# Should I use a lower threshold for calculations involving SOC?

I'm studying a compound $$\ce{VSe_2}$$ using the DFT package Quantum ESPRESSO (QE). The following is the input file.

&CONTROL
calculation = 'scf'
outdir = './outdir'
verbosity = 'high'
tprnfor = .true.
tstress = .true.
pseudo_dir = '.'
/
&SYSTEM
ibrav = 0
A =    3.35381
nat = 3
ntyp = 2
ecutwfc = 55
ecutrho = 650
starting_magnetization(1) = 1
starting_magnetization(2) = 0
lspinorb = .true.
noncolin = .true.
occupations = 'smearing'
smearing = 'mv'
degauss = 0.005d0
/
&ELECTRONS
conv_thr = 1e-8
mixing_beta = 0.7d0
/
CELL_PARAMETERS {alat}
1.000000000000000   0.000000000000000   0.000000000000000
-0.500000000000000   0.866025403784439   0.000000000000000
0.000000000000000   0.000000000000000   2.087308375869810
ATOMIC_SPECIES
V   50.94150  V.rel-pbe-spnl-kjpaw_psl.1.0.0.UPF
Se   78.96000  Se.rel-pbe-dn-kjpaw_psl.1.0.0.UPF
ATOMIC_POSITIONS {crystal}
Se   0.666666666666667   0.333333333333333   0.774935000000000
Se   0.333333333333333   0.666666666666667   0.225065000000000
V   0.000000000000000   0.000000000000000   0.000000000000000
K_POINTS {automatic}
6 6 6 0 0 0


When I'm using a threshold of $$10^{-8}$$, the SCF loop doesn't converge after 100 iterations. Is it okay to use a lower threshold of $$10^{-4}$$ so as to aid convergence?

SCF output:

 estimated scf accuracy    <       0.00015001 Ry
estimated scf accuracy    <       0.00017012 Ry
estimated scf accuracy    <       0.00023876 Ry
estimated scf accuracy    <       0.00015530 Ry
estimated scf accuracy    <       0.00010229 Ry
estimated scf accuracy    <       0.00030521 Ry
estimated scf accuracy    <       0.00038586 Ry
estimated scf accuracy    <       0.00035244 Ry
estimated scf accuracy    <       0.00018834 Ry
estimated scf accuracy    <       0.00026472 Ry
estimated scf accuracy    <       0.00026759 Ry
estimated scf accuracy    <       0.00035791 Ry
estimated scf accuracy    <       0.00014860 Ry
estimated scf accuracy    <       0.00023776 Ry
estimated scf accuracy    <       0.00031980 Ry
estimated scf accuracy    <       0.00023777 Ry
estimated scf accuracy    <       0.00029549 Ry
estimated scf accuracy    <       0.00031407 Ry
estimated scf accuracy    <       0.00009668 Ry
estimated scf accuracy    <       0.00019794 Ry

• Why don't you taking $10^{-6}$? – Jack Jan 23 at 14:09
• @Jack The scf accuracy values presented belong to the last part of the 100 iteration loop. And you could see that the lowest it goes is 0.00009668. Thats why I thought of taking $10^{-4}$ – Atom Jan 23 at 14:42
• You can set the maximum iteration steps of the SCF loop to 300 with a threshold of $10^{-6}$. If you can't obtain the converged results, check your input files is OK? If so, tune mixing paramters. – Jack Jan 30 at 1:18

I wouldn't recommend that you go above $$10^{-8}$$, what you have in your current calculation. With SCF calculations, you definitely need a lower convergence threshold compared to say relaxing procedures. I suspect your problem is rooted elsewhere. Try:

1. Lowering 'mixing-beta'
2. Removing 'verbosity' as it just prints energy levels and takes up time, which you can after you've made you can achieve convergence.

Also, did you do an SCF calculation with SOC turned off? Did you achieve convergence then? The usual procedure is to make sure that your non-relativistic calculation can converge properly first. This means tuning 'ecutwfc' and 'K-POINTS' so that proper convergence has been achieved. Once this is done, turn on SOC and see if you still encounter the same problem.