[Wien] total energy when using -orbc for constant V-matrix

[Stefaan Cottenier]

>   I tried to use LDA+U with -orbc flag to stabilize various orbital polarized
> states in my calculations for some transition-metal oxides. In order to 
> compare their total energies, I made a simple test, namely, run 'LDA+U 
> with -orbc' for a well converged 'LDA+U with -orb' case. The total energies
> are expected to be the same. However,  it is surprising that in my current 
> calculations using the latest version, the total energies change by several 
> Rydbergs. I am wondering if any term is missing in the calculations of the
> total energy when using -orbc. In addition, can one use -orbc to do structural
> relaxation like doing LDA+U with -orb ?

-orbc is not meant for 'final' calculations, but rather for bringing the 
scf cycle close to a type of solution you want. Then you turn off -orbc 
(i.e. use plain -orb), after which the scf cycle will spontaneously 
stabilize in the nearest local energy minimum, which can (but not always 
should) have the qualitative properties of the solution you wanted to 
enforce.

I did notice indeed that when used that way, there are several Ry's of 
difference between energies with and without -orbc. I cannot give you 
the technical explanation for this (others ?), but never cared about it 
as I always use the energies with plain -orb anyway.

Note: -orbc freezes the orbital potential that is consistent with the 
initial density matrices. It does not freeze the density matrices 
themselves: they can evolve during the scf cycle. Hence, when using 
-orbc as described above, you better check at the end of the -orbc scf 
cycle whether the final density matrices are not too different from your 
initial ones. If they are very different, you probably did not end up 
close the type of solution you wanted to enforce, and the subsequent 
-orb scf cycle might result into a selfconsistent solution that you do 
not want.

Stefaan