1
General Questions and Answers / Re: Constraints_data parameter in function "writePOSCAR"
« on: April 15, 2024, 11:48 »
Thanks for your reply.
This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.
You can read more about it here: https://docs.quantumatk.com/tutorials/dft_half_pps/dft_half_pps.html. In short, it's an attempt to bring some aspect of empirical pseudopotentials into DFT, but instead of a scissor operator acting on the final band structure (which wouldn't work in Ge for instance) there are shifts being applied within the self-consistent cycle. It does require empirical tuning, i.e. you should know the band gap from experiments to tune the parameters.
We implemented this method quite a long time ago, and honestly have not found much use for it. Since then, we have added DFT+1/2 which seems to be a more robust and general scheme, where the same parameters seem to work for many different systems, although you can fine tune them if you do know the band gap. And of course HSE06 and other hybrid functionals which perform really well in LCAO, and which can be truly predictive of the band gaps of unknown materials.
Great.ranging from 100 to 500 atoms. Using 2-4 process/ 2-4 nodes per k point
How many cores you are using & how many atoms?
WelcomeHi, I found out the slow convergence is caused by the ghost atoms . After fixing the coordinates of those ghost atoms (in between VDW layers), the optimization of geometry becomes significantly more efficient.
Let me know which one works for you.
Yes, you are right iteration control affects electronic convergence.Thank you for your useful guide.
While increasing mesh and K point do help a system converge but it makes it converge slower.
A tip for fast convergence would be:
1. If you are suppose converging an unpolarized system, try first with low k point while keeping mesh cutoff higher or normal is what to be kept. After your system is converged at low K point increase K point to what u want. This can help achieve quicker optimization
2. If you are stuck with polarized optimization, then best is to perform unpolarized optimization first then use converged file to perform polarized optimization. This surely has worked in my case & is well reported method.
3. Another way is to increase your cores or nodes to achieve quicker optimization.
4. If still you are not getting quick optimization, perform a small atomic rattling for your structure in builder and perform any of the above steps. This should help too.
5. If still you are not getting quick optimization, then contact synopsys at solvnet.
Try playing with parameters in Iteration control. Look into manual.Thank you for the suggestion. From what I understand, appropriate iteration control parameter would improve electronic convergence, but having little effect towards ionic (relaxation) convergence. Do you think increase density_mesh_cutoff and kpoints density would help ionic convergence?