Errors & Solutions and Tips

[nguyen-group]

Errors & Solutions

• ERROR: Too many bands are not converged from scf or nscf calculations

Solution: increase ecutwfc or decrease conv_thr, or try to do both of them.

• ERROR: S matrix not positive definite

Solution: Run with 'cg' diagonalization might work.

• ERROR: No convergence has been achieved from ph.x

Solution: Reduce alpha_mix(1) in ph.in to 0.3 or 0.1, or the readers can reduce alpha_mix gradually as
alpha_mix(10) = 0.3
alpha_mix (15) = 0.1
That means that you use the default value for alpha_mix (= 0.7) in the first 10 iterations, then you reduce it to 0.3. And after the 15th iteration, you reduce it to 0.1.

• Run linear response calculations (phonons, calculation of Hubbard parameters, etc.) for magnetic insulators with occupations = 'fixed'

Solution: perform a second SCF calculation with fixed occupations by setting total_magnetization and starting from the previous converged SCF calculation (see the procedure described in q-e/HP/examples/example02/README)

• ERROR: Energy error = 0.0E+00 Ry Gradient error = xxx Ry/Bohr

Solution: Increasing the etot_conv_thr and forc_conv_thr values

Tips

• Control the band gap values on the HSE (Heyd-Scuseria-Ernzerhof) hybrid functional calculation:

In HSE, the screening parameter μ is used to determine the range separation between the short-range and long-range parts of the exchange interaction. The default screening parameter is 0.106 in units of 1/bohr. This value can be changed in "screening_parameter = 0.106" in &SYSTEM to modify the band gap value. Usually, a smaller screening_parameter will give a higher energy band gap.

Error installing:

• ERROR: F90-F-0949-#error -- No fft_scalar backend selected! (fft_scalar.f90: 37)

Solution: Adding to flags: DFLAGS="-D__FFTW" and FFLAGS="-O3 -D__FFTW -cpp" when ./configure (note: it works in my case to install QE with AOCL_FFLW of AMD)

• ERROR: F90-F-0004-Corrupt or Old Module file /opt/q-e-qe-7.4.1//MBD/mbd.mod (mbdlib.f90: 24)
  This happens when the .mod file mbd.mod is incompatible with the current compiler or was generated with different flags or another compiler entirely (e.g., gfortran, ifort, etc.).

Solution: Check make.inc to ensure that F90 = gfortran is same with current compiler. In my case, when I used F90=mpif90 for ./configure, but it was F90 = gfortran in make.inc. Thus, I need to fix it manually.