Boxworld: A Generalized Probabilistic Theory

Updated 12 April 2026

Boxworld theory is a generalized probabilistic framework that realizes maximal nonlocal but nonsignalling correlations, exemplified by PR boxes.
It rigorously tests the foundations of contextuality and information processing by analyzing state-space geometry and operational constraints.
Its unique algebraic and geometric structures challenge classical and quantum models, revealing anomalies in communication complexity and thermodynamics.

Boxworld theory is a generalized probabilistic theory (GPT) that embodies the ultimate nonlocal but nonsignalling correlations constrained only by basic operational consistency and the no-signalling principle. Notably, boxworld realizes the extremal points of the nonsignalling polytope, such as Popescu–Rohrlich (PR) boxes, saturating Bell-type inequalities beyond quantum theory while respecting nonsignalling. Its operational content plays a central role in the study of fundamental limits of correlations, contextuality, information processing, and the structure of physical theories beyond quantum mechanics.

1. Structural Foundations of Boxworld

Boxworld is defined at both the single-system and multipartite levels by finite sets of preparations, measurements, and outcome statistics:

Elementary system (gbit): The minimal system, or "gbit," has binary measurements $x \in \{0,1\}$ and binary outcomes $a \in \{0,1\}$ . The set of states comprises all conditional probability distributions $P(a|x)$ with normalization and positivity. The pure states are deterministic response functions $a = \alpha x \oplus \beta$ with $\alpha, \beta \in \{0,1\}$ , yielding a state-space polytope that is a square (hypercube in higher dimensions) (Plávala, 2021, Janotta et al., 2013, Brunner et al., 2014).
Effects: Effects are affine linear functionals $e: \Omega \to [0,1]$ . In the no-restriction hypothesis variant, the allowed effect set is the dual polytope (diamond/octahedron for the gbit) (Janotta et al., 2013).
No-signalling composites: Multipartite states are conditional distributions $P(a_1, ..., a_N|x_1, ..., x_N)$ subjected to linear no-signalling constraints ensuring that measurement choices on one subset of parties cannot influence marginal statistics elsewhere (Eftaxias et al., 2022).

The maximal tensor product supplies all such joint distributions consistent with local tomography and no-signalling.

2. COPE Matrices, Contextuality, and Nonclassicality

Boxworld’s operational data is fully encoded in its Conditional Outcome Probabilities of Events (COPE) matrix $C^{\text{BW}}$ , whose structure determines whether a noncontextual ontological model may exist (Shahandeh et al., 10 Dec 2025):

COPE matrix for minimal boxworld:

$C^{\rm BW} = \begin{pmatrix} 1 & 0 & 0 & 1 \ 0 & 1 & 1 & 0 \ 1 & 0 & 1 & 0 \ 0 & 1 & 0 & 1 \end{pmatrix}$

representing conditional probabilities $p(k|P_i,M_j)$ for four preparations and two binary measurements.

ENMF Criterion: An operational theory admits a noncontextual ontological model if and only if its COPE matrix admits an equirank nonnegative matrix factorization (ENMF): $a \in \{0,1\}$ 0 with $a \in \{0,1\}$ 1 and $a \in \{0,1\}$ 2. For boxworld, $a \in \{0,1\}$ 3, but no factorization with $a \in \{0,1\}$ 4 and $a \in \{0,1\}$ 5 exists (Shahandeh et al., 10 Dec 2025).
Geometric interpretation: The columns of $a \in \{0,1\}$ 6 are points forming a square in a 2-dimensional subspace of the standard simplex. There is no way to express these as mixtures of three ontic extremal states without leaving the convex hull—thus, no ENMF exists.

Consequence: Boxworld is operationally and ontologically contextual; its correlations cannot be replicated by a classical simplex model without inaccessible (contextual) degrees of freedom (Shahandeh et al., 10 Dec 2025).

3. Nonlocality, PR Boxes, and Supra-Quantum Correlations

Boxworld realizes all non-signalling boxes, including those that go beyond quantum Tsirelson bounds:

PR-box correlations: For two parties with binary measurements and outcomes, boxworld admits extremal states with

$a \in \{0,1\}$ 7

saturating the algebraic (no-signalling) CHSH bound $a \in \{0,1\}$ 8, exceeding the quantum maximum $a \in \{0,1\}$ 9 (Plávala, 2021, Janotta et al., 2013, Gutt et al., 2016).

Geometry and tensor products: The maximally entangled boxworld states are extremal in the maximal tensor product, and the effect cone is minimal—only separable effects are present in the standard version. This absence of entangled measurements is structurally responsible for the absence of entanglement swapping (Janotta, 2012, Eftaxias et al., 2022).
Polytope structure: The state and effect spaces are polygons (hypercubes and crosspolytopes), lacking the additional self-duality and spherical symmetry of quantum state spaces. This allows boxworld to realize vertices of the no-signalling polytope that quantum theory cannot approach (Janotta et al., 2013, Plávala, 2021).

4. Measurements, Effect Structure, and Multi-Box Phenomena

Boxworld exhibits a rich but constrained measurement structure, with significant implications for information processing (Eftaxias et al., 2022):

Wirings vs. non-wirings: All bipartite measurements are implementable by local wirings (sequential local measurements and classical feed-forward); genuine non-wiring effects arise in the tripartite and larger settings. For $P(a|x)$ 0 there are $P(a|x)$ 1 non-wiring extremal effects, in $P(a|x)$ 2 symmetry classes (Eftaxias et al., 2022).
Enumerative results: The enumeration of extremal effects reflects a tradeoff: the enlarged state polytope due to maximal no-signalling is accompanied by a reduced set of admissible effects.
Information-theoretic utility: Non-wiring joint measurements allow for perfect discrimination between certain states that wirings cannot distinguish, enable nonlocality distillation, and yield a boxworld analogue of nonlocality without entanglement. For instance, certain tripartite boxworld states are perfectly distinguished only by non-wiring measurements, and 3-box "nonlocality without entanglement" perfectly discriminates 8 orthogonal product states via a non-wiring global measurement (Eftaxias et al., 2022).

5. Information, Dimension Mismatch, and Thermodynamics

Boxworld features a pronounced dimension mismatch between the maximum number of perfectly distinguishable states in a single measurement (measurement dimension $P(a|x)$ 3) and the (potentially exponentially larger) number of pairwise distinguishable states (information dimension $P(a|x)$ 4):

Dimension mismatch: For a gbit, $P(a|x)$ 5 (binary output measurements), $P(a|x)$ 6 (all pure states pairwise distinguishable by some measurement) (Brunner et al., 2014). Generalizing to a $P(a|x)$ 7-hypercube bit, $P(a|x)$ 8, $P(a|x)$ 9. This mismatch can be made arbitrarily large (Brunner et al., 2014).
Collapse of communication complexity: Communication complexity collapses as Alice can encode an exponential amount of information in a hypercube bit and Bob retrieves any single bit by choosing the measurement—this is operationally equivalent to sharing PR boxes (Brunner et al., 2014).
Violation of information causality: The mutual-information bound $a = \alpha x \oplus \beta$ 0 is violated in boxworld, where $a = \alpha x \oplus \beta$ 1 can reach $a = \alpha x \oplus \beta$ 2 for an $a = \alpha x \oplus \beta$ 3-bit hypercube, exceeding the limit for classical and quantum systems (Brunner et al., 2014).
Thermodynamic anomalies: Since the minimal erasure cost is set by $a = \alpha x \oplus \beta$ 4, but $a = \alpha x \oplus \beta$ 5 can be arbitrarily larger, one can in principle erase an exponential amount of information while expending only $a = \alpha x \oplus \beta$ 6, challenging Landauer’s principle and the second law unless further restrictions are imposed (Brunner et al., 2014).

6. Causality, Process Theory, and Contextuality

Boxworld serves as an extremal testbed for indefinite causal order and contextuality:

Indefinite causal order: In higher-order generalizations (“process tensor” approach), unconstrained boxworld trivializes to the full polytope of bipartite conditional distributions, including perfect two-way signalling. Imposing physical principles such as Nonsignaling Preservation (NSP) and No Signaling Without System Exchange (NSWSE) restricts the set of allowed process tensors to a convex polytope that encompasses all quantum process matrices but reaches strictly higher bounds on causal inequalities (GYNI, LGYNI, OCB) (Bavaresco et al., 2024).
Contextuality and ENMF: The failure of ENMF for the boxworld COPE matrix provides a direct algebraic certificate of strong contextuality (Shahandeh et al., 10 Dec 2025).
Bohrification/topos-theoretic formulation: Non-signalling boxworld states are precisely probability valuations on an internal frame (Heyting algebra) in a Kripke topos of measurement contexts. This unifies classical, quantum, and boxworld theories as instances of the same categorical construction, with boxworld occupying a “super-quantum” regime where the logical structure is strictly finer than in quantum theory (Gutt et al., 2016).

7. Reversible Dynamics and Generalizations

Boxworld is highly rigid in reversible transformations:

Reversible maps: Apart from subsystem permutations and local relabellings of measurement settings and outcomes, there are no nontrivial reversible dynamics, in contrast to the continuous unitary evolutions in quantum mechanics (Al-Safi et al., 2013). This reflects the polyhedral geometry and lack of a rich symmetry group in the state space and effect algebra.
Generalizations: Variants of boxworld restore entangled measurements by constructing intermediate tensor products (weak self-duality), or interpolate continuously between classical and boxworld cases, revealing families of theories with continuously variable CHSH-type violations between the classical ( $a = \alpha x \oplus \beta$ 7), quantum ( $a = \alpha x \oplus \beta$ 8), and PR-box ( $a = \alpha x \oplus \beta$ 9) bounds (Janotta, 2012).

Boxworld is thus the canonical model for maximal nonsignalling nonclassicality—its operational, geometric, and algebraic properties cast sharp light on distinctions among classical, quantum, and post-quantum theories, the role of contextuality, and the ultimate constraints on information processing in physical reality (Plávala, 2021, Janotta et al., 2013, Shahandeh et al., 10 Dec 2025, Brunner et al., 2014, Eftaxias et al., 2022, Bavaresco et al., 2024, Gutt et al., 2016, Janotta, 2012, Al-Safi et al., 2013).