I'm trying to get some reasonable measure of the entropy of a system modelled by the truncated Wigner method. The Wigner function contains all the information about a density matrix. So, I figure it should be possible (but not necessarily easy) to calculate the von Neumann entropy: $$S=-\text{Tr}\left\{\hat{\rho}\log(\hat{\rho})\right\}$$ using a Wigner function $W(x,p)$. Looking at this document (section 2), it seems like the linear entropy (a linear expansion of von Neumann entropy about a pure state) can be calculated readily from the quasidistribution itself: $$S_2=1-\text{Tr}\left\{\hat{\rho}^2\right\}=1-(2\pi\hbar)\int dx\int dp \left(W(x,p)\right)^2$$ Suppose we apply the truncated Wigner method, in which one derives an equation of motion for $W(x,p)$, truncates higher-order terms based on a physical argument in order to get the equation of motion into Fokker-Planck form, and then stochastically unravels this Fokker-Planck equation. The observables of the system can be readily recovered from averages over the stochastic fields. But neither of these entropies are observables, even if they can be calculated from the full density matrix. Is there some method by which one or both of these entropies could be calculated from the stochastic fields in truncated Wigner?


This might be useful. Indeed, the impurity you write, $$ S_2=1-h\langle W\rangle= \int\! dx dp ~(W-hW^2) $$ is exact, by an exceptional feature of the Wigner representation, and it vanishes only for pure states, thus quantifying departures from them.

It is the leading h expansion of the full quantum entropy, the exact transcription of the von Neumann entropy in phase space, $$ S=- \langle \ln ~\hat \rho \rangle= -\int \! dx dp ~ W \ln _\star (hW)= \sum_{n=1}^\infty \frac{1}{n}\langle (1-hW)^n_\star\rangle , $$ using a common notation for the star-logarithm and star-powers. (Only the leading, n =1, power lacks a star the the concomitant extra h.)

A consistent systematic approximation in h is generally hard to devise. However, if you are willing to sacrifice some hs in your expansion, but not others, you could try the Wehrl entropy, living heuristically--as long as you were aware. Some like it on account of its simplicity, but the capricious Gaussian smoothing underlying it hides important features of the distribution, for which few practitioners have any sound intuition at all.

The talk linked above corresponds to this paper of mine.

  • $\begingroup$ Forgive my ignorance, but how is this likely to be easier to calculate from the stochastic unravelling in truncated Wigner? $\endgroup$ – user502382 Mar 27 at 5:33
  • $\begingroup$ Detail your specific difficulty? Your last paragraph is schematic at best. If you have found the W s to arbitrary accuracy, you plug them in consistently to the above expressions. It is not particularly clear what you are doing, how you are doing it, and what blocks you. $\endgroup$ – Cosmas Zachos Mar 27 at 14:59
  • $\begingroup$ Barking up this tree? $\endgroup$ – Cosmas Zachos Mar 27 at 15:47

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.