Quantum Invariant Filtering

• Quantum Invariant Filtering is defined by exploiting symmetries and invariant structures for robust quantum state estimation and control.
• It maps frequency-domain filter responses onto continuous control Hamiltonians, achieving high-fidelity noise suppression and extended coherence times.
• QIF enables adaptive error mitigation and practical implementations across platforms such as NV centers, superconducting qubits, and trapped ions.

Quantum Invariant Filtering (QIF) refers to a theoretical and experimental suite of frameworks that exploit symmetries and dynamical invariants to design or analyze quantum filtering procedures—continuous or discrete-time update rules for estimating quantum states from partial, noisy measurements—such that the filtering equations and processes respect specified symmetry constraints or invariant structures. In quantum control, QIF also encompasses the direct synthesis of continuous-time control Hamiltonians to realize prescribed frequency-domain responses, thereby mapping finite-impulse-response (FIR) filters onto physically implementable quantum dynamics. QIF has applications spanning quantum control, measurement, error mitigation, and quantum information, with recent experimental validation demonstrating noise suppression, error robustness, and coherence preservation beyond established pulsed control protocols (Cangemi et al., 18 Jun 2025, Gough, 2022, Sritharan et al., 2023, Gough, 2018).

1. Mathematical Formulation of Quantum Invariant Filtering

Quantum filtering problems are formulated for a quantum system (Hilbert space $\mathcal{H}_s$) in contact with one or more monitored bosonic fields (Fock space $\mathcal{F}$). The global unitary evolution $U(t)$ is governed by a singular Hamiltonian,

$$\Upsilon(t) = H \otimes I_f + i \sum_k L_k \otimes b_k^*(t) - i \sum_k L_k^* \otimes b_k(t)$$

where $H$ is the system Hamiltonian, $\{L_k\}$ are Lindblad operators, and $b_k(t), b_k^*(t)$ are input quantum white-noise operators (Gough, 2018).

The system evolution is described by a quantum stochastic differential equation (QSDE) in Itō form,

$$dU(t) = \left\{ \sum_{j,k}(S_{jk}-\delta_{jk})\otimes d\Lambda_{jk}(t) +\sum_j L_j\otimes dB_j^*(t) -\sum_{j,k}L_j^* S_{jk}\otimes dB_k(t) -(\tfrac{1}{2}\sum_\ell L_\ell^* L_\ell + i H)\otimes dt \right\} U(t)$$

with $S$ a unitary scattering matrix.

Continuous measurement of an output quadrature $Y^\mathrm{out} (t)$ allows the application of conditional expectation onto the observed algebra, enabling the derivation of the nonlinear quantum filtering (Belavkin) equation for the conditioned state $\rho_t$,

$$d\rho_t = -\frac{i}{\hbar}[H,\rho_t]\,dt + \mathcal{D}[L](\rho_t)\,dt + \mathcal{H}[L](\rho_t)\,dW_t$$

where $\mathcal{D},\mathcal{H}$ are Lindblad superoperators and $dW_t$ is the innovations process (Gough, 2018, Gough, 2022).

QIF imposes invariance constraints by considering a symmetry group $G$ with a unitary representation $U(g)$ on $\mathcal{H}_s$. The filtering equations are said to be invariant if,

$$U(g) H U(g)^\dagger = H, \quad U(g) L_k U(g)^\dagger = L_k\ \forall\,k,g\in G$$

ensuring the filter (the conditional state or observable process) transforms covariantly under the group's action (Gough, 2018, Gough, 2022).

2. Dynamical-Invariant Synthesis and Frequency-Domain QIF

A distinct operationalization of QIF arises in the context of quantum control, where it is leveraged to synthesize control fields enforcing desired spectral filter characteristics. Specifically, starting from a target frequency-domain filter, the Lewis–Riesenfeld invariant formalism is used to derive continuous control Hamiltonians capable of realizing arbitrary FIR responses on a qubit: $$H(t) = \frac{\Delta(t)}{2} \sigma_z - \frac{\epsilon(t)}{2} \sigma_x$$ with the invariant $I(t)$ parameterized as (Cangemi et al., 18 Jun 2025): $$I(t) = \frac{1}{2}\bigl[ -\cos\alpha(t)\,\sigma_x + \sin\alpha(t)\sin\beta(t)\,\sigma_y + \sin\alpha(t)\cos\beta(t)\,\sigma_z \bigr]$$ Boundary conditions and the invariance equation $i[H(t),I(t)] + \partial_t I(t) = 0$ allow analytic derivation of $\epsilon(t)$ and $\Delta(t)$ for the desired impulse response $\mathcal{H}(t) = \cos\beta(t)$, yielding a direct and invertible mapping from frequency domain to time-dependent drive (Cangemi et al., 18 Jun 2025).

This approach supports construction of single-band, multi-band, and phase-sensitive filters via proper selection of $\mathcal{H}(t)$, which in the frequency domain corresponds to an analytically prescribed transfer function $F(\omega)$. The resulting modulations can be implemented as continuous microwave amplitude controls in physical qubit platforms.

3. Symmetry, Reduction, and Invariant Measures

QIF exploits invariance to enable dimension reduction and statistical simplification. When $G$ is a symmetry of the system-measurement pair, the conditional state (or its equivalent filter representation) is supported on the fixed-point subalgebra or the invariant subspace. For instance, for a spin-$1/2$ system under $U(1)$ symmetry, one can reduce the filtering problem to a classical SDE for the excited-state probability (Gough, 2018): $$dp_t = -\kappa p_t\,dt + 2\sqrt{\kappa}p_t(1-p_t)\,dW_t$$ In systems exhibiting phase or permutation symmetry, filters inherit corresponding simplifications, enabling practical implementation in high-dimensional or networked systems (Gough, 2018, Gough, 2022).

For infinite-dimensional quantum spin systems modeled by stochastic partial differential equations (SPDEs) on separable Hilbert spaces $H$, the invariant filtering dynamics yield measure-valued solutions $\Pi_t \in \mathcal{P}(H)$. These solutions are characterized by Markov and Feller properties, and, under suitable ergodicity and backward-forward $\sigma$-field conditions, admit unique invariant measures $\mathfrak{M}$ whose barycenter is the ergodic measure of the signal process. This structure ensures stability and well-posedness for long-time estimation (Sritharan et al., 2023).

4. Quantum Filtering Equations and Error Analysis

Quantum filtering admits both normalized (Belavkin–Kushner–Stratonovich) and unnormalized (Belavkin–Zakai) forms. For a filtered (conditional) observable $X$,

$$d\pi_t(X) = \pi_t(\mathscr{L}X)\,dt + \big[ \pi_t(XL+L^*X) - \pi_t(X)\pi_t(L+L^*) \big] dI(t)$$

where $dI(t) = dY^{\text{out}}(t) - \pi_t(L+L^*)dt$ is the innovations process (Gough, 2022).

In more general, infinite-dimensional settings, the Fujisaki–Kallianpur–Kunita (FKK) and Zakai-type filtering equations provide measure-valued, continuous-time update laws: $$d\Pi_t[f] = \Pi_t[Af]\,dt + \sum_i (\Pi_t[f h_i + D_i f] - \Pi_t[f]\Pi_t[h_i])\,d\nu_i(t)$$

$$d\Gamma_t[f] = \Gamma_t[Af]\,dt + \Gamma_t[f h] \cdot dY(t)$$

where $\Pi_t[f]$ are conditional expectations, $\Gamma_t$ their unnormalized counterparts, and the noise processes $d\nu_i$, $dY(t)$ capture innovations from continuous measurement (Sritharan et al., 2023).

QIF also yields Riccati-type equations for the evolution of the error covariance, extending the classical Kalman–Bucy approach to quantum and infinite-dimensional settings. This facilitates uncertainty quantification and robust state estimation (Sritharan et al., 2023).

5. Experimental Realizations and Comparative Performance

QIF—specifically the frequency-domain control variant—has been experimentally implemented on a single nitrogen-vacancy (NV) center in diamond. The synthesized control fields realize targeted passbands with high fidelity, as confirmed by fluorescence contrast measurements tracing $|F(\omega)|^2$ in the passband. In direct comparison to Carr-Purcell-Meiboom-Gill (CPMG) pulsed dynamical decoupling, QIF produces smoother, single-band frequency responses without harmonic sidelobes, achieving noise suppression and coherence times ($T_2 \approx 2.7$ ms) nearly two orders of magnitude longer than CPMG sequences of equal duration ($\sim 0.07$–$0.14$ ms), and maintains greater than $80\%$ contrast even under $\pm 50\%$ amplitude calibration errors (Cangemi et al., 18 Jun 2025).

QIF approaches are shown to be platform-agnostic: their requirements are limited to the ability to generate smooth, continuous control fields and access to a $\sigma_z$-coupled signal (or analogues), enabling straightforward adaptation to superconducting qubits, trapped ions, and NMR systems (Cangemi et al., 18 Jun 2025).

6. Extensions, Robustness, and Limitations

The QIF paradigm generalizes to multi-qubit systems, with ongoing research focused on extending invariant filtering to collective invariants supporting entangling gates and multi-spin quantum sensing. Embedding QIF within adaptive, closed-loop control architectures promises real-time noise spectrum estimation and mitigation.

Fundamental limitations arise from the perturbative treatment's range of validity (weak driving $\delta t_f \lesssim 1$), bandwidth constraints and smoothing requirements of experimental hardware, and the increased complexity of synthesizing sharply selective filters, which require longer interaction times and higher control resolution. For stronger signals or target filters with high time-frequency selectivity, numerical optimal control methods may become necessary (Cangemi et al., 18 Jun 2025, Sritharan et al., 2023).

7. Connections, Implications, and Open Problems

QIF unites core ideas from quantum probability, stochastic calculus (Hudson–Parthasarathy), conditional expectations on operator algebras (Takesaki), and classical filtering theory. Its symmetry-driven structure supports dimension reduction, robust state estimation, and stabilization protocols, with applications in quantum error correction, feedback stabilization, quantum control networks, and information processing (Gough, 2022, Gough, 2018, Cangemi et al., 18 Jun 2025).

Open directions include extension to non-Markovian baths, exploration of invariant filtering in relativistic and networked quantum systems, the development of white-noise formulations for pathwise analysis, and the study of QIF-induced dynamics at quantum phase transitions. QIF stands as a mathematically rigorous and experimentally validated framework for state estimation and control in symmetry-rich, noisy quantum environments (Sritharan et al., 2023, Cangemi et al., 18 Jun 2025).