26.11.2014 Views

Maximally localized Wannier functions: Theory and applications

Maximally localized Wannier functions: Theory and applications

Maximally localized Wannier functions: Theory and applications

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

8<br />

1. Real-space representation<br />

An interesting consequence stemming from the choice<br />

of (18) as the localization functional is that it allows<br />

a natural decomposition of the functional into gaugeinvariant<br />

<strong>and</strong> gauge-dependent parts. That is, we can<br />

write<br />

where<br />

<strong>and</strong><br />

Ω I = ∑ n<br />

[<br />

Ω = Ω I + ˜Ω (19)<br />

⟨ 0n | r 2 | 0n ⟩ − ∑ Rm<br />

˜Ω = ∑ n<br />

∑<br />

Rm≠0n<br />

∣<br />

∣⟨Rm|r|0n⟩ ∣ ]<br />

2<br />

(20)<br />

∣<br />

∣⟨Rm|r|0n⟩ ∣ ∣ 2 . (21)<br />

It can be shown that not only ˜Ω but also Ω I is positive<br />

definite, <strong>and</strong> moreover that Ω I is gauge-invariant,<br />

i.e., invariant under any arbitrary unitary transformation<br />

(10) of the Bloch orbitals (Marzari <strong>and</strong> V<strong>and</strong>erbilt, 1997).<br />

This follows straightforwardly from recasting Eq. (20) in<br />

terms of the b<strong>and</strong>-group projection operator P , as defined<br />

in Eq. (15), <strong>and</strong> its complement Q = 1 − P :<br />

Ω I = ∑ ⟨0n|r α Qr α |0n⟩<br />

nα<br />

= ∑ α<br />

Tr c [P r α Qr α ] . (22)<br />

The subscript ‘c’ indicates trace per unit cell. Clearly<br />

Ω I is gauge invariant, since it is expressed in terms of<br />

projection operators that are unaffected by any gauge<br />

transformation. It can also be seen to be positive definite<br />

by<br />

∑<br />

using the idempotency of P <strong>and</strong> Q to write Ω I =<br />

α Tr c [(P r α Q)(P r α Q) † ] = ∑ α ||P r αQ|| 2 c.<br />

The minimization procedure of Ω thus actually corresponds<br />

to the minimization of the non-invariant part<br />

˜Ω only. At the minimum, the off-diagonal elements<br />

|⟨Rm|r|0n⟩| 2 are as small as possible, realizing the best<br />

compromise in the simultaneous diagonalization, within<br />

the subspace of the Bloch b<strong>and</strong>s considered, of the three<br />

position operators x, y <strong>and</strong> z, which do not in general<br />

commute when projected onto this space.<br />

2. Reciprocal-space representation<br />

As shown by Blount (1962), matrix elements of the<br />

position operator between WFs take the form<br />

∫<br />

V<br />

⟨Rn|r|0m⟩ = i<br />

(2π) 3 dk e ik·R ⟨u nk |∇ k |u mk ⟩ (23)<br />

<strong>and</strong><br />

⟨Rn|r 2 |0m⟩ = −<br />

V ∫<br />

(2π) 3 dk e ik·R ⟨u nk |∇ 2 k|u mk ⟩ . (24)<br />

These expressions provide the needed connection with<br />

our underlying Bloch formalism, since they allow to express<br />

the localization functional Ω in terms of the matrix<br />

elements of ∇ k <strong>and</strong> ∇ 2 k<br />

. In addition, they allow to<br />

calculate the effects on the localization of any unitary<br />

transformation of the |u nk ⟩ without having to recalculate<br />

expensive (especially when plane-wave basis sets are<br />

used) scalar products. We thus determine the Bloch orbitals<br />

|u mk ⟩ on a regular mesh of k-points, <strong>and</strong> use finite<br />

differences to evaluate the above derivatives. In particular,<br />

we make the assumption that the BZ has been discretized<br />

into a uniform Monkhorst-Pack mesh, <strong>and</strong> the<br />

Bloch orbitals have been determined on that mesh. 4<br />

For any f(k) that is a smooth function of k, it can<br />

be shown that its gradient can be expressed by finite<br />

differences as<br />

∇f(k) = ∑ b<br />

w b b [f(k + b) − f(k)] + O(b 2 ) (25)<br />

calculated on stars (“shells”) of near-neighbor k-points;<br />

here b is a vector connecting a k-point to one of its neighbors,<br />

w b is an appropriate geometric factor that depends<br />

on the number of points in the star <strong>and</strong> its geometry<br />

(see Appendix B in Marzari <strong>and</strong> V<strong>and</strong>erbilt (1997) <strong>and</strong><br />

Mostofi et al. (2008) for a detailed description). In a<br />

similar way,<br />

|∇f(k)| 2 = ∑ b<br />

w b [f(k + b) − f(k)] 2 + O(b 3 ) . (26)<br />

It now becomes straightforward to calculate the matrix<br />

elements in Eqs. (23) <strong>and</strong> (24). All the information<br />

needed for the reciprocal-space derivatives is encoded in<br />

the overlaps between Bloch orbitals at neighboring k-<br />

points<br />

M (k,b)<br />

mn = ⟨u mk |u n,k+b ⟩ . (27)<br />

These overlaps play a central role in the formalism, since<br />

all contributions to the localization functional can be<br />

expressed in terms of them. Thus, once the M mn<br />

(k,b)<br />

have been calculated, no further interaction with the<br />

electronic-structure code that calculated the ground state<br />

wave<strong>functions</strong> is necessary - making the entire <strong>Wannier</strong>ization<br />

procedure a code-independent post-processing<br />

step 5 . There is no unique form for the localization functional<br />

in terms of the overlap elements, as it is possible<br />

4 Even the case of Γ-sampling – where the Brillouin zone is sampled<br />

with a single k-point – is encompassed by the above formulation.<br />

In this case the neighboring k-points are given by reciprocal lattice<br />

vectors G <strong>and</strong> the Bloch orbitals differ only by phase factors<br />

exp(iG · r) from their counterparts at Γ. The algebra does become<br />

simpler, though, as will be discussed in Sec. II.F.2.<br />

5 In particular, see Ferretti et al. (2007) for the extension to<br />

ultrasoft pseudopotentials <strong>and</strong> the projector-augmented wave<br />

method, <strong>and</strong> Freimuth et al. (2008); Kuneš et al. (2010); <strong>and</strong><br />

Posternak et al. (2002) for the full-potential linearized augmented<br />

planewave method.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!