ABINIT Discussion Forums

Hello,
Since hybrid functionals can't be used for GS calculations yet, and I'm trying to use some geometry opmtimizations using MetaGGA's in abinit. I'm starting with the lattice constants of graphite to work out problems with my input files. When I do, I find that the program crashes during the initialization of the 2nd broyden step. This happens in both the sequential and parallel versions, and both the internal and externally linked libxc. Does anyone have a suggestion? Is there a variable for reading in the kden files? The Hessian does get initialized and output to _o_DS1_TIM1_KDEN, and is the same size as the corresponding density file. Are these just not ready for geometry optimization yet?
Thanks a bunch for any suggestions.
James

Here's a copy of my log file
"--- Iteration: ( 2/30) Internal Cycle: (1/1)
--------------------------------------------------------------------------------

Cartesian coordinates (xcart) [bohr]
0.00000000000000E+00 0.00000000000000E+00 0.00000000000000E+00
8.88178419700125E-16 2.69050809574840E+00 -3.52356707465703E-45
3.84206680863613E-16 6.65465491863178E-16 6.27457126628040E+00
2.33004836000581E+00 1.34525404787420E+00 6.27457126628040E+00
Reduced coordinates (xred)
0.00000000000000E+00 0.00000000000000E+00 0.00000000000000E+00
3.33333333333333E-01 6.66666666666667E-01 -2.80791081688907E-46
0.00000000000000E+00 0.00000000000000E+00 5.00000000000000E-01
6.66666666666666E-01 3.33333333333333E-01 5.00000000000000E-01
Scale of Primitive Cell (acell) [bohr]
4.66009672001163E+00 4.66009672001163E+00 1.25491425325608E+01
Real space primitive translations (rprimd) [bohr]
4.66009672001163E+00 0.00000000000000E+00 0.00000000000000E+00
-2.33004836000581E+00 4.03576214362261E+00 1.80346993469317E-49
7.68413361727225E-16 1.33093098372636E-15 1.25491425325608E+01
Unitary Cell Volume (ucvol) [Bohr^3]= 2.36012249773366E+02
Angles (23,13,12)= [degrees]
9.00000000000000E+01 9.00000000000000E+01 1.20000000000000E+02
Lengths [Bohr]
4.66009672001163E+00 4.66009672001163E+00 1.25491425325608E+01

---SELF-CONSISTENT-FIELD CONVERGENCE--------------------------------------------

getcut: wavevector= 0.0000 0.0000 0.0000 ngfft= 20 20 60
ecut(hartree)= 20.000 => boxcut(ratio)= 2.13184

ewald : nr and ng are 3 and 33
forrtl: severe (174): SIGSEGV, segmentation fault occurred
"

and a copy of my input file:

#Graphite with FHI GGA Psuedopotential and M06-L Lattice constant optimization
chkprim 0

acell 2.456 2.456 6.695 angstrom
angdeg 90 90 120
nstep 50
optcell 3
ionmov 2
dilatmx 1.07
ecutsm 0.5

ntypat 1
znucl 6 #9
nshiftk 1
shiftk 0.0 0.0 0.0
natom 4
typat 4*1
nband 30

spinat 0 0 1
0 0 -1

0 0 1
0 0 -1
nsppol 2
nspinor 1
nspden 2

occopt 3
tsmear .001
xred 0 0 0
1/3 2/3 0
0 0 .5
2/3 1/3 0.5

ndtset 6
ecut: 20 ecut+ 5

ngkpt 14 14 1
ixc -203233 #M06-L MetaGGA

usekden 1
prtkden 1
timopt 2
prtden 1
#prtwf 1
toldfe 5d-10
#toldff 1d-6
#tolmxf 1d-5
ntime 30

Are you using a mGGA pseudopotential? I guess more specifically, what pseudopotentials are you using?

Hi Prof. Zwanzig,
It took a long time to get back to this, but I wanted to not just give you a silly answer. I've tried using a standard GGA pseudopotential (HGH PBE) and generating my own with APE and the M06-L functional with the output in the abinit6 format. In the run that I had posted, I had used toldfe as the stopping criteria, because all of the ion positions were fixed by symmetry. However, when I broke this symmetry and used tolmxf and toldff, I found that abinit was not computing the forces

cartesian forces (hartree/bohr) at end:
1 -0.00000000000000 -0.00000000000000 -0.00000000000000
2 -0.00000000000000 -0.00000000000000 -0.00000000000000
3 -0.00000000000000 -0.00000000000000 -0.00000000000000
4 -0.00000000000000 -0.00000000000000 -0.00000000000000
frms,max,avg= 0.0000000E+00 0.0000000E+00 0.000E+00 0.000E+00 0.000E+00 h/b

I know that there can be an issue with meta-GGA functional and analytic gradients, is that why abinit doesn't compute the forces?
James

which version of abinit are you using? the features around mGGA are in a state of active development at the moment.

Joe

I'm using the release version abinit 7.2.1

I just tried a similar test using the TB09 mGGA as done in the abinit test suite (libxc test 09) and forces and stresses seem to work there. Given that there seem to be about 10! different functionals around in the literature and in libxc, I'm not surprised that all features of all possibilities have not been coded yet.