Different energies from the same input files with isif=3

[felixvasp]

Dear Users and developers,
I have encountered some problems calculating the energies with isif=3.
First task: a geometric optimization with isif-3 is done and energy E0 is obtained.
Second task: use the optimized structure from task 1 as starting structure to do another geometric optimization with the same input files as task 1

The first step of the geometric optimization in task 2 gives a different energy from the one from the final step in task 2. The difference is ~0.6 eV for a 256-atom system, any ideas why this is happening? thanks.

<span class='smallblacktext'>[ Edited ]</span>

[giacomo giorgi]

I suppose that changing the lattice parameter in the optimization has lead to a change in the PW cut off.

Thus, new lattice parameter in the restarted job, different cut off, new energy....

[felixvasp]

Hi, glacomo
Thanks for reply, In fact, to avoid the pulay stress error I used a pretty high energy cutoff (600 eV), but still the energies of two tasks are different. Also, I tried copying the WAVECAR of task 1 to task2 and run task 2 with restart=1 (trying to keep constant energy cutoff), but the energy is the same as the task 2 with restart=0.Thanks.
[quote="'smallblacktext'>[ Edited Wed Feb 06 2013, 09:15PM "]</span>

[dem]

Hi guys,
I've got similar results with vasp 5.2.12. The energies calculated in a run with isif=3 and then again (either with isif=2 or isif=3) for the previously equilibrated unit cell do not match. In fact, I've got two similar systems and for one system the new energy is 0.06eV lower than the old one, while for the other system it is 0.09eV higher. Both unit cells contain ~100 atoms and ENCUT=600eV.

I guess this means that the energy in runs with isif=3 tends to drift (the more the cell dimensions change the worse is the drift) and it is worth recalculating the energy in a separate run starting from an equilibrated unit cell.

[admin]

The larger is the cell the more difficult is to find the global minimum and the higher is the probability to fall into a local minimum. If the crystal structure is known, perhaps is it for large systems better to fix the shape of the cell and using ISIF=2 construct the E vs. Volume curve.

[murphy]

Dear all,

I met the same problem. My system is a large ~100 atom supercell, but with a lower ENCUT=350 eV.
Thank you for the discussion.

So is this energy jump caused by the change of energy cutoff, or by entering into a local minimum?

Really hope some one can make it clear.

Many thanks,
Murphy