1
$\begingroup$

I'm going through Marzari's paper about maximally localized wannier functions. There is a passage I'm trying to understand but can't seem to get. On the third page, between equation 5 and 6, they have the following:

$$ \left\langle u_{nk}|u_{mk+b}\right\rangle = \left\langle\psi_{nk}|e^{-ib\cdot r}|\psi_{mk+b}\right\rangle $$ which is trivial to prove using Bloch's theorem: $\left.|\psi_{nk}\right\rangle = e^{ik\cdot r}\left.|u_{nk}\right\rangle$. However, what I don't get is the following part where they state that the above equation is equal to:

$$ \left\langle\psi_{nk}|e^{-ib\cdot r}|\psi_{mk+b}\right\rangle = \sum_R e^{-ik\cdot R}\left\langle Rn|e^{-ib\cdot r}|0m\right\rangle $$

where $\left.|Rn\right\rangle=\int dk e^{-ik\cdot R}\left.|\psi_{nk}\right\rangle$ is a Wannier function centered in cell R (the inverse transform is $\left.|\psi_{nk}\right\rangle = \sum_R e^{ik\cdot R}\left.|Rn\right\rangle$). I'm trying to prove this to myself but can't figure out how they just "note" this. As per the definition of Wannier functions, I would believe there would be a double sum on Wannier centers like so:

$$ \left\langle\psi_{nk}|e^{-ib\cdot r}|\psi_{mk+b}\right\rangle = \sum_{RR'} e^{-ik\cdot R}\left\langle Rn|e^{-ib\cdot r}|R'm\right\rangle e^{i(k+b)\cdot R'} $$

and I believe it would be possible to use translation/symmetry arguments to rearrange this equation such that the rightmost ket is expressed from the wannier function at the origin $\left.|0m\right\rangle$. However this would not kill one of the sum as the statement suggests.

So my question is: How did they arrive at this equation? My guess is that one could use the properties of the Bloch function to deduce this but I'm not sure how.

As a follow up: this result is somewhat applied to any operator matrix element (like the Hamiltonian). Therefore I believe the proof of this statement should hold for any operator.

N.B.: In all the above, R, r, k and b are all 3D vectors.

$\endgroup$

1 Answer 1

0
$\begingroup$

Ok I believe I got the answer right after posting the question. The trick is to revert back to real space (using the identity $1=\int dr\left.|r\right\rangle\left\langle r|\right.$:

$$ \sum_{RR'}e^{-ik\cdot R}e^{i(k+b)\cdot R'}\int dr\left\langle Rn|r\right\rangle e^{-ib\cdot r}\left\langle r|R'm\right\rangle $$

but wannier functions are related by translation operation $\left\langle r|R'm\right\rangle = \left\langle r - R'|0m\right\rangle$. Then, doing a change of variable $r\rightarrow r+R'$, you get $$ \sum_{RR'}e^{-ik\cdot R}e^{i(k+b)\cdot R'}\int dr\left\langle Rn|r+R'\right\rangle e^{-ib\cdot (r+R')}\left\langle r|0m\right\rangle = \sum_{RR'}e^{-ik\cdot (R - R')}\int dr\left\langle R-R'n|r\right\rangle e^{-ib\cdot r}\left\langle r|0m\right\rangle$$

where an exponential got cancelled out. From then, we can get rid of the temporary real space integral: $$ = \sum_{RR'}e^{-ik\cdot (R - R')}\left\langle R-R'n|e^{-ib\cdot r}|0m\right\rangle $$

and now we can use a simple symmetry argument. Since we sum on all possible values of $R-R'$ and since that $R-R'$ is itself a wannier center as well (R and R' are part of a Bravais lattice), this double sum is equivalent to a single sum on all bravais lattice vectors. Therefore, we can get rid of the sum on $R'$ for instance and replace any occurance of $R-R'$ for $R$ only. Therefore,

$$ = N\sum_{R}e^{-ik\cdot R}\left\langle Rn|e^{-ib\cdot r}|0m\right\rangle. $$ where $N$ is the number of Wannier functions. Also, in the OP I did not include a factor of $1/\sqrt{N}$ in the definition of the wannier functions but usually it is included and these would cancel out the factor of $N$ appearing here.

As for a more general proof related to any operator, as long as it is diagonal in the real space and have a translational symmetry (which is true for the Hamiltonian) and that we compare matrix elements of the form $\left\langle \psi_{nk}|O(r)|\psi_{mk}\right\rangle$ (at the same k instead of k+b like above), the same argument holds.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.