Spin Kerr-cat Encoding

• Spin Kerr-cat encoding is a method that uses Kerr nonlinearity and two-photon drives to create a protected two-branch cat manifold behaving as an effective spin-½ system.
• It features strongly biased noise channels with exponentially suppressed interwell transitions, enhancing coherence in both bosonic and literal spin implementations.
• The approach enables advanced gate operations and error correction through Hamiltonian confinement, dissipative stabilization, and innovative control and readout techniques.

Spin Kerr-cat encoding denotes two closely related constructions. In most superconducting implementations, it refers to a bosonic Kerr-cat qubit: a nonlinear oscillator driven by a two-photon pump, whose stabilized cat manifold behaves as an effective spin-$\tfrac12$. In a stricter usage, it refers to a literal spin realization in which the logical basis is formed by cat-like superpositions of spin-coherent states or by the lowest parity-resolved levels of a $\mathbb Z_2$-symmetric spin Hamiltonian. In both cases, the essential structure is a nonlinear Hamiltonian with two semiclassical branches, a parity decomposition, and a protected two-dimensional logical manifold with strongly anisotropic error channels (Hajr et al., 2024, McIntyre et al., 21 Apr 2026).

1. Terminology, scope, and logical manifold

The standard bosonic Kerr-cat Hamiltonian is

$$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$

with cat size set by $\alpha=\sqrt{\epsilon_2/K}$ for real drive amplitude. Its computational manifold is spanned by even and odd cat states

$$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$

or, equivalently, by the approximately localized coherent states $|\pm \alpha\rangle$ obtained as superpositions of those parity eigenstates (Ding et al., 2024).

The pseudo-spin mapping is explicit after projection into the cat manifold. In the high-coherence 2D Kerr-cat architecture, the projected annihilation operator is

$$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$

so the physical bosonic operator acts predominantly as a logical Pauli operator, with exponentially small admixture outside the dominant axis. This is the cleanest sense in which a bosonic Kerr-cat qubit behaves as an effective spin-$\tfrac12$ (Hajr et al., 2024).

Axis conventions are not universal. One widely used convention takes the coherent-state pair $|\pm\alpha\rangle$ as the logical $Z$ basis and even/odd cats as the logical $\mathbb Z_2$0 basis, while another takes the even/odd cats as $\mathbb Z_2$1 and the coherent states as $\mathbb Z_2$2. The underlying protected manifold is the same; only the Pauli labeling changes (Grimm et al., 2019). This convention dependence is the main source of ambiguity in the phrase “spin Kerr-cat encoding.”

2. Protected dynamics, biased noise, and logical failure modes

The practical utility of Kerr-cat encoding is its strongly biased noise. Because

$\mathbb Z_2$3

single-photon loss mainly reveals which coherent-state branch is occupied rather than mixing the two branches. In the coherent-state basis, this means that interwell transitions are exponentially suppressed as the overlap $\mathbb Z_2$4 becomes small. In the stabilized 2D device, the phase-flip lifetime of cat states follows the large-cat scaling

$\mathbb Z_2$5

while the coherent-state lifetime increases strongly with cat size and exceeds $\mathbb Z_2$6 ms by $\mathbb Z_2$7 (Hajr et al., 2024).

The ideal exponential suppression is not the whole story. In experiment, the bosonic manifold is limited by heating into excited states, multiphoton effects, and dephasing noise. A later coherence study gave direct evidence that the dominant limit on the bit-flip time is leakage out of the qubit manifold: the leakage population was measured to be $\mathbb Z_2$8, twelve times higher than in the undriven system. Frequency-selective engineered dissipation then cooled this leakage back into the computational manifold and increased the bit-flip time up to $\mathbb Z_2$9 ms without degrading equatorial coherence (Adinolfi et al., 2 Nov 2025).

Detuning adds a further layer of structure. In the “critical cat” framework, the slow odd-sector Liouvillian rate

$$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$0

acts as the logical bit-flip rate, while

$$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$1

quantifies leakage out of a metastable logical manifold. Large positive detuning and small but nonzero two-photon loss can suppress $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$2 beyond all resonant operating points, but the resulting code may be metastable rather than steady-state, so logical protection must be assessed jointly through bit flips and leakage (Gravina et al., 2022).

This suggests that “protection” in spin Kerr-cat encoding is not a single mechanism. In the bosonic literature it can arise from Hamiltonian confinement, dissipative stabilization, or a hybrid Liouvillian structure, and the dominant failure mode can shift from direct logical flips to leakage-mediated corruption depending on drive, detuning, and reservoir engineering.

3. Control, readout, and hardware realizations

The original experimental Kerr-cat demonstration used a superconducting nonlinear resonator in a 3D cavity, stabilized by the interplay of Kerr nonlinearity and a two-photon drive. It achieved a protected-axis coherence time of about $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$3 ms at $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$4, with single-qubit control via a resonant $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$5 drive and a $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$6 gate obtained by briefly turning off the stabilization drive. The same work demonstrated single-shot cat-quadrature readout under stabilization, with readout QNDness $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$7 (Grimm et al., 2019).

A major architectural advance was the 2D superconducting implementation using a SNAILmon oscillator and an on-chip wide band-block filter. The filter provided $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$8 dB isolation at the qubit frequency while leaving stabilization and readout frequencies essentially unaffected, removing the usual trade-off between strong drive-port coupling and Purcell decay. In that device, cat quadrature readout reached $$\hat H_{\mathrm{KCQ}}/\hbar = -K \hat a^{\dagger 2}\hat a^2 + \epsilon_2 \hat a^{\dagger 2} + \epsilon_2^* \hat a^2,$$9 QNDness at $\alpha=\sqrt{\epsilon_2/K}$0, and universal single-qubit control was obtained by combining fast $\alpha=\sqrt{\epsilon_2/K}$1 rotations with an $\alpha=\sqrt{\epsilon_2/K}$2 gate generated by phase modulation of the stabilization tone rather than abrupt pump shutoff (Hajr et al., 2024).

Kerr-cat encoding has also been developed as an ancilla architecture for bosonic memories. A parametrically coupled Kerr-cat qubit and high-$\alpha=\sqrt{\epsilon_2/K}$3 cavity realize a conditional-displacement interaction

$\alpha=\sqrt{\epsilon_2/K}$4

with rate $\alpha=\sqrt{\epsilon_2/K}$5. This gives a fast ancilla-controlled oscillator primitive for syndrome extraction and characteristic-function tomography. The same work showed that idling dephasing of the storage cavity is not set by the exponentially suppressed residual cross-Kerr alone, but by heating of the Kerr-cat into excited states; frequency-selective dissipation removes this dephasing and restores the cavity $\alpha=\sqrt{\epsilon_2/K}$6 to the $\alpha=\sqrt{\epsilon_2/K}$7 limit (Ding et al., 2024).

A complementary line of work uses DV–CV hybridization in Kerr parametric oscillators. Bell states prepared in the Fock basis can be adiabatically converted into Bell states in the even/odd cat basis, and a $\alpha=\sqrt{\epsilon_2/K}$8-type gate can be implemented directly between cat states. This places the pseudo-spin description of the Kerr-cat manifold on the same footing as ordinary two-level superconducting control, but with the encoded basis living inside a nonlinear oscillator (Hoshi et al., 2024).

4. Gate deformation, diagnostics, and generalized bosonic extensions

Gate performance in Kerr-cat encoding depends on the full time profile of the control, not just its amplitude. In a preparation-space study of gate-induced corruption, a Gaussian perturbation of the pump was analyzed over a disk of coherent-state initial conditions. A quench-like pulse centered at $\alpha=\sqrt{\epsilon_2/K}$9 produced a classical leak fraction of $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$0 and almost complete logical erasure, with trace distance $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$1 and idle-normalized retention $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$2. The same peak amplitude delivered as a smooth Gaussian centered at $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$3 reduced the classical leak fraction to $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$4 and raised the logical distinguishability to $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$5 and $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$6 (Wiggins, 29 Jun 2026).

That work also sharpened the relation between classical transport and quantum logical loss. A finite-time sensitivity field accurately locates the classical transport boundary in preparation space, while a Loschmidt echo evaluated near the end of the pulse predicts the much later quantum outcome with correlation $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$7 in the quench regime. By contrast, the OTOC is only transiently useful and does not provide a robust practical bridge in the studied parameter range (Wiggins, 29 Jun 2026). A plausible implication is that Kerr-cat gate design must treat separatrix transport and open-system scrambling as coupled but nonidentical phenomena.

Generalized Kerr-cat codes extend the ordinary two-component construction by replacing ordinary coherent states with Kerr-deformed coherent states. The resulting families interpolate continuously between $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$8 and $$| \mathcal{C}^\pm_\alpha \rangle = \mathcal{N}_\alpha^\pm (|+\alpha \rangle \pm |-\alpha \rangle),$$9 coherent states, depending on the sign of the Kerr nonlinearity. In the negative-$|\pm \alpha\rangle$0 branch, the displaced Kerr coherent states have finite support up to $|\pm \alpha\rangle$1 and become $|\pm \alpha\rangle$2-coherent states at $|\pm \alpha\rangle$3; in that branch, the overlap of opposite-amplitude states can vanish exactly at

$|\pm \alpha\rangle$4

which gives a compact, explicitly spin-like limit of the bosonic Kerr-cat framework. Recovery maps for these generalized cats are optimized by semidefinite programming, and the codes outperform ordinary two-component cats under combined loss and dephasing (Viladomat et al., 12 Jun 2026).

Higher-order nonlinearities lead to still richer phase-space organization. The chiral cat code adds a $|\pm \alpha\rangle$5 term to the Kerr-cat setting and creates a four-lobe structure with a low-amplitude code manifold and a high-amplitude error manifold. In that construction, bit flips can be trapped as same-sign transfers into the high-amplitude sector and then actively corrected, rather than appearing immediately as direct logical swaps between the low-amplitude code states (Labay-Mora et al., 14 Mar 2025).

5. Literal spin Kerr-cat qubits

A literal spin Kerr-cat qubit has now been proposed for a single quadrupolar nucleus with spin length $|\pm \alpha\rangle$6. The underlying Hamiltonian is

$|\pm \alpha\rangle$7

Here the $|\pm \alpha\rangle$8 term is the spin analogue of Kerr nonlinearity, while the $|\pm \alpha\rangle$9 term breaks the continuous $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$0 symmetry down to $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$1. The parity operator

$$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$2

commutes with the Hamiltonian, so the eigenstates have definite $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$3-parity (McIntyre et al., 21 Apr 2026).

The encoded basis is formed by the two lowest eigenstates at a clock-transition field $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$4, and those states are well approximated by parity-projected spin-coherent cats

$$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$5

The logical qubit is then defined as

$$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$6

with $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$7 chosen variationally to maximize fidelity with the two lowest exact eigenstates (McIntyre et al., 21 Apr 2026).

The protection mechanism differs from the bosonic case. Rather than relying primarily on exponentially weak interwell tunneling, the nuclear spin qubit is tuned to a clock transition: $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$8 where $$\hat{P}_{\rm C} \hat{a} \hat{P}_{\rm C}=\alpha \hat{Z}-i \alpha e^{-2 \alpha^2} \hat{Y},$$9 is the encoded splitting and $\tfrac12$0. This suppresses first-order dephasing from magnetic and quadrupolar fluctuations. Using measured parameters for $\tfrac12$1 donors in silicon, the work estimates that the encoded coherence time can reach

$\tfrac12$2

and that a two-qubit gate mediated by hopping electrons could achieve $\tfrac12$3 fidelity if quadrupolar splittings were enhanced by a factor of about $\tfrac12$4, neglecting shuttle and readout errors on the electron (McIntyre et al., 21 Apr 2026).

An earlier precursor to literal spin Kerr-cat physics appeared in two-component Bose–Einstein condensates. There, the effective nonlinear evolution is governed by a number-space expansion whose quadratic term produces an effective Kerr Hamiltonian $\tfrac12$5. Starting from a coherent spin state, evolution for

$\tfrac12$6

creates a superposition of two coherent spin states, and the feasibility analysis concluded that cat sizes of hundreds of atoms should be realistic. Optical readout was proposed through the Husimi $\tfrac12$7-function of light retrieved from the condensate (Lau et al., 2014). This was not a logical qubit architecture in the modern fault-tolerant sense, but it established the core dynamical mechanism of a Kerr-generated spin cat.

6. Related spin-cat encodings and taxonomy

Not every spin-cat code is a spin Kerr-cat code. The distinction is central to the modern taxonomy.

The spin-$\tfrac12$8-Cat code generalizes bosonic cat-code ideas to permutationally symmetric spin ensembles by encoding logical states as superpositions of equatorial spin-coherent states distributed around the Bloch-sphere equator. Its protection comes from modular decomposition of the Dicke ladder and syndrome-resolved correction, and its gates are implemented using only first-order central-spin interactions. The paper is explicit that this is not an explicit Kerr-stabilized spin-cat construction; the cat structure is generated kinematically and controlled dynamically rather than by a spin Kerr Hamiltonian (Franke et al., 16 Mar 2026).

A different finite-dimensional route is the large-spin cat code for fault-tolerant computation, which encodes a qubit into the extremal states $\tfrac12$9 of a spin-$|\pm\alpha\rangle$0 qudit and uses rank-preserving gates tailored to errors linear or quadratic in angular momentum. This architecture is directly analogous to continuous-variable cat fault tolerance at the level of biased-noise logic, but again it does not rely on Kerr stabilization of a double-well manifold (Omanakuttan et al., 2024).

Spin GKP codes clarify the relation between cat and Kerr ideas in collective-spin systems. In that framework, one-axis twisting is identified as mathematically analogous to the bosonic Kerr interaction, since both generate multipronged cat states at specific times, while two-axis countertwisting plays the role of squeezing. The resulting spin GKP codes outperform spin cat codes under isotropic ballistic dephasing, whereas ordinary spin cat codes are best when dephasing is concentrated along a single axis (Omanakuttan et al., 2022). This suggests that a genuine spin Kerr-cat encoding is best viewed not as the universal endpoint of spin-bosonic analogies, but as one point in a broader landscape that includes modular spin cats, twisted-spin grid codes, and large-spin biased-noise encodings.

In this taxonomy, “spin Kerr-cat encoding” has a narrow and a broad meaning. Narrowly, it denotes a literal spin system whose nonlinear Hamiltonian contains the finite-dimensional analogues of Kerr and two-photon terms, as in quadrupolar nuclei (McIntyre et al., 21 Apr 2026). Broadly, it denotes any encoding in which a parity-resolved cat manifold behaves as an effective spin-$|\pm\alpha\rangle$1, including bosonic Kerr-cat qubits whose protected manifold is routinely described in pseudo-spin language (Hajr et al., 2024). The common invariant across these usages is the emergence of a two-branch cat manifold with parity structure and anisotropic logical noise; the main differences lie in Hilbert-space dimension, stabilization mechanism, and the way leakage competes with logical protection.