CV Cluster States in Quantum MBQC

• CV cluster states are multipartite entangled Gaussian states defined by graph-structured correlations between qumodes, serving as foundational resources for MBQC.
• Finite squeezing converts ideal nullifier equations into precise Gaussian constraints, underscoring challenges in experimental realization and computational fidelity.
• Diverse platforms—including optical OPOs, microwave JPAs, and Kerr-microcombs—demonstrate scalable CV cluster state generation for quantum computation and communication.

Searching arXiv for additional recent context on CV cluster states. Continuous-variable (CV) cluster states are multipartite entangled Gaussian states of bosonic modes (“qumodes”), defined by graph-structured correlations between canonical quadratures and used as the pre-entangled resource for measurement-based quantum computation (MBQC), where computation proceeds by local measurements and feedforward rather than by an online sequence of entangling gates (Lingua et al., 8 Apr 2026). In the idealized construction, one begins from momentum eigenstates and applies controlled-phase couplings according to a graph adjacency matrix; in realistic settings, the same resource is described as a finitely squeezed Gaussian graph state with a complex adjacency matrix and nonzero nullifier variance (Alexander et al., 2015). Across the literature, CV cluster states appear both as a foundational MBQC resource and as a broader Gaussian entanglement substrate for quantum networking, cryptography, bosonic error-correction resource generation, and CV-to-qubit interfaces (Oruganti, 2024).

1. Graph-state definition and Gaussian formalism

In the graph-state formulation, an ideal CV cluster state on $N$ modes is written as

$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$

where $A$ is the real adjacency matrix specifying the graph connectivity (Lingua et al., 8 Apr 2026). The associated graph nullifiers are

$\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$

and the ideal cluster-state condition is

$$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$

Equivalent formulations appear throughout the optical and microwave literature, sometimes with $\hat q,\hat p$ notation and sometimes with $X,Y$ amplitude/phase quadratures (Rigas et al., 2012).

Because ideal quadrature eigenstates are unphysical, the experimentally relevant object is a finitely squeezed Gaussian graph state. A standard representation uses the complex adjacency matrix

$Z=A+iU,$

where $A$ is the canonical graph and $U$ encodes finite squeezing (Lingua et al., 8 Apr 2026). In the Menicucci-style graphical calculus, a pure Gaussian state is represented by

$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$0

and the covariance matrix is

$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$1

so graph structure and finite-squeezing noise are treated within one object (Alexander et al., 2015).

A complementary phase-space description makes the finite-squeezing approximation explicit. For a finitely squeezed $$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$2-mode cluster state with nullifier coordinates $$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$3, the Wigner function can be written as

$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$4

which reduces in the infinite-squeezing limit to a product of delta constraints,

$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$5

This makes precise the usual statement that finite squeezing turns exact nullifier equations into narrow Gaussian constraints around the target graph correlations (Lau et al., 2013).

2. Geometry, universality, and macronode structure

In CV-MBQC, graph geometry is not a superficial detail: it determines what information can flow and what computations can be implemented by adaptive measurement. A central distinction in the literature is that a two-dimensional square-lattice graph is the theoretically established universal resource geometry for CV-MBQC, whereas 1D cluster states or quasi-1D structures such as ladders do not provide the same computational universality (Lingua et al., 8 Apr 2026). This is why the move from large 1D resources to genuine 2D connectivity is repeatedly treated as a structural milestone rather than a mere increase in entangled mode count.

One important family of architectures uses dual-rail and bilayer constructions rather than canonical one-mode-per-site square lattices. The bilayer square lattice generated from a single optical parametric oscillator combines frequency and time multiplexing into a resource that is computationally universal, with each lattice site realized as a macronode containing two physical qumodes, one in each polarization (Alexander et al., 2015). Related work on the quad-rail lattice shows that computing directly on a macronode resource, instead of first reducing it to an ordinary square lattice, provides additional routing flexibility, compactness, and control over multimode Gaussian circuits because each lattice location contains several physical modes and therefore extra local measurement degrees of freedom (Alexander et al., 2016).

This macronode viewpoint recurs across several platforms. In optical settings it underlies bilayer square-lattice and quad-rail protocols; in the integrated microcomb architecture it organizes the 3D lattice into four-mode macronodes and distributed modes; in dual-rail cryptographic constructions it motivates quotienting the infinite graph into a smaller effective state (Wu et al., 2019). A consistent theme is that scalable CV cluster-state generation often produces richer graph objects than the canonical square lattice, and that this additional structure can be exploited rather than discarded.

3. Physical realizations and scaling architectures

CV cluster states have been pursued in optical, microwave, integrated photonic, spatial-mode, polarization-spatial, and phononic settings. The common strategy is to generate Gaussian entanglement efficiently in a large mode basis—frequency, time, spatial mode, polarization-spatial hybrids, or mechanical normal modes—and then shape the resulting graph with passive interference, delay, pump engineering, or local measurement choices. Representative examples are summarized below (Alexander et al., 2015, Lingua et al., 8 Apr 2026, Wu et al., 2019, Govindarajan et al., 16 Sep 2025, Pooser et al., 2014, Rigas et al., 2012).

Platform	Mode structure	Representative capability
Single-OPO time-frequency optics	Frequency $$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$6 time bilayer square lattice	$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$7 (Alexander et al., 2015)
Microwave JPA	Propagating microwave frequency modes	2D square and honeycomb CV cluster states across 191 modes (Lingua et al., 8 Apr 2026)
Kerr-microcomb silicon nitride photonics	Frequency $$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$8 time and frequency$$\ket{\Psi}=C_\text{z} \ket{0}_{p_1}\ket{0}_{p_2}\dots\ket{0}_{p_N}, \qquad C_z = \prod_{i,j}^N e^{i A_{ij} x_i x_j}=e^{i\vec{x}^T A\vec{x}},$$9frequency$A$0time	$A$1 (Wu et al., 2019)
Phononic quantum network	Mechanical normal modes in waveguide-coupled resonators	Scalable pulsed generation with $A$2 driving tones for $A$3 resonators (Govindarajan et al., 16 Sep 2025)
Spatial and polarization-spatial optics	Optical spatial mode comb; co-propagating vector-beam modes	Multimode spatial-comb proposal; four-node cluster from a squeezed cylindrically polarized beam (Pooser et al., 2014, Rigas et al., 2012)

The optical literature developed two especially scalable directions. One combines time and frequency multiplexing so that each frequency rail supports a very long temporal dual-rail cluster, and then fuses multiple such resources into higher-dimensional states (Du et al., 2022). Another uses a single type-II optical parametric oscillator plus simple interferometry to generate a bilayer square lattice that is universal for Gaussian MBQC and compatible with non-Gaussian extensions (Alexander et al., 2015). Integrated silicon-nitride Kerr-microcomb proposals bring the same logic onto a chip, using below-threshold microrings for quantum two-mode squeezing and above-threshold soliton combs as phase-coherent local-oscillator references (Wu et al., 2019).

The microwave realization introduced a distinct route to large graph synthesis. A single Josephson Parametric Amplifier pumped by a multitone waveform near $A$4 can engineer interference between mixing products so that desired graph edges add constructively and unwanted ones cancel destructively, enabling square and honeycomb 2D cluster states directly in propagating microwave radiation (Lingua et al., 8 Apr 2026). In the phononic proposal, pulsed cavity-mediated Gaussian operations on waveguide-coupled mechanical normal modes generate weighted cluster-state graphs with only local, modular controls (Govindarajan et al., 16 Sep 2025).

4. Verification, nullifiers, and resource quality

The principal diagnostic across the literature is the nullifier test. For a target graph with nullifiers $A$5, one compares the measured nullifier variance $A$6 against the vacuum reference $A$7; for a finitely squeezed cluster state one expects

$A$8

Below-vacuum nullifier variance is therefore the standard signature of graph-structured CV cluster correlations (Lingua et al., 8 Apr 2026).

In the first microwave-domain realization of genuine 2D CV cluster states, nullifier squeezing reached up to $A$9 dB below vacuum for the square lattice and up to $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$0 dB below vacuum for the honeycomb lattice, corresponding respectively to $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$1 and $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$2 standard deviations below vacuum (Lingua et al., 8 Apr 2026). The same work also introduced an explicit hidden-entanglement analysis. In that formalism, unwanted correlations appear as off-diagonal elements of the finite-squeezing matrix $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$3, and a Hidden Entanglement Ratio quantifies their weight relative to the desired canonical graph correlations. The reported conclusion was that there was no apparent hidden entanglement up to about $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$4 dB of nullifier squeezing, and that at optimal squeezing the hidden correlations were “a factor of 5 weaker,” with $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$5 (Lingua et al., 8 Apr 2026).

Nullifier thresholds also appear in optical proposals. In a hybrid time-frequency architecture, the 1D dual-rail nullifiers satisfy

$\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$6

for full inseparability, leading to the statement that squeezing stronger than about $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$7 is sufficient for the 1D cluster state, while the 3D construction requires about $\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$8 (Du et al., 2022). In phononic networks, the average nullifier

$\hat{\delta}_j=\hat{p}_j-\sum_k A_{jk}\hat{x}_k,$9

serves as the cluster-state quality metric, and under dissipation the authors find a non-monotonic dependence on squeezing because stronger squeezing also increases phonon occupation and therefore vulnerability to loss (Govindarajan et al., 16 Sep 2025).

An important methodological point is that nullifier squeezing is not the whole story. The microwave work explicitly emphasizes that hidden correlations can compromise how cleanly an experimental Gaussian state matches the intended canonical graph even when the target nullifiers are squeezed (Lingua et al., 8 Apr 2026). This has made resource-quality analysis increasingly graph-aware rather than relying on mode-by-mode squeezing figures alone.

5. Computation, gates, and derived resource states

Small CV cluster states already support explicit one-way gate implementations. A four-mode linear optical cluster was analyzed as a resource for phase-space displacement, a single-mode squeezing operation, and a controlled-$$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$0 gate, with finite squeezing entering directly into gate precision and fidelity (0811.2887). In a related four-partite optical experiment, a linear cluster state together with two CV teleportation elements was used toward a measurement-based controlled-$$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$1 gate; with cluster correlation squeezing about $$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$2 dB below SNL and feedforward gains $$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$3, the reported fidelities were $$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$4 with the experimental cluster state versus $$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$5 without cluster resources (Wang et al., 2010). These small-scale studies established the canonical MBQC workflow: prepare the cluster offline, couple inputs by Bell-type interfaces, measure selected quadratures, and complete the gate by feedforward.

The operational scope of CV cluster states has expanded well beyond Gaussian gate teleportation. In quantum secret sharing, the cluster-state formalism provides the graph, measurement, and security structure for CC, CQ, and QQ protocols, with finite squeezing treated explicitly as a leakage mechanism absent in the ideal infinite-squeezing limit (Lau et al., 2013). In quantum cryptography, an infinite dual-rail cluster state can be quotient-reduced by a node-coloring scheme to a six-mode pure Gaussian graph state for three-user conference key agreement, with the notable feature that the dual-rail-derived quotient state can generate bipartite keys post-QCKA, a feature not achievable with GHZ states (Oruganti, 2024).

Several recent works reinterpret CV cluster states as parent resources for non-Gaussian or discrete-variable structures. One proposal shows that a CV cluster state contains, in the displaced GKP basis, a superposition of qubit cluster states on the same graph, and gives a download protocol that transfers the embedded entanglement to physical qubits. In that analysis, finite squeezing maps to qubit erasure with probability

$$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$6

leading to squeezing targets of about $$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$7 dB for robust qubit memory or non-fault-tolerant quantum computation and $$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$8 dB for fault-tolerant quantum computation (Han et al., 11 Apr 2025). Another work introduces PhANTM, a photon-counting-assisted node-teleportation method that applies polynomial filters during cluster teleportation and thereby generates embedded cat states and approximate grid/GKP states inside an otherwise Gaussian cluster (Eaton et al., 2021).

These developments suggest a broader interpretation of CV cluster states as entanglement backbones. They are not only substrates for Gaussian MBQC, but also intermediaries from which one can derive cryptographic graph states, many-qubit cluster states, cat states, and grid states by suitably chosen local measurements and feedforward (Oruganti, 2024, Han et al., 11 Apr 2025, Eaton et al., 2021).

6. Finite squeezing, excessive entanglement, and computational limits

The dominant practical limitation is finite squeezing. Across optical, microwave, and mechanical implementations, finite squeezing converts exact nullifier equations into approximate ones, introduces additive Gaussian noise into teleportation-based gates, and amplifies the impact of loss and mode mismatch (Wang et al., 2010, Lingua et al., 8 Apr 2026). Large graph size does not by itself solve this problem. The microwave 2D experiment explicitly identifies finite squeezing, loss, residual higher-order mixing, and mode nonuniformity away from resonance as practical constraints, and notes that fault-tolerant CV-MBQC will likely require stronger squeezing together with non-Gaussian resources or bosonic encodings such as GKP-type approaches (Lingua et al., 8 Apr 2026).

A related conceptual limit is that “cluster state” and “large entangled Gaussian state” are not interchangeable notions of computational usefulness. The literature repeatedly distinguishes large 1D or quasi-1D resources from genuinely universal 2D geometries (Lingua et al., 8 Apr 2026). It also distinguishes generic multipartite entanglement from MBQC suitability. In one recent theoretical development, absolutely maximal entanglement (AME) is defined for CV systems and shown to be generic among infinitely squeezed Gaussian states; in particular, CV cluster states are generically AME. The same work argues that this does not make them useless overall, but suggests that generic highly entangled CV cluster states are unlikely to be useful in the specific way required for generic MBQC speedup, while being highly useful for multiparty teleportation, secret sharing, key distribution, Gaussian perfect-tensor networks, and Gaussian multi-unitary circuits (Kwon et al., 19 Mar 2025). Exact CV AME states, however, are non-normalizable and therefore unphysical, so the operational discussion necessarily returns to finitely squeezed approximations (Kwon et al., 19 Mar 2025).

A more recent complexity-theoretic analysis addresses the intrinsic computational power of finitely squeezed CV cluster states through measurement-based linear optics. It identifies a squeezing-driven complexity phase transition: if

$$\hat{\delta}_j\ket{\Psi}=0, \qquad \Delta \delta_j^2=0.$$9

then MBLO-based sampling is classically simulable in polynomial time, whereas under an approximate-GBS hardness assumption there exists a regime $\hat q,\hat p$0 in which the sampling problem is classically intractable (Go et al., 9 Apr 2026). This sharpens the usual finite-squeezing caveat into a complexity statement: low squeezing is not only noisy, but can render the model efficiently classically simulable.

For that reason, the modern picture of CV cluster states is bifurcated. On one side, they remain the most scalable Gaussian resource known for one-way optical and microwave architectures. On the other, finite squeezing, loss, and the eventual need for non-Gaussian processing remain decisive. The field has therefore moved toward hybrid interpretations in which CV cluster states provide large-scale deterministic Gaussian entanglement, while universality, error correction, or discrete-variable extraction are supplied by GKP states, cat states, photon counting, or other non-Gaussian ingredients (Eaton et al., 2021, Han et al., 11 Apr 2025).