Fermionic Antiflatness (FAF)

• Fermionic antiflatness (FAF) is a covariance-matrix-based measure that quantifies fermionic non-Gaussianity by capturing deviations from free-fermion Gaussian states using Majorana two-point correlators.
• It remains invariant under fermionic Gaussian unitaries and exhibits additive properties on tensor products, enabling clear comparisons with other quantum magic measures.
• FAF is practically useful in experimental protocols and large-scale numerics, serving as a diagnostic tool in probing many-body localization, quantum chaos, and interacting regimes.

Fermionic antiflatness (FAF) denotes a family of covariance-matrix-based measures of fermionic non-Gaussianity for pure many-body states. In the formulation used in recent work, FAF quantifies deviation from the class of fermionic Gaussian states—equivalently the free-fermion or matchgate-simulable manifold—and thus measures a fermionic magic resource or, in the language of disordered spin chains, the amount of “complexity beyond free fermions” (Sierant et al., 30 May 2025, Falcão et al., 30 Jan 2026). Its appeal is that it is faithful to Gaussianity, invariant under fermionic Gaussian unitaries, computable directly from Majorana two-point correlators, and sufficiently tractable to be used in analytical estimates, large-scale numerics, and laboratory protocols (Haug et al., 25 May 2026).

1. Definition and basic meaning

For a pure state $|\Psi\rangle$ with Majorana covariance matrix $M$, recent many-body work defines the $k$-th fermionic antiflatness as

$$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$

where $L$ is the number of physical spins or fermionic modes in the chosen encoding (Falcão et al., 30 Jan 2026). In an equivalent $n$-mode notation with antisymmetric covariance matrix $\Gamma_\rho$, one writes

$$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$

with $\nu_j$ the singular values of the covariance matrix (Haug et al., 25 May 2026). For $k=1$, the measure reduces to a quadratic function of bilinear Majorana correlators,

$M$0

in one common normalization, or equivalently

$M$1

in another (Falcão et al., 30 Jan 2026, Haug et al., 25 May 2026).

The reference class is the set of fermionic Gaussian states. These are the states generated by quadratic Majorana Hamiltonians, fully characterized by their two-point Majorana correlators, and obeying Wick factorization (Falcão et al., 30 Jan 2026, Sierant et al., 30 May 2025). For pure Gaussian states the covariance matrix saturates

$M$2

so FAF vanishes exactly on that class (Falcão et al., 30 Jan 2026, Haug et al., 25 May 2026). A positive value therefore signals that the state cannot be described as a free-fermion state.

The term “antiflatness” refers to the covariance-spectrum viewpoint. In Williamson form, the covariance matrix has mode-resolved singular values $M$3 or $M$4; pure Gaussian states have a perfectly flat spectrum at value $M$5, whereas non-Gaussianity appears as a downward deformation, giving

$M$6

in the notation of the general many-body framework (Sierant et al., 30 May 2025). This makes FAF a spectral deficit from Gaussian saturation rather than an optimization distance over the Gaussian manifold.

2. Structural properties and relation to Gaussian distance

Three properties are central in the literature. First, FAF is faithful: $M$7 Second, it is invariant under fermionic Gaussian unitaries $M$8,

$M$9

Third, it is additive on tensor products when one factor has fixed fermionic parity, and subadditive otherwise (Falcão et al., 30 Jan 2026). In the broader commutant-based construction, these properties arise because FAF is built from a replica observable lying in the fermionic commutant of the Gaussian unitary group (Sierant et al., 30 May 2025).

A useful operational comparison is with trace distance to the pure Gaussian set. For a pure state $k$0, define

$k$1

Then the practical-testing work proves the two-sided bound

$k$2

showing that small $k$3 implies closeness to the Gaussian manifold, while states that are $k$4-far from every pure Gaussian state necessarily have $k$5 (Haug et al., 25 May 2026). This does not make FAF identical to a Gaussian distance, but it makes it a quantitatively controlled proxy.

The computational structure is unusually simple. Once the pure state is known, one computes all two-point Majorana correlators, assembles the covariance matrix, and evaluates $k$6; for $k$7, no diagonalization or optimization is required (Falcão et al., 30 Jan 2026). This contrasts with other fermionic magic measures such as fermionic rank or Gaussian extent, which the general resource framework describes as harder to compute because they require minimization or decomposition problems (Sierant et al., 30 May 2025).

3. Measurement protocols and mixed-state witnessing

The most direct experimental development uses $k$8 as a test for leaving the pure Gaussian regime. One two-copy protocol measures commuting observables $k$9 and constructs

$$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$0

with

$$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$1

This yields an $$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$2-tester that distinguishes pure Gaussian states from pure states $$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$3-far from the Gaussian set using

$$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$4

two-copy Bell measurements (Haug et al., 25 May 2026).

A complementary single-copy protocol partitions all Majorana bilinears into $$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$5 commuting matchings and estimates the same quantity from repeated measurements of those commuting layers. The resulting pure-state Gaussianity tester has complexity

$$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$6

and the corresponding additive estimator for $$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$7 scales as $$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$8 up to logarithmic factors (Haug et al., 25 May 2026). The Bell version is one-sided on pure Gaussian inputs, whereas the single-copy version trades that feature for reduced hardware demands.

Because raw $$\mathcal F_k(|\Psi\rangle)=L-\frac12 \mathrm{tr}\!\left[(M^TM)^k\right],$$9 is not by itself a mixed-state non-Gaussianity witness—mixed Gaussian states also have nonzero covariance-spectrum shrinkage—the same work introduces a purity-corrected witness

$L$0

If $L$1, the state is not a mixed fermionic Gaussian state (Haug et al., 25 May 2026). The same Bell data provide both $L$2 and $L$3, so the witness is experimentally accessible. On the IQM Garnet quantum computer, this witness was used to show that noise can both reduce and enhance non-Gaussianity, depending on the circuit and parameter regime (Haug et al., 25 May 2026).

4. Many-body phenomenology in spin chains, localization, and gauge theories

In equilibrium many-body systems, FAF is zero throughout the free-fermion transverse-field Ising chain because every eigenstate is fermionic Gaussian after Jordan–Wigner fermionization (Sierant et al., 30 May 2025). Adding a single quartic impurity produces only $L$4 ground-state FAF away from criticality, whereas an extensive quartic perturbation in the ANNNI model produces

$L$5

with a nonzero density $L$6 throughout the interacting regime (Sierant et al., 30 May 2025). Along the Peschel–Emery line,

$L$7

the periodic-chain ground state becomes exactly fermionic Gaussian again, so $L$8 despite the interacting Hamiltonian (Sierant et al., 30 May 2025). This is one of the clearest examples of FAF detecting hidden free-fermion structure that is not singled out by entanglement alone.

Critical behavior appears in several forms. In the impurity model, FAF itself remains finite but its derivative develops a logarithmic peak,

$L$9

with the peak position approaching the Ising critical point as $n$0 (Sierant et al., 30 May 2025). In the ANNNI model, the subleading contribution obeys

$n$1

for open boundaries at criticality, while for periodic boundaries the corresponding term remains $n$2 (Sierant et al., 30 May 2025). The paper attributes these forms to universal behavior of Majorana correlators and to boundary effects.

In disordered spin chains, FAF has been used as a direct probe of ergodic and many-body localized regimes. For highly excited eigenstates of the disordered XXZ chain and its impurity variant, the disorder-averaged FAF density $n$3 crosses from typical-state behavior at weak disorder to strongly suppressed values deep in the MBL regime (Falcão et al., 30 Jan 2026). The asymptotic scaling depends on interaction support: in the XXZ chain FAF remains extensive, with $n$4, whereas in the impurity model it becomes $n$5-independent at strong disorder, yielding an area-law bound (Falcão et al., 30 Jan 2026). Rare cat-like resonant eigenstates exhibit strongly enhanced FAF; because the cat component has vanishing covariance matrix, FAF directly counts the number of spins participating in the resonance (Falcão et al., 30 Jan 2026).

The measure has also been applied to pure-gauge ladder systems. There the second-order quantity

$n$6

was computed for effective one-dimensional encodings of truncated SU(2) and qubit $n$7 theories. In the Abelian $n$8 case, the FAF density plateau scales as $n$9, so total FAF remains $\Gamma_\rho$0, consistent with a near-free-fermion description after Jordan–Wigner mapping. In the truncated SU(2) case, the weak-coupling FAF density converges rapidly with system size, indicating finite-density fermionic non-Gaussianity (Santra et al., 8 Oct 2025). The same study stresses that this conclusion is encoding-dependent and that FAF was not computed for $\Gamma_\rho$1, because the Majorana mapping was not straightforward in that representation (Santra et al., 8 Oct 2025).

5. Quantum chaos, eigenstate typicality, and SYK

FAF has become a diagnostic of chaotic fermionic states. In generic ergodic many-body systems, highly excited eigenstates exhibit nearly maximal fermionic non-Gaussianity, with the average mid-spectrum behavior approaching the typical-state value

$\Gamma_\rho$2

in the normalization used for spin-chain studies (Sierant et al., 30 May 2025). Under local random circuits initialized in Gaussian states, FAF grows rapidly, becomes extensive at $\Gamma_\rho$3, and approaches the typical value exponentially, while the saturation time to fixed accuracy scales as $\Gamma_\rho$4 (Sierant et al., 30 May 2025). Under local Hamiltonian evolution, by contrast, the saturation time scales linearly in system size, which the same work interprets as a consequence of locality and conservation laws (Sierant et al., 30 May 2025).

The Sachdev–Ye–Kitaev model has supplied a more explicit chaotic and holographic setting. In one normalization for $\Gamma_\rho$5 Majoranas,

$\Gamma_\rho$6

with $\Gamma_\rho$7, Gaussian states have $\Gamma_\rho$8 while Haar-random-like states satisfy $\Gamma_\rho$9 at large $$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$0 (García-García et al., 2 Jul 2026). For Kourkoulou–Maldacena states,

$$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$1

the large-$$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$2 result is

$$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$3

The coefficient of the linear-in-$$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$4 term is tunable by the inverse temperature, interpolating from $$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$5 at $$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$6 to $$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$7 as $$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$8 (García-García et al., 2 Jul 2026). In the holographic interpretation of these states, this corresponds to tuning the magic content of a boundary state dual to a near-AdS$$FAF_k(\rho)=n-\frac12 \mathrm{tr}[(-\Gamma_\rho^2)^k]=n-\sum_{j=1}^n \nu_j^{2k},$$9 black hole with an end-of-the-world particle behind the horizon (García-García et al., 2 Jul 2026).

For Gaussian states evolved in real time under SYK, the same quantity obeys

$\nu_j$0

At $\nu_j$1, $\nu_j$2 starts at zero and approaches $\nu_j$3 exponentially, with a rate given by four times the leading Ruelle–Pollicott resonance in the large-$\nu_j$4 analysis (García-García et al., 2 Jul 2026). Numerical results for exact SYK eigenstates further show $\nu_j$5 throughout the spectral bulk, with subleading corrections decaying exponentially in dense SYK but only as a power law in sparse variants near the ground state (García-García et al., 2 Jul 2026). This places FAF within the modern correspondence between non-Gaussianity, eigenstate typicality, scrambling, and low-dimensional gravity.

6. Adjacent meanings, related interpretations, and limitations

A distinct line of work uses “antiflatness” to describe fluctuations of the entanglement spectrum rather than departure from fermionic Gaussianity. There the basic objects are Rényi-entropy spreads

$\nu_j$6

the associated partial order of antiflat majorization, and derived quantities such as Capacity of Entanglement, linear Rényi spread, and logarithmic antiflatness (Jasser et al., 20 May 2026). This is a spectral theory of reduced density operators; it is not the same notion as covariance-based fermionic antiflatness, although it suggests a possible extension of the term to fermionic entanglement spectra.

Another mathematically related, but terminologically distinct, construction is the parity-graded representation of fermionic wavefunctions summarized by the slogan “fermions = bosons + one.” That work does not define FAF, but it shows that antisymmetric fermionic wavefunctions can be represented as symmetric functions on an enlarged space with one auxiliary odd sector, so that fermionic statistics appear as a two-sheeted cover over bosonic configuration data (Fu, 13 Oct 2025). This suggests a structural reading of antiflatness as the irreducible parity-graded or double-cover aspect of fermionic states, but that interpretation remains separate from the standard covariance-matrix measure.

The covariance-based FAF literature also has clear limitations. The general many-body framework and the disordered-chain analysis focus on pure full-system states and do not define a mixed-state FAF in the body of those works (Sierant et al., 30 May 2025, Falcão et al., 30 Jan 2026). In lattice gauge theory, the reported FAF values depend on the effective encoding and on the availability of a clean Jordan–Wigner map (Santra et al., 8 Oct 2025). In localization studies, conclusions are based on finite-size numerics, and rare long-range resonances remain low-probability at accessible sizes (Falcão et al., 30 Jan 2026). These points suggest that current usage of FAF is best understood as a precise and productive pure-state diagnostic of fermionic non-Gaussianity, with mixed-state resource theory, encoding independence, and broader universality still under development.