SciPost logo

SciPost Submission Page

A comprehensive study of the velocity, momentum and position matrix elements for Bloch states: application to a local orbital basis

by Juan José Esteve-Paredes and Juan José Palacios

This Submission thread is now published as

Submission summary

Authors (as registered SciPost users): Juan José Esteve-Paredes
Submission information
Preprint Link: scipost_202208_00050v2  (pdf)
Date accepted: 2022-10-12
Date submitted: 2022-10-08 21:03
Submitted by: Esteve-Paredes, Juan José
Submitted to: SciPost Physics Core
Ontological classification
Academic field: Physics
Specialties:
  • Condensed Matter Physics - Theory
  • Condensed Matter Physics - Computational
Approaches: Theoretical, Computational

Abstract

We present a comprehensive study of the velocity operator, $\hat{\boldsymbol{v}}=\frac{i}{\hbar} [\hat{H},\hat{\boldsymbol{r}}]$, when used in crystalline solids calculations. The velocity operator is key to the evaluation of a number of physical properties and its computation, both from a practical and fundamental perspective, has been a long-standing debate for decades. Our work summarizes the different approaches found in the literature, but never connected before in a comprehensive manner. In particular we show how one can compute the velocity matrix elements following two different routes. One where the commutator is explicitly used and another one where the commutator is avoided by relying on the Berry connection. We work out an expression in the latter scheme to compute velocity matrix elements, generalizing previous results. In addition, we show how this procedure avoids ambiguous mathematical steps and how to properly deal with the two popular gauge choices that coexist in the literature. As an illustration of all this, we present several examples using tight-binding models and local density functional theory calculations, in particular using Gaussian-type localized orbitals as basis sets. We show how the the velocity operator cannot be approximated, in general, by the $k$-gradient of the Bloch Hamiltonian matrix when a non-orthonormal basis set is used. Finally, we also compare with its real-space evaluation through the identification with the canonical momentum operator when possible. This comparison offers us, in addition, a glimpse of the importance of non-local corrections, which may invalidate the naive momentum-velocity correspondence.

Author comments upon resubmission

Dear Editor,

Please find the resubmission of our manuscript after receiving the feedback of our referee. The changes in the text are written in red color.

Sincerely,

J.J. Esteve-Paredes and J.J. Palacios

List of changes

1. In the Introduction, we have enlarged the discussion below Eq. (1). (Referee's suggestion)
2. In Sec. II.A, we have slightly changed the notation for the operators to avoid confusions regarding the local or nonlocal character of operators. We have also elaborated a bit more below Eq. (6). (Referee's suggestion)
3. In Sec. II.B, we have removed the definition of a $\Delta$ term arising due to the incompletness of the Hilbert Space. Now we just comment this effect in the text, and also this paragraph has been moved to the end of the section. The discussion is maintained as it is relatable to our results in Section IV. (Referee's suggestion)
4. In Sec. II.D, a minor comment has been added below Eq. (12), and the wording has been slightly changed. (Referee's suggestion)
5. In Sec. II.D, the diagonal case for Eq (11) is now discussed. (Referee's suggestion)
6. In Sec. III, a comment has been added below Eq. (25) to discuss Berry connection properties. (Referee's suggestion)
7. In Sec. III, Ref. [15] was incorrectly cited below Eq. (27). It has been replaced by Ref. [8].
8. In Sec. IV.A, we have changed the acronym CREN to CRENBL, as listed in the repository www.https://www.basissetexchange.org/. We have also include a citation to this URL.
9. In Sec. IV.A, we have reacommodate the wording in the discussion based on the change number 3 in this list. Some other sentences has been rewritten in a clearer and relaxed way.
10. In Sec. IV.A, the plots now include the previous notation change in the legends.
11. In Sec. IV.A, an inset in Fig. 2(b) has be included showing how we arrange the atoms in our numerical calculations. The reader could now recreate our results easier.
12. In Appendix B, the overlap matrix gradient has been explicitely shown in terms of the local orbitals. A identification with the former is made in the text below. (Referee's suggestion)

Published as SciPost Phys. Core 6, 002 (2023)


Reports on this Submission

Anonymous Report 1 on 2022-10-10 (Invited Report)

  • Cite as: Anonymous, Report on arXiv:scipost_202208_00050v2, delivered 2022-10-09, doi: 10.21468/SciPost.Report.5859

Strengths

See previous report. Overall I judge that the authors have done a conscientious job of replying to my comments and making corresponding changes to the manuscript. I have additional comments concerning my comments 1(b) and 1(d) that are detailed in the attachment, which may lead to optional revisions.

Weaknesses

See previous report

Report

I think the manuscript now meets the journal's standards and should be accepted for publication.

Requested changes

See attached PDF.

Attachment


  • validity: high
  • significance: good
  • originality: good
  • clarity: high
  • formatting: excellent
  • grammar: good

Author:  Juan José Esteve-Paredes  on 2022-10-14  [id 2921]

(in reply to Report 1 on 2022-10-10)

We acknowledge the referee the recommendation for publication. We also include below a small response to the last comment by our referee. We count the equations the referee's last response from 1 to 4, as seen in the webpage for the manuscript submission.

Overall I judge that the authors have done a conscientious job in replying to my comments and making corresponding changes in the manuscript. I still have my reservations about the way my comments 1(b) and 1(d) were answered. (Incidentally, the authors misquoted 1(b) with a cut-and-paste typo; the sentence should have read ¨First, isn't it true that $\langle \boldsymbol{k}|\boldsymbol{v}|n'\boldsymbol{k} \rangle_v$ in Eq. (1) contains a $\delta_{\boldsymbol{k}\boldsymbol{ k}'}$?".) If we accept from other arguments that $\langle n \boldsymbol{k}|\boldsymbol{v}|n'\boldsymbol{k}'\rangle_v$ is diagonal in $\boldsymbol{k}$ (e.g., using that $\boldsymbol{v}$ is a periodic operator), and in view of the $\omega_{n\boldsymbol{k},n'\boldsymbol{k}}$ prefactor, we only have to evaluate $\langle \boldsymbol{k}|\boldsymbol{r}|n'\boldsymbol{k} \rangle_v$ for $\boldsymbol{k}=\boldsymbol{k}'$ and $n\neq n'$. Here I adopt the notation that $\tilde{\psi}$ are the wave functions normaized to volume $L=Na$ while $\psi=\sqrt{N}\tilde{\psi}$ are normalized to a primitive cell. Then in 1D

$$ \begin{split} \langle \boldsymbol{k}|\boldsymbol{r}|n'\boldsymbol{k}\rangle_v&=\int_{0}^{Na}dx \ x \tilde{\psi}_{nk}^{\ast}(x) \tilde{\psi}_{n'k}(x) \\ &=\frac{1}{N}\int_{0}^{Na}dx \ x \psi_{nk}^{\ast}(x) \psi_{n'k}(x) \\ &=\frac{1}{N}\sum_{j=0}^{N-1}\int_{ja}^{(j+1)a}dx \ x \psi_{nk}^{\ast}(x) \psi_{n'k}(x) \\ &=\frac{1}{N}\sum_{j=0}^{N-1}\int_{0}^{a}dx (x+ja) \psi_{nk}^{\ast}(x) \psi_{n'k}(x) \end{split} $$

where I have used that

$$ \psi^{\ast}_{nk}(x)\psi_{n'k}(x) $$

is periodic under $x \rightarrow x+a$. Then

$$ \begin{split} \langle n\boldsymbol{k}|\boldsymbol{r}|n'\boldsymbol{k} \rangle_v=& \Big( \frac{1}{N} \sum_{j=0}^{N-1}\Big) \int_{0}^{a}dx \ x \psi_{nk}^{\ast}(x) \psi_{n'k}(x) \\ & \Big( \frac{a}{N} \sum_{j=0}^{N-1}j\Big) \int_{0}^{a}dx \psi_{nk}^{\ast}(x) \psi_{n'k}(x) \end{split} $$

The second integral vanishes by orthogonality of the wave functions and the first factor in the first term is unity, so

$$ \langle n\boldsymbol{k} |\boldsymbol{r}|n'\boldsymbol{k}\rangle_v=\int_{0}^{a}dx \ x \psi_{nk}^{\ast}(x) \psi_{n' \boldsymbol{k}}(x) $$

I believe another argument along these lines allows to show that $C_{nk,n'k}$ can be similarly writen in terms of the boundaries of the primitive cell. This kind of development is that I had in mind when I wrote ¨I suspect Eqs. (11-12) can be recast in terms of integrals over a the interior and the boundary of a single primitive cell, with wave functions normalized to the unit cell''. I leave it as an option for the authors to discus this somehow in their revised manuscript.

We thank the referee again for providing such detailed feedback. First, let us write the general version of Eq. (3) in this response for the general, nondiagonal case, (we use $\tau$ as the integration variable in the unit cell)

$$ \begin{split} \langle n\boldsymbol{k}|\boldsymbol{r}|n'\boldsymbol{k}' \rangle_v=& \delta_{kk'} \int_{\tau_0}^{\tau_0+a}d\tau \ \tau u_{nk}^{\ast}(\tau) u_{n'k}(\tau) \\ & \Big( \frac{a}{N} \sum_{j=0}^{N-1}j \Big) \int_{\tau_0}^{\tau_0+a}d\tau \psi_{nk}^{\ast}(\tau) \psi_{n'k'}(\tau). \end{split} $$

Note that there is still a dependance with $\tau_0$, that can be related to the original $x_0$ origin for the integration volume $v$. Eq. (3) is a special case of this with $\tau_0=0$. Note also that the second term only includes the orthogonality condition for the $k$-diagonal case, in which we obtain the normalization condition for the periodic part: \begin{equation} \int_{\tau_0}^{\tau_0+a}d\tau \psi_{nk}^{\ast}(\tau) \psi_{n'k}(\tau)= \int_{\tau_0}^{\tau_0+a}d\tau u_{nk}^{\ast}(\tau) u_{n'k}(\tau) =\delta_{nn'}.
\end{equation} With this, we see that the simple form of Eq. (3) can only be obtained for the $k-$diagonal case, in which the position matrix elements still depend on the origin, as in the general case covered with Eqs (11) and (12) in the manuscript. The advantage of this reduction may be in terms of computational terms, where it could be desired to have a smaller integration grid. As nothing fundamental is added in this step, we prefer to avoid including this discussion in the manuscript.

Login to report or comment