Cat Breeding for GKP States

• Cat breeding for GKP states is a method that converts non-Gaussian optical primitives into large-amplitude, highly squeezed cat states via iterative interference, measurement, and conditioning.
• This technique enables the scalable preparation of approximate GKP grid states, which are essential for implementing fault-tolerant continuous-variable quantum computation.
• Experimental protocols range from Fock catalysis to cluster-state approaches, achieving cat amplitudes between 1.5 and 3.5 with squeezing up to 13 dB while improving error correction metrics.

Cat breeding is a methodology for engineering squeezed Schrödinger-cat states—superpositions of oppositely displaced, highly squeezed coherent states—as scalable resources for the preparation of Gottesman-Kitaev-Preskill (GKP) code states in continuous-variable (CV) quantum information processing. The "breeding" process employs iterative interference, measurement, and conditioning of non-Gaussian optical or bosonic states to systematically increase the size and macroscopicity of the superpositions, ultimately enabling the construction of approximate GKP grid states. Cat breeding underpins experimental and theoretical protocols across optical, superconducting, and general bosonic platforms for fault-tolerant quantum computation.

1. Foundations of Cat State Breeding

At its core, cat-state breeding transforms basic non-Gaussian primitives—typically single-photon, two-photon, or small-amplitude cat/Fock resources—into large-amplitude, highly squeezed coherent-state superpositions via controlled linear-optical mixing and heralded conditioning. The canonical target output state is the squeezed odd cat

$$\left|\mathrm{SCSS}(\alpha, z)\right\rangle = \hat{S}(z)\, \frac{|\alpha\rangle - |-\alpha\rangle}{\sqrt{2(1 - e^{-2|\alpha|^2})}}$$

where $\hat S(z)$ is the squeezing operator and $\alpha$ the coherent amplitude (Caron et al., 14 Jan 2026).

Iterative breeding is achieved by interfering prepared cats on symmetric (50:50) beam splitters, followed by strategic homodyne or photon-number-resolved measurements on ancillary output ports; surviving modes "inherit" a superposed and phase-conditioned version with increased effective amplitude $\alpha \rightarrow \sqrt{2}\,\alpha$ per round. This hierarchical construction is scalable through concatenation, temporal multiplexing, and storage in quantum memory cavities (Simon et al., 2024, Winnel et al., 2023).

2. Cat Breeding Protocols and Experimental Platforms

Protocols vary in the resource states, measurement modality, and experimental control:

• Fock/Photon catalysis breeding: Small Fock states (e.g., $|1\rangle$, $|2\rangle$) are mixed via a tunable beam splitter, heralded by homodyne detection, yielding large-amplitude cats with controlled squeezing; experimentally achieved amplitudes $\alpha\simeq2.47$ and squeezing $4.8\,\mathrm{dB}$ have been realized (Caron et al., 14 Jan 2026). Photon catalysis using coherent-state and ancillary single-photon interference followed by PNR detection enables cascaded amplitude growth and generalized superposition control (Eaton et al., 2019).
• Quantum memory–assisted iterative breeding: Successive cat-growing steps are enabled by storing intermediate states in low-loss cavities, enabling asynchronous multiplexing and significant rate improvements over fully parallel schemes; this allows kHz cat generation rates and linear time scaling with the number of breeding rounds (Simon et al., 2024).
• Cluster-state and node-teleportation based breeding: Cat states are "injected" and bred entirely within continuous-variable cluster states using Photon-counting Assisted Node-Teleportation (PhANTM) gadgets—local non-Gaussian gates via photon subtraction and homodyne detection (Eaton et al., 2021, Renault et al., 18 Nov 2025). This approach enables deterministic integration of cat breeding into measurement-based quantum computation.
• Deterministic all-Gaussian+PNRD schemes: All-Gaussian operations supplemented by photon-number resolving detectors (PNRDs) can deterministically prepare squeezed cats, which can then be bred into GKP states with arbitrarily high fidelity provided sufficient squeezing, using pure linear optics, feed-forward, and adaptive amplitude-rescaling (Winnel et al., 2023, Renault et al., 18 Nov 2025).

Protocol Variant	Primary Non-Gaussian Resource	Measurement Type
Fock mixing/homodyne (Caron et al.)	Heralded $|1\rangle$, $|2\rangle$	Homodyne
Photon catalysis (Eaton et al.)	Single photon + coherent	PNR
Quantum memory iterative (Simon et al.)	Single photons	Homodyne
CV cluster (PhANTM)	Cluster state + photon subtraction	Homodyne + PNR
All-Gaussian+PNRD (Winnel et al.)	Fock state	PNRD

3. Cat Breeding as a Pathway to GKP States

Once large-amplitude, squeezed Schrödinger-cat states are prepared, these serve as building blocks for approximate GKP codewords:

• GKP states are multidimensional grid states stabilized by displacement operators $\hat S(z)$0, $\hat S(z)$1, with ideal position-basis wavefunctions as combs of Dirac deltas spaced $\hat S(z)$2 apart.
• Cat-breeding for GKP: Pairs (or larger subsets) of cats are interfered (e.g., via beam splitter), homodyned, and conditioned to yield superpositions with multiple, evenly spaced coherent components (“grids”), matching the GKP lattice structure in phase space (Banic et al., 14 Apr 2025, Pizzimenti et al., 2024, Eaton et al., 2021, Winnel et al., 2023).

Hierarchical breeding (multiple rounds) sequentially doubles the number of peaks and sharpens the peaks’ widths—each step producing a closer approximation to a GKP codeword. Explicit Wigner-function analysis and sum-of-Gaussians techniques provide quantitative phase-space characterization (Banic et al., 14 Apr 2025, Solodovnikova et al., 8 Aug 2025).

4. Performance, Resource Scaling, and Limitations

Fidelity and squeezing: Experimental and simulated breeding protocols yield cat amplitudes $\hat S(z)$3, squeezing up to $\hat S(z)$4–$\hat S(z)$5 dB, and state fidelities with ideal cats or GKP qunaughts exceeding $\hat S(z)$6 for current-generation free-space optics, and $\hat S(z)$7 in deterministic, loss-free simulations (Caron et al., 14 Jan 2026, Winnel et al., 2023). Fault-tolerance for CV codes typically requires GKP effective squeezing $\hat S(z)$8 10 dB.

Success probability: The probability per breeding round depends on photon source/detector efficiency, loss, and homodyne window; temporal (asynchronous) multiplexing substantially increases rates, and high-efficiency SNSPDs, cavity engineering, and feed-forward are critical for scaling (Simon et al., 2024, Renault et al., 18 Nov 2025).

Resource scaling: Each round multiplies photon number and cat amplitude, such that for $\hat S(z)$9 levels (and final GKP with $\alpha$0 peaks), initial cat amplitude must be scaled as $\alpha$1 (Weigand et al., 2017, Banic et al., 14 Apr 2025). Resource-efficient protocols exploit dynamic resetting, amplitude-adaptive squeezing, and classical post-processing to avoid exponential resource blowup.

Loss tolerance: The threshold for optical loss is stringent: exceeding $\alpha$2 total loss in the cat-breeding chain prohibits reaching the fault-tolerance effective squeezing threshold (9.75 dB), even with high-quality seed cats (Solodovnikova et al., 8 Aug 2025). This places strong requirements on quantum memory, detection, and mode-matching fidelity.

5. Theoretical and Algorithmic Developments

• Sum-of-Gaussians simulation: The phase-space representation of cats and bred grid states as explicit sums of a small number of Gaussians enables efficient and exact simulation of high-dimensional protocols, tracking stabilizer expectation values, state fidelities, and effective squeezing under arbitrary optical circuits and measurement outcomes (Banic et al., 14 Apr 2025, Solodovnikova et al., 8 Aug 2025).
• Analytical bounds and post-selection-free breeding: Classical post-processing of homodyne data—i.e., adaptive phase estimation—enables deterministic, post-selection-free breeding, with every experimental run yielding a grid state up to a corrective displacement. Statistical bounds confirm quadratic squeezing improvement per round in the absence of loss (Weigand et al., 2017).
• Error channels: Photon loss, finite detection efficiency, dephasing, and phase drift are the primary error sources. The resilience to each is protocol-dependent, with typical tolerances of $\alpha$3 loss and fidelity degradation scaling linearly (first few rounds) or exponentially (long chains) (Solodovnikova et al., 8 Aug 2025, Caron et al., 14 Jan 2026, Kiryu et al., 20 Mar 2026).

6. Applications and Outlook for Fault-Tolerant CV Quantum Computing

Cat breeding for GKP state synthesis is a linchpin for scalable, fault-tolerant quantum error correction in CV photonics, superconducting circuits, and general bosonic settings. Current demonstrations have established $\alpha$4–$\alpha$5 cats at kHz rates on free-space platforms (Caron et al., 14 Jan 2026, Simon et al., 2024), and deterministic all-Gaussian+PNRD or dynamic cluster-state architectures have closed the gap to deterministic, threshold-level GKP state production (Winnel et al., 2023, Renault et al., 18 Nov 2025).

Hybrid entanglement breeding supports encoded qubit/qudit integration, error-corrected memory modules, and ancilla-based universal gate sets (Kiryu et al., 20 Mar 2026). Further improvements—higher squeezing, loss-insensitive topologies, integrated feed-forward, and improved detector technology—are anticipated to push cat breeding to commercial-scale GKP ancilla factories, enabling robust CV measurement-based quantum computing (Caron et al., 14 Jan 2026, Winnel et al., 2023, Renault et al., 18 Nov 2025, Kiryu et al., 20 Mar 2026).

7. Summary Table: Protocols and Achieved Metrics

Reference	Cat Amplitude $\alpha$6	Squeezing (dB)	GKP Fidelity	Effective Squeezing	Success Rate	Notable Innovations
(Caron et al., 14 Jan 2026)	2.47	4.8	0.53	—	Hz-kHz	Tunable BS, Fock mixing, QMC storage
(Simon et al., 2024)	1.63	3.6	>0.60	—	$\alpha$7/s	Event-ready cavity, temporal breeding
(Winnel et al., 2023)	2–3	9–13	>0.99 (theory)	>10 dB	Det	All-Gauss+PNRD, deterministic
(Solodovnikova et al., 8 Aug 2025)	—	up to 18 (seed)	—	9.75 dB (threshold)	—	Loss-threshold analysis, code toolkit

Here "Det" denotes deterministic, lossless theoretical protocols. For full device performance, losses, actual PNRD efficiency, and multiphoton source rates must be included, as in (Simon et al., 2024, Caron et al., 14 Jan 2026, Solodovnikova et al., 8 Aug 2025).

---

The field continues to evolve rapidly with a focus on pushing the achievable squeezing, macroscopicity, and deterministic operation of cat breeding modules, as well as integrating bred cats into scalable error-corrected CV architectures for quantum computing (Caron et al., 14 Jan 2026, Simon et al., 2024, Renault et al., 18 Nov 2025).