Electronic Convergence

Quick Description: We can treat electronic and ionic convergence as separate steps so as to mathematically simplify solving Schrodinger's Equation. More elaborately put: "The forces on both electrons and nuclei due to their electric charge are of the same order of magnitude, and so the changes which occur in their momenta as a result of these forces must also be the same. One might, therefore, assume that the actual momenta of the electrons and nuclei were of similar magnitude. In this case, since the nuclei are so much more massive than the electrons, they must accordingly have much smaller velocities. Thus it is plausible that on the typical time-scale of the nuclear motion, the electrons will very rapidly relax to the instantaneous ground-state configuration, so that in solving the time-independent Schrödinger equation... we can assume that the nuclei are stationary and solve for the electronic ground-state first, and then calculate the energy of the system in that configuration and solve for the nuclear motion." - Peter Haynes

The Point: Sometimes you will have problems reaching electronic convergence in a single ionic step. Here are a few things to try.

Notes: These notes are collected from quite a few different forum posts (here, here, and here) with some help from my labmate, Jessica McKinley. The vasp manual is a really great resource for this problem too.

• Increase NELM, which sets the max number of electronic self consistency steps. However, "Normally, there is no need to change the default value: if the self-consistency loop does not converge within 40 steps, it will probably not converge at all." - VASP guide
• Set "ALGO = Normal" in your INCAR.
• Add a delay by setting NELMDL. This will increase the number of non-self-consistent to improve the pre-convergence of the wavefunction. You can either set this at the beginning of the job (by using a negative number) or by adding this delay at every ionic step (by using a positive number). This delay is especially important when the SC-convergence is bad, as is the case surfaces, molecules/clusters, chains, or charge density difference plots.  This delay can speed up convergence and can be essential to achieving convergence at all.
• Play with mixing parameters AMIX and BMIX, and/or AMIX_MAX and BMIX_MAX (for the spin-polarized runs). Reduce these parameters (perhaps a few steps with linear mixing first to generate an initial WAVECAR), then rerun from there with the default mixing parameters. Make the preconvergence easier and then attempt to reach true convergence. It may help to use L(S)DA+U. If you have f-elements or d-elements, set LMAXMIX to 6 or 4, respectively.