Central Limit in Quantum Mixtures

• Central Limit Theorem for Quantum Mixtures is a framework showing how rescaled sums or convolutions of quantum operations converge to Gaussian-like distributions.
• It leverages operator algebra, spectral analysis, and quantum stochastic processes to provide explicit rates, mixture weights, and large deviation principles in discrete and continuous systems.
• The theorem unifies results from open quantum walks, discrete qudit systems, bosonic modes, and quantum neural networks, offering practical tools for error estimation and device benchmarking.

The central limit theorem (CLT) for quantum mixtures describes the convergence of rescaled sums, convolutions, or compositions of quantum states or channels to a limiting distribution, often characterized as a mixture of Gaussian (or "Gaussian-like") quantum states. This unifies several phenomena in open quantum systems, discrete and continuous variable settings, non-commutative probability, and quantum neural mixture models, and generalizes classical limit theorems into the quantum domain. The theory leverages operator algebra, quantum stochastic processes, and the spectral theory of quantum channels, providing precise asymptotic characterizations (including rates and large deviations) of quantum mixtures.

1. Central Limit Theorem for Homogeneous Open Quantum Walks

For a homogeneous open quantum walk (OQW) on a lattice $V\subset\mathbb{R}^d$ with a finite-dimensional Hilbert space $\mathcal{H}$ and Kraus operators $\{L_i\}_{i=1}^v$, the classical position process $(X_n)$ associated with quantum trajectories and a fixed initial state $\rho$ displays a remarkable central limit mixing phenomenon. The law of the rescaled position,

$$\mu_{\rho,n} = \text{Law}_\rho\left(\frac{X_n - X_0}{\sqrt{n}}\right)$$

converges, in the weak sense, to a finite convex combination of Gaussian measures: $$\frac{X_n-X_0}{\sqrt{n}}\xrightarrow{d} \sum_{\alpha\in A} a_\alpha(\rho)\, \mathcal{N}(m_\alpha, D_\alpha)$$ where $A$ is a finite index set determined by the spectral decomposition of the local channel $\Phi(\sigma)=\sum_i L_i\sigma L_i^*$, $a_\alpha(\rho)$ are weights depending on the initial state, and $\mathcal{H}$0 denotes a Gaussian with mean $\mathcal{H}$1, covariance $\mathcal{H}$2. The means and covariances are determined by spectral data of the deformed channel $\mathcal{H}$3 and its restriction to minimal enclosures in the decomposition of $\mathcal{H}$4 (Carbone et al., 2021).

This mixture structure arises purely from quantum reducibility; if the local channel is irreducible, the limit reduces to a single Gaussian. For reducible or periodic quantum channels, the limit remains a mixture, with no irreducibility required. The explicit formulas for weights, means, and covariances involve absorption operators, invariant states, and derivatives of the logarithm of the spectral radius.

A full large deviation principle is established in the fast-recurrent case ($\mathcal{H}$5), with good rate functions formulated via the Legendre transform of the logarithm of the spectral radius. For general cases, one still obtains matching upper and lower bounds for large deviations.

2. Quantum Central Limit Theorem for Discrete Mixtures

In discrete-variable (DV) quantum systems (qudits with prime local dimension), the central limit theorem for quantum mixtures employs a Clifford-based quantum convolution operation, generalizing addition to the noncommuting operator regime. Iterating this convolution (e.g., beam-splitter or Hadamard-type) for an initial state $\mathcal{H}$6, the N-fold convolution $\mathcal{H}$7 converges exponentially fast (in 2-norm, trace norm, or relative entropy) to its associated minimal stabilizer-projection state (MSPS): $\mathcal{H}$8 where $\mathcal{H}$9 is the MSPS and the “magic gap” $\{L_i\}_{i=1}^v$0 quantifies the spectral gap in the Fourier space (the analog of the spectral gap in Markov theory). Stabilizer states act as finite-dimensional analogs of Gaussians in this setting; they extremize entropy and Fisher information relative to convolution.

This framework extends to quantum channels and their Choi–Jamiołkowski representations, showing that repeated convolution of quantum channels with zero-mean Choi states drives convergence to the stabilizer channel, at an exponential rate determined by the magic gap (Bu et al., 2023, Bu et al., 2023, Bu et al., 2024).

3. Quantum Central Limit Theorem for Continuous Variables and Optimal Convergence Rates

For continuous-variable systems (bosonic modes), the standard quantum CLT generalizes the Cushen–Hudson result: the n-fold symmetric convolution (implemented by a network of beam splitters) of a centered m-mode state $\{L_i\}_{i=1}^v$1 converges to its Gaussian Gaussification $\{L_i\}_{i=1}^v$2 with the same first and second moments: $\{L_i\}_{i=1}^v$3 For finite third moments, the convergence in trace distance is optimal at $\{L_i\}_{i=1}^v$4; for finite fourth moments (plus a correction for $\{L_i\}_{i=1}^v$5), the relative entropy converges at the optimal rate $\{L_i\}_{i=1}^v$6. These rates are tight and directly parallel the classical Berry–Esseen and entropic CLTs. Technical tools include quantum analogues of Edgeworth expansions and operator-level bounds (Beigi et al., 2024, Becker et al., 2019).

In non-i.i.d. cascades (e.g., repeated interaction with an environment through beam splitters), a similar CLT and explicit convergence rates hold, with the effective state of the environment channel converging in trace norm (and hence in diamond norm for the channel) to the corresponding Gaussian attenuator.

4. Mixtures in Quantum Neural Networks and Non-Communtative Probability

Central limit behavior also governs mixtures in quantum neural network (QNN) architectures, specifically in mixtures of experts (MoE) paradigms. For an MoE generated by N quantum experts with parameters evolved via gradient flow, the fluctuations of the empirical measure about its mean-field limit converge to a Gaussian process on the dual of a Sobolev space, solving a linearized transport equation. The covariance structure of the limit is governed by the evolution of the quantum neural tangent kernel (QNTK), providing a dynamical analogue of the Gaussian CLT in parameter space and function space. A complementary Sanov principle yields a full large deviation principle for the pathwise empirical measure, paralleling quantum Sanov and mean-field results (Hernandez, 19 Jun 2026).

In abstract non-commutative probability, quantum mixtures with generalized commutation relations (including $\{L_i\}_{i=1}^v$7-deformations) yield non-commutative CLTs whose limits ("quantum Gaussians") are characterized by combinatorial parameters (crossings, nestings) and explicit Fock space realizations. The classical, free, and Boolean probability limits are unified within this framework (Blitvić, 2012).

5. Mixtures, Partial Distinguishability, and Open Quantum Systems

The quantum CLT on the level of mixtures extends to physical setups involving partial distinguishability, as in bosonic linear optics. If each boson carries an internal degree of freedom (e.g., polarization, arrival time), then the reduced state on a subsystem after an unbiased interferometer converges to a multimode Gaussian state parameterized by the Gram (overlap) matrix defining mode distinguishability. The photon-number statistics in output ports interpolate between geometric (fully indistinguishable) and Poissonian (fully distinguishable) laws, providing an experimentally accessible witness of residual distinguishability (Robbio et al., 2024).

In homogeneous open quantum walks and measurement records, CLT-type behavior appears both in the trajectory statistics (position increments) and in the mixture decomposition arising from reducible dynamics. When the system possesses block-diagonal Kraus operators, the asymptotic law is a mixture of Gaussians, with weights determined by initial block-occupations; each block evolves independently toward a classical limit characterized by its block-specific drift and covariance (Attal et al., 2012, Carbone et al., 2021).

6. Spectral Origins, Rates of Convergence, and Large Deviations

The spectral theory of quantum channels and convolutions underpins the origin of Gaussian mixtures in quantum CLTs. Minimal enclosures, invariant subspaces, and spectral properties—such as the presence or absence of irreducibility—control the decomposition into mixture components. Concrete formulas for rates and large deviations, including precise spectral radius differentiations and Legendre transforms, yield a quantitative, non-asymptotic theory of convergence.

A key feature of the quantum CLT in discrete settings is the appearance of exponential convergence rates—controlled by the magic gap—in contrast to the polynomial rates (trace distance $\{L_i\}_{i=1}^v$8, relative entropy $\{L_i\}_{i=1}^v$9) in the continuous-variable case, where mixing phenomena are less pronounced due to infinite dimensionality and smoothness requirements (Bu et al., 2023, Bu et al., 2023, Bu et al., 2024, Beigi et al., 2024).

System/Model	Limiting Law	Rate/Mechanism
Open quantum walks (finite $(X_n)$0)	Mixture of Gaussians	Spectral analysis of channel
Discrete-variable qudits	Stabilizer (MSPS, "DV-Gaussian")	Exponential (magic gap)
Bosonic modes (CV Gaussian)	Gaussian state	$(X_n)$1
Quantum NNs (MoE parameter laws)	Gaussian process	Linearized flow (NTK)
Noncommutative probability ($(X_n)$2)	$(X_n)$3-Gaussian law	Wick rule, combinatorics

7. Implications and Applications

The CLT for quantum mixtures provides a rigorous foundation for the emergence of classicality from quantum trajectories, the Gaussification of quantum states in both DV and CV settings, and the universality of quantum fluctuations in multipartite, open, and totally disconnected regimes. It supplies theoretical tools for benchmarking quantum devices, diagnosing partial distinguishability, understanding decoherence, and quantifying the resource-theoretic aspects of "magic" and stabilizer/non-stabilizer mixtures.

In quantum information theory, these results clarify the stability of channel capacities under mixing and reveal how quantum mixtures approach entropy maximizers. In quantum computation, the stabilizer/“DV-Gaussian” fixed points act as fenceposts for resource conversion and magic distillation protocols.

The detailed convergence rates, especially the explicit dependence on spectral gaps and moments, enable scalable error estimation and performance prediction in quantum networks, boson sampling, quantum neural backends, and quantum measurement record analysis.