Symmetric Quantum Feedback Control

• Symmetric quantum feedback control is a set of protocols that impose symmetry on measurement operators, feedback Hamiltonians, and channels to streamline quantum state manipulation.
• It integrates both measurement-based and coherent feedback methods, utilizing invariant structures like permutation and phase-space symmetry to optimize state evolution.
• These approaches enable efficient entanglement generation and robust state preparation, achieving high fidelities even under noise and decoherence.

Symmetric quantum feedback control comprises feedback protocols in which symmetry is imposed on the measurement operators, feedback Hamiltonians, admissible state space, or the induced completely positive trace-preserving map. In the recent literature, symmetry appears in at least three distinct technical senses: permutation- and bit-flip-symmetric entanglement generation under continuous measurement, phase-space-covariant control of canonically conjugate quadratures, and symmetry classification of discrete feedback channels through their non-Hermitian superoperator or Bloch-matrix structure (Zhang et al., 2018, Rouillard et al., 2022, Wen et al., 16 Sep 2025). The topic therefore spans both measurement-based and coherent feedback, but its most explicit formulations are presently measurement-conditioned and channel-theoretic rather than purely Hamiltonian.

1. Meanings of symmetry in quantum feedback

The literature distinguishes two broad feedback architectures. In measurement-based feedback control, the system is monitored and the resulting information is used to choose a control action; in coherent feedback control, plant and controller remain quantum and are coupled without an explicit measurement step (Zhang et al., 2014). A precise structural correspondence exists between coherent feedback control and non-selective measurement-based feedback control: both reduce to the same operator-sum evolution

$$\rho_{k+1}=\sum_i K_i \rho_k K_i^\dagger,\qquad \sum_i K_i^\dagger K_i=I,$$

once the controller or measurement outcomes are averaged over (Qi et al., 2014). This correspondence is broken when one retains the real-time conditional state or trajectory, because selective measurement-based feedback exploits information that is absent from the averaged map (Qi et al., 2014).

Within this general setting, “symmetry” is used in several non-identical ways. In multipartite entanglement protocols, symmetry means permutation invariance, bit-flip invariance, or restriction to symmetric and antisymmetric invariant subspaces (Zhang et al., 2018, Zhang et al., 2014). In channel-based formulations, symmetry means Bernard–LeClair-type constraints on the feedback CPTP map or on its doubled-space Bloch matrix (Nakagawa et al., 2024, Wen et al., 16 Sep 2025). In continuous-variable settings, symmetry can mean a phase-space-covariant treatment of an arbitrary quadrature $\hat Q=\alpha \hat q+\beta \hat p$ and its canonically conjugate observable $\hat P$ (Rouillard et al., 2022). A coherent-feedback analogue appears in the driven $\Lambda$-system with two balanced radiative channels,

$$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$

where a microwave pump provides structured phase control in a multichannel delayed loop (Barkemeyer et al., 2019).

2. Feedback architectures and dynamical models

A standard discrete measurement-based model treats the conditional density operator $\rho_k$ as the feedback state. For a chosen measurement $u_k$ with operators $\mathsf M_{u_k}(y)$, the update is

$$\mathbb P(y_k=y\mid u_k,\rho_k)=\operatorname{tr}\!\left(\mathsf M_{u_k}(y)\rho_k \mathsf M_{u_k}(y)^\dagger\right),$$

$$\rho_{k+1}=\mathcal M_{u_k}^{y_k}(\rho_k)=\frac{\mathsf M_{u_k}(y_k)\rho_k \mathsf M_{u_k}(y_k)^\dagger}{\operatorname{tr}\!\left(\mathsf M_{u_k}(y_k)\rho_k \mathsf M_{u_k}(y_k)^\dagger\right)}.$$

Because $\hat Q=\alpha \hat q+\beta \hat p$0 is a sufficient statistic, the measurement-selection problem becomes a controlled Markov process on density matrices, and optimal finite-horizon policies are Markovian in $\hat Q=\alpha \hat q+\beta \hat p$1 (Fu et al., 2014). The same work notes that, if a symmetry group $\hat Q=\alpha \hat q+\beta \hat p$2 acts on states and measurements, one could restrict the state space to symmetry-reduced sufficient statistics or require the control law to be equivariant or invariant under that action (Fu et al., 2014).

Continuous-time feedback is often written in Wiseman–Milburn form. In the harmonic-potential setting, an arbitrary quadrature

$\hat Q=\alpha \hat q+\beta \hat p$3

is monitored, and feedback proportional to the measured signal is generated by

$\hat Q=\alpha \hat q+\beta \hat p$4

The resulting feedback master equation shows that direct Markovian feedback can simultaneously damp the measured quadrature and its conjugate momentum, compensate noise introduced by the measurement, and add a quadratic term $\hat Q=\alpha \hat q+\beta \hat p$5 to the effective Hamiltonian (Rouillard et al., 2022). For the harmonic oscillator, the transformed Hamiltonian remains of oscillator form in the $\hat Q=\alpha \hat q+\beta \hat p$6 variables when

$\hat Q=\alpha \hat q+\beta \hat p$7

which is the clearest phase-space symmetry statement in that work (Rouillard et al., 2022).

3. Symmetry-preserving state preparation and entanglement generation

The most explicit symmetric-control protocols are continuous-measurement schemes for multipartite entanglement generation. The PaQS framework chooses a feedback unitary $\hat Q=\alpha \hat q+\beta \hat p$8 and optimizes the next-step fidelity to a target state $\hat Q=\alpha \hat q+\beta \hat p$9 by setting

$\hat P$0

or equivalently

$\hat P$1

so that the controller contains both a proportional term and a quantum-state-based term (Zhang et al., 2018). In that setting, the symmetry of both measurement and feedback operators is essential for construction of effective protocols: it creates invariant reduced subspaces in which the target becomes a nondegenerate measurement eigenstate, while also preserving locality when only local feedback Hamiltonians are allowed (Zhang et al., 2018).

For Dicke and W states, the natural symmetric measurement and feedback generators are

$\hat P$2

These collective operators preserve the permutation-symmetric subspace of dimension $\hat P$3, spanned by Dicke states $\hat P$4. Inside that reduced sector, the degeneracy of $\hat P$5 is removed, which is why the measurement record becomes directly informative for symmetric target preparation (Zhang et al., 2018). Under perfect measurement efficiency, entangled states can be reached with fidelity approaching unity under non-Markovian feedback control protocols, while Markovian protocols resulting from optimizing the feedback unitaries on ensemble averaged states still yield fidelities above $\hat P$6; the PaQS approach generates W, general Dicke, and GHZ states efficiently, for up to $\hat P$7 in some cases (Zhang et al., 2018).

For GHZ states, permutation symmetry alone is insufficient. One also needs bit-flip symmetry. The symmetric two-body measurement

$\hat P$8

and the symmetric feedback unitary

$\hat P$9

restrict the dynamics to a GHZ-symmetric subspace in which the GHZ target is a nondegenerate eigenstate of the measured observable (Zhang et al., 2018). By contrast, a non-symmetric one-body observable leaves the target in a degenerate eigenspace and produces poor protocols (Zhang et al., 2018).

A broader PaQS formalism for continuous measurement and feedback reproduces adaptive phase measurement, half-parity and full-parity entanglement generation, Dicke-state preparation, and GHZ protocols (Martin, 2020). Its basic controlled SME,

$\Lambda$0

covers most known feedback constructions in that thesis (Martin, 2020). The same work uses symmetric collective measurements

$\Lambda$1

and symmetric feedback Hamiltonians

$\Lambda$2

for permutation-invariant Dicke targets, and introduces balanced pairwise channels

$\Lambda$3

for GHZ-type constructions (Martin, 2020).

4. Channel symmetries, AZ$\Lambda$4 classes, and topological feedback

A discrete feedback cycle can be written as a quantum channel

$\Lambda$5

or, for successive feedback,

$\Lambda$6

The symmetry object is therefore the feedback channel itself, or its doubled-space representation $\Lambda$7, not merely the feedback Hamiltonian (Nakagawa et al., 2024, Wen et al., 16 Sep 2025).

For discrete quantum feedback control with projective measurements, one obtains ten symmetry classes of channels rather than the full Bernard–LeClair classification of generic non-Hermitian matrices (Nakagawa et al., 2024). In the translationally invariant single-particle setting, the relevant topological object is the Bloch matrix $\Lambda$8, and point-gap topology is characterized by

$\Lambda$9

Topological Maxwell demons then realize robust feedback-controlled chiral or helical transport against noise and decoherence (Nakagawa et al., 2024).

A sharper 2025 result establishes that for non-adaptive successive feedback control with bare measurements,

$$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$0

the possible Bernard–LeClair symmetry classes collapse to the ten-fold AZ$$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$1 subclass (Wen et al., 16 Sep 2025). The key spectral reason is Hilbert–Schmidt contractivity of bare measurements: $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$2 with equality only when $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$3; consequently the only unit-modulus eigenvalue is $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$4, which excludes $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$5 symmetry and the $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$6 cases of $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$7, $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$8, and $$\hbar \int \mathrm{d}k \left[ g_k r_k^\dagger \left( \sigma_{13} + \sigma_{23} \right) + H.c. \right],$$9 (Wen et al., 16 Sep 2025). For general non-bare measurements this restriction can be violated. The explicit spin-flipping protocol

$\rho_k$0

realizes a $\rho_k$1-symmetric channel outside the ten-fold classification (Wen et al., 16 Sep 2025).

The chiral Maxwell demon with Gaussian measurement errors shows that topological invariants survive realistic readout imperfections. In that model, the projective case gives

$\rho_k$2

while with Gaussian error $\rho_k$3 the winding about the origin becomes

$\rho_k$4

and the winding near the steady-state eigenvalue remains

$\rho_k$5

showing that the topology around the steady-state mode is robust even when measurement errors preserve some off-diagonal coherence (Wen et al., 16 Sep 2025).

5. Symmetry reduction, optimal policies, and performance limits

Several control theories are not symmetry-specific in their original formulation but are directly reusable once a symmetry-reduced state representation is available. In measurement-selection control, the posterior density matrix is the sufficient statistic, and the optimal finite-horizon arrival-probability objective obeys the Bellman recursion

$\rho_k$6

with an optimal Markov law depending only on the current conditional state (Fu et al., 2014). The same work notes that in symmetric problems one may replace the full density matrix by a reduced symmetric descriptor, such as a permutation-invariant marginal, irreducible-representation block coordinates, or orbit labels, without changing the dynamic-programming architecture (Fu et al., 2014).

For continuous monitoring, performance certification can be formulated through quantum filtering and moment-sum-of-squares relaxations. The hierarchy of convex optimization problems in

$\rho_k$7

furnishes monotonically improving computable bounds on the best attainable performance (Holtorf et al., 2023). A plausible implication is that symmetry-constrained policies or invariant coordinates can be inserted once the symmetry-reduced filtered dynamics retain the polynomial structure required by the SOS program (Holtorf et al., 2023).

Learning-based state feedback offers another symmetry-adaptable route. The QGASS framework trains feedback laws of the form

$\rho_k$8

on the conditioned state produced by a stochastic master equation, and its main demonstration targets the symmetric Bell state

$\rho_k$9

The method does not impose symmetry by construction, but the same work notes that symmetry could be incorporated through invariant state representations, tied parameters, equivariant policy architectures, and symmetric costs (Evans et al., 2021).

Precision and speed limits under feedback are now available in channel-resolved form. For jump feedback, the thermodynamic uncertainty relation becomes

$u_k$0

and for homodyne feedback

$u_k$1

with feedback-dependent quantum dynamical activity $u_k$2 determined from two-sided generators and continuous matrix product states (Hasegawa, 2023, Yunoki et al., 13 Feb 2025). Because the feedback acts channel by channel through weights $u_k$3 and a Hermitian operator $u_k$4, these formulas directly accommodate balanced or symmetry-related gain assignments, even though the cited works do not impose such constraints explicitly (Hasegawa, 2023, Yunoki et al., 13 Feb 2025).

6. Experimental realizations, deliberate asymmetry, and open problems

Recent large-scale experiments provide a sharp contrast case: feedback can also be designed to break symmetry deliberately. In a one-dimensional chain of superconducting qubits, a monitored random circuit with branch Kraus operator

$u_k$5

and real-time conditional operations was implemented on IBM hardware, with large low-noise chains selected up to $u_k$6 qubits (Shen et al., 13 Apr 2026). The key observation is that mid-circuit measurements without feedback do not generate intrinsic directionality in the local density observable, even with a spatially varying measurement profile: $u_k$7 Asymmetry appears only when outcomes are converted into control actions, either through position-dependent feedback strength $u_k$8 or through directional operators such as a conditional SWAP acting from $u_k$9 to $\mathsf M_{u_k}(y)$0 (Shen et al., 13 Apr 2026). The same channel/Kraus formalism therefore provides a template for symmetry-preserving design by choosing inversion-symmetric $\mathsf M_{u_k}(y)$1, mirrored left/right actions, or alternating rules that cancel net drift, although such protocols were not implemented in that work (Shen et al., 13 Apr 2026).

More broadly, the review literature already treats symmetric and asymmetric feedback as distinct control resources. In two-atom entanglement control, symmetric feedback and local asymmetric feedback were explicitly contrasted, and coherent field-mediated feedback was noted to rely in some cases on elements that break time-reversal symmetry (Zhang et al., 2014). The current symmetry-centered channel theory leaves several questions open. The non-adaptive assumption in the AZ$\mathsf M_{u_k}(y)$2 classification remains essential, and the adaptive case was explicitly identified as open (Wen et al., 16 Sep 2025). Large-scale experiments have demonstrated feedback-induced asymmetry, but not a symmetry-preserving counterpart on the same hardware class (Shen et al., 13 Apr 2026). Symmetry-equivariant learning architectures and symmetry-adapted SOS reductions are technically plausible extensions of existing methods, but they are not part of the formal developments in the present performance-learning literature (Evans et al., 2021, Holtorf et al., 2023).

Symmetric quantum feedback control is therefore best understood as a family of constrained feedback-design problems rather than a single protocol class. Its mature components are already visible: symmetry-adapted continuous-measurement laws for Dicke, W, and GHZ targets; phase-space-covariant quadrature control; and a channel-theoretic symmetry/topology classification for successive discrete feedback maps. What remains under active development is the synthesis of these strands into scalable many-body controllers that preserve symmetry at the level of state, policy, and CPTP dynamics simultaneously (Zhang et al., 2018, Rouillard et al., 2022, Wen et al., 16 Sep 2025).