Genuine Multipartite Nonlocality Is Fermionic Magic

Technical Summary: Genuine Multipartite Nonlocality Is Fermionic Magic

Problem Statement
The paper addresses a fundamental question in quantum information theory: whether the boundary between classically simulable quantum systems and universal quantum computers also constitutes a boundary for quantum correlations. Specifically, it investigates the limits of genuine multipartite nonlocality (GMNL) in systems of free fermions.

While it is established that free fermions (and Clifford circuits) can be efficiently simulated classically, a single non-Gaussian state (magic state) renders fermionic systems universal. However, it remained an open question whether free fermions could exhibit GMNL. Previous work showed that free fermions are nonlocal (violating CHSH) and that certain topological subgroups cannot violate CHSH or reach the GHZ paradox. The central problem is determining how much GMNL the continuous manifold of fermionic Gaussian states can carry and whether this capacity is bounded by the same constraints that limit their classical simulability.

Methodology
The author employs a combination of analytical proofs, numerical optimization, and experimental verification:

1. Analytical Framework:

   • Formalism: The study utilizes Majorana operators ($c_i$) and fermionic Gaussian states defined by their covariance matrix $\Gamma$. Measurements are modeled as local bilinear observables $O = i(u \cdot c)(v \cdot c)$, representing occupation numbers after local Gaussian rotations.
   • Inequalities: The author analyzes the Mermin polynomial ($M_n$) and the Svetlichny polynomial ($S_n$), which serve as witnesses for genuine multipartite entanglement and nonlocality, respectively.
   • Proof Techniques: The proofs rely on the multilinearity of the Pfaffian (Wick's theorem), the properties of fermion parity, and linear programming over pair weights. A key lemma establishes a "vacuum-full" inequality bounding the product of occupation probabilities for complementary patterns.

2. Numerical Verification:

   • The author performs direct maximization over Gaussian states (pure and mixed) using L-BFGS-B optimization on covariance matrices.
   • They employ interval branch-and-bound methods to rigorously certify upper bounds for settings that do not strictly share a direction, addressing the "general plane" case.

3. Experimental Implementation:

   • Hardware: Experiments were conducted on three IBM Quantum processors: ibm fez (Heron r2), ibm marrakesh (Heron r2), and ibm berlin (Nighthawk r1).
   • Encoding: Logical qubits were encoded in a dual-rail scheme ($|01\rangle, |10\rangle$), where logical Pauli operators correspond to Majorana bilinears.
   • Circuit: The protocol uses a constant-depth matchgate circuit with a single non-Gaussian element: a controlled-phase gate $CP(\phi)$. By varying $\phi$, the system transitions from a Gaussian (simulable) regime to a non-Gaussian (magic) regime.
   • Metrics: The study measures the Mermin value $M$ and the joint success probability of multiple independent "cells" (6-qubit blocks) to test monogamy constraints.

Key Contributions and Results

1. Theoretical Bound on Gaussian States:

   • Theorem 1: For any fermionic Gaussian state (pure or mixed) and local bilinear settings where each party's two measurement planes share a direction (or commute), the three-party Mermin correlator is bounded by $|\langle Q_A Q_B Q_C \rangle| \le 2\sqrt{2}$.
   • Implication: This bound equals the biseparable limit of the Mermin inequality. Consequently, no fermionic Gaussian state can certify genuine tripartite entanglement via the Mermin inequality, nor can it violate the Svetlichny inequality (bound 4).
   • Generalization (Theorem 3): For $n$ parties, the bound is $|\langle \prod Q_j \rangle| \le 2^{n/2}$. Free fermions violate the Mermin polynomial by exactly $\sqrt{2}$ (the Tsirelson bound of a single pair) but never exceed the biseparable bound for $n \ge 4$. Thus, free fermions are never genuinely multipartite nonlocal, regardless of the number of parties.

2. Identification of the Resource:

   • The paper establishes that genuine multipartite nonlocality is the exact resource required to leave the class of simulable fermionic systems.
   • Theorem 4: In a device-independent setting, a Mermin value exceeding $2\sqrt{2}$ under fermionic linear-optical readout certifies the presence of fermionic magic (non-Gaussianity). This provides a direct link: GMNL is the signature of universality in fermionic computation.

3. Experimental Certification:

   • The author successfully demonstrated violations of the Gaussian bound on hardware.
   • On ibm fez, a single cell achieved a raw Mermin value of $M_{raw} = 3.305 \pm 0.022$ at $\phi = \pi$, exceeding the $2\sqrt{2} \approx 2.828$ bound by 22 standard deviations.
   • Monogamy Test: Using 48 qubits (8 independent cells) on ibm fez, the joint success probability was $0.503 \pm 0.002$, significantly exceeding the Gaussian bound of $\beta^8 \approx 0.282$. This confirms that the nonlocality is not merely a collection of pairwise correlations but a genuine multipartite resource enabled by the non-Gaussian gate.
   • The results held across different architectures (Heron and Nighthawk) and were robust against noise, with visibility $A \approx 0.83-0.87$.

Significance and Claims
The paper claims to resolve the open question of whether the computational boundary of free fermions (Gaussian states) coincides with a correlation boundary. The findings are:

• Parity as the Obstruction: The inability of free fermions to exhibit GMNL is fundamentally due to fermion parity. The parity conservation inherent in quadratic Hamiltonians restricts the correlations to the biseparable regime.
• Magic as GMNL: Genuine multipartite nonlocality is identified as the specific resource that distinguishes universal fermionic computation from simulable free-fermion dynamics. Just as magic states enable universal Clifford computation, fermionic magic (non-Gaussianity) enables GMNL.
• Device-Independent Certification: The work provides a method to certify fermionic magic and the presence of interactions beyond mean-field theory (e.g., in BCS superconductors or Kitaev chains) using only the statistics of occupation number measurements, without requiring full state tomography.
• Hardware Validation: The experimental results on IBM processors demonstrate that current quantum hardware can generate and certify this specific form of nonlocality, validating the theoretical bounds and the monogamy of the resource.

The author remains modest regarding the "general plane" case (where measurement planes do not share a direction), noting that while numerical searches return $2\sqrt{2}$, a rigorous analytical proof for the general case (certified constant 3.26) relies on branch-and-bound methods and a second-order argument at the maximizing corner, which is still being refined. However, for the physically relevant case of encoded qubits (dual-rail or Majorana tetrons), the hypothesis of shared directions is automatically satisfied, making the $2\sqrt{2}$ bound rigorous and the certification of magic definitive.