optimization of a large grahene cluster
Clicked A Few Times
3:11:28 PM PDT - Fri, Oct 25th 2019
Hello all,
I am trying to optimize and calculate the frequencies for a relatively large graphene cluster. I am using the criteria below. I am having a hard time optimizing it. Can someone recommend me better parameters for optimization?

dft
 odft

  mult 2

  xc b3lyp

  maxiter 2000

  CONVERGENCE energy 1e-4

  CONVERGENCE gradient 1e-3

  grid medium

  direct

end



driver

  clear

  LOOSE

  eprec   1e-03

  maxiter 2000

end

basis spherical
 * library cc-pvdz
end

  1. task dft optimize

task dft freq numerical


Best,
Gangotri

Clicked A Few Times
2:22:32 PM PDT - Mon, Oct 28th 2019
Is the SCF not converging for the initial geometry? Or if the SCF converges, but the geometry optimizations are not converging, are the energies fluctuating? Or is the energy monotonically decreasing, but the calculation is just very slow? These scenarios would each lead to different suggestions.

Also, how many atoms are you looking at and what level of theory do you need? Perhaps open-shell DFT is just too expensive a method to use for your system. The frequency calculations will take much more time than the geometry optimization... Maybe you can optimize the geometry and calculate vibrational frequencies using molecular mechanics, semiempirical methods, or restricted open-shell DFT, and then do a single-point electronic energy evaluation using open-shell DFT?

Clicked A Few Times
2:12:33 PM PDT - Tue, Oct 29th 2019
Hello! Thank you for your reply. The scf is not optimizing even for the first run i.e. your very first point. There are more than 500 iterations done but the energies are just fluctuating. I have optimized the geometry using empirical methods before I put them in the NWCHEM. Is there any changes I can do in the dft and driver module that can help in the optimization?

G

Clicked A Few Times
6:35:53 AM PDT - Wed, Oct 30th 2019
Having not ran large graphene simulations myself, I can only give general SCF convergence suggestions. Some possible strategies for SCF convergence are conveniently listed on Orca Input Library and can be adapted for NWChem:

1) Use the quadratic convergence algorithm: cgmin (analogous to !NRSCF keyword in orca). Here's a testimonial for cgmin with metallic systems in NWChem.
2a) Run a preliminary HF calculation, and use those vectors to start DFT calculation (example code), and/or
2b) Run a preliminary calculation for a cationic, closed-shell system, and then read those vectors into your desired neutral, open-shell system (example code).
3) Use a finer grid
4) Use more DIIS Fock matrices (default 10, but Orca Input Library suggests that 15-40 may be necessary for open-shell systems).
5) Increase damping and number of damped cycles for DIIS (somewhat analogous to !TIGHTSCF keyword in Orca)

Example input to implement suggestions 3, 4, and 5:

dft
  grid fine
  convergence diis 20
  convergence damp 90
  convergence ncydp 30
end


Forum >> NWChem's corner >> General Topics