Si SCF

Goal

Run one SCF on the simplest system there is (diamond-structure silicon, two atoms) and read every block of the output. This input is the skeleton for every later example.

New cards and variables

Input file

Download si.scf.in

! E01: the simplest possible SCF, on diamond-structure silicon.
! Two atoms, the three mandatory namelists, the three mandatory cards.

&CONTROL
  calculation  = 'scf'        ! one self-consistent run, no ionic motion
  prefix       = 'si'         ! basename under outdir; follow-up runs must reuse it
  outdir       = './tmp/'     ! scratch: charge density + wavefunctions land here
  pseudo_dir   = './pseudo/'  ! where the .UPF pseudopotential files live
  verbosity    = 'high'       ! print symmetry ops, k-point list (always 'high' while learning)
  tprnfor      = .true.       ! print forces (zero here by symmetry, but learn to check)
  tstress      = .true.       ! print the stress tensor (it reports on the lattice constant)
/
&SYSTEM
  ibrav        = 2            ! fcc Bravais lattice; primitive vectors follow QE's convention
  celldm(1)    = 10.26        ! cubic lattice constant a, in BOHR (= 5.43 Angstrom, experimental)
  nat          = 2            ! atoms in the cell
  ntyp         = 1            ! atomic species
  ecutwfc      = 30           ! wavefunction cutoff in Ry (converge it yourself in E3)
  ecutrho      = 240          ! density cutoff: 8 x ecutwfc, the PAW/US rule (4x default is for NC)
  occupations  = 'fixed'      ! integer occupations: a semiconductor with a gap
/
&ELECTRONS
  conv_thr     = 1.0d-8       ! SCF threshold in Ry (1e-6 default is too loose once forces matter)
  mixing_beta  = 0.7          ! aggressive mixing; fine for a simple insulator
/

! label | mass in amu (physical only in MD and phonons) | UPF file
ATOMIC_SPECIES
  Si  28.0855  Si.pbe-n-kjpaw_psl.1.0.0.UPF

! coordinates in units of celldm(1) ('alat').
! diamond = fcc lattice + two-atom basis at (0,0,0) and (1/4,1/4,1/4)
ATOMIC_POSITIONS (alat)
  Si  0.00  0.00  0.00
  Si  0.25  0.25  0.25

! Monkhorst-Pack grid nk1 nk2 nk3, then three 0/1 shifts (unshifted here).
! symmetry reduces 8x8x8 = 512 points to 29 irreducible ones.
K_POINTS (automatic)
  8 8 8  0 0 0

Run

mkdir -p pseudo tmp        # put the Si PAW UPF in pseudo/ (Chapter 01)
pw.x -in si.scf.in > si.scf.out                     # serial
mpirun -np 6 pw.x -nk 6 -in si.scf.in > si.scf.out  # the parallel setup used for the measurements

What to check in the output

A few reading notes:

• Only the total-energy line marked ! is the converged value. The absolute number differs wildly from ultrasoft results because this is PAW, and absolute total energies are not comparable anyway.
• With occupations='fixed' and the default nbnd (occupied bands only), the output shows highest occupied level but no gap estimate (lowest unoccupied is missing). To see the gap, raise nbnd (Exercise 3).
• P = +20 kbar means this lattice constant (5.43 Å, the experimental value) is compressed relative to the PBE equilibrium. The PBE equilibrium constant is found in E6.

Exercises

1. Switch verbosity to 'low', watch how much output disappears, and explain why 'high' is the right choice while learning.
2. Change occupations to 'smearing'. How does the energy move, and why is smearing wrong for a semiconductor?
3. Add nbnd = 8 so that highest occupied, lowest unoccupied level appears, and read off the gap estimate.
4. Delete ecutrho (falling back to the 4x default) and see what happens.

Related chapters

02 Input file structure · 03 Units and coordinates · 08 SCF and NSCF