The manual for VASP recommends using ISTART = 1 for calculations wherein the cell shape/volume changes, but it recommends ISTART = 2 for the same scenario as well. A little clarification (+ discussion) is appreciated.
First of all, you have to remember that this flag is used to read the WAVECAR. So, if you want to read the information from the WAVECAR to start a new calculation you have to put ISTART = 1.
http://cms.mpi.univie.ac.at/vasp/guide/node101.html
Dear Sinhue, I read in the manual that if we restart the calculation as a continuation of the previous relaxation with the WAVECAR file from this previous run and an empty INCAR then it works. Does this mean that the default value for ISTART in that case becomes either 1, 2, or 3?
Default is always 0 for ISTART but if WAVECAR is present then default ISTART is 1. If you want to use any other value of ISTART then you have to specify it in the INCAR.
To quote Georg Kresse, "fixed basis set calculations are obviously a very bad idea"
see Slide 7 of this presentation
https://www.vasp.at/vasp-workshop/slides/accuracy.pdf
Erik Nykwest I have a lot of confusion with basis sets. The last line for ISTART = 2 in http://cms.mpi.univie.ac.at/vasp/guide/node101.html is a bit contradictory with the workshop slide that you shared. Can you please help me understand basis sets in terms of cell shape/volume relaxation?
Hitanshu Sachania, so the last line reads:
"There is only one exception to this general rule: All volume/cell shape relaxation algorithms implemented in VASP work with a constant basis set, so continuing such jobs requires to set ISTART=2 to get a 'consistent restart' with respect to the previous runs (see section 9.6)!"
There is a difference between "correct" and "consistent". There is no formal contradiction between the above statement and Kresse's statement. The guide is saying "if you want to be consistent use a fixed basis set", while Kresse is saying that "using a fixed basis set is a bad idea because it's not accurate". Both statements can be true at the same time. However, that leads us to the question "if it's a bad idea, then why does VASP do it that way during relaxations?"
If you look at the 240 eV energy volume curves on page 7 you can see that the "basis set fixed" calculations vary smoothly but lead to the wrong equilibrium volume, while the "cutoff fixed" calculations are jagged but lead to the correct equilibrium volume. If the energy changes smoothly then it is easy to extrapolate the minima. I suspect this is why VASP keeps the basis set fixed during minimization.
But why do fixed basis set calculations lead us to the incorrect answer?
and on page 8
"for the expanded lattice the basis set corresponds effectively to a lower cut-off Gcut and therefore a lower quality"
In other words, if you keep your basis set fixed, then the accuracy of you calculations change! So the energy vs volume curve loses (or gains) accuracy the further you get from the starting volume.
So what does this mean for relaxations? If you have a cell whose volume is drastically different from the equilibrium volume and you do one long continuous relaxation (lets say it takes N steps), then it will follow a nice smooth energy vs volume curve that may lead to the WRONG equilibrium volume.
Now consider the same system, except you break the calculation into more, smaller, calculations (for example 2 calculations of N/2 steps), and when you restart VASP you use a fixed basis set. Then when you restart VASP the energy will be continuous (i.e. the energy of the last ionic step of the previous relaxation will be equal to the energy of the first ionic step of the current relaxation) and the relaxation will follow a the same energy vs volume curve to the same equilibrium volume.
However, if you break the calculation into more, smaller, calculations but you use a fixed energy cutoff when you restart VASP instead, then when you restart VASP the energy will jump discontinuously (i.e. the energy of the last ionic step of the previous relaxation will not be equal to the energy of the first ionic step of the current relaxation) and the relaxation will follow a DIFFERENT energy vs volume curve to a different equilibrium volume.
So how should you relax your structure? My suggestion is
1) Ensure your calculations are well converged in encut.
2) Once you have reached a relaxed structure resubmit your relaxation job. If you have reached a false equilibrium volume it may continue relaxing.
- Badges
- Science method
- Similar topics
- Methodology
- Crystallography