrav= 0, > nat= 96, > ntyp= 3, > nbnd=400, > nosym=.TRUE. > ecutwfc =40.0, > occupations='smearing', > smearing='gaussian', > degauss=0.01 > nspin=2, tot_magnetization=1, > input_dft='pbe0', nqx1 = 1, nqx1 = 1, nqx3 = 1, > exxdiv_treatment='gygi-baldereschi' > ecutvcut=0.7 > x_gamma_extrapolation = .TRUE. > / > &electrons > diagonalization='david' > mixing_mode = 'plain' > mixing_beta = 0.7 > conv_thr = 5.0d-6 > / > ATOMIC_SPECIES > Hf 178.49 hf_semi_NC_ebn_3.upf > O 15.999 o_pbe.upf > Ta 180.95 Ta.pbe-nsp-van.upf > > CELL_PARAMETERS > ... > _______________________________________________ > Pw_forum mailing list > [email protected] > http://pwscf.org/mailman/listinfo/pw_forum _______________________________________________ Pw_forum mailing list [email protected] http://pwscf.org/mailman/listinfo/pw_forum
