Author Topic: Force and stress tolerances for 2D monolayer optimization  (Read 3682 times)

0 Members and 1 Guest are viewing this topic.

Offline AD

  • Regular QuantumATK user
  • **
  • Posts: 11
  • Country: be
  • Reputation: 0
    • View Profile
Hi,

I have to optimize multiple supercells and I am wondering if I can follow some guidelines in deciding parameters that can
quickly(reasonably) optimize the systems. I have optimized the unit cells of various 2D materials monolayers with large enough
vacuum (almost 20 Angstrom) on top. When I optimize the supercells now with max_force = 0.01ev/Ang and max_stress=0.001eV/Ang3(same as for unit cells), I find that the supercell made of optimized unit cell still takes quite some time to finish. Since the 2D materials consist of different metals and chalcogens, is there a safe cut-off that can help optimize the supercells rather quickly? Say max_force=0.05 and max_stress=0.05? Will it ber reasonable? I want to calculate total energies in the end and later dos.  K points used are 1x1x1 (for unit cells it was 5x5x1). GGA.PBE with vdW correction and DZP basis is used. Can I fix the z coordinate for supercells to save time in relaxation? I didn't do it earlier because sometimes (for some systems) I see the z direction change during relaxation even with 20 Ang vacuum. Since this is just a first step and later  I have to model interactions of molecules with these monolayers, I would like to find a way that can give reasonably accurate energies and save time.

Thanks for your inputs.

Offline lknife

  • QuantumATK Guru
  • ****
  • Posts: 214
  • Country: us
  • Reputation: 1
    • View Profile
Re: Force and stress tolerances for 2D monolayer optimization
« Reply #1 on: June 21, 2017, 02:01 »
"I have optimized the unit cells of various 2D materials monolayers with large enough
vacuum (almost 20 Angstrom) on top. "

How did you optimize the unit cell of 2D monolayers? What's your parameters for the geometry optimization? Normally, if you well did the optimization, you can change it to a supercell without additional optimization.

"K points used are 1x1x1 (for unit cells it was 5x5x1). "
In my opinion, at least, the K point 1x1x1 is not enough for any material. Maybe the simulation on quantum dot can use this setting.


Offline AD

  • Regular QuantumATK user
  • **
  • Posts: 11
  • Country: be
  • Reputation: 0
    • View Profile
Re: Force and stress tolerances for 2D monolayer optimization
« Reply #2 on: June 21, 2017, 11:27 »
Hi,
Thank you for your reply.

I have given the parameters in my question. When I asked this question my main concern was If I can go for a higher stress and force settings to improve time.The unit cells in most cases consist of 3 atoms (take MoS2 for example) and can be obtained from materials project. Inside geometry optimization I use 0.01eV/Ang max_force, 0.001 eV/Ang max_stress. Steps 200 and nothing is constrained (neither x,y,z nor bravais lattice). I guessed that the cut-offs above are already rigorous so that shouldn't be a problem. One thing to clarify though that the supercells are very large (8x8x1). As far as I have read in journals, people don't use such large cells (mostly 5x5x1 supercells). I had to use bigger for consistency and because of previous experiences with certain materials. So using 1x1x1 k points for supercells won't be a problem. In case of single unit cells it will cause problems and ATK detects that I should use 4x3x1 k-points in general, but to be safe across all materials I go with 5x5x1 k-points (With 20 Ang vacuum, the z direction is way too big to use more k points. Of course I can use but I shouldn't have to). Other settings in Geometry optmization are standard LBFGS and 0.2 Ang step size.


Thanks

Offline lknife

  • QuantumATK Guru
  • ****
  • Posts: 214
  • Country: us
  • Reputation: 1
    • View Profile
Re: Force and stress tolerances for 2D monolayer optimization
« Reply #3 on: June 21, 2017, 19:41 »
OK, I see. Your 8x8x1 supercell is just like the supercell in tutorial "mobility" when calculating the Hamiltonian derivatives. Thus, 1x1x1 k-point sampling is enough. I made a mistake.

As to your main concern "If I can go for a higher stress and force settings to improve time", it can certainly speedup your calculation. However, for a serial of similar calculations, I think you'd better at least compare the results obtained with different settings once or more to make sure there is no significant difference between them. After all, the results obtained with higher-accuracy settings must be more accurate than the results obtained with lower-accuracy settings.


Offline AD

  • Regular QuantumATK user
  • **
  • Posts: 11
  • Country: be
  • Reputation: 0
    • View Profile
Re: Force and stress tolerances for 2D monolayer optimization
« Reply #4 on: June 21, 2017, 22:09 »
Thanks for the reply. Yes, I guess a benchmark would be helpful. I was just hoping that there should be some general settings
that I can use safely specially in the context of supercells (maybe a lower restraint).

Thanks for your input.