Quantum computing proposal puts ‘over-squeezed’ atoms to work

When worse squeezing becomes useful

Soi-Chan Lei, an independent researcher in Macau, proposes turning an apparent drawback of atomic spin squeezing into a quantum-computing resource. The theoretical and numerical study considers rubidium-87 atoms in a Bose–Einstein condensate inside an optical cavity. Rather than treating each atom as a separate computing unit, it uses their collective spin as an oscillator-like information carrier.

The paper’s first version was submitted to arXiv on November 8, 2026. The supplied record does not establish peer-reviewed publication, and the work reports calculations and simulations—not an experimental quantum processor.

Its central question is whether evolution beyond the best time for precision sensing can generate useful non-Gaussian states. Gaussian states have smooth, bell-shaped phase-space descriptions. Squeezing redistributes their uncertainty, reducing fluctuations in one direction while increasing them in another. The proposal seeks more complex states with interference features that ordinary Gaussian squeezing cannot produce.

Engineering the interaction

The proposed hardware combines cavity-mediated atomic interactions with two optical drives. A repeating control sequence uses rapid collective rotations and phase changes to engineer an effective interaction called two-axis twisting. This periodic-control approach is known as Floquet engineering.

The effective Hamiltonian is derived in a fast-pulsing approximation: the control cycle must be short compared with the interaction timescales. The intended state preparation proceeds through controlled dynamics rather than retaining only runs associated with a successful photon-detection event.

To explain how non-Gaussian behavior emerges, Lei maps the collective spin onto an oscillator description using the Holstein–Primakoff transformation. Near a fully polarized state, the collective spin behaves approximately like an oscillator on a flat phase plane. But a finite atomic ensemble has a curved spin geometry, represented by a Bloch sphere.

In the large-spin, low-excitation limit, two-axis twisting becomes an ordinary Gaussian squeezing operation. At finite atom number, curvature adds nonlinear shearing. Continued evolution can fold the stretched state and produce quantum interference. An asymmetric control sequence can additionally retain a Kerr interaction, whose dependence on excitation number provides another nonlinear contribution.

The paper explicitly limits its analytic expansion to low excitation. Deep in the over-squeezed regime, it instead relies on simulations of the microscopic spin Hamiltonian to include higher-order curvature effects.

What the simulation reports

The proof-of-principle calculation uses 50 atoms, an effective interaction strength of 1.0 MHz and a local relaxation rate equal to 0.0005 times that strength.

Its best open-system spin squeezing occurs at approximately 0.64 microseconds, when the Wineland squeezing parameter reaches about 0.124. Smaller values indicate better metrological squeezing. By 2.00 microseconds, the state has evolved beyond that optimum, and the reconstructed Wigner function develops alternating interference features, with a reported minimum near −0.214.

A Wigner function is a phase-space representation of a quantum state, not an ordinary probability distribution. Negative regions reveal nonclassical structure. Here, they support the proposed mechanism for generating non-Gaussian states; they do not demonstrate an executed computation or fault-tolerant operation.

There is an important qualification: the reconstruction uses the normalized component remaining in the fully symmetric collective-spin subspace. The reported retained weight is approximately 0.946 at 2.00 microseconds. Population outside that subspace is excluded from the displayed oscillator-state reconstruction. This numerical conditioning is distinct from the proposal’s avoidance of photon-heralding post-selection during preparation.

A compact model is not yet a large processor

The computational method separates state evolution from collective-state reconstruction. An exact matrix product operator represents the evolution generator with an internal bond dimension of six, independent of atom number. A dynamic-programming projector then extracts the symmetric density matrix without truncating its description to a limited set of spin moments.

Those properties remove particular representation bottlenecks, but do not make every numerical step exact or cost-independent. The evolving state is compressed, and the paper attributes changes in the squeezing optimum to both dissipation and bond compression.

Lei proposes extending the architecture to ensembles of 1,000–100,000 atoms. This paper’s detailed over-squeezed benchmark, however, remains at 50 atoms. Larger-scale non-Gaussian dynamics, experimental control and fault-tolerant computing are not demonstrated. Coupling multiple modes is identified as future work. The present result is a proposed physical mechanism and a numerical example—not a completed quantum computer.