2
$\begingroup$

In Statistical Physics (Vol 5 of Landau's books) section 11, Landau derives an important relation: $\overline{\frac{\partial E(p, q;\lambda)}{\partial \lambda}} = \left(\frac{\partial E}{\partial \lambda}\right)_S$.

In his derivation, there is an important step: enter image description here

He argues that averaging over statistical distribution and differentiating with respect to time can be interchanged.

But since formally, $E = \int f(p, q; \lambda(t)) E(p, q; \lambda(t)) d^3 p d^3 q$, where $f(p, q; \lambda(t))$ is the statistical distribution, it seems that the interchange is illegal? Or we must assume that under the time $dt$, the change in $f(p, q; \lambda(t))$ is small and can be ignored?(Is it a consequence of adiabatic condition?)

Am I not getting the point of adiabatic condition?

$\endgroup$

1 Answer 1

4
$\begingroup$

Note that the derivative with respect to $\lambda$ in Landau is being taken at fixed entropy $S$. This means that the probability distribution is not changing. This will happen if the change is slow eneough that the quantum adiabatic theorem holds --- meaning that the system stays in the same quantum state.

When you include the change in ditribution this gives rise to the $TdS$ part of $$ dU = T\,dS-P\, dV. $$

For example, given a Hamiltonian $\hat H$ whose eigenstates $|n\rangle$ have energy $E_n$, the canonical ensemble assigns a probability
$$ \rho_n = \frac 1 Z e^{-\beta E_n} $$ to the state $|n\rangle$. Here $\beta=1/T$ and $$ Z= \sum_n e^{-\beta E_n}={\rm tr}\left\{e^{-\beta H}\right\} $$ is the partition function. The thermodynamic internal energy $U$ is then $$ U= \frac 1 Z \sum_nE_n e^{-\beta E_n}. $$

For any probability distribution $\rho_n$, the associated entropy is defined to be $$ S= - \sum_n \rho_n \ln \rho_n $$ For the canonical distribution, this leads to $$ S= \frac 1 Z \sum_n \left( \frac{E_n}{T} +\ln Z\right) e^{-\beta E_n}= \frac U T +\ln Z. $$ A change in the distribution leads to a change in the entropy of $$ \delta S = -\sum_n (\ln \rho_n) \delta \rho_n. $$ Here we have use the fact that $\sum_n \rho_n=1$ implies that $\sum_n\delta \rho_n=0$.

Let the Hamiltonian depend on a parameter $\lambda$ (the system volume, for example). The energy levels $E_n(\lambda)$ then also depend on $\lambda$, and we use the symbol ``$P$'' to stand for the generic thermodynamic average $$ P= -\left\langle \frac{ d\hat H}{d\lambda}\right\rangle \nonumber\\ =- \frac 1 Z \sum_n \left( \frac{d E_n}{d\lambda}\right) e^{-\beta E_n}.\nonumber $$

Thus $$ \delta U = \frac 1 Z \sum_n \left(\frac{d E_n}{d\lambda}\right)\, \delta \lambda e^{-\beta E_n} + \sum_n E_n\, \delta\left(\frac 1 Z e^{-\beta E_n}\right)\nonumber\\ = -P\, d\lambda + T \sum_n \left( \frac{E_n}{T} +\ln Z\right) \delta \left(\frac 1 Z e^{-\beta E_n}\right)\nonumber\\ = -P\, d\lambda + T \sum_n (-\ln \rho_n) \delta \rho_n\nonumber\\ = -P\, d\lambda +T dS.\nonumber $$ In passing from the first line to the second, we are allowed to insert the ``$T\ln Z$'' because $\sum \delta\rho_n=0$.

$\endgroup$

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