GKP-Encoded Photonic Qubits

Updated 28 January 2026

GKP-encoded photonic qubits are a representation of quantum information using a lattice grid of peaks in continuous-variable optical modes, enabling effective error correction against displacement errors.
They achieve robustness through deterministic Gaussian operations and have been experimentally realized using various state preparation protocols with specific trade-offs in squeezing and fidelity.
Integration into cluster-state architectures and repeater networks supports scalable fault-tolerant quantum computing and communication, as evidenced by quantitative benchmarks in recent studies.

Gottesman-Kitaev-Preskill (GKP) encoding is a paradigm for representing discrete-variable quantum information in the continuous-variable phase space of a bosonic mode—prominently, an optical field—via a grid of equally spaced peaks corresponding to eigenstates of lattice stabilizers. GKP-encoded photonic qubits are designed to enable robust error correction against small displacement errors while being compatible with deterministic Gaussian operations, making them a foundational primitive for scalable fault-tolerant quantum information processing in optics (Wayo, 22 Jan 2026, Takase et al., 2022, Pizzimenti et al., 2024). The successful realization of GKP states and their stabilizer-based logical operations in photonic systems is central to emerging quantum computing, communication, and metrology architectures.

1. GKP Code Structure and Physical Realizations

The GKP code encodes a logical qubit in a single oscillator via two mutually commuting stabilizers, typically chosen as

$\hat S_q = \exp(i\,2\sqrt{\pi}\,\hat q)\,, \qquad \hat S_p = \exp(-i\,2\sqrt{\pi}\,\hat p)\,,$

where $\hat q, \hat p$ obey the canonical commutation relation. Logical Pauli operators are

$\bar Z = \exp(i\sqrt{\pi}\hat q)\,, \qquad \bar X = \exp(-i\sqrt{\pi}\hat p)\,.$

The ideal logical $\lvert 0_L \rangle$ is a Dirac comb in $q$ : $|0_L\rangle \propto \sum_{n\in\mathbb{Z}} \delta(q - 2n\sqrt{\pi})\,,$ and $\lvert 1_L \rangle$ is shifted by $\sqrt{\pi}$ [$2601.16244$, $2212.05436$].

Physical GKP states require finite energy and are realized as combs of narrow Gaussians with envelope: $\psi_0^{(r)}(q) \propto \sum_{n} e^{-\frac{(q-2n\sqrt{\pi})^2}{2\sigma^2}}\,,$ where $\sigma^2 = \frac{1}{2}e^{-2r}$ and $r$ (in dB) is the squeezing parameter [$2601.16244$]. Approximation fidelity improves with increasing squeezing, but hardware limits presently cap optical implementations at $10$–$12$ dB.

Multiple experimental platforms have demonstrated methods for preparing approximate GKP states, including:

Interferometric protocols based on photon-subtracted squeezed states and homodyne postselection (Konno et al., 2023).
Random-walk mechanisms in spatial or motional degrees of freedom (Sakuldee, 2024).
Gaussian breeding and superposition breeding protocols employing iterated photon subtraction and feed-forward (Takase et al., 2022, Pizzimenti et al., 2024).
Use of atomic ensemble ancillae via Faraday QND interactions (Motes et al., 2017).
Free-electron interactions via phase-matched PINEM (Dahan et al., 2022).

2. State Preparation Protocols and Tradeoffs

GKP state engineering in photonic modes is ultimately constrained by the achievable squeezing and efficiency of ancillary resource generation and measurement. Dominant methods include:

Measurement-Based Breeding: Iterates photon subtraction and beam splitter mixing, postselecting on photon-number-resolved detection or high-efficiency homodyne, yielding grid states with heralding probability $P \sim 10^{-6}$ – $10^{-5}$ for $10$ dB squeezing and state fidelities $F > 0.98$ [$2212.05436$, $2409.06902$].
Random Walk Mechanisms: Implement quantum walks in position space, accumulating controlled displacements conditioned on a two-level coin, with fidelity tradeoff between walk steps $N$ and spike width $\Delta$ (Sakuldee, 2024).
Quantum-Walk Encodings via Linear Optics: Employs cat states as a coin in Mach-Zehnder interferometry to realize conditional displacements on a squeezed vacuum grid, with each step’s fidelity limited by operational parameters $(\alpha, \phi, \zeta)$ (cat amplitude, phase-shift, squeezing) and overall postselection rate $2^{-N}$ for $N$ steps (Wu et al., 2024).
Atomic or Free-Electron Ancilla Methods: Use collective atomic spin projections or free-electron combs to induce GKP-compatible conditional displacements on optical modes, accessing grid state fidelities above 0.9 for $>10$ dB squeezing at $10\%$ – $30\%$ success probability (Motes et al., 2017, Dahan et al., 2022, Baranes et al., 2023).
Hybrid and Time-Frequency Encodings: Encode grid states in time-frequency collective modes of multiphoton fields, leveraging enhanced noise tolerance and photon-loss resilience (Descamps et al., 2023).

Resource-scaling is fundamentally polynomial in the width of the comb (squeezing), with tradeoffs among squeezing $r$ , number of ancillary states, postselection rates, and attainable logical fidelity.

3. Fault-Tolerance: Error Models, Correction, and Code Concatenation

The dominant error channel for GKP encoded photonic qubits is Gaussian random displacement (induced by finite squeezing, photon loss, or amplifier noise), which maps directly to effective logical $Z$ -flip channels after phase-space modular correction: $D_Z(\rho) = (1-p_Z)\,\rho + p_Z Z\rho Z\,,$ where $p_Z(r) = \min[0.5,\,\alpha_s \exp(-\beta r_{[\mathrm{dB}]})]$ with $\alpha_s\approx 0.15, \beta\approx 0.20$ per LiDMaS [$2601.16244$]. For $r$ rising from $8$ to $16$ dB, $p_Z$ falls from $10^{-1}$ to $10^{-3}$ .

Fault-tolerant architectures concatenate the inner GKP code with outer stabilizer codes—surface code or QLDPC—using extracted analog syndrome information. The surface code logical error rate under residual error $p_\mathrm{phys}$ is [$2601.16244$]: $p_L \approx A (p_\mathrm{phys}/p_\mathrm{th})^{(d+1)/2}\,,$ with $p_\mathrm{th}\approx 1\%$ , $A\sim 1$ , and $d$ the surface code distance.

GKP analog syndrome outputs $r_q, r_p$ (modulo shift), available from homodyne measurements, enable soft-information iterative decoding and contribute to surpassing the CSS Hamming bound for finite-rate QLDPC architectures (Raveendran et al., 2021). Error floors are significantly suppressed by leveraging the continuous syndrome distributions. Practical error correction also addresses photon loss via either heralded erasure (aborting injection cycles) or conversion to effective random displacement channels upon amplification (Fukui et al., 2020).

4. Logical Gate Implementations and Magic-State Injection

Single- and two-qubit Clifford gates in GKP codes are executable with Gaussian (linear-optical) operations (beam splitters, squeezers, phase shifters). For universality, non-Clifford operations (T-gate) are crucial. In photonic GKP, the dominant T-gate mechanism is injection and distillation of magic states.

LiDMaS presents a comprehensive architecture-level model of repeat-until-success (RUS) magic-state injection [$2601.16244$], with outer-code protection:

Key metrics: $P_\mathrm{success}$ (injection success probability), $\langle R \rangle$ (average rounds per success), $F_\mathrm{logical}$ (post-correction fidelity).
Under $12$–$14$ dB squeezing, $P_\mathrm{success} > 0.95$ , $F_\mathrm{logical} \sim 0.80$ for distance-5–7 surface code, with average overhead $\langle R \rangle \sim 1.16$ .
Loss sensitivity is weak due to heralded erasure; the dominant limiting factor is the residual dephasing from finite squeezing.

Architectural targets for squeezing are established by phase-boundary analysis: to achieve both $P_\mathrm{success} \geq 0.95$ and $F_\mathrm{logical} \geq 0.79$ at $2\%$ loss, a minimum squeezing $r_\mathrm{min} \gtrsim 13$ dB with $d=5$ is needed (see Table 1 in [$2601.16244$]).

For T-gate implementation, nonlinear feedforward circuits utilizing GKP-encoded ancillae outperform cubic-phase–based gates in both fidelity (attaining $F > 0.90$ for $r = 10$ dB) and circuit resource count, using only a single GKP magic state ancilla and dynamic-squeezer (Konno et al., 2021).

5. Large-Scale Cluster-State and Communication Architectures

High-fidelity, scalable cluster and graph states of GKP-encoded photonic qubits can be generated using Gaussian operations and probabilistic fusion protocols (Seshadreesan et al., 2021). The propagation of Gaussian noise and absolute displacement errors through measurement-based fusion and error-correction steps admits an exact representation by affine symplectic maps on the error covariance and mean, allowing for analytical control and optimization over the collective error model.

Applications to quantum communication leverage all-photonic GKP repeaters, where the key features are:

Deterministic entangling gates, room-temperature operation, and analog syndrome-aided multiplexed entanglement ranking (Rozpędek et al., 2023, Fukui et al., 2020).
Concatenation with outer codes, loss conversion to effective GKP-correctable noise via phase-insensitive or phase-sensitive amplification.
Secret-key rates surpassing repeaterless bounds at $1000$ km for squeezing above $12$–$14$ dB, with total resource counts $10^7$ – $10^8$ GKP qubits per protocol run (Rozpędek et al., 2023).

6. Experimental Progress and Outlook

Propagating GKP states have been demonstrated experimentally at telecom wavelengths, compatible with low-loss fiber networks and 5G photonic hardware (Konno et al., 2023). Experimental metrics on such states include:

Quadrature variances $\Delta^2 x \sim 1.45$ , $\Delta^2 p \sim 0.07$ , product $\ll 1/4$ .
Stabilizer expectation values exceed all-classical and all-Gaussian bounds.
Effective squeezing remains at $\sim 2.5$ dB in current generation, with target thresholds for full error correction at $>10$ dB.
Dominant limitations arise from optical loss, heralding inefficiency, and squeezing production.

Integration of high-fidelity GKP state preparation protocols (e.g., Gaussian breeding, PINEM electron-based, quantum walks) with large-scale time- and frequency-multiplexed photonic networks is advancing rapidly. Key remaining challenges are pushing squeezing $r$ beyond $12$ dB, achieving high-rate deterministic resource state generation, and integrating reliable analog syndrome extraction and feed-forward protocols. Free-electron and atomic-ensemble schemes offer promising routes for on-chip and cavity-based GKP qubit generation with comb structures spanning the optical spectrum (Dahan et al., 2022, Baranes et al., 2023).

7. Tables: Performance Benchmarks in Photonic GKP Architectures

Selected Metrics from LiDMaS RUS Magic-State Injection ( $p_\mathrm{base} = 0.02$ ):

Squeezing $r$ [dB]	$P_\mathrm{success}$	$\langle R \rangle$	$F_\mathrm{logical}$ (surface-code $d$ )
10	0.94	1.20	0.77 ( $d=3$ )
12	0.96	1.17	0.79 ( $d=5$ )
14	0.97	1.16	0.80 ( $d=7$ )

Minimum Squeezing for Design Targets (from LiDMaS):

Surface code distance $d$	$a_d$ [dB]	$b_d$ [dB]
1	14.8	1.1
3	13.6	0.9
5	12.8	0.7
7	12.1	0.6

These data quantitatively inform architectural requirements on squeezing and loss for achieving target logical fidelities and error correction rates in scalable photonic fault-tolerant computation (Wayo, 22 Jan 2026).

GKP-encoded photonic qubits, with their lattice stabilizer structure, provide a pathway to high-threshold, resource-efficient, and hardware-adapted error correction, supporting robust quantum computation and communication. Their state preparation, error correction paradigms, and integration into cluster- and network-based architectures are under active investigation, with both theoretical benchmarks and experimental demonstrations rapidly advancing toward architectural-scale deployment (Wayo, 22 Jan 2026, Takase et al., 2022, Pizzimenti et al., 2024, Konno et al., 2023).