forum@abinit.org
Subject: The ABINIT Users Mailing List ( CLOSED )
List archive
- From: "Scott P. Beckman" <sbeckman@ices.utexas.edu>
- To: forum@abinit.org
- Subject: stability of SCF loop with wfoptalg=1 on SP2
- Date: Fri, 19 Aug 2005 09:52:03 -0500
Hello,
I'm trying to parallize over bands, but the SCF loop seems to not converge, or it blows up (if I pick nbdblock to be large).
I've read the archives and I've looked at the Test_v3/t41.in example file. The only advice regarding the stability is to keep nbdblock small. Even when I set nbdblock 1 which is unparalized (right?) the convergence is a problem.
Are there other parameters that I can change to improve convergence? Do I need to tweak the parameters of the CG minimization? In the example Test_v3/t41.in the SCF loop is run using the default (CG minimization), is it better to switch to Anderson mixing (iscf=3)?
When I run the cell for nbdblock of 1, 2, and 4 it does seem to parallize because for a given wall-clock-time 4 times as many loops are calculated for nbdblock=4 than for nbdblock=1, but the ETOT ceases to reduce after the first 3 or 4 iterations.
Below I've attached the details of my calculation as well as a summary of the SCF loop run for nbdblock=2
I'd appreciate any advice.
Thank you,
Scott
=============
In my calculation I have 1 kpoint at gamma, no spin, and 448 states.
I'm running abinit 4.5.3 on the NERSC SP2 computer called, "seaborg"
Here is my input file (truncated):
#Definition of the unit cell
acell 3*30.78
rprim 0.0 0.0 1.0
1.0 0.0 0.0
0.0 1.0 0.0
#Definition of the atom types
ntypat 1 # There is only one type of atom
znucl 14 # It is Silicon
#number of states
nband 448
#parallelization parameters
localrdwf 0
wfoptalg 1
nbdblock 2
#mffmem 0
#mkmem 0
#Definition of the atoms
natom 216 # There are eight atoms
typat 216*1
# They all of type 1, that is, Silicon.
xcart
.000000 .000000 .000000
.000000 .000000 10.254159
# <214 other atoms which I've cut from this message...>
#Definition of the planewave basis set
ecut 16.0 Ry # Maximal kinetic energy cut-off, in Hartree
#Definition of the k-point grid
kptopt 0 # Pick one point at Gamma
#Definition of the SCF procedure
nstep 100 # Maximal number of SCF cycles
toldfe 2.0d-4 Ry # SCF tolerance
diemac 12.0 #For Silicon
==========
Here are the results from when I try using ndblock 2
-bash-2.05b$ grep 'ETOT' s*.out
s00509.0.out: ETOT 1 -751.47821637601 -7.515E+02 9.686E-03 4.012E+02 7.504E-03 7.504E-03
s00509.0.out: ETOT 2 -747.33285059714 4.145E+00 3.070E-03 7.908E+02 1.048E-02 7.148E-03
s00509.0.out: ETOT 3 -958.38383973537 -2.111E+02 5.495E-02 1.511E+02 6.166E-03 1.878E-03
s00509.0.out: ETOT 4 -956.70213692513 1.682E+00 3.084E-02 8.664E+02 3.306E-03 3.396E-03
s00509.0.out: ETOT 5 -956.62261883621 7.952E-02 2.473E-02 4.942E+02 4.401E-03 3.986E-03
s00509.0.out: ETOT 6 -955.91889202857 7.037E-01 4.237E-02 2.621E+02 6.614E-03 8.539E-03
s00509.0.out: ETOT 7 -957.68210559965 -1.763E+00 7.069E-02 2.466E+02 7.053E-03 4.627E-03
s00509.0.out: ETOT 8 -956.68684912222 9.953E-01 1.178E-01 6.697E+02 6.181E-03 6.405E-03
s00509.0.out: ETOT 9 -956.60608351456 8.077E-02 2.344E-02 1.446E+02 5.416E-03 4.662E-03
s00509.0.out: ETOT 10 -957.08065326697 -4.746E-01 2.926E-01 2.483E+01 5.151E-03 3.356E-03
s00509.0.out: ETOT 11 -956.36743550574 7.132E-01 2.268E-02 1.044E+02 3.775E-03 4.755E-03
s00509.0.out: ETOT 12 -957.56864369269 -1.201E+00 5.286E-02 4.373E+01 5.937E-03 4.396E-03
s00509.0.out: ETOT 13 -956.09735922438 1.471E+00 1.271E-02 3.193E+01 5.647E-03 4.728E-03
s00509.0.out: ETOT 14 -956.37022803816 -2.729E-01 2.906E-02 3.309E+01 5.269E-03 4.129E-03
s00509.0.out: ETOT 15 -956.67570978833 -3.055E-01 2.764E-02 3.440E+01 5.590E-03 5.024E-03
s00509.0.out: ETOT 16 -956.48829691245 1.874E-01 2.715E-02 5.744E+01 1.232E-02 7.456E-03
s00509.0.out: ETOT 17 -957.00396225354 -5.157E-01 7.132E-02 2.999E+01 5.378E-03 4.806E-03
s00509.0.out: ETOT 18 -956.76329776060 2.407E-01 2.874E-02 2.421E+01 6.049E-03 5.195E-03
s00509.0.out: ETOT 19 -955.60989295271 1.153E+00 2.172E-02 6.916E+01 4.324E-03 5.136E-03
s00509.0.out: ETOT 20 -841.59537695346 1.140E+02 1.282E-01 4.586E+04 1.667E-01 1.616E-01
s00509.0.out: ETOT 21 -949.31863557768 -1.077E+02 1.369E-01 4.715E+02 1.672E-01 1.295E-02
s00509.0.out: ETOT 22 -957.41192412422 -8.093E+00 1.197E-02 5.254E+01 1.418E-02 4.192E-03
s00509.0.out: ETOT 23 -838.38884562380 1.190E+02 2.301E-01 2.640E+04 6.352E-02 6.183E-02
s00509.0.out: ETOT 24 -944.16123848793 -1.058E+02 3.139E-01 3.868E+02 5.708E-02 1.139E-02
s00509.0.out: ETOT 25 -955.55830652936 -1.140E+01 1.534E-02 1.131E+01 1.333E-02 4.026E-03
- stability of SCF loop with wfoptalg=1 on SP2, Scott P. Beckman, 08/19/2005
Archive powered by MHonArc 2.6.16.