Fermionic non-Gaussianity via computable measures and experimentally accessible bounds

AI summary

Overview

Research area: Quantum information and quantum many-body physics — specifically the resource theory of fermionic non-Gaussianity, geometric distance measures to the fermionic Gaussian state manifold, and experimentally measurable witnesses for strongly correlated fermionic systems.

Technical level: Advanced. The paper combines resource-theoretic monotone proofs, an operator-inequality theorem, a self-consistent fixed-point iteration, and numerical benchmarks on the one-dimensional Fermi–Hubbard model.

Scope: The paper defines a fidelity-based geometric measure of fermionic non-Gaussianity, supplies an iterative algorithm to compute it beyond perturbation theory, and derives an experimentally accessible lower bound built only from two- and four-point fermionic correlators.

What This Paper Is About

Fermionic Gaussian states — ground states of quadratic Hamiltonians that obey Wick's theorem — are classically tractable, but the physically and computationally interesting behavior of interacting fermions lies precisely in how far a state departs from that tractable set. Measuring that departure geometrically is hard because the Gaussian manifold is non-convex and because existing approaches rely on perturbative expansions that fail at strong coupling. The authors construct a fidelity-distance measure to the closest Gaussian state, an iteration that finds that closest state numerically, and a mathematically rigorous lower bound on the distance that can be evaluated from correlators routinely measured in quantum simulators.

Key Contributions

1. A geometric non-Gaussianity measure with resource-theoretic standing. The authors define 𝒟(ρ) = min over Gaussian σ of [1 − ℱ(ρ,σ)] using the Uhlmann–Jozsa fidelity, and prove (Proposition 1) that it is a faithful monotone of the resource theory of fermionic non-Gaussianity: it vanishes exactly on the Gaussian set and does not increase under Gaussian operations.

2. A practical iteration for the closest Gaussian state. Lemma 1 shows that stationarity of the fidelity over the Gaussian manifold reduces to a self-consistency condition on the covariance matrix, and Theorem 1 turns this into an explicit update rule obtained by evaluating the relevant integral with the trapezoidal rule. The authors argue that exact and trapezoidal gradients vanish together, so the solutions are unchanged.

3. A sharpened operator Jensen inequality and a correlator-based bound. Theorem 2 proves a two-sided bound on the Jensen gap in terms of the variance of the operator, resolving the gap to leading order in fluctuations. Applied to the fidelity, this yields Theorem 3, an upper bound B(K) on the overlap, and Theorem 4, a lower bound 𝒟^L_ε on the distance that involves only two- and four-point Majorana correlators of the target.

4. A parameter-free witness requiring only one round of measurement. Corollary 1 removes the regulator by maximizing over ε, and Theorem 5 gives the stationarity condition and gradient of the relaxed bound. Because every iteration re-contracts the same fixed correlators classically, all quantum data — M^ψ, μ, Var_ψ(Γ) and S — needs to be measured only once.

Main Findings

• Non-Gaussianity grows monotonically with interaction. For the ground state of the one-dimensional Fermi–Hubbard chain at filling ⌊N/2⌋, the geometric measure 𝒟(|ψ⟩) increases monotonically with U/t for all system sizes considered, and at fixed U/t it increases with system size in the authors' data.

• The witness is non-perturbative and becomes positive past a threshold. The bound 𝒟^L is not guaranteed positive because it derives from a relaxation. It becomes a genuine, positive witness once U/t exceeds a threshold that depends on lattice parity — lower for even-site configurations — and that moves toward weaker interactions as the system size grows.

• The witness beats antiflatness at intermediate to strong coupling. Benchmarked on N = 8 and N = 10 sites at half filling, 𝒟^L vanishes below U/t ≈ 6.2 for N = 8 and below U/t ≈ 5.0 for N = 10. Below those thresholds fermionic antiflatness supplies the only non-trivial bound; above them the ordering reverses and 𝒟^L continues tracking 𝒟 up to the strongest couplings considered.

• Antiflatness has a structural ceiling. The antiflatness bound ℱ₁(|ψ⟩)/(4n), with n = 2N fermionic modes, cannot exceed 1/4 by construction and is nearly independent of system size — the two circle curves in Fig. 2 nearly coincide — whereas 𝒟^L sharpens with N.

• The iteration outperforms simulated annealing in benchmarks. Against simulated annealing over the Gaussian manifold, the fixed-point iteration reached equal or better fidelities at a fraction of the cost, and only a few iterations sufficed to converge.

• The regulator choice is mild. The maximum of 𝒟^L_ε over ε is found to be rather shallow, and the bound at the fixed value ε = 0.02 already lies close to the optimum ε⋆, which is why the benchmarks use ε = 0.02.

• Operational meaning of the measure. No measurement can distinguish the closest Gaussian state σ⋆ from the target with success probability exceeding ½[1 + √𝒟(ρ)], and the distance bounds observable error as |⟨O⟩_ρ − ⟨O⟩σ⋆| ≤ 2‖O‖∞ √𝒟(ρ).

Methodology in Plain English

The authors start from a natural definition of "how non-Gaussian is this state": take the quantum fidelity (a measure of overlap) between the target state and the best-matching Gaussian state, and convert it into a distance. The obstacle is that finding the best-matching Gaussian state means maximizing overlap over a curved, non-convex set, which has no closed-form solution.

Their first move is calculus: they ask what must be true at a maximum, and find that the covariance matrix of the optimal Gaussian state has to equal a certain imaginary-time-averaged expression built from the target state and the trial Hamiltonian. That self-consistency condition becomes an algorithm — start from the Gaussian state whose covariance matrix matches the target's, then repeatedly update the covariance matrix using the current trial state until it stops changing.

Their second move addresses experiment. Full knowledge of the wavefunction is unavailable in a lab, so they prove a general inequality that controls how far the expectation of a matrix exponential can stray from the exponential of the expectation, with the deviation governed by the variance. Applying this to the trial Hamiltonian gives an upper bound on the fidelity that depends only on the first two moments of that Hamiltonian in the target state — quantities assembled from two- and four-point fermionic correlators. Combining this with a regularization that excludes perfectly pure Gaussian states (which would make the bound diverge) produces a rigorous lower bound on the distance. The regularization parameter ε is then optimized away, leaving a parameter-free witness.

Finally, they test everything numerically on exact ground states of the one-dimensional Fermi–Hubbard model across interaction strengths from the free-fermion point upward, comparing the exact iterated distance, the new lower bound, and the existing antiflatness bound.

Why This Matters

Impact on research. The work fills a gap in the resource theory of fermionic non-Gaussianity by supplying a faithful, operationally meaningful distance measure together with a practical way to compute it, and it does so without any expansion in interaction strength. This matters because previous correlator-based approaches — expansions in two-mode entanglement energies or in correlations up to four-point — lose validity precisely in the strongly correlated regimes where non-Gaussianity is largest. The sharpened operator Jensen inequality is also presented as a result of independent interest.

Real-world applications (as discussed by the authors):

• Quantum gas microscopy of Fermi–Hubbard systems: such microscopes already measure two- and four-point correlators and have resolved them to steadily higher order, so the witness can be evaluated with existing capabilities.
• Digital quantum processors: Fermi–Hubbard models are realized on superconducting qubits and trapped ions at scales reaching beyond exact classical simulation, and off-diagonal pairing correlations have recently been measured directly on trapped-ion hardware.
• Matchgate shadow protocols: random matchgate shadows can supply the full set of correlations up to fourth order at polynomial cost, and the same snapshots yield ⟨ψ|σ_K|ψ⟩ for any Gaussian σ_K, showing non-Gaussianity can be certified using only Gaussian operations.
• Semiconductor quantum-dot arrays: proposed as a small-scale testbed where each quantity entering the derivative of 𝒟^L comes with a rigorous bound on the number of snapshots required, so error bars propagate into an error bar on 𝒟^L.

Industry relevance. The certified value indicates how far a prepared state has moved beyond the reach of classical matchgate calculations. As quantum simulators approach regimes where classical numerics can no longer serve as a benchmark, a scalable, measurement-efficient certificate of non-Gaussianity becomes a practical validation tool for hardware platforms aiming at quantum advantage in fermionic simulation.

Future Directions

• Going beyond the benchmarks. The demonstration is restricted to ground states of the one-dimensional Fermi–Hubbard model at or near half filling for even and odd N. Extending the benchmarks to higher dimensions, doped regimes, dynamics, and mixed states (the measure is defined and proven monotone for arbitrary mixed states, but the numerical study uses pure targets) is a natural next step.

• Characterizing the threshold behavior. The parity dependence of the interaction threshold at which 𝒟^L becomes positive, and its scaling with system size, are documented numerically but not derived analytically; understanding this scaling would clarify when the witness is useful in the thermodynamic limit.

• Closing the gap between bound and exact distance. The relaxed optimization landscape differs from the exact one, so the Gaussian state maximizing B need not coincide with the one maximizing the true fidelity. Whether a tighter relaxation or a better-behaved regularization can narrow this gap is open.

• Experimental demonstration. The authors lay out several platforms and protocols but report no experimental implementation; a first measurement of 𝒟^L on a quantum-gas microscope, trapped-ion device, or quantum-dot array would test the practical cost estimates for snapshot counts.

Target Audience

Quantum information theorists and resource-theory researchers interested in non-Gaussianity and matchgate computation; condensed-matter and cold-atom physicists studying strongly correlated fermions; and experimentalists working with quantum gas microscopes, trapped ions, superconducting qubits, or quantum-dot arrays who need scalable, measurement-efficient certificates of many-body state complexity. Readers should be comfortable with fermionic Gaussian states, covariance matrices, Majorana operators, and Uhlmann fidelity, since the paper moves quickly through these tools.

Authors’ abstract

Interacting fermions owe both their non-trivial correlations and their computational power to their departure from Gaussianity. Yet, quantifying that departure geometrically faces two challenges: the Gaussian manifold is non-convex, and perturbative treatments are confined to weak interactions. Here, we introduce a geometric measure of fermionic non-Gaussianity, the fidelity distance of a target state to the manifold of fermionic Gaussian states. We construct an iteration formalism to numerically access it, applicable also at strong coupling. Further, we prove a sharpened operator Jensen inequality, of interest in its own right. We use it to derive a witness that involves only two- and four-point fermionic correlators. Numerical benchmarks on the one-dimensional Fermi--Hubbard model show that the quality of the witness improves with system size and, for strong coupling, is tighter than antiflatness. Our results open a scalable route to charting non-Gaussianity of strongly-interacting fermionic quantum simulators.