[Pw_forum] Increase in number of diagonalization errors

Heather Whitley whitley3 at llnl.gov
Wed Jan 14 19:20:58 CET 2009


Hi,

I recently started using the new version of pwscf from QE 4.0.4; I had 
previously been using QE 3.2.3.  I've noticed that it seems like I get 
these Cholesky diagonalization errors much more frequently with the new 
version of the code.  I've read the advice on how to avoid these errors 
(running with the "-ndiag 1" flag or changing the option for 
diagonalization method). I am, however, curious if there is some reason 
why I would see these errors more frequently with the new version? 
Maybe it's something in my input?  I've attached the text from the error 
(typical of what other people have asked about in previous threads), as 
well as a sample input file.

Thank you,
Heather

 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
      task #        63
      from  pdpotf  : error #       422
       problems computing cholesky decomposition
 
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

  &control
     calculation='nscf'
     restart_mode='restart'
     tstress=.true.
     tprnfor=.true.
     etot_conv_thr=1.0d-5
     forc_conv_thr=1.0d-4
     pseudo_dir = './'
     outdir='./'
     title='cd15se15'
     prefix='cd15se15'
     wf_collect = .false.
     max_seconds=1.43e+04
     nstep=2500
     tefield=.false.
     dipfield=.false.
  /
  &system
     ibrav=8
     celldm(1) =60.261212118
     celldm(2) =1.0
     celldm(3) =1.0
     nat=36
     ntyp=5
     nelec=286
     ecutwfc =35.0
     emaxpos = .45
     eopreg = 0.1
     eamp = 0.001
     edir = 3
     occupations='smearing'
     degauss=0.001
     smearing='gaussian'
  /
   &electrons
     diagonalization='david'
     electron_maxstep=500
     conv_thr =  1.0d-10
     mixing_beta=0.5
     startingwfc='random'


-- 
***********************************************
Heather D. Whitley, Quantum Simulations Group
Lawrence Livermore National Laboratory, L-415
Livermore, CA 94551
http://qsg.llnl.gov



More information about the users mailing list