"G(tau) must not change sign" in DMFT calculation

Total energy, geometry optimization, DFT+U, spin....

Moderator: bguster

Locked
pipidog
Posts: 19
Joined: Wed Apr 15, 2015 1:05 am

"G(tau) must not change sign" in DMFT calculation

Post by pipidog » Fri Nov 18, 2016 10:01 pm

Dear all,

I performed a DMFT calculation and use OmegaMaxEnt to obtain the corresponding
spectra, but I always get this error message:

"G(tau) must not change sign."

while using OmegaMaxEnt. What's the reason for this error? Is it because my DMFT
calculation doesn't converge well? How can I solve this issue?

Thanks!

amadon
Posts: 46
Joined: Mon Aug 24, 2009 10:58 am

Re: "G(tau) must not change sign" in DMFT calculation

Post by amadon » Mon Nov 21, 2016 9:37 am

Dear,

It seems that your imaginary time Green's function changes sign which is not physically correct. Could you post a plot of your Green's function
in imaginary time (or the raw data) ?
Is your DMFT calculation correctly converged ?

Thanks
Bernard

PS: Could you precise your affiliation ?
Thanks
Bernard Amadon
CEA
France

pipidog
Posts: 19
Joined: Wed Apr 15, 2015 1:05 am

Re: "G(tau) must not change sign" in DMFT calculation

Post by pipidog » Wed Nov 23, 2016 1:12 am

Hello,

The DMFT calculation is hard to converge, so I only set a rather low criterion (tolvrs 1e-2) to perform
the calculation otherwise it could never finish. Do you think it is the main reason to make the sign
problem? (I set dmft_tollc=1e-5, but usually, it is not easy to achieve within a reasonable dmft_iter)

I took a look on your lecture http://ftp.abinit.org/ws11/Amadon.pdf and it looks possible to just make a
one shot DMFT calculation and a roughly picture can be obtained. If one-shot calculation is possible,
I suppose the convergence is not reached. If so, is convergence a critical issue that make the sign problem?

* Also a quick question:
Currently the dmft solver seems not support SOC case due to the spinor wavefunctions. If so,
do you think DMFT calculation reliable for 4f electrons? (I saw in your lecture, Ce2O3 works well
abinit DMFT calculation. As far as I know, Ce also has SOC. So do you think SOC effect important
in DMFT?)

Attached is a calculation for Eu metal where there is only one atom per unitcell, so should be a
rather simple case. Beacuse this is just a quick check, many parameters are not taken
to ensure high precision. Any suggestion to help me perform DMFT using abinit is appreciated !

Thanks!
Shu-Ting Pi
University of California, Davis

==================================================================
# << Project: Eu-LDA+DMFT >>

AUTOPARAL 1
ndtset 1 # how many running data set
jdtset 2
pawprtvol 3 # show more information

#Definition of the Structure ======================
chkprim 0 # check primitive cell, default: 1 (stop) / 0 (warn)
nsym 0 # number of symmetry, default: 0 (autocheck) / 1 (nosymm)
acell 3*4.018 angstrom # equal to 1 angstrom
angdeg 3*109.471 # angle betwee two basis vectors
ntypat 1 # Total types of atoms
znucl 63 # atom species (.files must in same order)
natom 1 # total atoms per unitcell
typat 1 # atom types
xred
0.0 0.0 0.0

# Spin-orbit =======================================
nspinor 1 # spinor
nspden 2 # spin density
nsppol 2 # spin polarizaton
spinat 0 0 0 # initial moments
pawspnorb 0 # turn on paw SOC
spnorbscl 1.0 # scaling of SOC

# Correlated Orbital ================================
usepawu 1 # Full localized doubling counting
dmatpuopt 1 # The density matrix: the simplest expression.
lpawu 3 # f-orb

#Convergence ==========================
ecut 14.7 # Maximal kinetic energy cut-off, in Hartree
pawecutdg 40 # ecut for paw double grid
occopt 3 # smearing of the energy levels
tsmear 1200 K
diemac 1.0e+6 # large value for metal
nband 40
# k-mesh ===========================================
kptopt 4 # kpt generation, if SOC, no TRS allowed
ngkpt 10 10 10 # k-mesh along each axis
nshiftk 1 # how many k-shift
shiftk # shift vector
0.5 0.5 0.5
istwfk *1

# DFT SCF ===========================================
nstep1 300 # Maximal number of SCF cycles
#nline1 10
#nnsclo1 10
upawu1 0.0 eV # U
jpawu1 0.0 eV # J
tolvrs1 1.0d-3

# DMFT ==============================================
usedmft2 1
nline2 20
nnsclo2 20

getwfk2 1
nstep2 100
tolvrs2 1.0d-2

upawu2 8.0 eV
jpawu2 1.0 eV

dmft_solv2 5 # CTQMC
dmftbandi2 6 # initial band
dmftbandf2 12 # final band

dmft_iter2 50
dmft_mxsf2 0.3
dmft_tollc 1e-5

dmft_dc2 1
dmft_nwlo2 100
dmft_nwli2 100000

dmftqmc_l2 50
dmftqmc_n2 1.d6
dmftqmc_therm2 10000 #10000
dmftctqmc_gmove2 0
dmftctqmc_order2 50 #50

amadon
Posts: 46
Joined: Mon Aug 24, 2009 10:58 am

Re: "G(tau) must not change sign" in DMFT calculation

Post by amadon » Wed Nov 23, 2016 9:47 am

Hello,

More generally, if you want more precise data, the first thing to do is to increase dmftqmc_n, which is rather low.
Could you please post your Green's function in imaginary time ?

Thanks
Bernard
Bernard Amadon
CEA
France

pipidog
Posts: 19
Joined: Wed Apr 15, 2015 1:05 am

Re: "G(tau) must not change sign" in DMFT calculation

Post by pipidog » Thu Nov 24, 2016 12:03 am

Hello Bernard,

I'm new to DMFT in abinit. The are many output files, which one are you talking about?
I got the following files with different names:

DS2_atom_01_Gtau_xx.dat
DS2_atom_01_Gw_xx.dat
DS2_atom__G0w_.dat
Gtau.dat
DS2Self-omega_iatom0001_isppol1
DS2Self-omega_iatom0001_isppol2

I can't find any document explain their meaning. Which file contains the GF in imaginary time?
(I guess, Gtau.dat? but which column? It's hard the understand the output of the dmft calculation).

Thanks!
--
Shu-Ting Pi
Postdoctoral Fellow
University of California, Davis

Locked