I'm studying the phase space formulation of quantum field theory for my final degree project, and I have found two very different ways to construct the Wigner funtion. In the first method, a phase space is constructed as the tensor product of the Hilbert space which represent fields defined in configuration space. More precisely, the Minkowski space is equipped with a cotangent bundle $T^{*}\mathbb{M} \equiv \Gamma$ with coordinates $(q^{\mu},p_{\mu})$, with $\mathcal{H}(\Gamma)$ being the Hilbert space of square-integrable functions $\psi(q,p)$. Then, the operators $Q^{\mu} = q^{\mu}+\frac{i\hbar}{2}\partial_{p_{\mu}}$ and $P_{\mu} = p_{\mu}-\frac{i\hbar}{2}\partial_{q^{\mu}}$ are introduced, which provide $\mathcal{H}(\Gamma)$ with a non-conmutative algebra. Lastly, the Wigner function is defined as $f_{W} = \phi(q,p)\star \phi^{\dagger}(q,p)$, where $\phi(q,p)$ is the solution to the corresponding equation (Klein-Gordon, Dirac, etc.) written in terms of $Q^{\mu}, P_{\mu}$. In summary, this approach uses the phase space as the labels where the different fields are defined, and constructs a "phase space wave function" which, from what I have read, is not very conventional. Of course, I have oversimplyfied the theory and skipped over numerous details which provide mathematical sense. This method was developed by Amorim et al. and is explained in their article Quantum Field Theory in Phase Space link to the article. The second method is the one developed by Mrówczy´nski et al. in their article Wigner Functional Approach to Quantum Field Dynamics, link to the article. In this one, the fields $\Phi(x)$ are defined in each point of the Minkowski space (like a standard field), and not in the cotangent bundle $T^*\mathbb{M}$. However, the Wigner function (or rather, the Wigner functional) is defined as $$W[\Phi(x),\Pi(x)] = \int \mathcal{D}\phi'(x)\ exp\left[-i\int dx \Pi(x)\phi'(x)\right] \langle\Phi(x)+\frac{1}{2}\phi'(x)|\hat{\rho}(t)|\Phi(x)-\frac{1}{2}\phi'(x)\rangle,$$ i.e. the Wigner functional depends on the field and its conjugate momentum, which contitute the "phase space" in this theory. To summarize, in the first method, we have the classical phase space, which constitutes the space in which the fields are defined and, in the second one, the phase space aspect is represented by the field operators and their conjugate momenta. As you can see, these two approaches are very different. My questions are:

  • What is the physical meaning behind the difference in how phase space is defined in each theory?
  • Are these methods used in different areas of Theoretical Physics?
  • Are both equally valid?

I know these topics are not known by many people, but I would appreciate if someone with experience in these field helped me out, since I am a bit lost and I need a little context.

  • 1
    $\begingroup$ Might peruse this and this. They are simpler than the second one. $\endgroup$ Commented Mar 1, 2021 at 22:40
  • $\begingroup$ @CosmasZachos Thank you very much for the references. Mrówczy´nski's article was actually easy to understand for me since it used the QFT notation I am accustomed to from my degree. Any thoughts on the first mehod though? I recall you saying it's not very convenient to define a phase space wave function in another post, so I was curious about this one. $\endgroup$
    – Marcosko
    Commented Mar 2, 2021 at 7:58
  • 1
    $\begingroup$ Unfortunately, I cannot access that paper, not even through work. The overall method of M Ban, Torres-Vega-Frederick, etc on symplectic Hilbert space adopted by Oliveira et al consists of plugging the dyadic density matrix with a reference state, cf. Chruscinski & Mlodawski. I have no clue, however, why such a twisted rewriting of the standard formulation could lead to insights; it has let to pretty bad mistakes and misconceptions in the past. (You really don't want to know.) The second approach is straightforward, albeit non-covariant. $\endgroup$ Commented Mar 2, 2021 at 14:41
  • $\begingroup$ @CosmasZachos that makes sense, since the article that developes the first method goes through a significant amount of trouble in order to construct the algebra in which the phase space wave function (i.e. the field) is defined. I'll look more into the second approach although, as yuggib pointed out, there is the problem of unambiguously determining the measure of the Wigner functional. $\endgroup$
    – Marcosko
    Commented Mar 2, 2021 at 19:44
  • $\begingroup$ If one had a rubber wrench, one would apply it on several gadgets to see if it could work.... For the 2nd approach, there are books, and a popular review by Jackiw on wave functional QFT... $\endgroup$ Commented Mar 2, 2021 at 19:46

1 Answer 1


There are some problems in trying to define a quasi-probability distribution in phase space for quantum fields, à la Wigner.

The foremost complication is given by the lack of a universal reference measure in an infinite dimensional vector (e.g. Hilbert) space. For finite dimensional vector spaces, there is the Lebesgue measure, and thus it makes sense to define a probability (or quasi-probability) distribution with respect to the Lebesgue measure. The Wigner function provides such a way, consistently with taking the classical limit (in the limit in fact the Wigner function becomes a true probability distribution).

The second method cited by the OP is trying to do something of this kind, however by using a mathematically ill-defined or at least ambiguous object in the form of the reference path measure "$\mathcal{D}\phi(x)$. What is the measure chosen? The Gaussian (free) one? Some equilibrium measure for an interacting field theory? These measures behave very differently, and are very tricky to define even in the simplest cases. Even if the measure is unambiguously chosen, it would not be good for all quantum states: roughly speaking, the gaussian free measure is not suited to describe interacting states (e.g. the interacting vacua).

There is however a way to define an object similar, at least in spirit, to the Wigner function also for quantum fields. A quantum state (be it of a theory with finitely or infinitely many degrees of freedom) is a noncommutative probability. The Wigner way of characterizing this noncommutativity is by defining a phase space distribution that is not necessarily positive. Another way, that can be generalized more easily to bosonic theories, is to look at Fourier transforms of measures. Let us remain finite dimensional for the moment. The well-known Bochner theorem states that there is a bijection between finite measures and the continuous functions of positive type whose value in zero coincides with the total mass of the measure. The bijection map is given by the Fourier transform (here written directly in phase space terms for convenience): $$ \hat{\mu}(\xi,\eta)= \int_{\mathbb{R}^{2d}} e^{i(\xi\cdot q + \eta\cdot p)} \mathrm{d}\mu(q,p)\;. $$ If one now quantizes the character $e^{i(\xi\cdot q + \eta\cdot p)}$ to an operator, and replaces the integral with a trace w.r.t. a quantum state $\rho_{\hbar}$ (density matrix) obtains a noncommutative Fourier transform: the quantization of the character is the so-called Weyl operator $W_{\hbar}(\xi,\eta)$, thus $$\hat{\rho}_{\hbar}(\xi,\eta)=\mathrm{tr}[\rho_{\hbar} W_{\hbar}(\xi,\eta)]\; .$$ It turns out that the noncommutative Fourier transform is very nice, in particular behaves very similarly to the Fourier transform of a measure: it is continuous, its value in zero is the trace of the state, and it is of almost positive type. With respect to positive type functions, there is a quantum correction in the form of relative phases of the type $e^{i\hbar \sigma(\xi,\eta)}$, where $\sigma(\xi,\eta)$ is the canonical symplectic form associated to the (dual of the) phase space. This is analogous to the non-positivity of the Wigner function, and the appearence of the symplectic form is a clear signature of the phase space structure. Finally, taking suitably the limit $\hbar\to 0$, there exists a classical measure $\mu$ such that the noncommutative Fourier transform converges to the classical one: $$\lim_{\hbar\to 0} \hat{\rho}_{\hbar}(\xi,\eta) = \hat{\mu}(\xi,\eta)\; .$$ Therefore, the quantum description correctly reduces to the classical one in the limit $\hbar\to 0$.

The advantage of this formulation with respect to Wigner's one is that the extension to infinitely many degrees of freedom is possible, without (too many) difficulties. Bosonic quantum fields are indeed defined starting from the Weyl operators, and their canonical commutation relations. In QFTs it is customary to take the complexification of the phase space as the reference space (roughly speaking, taking $z=q+ip$ and $\bar{z}=q-ip$ as canonical variables). In addition, as we have seen the character is defined with respect to the dual space with variables $(\xi,\eta)$ (let us call $\zeta = \xi +i\eta$ its complexification). However, while in finite dimensions any vector space is isomorphic to its dual this is not true for infinite dimensional spaces (in addition, there are several possible dual spaces). To distinguish between the phase space of fields, and its dual space, let us call $z$ the (classical) field and $\zeta$ a test function. So the algebra of Weyl operators, encoding the canonical commutation relations, is given by identifying the space of test functions. Such space shall be infinite dimensional, and endowed with a symplectic form that enters in the CCR (as it entered in the noncommutative Fourier transform). This symplectic form is usually defined starting from the canonical poisson bracket between a field and its momentum. In the complexified version, the standard choice is to define an inner product on the infinite dimensional complex space of test functions, from which the symplectic form is obtained by taking the imaginary part of the inner product (and considering the complex vector space as a real space with "twice many" variables). A standard example is to take the test functions in the Schwartz space of rapidly decreasing functions $\mathscr{S}$ (with inner product given by the standard $L^2$ inner product), and the tempered distributions $\mathscr{S}'$ as the (complexified) phase space of classical fields. This choice amounts to the canonical poisson bracket $$\{\bar{z}(x),z(y)\}= i \delta(x-y)\; .$$

Now, given a bosonic quantum field state $\rho_{\hbar}$, the noncommutative Fourier transform is again defined as $$\hat{\rho}_{\hbar}(\zeta)= \mathrm{tr}[\rho_{\hbar} W_{\hbar}(\zeta)]\;,$$ the same as for finite dimensions. Again, a noncommutative form of the Bochner theorem holds (with some technical restrictions): there is a bijection, given by the noncommutative Fourier transform, between regular$^{1}$ states and complex-valued functionals of test functions that are continuous when restricted to finite dimensional subspaces, whose value in zero coincides with the trace of the state, and that are of almost positive type. An analogous theorem, in the commutative case, holds for so-called cylindrical measures (an object that is more general than measures, and exists only in infinite dimensional spaces; this object is the correct classical counterpart of a quantum state in QFT). So again, the parallel between classical and noncommutative probability is done effectively. Furthermore, as in the finite dimensional case, taking suitably the limit $\hbar\to 0$ there exists a cylindrical measure $m$ such that $$\lim_{\hbar\to 0} \hat{\rho}_{\hbar}(\zeta) = \hat{m}(\zeta)\; .$$

Let me conclude with a small remark on fermionic theories. Since for fermions the Weyl operators (exponentials) are not available, one should use the noncommutative version of the moments of a probability distribution (essentially the trace of arbitrary products of smeared fields). In addition, for fermions there are currently no results available concerning the limit $\hbar\to 0$, and recovering a classical field theory.


Noncommutative Fourier Transform and Bochner Theorem:

  1. I. Segal. Foundations of the theory of dyamical systems of infinitely many degrees of freedom. II. Canad. J. Math., 13:1–18, 1961.
  2. M. Merkli. The ideal quantum gas, in Open Quantum Systems I. Lecture Notes in Mathematics. 2006

Classical limit of states in bosonic QFTs: limit of Fourier transforms:

  1. Z. Ammari, F. Nier. Mean field limit for bosons and infinite dimensional phase-space analysis. Ann. Henri Poincaré, 9(8):1503– 1574, 2008.
  2. M. Falconi. Cylindrical Wigner Measures. Documenta Math. 23:1677–1756, 2018.

1: Regular states are defined rather abstractly, but are essentially the only type of physically relevant quantum states, I won't make a lengthy and probably unnecessary discussion about it here.

  • $\begingroup$ For the case of a bosonic field, (q,p) would correspond to $(\phi(x),\pi(x))$, right? Also, can you link some papers where this formalism is developed? I think I understand almost everything, but it would be useful to see it applied to quantum fields. $\endgroup$
    – Marcosko
    Commented Mar 2, 2021 at 13:48
  • 1
    $\begingroup$ Indeed $(q,p)$ would correspond to the field and its momentum, and $(z,\bar{z})$ essentially to the annihilation and creation operators. I will post references later when I am at the computer. $\endgroup$
    – yuggib
    Commented Mar 2, 2021 at 17:30
  • 1
    $\begingroup$ @Marcosko I added some references. Let me mention that semiclassical analysis for infinite dimensional bosonic systems, using measures and their Fourier transforms, is a very active research topic in mathematics and especially mathematical physics nowadays. Therefore, the list of references is very far from being exhaustive, but it should give you a flavor of it in case you are interested. $\endgroup$
    – yuggib
    Commented Mar 3, 2021 at 22:34

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.