Bosonic Quantum Error Correction

• Bosonic quantum error correction is a hardware-efficient paradigm that encodes logical quantum information into the infinite-dimensional Hilbert spaces of bosonic systems to mitigate photon loss and dephasing.
• It employs distinct code families such as cat, binomial, and GKP codes, each using tailored syndrome extraction and recovery strategies for dominant error types.
• This approach reduces overhead compared to conventional multi-qubit codes, enabling scalable fault-tolerant quantum computing, robust long-distance communication, and advanced quantum sensing.

Bosonic quantum error correction (QEC) encodes logical quantum information directly into the infinite-dimensional Hilbert space of bosonic systems, such as electromagnetic cavity modes, phonons of trapped atoms, or optical fields. By using quantum states of a single or a small number of oscillators, these codes exploit the rich structure and redundancy inherent in the bosonic degree of freedom to efficiently detect and correct dominant physical errors, most notably photon loss and dephasing. This hardware-efficient paradigm fundamentally differs from standard multi-qubit codes, offering both lower overhead and unique noise-tailoring capabilities.

1. Code Families and Principles of Bosonic QEC

Three canonical families of bosonic QEC codes have emerged: cat codes, binomial codes, and Gottesman–Kitaev–Preskill (GKP) codes, with extensions and hybrids—such as squeezed Fock state codes, squeezed-cat codes, number-phase lattice codes, and symmetry-operator-based codes—broadening the landscape.

Cat Codes

Cat codes encode logical qubits in superpositions of coherent states (e.g., $|\alpha\rangle \pm |-\alpha\rangle$) in a single bosonic mode. These codes are stabilized against single-photon loss by harnessing photon-number parity; photon loss events toggle parity between even and odd manifolds, which can be efficiently detected (Cai et al., 2020, Noh, 2021, Hastrup et al., 2021).

Binomial Codes

Binomial codes use finite superpositions of photon-number (Fock) states, patterned to correct specific numbers of photon loss, gain, and dephasing errors exactly up to a certain order in the quantum master equation's step size. The lowest-order binomial code corrects single-photon loss using basis states $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$, $|1_L\rangle = |2\rangle$ (Michael et al., 2016, Cai et al., 2020, Chelluri et al., 27 Mar 2025). These codes generalize naturally to higher-order loss/gain and to multi-mode encodings.

GKP Codes

The GKP code embeds a logical qubit in the grid structure of position–momentum space: logical codewords are superpositions of equally spaced position (or momentum) eigenstates, stabilized by discrete modular translations. The code corrects small shift errors in both quadratures, and is particularly robust for Gaussian noise and bosonic loss (Brady et al., 2023, Noh et al., 2019). Experimental implementations have reached beyond break-even protection (Lachance-Quirion et al., 2023).

Extended and Hybrid Codes

Recent advances introduced codes based on squeezed Fock-state superpositions (Bashmakova et al., 30 May 2025, Zeng et al., 5 Oct 2025), general number-phase lattices (Hu et al., 17 Aug 2025), and codes built from symmetry operators (notably via $\chi^{(2)}$ interactions) (Niu et al., 2017). These provide new mechanisms for dephasing/loss correction or exploit specific hardware nonlinearities.

2. Error Models, Syndrome Extraction, and Recovery Maps

Bosonic channels are predominantly subject to amplitude damping (photon loss), dephasing, and small displacement errors.

• Loss: Generated by Lindblad dissipators $$\kappa \mathcal{D}[a]\rho = \kappa(a\rho a^\dagger - \frac{1}{2}\{a^\dagger a, \rho\})$$, mapping codewords between parity sectors or orthogonal subspaces.
• Dephasing: Modeled as $\gamma \mathcal{D}[n]\rho$; result in phase diffusion in Fock space.
• Displacements: Unitary errors $D(\alpha)$, with random shift distributions.

Codes correct errors by two key processes:

• Syndrome Measurement: Mapping error events to orthogonal subspaces via photon-number parity, modular number operators, phase measurements, or GKP lattice stabilizer extractions (Michael et al., 2016, Grimsmo et al., 2019, Hu et al., 17 Aug 2025). Examples include dispersive parity measurement (cat/binomial codes) and homodyne detection after GKP ancilla coupling (GKP codes) (Lachance-Quirion et al., 2023, Terhal et al., 2020).
• Recovery: Conditional unitary operations designed to map corrupted states back to the code space; autonomous error correction via engineered dissipation (continuous stabilization), feedforward based on syndrome, or teleportation-based protocols (Kudra et al., 2022, Hastrup et al., 2021, Putterman et al., 2024, Grimsmo et al., 2019).

Table: Representative Syndrome Extraction Mechanisms

Code Type	Syndrome Observable	Physical Mechanism
Cat/Binomial	Number parity/modulo	Dispersive ancilla, cavity QED
GKP	Modular $q$, $p$ shifts	GKP ancilla + SUM (QND) + homodyne
Squeezed Fock	Parity & phase	Qutrit ancilla, phase-rotation gates
Number-Phase	Phase measurement	Canonical phase, modular number
Symmetry-based	Mode permutations/parity	$\chi^{(2)}$ nonlinear optics

3. Performance Metrics, Code Optimization, and Experimental Benchmarks

Code performance is quantified by average gate fidelity, Knill–Laflamme cost, entanglement infidelity, and logical error rate per correction cycle. For continuous error channels:

• Break-even is achieved when the logical lifetime surpasses the best physical mode (Cai et al., 2020, Noh, 2021, Lachance-Quirion et al., 2023).
• Thresholds depend on hardware parameters, e.g., GKP squeezing $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$011–18 dB for concatenated surface–GKP codes to reach scalable quantum fault tolerance (Noh et al., 2019).
• Scaling: Logical error suppression scales as $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$1 for code distance $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$2, where $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$3 is the physical loss probability. For GKP codes, logical error rates decrease as $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$4 for displacement noise variance $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$5 (Brady et al., 2023).
• Cat and binomial codes exhibit minimum infidelity at optimal mean photon numbers; performance degrades with excess photon loss or hardware-induced Kerr.
• Squeezed Fock superposition codes achieve error scaling $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$6 in squeezing parameter $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$7 and maintain exact codeword orthogonality at all $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$8 (Zeng et al., 5 Oct 2025).
• Squeezed Fock codes outperform squeezed–cat codes of the same mean photon number for photon-loss errors, with comparable Petz-map fidelities (Bashmakova et al., 30 May 2025).

Experimental works have demonstrated hardware-efficient QEC using concatenated bosonic codes—cat codes stabilized via engineered dissipators combined with repetition codes, yielding logical error per cycle $|0_L\rangle = (|0\rangle + |4\rangle)/\sqrt{2}$91.7% at $|1_L\rangle = |2\rangle$0 (Putterman et al., 2024).

4. Universal Gate Sets, Logical Operations, and Syndrome Circuit Engineering

Universal control of bosonic logical qubits exploits the oscillator's phase space and bosonic nonlinearities:

• Single-qubit gates: Implemented via SNAP (selective number-dependent arbitrary phase) gates, phase-space displacements, and nonlinear Hamiltonians (engineered via circuit QED or optics) (Cai et al., 2020, Cai et al., 29 Jan 2026).
• Entangling gates: For GKP, bosonic controlled-phase gates via 3WM/4WM elements, Bloch–Messiah decompositions, beamsplitters, or Kerr nonlinearities (Terhal et al., 2020, Brady et al., 2023).
• Bias-preserving gates: In concatenated architectures with protected cat or binomial encodings, physical gates are tailored to commute with error syndromes to avoid error proliferation (Putterman et al., 2024).
• Teleportation-based correction: Logical state transfer and error correction via syndrome measurement and adaptive correction, e.g., in all-optical cat code error correction (Hastrup et al., 2021), and for universal number-phase lattice codes (Hu et al., 17 Aug 2025).

Recent strategies for error-detectable universal control use the full qutrit space of the superconducting ancilla, discarding error trajectories signaled by ancilla relaxation to achieve logical gate/process fidelities exceeding 99.6% and $|1_L\rangle = |2\rangle$1 QEC gain beyond break-even (Cai et al., 29 Jan 2026).

5. Fault Tolerance, Concatenation, and Large-Scale Architectures

Bosonic codes can be concatenated with higher-level qubit stabilizer codes or used as modules in topological QEC architectures. Notable schemes include:

• Surface–GKP Code: GKP code as an inner layer, concatenated with a qubit or oscillator surface code, achieving fault tolerance for physical error rates approaching 1% and GKP squeezing $|1_L\rangle = |2\rangle$211–18 dB (Noh et al., 2019).
• Cat/Repetition Codes: Cat-encoded physical qubits form the base layer; concatenation with an outer repetition or surface code corrects remaining asymmetric errors. Below-threshold scaling is observed experimentally in superconducting circuits (Putterman et al., 2024).
• Bacon–Shor and LDPC Concatenation: Number-phase codes or binomial codes concatenated with repetition (X or Z basis) or subsystem LDPC codes, confining error propagation and allowing for software-based, syndrome-tracked recovery (Grimsmo et al., 2019).
• Resource-efficient scaling: Hardware-efficient bosonic QEC reduces the number of physical qubits or photonic modes required to reach a logical error target compared to conventional qubit-based surface codes (Brady et al., 2023, Putterman et al., 2024).

6. Hardware, Implementation Modalities, and Outlook

Bosonic QEC codes are demonstrated and under active development across multiple hardware platforms:

• Superconducting microwave cavities (cQED): High-$|1_L\rangle = |2\rangle$3 3D resonators dispersively coupled to transmon ancillas, enabling precise control, syndrome extraction, and recovery unitaries via GRAPE-optimized pulses, SNAP/PA gates, or engineered dissipation (Lachance-Quirion et al., 2023, Putterman et al., 2024, Cai et al., 29 Jan 2026).
• Optical modes: One-way QEC in cat, binomial, and number-phase codes via canonical phase measurement, beamsplitters, high-efficiency PNRDs, and atom-cavity entanglement (Hastrup et al., 2021, Hu et al., 17 Aug 2025).
• Trapped ions and neutral atoms: Encoding GKP codes into collective motional states, with ancilla-controlled displacements via sideband transitions; proposals for scalable architectures in optical tweezer arrays and lattices are supported by realistic trap and control parameters (Bohnmann et al., 2024).
• Nonlinear optics ($|1_L\rangle = |2\rangle$4-based codes): Codes exploiting symmetry properties and hardware-native nonlinear evolution, providing efficient error detection and correction with reduced resource requirements (Niu et al., 2017).

Outlook challenges include pushing squeezing and coherence beyond the thresholds for concatenated fault tolerance, high-fidelity syndrome extraction with minimal ancillary-induced backaction, and extending code constructions to optimize for mixed error models and device-specific constraints.

7. Applications: Communication, Computation, and Beyond

Bosonic QEC architectures enable:

• Long-distance quantum communication: QEC-protected quantum repeaters using microwave or optical cavities and binomial, cat, or GKP codes increase secret-key rates and distance, outperforming unencoded memory approaches (Chelluri et al., 27 Mar 2025, Brady et al., 2023).
• Fault-tolerant quantum computing: Hardware-efficient logical encoding, universal gate sets, and scalable QEC with performance approaching or exceeding the break-even point (Cai et al., 2020, Putterman et al., 2024, Cai et al., 29 Jan 2026).
• Quantum sensing: GKP codes and number-phase lattice codes enable robust displacement and phase measurements with resilience to dominant bosonic noise, advancing the Heisenberg-limited sensing frontier (Brady et al., 2023, Hu et al., 17 Aug 2025).
• Resource state engineering: High-fidelity bosonic magic-state generation for continuous-variable computation universality (Brady et al., 2023, Zeng et al., 5 Oct 2025).

Future research directions include the design of multi-mode bosonic codes for collective noise, further optimization of syndrome extraction and feedback circuitry, and adaptation to new physical platforms with emerging bosonic quantum hardware capabilities.