Approximate GKP States in Quantum Error Correction

• Approximate GKP states are finite-energy, Gaussian-enveloped combs that encode logical qubits for continuous-variable quantum error correction.
• Preparation techniques include superconducting circuits, photonic schemes, and neural network optimizations to achieve high fidelity and fault tolerance.
• Robust error management involves controlling Gaussian noise and finite squeezing effects to meet the thresholds required for scalable quantum computation.

Approximate Gottesman-Kitaev-Preskill (GKP) States are finite-energy, experimentally realizable analogues of the ideal GKP codewords, crucial for continuous-variable (CV) quantum error correction and fault-tolerant quantum computation. While ideal GKP codewords form periodic Dirac combs in position (or momentum) with infinite sharpness and energy, all physically accessible versions are deeply modified by finite squeezing, envelope truncation, and discretization, resulting in non-orthogonal, normalizable codewords. Their mathematical forms, noise handling, resource requirements, preparation methods, complexity, and practical performance thresholds are central to both experiment and theory.

1. Mathematical Structure and Standard Forms

Approximate GKP states encode a logical qubit into a single bosonic mode via translational invariance under the operators $S_q = \exp(i2\sqrt{\pi}\hat{q})$ and $S_p = \exp(-i2\sqrt{\pi}\hat{p})$. The ideal basis states are

$$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$

but are replaced by approximate combs for physical implementation.

The universally adopted standard form is a Gaussian-enveloped, Gaussian-spiked comb,

$$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$

with $\Delta$ the squeezing parameter ($\Delta \ll 1$ is strong squeezing), and $N(\Delta)$ normalization. The width of each spike is $\Delta$, and the envelope is $1/\Delta$. The momentum-basis has dual structure, and the Wigner function is a doubly periodic sum of phase-space Gaussians enveloped by a two-dimensional Gaussian of width $\Delta$ and $S_p = \exp(-i2\sqrt{\pi}\hat{p})$0 (Mensen et al., 2020, Matsuura et al., 2019).

All major approximations—envelope-damped, thermal, Gaussian random displacement—are proven equivalent up to squeezing transformations, and there exists a unique "standard" parametrization mapping all forms onto three real parameters $S_p = \exp(-i2\sqrt{\pi}\hat{p})$1 that can be tuned for physical constraints (Matsuura et al., 2019). The average photon number diverges as $S_p = \exp(-i2\sqrt{\pi}\hat{p})$2 at strong squeezing, setting an energetic resource floor.

2. Physical Preparation Protocols

Preparation of high-fidelity approximate GKP states remains an experimental bottleneck. Multiple architectures and recipes have been advanced:

• Periodic Driving in Superconducting Circuits: Realizes GKP states as Floquet eigenstates of a time-periodic driven nonlinear oscillator. Using a symmetric SQUID with a superinductor and capacitance $S_p = \exp(-i2\sqrt{\pi}\hat{p})$3, and modulating with multi-harmonic flux, a near-ideal cosine Hamiltonian emerges in the Floquet-Magnus expansion. The adiabatic frequency ramp prepares approximate GKP magic states with squeezing $S_p = \exp(-i2\sqrt{\pi}\hat{p})$4 dB at $S_p = \exp(-i2\sqrt{\pi}\hat{p})$5, $S_p = \exp(-i2\sqrt{\pi}\hat{p})$6 dB at $S_p = \exp(-i2\sqrt{\pi}\hat{p})$7 in $S_p = \exp(-i2\sqrt{\pi}\hat{p})$82 μs (Kolesnikow et al., 2023).
• Photonic Schemes with PNR and Homodyne Detection: A dominant strategy uses measurement-based generation of squeezed coherent-state superpositions (SCSS) via generalized photon subtraction, "breeding" two such states with a beamsplitter and homodyne to produce a single-mode GKP output. For $S_p = \exp(-i2\sqrt{\pi}\hat{p})$9 PNR-resolved photons, states reach ~10 dB effective squeezing at heralding rates $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$0 and $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$1 to the ideal form (Pizzimenti et al., 2024).
• Random Walk and Free Electron Mechanisms: Coin-controlled random walks in transverse optical position or strongly coupled free-electron–photon interactions in photonic structures can also deterministically generate GKP states. These approaches leverage binomial or multinomial random walks whose central-limit envelope becomes Gaussian, achieving $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$2 (fidelity $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$3) with moderate iteration depth (Sakuldee, 2024, Dahan et al., 2022).
• Neural Network Optimization: Recent advances utilize neural networks to optimize the complex weights in the superposition of squeezed coherent states, reducing the number of required components (from 21 to 7 at 9.55 dB) while simultaneously lowering logical error under photon loss and dephasing channels (Zeng et al., 2024).
• Universal Circuit Constructions: Heralded circuits, such as those combining repeated qubit-controlled displacements and Gaussian couplings, prepare approximate $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$4 states with constant probability and trace-norm error $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$5, and circuit size linear in $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$6 (Brenner et al., 2024).

3. Noise, Error Models, and Thresholds

Approximate GKP states are intrinsically subject to finite squeezing noise, leading to nonorthogonality and leakage outside the code subspace. The dominant error models are:

• Gaussian Random Displacement (GRD): Each physical GKP state is modeled as a GRD of the ideal code, leading to independent $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$7 and $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$8 errors with variance $$|0_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = 2n\sqrt{\pi}\rangle, \quad |1_L\rangle \propto \sum_{n \in \mathbb{Z}} |q = (2n+1)\sqrt{\pi}\rangle,$$9. Logical errors are suppressed as $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$0 (Noh et al., 2019, Wan et al., 2019, Mensen et al., 2020).
• Envelope Damping and Twirling: The envelope operator $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$1 is equivalent under stabilizer-twirled decoding to a GRD channel in the $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$2 regime. Pauli-twirled logic enables completely positive, trace-preserving descriptions at finite energy (Jafarzadeh et al., 18 Apr 2025).

Thresholds for fault-tolerant logical operation are set by the achievable squeezing. In surface-GKP concatenation, with only GKP-state noise, the error threshold is at 11.2 dB ($$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$3); with both GKP and circuit-level noise, the threshold rises to 18.6 dB squeezing (Noh et al., 2019). Advantageous noise-protection features persist for experimentally accessible $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$410 dB squeezing.

4. Figures of Merit: Fidelity, Error Rates, and Logical Properties

Core metrics for approximate GKP states include:

• Physical and Logical Fidelity: Physical fidelity to the ideal code decays as $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$5. Logical fidelity after projection or noise channels is subject to modular subsystem decomposition, with embedded error tracking via variance and trace distance (Tzitrin et al., 2019, Pantaleoni et al., 2021).
• Error Probability: For $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$6-stabilizer (homodyne) measurement, the logical error rate is $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$7 (Mensen et al., 2020).
• Photon Number and Energy Cost: The mean photon number diverges as $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$8 as squeezing increases, limiting the accessible regime in practice (Hosseinynejad et al., 5 Nov 2025).
• Subsystem/Modular Decomposition: Every GKP state admits a decomposition into logical and gauge degrees of freedom, allowing precise analysis of logical content, error-propagation, and gauge-mode coupling under gates and teleportation (Tzitrin et al., 2019, Pantaleoni et al., 2021).

5. Fault-Tolerant Preparation and Experimental Constraints

Fault-tolerant preparation protocols account for both active and passive errors during the GKP state construction:

• Phase Estimation with Flag-Quibit Acceptance Windows: Using multi-round phase-estimation circuits with flag ancillae and acceptance windows, it is possible to prepare approximate GKP states with certified $$\psi_{0,\Delta}(q) = N(\Delta) \sum_{n \in \mathbb{Z}} e^{-2\pi \Delta^2 n^2} \exp\left[-\frac{(q-2n\sqrt{\pi})^2}{2\Delta^2}\right],$$9-fault tolerance. Output states are selected to guarantee that no single measurement or damping error pushes the encoded shift beyond the GKP correctability radius $\Delta$0 (Shi et al., 2019).
• Dynamical Decoupling and Floquet Engineering: Rapid sequences of controlled displacement kicks, interleaved with Gaussian unitaries, enable passive Hamiltonian engineering of GKP stabilizer subspaces. Trade-offs involve control speed, achievable gap, and energy costs under the desired squeezing (Conrad, 2020, Kolesnikow et al., 2023).
• Photonic and Electronic Resource Scaling: Heralding rates for photonic protocols, explicit gate count scaling, and component error budgets are analytically derived. For instance, neural-network-optimized GKP codes reach break-even fidelity ($\Delta$1 over conventional) after 50 QEC cycles with just 7 squeezed coherent components, and tolerances on circuit elements and losses are quantified in the context of surface-GKP and measurement-based luminous quantum computing (Zeng et al., 2024, Hosseinynejad et al., 5 Nov 2025, Brenner et al., 2024).

6. Applications: Computation, Channel Coding, and Graph States

Approximate GKP states have become the linchpin of multiple quantum computing and communication architectures:

• Universal Computation: Clifford gates, magic state injection, and even probabilistic T-gate teleportation are realizable using only Gaussian operations and adaptive measurements on realistic GKP ancillae, with fidelity and success rates determined by the available squeezing (Hosseinynejad et al., 5 Nov 2025).
• Hybrid and Concatenated Codes: Surface-GKP and concatenated codes leverage finite-energy GKP's to achieve robust error correction at accessible resource cost, provided the squeezing threshold is met (Noh et al., 2019).
• Measurement-Based Quantum Computing: High-fidelity graph (cluster) states can be deterministically constructed from approximate GKP nodes, with systematic propagation of displacement and covariance errors characterized by Gaussian-ensemble techniques (Seshadreesan et al., 2021).
• Experimental Demonstrations: Time-frequency encoded GKP states with traveling photons, random-walk optics, and free-electron–photon interactions are all under active study or demonstration in state-of-the-art experiments, with theoretical treatments capturing the error scaling with photon number, envelope parameters, and post-selection probabilities (Descamps et al., 2023, Dahan et al., 2022).

7. Recent Advances and Outlook

Recent research has yielded rigorous characterizations of the complexity of preparing approximate GKP states: heralded circuit constructions with depth linear in $\Delta$2 are proven optimal (Brenner et al., 2024). Neural network approaches have enabled efficient squeezing-coherent expansions with minimal components (Zeng et al., 2024). There is now explicit analytical equivalence (parameter mapping) among all major classes of approximations, and thresholds for practical computation are now quantitatively established in both error-correction and computational models (Matsuura et al., 2019, Pizzimenti et al., 2024). Hybrid hardware realization approaches continue to converge toward the $\Delta$3 dB squeezing benchmark required for scalable fault-tolerant operation.

The field is now at the stage where high-quality approximate GKP states can be prepared, manipulated, and logically characterized with quantitative precision, forming a foundation for next-generation quantum information processing in both superconducting and photonic architectures.