LP Witness for Network Nonlocality

• The paper introduces an LP witness for network nonlocality that formulates certification as a feasibility problem using tailored linear constraints.
• The method leverages system-specific constraints based on network topology to bypass the non-convexity inherent in multi-source quantum scenarios.
• Experimental protocols in ring, star, and chain networks demonstrate its applicability, providing bounds on nonlocality cost and multipartite monotones.

A linear programming (LP) witness for network nonlocality is a computational tool that certifies the presence of nonlocal correlations in quantum networks—correlations incompatible with any network-local hidden variable model—by formulating network-locality as a feasibility or optimization problem in linear programming. This approach adapts Bell-type nonlocality certification, originally designed for single-source scenarios, to the complex, non-convex setting of quantum networks involving multiple independent sources. LP witnesses use system-specific linear constraints, which are tailored to the topology and operational protocol of the network, to define a polytope whose infeasibility for observed data certifies network nonlocality. This method circumvents the combinatorial explosion of variables and constraints typical of non-convex and polynomial-based techniques, and it allows systematic device-independent certification across diverse quantum network architectures (Hayes-Shuptar et al., 26 Dec 2025).

1. Network-Locality and the Challenge of Non-convexity

In the Bell scenario, the set of local (i.e., classical) correlations is the convex polytope defined by all probabilistic mixtures of deterministic local response functions, allowing linear programming methods to efficiently determine classical bounds and certify quantum violations. For multi-source quantum networks, the set of network-local correlations is defined by behaviors

$$p(a_1,\dots,a_N)\;=\;\int\!\left[\prod_{m=1}^M d\lambda_m\,p_m(\lambda_m)\right]\prod_{n=1}^N p_n\bigl(a_n|\{\lambda_m : m \in \text{neigh}(n)\}\bigr),$$

with $M$ independent sources and $N$ parties. Imposing statistical independence of source distributions $p_m(\lambda_m)$ renders the network-local set non-convex, making the membership decision problem both nonlinear and computationally hard (Hayes-Shuptar et al., 26 Dec 2025). Traditional nonlinear or polynomial-based characterizations (e.g., through inflation or semidefinite programming hierarchies) become computationally prohibitive as network size increases.

2. Construction of Linear Programming Witnesses

The LP witness formalism introduces an auxiliary distribution $q(o,\lambda)$ over a product space of a reduced outcome subset $\mathcal{O}_S$ and compatible joint “strategy” assignments $\mathcal{S}$ of local variables, such that $q$ is consistent both with observed statistics and the network-locality structure. Five classes of linear constraints define the polytope:

Constraint Classes:

Class	Operational Role	Mathematical Form
1. Distribution Validity	$q$ is a bona fide probability	$q(o,\lambda)\ge0$, $\sum_{o,\lambda}q(o,\lambda)=1$
2. Marginal Agreement	$q$ reproduces observed marginal data	$$\sum_{\lambda}q(o,\lambda) = \frac{p(o)}{\sum_{o'}p(o')}$$
3. Strategy Distribution	$q(\lambda)$ matches network structure	$$\sum_{o}q(o,\lambda) = \frac{\mu(\lambda)}{\sum_{\lambda'}\mu(\lambda')}$$
4. Conditional Independence	Local output only depends on neighbors	Linear equalities on $q(o,\lambda)$ for strategies sharing neighbors
5. Domain Asymmetry	Event-strategy crosschecks	$$q(\mathcal{S}_p^{(1)})-q(\mathcal{S}_p^{(2)})=\Gamma_{o_p}$$

Feasibility of this system for observed data is necessary for the data to admit a network-local model. If the LP is infeasible, nonlocality is certified (Hayes-Shuptar et al., 26 Dec 2025).

3. General Procedure and Network Examples

The general protocol for constructing an LP witness proceeds as follows: (1) Enumerate the set of strategy configurations $\mathcal{S}$ compatible with the network graph and with the chosen subset of outcomes $\mathcal{O}_S$; (2) Define the decision variables $q(o,\lambda)$ on $\mathcal{O}_S\times\mathcal{S}$; (3) Express all five classes of constraints based on observed data and network structure; (4) Formulate and solve the resulting LP. A concrete example is the 6-party, 4-source ring network, where $\mathcal{O}_S$ consists of all events with four single-clicks and two no-clicks, and $\mathcal{S}$ of all strategy assignments consistent with these events. The LP has $240\times 30$ variables and constraints derived explicitly from the observed frequencies and network topology. Solving this feasibility LP for empirical data demonstrates network nonlocality by confirming lack of solutions for certain parameter regimes (e.g., when the beamsplitter transmissivity $t$ is in $(0,0.292)\cup(0.708,1)$) (Hayes-Shuptar et al., 26 Dec 2025).

4. Comparison with Alternative Network Nonlocality Approaches

Traditional inflation methods for network nonlocality certify incompatibility with network-local models by replicating sources and imposing higher-order symmetry constraints, resulting in variable and constraint counts that grow combinatorially with network size and inflation order. Rigidity-based schemes exploit uniqueness or extremality of quantum strategies but generally require strong a priori knowledge of the network and specific measurement assignments. By contrast, the LP witness framework requires enumeration only of feasible outcome-strategy pairs and imposes a fixed system of linear constraints—significantly reducing computational overhead, enhancing scalability, and enabling device-independent certification from real empirical data (Hayes-Shuptar et al., 26 Dec 2025).

5. Graphic Games and Special Cases: From Bell Inequalities to General Networks

A highly structured subclass of LP witnesses emerges in so-called “graphic games,” where the quantum network is represented by a graph $G=(V,E)$ with each vertex corresponding to an EPR pair distributed among “players” (network nodes) (Luo, 2019). Each player’s measurement is specified by subsets $V^{x_i}\subset V$, and local outputs are determined by assignment vectors on these subsets. The classical polytope (local model) is defined by classical consistency conditions on overlapped vertices and local deterministic policies, giving rise to an LP over the network-local polytope. The LP constraints capture normalized probability assignment, positivity, and node-overlap consistency. This construction generalizes standard Bell inequalities to arbitrary networks and allows explicit calculation of classical bounds, as well as analytic characterization of when quantum correlations lead to a gap above the classical optimum (e.g., via Theorem 1 in (Luo, 2019)). Special instances, such as star networks, reduce to collections of CHSH-type games with analytically computable LP-bounds.

6. Wiring, Multipartite Monotones, and Lower Bounds via LP

For multipartite non-signaling boxes, LP-based witnesses quantify network nonlocality emerging from “wirings”: procedures that locally map outputs of several parties to effective inputs and outputs for merged super-parties (Tuziemski et al., 2014). The maximal wireable nonlocality (MWN) is determined via an LP that maximizes a bipartite Bell expression (e.g., CHSH functional) over boxes with a constrained bilocal (or more generally, $k$-local) decomposition. The LP variables encode the distribution and convex weights of extremal local/signal strategies. The optimal LP solution bounds multipartite monotones such as the nonlocality cost and robustness relative to partially local classes. Furthermore, analytic results relate the LP optimal value to weights assigned to signaling vertices in the decomposition, establishing rigorous lower and upper bounds for MWN and associated monotones.

7. Applications and Experimental Protocol

The LP witness provides a practical scheme for experimental certification of network nonlocality. The protocol is as follows: (1) empirically measure the relevant joint distribution $p(a_1,\dots,a_N)$ for a suitably chosen outcome subset; (2) enumerate strategies and set up the $q(o,\lambda)$ variables; (3) formulate the LP with the full system of linear constraints; (4) solve the LP—if infeasible, network nonlocality is certified. This method is widely applicable, as shown in analyses of ring, star, and chain networks. Applications include device-independent entanglement certification, quantum internet architectures, systematic generation of new nonlocal games beyond CHSH, and complexity-theoretic separations in interactive proofs and computational settings (Hayes-Shuptar et al., 26 Dec 2025, Luo, 2019, Tuziemski et al., 2014).