Author Topic: Bad to no SCF convergence in higher temperature Molecular Dynamics  (Read 2922 times)

JakubV

  • Full Member
  • ***
  • Posts: 39
  • Karma: +0/-0
Dear Turbomole Experts,

I would like to ask about hints/workarounds to bad to no SCF convergence (RI-DFT (marij) with PBE and SV/def2-SVD basis set) in Molecular Dynamics for higher temperature. I am using the default time step=80 a.u. and thermometer relaxation time 800 a.u.
Temperature 400 K. Jobex script with -md

My problem is that MD often does only ca. three steps and end as SCF in some single-point fails to converge. What I blame on the fact that in that single point, the molecular geometry is a bit too far from equilibrium. I need to proceed and I am using the MD mostly just to search through extensive configuration space, I definitely do not need very accurate energy values. I was even considering the use of UFF (universal force field) and CMD (just run inside define routine, for example)... but for my system*, the UFF gives strange results...

So my take was to significantly increase the threshold value of SCF convergence... or if it would be possible to somehow force the program to accept the orbital data and force it to somehow compute gradient even if SCF cycles after given number of iterations don't converge?   

* 2D array of Carboranes on metallic surface (monoatomic layer). In total, one or two hundred of atoms (but I do consider upscale to several hundred).

Best regards,
JakubV

uwe

  • Global Moderator
  • Hero Member
  • *****
  • Posts: 558
  • Karma: +0/-0
Re: Bad to no SCF convergence in higher temperature Molecular Dynamics
« Reply #1 on: June 08, 2021, 05:10:01 PM »
Hi,

for the SCF convergence in such cases it might be helpful to use the $fermi option to enable fractional occupations and to not restrict the spin state. Note that higher DIIS damping is required in that case.

Did you try xtb for the MD simulation instead? With newer Turbomole versions, try 'jobex -md -level xtb'.


Arnim

  • Developers
  • Sr. Member
  • *
  • Posts: 253
  • Karma: +0/-0
Re: Bad to no SCF convergence in higher temperature Molecular Dynamics
« Reply #2 on: June 14, 2021, 04:40:04 PM »
Hello,

a time step of 80 a.u. is rather large. For high frequency modes it is often necessary to use smaller time steps, e.g. 40 a.u. in order to sample the motions.

Best
Arnim