Bosonic Grid States and QEC

• Bosonic grid states, also known as GKP or comb states, are non-Gaussian quantum states that encode discrete information in the continuous phase space of harmonic oscillators.
• They leverage structured displacement operators and modular decomposition to facilitate quantum error correction with reduced error rates in hardware-efficient platforms.
• Experimental realizations on superconducting circuits, trapped ions, and optical setups use squeezing and beam splitter protocols to achieve high-fidelity logical operations.

Bosonic grid states, also known as comb states or Gottesman–Kitaev–Preskill (GKP) states, are highly structured non-Gaussian quantum states of bosonic (harmonic oscillator) modes. These states enable the encoding of discrete-variable (qubit or qudit) information in the infinite-dimensional Hilbert space of continuous-variable (CV) systems. Grid states play a central role in bosonic quantum error correction (QEC), offering a hardware-efficient path to fault-tolerant quantum information processing, particularly in superconducting circuits, trapped ions, and optical platforms. Their implementation, manipulation, and error correction harness both the algebraic structure of phase-space lattices and the physical accessibility of oscillator modes.

1. Mathematical Structure and Encoding

A (canonical) single-mode GKP grid state is defined as a simultaneous $+1$ eigenstate of two commuting displacement operators: $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$ where $\hat q,\,\hat p$ are canonical quadrature operators with $[\hat q,\hat p]=i$. For the square-lattice code, $\xi^2=2\pi$ sets the grid periodicity. The ideal codewords in position (or $q$) space are infinite-energy combs: $$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$ corresponding to a Dirac comb of narrow Gaussians at $q=2s\alpha$. In momentum space, the structure is analogous, due to the duality between $q$ and $p$ (Weigand et al., 2017, Lemonde et al., 2024).

Practical realizations use finite-energy analogues with a Gaussian envelope: $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$0 where $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$1 denotes squeezing of the vacuum. The “grid-likeness” is quantified by effective squeezing parameters $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$2, representing the rms width of the comb peaks in $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$3 and $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$4 (Weigand et al., 2017). A perfect grid state exhibits $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$5, unattainable in practice, so finite $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$6 and $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$7 are chosen to optimize fidelity and energy constraints.

Logical information is embedded in displacements of the grid. The canonical GKP code encodes a qubit with logical operators $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$8 and $$S_p = \exp(i \xi \hat p), \qquad S_q = \exp(i \xi \hat q),$$9, implementing Pauli $\hat q,\,\hat p$0 and $\hat q,\,\hat p$1 as half-grid displacements (Lemonde et al., 2024).

2. Physical Generation Protocols

Squeezed Cat-State Breeding

Grid states are non-Gaussian and require non-Gaussian resources for preparation. Early strategies employed squeezed Schrödinger cat states passed through linear-optical circuits. In the post-selected “breeding” protocol, two squeezed cat states are mixed on a beam splitter, a homodyne measurement is performed on one output, and only measurement outcomes near $\hat q,\,\hat p$2 are kept. Iterating this process increases the comb density and sharpens the envelope but at an exponentially decreasing success rate (Weigand et al., 2017).

A critical advance is the removal of post-selection via classical post-processing. Each homodyne outcome is recorded, and the net phase shift of the emerging comb is calculated. A corrective displacement is applied at the end based on the measurement record, obviating the need to discard any data. Numerics confirm this “no post-selection” protocol achieves effective squeezing $\hat q,\,\hat p$3 close to the post-selected ideal, enabling deterministic preparation of high-quality grid states sufficient for QEC applications (Weigand et al., 2017).

Quantum Walk and Linear-Optical Realizations

Alternative schemes realize grid states via linear-optical quantum walks, where a cat state serves as a quantum coin and a squeezed vacuum as the walker. Conditional phase-space displacements are implemented via Mach-Zehnder interferometers (MZIs) with post-selection on the ancillary coin mode’s parity. The resulting state after $\hat q,\,\hat p$4 steps is an approximate GKP codeword, with fidelity controlled by resource parameters. Post-selection success per step scales as $\hat q,\,\hat p$5, but for small $\hat q,\,\hat p$6 and in superconducting circuit platforms with large cat amplitudes $\hat q,\,\hat p$7, overall codeword fidelities $\hat q,\,\hat p$8 are attainable (Wu et al., 2024).

Modular-Subsystem Decomposition

Grid states can also be interpreted as embedding a virtual qubit in the modular structure of the oscillator’s Hilbert space. Any $\hat q,\,\hat p$9 can be decomposed as $[\hat q,\hat p]=i$0, with $[\hat q,\hat p]=i$1 and $[\hat q,\hat p]=i$2. Parity of $[\hat q,\hat p]=i$3 identifies the logical qubit, while $[\hat q,\hat p]=i$4 and the residual integer form an auxiliary “gauge” mode. This decomposition underpins syndrome extraction and error correction, with measurements of $[\hat q,\hat p]=i$5 isolating logical noise from gauge-mode entanglement (Pantaleoni et al., 2019).

3. Multi-Mode and Lattice-Structured Grid Codes

The grid code paradigm generalizes to multiple oscillator modes, where the code space is the simultaneous $[\hat q,\hat p]=i$6 eigenspace of a set of commuting displacement operators associated with a symplectically integral lattice $[\hat q,\hat p]=i$7 in multimode phase space. The stabilizer group

$[\hat q,\hat p]=i$8

is defined by lattice generators, with logical operators drawn from the dual lattice $[\hat q,\hat p]=i$9. Multimode codes, such as the tesseract (hypercubic) code and the $\xi^2=2\pi$0 root-lattice code, offer increased error-correction regions (larger Voronoi cells), suppressed logical-fault propagation from dissipation in ancillary qubits, and richer syndrome information. The code dimension is set by $\xi^2=2\pi$1 (Royer et al., 2022, Lemonde et al., 2024).

Logical operators correspond to specific displacements in phase space, and Clifford operations correspond to symplectic transformations preserving the lattice. Non-Clifford operations can be realized via envelope-preserving nonlinear evolutions, often Kerr-type interactions, exploiting certain finite-order lattice symmetries (Royer et al., 2022).

A key result in the multimode setting is the exponential suppression ($\xi^2=2\pi$2) of logical error rates arising from ancilla-induced faults, in contrast to the constant ($\xi^2=2\pi$350\%) error rate for single-mode codes, given equal code sizes. This stems from the denser lattice packing and redundancy in the higher-dimensional phase space (Royer et al., 2022).

4. Error Models, Syndrome Extraction, and Correction

Bosonic grid codes are intrinsically tailored to correct small displacement errors in phase space, modeling decoherence from photon loss or generic Gaussian noise as random shifts $\xi^2=2\pi$4: $\xi^2=2\pi$5 where $\xi^2=2\pi$6 is typically a Gaussian (Lemonde et al., 2024, Royer et al., 2022). The correctable region is bounded by half the grid spacing: $\xi^2=2\pi$7 for the square code.

Syndrome extraction is performed by measuring $\xi^2=2\pi$8 and $\xi^2=2\pi$9 modulo the grid spacing (modular variables), often via echoed conditional-displacement gates between oscillator and ancillary qubit. Error diagnosis and correction employ either closest-integer decoding—shifting back into the fundamental cell—or maximum-likelihood decoding using historical syndrome data. The per-round logical error rate is

$q$0

Repeating $q$1 rounds yields logical fidelity $q$2 (Lemonde et al., 2024). In practice, extraction of both syndrome value and “confidence” information in real-time allows tailoring outer code strategies and leveraging the erasure channel structure induced by amplitude damping.

Gauge subsystems arising from the modular decomposition lead to possible logical–gauge entanglement; syndrome measurements project the gauge, and corrective displacements restore logical purity (Pantaleoni et al., 2019).

5. Logical Gates and Clifford/Non-Clifford Operations

Logical operations in bosonic grid codes are naturally implemented as phase-space displacements or symplectic transformations. Pauli and Clifford gates correspond to half-grid displacements and phase-space rotations, respectively:

• $q$3, $q$4.
• Hadamard gate by $q$5 rotation: $q$6 (Lemonde et al., 2024).

In the multimode context, Gaussian logical gates are realized by the action of lattice-symmetry-preserving symplectic transformations, while universal sets require supplementing with non-Gaussian, envelope-preserving nonlinear operations (e.g., Kerr or cross-Kerr interactions) (Royer et al., 2022). Gates such as SNAP (selective number-dependent arbitrary phase) enable fine control at the Fock-state level.

Operations may be performed virtually (by adjusting the frame and updating syndrome signs), directly (physical gates), or through measurement-based protocols.

6. Experimental Realization and Performance Benchmarks

Superconducting Circuits

In three-dimensional cavity QED (cQED) systems, high-coherence superconducting cavities support long-lived bosonic modes. A dispersively coupled nonlinear element (such as a transmon or fluxonium qubit) enables universal control. State-of-the-art demonstrations have achieved cavity lifetimes $q$7 ms and MHz-scale logical gate rates. Multimode architectures exploiting multi-post cavities permit encoding several grid-state qubits per physical resonator (Lemonde et al., 2024).

State stabilization is performed by “small–big–small” (sBs) echoed conditional-displacement sequences and autonomous feedback, robustly preparing grid states with $q$8. Mid-circuit syndrome measurement and ancilla reset support real-time QEC with confidence readout, directly integrating error tracking and correction with logical operations.

Linear-Optical and Trapped-Ion Platforms

Linear-optical schemes use squeezed vacuum and cat states, combined via Mach-Zehnder interferometers and post-selection, to approximate grid states. Optical squeezing levels of $q$9–$$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$0 dB and cat states with $$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$1 are experimentally feasible. Superconducting circuits achieve higher $$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$2 ($$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$3–$$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$4), enabling larger codeword fidelities and lower post-selection thresholds (Wu et al., 2024).

Trapped-ion systems use motional modes coupled by state-dependent forces for direct grid state preparation and manipulation; the GKP code has been deterministically demonstrated in ion traps (Royer et al., 2022).

Comparison and Scalability

GKP-based bosonic codes drastically reduce hardware overhead compared to surface codes, as QEC operates at the single physical unit level. Multimode grid codes further enhance robustness to dissipation and correlated errors, reduce error propagation from ancilla faults, and allow richer decoding strategies. Logical clock rates at the MHz scale are feasible, supporting both rapid computation and low-latency error correction (Lemonde et al., 2024).

7. Outlook and Theoretical Connections

Bosonic grid-state encoding provides a unifying interface between continuous-variable and qubit-based quantum information, enabling systematic construction and analysis of CV codes via modular decompositions and phase-space lattice theory. The underlying framework extends to higher-dimensional and qudit codes, supports various lattice geometries (including $$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$5, $$\ket{\Psi_{\rm grid}} \propto \sum_{s\in\mathbb Z} e^{-\tfrac{1}{2\Delta^2}(q-2s\alpha)^2}\ket{q}, \qquad \alpha = \frac{\xi}{2}, \quad \Delta\ll1,$$6), and enables concatenation with discrete-variable QEC codes. The modular-subsystem perspective clarifies logical-gauge structure, noise channels, and the purification role of syndrome extraction (Pantaleoni et al., 2019, Royer et al., 2022).

These developments establish bosonic grid codes as the core primitive for hardware-efficient fault-tolerant quantum computing in superconducting, optical, and trapped-ion architectures, with concrete pathways toward large-scale implementation and universal continuous-variable quantum computation (Weigand et al., 2017, Lemonde et al., 2024, Royer et al., 2022).