Quasi-Probability Distribution Methods

• Quasi-probability distribution methods are frameworks that assign normalized, additive—but potentially negative—distributions to quantum states and observables, retaining noncommutativity and coherence.
• They employ Fourier transforms of characteristic functions and coherent-state kernels to reveal ordering-dependent statistical moments, enabling transparent analyses in quantum optics.
• These methods extend to finite-dimensional systems, polarization tomography, and dynamic scenarios, offering insights into state reconstruction and interference in diverse quantum applications.

Quasi-probability distribution methods are representation schemes that assign additive distributions to quantum states, observables, or dynamical processes on a classical-looking domain—such as phase space, Stokes space, tuples of eigenvalues, or trajectory space—while relaxing positivity, and in some constructions reality, so that noncommutativity, coherence, discrete spectra, or support constraints can be retained (Harada, 2011, Hofer, 2017). The resulting object is typically normalized and reproduces marginals, ordered moments, or reconstruction formulas, but it need not be an ordinary probability measure. In that sense, quasi-probability is not a single formalism but a family of transform methods and symbol calculi, ranging from Cahill–Glauber phase-space distributions and Stratonovich–Weyl kernels to Keldysh contour constructions, full counting statistics, polarization tomography, and finite-state Kirkwood–Dirac representations (Alexanian, 2021, Umekawa et al., 2023).

1. Definitional framework and characteristic-function constructions

A standard route to quasi-probability is to begin with a characteristic function and then Fourier transform it. In the Cahill–Glauber setting for bosonic modes, the $p$-ordered characteristic function is defined by

$$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$

and the corresponding quasidistribution is

$$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$

The parameter $p$ selects the ordering convention: $p=1$ gives the Glauber–Sudarshan $P$-representation, $p=0$ the Wigner function, and $p=-1$ the Husimi $Q$-function. In this formulation, moments are generated either by derivatives of $\chi_a$ or by phase-space integrals against $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$0, and the Gaussian closed forms make thermal broadening, squeezing, displacement, and ordering offsets explicit (Alexanian, 2021).

A more abstract formulation starts from a generalized coherent-state system on a homogeneous phase space $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$1. There one takes the coherent-state expectation

$$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$2

as the generalized Husimi distribution and then generates a full $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$3-parameterized family by integral transforms built from the coherent-state overlap kernel. In this scheme, Husimi distributions are operationally distinguished because they coincide with coherent-state POVM statistics, while Wigner- and Glauber–Sudarshan-type symbols appear as deformations of that basic positive object. The corresponding kernels satisfy generalized Stratonovich–Weyl conditions, and the duality relation

$$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$4

plays the role of the symbol calculus (Harada, 2011).

These two starting points—ordered characteristic functions and coherent-state kernels—already exhibit the central methodological theme of quasi-probability methods: the representation is chosen so that certain algebraic structures become simple, even if positivity is sacrificed.

2. Ordered phase-space families and coherent-state geometry

In quantum optics, the Cahill–Glauber hierarchy is the canonical example of an ordering-sensitive quasi-probability family. For displaced-squeezed bosonic states at finite temperature, the quasidistributions remain Gaussian, with the parameters $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$5, $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$6, and $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$7 encoding anisotropic widths and phase-sensitive correlations. The same formalism yields explicit number moments such as

$$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$8

which separates the thermal, squeezing, displacement, and ordering contributions. The method is efficient precisely because the full state information is compressed into the characteristic function (Alexanian, 2021).

The same logic reappears in coherent-state formulations of optical polarization. Rather than working on the Poincaré sphere from the outset, one may adapt the Cahill–Glauber kernel to a two-mode bosonic field by imposing the index of polarization $$\chi_a(\xi,p)=\mathrm{Tr}\,\rho_a \Big[\exp(\xi B^\dagger-\xi^* B)\exp\!\big(p|\xi|^2/2\big)\Big],$$9 directly at the kernel level. This yields polarization-sensitive quasi-probability distribution functions in ordinary bosonic phase space and, for bi-modal quadrature coherent states, a Gaussian Wigner-function-like expression whose dependence on the polarization parameters is explicit. A plausible implication is that polarization can be treated either as a constrained two-mode phase-space structure or as an $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$0-type observable sector, depending on which representation makes the target calculation more transparent (Singh et al., 2012).

A related but more general point is that quasi-probability families often interpolate between smoothing and desmoothing operations. In the coherent-state-based scheme, moving from one $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$1-parameter to another is realized by a kernel transform generated by coherent-state overlaps; in the Cahill–Glauber language this is the familiar $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$2-to-Wigner-to-$$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$3 hierarchy. This suggests that ordering is not merely a technical label but a way of tuning how much noncommutative information is retained in the representation (Harada, 2011).

3. Finite-dimensional, group-theoretic, and discrete phase spaces

For finite-dimensional quantum systems, quasi-probability methods split into several distinct lines. One line is based on the Kirkwood–Dirac distribution. Within the general quasi-classicalization framework, the KD distribution

$$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$4

is singled out by two properties in low dimensions: it is supported only on tuples of actual eigenvalues, and for two-state and three-state systems it can distinguish all states from only two noncommuting observables. In the qubit case, the paper proves that the imaginary part is essential: real-valued two-observable quasi-probabilities cannot be informationally complete in that setting. This sharply contrasts with Wigner-type constructions, which remain real but need not respect the “possible values” criterion in finite-state systems (Umekawa et al., 2023).

A second line uses $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$5-symmetric coherent states on $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$6. In that framework, one constructs finite-dimensional $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$7-, Wigner-, and $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$8-functions for $$W_a(\eta,p)=\frac{1}{\pi^2}\int d^2\xi\, \chi_a(\xi,p)\exp(\eta\xi^*-\eta^*\xi).$$9-level systems by generalized $p$0 coherent-state kernels. For the fundamental representation, the three families differ mainly by the coefficient $p$1 multiplying the generator expectation values inside the kernel, and the density operator can be reconstructed from the resulting phase-space function. Here the analogue of classical phase space is not $p$2 but the compact coherent-state manifold $p$3 (Tilma et al., 2011).

A third line extends this idea to fully symmetric $p$4-quDit systems. There the phase space is

$p$5

and the one-parameter family

$p$6

is built from generalized Fano multipole operators and Stratonovich–Weyl kernels. The corresponding heat kernel

$p$7

connects two orderings, with $p$8 playing the role of time, and in the thermodynamic limit $p$9 the smoothing reduces to the usual Gaussian smoothing for $p=1$0. The same formalism also yields a finite-dimensional twisted Moyal product through a trikernel (Calixto et al., 20 Jul 2025).

Discrete phase-space positivity leads in a different direction. In a $p=1$1-dimensional system, minimum-uncertainty states generated from the top eigenvector of the Harper operator produce a non-negative, translationally covariant quasi-probability

$p=1$2

Its marginals are necessarily smeared, but the paper proves that—under translational covariance—no other non-negative quasi-probability can have sharper marginals. It is complete for odd $p=1$3 and incomplete for even $p=1$4, so positivity, covariance, and completeness coexist only in the odd-dimensional case (Hashimoto et al., 2017).

Loop Quantum Cosmology provides yet another discrete/non-Schrödinger variant. There the $p=1$5-parametrized quasi-distributions live on $p=1$6, the Bohr compactification and its discrete dual, and are given by

$p=1$7

Their marginals are ordering-independent, and for a single character the quasi-distribution is positive for every $p=1$8, a behavior the paper explicitly contrasts with the standard Schrödinger representation (Berra-Montiel et al., 2020).

4. Dynamical quasi-probabilities, counting statistics, and fluctuating work

Quasi-probability methods become especially distinctive when the target object is not a state at one instant but a set of noncommuting observables along a dynamical history. The Keldysh quasi-probability distribution is defined from the generating function

$p=1$9

and its Fourier transform gives a distribution over the corresponding classical Keldysh fields. In this representation, positivity is lost precisely because forward and backward trajectories on the Keldysh contour interfere. The quantum field $P$0 measures the mismatch between the two branches; whenever $P$1, negative contributions can appear. The same formalism reproduces the Wigner function, full counting statistics, weak-value distributions, and the quasi-distributions relevant to Leggett–Garg inequalities (Hofer, 2017).

Measurement models enter through detector Wigner functions. The measured positive distribution is obtained by convolving the intrinsic KQPD with detector kernels, so imprecision broadens the variable while detector back-action damps the off-diagonal interference terms. This is why negativity is not directly observed as a literal negative frequency count, even though it remains a diagnostic of non-classical dynamics (Hofer, 2017).

The full counting statistics formulation of work in a measurement-based quantum Otto engine provides a particularly transparent thermodynamic example. There the FCS characteristic function is

$P$2

and the work quasi-distribution is

$P$3

Because the pair $P$4 need not be diagonal, the distribution contains off-diagonal interference terms rather than positive transition probabilities. In the measurement-based Otto engine this negativity is tied to coherence in the energy eigenbasis at the start of the compression stroke, and that coherence is created by the non-selective Bell-basis measurement together with anisotropic spin coupling. In the standard two-bath Otto engine, complete thermalization makes the relevant states diagonal, the off-diagonal terms vanish, and the FCS distribution collapses to the TPM distribution (Purkait et al., 2024).

The same work also emphasizes a performance connection: coherence modifies the finite-time average work and its fluctuations. A plausible implication is that, in quantum thermodynamics, quasi-probability methods are not merely descriptive; they isolate the exact interference terms responsible for both nonclassical work statistics and finite-time advantages.

5. Polarization quasi-probabilities and observable-specific negativity

Polarization quasi-probability distributions are defined not on canonical phase space but on Stokes space $P$5. The polarization characteristic function

$P$6

generates the PQPD through a three-dimensional Fourier transform. Because each Stokes operator has an integer-valued spectrum, the one-dimensional marginals are sums of $P$7-functions at integer values, and the full PQPD is generically singular and negative (Chekhova et al., 2013).

A central misconception addressed by this literature is that quasi-probability negativity always witnesses nonclassicality of the underlying state in the Wigner sense. For PQPD that inference fails. The negativity can arise even for coherent light because the Stokes observables themselves have discrete spectra. In the weak coherent-state regime, the distribution contains $P$8- and $P$9-type shell terms in Stokes space, and those singular structures generate negative regions after reconstruction. The experimental observation of negative-valued PQPD therefore requires photon-number resolving detectors; photon-number averaging washes out precisely the discrete structure that produces the negativity (Spasibko et al., 2015).

For bright many-photon states, detector smoothing regularizes the PQPD into a positive Gaussian mixture, even for nonclassical inputs such as single-photon seeded squeezed vacuum. To recover Wigner-sensitive negativity, the paper introduces “highlighting”: a strong coherent beam is added in the orthogonal polarization mode. In the strong-field limit, the polarization characteristic function reduces to the symmetrically ordered characteristic function of the signal mode, so the smoothed PQPD becomes a Gaussian-blurred Wigner function. The effective inefficiency parameter

$p=0$0

then controls whether negativity survives; for the squeezed single-photon example, the condition is $p=0$1 (Chekhova et al., 2013).

Taken together, the polarization papers show that quasi-probability negativity can be observable-specific. In Stokes-space tomography it may arise from the discrete nonclassicality of the measured observables rather than from the state alone.

6. Foundations, interpretive limits, and broader extensions

Two sharply different foundational positions appear in the literature summarized here. One is critical: from a Coxian/Bayesian viewpoint, a probability is a tool for ranking plausibility, so any complex-valued candidate must lie on a non-self-intersecting curve in $p=0$2 that can be mapped order-preservingly to $p=0$3. On this basis, the Kirkwood distribution

$p=0$4

and the associated Hofmann conditional complex probabilities are argued not to qualify as probabilities in the Coxian sense, because their values do not support a unique transitive plausibility ordering (Vanslette, 2016).

The second position reconstructs quasi-probabilities from structural principles. Starting from a universal valuation $p=0$5 on finite Boolean algebras of statements and imposing Syntactic Locality, Local Deducibility, Universality, Maximum realisability, and symmetry constraints, the paper proves that there exists a bijection $p=0$6 such that

$p=0$7

is finitely additive on mutually exclusive statements: $p=0$8 These additive representatives are called pre-probabilities. Because the representation is unique only up to additive regraduation, quasi-probabilities arise when this freedom can be fixed canonically, while ordinary finite probabilities appear as the positive subclass stable under relativisation to sub-universes (Surace, 12 Feb 2026).

Beyond quantum state and observable representations, the same quasi-probability logic appears in other physics problems. In nonconvex geometry, the generalized chord distribution

$p=0$9

is a signed distribution rather than an ordinary density; its alternating-sign histogram reproduces Dirac-type integral formulas for arbitrary shapes and multi-body systems (Vlasov, 2010). In Bayesian cosmological inference, the “quasi-Gaussian” approximation for correlation functions is not a signed measure, but it is a support-adapted transformed distribution that plays an analogous role: one maps constrained observables into unconstrained variables, imposes Gaussianity there, and transforms back with the Jacobian, thereby respecting the admissible domain that a naive Gaussian likelihood would violate (Wilking et al., 2013).

These developments suggest a broad methodological conclusion. Quasi-probability distribution methods are best understood as additive representation strategies for situations in which direct positive probability either fails, obscures the relevant algebra, or ignores nontrivial support geometry. Their strength lies in preserving exactly those structural features—operator ordering, interference, discrete spectra, covariance, or admissible support—that ordinary probability would erase.