Bures–Hall Ensemble in Quantum Information

• Bures–Hall ensemble is a probability ensemble of random density matrices induced by the Bures measure, serving as a key model in quantum entanglement studies.
• It employs fixed-trace and unconstrained formulations with a Pfaffian structure, enabling exact calculations of purity and von Neumann entropy.
• Its mapping to the Cauchy–Laguerre model and use of biorthogonal techniques facilitate precise spectral analysis and extensions to deformed ensembles.

Searching arXiv for the core Bures–Hall papers to ground the article in the literature. arXiv search query: Bures-Hall ensemble entanglement entropy purity Cauchy-Laguerre The Bures–Hall ensemble is a probability ensemble of random density matrices induced by the Bures measure, equivalently by the reduced state of a bipartite random pure state drawn according to a Bures-type construction. In the bipartite setting $\mathbb C^m\otimes\mathbb C^n$ with $m\le n$, the reduced density matrix $\rho_A$ has eigenvalues distributed by a fixed-trace Bures–Hall law featuring the characteristic factor $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$, together with a Dirac constraint $\delta(1-\sum_i\lambda_i)$ and a boundary exponent $\alpha=n-m-\tfrac12$ (Wei, 2020). The ensemble occupies a central position in quantum-information geometry because the Bures measure is the “minimal–disturbance” or “quantum Jeffreys prior” on density matrices, and in random-matrix theory because its spectral statistics form a Pfaffian point process closely tied to the Cauchy–Laguerre biorthogonal ensemble (Wei, 2020, Forrester et al., 2014).

1. Definition and geometric origin

For a bipartite pure state $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$ chosen according to the Bures measure, the reduced density matrix

$$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$

has eigenvalues $\{\lambda_i\}_{i=1}^m$ with joint probability density

$$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$

where

$m\le n$0

and

$m\le n$1

The Dirac delta enforces $m\le n$2 (Wei, 2020).

An equivalent sampling prescription is to draw a random pure state on the $m\le n$3-dimensional Hilbert space from the Bures measure, or via a Ginibre–unitary construction, and then trace over subsystem $m\le n$4 (Wei, 2020). A closely related purification-based formulation writes a random pure state as

$m\le n$5

with $m\le n$6 i.i.d. complex Gaussians, and then forms

$m\le n$7

where $m\le n$8 is distributed with weight proportional to $m\le n$9, $\rho_A$0; the reduced state $\rho_A$1 defines the generalized Bures–Hall ensemble (Li et al., 2021).

The ensemble is induced by the Bures distance, and the associated volume element factorizes into eigenvalue and Haar-unitary parts. In eigenvalue coordinates, the Bures measure contains the determinant factor $\rho_A$2 and the eigenvalue-repulsion factor $\rho_A$3, in contrast with the Hilbert–Schmidt measure, which retains only $\rho_A$4 (Slater, 2021). This places the Bures–Hall ensemble at the intersection of quantum-state geometry and invariant random-matrix theory.

2. Fixed-trace and unconstrained forms

A standard structural distinction is between the fixed-trace ensemble for normalized eigenvalues $\rho_A$5 and the unconstrained ensemble for positive variables $\rho_A$6. Forrester and Kieburg showed that the “unconstrained” Bures–Hall measure is

$\rho_A$7

with

$\rho_A$8

and that it is exactly the marginal of a two-matrix “Cauchy–Laguerre” model with weight functions

$\rho_A$9

and interaction kernel $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$0 (Wei, 2020).

The fixed-trace and unconstrained ensembles are related by introducing the scale variable $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$1. In the unconstrained formulation $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$2 has gamma law proportional to $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$3, with

$$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$4

and $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$5 is independent of the normalized eigenvalues $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$6 (Wei et al., 7 Jun 2025). This factorization underlies the moment-relation technique used in exact entropy calculations (Wei, 2020).

The fixed-trace and unrestricted-trace versions are also related by Laplace transform. In the fixed-trace formulation, $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$7-point correlations can be obtained from those of the unrestricted ensemble by an inverse Laplace transform in the total trace variable (Sarkar et al., 2019). This relation is important because many calculations are technically simpler in the unconstrained ensemble and can then be transferred back to the normalized density-matrix setting.

3. Pfaffian structure and the Cauchy–Laguerre correspondence

The decisive structural result is the mapping between the Bures–Hall ensemble and the Cauchy two-matrix model. Forrester and Kieburg established that the Bures partition function squares to the Cauchy two-matrix partition function, specifically

$$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$8

for the Laguerre-weighted case $$\prod_{i<j}(\lambda_i-\lambda_j)^2/(\lambda_i+\lambda_j)$$9 (Forrester et al., 2014). This explains how a Pfaffian point process can be derived from a determinantal point process.

The Cauchy two-matrix ensemble has joint density

$\delta(1-\sum_i\lambda_i)$0

and its correlation functions are determinantal (Forrester et al., 2014). By contrast, the Bures ensemble is Pfaffian: $\delta(1-\sum_i\lambda_i)$1 with a $\delta(1-\sum_i\lambda_i)$2 matrix kernel $\delta(1-\sum_i\lambda_i)$3 built explicitly from the four Cauchy kernels $\delta(1-\sum_i\lambda_i)$4 (Forrester et al., 2014).

In the Laguerre-weighted Bures–Hall case, the one-point density can be expressed through Meijer $\delta(1-\sum_i\lambda_i)$5-functions: $\delta(1-\sum_i\lambda_i)$6 where

$\delta(1-\sum_i\lambda_i)$7

with parameters inherited from the biorthogonal construction (Wei, 2020). This representation is not merely formal; it is the basis of exact Mellin-transform evaluations of spectral moments and entropy averages.

A related line of work generalizes the ordinary Bures–Hall ensemble to a $\delta(1-\sum_i\lambda_i)$8-deformed Laguerre–Bures ensemble, in which the kernels are expressed in terms of Fox $\delta(1-\sum_i\lambda_i)$9-functions and reduce to Meijer $\alpha=n-m-\tfrac12$0-functions when $\alpha=n-m-\tfrac12$1 (Forrester et al., 2018). This situates the classical Bures–Hall model as a distinguished point inside a broader integrable family.

4. Exact average purity and von Neumann entropy

The entanglement content of a random bipartite pure state may be measured through the reduced density matrix $\alpha=n-m-\tfrac12$2. For the smaller subsystem, the purity is

$\alpha=n-m-\tfrac12$3

and the von Neumann entropy is

$\alpha=n-m-\tfrac12$4

(Wei, 2020).

Sarkar and Kumar conjectured simple closed forms for the average purity and average von Neumann entropy, based on exact evaluations from the fixed-trace level density and finite-sum Pfaffian formulas (Sarkar et al., 2019). These conjectures were later proved: $\alpha=n-m-\tfrac12$5 and

$\alpha=n-m-\tfrac12$6

(Wei, 2020).

The proof strategy rests on two ingredients. First, one uses moment-relations that trade the fixed-trace constraint $\alpha=n-m-\tfrac12$7 for an independent gamma integral, reducing the problem to averages over the unconstrained ensemble. Second, one invokes the Forrester–Kieburg mapping to the Cauchy–Laguerre biorthogonal ensemble, where one-point functions are explicitly available in Meijer $\alpha=n-m-\tfrac12$8-function form (Wei, 2020). The central Mellin-transform identity for Meijer $\alpha=n-m-\tfrac12$9-functions then converts integrals such as $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$0 and $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$1 into gamma and digamma expressions (Wei, 2020).

These exact formulas clarify the relation to the Hilbert–Schmidt case. Comparisons show that Bures–Hall average entropies differ from the Hilbert–Schmidt case, sometimes identified with Page’s law, by $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$2 corrections that reflect greater weight on mixed states (Wei, 2020). The data also state that the Bures–Hall ensemble tends to favor less entangled states, with larger purity and smaller von Neumann entropy, than the Hilbert–Schmidt ensemble (Wei et al., 2021). Taken together, these statements indicate that the Bures prior modifies typical entanglement in a systematic finite-size manner.

5. Higher moments, cumulants, and refined statistics

Beyond first moments, the Bures–Hall ensemble supports exact finite-size formulas for higher purity moments. Li and Wei obtained exact expressions for the second and third moments of quantum purity for arbitrary subsystem dimensions $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$3, extending earlier results that were limited to equal subsystem dimensions (Li et al., 2021). Their approach uses the unconstrained ensemble, Pfaffian $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$4-point densities, and recurrence relations for Cauchy–Laguerre biorthogonal polynomials. The first three purity moments are expressed through

$\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$5

with $\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$6 (Li et al., 2021).

For the von Neumann entropy, the first two cumulants were already known, and the third cumulant was derived in closed form in later work. The mean is

$\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$7

the variance is

$\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$8

and the third cumulant is

$\lvert\psi\rangle\in\mathbb C^m\otimes\mathbb C^n$9

with explicit rational coefficients $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$0 and $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$1 in $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$2 (Wei et al., 7 Jun 2025).

The derivation of $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$3 is technically notable because it involves Pfaffian correlation kernels $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$4, nested finite sums of rational functions times polygamma values, and a collection of 18 single-sum “anomalies” $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$5 whose contributions cancel only after new re-summation identities are established (Wei et al., 7 Jun 2025). The resulting standardized skewness

$$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$6

satisfies $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$7 as $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$8 with $$\rho_A=\operatorname{Tr}_B\bigl(\lvert\psi\rangle\langle\psi\rvert\bigr)$$9, consistent with a central-limit conjecture (Wei et al., 7 Jun 2025).

These cumulants also yield an Edgeworth-type approximation for the standardized entropy variable $\{\lambda_i\}_{i=1}^m$0: $\{\lambda_i\}_{i=1}^m$1 where $\{\lambda_i\}_{i=1}^m$2 (Wei et al., 7 Jun 2025). This gives a more accurate approximation to finite-$\{\lambda_i\}_{i=1}^m$3 entropy distributions than a purely Gaussian approximation.

A more recent development establishes a recurrence relation for the $\{\lambda_i\}_{i=1}^m$4-th spectral moment valid for real-valued $\{\lambda_i\}_{i=1}^m$5, obtained via Christoffel–Darboux formulas that avoid lengthy summations. As an application, the average von Neumann entropy and quantum purity are re-derived from spectral moments (Wei et al., 1 Feb 2026). This suggests a unifying route from kernel identities to entanglement statistics.

6. Extensions, deformations, and related problems

Several extensions place the Bures–Hall ensemble inside wider analytic and physical frameworks. One is the quantum interpolating ensemble, a specialization of the $\{\lambda_i\}_{i=1}^m$6-deformed Cauchy–Laguerre two-matrix model. At $\{\lambda_i\}_{i=1}^m$7 and $\{\lambda_i\}_{i=1}^m$8, the interpolating model reduces to the Bures–Hall ensemble, while $\{\lambda_i\}_{i=1}^m$9 corresponds to the Hilbert–Schmidt endpoint (Wei et al., 2021). In this setting, average purity and von Neumann entropy can be computed for the entire interpolating family, and the Bures–Hall case emerges as the point where the finite sums collapse to the familiar simple closed forms (Wei et al., 2021).

Another direction concerns local spectral statistics and universality. The $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$0-deformed Laguerre–Bures ensemble is a Pfaffian point process with kernels expressible through Fox $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$1-functions; at $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$2 these reduce to the Meijer $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$3-function formulas of the ordinary Bures ensemble (Forrester et al., 2018). Hard-edge scaled limits are likewise given by explicit $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$4-kernel formulas (Forrester et al., 2018). This extends the Bures–Hall model from a single ensemble to a continuous family of integrable kernels.

Gap probabilities constitute a further extension. For the unconstrained Bures–Hall ensemble, the gap generating function $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$5 satisfies

$$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$6

where $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$7 is the gap generating function of the Cauchy–Laguerre two-matrix model (Witte et al., 2022). Through Laplace inversion, one obtains bottom-gap and top-gap probabilities for the fixed-trace Bures–Hall ensemble, leading to a rank-3 isomonodromic deformation problem and a compatible Lax triplet (Witte et al., 2022). This embeds Bures–Hall spectral gaps in the theory of integrable systems.

The ensemble also appears in applied contexts. Because the Bures and Hilbert–Schmidt measures have explicitly known Jacobians, Monte Carlo samples from one ensemble can be reweighted to estimate observables under the other. In the two-qubit case, this strategy has been used to recover the separability probabilities $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$8 for Hilbert–Schmidt and $$f(\lambda_1,\dots,\lambda_m) =\frac{1}{c}\; \delta\!\Bigl(1-\sum_{i=1}^m\lambda_i\Bigr)\, \prod_{1\le i<j\le m}\frac{(\lambda_i-\lambda_j)^2}{\lambda_i+\lambda_j} \;\prod_{i=1}^m \lambda_i^\alpha,$$9 for Bures, and to construct quantum-steering-ellipsoid-based estimates under Bures weighting (Slater, 2021). A different extension imposes a fixed-energy constraint on random mixed states, producing a microcanonical ensemble with Bures–Hall volume element and an associated density of states $m\le n$00, entropy $m\le n$01, and temperature $m\le n$02 (Miller, 1 Aug 2025). This suggests that Bures–Hall geometry can serve not only as a prior on density matrices but also as a foundation for a statistical mechanics of mixed states.

A recurring misconception is to regard the Bures–Hall ensemble as merely a minor variant of the Hilbert–Schmidt ensemble. The exact formulas above show otherwise: the additional $m\le n$03 and $m\le n$04 factors alter both global entanglement averages and higher-order fluctuation statistics (Slater, 2021, Wei, 2020). Another misconception is that the Pfaffian structure obstructs explicit analysis. The mapping to the determinantal Cauchy–Laguerre model shows the opposite: the Pfaffian character is precisely what becomes tractable through the biorthogonal two-matrix correspondence (Forrester et al., 2014).