CV Photonic Quantum Computing

• Continuous-variable photonic quantum computing is a technique that uses infinite-dimensional bosonic modes to encode and process quantum information via quadrature amplitudes.
• It employs deterministic Gaussian gates alongside essential non-Gaussian operations to achieve universal quantum computation and drive advanced applications.
• Scalable implementations leverage integrated optical circuits and squeezed-light sources to enable room-temperature, high-speed quantum simulation and error-corrected processing.

Continuous-Variable Photonic Quantum Computing (CVPQC) refers to the use of photonic systems that encode, process, and read out quantum information in bosonic modes with infinite-dimensional Hilbert spaces. Unlike discrete-variable (DV) platforms based on qubits, CVPQC manipulates quantum states via the quadrature amplitudes of the electromagnetic field. This paradigm offers high gate flexibility, deterministic Gaussian state preparation, and the possibility of room-temperature, large-scale integration (Choe, 2022, Romero et al., 2024). Theoretical and experimental advances position CVPQC as a central architecture for scalable quantum computing, quantum simulation, and quantum-enhanced machine learning.

1. Continuous-Variable Quantum State Space and Theoretical Foundation

Each photonic mode (“qumode”) is associated with an infinite-dimensional Hilbert space $\mathcal{H}$ spanned by Fock states $|n\rangle$ ($n=0,1,2,\ldots$). Physical realizations may also employ quadrature eigenstates $|x\rangle$ or $|p\rangle$, associated with the canonical operators $\hat{x}$ (position) and $\hat{p}$ (momentum), satisfying $[\hat{x},\hat{p}] = i\hbar$ (Choe, 2022).

Annihilation and creation operators are defined: $$\hat{x} = \sqrt{\hbar/2}(a + a^\dagger),\quad \hat{p} = i\sqrt{\hbar/2}(a^\dagger - a),\quad [a,a^\dagger]=1.$$ Quantum states are represented in phase space, often via the Wigner quasi-probability function $W(x,p)$. Gaussian states (vacuum, squeezed, coherent) have Gaussian Wigner functions; non-Gaussian states (e.g. cubic-phase, photon-subtracted) have Wigner functions exhibiting negativity—a necessary condition for quantum computational advantage (Choe, 2022, Chabaud, 2021).

2. CV Gate Sets: Gaussian and Non-Gaussian Operations

Universal quantum computation in the CV paradigm hinges on two classes of gates:

• Gaussian gates: These transform Gaussian states to Gaussian states and form the symplectic affine group. Key Gaussian gates include:
  • Displacement: $|n\rangle$0
  • Phase rotation: $|n\rangle$1
  • Single-mode squeezing: $|n\rangle$2
  • Beamsplitter: $|n\rangle$3

These operations can be decomposed for multimode systems via the Bloch–Messiah reduction theorem: a sequence of passive linear optics (interferometers), single-mode squeezers, and further passive optics (Choe, 2022, Andersen et al., 2010).

• Non-Gaussian gates: Essential for universality due to the classical simulability of pure-Gaussian circuits (Chabaud, 2021, Romero et al., 2024). The canonical non-Gaussian gate is the cubic-phase gate:

$|n\rangle$4

Additional non-Gaussian elements include photon-number addition/subtraction and Kerr-type nonlinearities (Shrivastava et al., 2019).

Universality: Any combination of Gaussian gates with at least one non-Gaussian gate, such as $|n\rangle$5, suffices to approximate arbitrary unitaries on CV systems (Lloyd–Braunstein theorem) (Choe, 2022).

3. Photonic Implementations, Cluster States, and Integration

Squeezed-Light Sources and Optical Circuits

Squeezed states are typically generated via optical parametric oscillators (OPOs) exploiting $|n\rangle$6 or $|n\rangle$7 nonlinearities. For example, below-threshold OPOs on lithium niobate or silicon nitride chips generate vacuum squeezing with up to $|n\rangle$8 dB inferred on-chip (Clark et al., 5 Jun 2025, Lenzini et al., 2018). Time- and frequency-multiplexed cluster-state generation leverages Kerr microresonator frequency combs producing thousands of entangled modes (Wu et al., 2019).

Integrated photonic circuits realize beam splitter networks, Mach–Zehnder interferometers, and programmable phase shifters with low loss (e.g., <0.1 dB/m in SiN) and high clock rates (GHz–THz bandwidth) (Masada et al., 2015, Clark et al., 5 Jun 2025). Fast electro-optic modulators and on-chip detectors provide real-time control and readout.

Continuous-Variable Cluster States and MBQC

Measurement-based quantum computation (MBQC) exploits large entangled Gaussian cluster states. Cluster state nodes are prepared by interfering squeezed vacua on beamsplitters according to the desired graph topology, often exploiting time- or frequency-multiplexed architectures for scalable resources (García-Beni et al., 2024, Wu et al., 2019). The nullifier relations,

$|n\rangle$9

define the stabilization of ideal cluster states.

MBQC universality is attained by combining homodyne measurements for Gaussian gates and (at least one) non-Gaussian gate or measurement (e.g., cubic-phase ancilla or photon counting) (Andersen et al., 2010, García-Beni et al., 2024).

Table: Core Hardware Components and Functions

Component	Function	Reference
OPO (LiNbO₃, Si₃N₄, Si)	Squeezed state generation	(Clark et al., 5 Jun 2025, Lenzini et al., 2018)
Integrated interferometer (MZI, BS)	Unitary mode mixing	(Masada et al., 2015)
Delay line / multiplexing	Time/frequency cluster extension	(Wu et al., 2019)
Homodyne detector (on-chip)	Quadrature (Gaussian) measurement	(Clark et al., 5 Jun 2025)
PNR detector (SNSPD, TES)	Fock-state, non-Gaussian detection	(Masada et al., 2015)

4. Measurement, Readout, and Error Models

• Homodyne detection measures $n=0,1,2,\ldots$0 with near-unit efficiency and $n=0,1,2,\ldots$1 GHz bandwidth on integrated platforms. This readout is deterministic for Gaussian circuits and serves as the principal method for quantum state tomography and MBQC (Andersen et al., 2010, Masada et al., 2015).
• Heterodyne detection enables simultaneous, but noisy, estimation of both quadratures.
• Photon-number-resolving detectors (PNR) permit measurement in the Fock basis, essential for heralding non-Gaussian states (e.g., photon subtraction) and universal computation (Choe, 2022).
• On–off ("Geiger") detectors provide “click/no-click” information for non-Gaussian event heralding; these are simpler but suffer from lower resolution.

Error sources include:

• Finite squeezing ($n=0,1,2,\ldots$2 < 15 dB), leading to residual Gaussian noise;
• Optical loss from waveguides, couplings, and detectors;
• Dark counts and quantum efficiency limitations in PNR detectors;
• Mode mismatch and phase noise (Choe, 2022, Lenzini et al., 2018).

Error correction leverages bosonic codes, notably the Gottesman–Kitaev–Preskill (GKP) code, which encodes logical qubits in oscillator superpositions and allows correction of small displacement errors in both quadratures. Fault tolerance requires $n=0,1,2,\ldots$3–13 dB of cluster squeezing and low overall loss (Renault et al., 2024, Wu et al., 2019).

5. Quantum Algorithms, Advantage, and Applications

Computational Complexity and Quantum Advantage

CV circuits with only Gaussian resources and homodyne detection are classically efficiently simulable (via the covariance matrix formalism) (Chabaud, 2021). Quantum computational supremacy emerges when non-Gaussian operations are included, specifically for tasks such as boson sampling, grounded in the complexity of matrix Hafnians or permanents (Chabaud, 2021). Sampling from non-Gaussian CV circuits with Gaussian measurements is provably classically hard under standard complexity assumptions.

Variational and Optimization Algorithms

Variational quantum algorithms (VQAs), such as CV-QAOA and photonic counterdiabatic quantum optimization, exploit the infinite-dimensionality and native continuous encoding to perform optimization over both continuous and integer variables. These algorithms demonstrate practical convergence on polynomial and integer programming tasks using generalized gate ansätze that combine Gaussian layers with non-Gaussian measurements, often commercialized via silicon nanophotonic chips with 8+ modes (Enomoto et al., 2022, Chandarana et al., 2023).

Quantum Machine Learning

Continuous-variable quantum neural network architectures leverage the natural affinity of CV bosonic modes for vector-like or function-like representations. CV-QONNs and MBQC-based quantum reservoir computers utilize combinations of Gaussian optics, measurement-induced nonlinearities (e.g., photon subtraction or feedforward conditioned on ancilla measurement), and large entangled resource states for function approximation, classification, and distributed learning—capable of universal approximation in a single layer for sufficiently rich non-Gaussian activation (Krasimirov-Ivanov et al., 4 Dec 2025, García-Beni et al., 2024, Bangar et al., 2023).

Quantum Simulation and Classical Dynamics

CVPQC is an efficient platform for simulating quantum dynamics and even certain classical systems in the Koopman–von Neumann formalism. Gaussian operations implement arbitrary quadratic (Liouvillian) flows, while non-Gaussian gadgets permit nonlinear evolution, with resource-efficient scaling for large phase-space problems and quantum field theories, circumventing the need to discretize field amplitudes (Abel et al., 2024, Gao et al., 15 Dec 2025).

6. Fault-Tolerance, Scalability, and Engineering Prospects

Architecture-level designs explicitly target passive, scalable photonic integration, employing high-Q SiN rings for generating large multidimensional CV cluster states, time/frequency multiplexing for resource efficiency, and on-chip homodyne and PNR detection for universal measurement patterns (Renault et al., 2024, Wu et al., 2019, Masada et al., 2015). Threshold theorems and simulation confirm that logical error rates below the surface-code threshold are achievable at ≈ 12–13 dB cluster squeezing (Renault et al., 2024). Magic state generation for universality is feasible within the same squeezing budget using cat-assisted protocols that significantly improve success rates relative to DV schemes.

A direct comparison with DV photonics highlights the advantages in deterministic gate set, exponential Hilbert-space growth per mode ($n=0,1,2,\ldots$4 for $n=0,1,2,\ldots$5 modes, truncated in practice to cutoff$n=0,1,2,\ldots$6), and room-temperature, GHz-scale operation (Choe, 2022, Romero et al., 2024). Remaining technical obstacles include improvement of squeezing and loss, deterministic non-Gaussian gate synthesis, and integration of low-noise bosonic codes for fault-tolerance.

Ongoing advances point toward wafer-scale foundry fabrication of CV quantum photonic processors, supporting application domains from cryptography and optimization to high-dimensional quantum machine learning and simulation (Clark et al., 5 Jun 2025, Krasimirov-Ivanov et al., 4 Dec 2025).

---

References:

(Choe, 2022, Chabaud, 2021, Romero et al., 2024, Andersen et al., 2010, Masada et al., 2015, Clark et al., 5 Jun 2025, Lenzini et al., 2018, Renault et al., 2024, Wu et al., 2019, García-Beni et al., 2024, Chandarana et al., 2023, Enomoto et al., 2022, Shrivastava et al., 2019, Krasimirov-Ivanov et al., 4 Dec 2025, Bangar et al., 2023, Abel et al., 2024, Gao et al., 15 Dec 2025)