Hilbert–Schmidt Ensemble Overview

• The Hilbert–Schmidt ensemble is a unitarily invariant measure on density matrices defined by the Hilbert–Schmidt metric and realized through induced Wishart constructions.
• It employs two equivalent sampling procedures: one by normalizing a Wishart matrix from a Ginibre ensemble and another via partial tracing of a Haar-random pure state.
• Applications include quantifying entanglement and purity, with exact spectral and distance formulas offering insights into quantum state geometry.

Searching arXiv for relevant papers on Hilbert–Schmidt ensembles, induced measures, and related random density matrix results. Using arXiv search to retrieve recent and foundational papers relevant to the Hilbert–Schmidt ensemble. The Hilbert–Schmidt ensemble is the unitarily invariant, flat Euclidean measure on the convex set of density matrices, defined by the Hilbert–Schmidt metric

$$d_{HS}(\rho,\sigma)=\sqrt{\operatorname{Tr}\big[(\rho-\sigma)^2\big]}.$$

In random-matrix and quantum-information formulations, it is the square case of the induced ensemble $\mu_{N,K}$, obtained either by partial tracing a Haar-random pure state on $\mathbb{C}^N\otimes \mathbb{C}^K$ or, equivalently, by normalizing a Wishart matrix $W=GG^\dagger$ built from a Ginibre matrix $G$. The Hilbert–Schmidt specialization corresponds to $K=N$, so that

$$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$

is distributed according to the fixed-trace Wishart–Laguerre measure and provides the canonical “structureless” ensemble of random mixed states (Zyczkowski et al., 2010, Kumar, 2020).

1. Definition as an induced and fixed-trace Wishart ensemble

For a bipartite Hilbert space $\mathcal H_A\otimes\mathcal H_B$ of dimensions $N\times K$, one may draw a Haar-random pure state $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$ and define the reduced state

$\mu_{N,K}$0

This construction yields the induced measures $\mu_{N,K}$1 on the set $\mu_{N,K}$2 of $\mu_{N,K}$3 density matrices. The Hilbert–Schmidt ensemble is the square case $\mu_{N,K}$4. An equivalent formulation expands $\mu_{N,K}$5 in a product basis with coefficients $\mu_{N,K}$6, arranges them into a matrix $\mu_{N,K}$7, and sets $\mu_{N,K}$8. The reduced state then has normalized Wishart form

$\mu_{N,K}$9

This equivalence is central both conceptually and computationally (Zyczkowski et al., 2010).

In the formulation used for exact distance calculations, the induced fixed-trace measure on density matrices is

$\mathbb{C}^N\otimes \mathbb{C}^K$0

where $\mathbb{C}^N\otimes \mathbb{C}^K$1 is the system dimension, $\mathbb{C}^N\otimes \mathbb{C}^K$2 is the environment or ancilla dimension, and $\mathbb{C}^N\otimes \mathbb{C}^K$3 is the Dyson index specifying real-symmetric ($\mathbb{C}^N\otimes \mathbb{C}^K$4) or complex-Hermitian ($\mathbb{C}^N\otimes \mathbb{C}^K$5) ensembles. The associated Wishart construction uses

$\mathbb{C}^N\otimes \mathbb{C}^K$6

with $\mathbb{C}^N\otimes \mathbb{C}^K$7 an $\mathbb{C}^N\otimes \mathbb{C}^K$8 real or complex Ginibre matrix distributed as

$\mathbb{C}^N\otimes \mathbb{C}^K$9

With this convention, the real variance of each independent entry is $W=GG^\dagger$0 for $W=GG^\dagger$1, and the complex variance is $W=GG^\dagger$2 per complex component for $W=GG^\dagger$3 (Kumar, 2020).

Two equivalent sampling procedures therefore generate Hilbert–Schmidt random density matrices. The first is the Ginibre/Wishart recipe: sample $W=GG^\dagger$4 with i.i.d. complex normal entries, set $W=GG^\dagger$5, and output $W=GG^\dagger$6. The second is the Haar-pure-state construction: sample $W=GG^\dagger$7, set $W=GG^\dagger$8, compute $W=GG^\dagger$9, and output $G$0. Proposition 1 in the structured-ensemble framework further shows stability under partial traces: tracing out a factor from an induced ensemble $G$1 yields $G$2 when $G$3 (Zyczkowski et al., 2010).

2. Spectral form, unitary invariance, and basic moments

The induced and Hilbert–Schmidt ensembles are unitarily invariant. In eigenvalue–eigenvector coordinates, the measure factorizes into Haar measure on eigenvectors and a joint density on eigenvalues. In the complex case $G$4, the induced ensemble has joint eigenvalue density

$G$5

For the Hilbert–Schmidt ensemble, $G$6, so the factor $G$7 becomes unity. Equivalently, in the matrix formulation, the dependence on a fixed comparator reduces to spectral invariants because of unitary invariance (Zyczkowski et al., 2010, Kumar, 2020).

The ensemble averages that organize many exact calculations are especially simple. One has

$G$8

and the average purity is

$G$9

For $K=N$0,

$K=N$1

and in the Hilbert–Schmidt specialization $K=N$2,

$K=N$3

If $K=N$4 is fixed with $K=N$5, then

$K=N$6

and for two independent induced density matrices $K=N$7,

$K=N$8

These identities are the basis of the compact Hilbert–Schmidt distance formulas (Kumar, 2020).

In the large-dimension regime $K=N$9 with fixed aspect ratio $$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$0, the empirical eigenvalue density of the rescaled eigenvalues $$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$1 converges to the Marchenko–Pastur law

$$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$2

For the Hilbert–Schmidt case $$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$3, this reduces to

$$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$4

The average von Neumann entropy of the reduced state for a Haar-random pure state on $$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$5 is

$$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$6

so that in the Hilbert–Schmidt specialization

$$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$7

This suggests that Hilbert–Schmidt random states are typically highly mixed, with entropy close to $$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$8 but purity of order $$\rho=\frac{GG^\dagger}{\operatorname{Tr}(GG^\dagger)}$$9 (Zyczkowski et al., 2010).

3. Exact mean-square Hilbert–Schmidt distances

The Hilbert–Schmidt distance between two density matrices is

$\mathcal H_A\otimes\mathcal H_B$0

and its squared form is

$\mathcal H_A\otimes\mathcal H_B$1

The exact results available for the Hilbert–Schmidt ensemble concern the mean of the squared distance, $\mathcal H_A\otimes\mathcal H_B$2, not the mean of the unsquared distance $\mathcal H_A\otimes\mathcal H_B$3. Since $\mathcal H_A\otimes\mathcal H_B$4 in general, this distinction is essential (Kumar, 2020).

For a random density matrix $\mathcal H_A\otimes\mathcal H_B$5 from the induced ensemble and a fixed density matrix $\mathcal H_A\otimes\mathcal H_B$6, the exact mean-square Hilbert–Schmidt distance is

$\mathcal H_A\otimes\mathcal H_B$7

Equivalently,

$\mathcal H_A\otimes\mathcal H_B$8

For $\mathcal H_A\otimes\mathcal H_B$9,

$N\times K$0

and in the Hilbert–Schmidt case $N\times K$1,

$N\times K$2

If $N\times K$3 is pure, $N\times K$4; if $N\times K$5 is maximally mixed, $N\times K$6. A notable structural feature is that no spectral detail of $N\times K$7 beyond $N\times K$8 enters the final formula (Kumar, 2020).

For two independent induced density matrices $N\times K$9, possibly with different ancilla dimensions $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$0,

$|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$1

In the complex Hilbert–Schmidt specialization,

$|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$2

For large $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$3, this is asymptotically $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$4, matching known large-$|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$5 behavior (Kumar, 2020).

The same paper derives corresponding exact formulas for non-normalized Wishart matrices. If $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$6 is Wishart with $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$7 degrees of freedom and $|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$8 is a fixed Hermitian matrix,

$|\psi\rangle\in\mathcal H_A\otimes\mathcal H_B$9

using

$\mu_{N,K}$00

For two Wishart matrices $\mu_{N,K}$01,

$\mu_{N,K}$02

These Wishart results are mapped to the fixed-trace formulas through an exact Laplace-transform procedure: one introduces an auxiliary variable in $\mu_{N,K}$03, takes the Laplace transform, rescales $\mu_{N,K}$04, and then inverts the transform at $\mu_{N,K}$05. No Weingarten calculus is required; unitary invariance and Gaussian/Wishart moments suffice (Kumar, 2020).

4. Asymptotic regimes and relation to other invariant ensembles

In the Hilbert–Schmidt ensemble $\mu_{N,K}$06, the average purity satisfies

$\mu_{N,K}$07

Consequently, for a fixed comparator $\mu_{N,K}$08,

$\mu_{N,K}$09

If $\mu_{N,K}$10 is pure, the mean-square distance approaches $\mu_{N,K}$11; if $\mu_{N,K}$12, it behaves as $\mu_{N,K}$13 and therefore vanishes as $\mu_{N,K}$14. For two independent Hilbert–Schmidt states,

$\mu_{N,K}$15

A plausible implication is that high-dimensional Hilbert–Schmidt states are typically close to maximally mixed and only weakly separated from each other in Hilbert–Schmidt geometry (Kumar, 2020).

The Hilbert–Schmidt ensemble is also the baseline from which several structured invariant ensembles are defined. In the broader family

$\mu_{N,K}$16

the choice $\mu_{N,K}$17, $\mu_{N,K}$18 produces the Hilbert–Schmidt ensemble, whereas $\mu_{N,K}$19, $\mu_{N,K}$20, $\mu_{N,K}$21 yields the Bures ensemble. This places Hilbert–Schmidt measure at the simplest point in a hierarchy of unitarily invariant random-state models (Zyczkowski et al., 2010).

The distinction from the Bures ensemble is explicit at the level of eigenvalue statistics. The Bures eigenvalue joint density is

$\mu_{N,K}$22

whereas the Hilbert–Schmidt density lacks both the $\mu_{N,K}$23 factor and the $\mu_{N,K}$24 denominator. The constructive Bures algorithm correspondingly uses both a Ginibre matrix and a Haar unitary,

$\mu_{N,K}$25

This suggests that the Hilbert–Schmidt ensemble should be regarded not merely as another unitary-invariant measure, but as the Euclidean benchmark against which more strongly purity-biased monotone-metric ensembles are compared (Zyczkowski et al., 2010).

Beyond Bures, the structured-ensemble paper describes superpositions of $\mu_{N,K}$26 random maximally entangled states and product-Ginibre constructions obtained from selective measurements in maximally entangled bases. Their limiting spectral laws include the arcsine law, the family $\mu_{N,K}$27, and the Fuss–Catalan distributions $\mu_{N,K}$28, while the Hilbert–Schmidt ensemble reappears as the Marchenko–Pastur baseline at $\mu_{N,K}$29 (Zyczkowski et al., 2010).

5. Low-dimensional determinantal moments and separability

For $\mu_{N,K}$30 bipartite systems, the Hilbert–Schmidt ensemble has a particularly explicit determinantal theory. The two-rebit state space is $\mu_{N,K}$31-dimensional, the two-qubit state space is $\mu_{N,K}$32-dimensional, and the Hilbert–Schmidt measure is the unitarily invariant flat measure on these convex bodies. In eigenvalue variables $\mu_{N,K}$33, $\mu_{N,K}$34, the Hilbert–Schmidt density is proportional to

$\mu_{N,K}$35

with $\mu_{N,K}$36 for two-rebits and $\mu_{N,K}$37 for two-qubits. The normalization constants of the eigenvalue-simplex densities are $\mu_{N,K}$38 for $\mu_{N,K}$39 and $\mu_{N,K}$40 for $\mu_{N,K}$41 (1207.1297).

In $\mu_{N,K}$42, the Peres–Horodecki criterion is necessary and sufficient: a state is separable if and only if its partial transpose is positive semidefinite, equivalently if and only if $\mu_{N,K}$43. The determinant range is

$\mu_{N,K}$44

while the “balanced” variable satisfies

$\mu_{N,K}$45

These bounds permit moment-based reconstruction of the $\mu_{N,K}$46 distribution under Hilbert–Schmidt measure (1207.1297).

Exact determinantal moments are available in closed form. For two-qubits $\mu_{N,K}$47,

$\mu_{N,K}$48

More generally, the mixed ratios

$\mu_{N,K}$49

are rational functions of $\mu_{N,K}$50 that can be written as ratios of degree-$\mu_{N,K}$51 polynomials. For $\mu_{N,K}$52, the two-qubit and two-rebit cases are

$\mu_{N,K}$53

and

$\mu_{N,K}$54

The first Hilbert–Schmidt moments of $\mu_{N,K}$55 are

$\mu_{N,K}$56

For the balanced variable,

$\mu_{N,K}$57

A central contrast is that the analogous Bures ratios appear as degree-$\mu_{N,K}$58 polynomials in $\mu_{N,K}$59, not degree-$\mu_{N,K}$60 (1207.1297).

These exact Hilbert–Schmidt moments support inverse moment reconstructions of the $\mu_{N,K}$61 density, for example via Legendre-polynomial reconstruction of Provost. In this way one obtains exact Hilbert–Schmidt separability probabilities: $\mu_{N,K}$62 This makes the Hilbert–Schmidt ensemble a quantitatively tractable reference measure for entanglement geometry in low-dimensional bipartite state spaces (1207.1297).

6. Qutrit Hilbert–Schmidt geometry and classicality indicators

For qutrits, the Hilbert–Schmidt ensemble can be developed directly from the metric-induced measure on

$\mu_{N,K}$63

with invariance under the adjoint action of $\mu_{N,K}$64. The Hilbert–Schmidt metric is

$\mu_{N,K}$65

For the regular stratum of nondegenerate spectra,

$\mu_{N,K}$66

and the Hilbert–Schmidt eigenvalue joint density on the ordered simplex is

$\mu_{N,K}$67

For degenerate strata of partition type $\mu_{N,K}$68, the density becomes

$\mu_{N,K}$69

with Haar-distributed angular variables on the corresponding coset $\mu_{N,K}$70 (Khvedelidze et al., 2022).

The qutrit state space is stratified as

$\mu_{N,K}$71

The regular stratum $\mu_{N,K}$72 has nondegenerate spectra and orbit space equal to the interior of an ordered simplex face. The degenerate stratum $\mu_{N,K}$73 corresponds to spectral types $\mu_{N,K}$74 and $\mu_{N,K}$75. The maximally mixed state is the single-point stratum $\mu_{N,K}$76 (Khvedelidze et al., 2022).

Within this geometry, “classical states” are those with everywhere positive qutrit Wigner function: $\mu_{N,K}$77 The associated geometric-probability indicator is

$\mu_{N,K}$78

with a stratum-wise version $\mu_{N,K}$79. Because of $\mu_{N,K}$80 invariance, the ratios reduce to eigenvalue integrals over orbit-space regions determined by Wigner-function positivity. For the Hilbert–Schmidt ensemble, the paper derives closed forms on both the regular and degenerate strata. On the regular stratum,

$\mu_{N,K}$81

with representative values

$\mu_{N,K}$82

On the degenerate stratum,

$\mu_{N,K}$83

with

$\mu_{N,K}$84

and

$\mu_{N,K}$85

The global Hilbert–Schmidt indicator has minimum

$\mu_{N,K}$86

and satisfies

$\mu_{N,K}$87

Thus the Hilbert–Schmidt qutrit indicator is exactly symmetric in the moduli parameter $\mu_{N,K}$88, while the corresponding Bures and Bogoliubov–Kubo–Mori indicators display mild symmetry breaking (Khvedelidze et al., 2022).

The qutrit comparison also shows a consistent ordering across regular and degenerate strata: $\mu_{N,K}$89 for the probability of Wigner-function positivity. Degenerate strata are more classical than the regular stratum for all three ensembles. This suggests that, within the geometric-probability framework used there, the Hilbert–Schmidt ensemble places comparatively more weight on regions of state space compatible with positive Wigner functions, while higher-symmetry strata amplify that tendency (Khvedelidze et al., 2022).