Generate WAVEDER file in VASP

In VASP calculations use

LOPTICS = .TRUE.

and increase (at least x3) the default VASP total number of bands.

NBANDS = XXXX

The aim is to increase the number of terms in the perturbation series (it converges as 1/energy) and reach empty bands with the energy ~7 Ry above the Fermi energy. There is no universal number, but this is a good/conservative starting point for s-p chemical bonds. Eventually you need to check the convergence of calculated m* with respect to NBANDS. States with a significant contribution of d electrons can be tough (even impossible) to converge. In WIEN2k for d electrons we recommend to use an extended basis set by adding high-energy local orbitals, but there is no such option in VASP, unfortunately. The error in m* for bands dominated by d electrons can be ~40%. It is recommended to verify the accuracy using a finite difference for a specific material.

Also, it is important to disable a finite differences derivative of the cell-periodic part of the orbitals

LPEAD =.FALSE.

Tests show that m* values calculated with LPEAD =.TRUE. are inaccurate, since this option implies a finite difference on a regular k-point grid. It is too sparse for accurate effective mass calculations even at the very high k-mesh density.