[gpaw-users] energy discontinuity with fixed magnetic moment

[Jens Jørgen Mortensen]

On Sat, 2011-03-19 at 02:46 +0100, Jason Gruber wrote:
> Hi all,
> 
> This issue came up in my work and I couldn't find any previous posts
> on this topic, so here goes.  I've been doing some calculations with
> bulk metallic solids at large lattice spacings, some of which tend to
> favor having a non-zero magnetic moment, hence I've been using spin
> polarized calculations.  What I find is that as I use smaller atomic
> volumes, the converged magnet moment tends toward zero.
> 
> To check whether I'm converging to the right magnetic moment, I've
> performed calculations where I set a fixed atomic volume and compute
> total energies as a function of magnetic moment using fixmagmom=True.
> In some cases, the total energy and magnetic moment change smoothly
> until the moment converges to zero, at which point I get a
> discontinuity in the energy.  Wondering why this is, I'm asking
> 
> 1.  When the magnetic moment is fixed to exactly zero with
> fixmagmom=True, does GPAW switch to spin-paired mode?

No - only if all the initial magnetic moments are zero.  If you have the
"magnetic moment column" then it is running in spin-polarized mode.

>   I still get the magnetic moment column of output during SCF either
> way, but, the energy computed this way matches the energy computed
> with a spin-paired calculation.
> 
> 2.  Is there a simple reason why the energy at zero magnetic moment
> jumps discontinuously above the smooth curve I get with 0.005, 0.010,
> 0.015, etc. fixed magnetic moments?

Could you send us an example (input script) that shows this
discontinuity?

Jens Jørgen

>   I'm using the PBE functional, but I don't see anything in PBE that
> suggests the limit of the spin-polarized XC functional doesn't
> converge to the spin-paired value.  Is it just that having the
> additional eigenstates allows for more "variational freedom" to find
> lower energies despite the fact that there are all paired electrons?
> 
> Thanks for any advice,
> 
> Jason