Recently I've been studying quantum dynamics with Sakurai's modern quantum mechanics, but I am confused with why the time evolution operator is written as

$$U(t,t_0)=\exp\left[\frac{-iH(t-t_0)}{\hbar}\right]$$ for time-independent Hamiltonian, while $$U(t,t_0)=\exp\left[-\left(\frac{i}{\hbar}\right)\int_{t_0}^tdt'H(t')\right]$$ for time-dependent (commuting case). My thought is that, since we can Taylor expand the wave function at $t=t_0$

$$\begin{align}\psi(x,t)&=\sum_{n=0}^\infty \frac{1}{n!}\left(\left(\frac{\partial}{\partial t}\right)^n\psi(x,t)\bigg|_{t=t_0}\right) (t-t_0)^n\\ &= \sum_{n=0}^\infty \frac{1}{n!}\left(\left(\frac{-iH}{\hbar}\right)^n\psi(x,t)\right)(t-t_0)^n\\ &=e^{\frac{-iH(t-t_0)}{\hbar}}\psi(x,t) \end{align} $$

we only need to know the value of $H$ at $t=t_0$. If this is true, then $U(t,t_0)=\exp\left[\frac{-iH(t-t_0)}{\hbar}\right]$ should hold whether $H$ is time-dependent or not. What have I done wrong here?

  • 1
    $\begingroup$ Where are you getting that the $n^\text{th}$ derivative is $(-i H / \hbar)^n$? $\endgroup$
    – DanielSank
    Commented Oct 27, 2015 at 7:48
  • $\begingroup$ Since $H\psi=i\hbar\frac{\partial}{\partial t}\psi$, we have $H=i\hbar\frac{\partial}{\partial t}$, then $\left(\frac{-iH}{\hbar}\right)^n=\left(\frac{\partial}{\partial t}\right)^n$. Correct me if I'm wrong, thanks. $\endgroup$
    – Rick Pan
    Commented Oct 27, 2015 at 7:56
  • $\begingroup$ @Rick Pan, the equation for derivative $\partial_t \psi =1/(i\hbar)H\psi$ holds only for $\psi$. It does not necessarily hold for its derivatives. $\endgroup$ Commented Oct 27, 2015 at 8:00
  • $\begingroup$ @JánLalinský Thanks for the comment. If we assume that the eigenstates of $H$ forms a complete basis, then we should be able to expand its derivatives with the complete basis. If this is the case, the relation should also hold for its derivatives right? $\endgroup$
    – Rick Pan
    Commented Oct 27, 2015 at 8:24
  • $\begingroup$ @RickPan, no, the relation does not hold because of the reason yuggib pointed out - the expansion coefficients are functions of time as well. Canonical momentum operator $p_x$ is always expressible as $-i\hbar\partial_x$, whatever the function $\psi$ may be, but the Hamiltonian is not always given by $i\hbar\partial_t$; it only holds for special functions - solution to time-dependent Schroedinger equation. $\endgroup$ Commented Oct 27, 2015 at 18:51

1 Answer 1


The first remark is that, at a rigorous level, you are not allowed to do all those manipulations freely. However, let's suppose for a moment that you would, for everything is extremely regular and well-behaved.

The (omitted) starting hypothesis is that $$i\partial_t\psi(t)=H(t)\psi(t)\; .$$ If we iterate the derivation, we do not get simply $H(t)^2\psi(t)$, but rather (this is a simple application of the product rule, that actually works also in this case) $$(i\partial_t)^2\psi(t)=i\dot{H}(t)\psi(t)+H(t)^2\psi(t)\; .$$ As we can easily see, this is where the OP's argument goes wrong, since the derivative of $H(t)$ does not vanish in general for time dependent operators.

I want to remark again, however, that this is not the proper way of dealing with these type of time-dependent equations. The proper way is, however, very complicated and it requires a lot of advanced functional analysis. If you are curious, the most common method is due to T.Kato, and can be found e.g. in this book.

  • $\begingroup$ Thanks for the answer, but I'm curious that if we expand $\partial_t \psi$ with eigenstates of $H$ (which is usually assumed to form a complete basis), don't we get $H^2\psi=(i\partial_t)^2 \psi$? $\endgroup$
    – Rick Pan
    Commented Oct 27, 2015 at 9:36
  • $\begingroup$ You could only use eigenstates for a fixed $t$, but they would not be eigenstates, in general, for another $t'\neq t$ (since $H(t')\neq H(t)$), and as well not eigenstates for $\dot{H}(t)$. Really, this is not the good way of thinking about that... ;-) $\endgroup$
    – yuggib
    Commented Oct 27, 2015 at 9:40
  • $\begingroup$ Thanks for the reply. It is true that eigenstates can change with respect to $t$, but I'm expanding the series at a fixed $t$, how is that not applicable? $\endgroup$
    – Rick Pan
    Commented Oct 27, 2015 at 10:01
  • $\begingroup$ Let $\psi_{n,t}$ be an eigenvector of $H(t)$; however in the expansion you have $\dot{H}(t)\psi_{n,t}$, and $\psi_{n,t}$ is not an eigenvector for it. $\endgroup$
    – yuggib
    Commented Oct 27, 2015 at 10:04
  • 1
    $\begingroup$ @RickPan Yes, for in general for a time dependent $H(t)$, the coefficients $c_k$ would depend on time as well as the functions $\psi_k$...believe me, this is not true/correct ;-) $\endgroup$
    – yuggib
    Commented Oct 27, 2015 at 12:42

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.