Phonon frequencies change with nband for insulating

Running Abinit Response calculations
1

Dear Abinit community,
I am calculating phonons for cubic BiCoO3 with Paw-GGA. When I increase nband the phonons frequencies change drastically. DOS calculation showed an insulating behavior. I need help to figure out what is happening. here are the three first modes at Gamma in cm-1:

nband 26 9.548265E+01 9.548265E+01 9.548265E+01
nband 28 5.988331E+01 5.988331E+01 5.988331E+01
nband 30 -1.873508E+02 -1.873508E+02 -1.873508E+02
nband 32 -1.879885E+02 -1.879885E+02 -1.879885E+02
nband 40 -1.888686E+02 -1.888686E+02 -1.888686E+02
nband 50 -1.884627E+02 -1.884627E+02 -1.884627E+02

here is my input:
pawxcdev 0

rfphon 1
rfatpol 1 5
irdwfk 1
tolvrs 1.0E-10
rfdir 1 1 1
nqpt 1
qpt 0.0 0.0 0.0
kptopt 3

####################################################################
chkprim 0
chksymbreak 0

optforces 1
ngkpt 6 6 6

nshiftk 1
shiftk 0 0 0

prtden 0
nshiftk 1
shiftk 0 0 0

prtden 0

#Definition of the atom types

acell 7.1907671831E+00 7.1907671831E+00 7.1907671831E+00 Bohr

pp_dirpath “/home/hdjani/codes/abinit-9.6.2/build/tests/WORK/BCO/dfpt/”
pseudos “Bi.GGA_PBE-JTH.xml, Co.GGA_PBE-JTH.xml, O.GGA_PBE-JTH.xml”

#Definition of the atoms
natom 5 # There are two atoms
ntypat 3 # There is only one type of atom
znucl 83 27 8 # The keyword “znucl” refers to the atomic number of the

typat 1 2 3 3 3 # They both are of type 1, that is, Silicon.
xred

0.00000000000000E+00 0.00000000000000E+00 0
5.00000000000000E-01 5.00000000000000E-01 0.5
5.00000000000000E-01 0.00000000000000E+00 0.5
5.00000000000000E-01 5.00000000000000E-01 0.0
0.00000000000000E+00 5.00000000000000E-01 0.5

spinat 0 0 0
0 0 0
0 0 0
0 0 0
0 0 0

occopt 7
nnsclo 4
npulayit 18
nline 4
nbdbuf 4
nsppol 2
nspinor 1
nspden 2

#Definition of the planewave basis set
ecut 20 # Maximal kinetic energy cut-off, in Hartree
ecutsm 0.5
pawecutdg 50

#DFT+U
usepawu 1
lpawu -1 2 -1
upawu 0 6 0 ev
jpawu 0 0.2 0 ev

nband 26
abinit.abo (158.5 KB)

nstep 100

Many Thanks,
Hania Djani

2

Hi Hania!

Indeed this looks strange. Do you expect the cubic phase to be unstable? Probably. No literature values to compare to? U and J values chosen randomly?

A few things to try:

  1. you definitely have too few bands. For simple insulators you can use occopt 1 and no smearing, with exactly nband=nelect, but for metallic occupation you need at least 10% empty bands (or more, has to be checked as you are doing here): the last nbdbuf = 4 wfk bands will be garbage, and in your abo file this includes some filled bands.
  2. your smearing is 0.01. Probably ok, but if the gap is small the phonons may be very sensitive. Test 0.001 as well (possibly with more nkpt)
  3. I would say your values for band >=32 are ok and converged, and the jump from the low nband case is normal. Do they look wrong to you? For small nband you don’t have any 0 freq acoustic modes. Do they appear for larger nband?
  4. Why set chkprim and chksymbreak to 0? Your cell is very symmetric and this prevents the usage of some symmetries to clean up the density etc…
  5. If the system wants to be magnetic, it might be unstable or have a hard time converging. Why set spinat to 0 explicitly? This will keep your system in an artificial unstable equilibrium
  6. phonons are more delicate than the GS, so the ecut may not be converged. Need to recheck it.
  7. your fftalg = 112 is slow. You should compile with fftw3 or mkl/DFTI

have fun