Consider the following derivation in David Vanderbilt's book "Berry Phases in Electronic Structure Theory - Electric Polarization, Orbital Magnetization and Topological Insulators" (2018, Cambridge University Press (page 100).

The wave function for the adiabatic approach is as follows: \begin{equation} |\psi(t)\rangle=e^{i\phi(\lambda(t))}e^{-i\gamma(t)}|n(t)\rangle \tag{1}\label{1} \end{equation}

where $e^{i\phi(\lambda(t))}$ is the geometric phase or Berry phase and $e^{-i\gamma(t)}$ is dynamic phase.

Berry phase has the following mathematical expression: \begin{equation} \phi(t)=\int_{\lambda(0)}^{\lambda(t)}A_n(\lambda)d\lambda \tag{2}\label{2} \end{equation} where $A_n$ is Berry connection.

Dynamic phase has the following mathematical expression: \begin{equation} \gamma(t)=\frac{1}{\hbar}\int_{0}^{t}E_n(t')dt' \tag{3}\label{3} \end{equation}

Expanding equation number (1) with first order $\lambda$ terms:

\begin{equation} |\psi(t)\rangle=e^{i\phi(\lambda(t))}e^{-i\gamma(t)}\big[|n(\lambda)\rangle+\dot{\lambda}|\delta n(t)\rangle\big] \tag{4}\label{4} \end{equation}

Equation (4) solves the time-dependent Schrödinger equation to order zero in $\dot{\lambda}$, but we now require that it should also do so at first order.

For this purpose we can discard terms that scale like $\ddot{\lambda}$ or $\lambda^2$. You get: \begin{equation} (E_n-H_{\lambda})|\delta n\rangle=-i\hbar (\partial_{\lambda}+iA_n)|n\rangle. \tag{5}\label{5} \end{equation}

which can also be written like:

$$\begin{equation} (E_n-H_{\lambda})|\delta n\rangle=-i\hbar \mathcal{Q}_{n}|\partial_{\lambda}n\rangle \tag{6}\label{6} \end{equation} $$

where $\mathcal{Q}_n=1-|n\rangle\langle n|$.

How did they get equation (5) and (6) from equation (4)?

  • 1
    $\begingroup$ I don't know what all the symbols mean here, and don't have that book, so can't look up what they mean easily. But Eq. 6 seems to be somewhat common: I've been working today on describing p-DMRG for my types of DMRG question, and you can see that Eq. 9 in this: arxiv.org/abs/1803.07150 is very similar to Eq. 6 in your question. Eq. 5 is similar. Maybe we just need to know what all these symbols mean? $\endgroup$ Jul 29, 2020 at 20:17
  • $\begingroup$ Eq. 5 is a direct application of the expression for first-order correction from perturbation theory. I don't totally understand eq. 6, but it is somehow leveraging the completeness of the basis. In particular, I just don't know what the notation $|\partial_{\lambda}n\rangle$ means. $\endgroup$
    – jheindel
    Sep 1, 2020 at 20:59
  • $\begingroup$ @jheindel: We have a policy not to ask 2 questions in one post, and combined with the fact that the notation was not made clear, the question was closed. However, I have deleted the question about Eq. 6, so if you can write an answer about Eq. 5 only, it would be very appreciated as it would help clean up our unanswered queue for the upcoming new year! $\endgroup$ Dec 31, 2020 at 16:53
  • 1
    $\begingroup$ Hi @jheindel ! Since this is now one of our longest standing unanswered questions, and has been left up for over 6 months, I wonder if you might be able to spare a couple minutes to answer the part about Eq. 5? The part about Eq. 6 was deleted. I'm trying to clear up the unanswered queue! $\endgroup$ Mar 19, 2021 at 15:56

1 Answer 1


The TD-Schrodinger equation is $$H(\lambda)|\psi(t)\rangle=i\hbar\frac{\partial}{\partial t}|\psi(t)\rangle$$

So, we just need to plug in \eqref{4} and remove all terms that aren't first-order in $\dot\lambda$. Let's start with the left-hand side, since its a little simpler.

$$\begin{align}H(\lambda)|\psi(t)\rangle&=e^{i\phi(\lambda(t))}e^{-i\gamma(t)}\bigg[H(\lambda)|n(t)\rangle+\dot\lambda H(\lambda)|\delta n(t)\rangle\bigg]\\ &\overset{\text{First order}}{\to}e^{i\phi(\lambda(t))}e^{-i\gamma(t)}\bigg[\dot\lambda H(\lambda)|\delta n(t)\rangle\bigg]\end{align}$$

Now for the right side, which is a bit more involved: $$\begin{align} i\hbar\frac{\partial}{\partial t}|\psi(t)\rangle&=i\hbar\frac{\partial}{\partial t}e^{i\phi(\lambda(t))}e^{-i\gamma(t)}\bigg[|n(t)\rangle+\dot\lambda|\delta n(t)\rangle\bigg]\\& =i\hbar \Bigg[iA_n(t)|\psi(t)\rangle-iE_n|\psi(t)\rangle+e^{i\phi(\lambda(t))}e^{-i\gamma(t)}\bigg[\frac{\partial}{\partial t}|n(t)\rangle+\ddot\lambda|\delta n(t)\rangle+\dot\lambda\frac{\partial}{\partial t}|\delta n(t)\rangle\bigg]\Bigg]\\ &\overset{\text{First order}}{\to}i\hbar e^{i\phi(\lambda(t))}e^{-i\gamma(t)}\dot\lambda \Bigg[iA_n(\lambda)|n(t)\rangle-iE_n|\delta n(t)\rangle\rangle+\frac{\partial}{\partial \lambda}|n(\lambda)\rangle\Bigg] \end{align}$$

As mentioned in the text just before Eqn 3.53, we can make the substitution $\frac{\partial}{\partial t}|n(t)\rangle=\dot\lambda\frac{\partial}{\partial \lambda}|n(\lambda)\rangle$ using the chain rule to express the derivative in terms of $\dot\lambda$. The inconsistent switching between $|n(t)\rangle$ and $|n(\lambda(t))\rangle$ can make the derivation a bit confusing, but these expression are defined to be equivalent and you can verify that the equations are consistent with either expression for the functional form of $|n\rangle$.

Combining these expressions, removing the common factor of $\dot\lambda e^{i\phi(\lambda(t))}e^{-i\gamma(t)}$, and separating $|n\rangle$ and $|\delta n\rangle$ terms, we obtain:

$$(H(\lambda)-E_n)|\delta n\rangle=i\hbar\bigg(\frac{\partial}{\partial \lambda}+iA_n\bigg)|n\rangle$$

Which is just \eqref{5} with the sign of both sides flipped. To get \eqref{6} just requires using the definition $A_n(\lambda)=\langle n(\lambda)|i\frac{\partial}{\partial \lambda}|n(\lambda)\rangle$. We can then rearrange the right hand side to be in terms of $\frac{\partial}{\partial \lambda}|n\rangle$, which you can succinctly express by introducing $\mathcal{Q}_n=1-|n\rangle\langle n|$

$$(H(\lambda)-E_n)|\delta n\rangle=i\hbar\mathcal{Q}_n\frac{\partial}{\partial \lambda}|n\rangle$$

  • $\begingroup$ Wow! What a complete answer! Thanks for the effort Tyberius ♦. $\endgroup$ Jul 13, 2021 at 16:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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