Consider a pair of equal mass (set $m=1$) harmonic oscillators of frequencies $\omega_1$ and $\omega_2$ coupled through a positive constant $g$. The Hamiltonian is


It is then pretty straightforward to diagonalize the system into normal modes and write (with $\hbar=1$)

$$H=\omega_-\left(a_-^\dagger a_-+\frac{1}{2}\right)+\omega_+\left(a_+^\dagger a_++\frac{1}{2}\right).$$

The expressions for the normal modes are ridiculously long and nasty so I won't bother writing them here. But the nice part is that $a_\pm$ commute, and therefore are easy to trace over independently. We then write down the partition function

$$Z=\mathrm{tr}\exp(-\beta H),$$

And in principle compute every expectation value imaginable using $Z$.

And so I did, and it worked out for a while. I then decided to compute moments. In particular, $\langle q_1^2\rangle$. My intuition then pointed me to

$$\frac{\partial Z}{\partial\omega_1},$$

Since $\omega_1$ shows up with $q_1^2$ in the Hamiltonian. But when I compute that, I don't get what I wanted. Instead, using some operator identities that I can elaborate on if necessary, I get

$$\frac{\partial Z}{\partial\omega_1}=-\beta\omega_1\mathrm{tr}\left(\left(q_1^2+i\frac{\beta}{2}(q_1p_1+p_1q_1)-\frac{\beta^2}{6}p_1^2\right)\exp(-\beta H)\right).$$

i.e. not $\langle q_1^2\rangle$. Can anyone help me extract this expectation value, and maybe other moments from this partition function?

  • $\begingroup$ $q^{2}$ is related to $a$ and $a^{\dagger}$. Would it be easier to first compute the moment of $a$ and $a^{\dagger}$? $\endgroup$
    – K_inverse
    Sep 19, 2018 at 9:19
  • 1
    $\begingroup$ @K_inverse it might be, but I'm not sure how to do that either $\endgroup$ Sep 19, 2018 at 23:30

1 Answer 1


I just realized there's a mistake in my calculations. Not sure where, but I'm sure it's there. The calculation goes as follows. The Free Energy is defined as

$$F=-\frac{1}{\beta}\log\mathrm{tr}\,e^{-\beta H}.$$

Suppose then the Hamiltonian depends on some parameter $\lambda$ as $H(\lambda)$. If we differentiate w.r.t. $\lambda$, making sure to keep temperature constant, we get

$$\frac{\partial F}{\partial\lambda}=-\frac{1}{\beta}\frac{\partial}{\partial\lambda}\log\mathrm{tr}\,e^{-\beta H}. $$ We now need to compute the right-hand side of the equation. The derivative of the log of something is easy:

$$\frac{\partial}{\partial\lambda}\log\mathrm{tr}\,e^{-\beta H}=\frac{\frac{\partial}{\partial\lambda}\mathrm{tr}\,e^{-\beta H}}{\mathrm{tr}\,e^{-\beta H}}$$ The derivative infiltrates the trace and we get

$$=\frac{\mathrm{tr}\,\frac{\partial}{\partial\lambda}e^{-\beta H}}{\mathrm{tr}\,e^{-\beta H}}.$$

Now we just need to compute that operator inside the trace. To do so, note that

$$e^{\beta H}\frac{\partial}{\partial\lambda}e^{-\beta H}=-\int\limits_0^\beta e^{tH}\frac{\partial H}{\partial\lambda}e^{-tH}\,dt. $$

$$\frac{\partial}{\partial\lambda}e^{-\beta H}=-e^{-\beta H}\int\limits_0^\beta e^{tH}\frac{\partial H}{\partial\lambda}e^{-tH}\,dt. $$

This is easy to check by noting that it's trivially true for $\beta=0$ and checking that the derivatives are equal. Now, we plug that into the thing we want to find out:

$$\frac{\partial}{\partial\lambda}\log\mathrm{tr}\,e^{-\beta H}=-\frac{\mathrm{tr}\,e^{-\beta H}\int\limits_0^\beta e^{tH}\frac{\partial H}{\partial\lambda}e^{-tH}\,dt}{\mathrm{tr}\,e^{-\beta H}}. $$

We pull the integral out of the trace and get

$$=-\int\limits_0^\beta \frac{\mathrm{tr}\,e^{-\beta H}e^{tH}\frac{\partial H}{\partial\lambda}e^{-tH}}{\mathrm{tr}\,e^{-\beta H}}\,dt$$ And using the cyclic property and the fact that all the exponentials commute with each other we get $$=-\int\limits_0^\beta\frac{\mathrm{tr}\,e^{-\beta H}\frac{\partial H}{\partial\lambda}}{\mathrm{tr}\,e^{-\beta H}}\,dt=-\beta\frac{\mathrm{tr}\,e^{-\beta H}\frac{\partial H}{\partial\lambda}}{\mathrm{tr}\,e^{-\beta H}}.$$

If we remember that the density operator in a thermal state is $\rho=\frac{e^{-\beta H}}{\mathrm{tr}\,e^{-\beta H}}$, we can write the final version of the expression for the derivative of the log:

$$\frac{\partial}{\partial\lambda}\log\mathrm{tr}\,e^{-\beta H}=-\beta\,\mathrm{tr}\,\rho\frac{\partial H}{\partial\lambda}=-\beta\left\langle\frac{\partial H}{\partial\lambda}\right\rangle. $$

We get the final expression, what we expected from intuition:

$$\frac{\partial F}{\partial\lambda}=\left\langle\frac{\partial H}{\partial\lambda}\right\rangle. $$


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.