---
title: Gaussian-Branched Cat States (GCSs)
url: https://www.emergentmind.com/topics/gaussian-branched-cat-states-gcss
type: topic
---

# Gaussian-Branched Cat States (GCSs)

Gaussian-Branched Cat States (GCSs) are hybrid or bosonic non-Gaussian states built from coherent superpositions of Gaussian branches. In the explicit terminology of the hybrid continuous-variable/discrete-variable formalism, the name denotes the \(N\)-qubits \(n\)-modes entangled states arising during “superpositions of Gaussian processes,” fully characterised by superposed phase-space quantities—generalised complex first moments, covariance matrices, and the qubit reduced density matrix (QRDM) [2510.01156]. In a closely related bosonic formulation, a GCS is any pure or mixed state whose support lies in the span of a finite set of pure Gaussian branches, so that cat states, squeezed-cat states, Gaussian multiplets, and related bosonic encodings all appear as special cases [2603.15258].

## 1. Definition and state-space structure

For a single qubit coupled to \(n\) bosonic modes, the hybrid density operator can be decomposed as
\[
\hat\varrho = \sum_{j,k\in\{\pm1\}} \hat\varrho_{jk}\otimes |j\rangle\langle k| ,
\]
and each matrix element is assigned a branched characteristic function
\[
\chi_{jk}(\tilde r)=
\exp\!\left[
-\frac14 \tilde r^{\rm T}\sigma_{jk}\tilde r
+i\,\tilde r^{\rm T}r_{jk}
+r^{(0)}_{jk}
\right].
\]
For \(N\) qubits, the branch labels become bit strings \(J,K\in\mathcal M\), and the general GCS ansatz is
\[
\chi_{JK}(\tilde r)=
\exp\!\left[
-\frac14 \tilde r^{\rm T}\sigma_{JK}\tilde r
+i\,\tilde r^{\rm T}r_{JK}
-\mathcal C_{JK}+i\phi_{JK}
\right].
\]
Diagonal branches \(J=K\) are physical Gaussian states with real covariance matrices and first moments; off-diagonal branches encode interference and may have complex phase-space data. The QRDM is recovered at \(\tilde r=0\), so \(\varrho^q_{JK}=e^{-\mathcal C_{JK}+i\phi_{JK}}\) [2510.01156].

In the finite-manifold bosonic formulation, one starts from a finite set of normalized, linearly independent, pure Gaussian states \(\mathcal B_D=\{\ket{g_k}\}_{k=0}^{D-1}\) and defines
\[
(\mathcal B_D):=\mathrm{span}\{\ket{g_0},\dots,\ket{g_{D-1}}\}.
\]
Any pure state on this manifold has the form
\[
\ket{c}=\sum_{k=0}^{D-1} c_k\ket{g_k}\equiv \Psi c,
\qquad
\braket{c}{c}=c^\dagger G c,
\qquad
G_{jk}=\braket{g_j}{g_k}.
\]
Standard cat states, such as \(\ket{\mathrm{cat}_\pm}\propto \ket{\alpha}\pm\ket{-\alpha}\), and squeezed-cat states are exactly \(D=2\) Gaussian-branch superpositions. In this sense, “two-branch Gaussian superpositions,” “Gaussian multiplets,” and finite Gaussian branch manifolds provide a bosonic realization of GCSs [2603.15258].

A recurrent misconception is to identify GCSs only with superpositions of two coherent states. The finite-span formulation is broader: the branches may be displaced squeezed vacua, generic pure Gaussian states, or multimode tensor-product Gaussians; the branch number may be \(D=2\) or larger; and the state may be pure or mixed, provided its support remains inside the finite Gaussian manifold [2603.15258].

## 2. Gaussian formalism for dynamics and measurements

The hybrid formalism is designed for Hamiltonians that are Gaussian in the bosonic variables and diagonal in the qubit basis. For one qubit, the operator-valued Gaussian Hamiltonian is
\[
\hat H_g(\hat r,\hat\sigma_z)
=
\frac12\hat r^{\rm T}H_{\hat\sigma}\hat r-r_{\hat\sigma}^{\rm T}\hat r+\frac12 H_q^0\,\hat\sigma_z,
\]
with
\[
H_{\hat\sigma}=H_m+H_q\hat\sigma_z,\qquad
r_{\hat\sigma}=r_m+r_q\hat\sigma_z.
\]
Conditioned on a qubit eigenvalue \(j=\pm1\), the modes evolve under a branch Hamiltonian
\[
H_j(\hat r)=\frac12\hat r^{\rm T}H_j\hat r-r_j^{\rm T}\hat r+\frac12H_q^0j,
\qquad
H_j=H_m+jH_q,\quad r_j=r_m+jr_q.
\]
This is the precise sense in which the theory describes superpositions of Gaussian processes: each branch is a standard Gaussian evolution, while the full hybrid state is their coherent superposition [2510.01156].

Imposing the GCS ansatz on the von Neumann equation yields closed ODEs for the complex branch covariances, first moments, and QRDM exponents. In the generic unitary case, the covariance matrices obey Riccati-type equations; in the important linear operator-valued-force case \(H_q=0\), the solution closes analytically in terms of a single mode propagator \(S_m(\tau)=e^{\tau\Omega H_m}\), and all branches share the same covariance evolution while contrasts \(\mathcal C_{JK}\) and phases \(\phi_{JK}\) encode separation and dephasing between branches [2510.01156].

The same closure persists under Markovian Gaussian noise. The most general open dynamics treated combines quadratic Hamiltonians, linear drift \(d^{\rm T}\hat r\), Gaussian diffusion \(D\), antisymmetric drift \(E\), and qubit dephasing \(\Gamma_z\). The branch covariances then satisfy the open-system generalization of the Riccati/Lyapunov equations, while the QRDM exponents acquire the additional term \(\frac{\Gamma_z}{2}(jk-1)\). This gives an exact, truncation-free treatment of unitary and open hybrid dynamics within the GCS family [2510.01156].

Measurements are equally natural in this language. Projective qubit measurements produce conditional bosonic states whose characteristic functions are linear combinations of branch Gaussians, hence generally non-Gaussian. Gaussian measurements on the modes are described by a reference Gaussian POVM with covariance \(\sigma_m\); the post-measurement QRDM element is
\[
\varrho^{q,\text{post}}_{JK}(r_m)
=
\varrho^q_{JK}\,
\frac{
\exp\!\big[-(r_{JK}-r_m)^{\rm T}(\sigma_{JK}+\sigma_m)^{-1}(r_{JK}-r_m)\big]
}{
\pi^n\sqrt{\det(\sigma_{JK}+\sigma_m)}
}.
\]
Noisy homodyne and heterodyne detection follow as special cases. The formalism therefore covers ideal and noisy measurements without leaving the branch representation [2510.01156].

## 3. Canonical families and antecedents

Several earlier state families can be naturally interpreted as GCSs. One major example is the family of superpositions of coherent states determined by quadratic Gauss sums. Starting from the harmonic oscillator and the fractional evolution operator
\[
U(\varphi)=e^{-i\varphi(\hat L-1/2)},
\]
one obtains
\[
|\alpha\rangle_\varphi
=
e^{-|\alpha|^2/2}
\sum_{n=0}^\infty
\frac{\alpha^n e^{-i\varphi n(n-1)/2}}{\sqrt{n!}}\ket n.
\]
For rational angles \(\varphi=2\pi M/N\), this becomes a finite superposition of coherent states at the vertices of a regular \(N\)-gon,
\[
|\alpha\rangle_\varphi=
\sum_{k=0}^{N-1} C_k\ket{\alpha_k},
\]
with coefficients \(C_k\) determined by quadratic Gauss sums. The \(N=2\) case recovers the Yurke–Stoler coherent state, while \(N>2\) yields multi-component “kittens.” The paper explicitly notes that there is no Gaussian envelope in \(k\)-space; the structured object is the phase pattern across branches, not a Gaussian weight over branches [2112.07613].

A second antecedent is the group-theoretic construction of “crystallized” Schrödinger cat states from cyclic and dihedral symmetries. There the seed is a generic Gaussian \(\phi(x)\), and the branches are generated by phase-space rotations \(\hat R(\theta_j)\) or by dihedral operations combining rotations with conjugation/reflection. The resulting states are superpositions of Gaussians whose centers form regular polygons or dihedral completions in phase space, and whose Wigner functions and symplectic tomograms are explicit sums of Gaussian terms. The number and pattern of tomographic maxima and minima reflect the order of the underlying symmetry group [2204.03373].

A third antecedent is the theory of generalized Gaussian cat states, where one superposes arbitrary pure Gaussian states,
\[
|\Psi\rangle = a\,|\mathsf U,u\rangle + b\,|\mathsf V,v\rangle.
\]
Its Wigner function has the standard decomposition into two Gaussian hills plus an interference term,
\[
W_\Psi(x)=|a|^2W_{|\mathsf U,u\rangle}(x)+|b|^2W_{|\mathsf V,v\rangle}(x)+|ab|\,\mathcal I(x),
\]
but the interference phase is no longer generically linear. In arbitrary dimension it is governed by a quadratic form; in one degree of freedom the phase is hyperbolic. The same phase-space structure survives evolution in a thermal reservoir, and the paper also discusses mixed Gaussian superpositions generated by conditional Gaussian operations or by Kerr-type dynamics on thermal states [1002.2248].

These precursors differ in motivation—fractional revivals, finite groups, phase-space geometry—but they share the defining GCS feature: non-Gaussian states assembled from a finite number of Gaussian branches with structured coherence relations.

## 4. Finite-manifold encoding and quantitative characterization

The finite-branch bosonic formalism yields an exact finite-dimensional representation of any state supported on a Gaussian branch manifold. Given \(\Psi=[\,\ket{g_0}\ \cdots\ \ket{g_{D-1}}\,]\) and Gram matrix \(G=\Psi^\dagger\Psi\), one defines the Löwdin-orthonormalized basis
\[
\Phi=\Psi G^{-1/2},
\qquad
\Phi^\dagger\Phi=I_D.
\]
Any mixed state on the manifold can then be represented by an effective \(D\times D\) density matrix
\[
\rho_{\rm eff}
=
\sum_{\mu=1}^M p_\mu\,\tilde c^{(\mu)}\tilde c^{(\mu)\dagger},
\qquad
\tilde c^{(\mu)}=
\frac{G^{1/2}c^{(\mu)}}{\sqrt{c^{(\mu)\dagger}Gc^{(\mu)}}}.
\]
The map is an exact isospectral encoding onto a finite \(D\)-dimensional Hilbert space, so all spectral quantities can be computed without Fock-space truncation [2603.15258].

This immediately gives exact formulas for von Neumann and Rényi entropies from the eigenvalues of \(\rho_{\rm eff}\), and for relative-entropy non-Gaussianity
\[
\delta_{\rm nG}(\rho)=S(\tau(\rho))-S(\rho),
\]
where \(\tau(\rho)\) is the Gaussian state with the same first and second moments as \(\rho\). For single-mode or multimode two-branch GCSs, the necessary first and second moments are assembled from branch data and Gaussian cross-moments, all obtainable analytically from Gaussian integrals [2603.15258].

In the bipartite two-branch case, the effective manifold is \(2\times2\), and after local Löwdin orthogonalization one obtains an exact two-qubit density matrix. For the pure Bell-like state
\[
\ket{\Psi_\varphi}
=
\frac{1}{\sqrt{\mathcal Z_\varphi}}
\bigl(\ket{\Gamma_1}+e^{i\varphi}\ket{\Gamma_2}\bigr),
\]
with local overlaps \(a=\langle g_1^A|g_2^A\rangle\) and \(b=\langle g_1^B|g_2^B\rangle\), the entanglement negativity has the closed form
\[
\mathcal N(\Psi_\varphi)
=
\frac{\sqrt{(1-a^2)(1-b^2)}}{2(1+ab\cos\varphi)}.
\]
The same framework also gives an exact “which-branch dephasing” model, both for single-mode cat decoherence and for Bell-like two-branch GCSs, while keeping the support inside the same finite branch span [2603.15258].

A complementary computational development appears in the exact simulation of realistic GKP cluster states, where resource states are represented as sums of Gaussian distributions in phase space and propagated analytically through Gaussian circuits and homodyne measurements. That formalism exactly simulates cat-bred GKP states and multimode GKP clusters more efficiently than standard Fock-basis truncations, while also clarifying where heuristic Gaussian random noise models fail at the level of conditional stabilizer expectation values [2504.10606].

## 5. Generation and control protocols

A general route to GCSs is conditional measurement on multimode Gaussian resources. For an arbitrary \(N\)-mode Gaussian input, photon-number-resolving detection on \(N-1\) modes leaves the heralded mode in the form
\[
|\psi_1\rangle
=
\hat D(\beta)\,\hat S(\zeta)\,
\sum_{n=0}^{n_{\max}} c_n \ket n,
\]
with \(n_{\max}\le n_T\), the total detected photon number. Thus the output factorizes into a Gaussian gate acting on a finite Fock superposition. Cat states are treated explicitly as targets of this architecture: for small amplitudes, an even cat is well approximated by \(c_0\ket0+c_2\ket2\); for larger amplitudes, by \(\hat S(\zeta_1)(c_0\ket0+c_2\ket2)\). The same formalism extends to mixed Gaussian inputs and experimental imperfections such as photon loss [1902.02331].

In fully optical state engineering, deterministic protocols based on Gaussian operations and photon-number measurements generate large-amplitude squeezed cat states of the form
\[
\ket{\mathcal C_{\alpha,r}^{\pm}}
\equiv
\mathcal N_\pm\bigl(\ket{\alpha,r}\pm\ket{-\alpha,r}\bigr),
\qquad
\ket{\alpha,r}\equiv \hat D(\alpha)\hat S(r)\ket0.
\]
The preparation uses only Gaussian unitaries—beam splitters, squeezers, displacements, phase rotations—and photon-number-resolving detection as the only explicit non-Gaussian ingredient. Scheme I deterministically maps a Fock state \(\ket n\) to a squeezed cat with
\[
|\alpha|\approx \sqrt{n\,\eta^k},
\qquad
r\approx -\frac12\ln(1-\eta^k),
\]
and the resulting cats can be bred into approximate GKP states with
\[
\Delta=\sqrt{\frac{\pi(1-\eta^k)}{2n\eta^k}}.
\]
Scheme II deterministically approaches a two-component cat from arbitrary even or odd parity input states by repeated weak tapping and PNR on squeezed-vacuum ancillas [2311.10510].

Remote preparation furnishes a distributed version of the same paradigm. Starting from a two-mode squeezed vacuum sent through lossy Gaussian channels, single- or multi-photon subtraction at one node and homodyne conditioning remotely prepare optical cat states at the other node. The measurement angle rotates the remotely prepared cat in phase space, and the protocol shows marked asymmetry under loss: the remotely prepared cat state is much more tolerant of loss in Alice’s channel than in Bob’s channel. In the ideal model with 3 dB squeezing and no loss, the fidelity can reach \(\sim99\%\); experimentally, for \(\eta_A=\eta_B\simeq0.9\), the remotely prepared state had \(|\alpha|\simeq0.65\), \(F\simeq0.67\), and central negativity \(W(0,0)\approx -0.10\) [2304.08863].

Within continuous-variable cluster states, the Photon-counting-Assisted Node-Teleportation Method (PhANTM) embeds and preserves cat states directly inside an otherwise Gaussian graph. The realistic PhANTM Kraus operator has the form
\[
\hat K_n
=
\pi^{1/4}\sqrt{\frac{2}{s}}\,
e^{-\hat Q^2/2s^2}\,
\hat R\!\left(\frac{\pi}{2}\right)\,
f_n(\hat Q),
\]
and in the weak-reflectivity/high-squeezing limit reduces to a Hermite-polynomial gate,
\[
\hat K_n \propto
\hat X(m)\hat R\!\left(\frac{\pi}{2}\right)
H_n\!\left(\frac{i\hat Q-m}{\sqrt2}\right).
\]
Repeated PhANTM steps generate cat states from Gaussian nodes, stabilize cat amplitudes against Gaussian noise during cluster teleportation, and embed standard cat-breeding and GKP-breeding protocols inside cluster-state processing [2112.10311].

## 6. Phase-space signatures, applications, and computational uses

The characteristic phase-space signature of a GCS is the coexistence of Gaussian branch peaks and interference terms. In generalized two-branch Gaussian cats, the interference term is a Gaussian envelope multiplied by an oscillatory factor with a quadratic phase, and in one degree of freedom this phase is hyperbolic rather than linear. In cyclic and dihedral Gaussian multiplets, Wigner functions exhibit polygonal or dihedral symmetries and tomograms whose fringe counts reflect the order of the underlying group. In Gauss-sum coherent superpositions, rational fractional-evolution times produce regularly spaced branches whose relative phases are controlled by number-theoretic data [1002.2248], [2204.03373], [2112.07613].

These phase-space structures support several application domains. In bosonic encoding theory, two-branch GCSs realize cat qubits, while larger finite Gaussian multiplets provide cat qudits and related encodings; exact finite-manifold methods then supply entropies, non-Gaussianity measures, and entanglement negativities directly from Gaussian overlaps [2603.15258]. In optical and cavity platforms, deterministic squeezed-cat generation and breeding furnish primitive GKP resources and support error-correction studies that surpass break-even under pure loss in the simulated regimes [2311.10510]. Exact Gaussian-sum simulation methods extend this perspective to multimode GKP Bell pairs and cluster states, including stabilizer and entanglement-witness calculations beyond the range of practical Fock truncations [2504.10606].

The hybrid GCS formalism also captures qubit-mediated Gaussian-process superpositions that are not naturally expressible as finite coherent-state sums. Two paradigmatic examples are treated explicitly. First, a squeezed, leaking, and homodyned resonator dispersively coupled to two qubits yields measurement-based entanglement between the qubits through branch-dependent Gaussian resonator trajectories. Second, a levitated nanoparticle undergoing Stern–Gerlach interferometry in a diffusive environment develops conditional motional Wigner negativity after spin measurement, while the contrast function cleanly separates branch distinguishability, diffusion, and qubit dephasing [2510.01156].

## 7. Limits, equivalences, and interpretive issues

Deterministic Gaussian conversion protocols clarify which non-Gaussian resources lie in the same Gaussian orbit as cats and which do not. A central result is that cat and binomial code states are approximately equivalent already at finite energy: for many parameter pairs \((N,K,\alpha)\), there exists a cat codeword with fidelity \(\gtrsim0.97\) to the corresponding binomial codeword, and the identity map is already optimal in the high-fidelity region. Photon-added and photon-subtracted squeezed states can also be converted to cats with substantially improved performance by adding an optimized single-mode squeezing operation. By contrast, several seemingly favorable conversions fail qualitatively despite moderate or even high fidelities: fidelity alone can be misleading when the target resource depends on specific Wigner-negativity patterns or rotational symmetries [2204.03373].

A second interpretive issue concerns the adjective “Gaussian” in GCS. In the finite-branch literature it refers to the branch states themselves—coherent states, squeezed states, or generic pure Gaussian states—not to a Gaussian weight over branch labels. This is explicit in the Gauss-sum construction, where the envelope in phase space is entirely determined by the coherent-state Gaussians themselves, while the nontrivial structure lies in the phase pattern across branches [2112.07613].

A third limit is exactness. The finite-manifold encoding is exact only for states whose support is strictly inside a finite branch span. Generic Gaussian noise channels such as loss or thermal noise typically move a state outside that manifold, requiring either an enlarged branch set or approximation. Similarly, the exact hybrid GCS formalism assumes Gaussian continuous-variable dynamics and couplings diagonal in the discrete-variable basis; it is exact within that class, but not a universal representation of arbitrary hybrid non-Gaussian dynamics [2603.15258], [2510.01156].

Taken together, these developments position GCSs as a unifying technical category for non-Gaussian states assembled from Gaussian components: finite coherent-state polygons, group-generated Gaussian multiplets, two-branch and multimode bosonic encodings, cluster-embedded cats and grids, and qubit-conditioned superpositions of Gaussian processes. The concept is broad enough to connect cat-state physics, bosonic coding, number-theoretic coherent superpositions, hybrid interferometry, and exact simulation techniques, while remaining concrete enough to admit closed-form dynamics, exact finite-dimensional reductions, and experimentally grounded preparation protocols.

Source: https://www.emergentmind.com/topics/gaussian-branched-cat-states-gcss