Wrong total energies for charged systems
Posted: Mon Jun 27, 2011 10:59 pm
I've been trying to use the delta-SCF approach to calculating positions of impurity levels and have found a problem with VASP total energies.
The simplest example that demonstrates this problem is the primitive unit cell of diamond. I converge the ground state geometry and total energy and then freeze the geometry and calculate the total energy of diamond with one hole per unit cell (admittedly not a physically relevant calculation) and take the total energy difference, E(N) - E(N-1). PWSCF and fhi-AIMS both give similar results for this, 10.5 eV and 10.6 eV, but VASP gives different results, which are strongly pseudopotential dependent: 8.6 eV for the soft and hard PAWs, 7.2 eV for the standard carbon PAW. This error is purely coming from the E(N-1) term. The KS spectra all agree and based on the decomposition of total energy give in VASP, this looks like a problem in the V(G=0) potential energy term.
Using deltaSCF on molecules in large supercells doesn't seem to have this problem in VASP, which makes it all the more strange.
The simplest example that demonstrates this problem is the primitive unit cell of diamond. I converge the ground state geometry and total energy and then freeze the geometry and calculate the total energy of diamond with one hole per unit cell (admittedly not a physically relevant calculation) and take the total energy difference, E(N) - E(N-1). PWSCF and fhi-AIMS both give similar results for this, 10.5 eV and 10.6 eV, but VASP gives different results, which are strongly pseudopotential dependent: 8.6 eV for the soft and hard PAWs, 7.2 eV for the standard carbon PAW. This error is purely coming from the E(N-1) term. The KS spectra all agree and based on the decomposition of total energy give in VASP, this looks like a problem in the V(G=0) potential energy term.
Using deltaSCF on molecules in large supercells doesn't seem to have this problem in VASP, which makes it all the more strange.