Fermionic entropy: an efficiently measurable strong monotone for non-Gaussianity
Source: arXiv:2607.29670 · Published 2026-07-31 · By Lorenzo Leone, Lennart Bittel
TL;DR
This work addresses the problem of quantifying fermionic non-Gaussianity, a key quantum resource beyond free-fermion (fermionic Gaussian) states. The authors focus on the fermionic entropy, defined via the squared Frobenius norm of the state's correlation matrix, and prove it is a strong monotone under Gaussian protocols on pure states—meaning it never increases on average under free fermionic operations. This firmly establishes the fermionic entropy as a proper resource monotone in the resource theory of fermionic non-Gaussianity. Further, they develop an unbiased estimator for the associated fermionic purity that can be measured experimentally with sample complexity independent of system size, achieving O(ε^{-2}) scaling in additive error ε. They prove a continuity bound (Fannes-type inequality) showing asymptotic continuity of the fermionic entropy, implying it upper bounds the asymptotic distillation rate of non-Gaussianity resource. These results enable a tolerant testing algorithm for fermionic Gaussian states that requires only O(n log n) samples, a quadratic improvement over prior work. Finally, the authors apply their monotone to study t-doped Matchgate circuits (free fermion circuits supplemented by t Majorana-local non-Gaussian gates) and prove a linear lower bound on doping needed to form approximate state 2-designs, establishing near-optimal doping levels required for universal-like random dynamics in this architecture.
Key findings
- The fermionic entropy Mf is a strong pure-state Gaussian monotone, fulfilling the key property M(ψ) ≥ Σ_i p_i M(ϕ_i) for pure-state Gaussian protocols (Theorem 1).
- Fermionic purity Pf, related to Mf, can be unbiasedly estimated to additive error ε with O(ε^{-2} log(1/δ)) two-copy measurements, independently of the number of qubits n (Theorem 2).
- Mf obeys an asymptotic continuity bound |Mf(ρ) - Mf(ρ')| ≤ 2n ||ρ - ρ'||_1 which scales linearly with n, improving prior quadratic bounds (Theorem 3).
- Mf upper bounds the asymptotic distillation rate of fermionic non-Gaussianity under Gaussian protocols, limiting resource conversion rates (Corollary 1).
- A tolerant testing algorithm for determining if an unknown state is εA-close or εB-far from any pure fermionic Gaussian state uses O(n ε_B^{-2} log(n ε_B^{-2}) log(1/δ)) samples, quadratic improvement in n (Theorem 4).
- A linear number t ≥ n/(8κ) of Majorana-local non-Gaussian doping gates (with locality κ=3 or 4) in Matchgate circuits is necessary to generate approximate state 2-designs with error less than 0.4% (Theorem 5).
- Known constructions show nearly linear doping O(n log n) suffices to achieve relative-error unitary k-designs in doped Matchgate circuits, nearly matching the lower bound up to logarithmic factors.
Threat model
The adversary is restricted to Gaussian protocols composed of free fermion operations such as Matchgate unitaries, partial traces, computational basis measurements, and conditioning on measurement outcomes. They cannot create or increase fermionic non-Gaussianity. The threat is to correctly quantify and certify the resourcefulness of states despite any allowed free fermion manipulations. The adversary cannot perform arbitrary unitaries beyond these constraints, so resource monotones must be nonincreasing under these free operations.
Methodology — deep read
Threat Model and Assumptions: The paper assumes an adversary limited to Gaussian protocols which are composed of Matchgate unitaries, partial traces, computational basis measurements, and classical conditioning. These operations cannot increase fermionic non-Gaussianity. The resource theory framework categorizes fermionic Gaussian states as free and aims to quantify non-Gaussianity as a resource monotone that does not increase under these free operations.
Data and State Description: The system considered is n qubits with associated 2n Majorana modes. States considered include pure states, mixed states (via convex roof extension), free fermionic Gaussian states, and states obtained by doping Matchgate circuits with Majorana-local non-Gaussian gates. The correlation matrix Γ(ρ) of size 2n×2n encodes all two-point Majorana correlators for state ρ.
Architecture/Algorithm: The central object is the fermionic entropy Mf(ψ) defined as Mf(ψ) = n(1 - Pf(ψ)), where Pf(ψ) = ||Γ(ψ)||_2^2/(2n) is the fermionic purity, the normalized squared Frobenius norm of the correlation matrix. They prove strong monotonicity (Theorem 1) using a decomposition technique on pure states and covariance matrix analysis. For measurements, they represent Mf as the expectation value of a Hermitian operator Λ^2 acting on two copies of the state. The estimator for Pf is constructed leveraging Bell basis measurements and moment bounds on Λ^2.
Training Regime: No training as such, but measurement protocols require O(ε^{-2} log(1/δ)) copies of two identical states to achieve additive error ε and confidence 1-δ, independent of system size.
Evaluation Protocol: Analytical proofs show strong monotonicity, asymptotic continuity via a Fannes-type inequality, and operational upper bounds for distillation rates. Sample complexity and tolerant testing bounds are derived mathematically by bounding moments, applying Bernstein-Chernoff concentration inequalities. For unitary designs, they compare moments of doped Matchgate ensembles versus Haar ensembles, using fermionic purity as a distinguishing statistic.
Reproducibility: The paper includes full technical proofs and lemmas in appendices, with constructive measurement schemes and explicit bounds. Code and datasets are not provided, as work is theoretical and analytical.
A concrete example end-to-end is the estimation of fermionic purity: one prepares two copies of state ρ, performs Bell sampling measurements corresponding to the Hermitian operator Λ^2, collects measurement outcomes, and computes an unbiased estimator of Pf with sample complexity independent of n, bounded by O(ε^{-2} log(1/δ)). This allows experimentally feasible certification of fermionic non-Gaussianity.
Technical innovations
- Proof that fermionic entropy Mf defined via squared Frobenius norm of correlation matrix is a strong pure-state Gaussian monotone, resolving a previous open question.
- Development of a measurement scheme unbiasedly estimating fermionic purity Pf with sample complexity O(ε^{-2}) independent of the system size.
- Improved Fannes-type continuity bound on Mf scaling linearly with system size n, enabling operational interpretation as an upper bound on asymptotic resource distillation rate.
- Quadratic improvement on sample complexity for tolerant testing algorithms of fermionic Gaussian states, reducing complexity from O(n^2) to O(n log n).
- Identification of linear lower bound on doping level t of non-Gaussian gates required in Matchgate circuits to form approximate state 2-designs, tightly bounding resource cost for universal-like dynamics.
Baselines vs proposed
- Prior tolerant testing algorithms: sample complexity = O(n^2) vs proposed: O(n log n)
- Measurement complexity for fermionic purity estimation in [12]: scales with system size vs proposed: O(ε^{-2}) independent of system size
- Doped Matchgate circuits forming approximate state 2-designs require at least t = Ω(n) gates vs upper bound constructions requiring t = O(n log n)
Limitations
- The strong monotonicity proof applies explicitly to pure states; extension to a more intrinsic mixed-state monotone remains open.
- Although sample complexity improved to near-linear, it is unknown if O(1) sample tolerant testers for fermionic Gaussianity exist.
- Measurement schemes rely on two-copy access and Bell sampling, which may be experimentally challenging for large systems.
- Analysis focuses on limited notions of unitary designs (mainly state 2-designs); higher design orders and tighter error regimes remain unexplored.
- The continuity bound scales linearly with n, which although improved, still grows with system size, potentially limiting asymptotic robustness.
- No empirical or experimental validation provided—results are fully theoretical and asymptotic in nature.
Open questions / follow-ons
- Can fermionic entropy monotonicity be directly extended or intrinsically defined for mixed states beyond convex roof constructions?
- Is it possible to construct tolerant fermionic Gaussianity testers with constant (O(1)) sample complexity, further improving on near-linear results?
- Can the monotonicity results proven here be generalized to the full family of fermionic antiflatness measures beyond the fermionic entropy?
- What are practical experimental protocols or circuit constructions that minimize complexity for estimating fermionic purity in large-scale quantum platforms?
Why it matters for bot defense
For bot-defense and CAPTCHA practitioners interested in quantum-enabled or quantum-resistant security, this work is relevant as it deepens understanding of fermionic non-Gaussianity—a quantum resource enabling universal quantum computation beyond classically simulable free fermion models. Efficiently measurable monotones like the fermionic entropy allow practical assessment and certification of non-Gaussianity resources, which could underpin future quantum-proof cryptographic primitives or secure random number generation. The sample-efficient testers and measurement schemes established here provide rigorous tools for validating complex quantum states relevant to quantum advantage claims. Furthermore, the results on unitary designs in doped Matchgate circuits highlight the quantum resource costs to achieve universal-like randomness, informing assumptions about what quantum circuits can feasibly be classically simulated or distinguished. Overall, the paper provides theoretical foundations and experimentally approachable methods for quantifying a critical quantum resource, relevant to those monitoring or defending against emerging quantum capabilities.
Cite
@article{arxiv2607_29670,
title={ Fermionic entropy: an efficiently measurable strong monotone for non-Gaussianity },
author={ Lorenzo Leone and Lennart Bittel },
journal={arXiv preprint arXiv:2607.29670},
year={ 2026 },
url={https://arxiv.org/abs/2607.29670}
}