My Community

Dear all:

in my molecular dynamics calculation, after the electronic steps converge, the following information appears almost constantly:

Information: wavefunction orthogonal band 736..

Could you kindly tell me why the information appears. I have tried many options, such as reduce the POTIM, increase PREC from low to Medium, increase nbands, etc. I am simulating a liquid. Is the information hurt the system to reach equilibrium?

best wishes!
yigang

the line indicates that a band crossing occured between the old and the new prediceted wavefunction.
Please set IWAVPRE=11 in INCAR

Thanks for your kind answer. However, when I change the IWAVPR from 12 to 11, the number of electronic steps in each ionic step increases from 4 to 8. Is there a good way(s) to reduce the number of electronic steps for the simulation parameter (i.e. IWAVPR=11) so that the whole MD simulation time can become less?
<span class='smallblacktext'>[ Edited Wed Nov 28 2007, 03:57AM ]</span>

no, if the el-scf takes a certain number of steps, there is no rule of thumb about how to decrease it. However, you can try whether it helps to adjust the mixing parameters, or if a different BZ integration method (with different smearing, if this is reasonable) improves convergence.

Beside AMIX and BMIX, increasing the parameters MAXMIX and NBANDS may help?

Dear,

In my molecular dynamics simulation, I have also seen the same information:

Information: wavefunction orthogonal band 2238 ...

When I set set IWAVPR=11 in INCAR, it is solved and this information message disappeared.
But I am confused when I saw the following explanation in VASPwiki ( https://www.vasp.at/wiki/index.php/IWAVPR ):

IWAVPR=2|12 A second order extrapolation for the orbitals and the charge density is done (eq. (9.9) in thesis G. Kresse). A must for ab-initio MD-runs.

As I understand from this expression, IWAVPR must be 2 or 12 for MD simulations. If I set IWAVPR=11, does it cause any problem?

Thank you
Best Wishes

Selma

Dear Selma,

IWAVPR 2 or 12 is usually a better choice for the extrapolation of wave functions when doing ab-initio MD. Thus this is the default for MD. However, if band crossings are happening frequently, the simpler extrapolation based on atomic charge densities can be beneficial. Are your results feasible with IWAVPR = 11? If they are, then there is no problem.

If you have further questions, please provide a minimal example.

Cheers, Michael

Dear Selma,

another, more complete answer to your question after discussing the matter with a colleague and looking more closely at the routine that performs the wavefunction prediction:

Wavefunction prediction is done to start the electronic minimization from a more reasonable guess when the ionic positions are updated. This reduces the number of minimization steps necessary to converge the electronic system.

The code checks if the old wavefunctions are orthogonal to the new ones. If that happens, this means that something went wrong, since a switch to a different state has happened. In that case, the extrapolation to the new state is NOT done. This will result in a step that takes longer to converge since the electronic minimization starts without a wavefunction prediction. In the worst case, the SCF cycle will not converge and derail the simulation.

If this happens not too often and your MD run remains stable, you don't have to worry about the orthogonality information. If it happens a lot, it might be better to switch to the other extrapolation scheme, IWAVPR = 11. This will on average result in more electronic steps for each ionic step but will avoid having no extrapolation of wavefunctions at all if the predicted bands are orthogonal to the old ones.

I have also changed the wording on the wiki to reflect that IWAVPR = 11 does not invalidate MD runs but is just usually less efficient.

Dear Michael,

Thank you very very much for you reply.
IWACPR is quite clear for me now.

Best Wishes
Selma