ABINIT Discussion Forums

Dear all,

I'm a new user to ABINIT, having worked mostly with CASTEP and Elk in the past. I'm observing a kind of convergence behaviour which I've never seen before in other codes and I was wondering if anyone knows what it is likely to be. It consists of an oscillatory reset from what seems to be quite good SCF progress back to something that isn't quite as good - see attached figure for an example using a n-confused porphin molecule. I've plotted the abs(DFE) value as a function of step. My input file is as below. I'm running on linux x86_64 compiled using gfortran 4.6 etc with respect to fftw3, openblas (the successor to goto2) and mvapich2. This run was done on 12 cores. The behaviour I'm interested in is most obvious from steps 100 and onwards. The potential residual displays more or less the same behaviour except the residual converges to a value of about 3e-6 in between the spikes rather than getting lower and lower.

# SCF parameters
ecut 20.0
ecutsm 0.5
pawecutdg 40.0
#toldff 5.0e-6
tolvrs 2.0d-7 # Tighten up convergence slightly
nstep 200
istwfk 2 # Gamma-point only means real wfs.
occopt 7 # Gaussian smearing
tsmear 0.005
diemac 2.0
timopt 2

# Kpoints

kptopt 1
ngkpt 1 1 1
nshiftk 1
shiftk
0.0 0.0 0.0 # Need to use the true gamma point.
nband 72 # Done to fix an input bug.

# Parallelization

paral_kgb 1
npkpt 1
npband 12
bandpp 2

# Basic Geometry

acell 16.0 16.0 10.0 angstrom
natom 38
ntypat 3
znucl 6 7 1
typat 20*1 4*2 14*3
xangst
9.9197288121E+00 5.0922015244E+00 4.9515052354E+00
5.0598102696E+00 5.1123376601E+00 4.9516571439E+00
5.0497619836E+00 9.9395334845E+00 5.1199937764E+00
9.9828780414E+00 9.9234081849E+00 5.1025428583E+00
8.5735780039E+00 4.7412160273E+00 4.9136963817E+00
8.1576513981E+00 3.3600733779E+00 4.8371325915E+00
6.8068459747E+00 3.3664758704E+00 4.8415264453E+00
6.4041470441E+00 4.7510923257E+00 4.9180696978E+00
4.5467077594E+00 6.3961882654E+00 5.0268591457E+00
3.1952972564E+00 6.8387310843E+00 5.0590869362E+00
3.2085823947E+00 8.2068661587E+00 5.1088712171E+00
4.5652960534E+00 8.6412958540E+00 5.1166060972E+00
6.3976245027E+00 1.0283592157E+01 5.1367852124E+00
6.9537198692E+00 1.1593661206E+01 4.8266022832E+00
7.5190912360E+00 9.4740860601E+00 5.3448394923E+00
8.6433946837E+00 1.0283634215E+01 5.1214235835E+00
1.0451793508E+01 8.6169154346E+00 5.1065516879E+00
1.1801982796E+01 8.1694328792E+00 5.1336911671E+00
1.1802649930E+01 6.7992916893E+00 5.0930943795E+00
1.0448349859E+01 6.3712749185E+00 5.0314181118E+00
9.6759155975E+00 7.4905208944E+00 5.0489413365E+00
8.2537387438E+00 1.1601506449E+01 4.8169030398E+00
5.3310547508E+00 7.5063301037E+00 5.0731677914E+00
7.4919271091E+00 5.5778044859E+00 4.9591063645E+00
1.0721903117E+01 1.0729394663E+01 5.0182508921E+00
1.0637816998E+01 4.2650039800E+00 4.9174266509E+00
4.3334324138E+00 4.2925270315E+00 4.9157296683E+00
4.3013208559E+00 1.0738525486E+01 5.0464186480E+00
7.5254624477E+00 8.4804137192E+00 5.7949325471E+00
6.3680282414E+00 1.2490325838E+01 4.5863116249E+00
1.2661857867E+01 8.8388191090E+00 5.1901724506E+00
1.2664073408E+01 6.1295116087E+00 5.0980410836E+00
8.8363973007E+00 2.5057345156E+00 4.7921719523E+00
6.1212105852E+00 2.5175960558E+00 4.8001455654E+00
2.3500269446E+00 8.8798026070E+00 5.1448038176E+00
2.3287287527E+00 6.1762333548E+00 5.0389177280E+00
6.3539886859E+00 7.4114721866E+00 4.9816355305E+00
8.6542848049E+00 7.4071195335E+00 4.9349598644E+00

Hello,

You forgot to attach the figure in your post. If you have convergence problems, you might try other values of iscf, for example a simple mixing (iscf = 2) with a low mixing parameter (diemix 0.3 for example). You can of course start from your wavefunction or density that you have already reached (see irdden/irdwfk or getden/getwfk).

David

Yes, I am indeed silly. Photo is attached this time...

On a related note, is there a way to force abinit to stop with an error if convergence is not reached over the SCF portion of a structural optimization? Bad forces from a non-converged SCF output can really throw a structural optimization way off. I've had a few molecules explode over the last few weeks...