Different Lattice Constants and Atomic Positions between Materials Project Database and VC-Relax Calculation

Materials Project
1

Dear colleagues,

I found a monoatomic Cr crystal structure that is not stable in bulk but it is experimentally observed in thin film. The structure ID is mp-17 (https://next-gen.materialsproject.org/materials/mp-17). This structure has a simple cubic Bravais lattice with 8 atoms in a unit cell. I have tried to do a vc-relax calculation in Quantum ESPRESSO to check the structure stability by considering the spin-orbit interaction. Here is my input file:

&CONTROL
calculation = ‘vc-relax’
pseudo_dir = ‘/home/anin/pseudo’
outdir = ‘…/…/tmp’
prefix = ‘cr_sc’
etot_conv_thr = 1.0D-6
forc_conv_thr = 1.0D-6
restart_mode = ‘restart’
/
&SYSTEM
ibrav = 1
a = 4.51
nat = 8
ntyp = 1
occupations = ‘smearing’
smearing = ‘mv’
degauss = 0.02
ecutwfc = 100.0
noncolin = .true.
lspinorb = .true.
/
&ELECTRONS
mixing_beta = 0.7
conv_thr = 1.0D-6
/
&IONS
ion_dynamics = ‘bfgs’
/
&CELL
cell_dynamics = ‘bfgs’
press_conv_thr = 0.3
cell_dofree = ‘all’
cell_factor = 3.0
/
ATOMIC_SPECIES
Cr 51.9961 Cr.rel-pz-spn-kjpaw_psl.1.0.0.UPF
ATOMIC_POSITIONS (crystal)
Cr 0.000000 0.000000 0.000000
Cr 0.500000 0.500000 0.500000
Cr 0.250000 0.000000 0.500000
Cr 0.750000 0.000000 0.500000
Cr 0.250000 0.500000 0.000000
Cr 0.750000 0.500000 0.000000
Cr 0.000000 0.250000 0.500000
Cr 0.500000 0.250000 0.000000
K_POINTS (automatic)
6 6 6 0 0 0

In this input file, I used the lattice constant from the Materials Project database (a = 4.51 A). However, the resulting structure from the vc-relax calculation is very different compared to the Materials Project database. The structure becomes tetragonal with a = 5.58 A and b = c = 3.95 A. The resulting atomic positions are also differ:
ATOMIC_POSITIONS (crystal)
Cr 0.0000000000 -0.0000337418 0.0000000000
Cr 0.5000000000 0.5000337418 0.5000000000
Cr 0.2499869382 -0.0000045017 0.5000000000
Cr 0.7500130618 -0.0000045017 0.5000000000
Cr 0.2500130618 0.5000045017 0.0000000000
Cr 0.7499869382 0.5000045017 0.0000000000
Cr 0.0000000000 0.4999253946 0.5000000000
Cr 0.5000000000 0.0000746054 0.0000000000
End final coordinates

Why does my vc-relax calculation produce a different result compared to the Materials Project database? Which one is more proper to be used for DFT calculation?

2

Hi @Nabail_Azhiim, it looks like you’re using a different exchange-correlation functional (Perdew-Zunger LDA?) based on the pseudopotential name than the Materials Project uses (either PBE GGA or r2SCAN meta-GGA). Additionally, we don’t use spin-orbit corrections (SOC). Those alone explain the differences in relaxed lattice parameters.

For a geometry optimization of a Cr cell, I don’t see a strong need for SOC. For monatomic metallic systems, you would typically only use SOC for heavier elements (f-block) or magnetic anisotropy energies.

Hope that helps!

3

Hello, @Aaron_Kaplan. Yes, you are correct. The pseudopotential I used in this calculation is using Perdew-Zunger (LDA) exchange-correlation. Here I consider SOC because I want to calculate the Spin Hall Conductivity (SHC) and Orbital Hall Conductivity (OHC) of the structure. Thus, is it okay if I don’t consider the SOC (hence using non-relativistic pseudopotential) during the vc-relax calculation but then consider the SOC on the scf and nscf calculation?

4

I don’t think you should use SOC for the relaxation, it’s likely that Hellmann Feynman forces aren’t implemented for SOC (see this 2021 Quantum Espresso thread). You might also reach out to the QE developers too to double check

Either way, relaxing without SOC and then using SOC for the N/SCF calculations is physically well-motivated. The ground state structure of Cr is not dominated by relativistic effects, so that shouldn’t affect the geometry if coded correctly

5

I’m sorry for my late response. I have tried to do a vc-relax calculation using a non-relativitistic pseudopotential from the Materials Cloud. However, the result is still unmatched with the Materials Project database. May I know which pseudopotential is used to generate lattice constant as in the Materials Project database (e.g. for structure mp-17 which I asked above)?

6

Since the Materials Project uses VASP for calculations, the pseudpotentials are VASP-specific - VASP unfortunately doesn’t support any pseudopotentials from Materials Cloud or Pseudo Dojo like QE does

The majority of our calculation parameters can be found in our documentation and in pymatgen

7

I see. Thank a lot for the information.