---
title: Kirkwood–Dirac Quasiprobability Representation
url: https://www.emergentmind.com/topics/kirkwood-dirac-quasiprobability-representation
type: topic
---

# Kirkwood–Dirac Quasiprobability Representation

The Kirkwood–Dirac (KD) quasiprobability representation is a quasiprobability description of a quantum state relative to two observables, introduced by Kirkwood in 1933 and Dirac in 1945 and now used across quantum metrology, chaos, thermodynamics, weak-value theory, and foundations [2009.04468]. For nondegenerate observables with eigenbases \(\{|a_i\rangle\}_{i=1}^d\) and \(\{|b_j\rangle\}_{j=1}^d\), the standard KD array of a state \(\rho\) is
\[
Q_{ij}(\rho)=\langle b_j|a_i\rangle\,\langle a_i|\rho|b_j\rangle
=\mathrm{Tr}\!\bigl(\Pi_{b_j}\Pi_{a_i}\rho\bigr).
\]
It is normalized, reproduces the Born marginals of both observables, and is informationally complete when \(\langle a_i|b_j\rangle\neq 0\) for all \(i,j\), but its entries can be negative or nonreal. Those departures from classical probability are precisely what make the representation useful: they encode noncommutativity, disturbance, anomalous weak values, and several operational quantum advantages [2403.18899].

## 1. Definition, reconstruction, and generalizations

The KD representation is indexed by outcome pairs of two measurements. Its basic structural properties are the same across the recent literature: normalization,
\[
\sum_{i,j}Q_{ij}(\rho)=1,
\]
marginal recovery,
\[
\sum_j Q_{ij}(\rho)=\langle a_i|\rho|a_i\rangle,\qquad
\sum_i Q_{ij}(\rho)=\langle b_j|\rho|b_j\rangle,
\]
and, when \(\langle b_j|a_i\rangle\neq 0\ \forall i,j\), the inversion formula
\[
\rho=\sum_{i,j}Q_{ij}(\rho)\,\frac{|a_i\rangle\langle b_j|}{\langle b_j|a_i\rangle}.
\]
Equivalently, \(\{|a_i\rangle\langle b_j|/\langle b_j|a_i\rangle\}\) forms a dual operator basis to \(\{\Pi_{a_i}\Pi_{b_j}\}\) [2601.17788].

This dual-basis viewpoint places KD within the general theory of quasiprobability representations. For any operator \(\hat C\), one may introduce the dual symbol
\[
T_{i,j}(\hat C)=\frac{\langle a_i|\hat C|b_j\rangle}{\langle a_i|b_j\rangle},
\]
so that
\[
\mathrm{Tr}(\hat C^\dagger \rho)=\sum_{i,j}T_{i,j}(\hat C)^*\,Q_{i,j}(\rho).
\]
The formal role is therefore analogous to a classical expectation value, except that the “distribution” \(Q\) is generally complex [2403.18899].

The standard two-basis KD distribution admits several extensions. One is the ordered \(k\)-measurement or extended KD array,
\[
Q_{i_1,\dots,i_k}(\rho)=
\mathrm{Tr}\!\Bigl(\Pi^{(k)}_{i_k}\Pi^{(k-1)}_{i_{k-1}}\cdots \Pi^{(1)}_{i_1}\rho\Bigr),
\]
which is central in scrambling and multi-time problems [2403.18899]. Another is the POVM form
\[
q_{xy}=\mathrm{Tr}[\rho\,E_xE_y],
\]
which preserves normalization and marginals while allowing nonprojective measurements [2504.09238]. A further generalization replaces finite-dimensional basis pairs by Fourier-dual structures on second countable locally compact abelian groups, where
\[
KD_\rho(g,\chi)=\mathrm{Tr}[\,\rho\,W(g,\chi)\,]
\]
with \(W(g,\chi)=M_\chi T_g\) the Weyl operator; in that setting the KD transform is a unitary map \(T_2(H)\to L^2(G\times \widehat G)\) [2507.23628].

A recurrent theme in the literature is that, unlike Wigner-type constructions, KD is arbitrary-basis rather than canonically tied to position–momentum structure. This suggests why it has become useful in discrete systems, generalized measurements, and basis-dependent resource theories [2403.18899].

## 2. Classicality, nonclassicality, and what KD values mean

A state is called KD-positive, or KD-classical, when all entries satisfy \(Q_{ij}(\rho)\ge 0\). In that case the KD table is a bona fide joint probability distribution over the two outcome sets. If some entry is negative or nonreal, the state is KD-nonclassical [2306.00086].

The recent literature distinguishes two kinds of nonclassicality. Negative KD elements signify nonclassicality of the “Metrological” type: they bound how weak-value amplification can exceed classical metrology limits, and they signal operator-scrambling in out-of-time-ordered correlators. Nonreal elements encode a different kind of nonclassicality: they quantify measurement back-action and “thermodynamic” contextuality, for example in fluctuation-theorem settings involving successive noncommuting energy measurements [2009.04468].

A common misconception is that noncommutation alone guarantees KD nonclassicality. That claim is false. Pairwise noncommutation of \(A\), \(B\), and \(\rho\) does not suffice: explicit examples exist in which \(A\), \(B\), and \(\rho\) all pairwise fail to commute while the KD array remains real and nonnegative [2009.04468]. The sufficient-nonclassicality theorem for pure states therefore uses finer support data. If
\[
2N_A+2N_B>3d+n_\parallel-3\bar n_\parallel,
\]
where \(N_A\) and \(N_B\) count support sizes in the two bases and \(n_\parallel,\bar n_\parallel\) record aligned and orthogonal overlaps, then at least one KD entry is negative or nonreal [2009.04468]. In particular, if no KD entry vanishes, the KD array is automatically nonclassical [2009.04468].

Several quantitative measures separate real negativity from nonreality. The paper “Conditions tighter than noncommutation needed for nonclassicality” defines
\[
{\cal N}=-1+\sum |KD|,
\qquad
{\cal N}^{\Re^-}=-1+\sum |\Re\,KD|,
\qquad
{\cal N}^{\Im}=\sum |\Im\,KD|,
\]
and proves that, over all pure states and sequences of \(k\) bases in dimension \(d\),
\[
\max {\cal N}=d^{(k-1)/2}-1.
\]
This maximum is achieved iff each consecutive pair of bases is mutually unbiased and the initial and final bases have uniform overlap with the state [2009.04468].

KD also fits into a broader hierarchy of admissible quasiprobabilities. Classical joint distributions form a strict subset of KD quasiprobabilities, and KD quasiprobabilities form a strict subset of a larger postquantum set defined only by normalization and entrywise modulus bounds [2504.09238]. For KD tables \(q_{xy}\), one has \(|q_{xy}|\le \max\{p_x,p_y\}\), \(\sum_{x,y}|q_{xy}|\le \sqrt N\), and, more strongly,
\[
\sum_{x,y}|q_{xy}|^2\le 1,
\]
with the same \(\ell_2\) bound extending to arbitrary-length KD chains [2504.09238].

## 3. Geometry of KD-positive states

The set of KD-positive states,
\[
\mathcal K_+=\{\rho\in\mathcal S\mid Q_{ij}(\rho)\ge 0\ \forall i,j\},
\]
is a convex subset of the state space. By Krein–Milman,
\[
\mathcal K_+=\mathrm{conv}(\mathrm{ext}\,\mathcal K_+).
\]
Two natural subsets of pure states always lie inside \(\mathcal K_+\): the projectors onto the \(A\)-eigenbasis and onto the \(B\)-eigenbasis [2306.00086].

A central geometric question is whether these basis projectors already generate the entire KD-positive set. Langrenez, Arvidsson-Shukur, and De Bièvre show that this occurs exactly when
\[
\mathcal K_+=\mathrm{conv}(\mathcal A\cup \mathcal B)
\quad\Longleftrightarrow\quad
\dim(\mathrm{Im}\,Q)=2d-1.
\]
They identify three regimes where only convex combinations of the \(A\)- and \(B\)-eigenprojectors are KD-positive: \(d=2\) with \(m_{ab}>0\); an open dense subset of \(U(3)\), equivalently full Haar measure, in dimension \(3\); and discrete-Fourier-transform bases in prime dimension. They also prove a stability statement: if \(U'\) is close to such a \(U\), then the same convex-hull description persists [2306.00086].

In these minimal regimes, \(\mathcal K_+\) is the \(2(d-1)\)-dimensional convex polytope with exactly \(2d\) vertices, namely the \(A\)- and \(B\)-projectors, and its facets are labeled by leaving out one \(A\)-projector and one \(B\)-projector [2306.00086]. Outside these regimes the geometry is richer. For a specific spin-\(1\) example with
\[
U^*=\frac13
\begin{bmatrix}
-1&2&2\\
2&-1&2\\
2&2&-1
\end{bmatrix},
\]
the KD-positive set contains mixed extreme points that are not convex combinations of pure KD-positive states. Writing
\[
F_\perp=\frac{1}{\sqrt 6}
\begin{bmatrix}
0&1&1\\
1&0&1\\
1&1&0
\end{bmatrix},
\qquad
\rho(x)=\tfrac13 I+xF_\perp,
\]
one finds \(\rho(x)\in\mathcal K_+\) for \(-1/\sqrt6\le x\le 1/(2\sqrt6)\), while \(\rho(x)\notin\mathrm{conv}(\mathcal A\cup\mathcal B)\) for \(0<x\le 1/(2\sqrt6)\) [2306.00086].

For pure states, the DFT case is especially rigid. In the characterization of KD-classical pure states, the DFT relation implies that the only KD-classical pure states satisfy
\[
n_A(\psi)\,n_B(\psi)=d,
\]
where \(n_A\) and \(n_B\) are the support sizes of \(|\psi\rangle\) in the two bases. This resolves the conjecture of De Bièvre cited there [2210.02876]. A later DFT study extends this to mixed states in prime-power dimensions, proving that the KD-classical set is the convex hull of KD-classical pure states and organizing the general \(d\)-dimensional structure by a directed graph on coprime factorizations of \(d\) [2603.13863].

## 4. Conditional expectation, weak values, and measurement dynamics

The KD representation is tightly connected to conditional averages and weak values. In the standard two-basis setting, a weak value of \(A\) under postselection on \(|b_j\rangle\) can be written as a KD-conditional average,
\[
A_w=\sum_i a_i\,\tilde Q_{i|j},
\]
with \(\tilde Q_{i|j}\) the conditionalized KD weights [2403.18899]. This is one reason negative or nonreal KD values are directly associated with anomalous weak values.

Recent work sharpens that connection by characterizing KD among all Born-compatible quasiprobability representations on the joint spectrum of \(\hat A\) and \(\hat B\). For any such representation, one can define a conditional expectation of \(\hat X\) given \(\hat B\). What is special about KD is uniqueness: only the KD representation has the property that its associated conditional expectation of \(\hat X\) given \(\hat B\) coincides with the best predictor of \(\hat X\) by a function of \(\hat B\), for all \(\hat X\) [2511.01996]. In the left-KD case,
\[
E^\ell_{KD,\rho}[\hat X|\hat B]
=
\sum_b
\frac{\langle b|\hat X\rho|b\rangle}{\langle b|\rho|b\rangle}\,
\Pi_b^B.
\]
This least-squares characterization distinguishes KD from other Born-compatible frames [2511.01996].

The role of KD in measurement theory has also been extended beyond the weak-measurement limit. A von Neumann pointer model with coupling
\[
H=\frac{p^2}{2m}+g\,A\otimes P
\]
and Gaussian pointer width \(\sigma\) yields the decoherence factor
\[
F(t)=\exp\!\Bigl[-\frac{(g\,\Delta a\,t)^2}{8\sigma^2}\Bigr].
\]
In the resulting reduced dynamics, the time-dependent KD array obeys the linear interpolation
\[
Q_{ij}(\rho(t))
=
F(t)\,Q_{ij}(\rho)+(1-F(t))\,Q_{ij}^{\rm Wigner},
\]
where
\[
Q_{ij}^{\rm Wigner}=p(a_i|\rho)\,p(f_j|a_i)\ge 0.
\]
Thus \(F\approx 1\) gives the full complex KD distribution in the weak regime, whereas \(F\to 0\) yields the real nonnegative Margenau–Hill or “classical Wigner” formula in the strong regime, while informational completeness is maintained throughout the transition [2601.17788].

This measurement-strength interpolation suggests a unifying reading of KD: it is not merely a weak-measurement artifact, but a representation whose deformation under pointer-induced decoherence continuously connects weak values, ABL-type conditional averages, and projective statistics [2601.17788].

## 5. Operational roles in quantum information, thermodynamics, and foundations

The KD representation has become operationally significant because its negative and nonreal values track concrete tasks and phenomena. In measurement-disturbance theory, the imaginary part of KD controls the change of probabilities under unitary generation:
\[
\frac{d}{d\theta}P(f_j|\psi_\theta)=2\sum_i a_i\,\Im[Q_{i,j}(\psi_\theta)].
\]
The imaginary sector therefore quantifies disturbance in a basis-resolved way [2403.18899].

In direct state measurement, a weak measurement of position \(\hat\pi_x=|x\rangle\langle x|\) followed by a strong measurement of momentum \(\hat\pi_p=|p\rangle\langle p|\) yields
\[
\langle \hat\pi_p\hat\pi_x\rangle_w
=
\mathrm{Tr}(\hat\pi_p\hat\pi_x\rho)
=
Q(x,p;\rho),
\]
so scanning \((x,p)\) directly reconstructs a wavefunction or density matrix in KD form [2403.18899].

In thermodynamics, KD provides a quasiprobability of work that preserves coherence discarded by two-point-measurement protocols. For instantaneous energy measurements,
\[
q_{if}(\rho)=\mathrm{Tr}\bigl[U^\dagger(t)\,\Pi_f(t)\,U(t)\,\Pi_i(0)\,\rho\bigr].
\]
Interferometric reconstruction with an ancilla accesses the characteristic function \(\mathcal G(u)\), whose real and imaginary parts are read out through \(\langle \sigma_x\rangle_A\) and \(\langle \sigma_y\rangle_A\), and Fourier transformation yields the KD work distribution [2405.21041]. Its first moment equals the standard average work,
\[
\langle W\rangle_{\rm KD}
=
\mathrm{Tr}[H(t)\rho(t)]-\mathrm{Tr}[H(0)\rho],
\]
while the imaginary part of the second moment is
\[
\Im \mathrm{var}\,W_{\rm KD}
=
\mathrm{Tr}\bigl[i\,\rho\,[H(0),\widetilde H(t)]\bigr],
\]
which is also the commutator term entering the Robertson–Schrödinger uncertainty relation [2405.21041].

In chaos and scrambling, extended KD quasiprobabilities underlie out-of-time-ordered correlators, with negativity and nonreality distinguishing scrambling from decoherence [2403.18899]. In foundations, KD negativity appears in Leggett–Garg inequality violations, consistent-histories interference terms, and contextuality. One recent coherence paper states that whenever any weak value \(\Pi_x^{\rm w}(z|\rho)\) is “strange,” meaning it has a negative real part or nonzero imaginary part, it is known to witness quantum contextuality; since \(C_{\rm KD}>0\) iff at least one such strange weak value appears, the corresponding KD-based coherence measure is directly tied to contextuality [2309.09162].

A broader resource-theoretic claim is also available: for every quantum resource defined by a closed convex set of nonresourceful states, there exists a KD-type quasiprobability distribution whose negativity reveals that resourceful state, and one can choose the construction so that the negative value is proportional to the Frobenius distance to the closest nonresourceful state [2401.03508]. This suggests that KD-type negativity is not confined to one operational niche.

## 6. Resource-theoretic, temporal, and abstract extensions

Several lines of work turn KD nonclassicality into standalone resource measures. One defines basis-relative KD-nonclassicality coherence by
\[
C_{\rm KD}(\rho;\{\Pi_x\})
=
\sup_{\{|z\rangle\}}
\Bigl[\sum_{x,z}\bigl|\mathrm{Tr}(\Pi_z\Pi_x\rho)\bigr|-1\Bigr].
\]
This quantity vanishes iff \(\rho\) is diagonal in the reference basis, is upper bounded by \(\tfrac12 H_{1/2}(p)\), and for pure states satisfies equality,
\[
C_{\rm KD}(|\psi\rangle)
=
\sum_x \sqrt{p_x}-1
=
\tfrac12 H_{1/2}(p)
\]
[2309.09162]. A complementary construction based only on the imaginary part defines
\[
C_{KD}(\rho)
=
\max_{\{|b_j\rangle\}}
\sum_{i,j}|\Im KD_{ij}(\rho)|,
\]
proves convexity and several monotonicity properties, shows
\[
C_{KD}(\rho)\le \sqrt{\mathrm{Var}_\rho(A)},
\qquad
C_{KD}(\rho)\le C_{\ell_1}(\rho),
\]
and establishes equality with \(C_{\ell_1}\) for a single qubit [2309.09152].

For bipartite systems, KD nonreality over orthonormal product bases yields measures \(Q^A_{KD}(\rho_{AB})\) and \(Q_{KD}(\rho_{AB})\) of general quantum correlations. These vanish exactly on classical–quantum or classical–classical states, are convex and locally unitary invariant, and on pure states provide faithful witnesses of entanglement and measurement-induced nonlocality [2208.03442].

Recent work has also produced experimentally economical witnesses of KD nonpositivity using moments. Defining
\[
q_n=\sum_{i,j}(Q_{ij})^n,
\]
one has, for any KD-positive table,
\[
q_2^2\le q_3.
\]
More generally, the Hankel matrix \(H_m(\mathbf q)\) built from \((q_1,\dots,q_{2m+1})\) must satisfy \(H_m(\mathbf q)\succeq 0\), so any violation of \(\det H_m(\mathbf q)\ge 0\) detects KD nonpositivity [2506.08107].

The temporal extension is structurally significant. For a multi-time process, one defines right, left, and doubled temporal KD quasiprobabilities, with \(\overleftarrow Q=\overrightarrow Q^{\,*}\) and real parts equal to temporal Margenau–Hill quasiprobabilities. These objects are experimentally accessible through interferometric schemes and feed directly into temporal Bloch tomography. The associated reconstructed “temporal states” recover, as special cases, consistent histories, pseudo-density operators, process tensors, quantum combs, superdensity operators, and doubled-density-operator formalisms [2601.05294].

Finally, the abstract harmonic-analysis formulation on second countable locally compact abelian groups shows that KD is the natural standard-ordering, or Kohn–Nirenberg, quantization on \(G\times\widehat G\). In that setting, generalized pure KD-positive states are, up to Weyl–Heisenberg action, Haar measures on closed subgroups; the classical fragment is nontrivial iff the group has a compact connected component; and for connected compact abelian groups, KD-positive states are exactly classical mixtures of Fourier-basis projectors [2507.23628]. This places the finite-dimensional geometry of KD positivity within a broader phase-space and representation-theoretic framework.

The resulting picture is coherent across these developments. KD behaves like a joint probability distribution to the extent allowed by Born marginals and normalization, but its negative and nonreal sectors encode the obstructions to a classical joint model. Those obstructions are not incidental defects of the formalism; they are the features through which KD connects measurement disturbance, weak values, contextuality, coherence, work statistics, temporal correlations, and the geometry of the quantum state space.

Source: https://www.emergentmind.com/topics/kirkwood-dirac-quasiprobability-representation