Spin Polarization & Bond Breaking/Forming

[d-farrell2]

Hi All,

I have come across some papers and books talking about spin forbidden reactions and how they can show up in ab initio calculations where spin is not accounted for, and have a question as to if this is a concern in DFT (most of the examples were for HF, but 1 paper specifically details DFT issues).

The common example given is the dissociation of the neutral H2 molecule. I have also seen H2 with an extra electron shown.

When the two atoms are pulled apart, it should dissociate into 2 neutral H atoms - but according to the literature, if you don't allow for spin polarization, you end up with H- & H+ .

I have tried this calculation with VASP, and didn't see this behavior (or it wasn't obvious) - so my question is : is this actually a problem due to spin polarization or something else entirely?

What I did was set up a 'large' cell (10x10x10 angstrom), put 2 H atoms close together (.5 angstrom to start) and ran a constant-velocity dynamics (gave one of the atoms a velocity away from the other, which was held fixed) simulation with and without spin polarization. I did this with the PAW/GGA PPs - to compare I also did a no spin case with PAW/LDA.

What I found is that in the PAW/GGA case, regardless of the spin polarization, the energy vs distance, electronic density and bader charges associated with each atom were essentially identical.

Based on what I had read, I expected to see the no-spin case to have a clearly larger electronic density & charge around one atom than the other.

[admin]

if you look at the energetics (total energies) of a system with 2 H atoms sufficiently apart (starting with a large distance and making one single run), you will see that FM alignment (both H spin up) is more favorable (more than 2eV for a H-H distance slightly larger than 7Ã…) than the non-magnetic solution. Please note that if ISPIN=1 (or not set), the system does not converge to H+ + H-, but to spin distribution leading to an overall moment of the unit cell of 0. This has the same total energy and partial charges as 2 H atoms with AFM alignment, which of course is the less stable spin configuration for 2 non-interacting H atoms).

[d-farrell2]

so - if I follow, then in the case of H2 dissociation by separating the atoms after starting at a close distance, I should get these two cases after pulling the atoms apart to a large distance (order of 10 angstom should work, I would think):

1) ISPIN=1 : total energy and partial charges on the H atoms correspond to the AFM solution (1 spin up, 1 spin down)

2) ISPIN=2: total energy, partial charges on the H atoms correspond to the FM solution (2 spin up)

So I shouldn't expect to see a difference in the electronic density around each atom then, or at least not a big one (since they would each have 1 electron)?

[d-farrell2]

In doing some additional testing, it seems with constant velocity MD that the total spin magnetization (when running w/spin polarization) stays the same throughout my run (the initial value)... is this the expected behavior?

I would expect the spin-polarized calculations to be able to converge to the non-zero spin value at large separations.

Instead, when the initial state is 'bonded' (2 H atoms close together, so mag = 0), it stays zero as I pull them apart (to up to 5.5 Angstrom apart). Meanwhile, if I start with them unbonded in the ferromagnetic state (5.5 Angstrom apart, initial mag = 2), it stays 2 as I push the atoms together, so I get an increasing energy rather than something which has a minimum at the H-H bond distance (i.e. I get an anti-bonding state).

Is there any explanation for this? Am I not specifying something correctly?