4
$\begingroup$

Hi everyone I'm here again to ask you if you know what I can modify to improve my phonon calculation of rutile

I think my issues are rooted in not-so-good cell optimizations, I optimize my starting geometry with this input (after a scf calculation on pretty much the same input). I've already tested the convergence of the K-grid and the ecutoff.

&CONTROL
    calculation = "scf"
    max_seconds =  8.64000e+04
    pseudo_dir  = "./UPF/pseudo_lda_ncpp"
    prefix = "80"
    verbosity = "high"
/

&SYSTEM
    a                         =  4.59420e+00
    c                         =  2.95940e+00
    degauss                   =  1.00000e-02
    ecutrho                   =  5.60000e+02
    ecutwfc                   =  8.00000e+01
    ibrav                     = 6
    nat                       = 2
    ntyp                      = 2
    occupations               = "smearing"
    smearing                  = "cold"
    space_group               = 136

/

&ELECTRONS
    conv_thr         =  1.00000e-06
    electron_maxstep = 200
    mixing_beta      =  4.00000e-01
    startingpot      = "atomic"
    startingwfc      = "atomic+random"
/

&IONS
ion_dynamics="bfgs"
/

&CELL
cell_dynamics = 'bfgs'
press = 0.0
/

K_POINTS {automatic}
 9 9 12 0 0 0

ATOMIC_SPECIES
Ti   47.86700  Ti.upf
O    15.99940  O.upf

ATOMIC_POSITIONS {crystal_sg}
Ti 0 0 0
O 0.3045 0.3045 0

and then, I take the results of this optimization (new atomic position and lattice parameter) and I apply them in a scf-calculation without the smearing (since I'm interested in obtaining raman spectra), then I simply run the ph.x code with this input:

&Inputph
  outdir    = './'
  prefix    = '80'
  fildyn    = 'rutile.dmat'
  lraman    = .true.
  verbosity = 'high'
  tr2_ph    =  1.0d-16
/
0.0 0.0 0.0

I'm new to Quantum Espresso world, and even if I have read (also in this forum) how to optimize the calculation, improving electronic convergence or geometrical optimization, I'm not really able to traslate this qualitatively advice into practical operation.

Hope you'll tolerate my trivial question, but I'm really getting lost inside this subject!

Thank you all in advance!

$\endgroup$

1 Answer 1

2
$\begingroup$

Generally a castep user here, but I don't think this is especially a code-dependent problem!

Firstly, what do you mean by "not so good"? Judging by the title (negative phonon) I presume you are seeing negative frequency (imaginary) phonon modes? As you say, this is usually a sign that you are not at a minimum energy state with your cell structure, and the cure is better geometry optimisation!

In terms of convergence criteria, are you converging forces directly, or just the energy? As a rule of thumb, the forces convergence is roughly the square root of the energy convergence (i.e. if energy is converged to 10^-6, forces are converged to ~10^-3). As you want to get a particularly high-quality geom opt in order to calculate the phonons accurately, you may need to converge to higher tolerances than you are used to.

Additionally, when performing your geometry optimisation, your energy is only converged to conv_thr: which looks like you are using 1e^-6. This means that you will have a degree of noise in your forces and stresses that come from your SCF convergence. This both makes it impossible to optimise below this noise floor (the BFGS algorithm ends up trying to relax numerical noise away, which does not actually relax your system!), and any supposed convergence is probably just lucky.

While it may seem expensive, a tighter convergence (e.g. conv_thr: 1e-10) may help you -- the noise on the forces will have dropped by roughly 2 orders of magnitude, so you should be able to get a higher quality geom_opt that way. Then the phonons should take care of themselves.

Hope that helps,

Rob

$\endgroup$
2
  • $\begingroup$ Thank you for your valuable advices! I'm currently running new calculations trying to fix the immaginary frequencies. Thank you again! $\endgroup$ Commented May 29 at 9:18
  • $\begingroup$ Fingers crossed that it goes well! $\endgroup$ Commented May 30 at 15:34

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .