Reduced Quantum Filter in Quantum Systems

• A reduced quantum filter is a low-dimensional model that approximates key measurement outcomes in open quantum systems using symmetry and projection techniques.
• It significantly reduces computational complexity by compressing state representation from O(n^2) to O(n) dimensions, especially in QND measurement scenarios.
• This approach enhances real-time estimation and feedback control, providing scalable solutions for quantum optics, superconducting circuits, and other complex systems.

A reduced quantum filter is a low-dimensional dynamical model that approximates or exactly reproduces the evolution of relevant statistics (often measurement outcomes or expectation values of specific observables) for an open quantum system subject to continuous or discrete measurement and control. These filters are pivotal for scalable quantum control and estimation, offering substantial dimensionality and complexity savings over generic quantum filtering equations by exploiting physical symmetries, quantum non-demolition (QND) structure, and information-theoretic or geometric reduction techniques.

1. Mathematical Foundations and Problem Statement

The classical quantum filter—exemplified by the stochastic master equation (SME) or Belavkin filtering equation—propagates the conditional state $\rho_t$ of an open quantum system under measurement. For a quantum system of Hilbert space dimension $n$, the density operator $\rho_t$ is an $n\times n$ Hermitian, trace-one matrix, and the SME generically governs the evolution with stochastic increments driven by measurement records and quantum noise, e.g.,

$$d\rho_t = -i[H, \rho_t] dt + \mathcal{D}[L](\rho_t) dt + \text{measurement terms}$$

where $$\mathcal{D}[L](\rho) = L\rho L^\dagger - \frac{1}{2}\{L^\dagger L, \rho\}$$ is the Lindblad dissipator.

The high computational cost associated with integrating $n^2-1$ coupled nonlinear stochastic equations motivates dimensional reduction: given that in most applications, only a subset of the information in $\rho_t$ is required (e.g., the population of certain subspaces, or the expectation values of specific observables), reduced quantum filters are sought that propagate a lower-dimensional estimate sufficient for real-time inference or control (Grigoletto et al., 23 Jan 2025, Liang et al., 11 Nov 2025, Grigoletto et al., 19 Mar 2024).

2. Exact Reduction in QND Measurement and Model-Manifold Construction

The most transparent setting for exact reduction is continuous-time QND measurement, where both the Hamiltonian $H$ and measurement operator $L$ are diagonalizable in a common basis and $L=L^\dagger$, $[H, L] = 0$ (Ramadan et al., 2023). In this regime, the system’s evolution conditioned on continuous measurement remains confined to an invariant manifold $M$ parameterized by a set of real coordinates $(\theta_k, \alpha_{k\ell}, \gamma_j)$. The system’s (unnormalized) conditional state evolves as

$$\bar{\rho}_\phi = \exp\left(\frac{1}{2} L_\theta + \frac{i}{2} H_\gamma\right) \rho_\alpha \exp\left(\frac{1}{2} L_\theta - \frac{i}{2} H_\gamma\right)$$

with $L_\theta = \sum_k \theta_k P_k$, $H_\gamma = \sum_j \gamma_j Q_j$, and the projectors $\{P_k\}$, $\{Q_j\}$ spanning the spectral manifold. The dynamics of the coordinates obey closed-form stochastic differential equations (SDEs) such as

$$d\theta_k = - (1+\eta) \lambda_k^2 dt + 2\sqrt{\eta}\,\lambda_k\, dY_t$$

drastically reducing the system’s representation from dimension $n^2-1$ to $O(n)$, and allowing the reconstruction of $\bar{\rho}_t$ via low-dimensional integration (Ramadan et al., 2023).

3. Exponential Family Manifolds, Projection Filters, and Information Geometry

When the QND conditions do not strictly hold, further reduction can be achieved via projection filtering onto an exponential family manifold

$$S = \left\{ \bar{\rho}_\theta = \exp\left( \frac{1}{2} \sum_{i=1}^m \theta_i A_i \right) \rho_0 \exp\left( \frac{1}{2} \sum_{i=1}^m \theta_i A_i \right) : \theta \in \mathbb{R}^m \right\}$$

where the $A_i$ are commuting self-adjoint operators such as projectors onto the eigenbases of $L$. The filter dynamics are projected onto the tangent bundle using the Fisher–Rao (quantum Fisher information) metric. The resulting projected SDE for $\theta$ is

$$d\theta = G(\theta)^{-1} \big[ \Xi(\theta) dt + \Gamma(\theta) \circ dY_t \big]$$

where the Gram matrix $$G_{ij}(\theta) = \mathrm{Tr}[ \bar{\rho}_\theta A_i A_j ]$$, and the error of the approximation can be exactly quantified; for specific choices of $A_i$ and in the QND limit, the error vanishes (Ramadan et al., 2023, Gao et al., 2017).

4. Algebraic and Conditional Expectation-Based Model Reduction

A systematic algebraic reduction framework is provided by projecting the filter’s state and dynamical superoperators onto a minimal *-subalgebra $\mathcal{A}$ containing all observables of interest via the quantum conditional expectation $E_\mathcal{A}$. For both continuous- (Grigoletto et al., 23 Jan 2025) and discrete-time (Grigoletto et al., 19 Mar 2024) quantum systems, this yields a reduced filter propagating the relevant sub-algebra, which can have dimension $\sum_k n_k^2 \ll n^2$ (with $n_k$ the sizes of block-diagonal components).

If $\tau_t \in \mathcal{A}$ is the reduced (possibly unnormalized) state, its evolution is governed by projected Lindblad, measurement, and counting superoperators: $$d\check{\tau}_t = \check{Q}(\check{\tau}_t) dt + \sum_j \check{G}_{D_j}(\check{\tau}_t) dY_t^j + \sum_j (\check{K}_{C_j} - I)(\check{\tau}_t) dN_t^j$$ with explicit mappings from the full to reduced operators via $E_\mathcal{A}$ (Grigoletto et al., 23 Jan 2025, Grigoletto et al., 19 Mar 2024). The method preserves the exact evolution of prescribed expectation values and measurement statistics, revealing the reduced filter as an exact “observable-space” realization.

5. Robustness, Stability, and Feedback Control Applications

Reduced quantum filters have been shown to guarantee robust estimation and stabilization in high-dimensional and uncertain open quantum systems, particularly under time-varying Hamiltonian perturbations and non-ideal measurement parameters. In the feedback stabilization scenario, only the weights on the diagonal blocks in the QND basis need to be tracked: $$q_j(t) = \mathrm{Tr}[\rho(t)\Pi_j], \qquad \sum_j q_j(t) = 1$$ with a reduced filter SDE propagating this probability simplex, resulting in an $O(N)$ variable filter for Hilbert space of dimension $N$, in contrast to generic $O(N^2)$ cost (Liang et al., 11 Nov 2025). Convergence and robustness are rigorously established using Lyapunov techniques, and the approach scales efficiently to high-dimensional systems or online measurement-based feedback with time-varying uncertainties.

6. Related Constructions: Quantum Filtering in Compression, Data, and Photonic Domains

Compression and Quantum JPEG: The reduced quantum filter concept appears in amplitude-encoding and quantum data compression, with protocols such as “quantum JPEG” implementing a sharp spectral cutoff in the quantum Fourier basis, discarding high-frequency qubits, and reconstructing a compressed image state with analytically trackable fidelity and resource scaling. The procedure can surpass classical compression cost for certain parameter regimes (Roncallo et al., 2023).

Amplitude Reduction for Search and Preprocessing: In quantum algorithms for data filtering or searching, a reduced quantum filter can be realized as a data-dependent unitary that suppresses amplitudes of undesirable elements via parameterized rotation gates, offering one-shot probability amplification in small search spaces or efficient pre-processing for amplitude-amplification-based search (Zakharova et al., 23 Apr 2025).

Photonic and Mode Filtering: In single-photon experiments, reduced quantum filters correspond to single-mode quantum buffers that project noisy emission onto dominant temporal–spectral modes via tailored control pulses, maximizing output indistinguishability and brightness, a feat unattainable with incoherent (classical) filters (Gao et al., 2019).

7. Implementation, Complexity, and Outlook

Implementation of reduced quantum filters typically involves the following workflow (for continuous-time SME):

1. Specify the observables or measurement statistics to be preserved.
2. Identify and construct the minimal model manifold (QND-invariant subspaces, exponential manifold, or block-diagonal algebra).
3. Project the filter equations, using the Fisher–Rao metric or quantum conditional expectation, onto the reduced space.
4. Integrate the resulting low-dimensional SDE or update rule, reconstructing observables as needed.
5. Guarantee stability and error bounds via theoretical analyses, with explicit analytic error estimates available in several cases (Ramadan et al., 2023, Gao et al., 2017, Liang et al., 11 Nov 2025).

These reductions yield dominant order-of-magnitude savings in memory and computation per update step, enabling real-time feedback and control in quantum optics, superconducting circuits, and quantum information processing. For high-dimensional sensors or many-body systems, the systematic use of reduced quantum filters is an enabling technology for scalable estimation and robust operation (Grigoletto et al., 23 Jan 2025, Liang et al., 11 Nov 2025).

---

Key References:

• “Exact solution and projection filters for open quantum systems subject to imperfect measurements” (Ramadan et al., 2023)
• “Quantum model reduction for continuous-time quantum filters” (Grigoletto et al., 23 Jan 2025)
• “Exact model reduction for discrete-time conditional quantum dynamics” (Grigoletto et al., 19 Mar 2024)
• “An Exponential Quantum Projection Filter for Open Quantum Systems” (Gao et al., 2017)
• “Stabilization of Time-Varying Perturbed Quantum Systems via Reduced Filters” (Liang et al., 11 Nov 2025)
• Additional applications: (Roncallo et al., 2023, Zakharova et al., 23 Apr 2025, Gao et al., 2019)