Generalized Probabilistic Theories (GPTs)

• Generalized Probabilistic Theories are defined through convex sets of states and effects, operationally linking preparation and measurement via probability laws.
• GPTs model composite systems using tensor products and local tomography, distinguishing quantum correlations from supra-quantum nonlocal effects.
• By incorporating principles like purification and self-duality, GPTs reconstruct key features of quantum theory and guide experimental tests of nonclassicality.

Generalized Probabilistic Theories (GPTs) offer an operational framework for describing a wide landscape of possible physical theories, including classical probability theory, quantum theory, and various nonclassical and supra-quantum alternatives. GPTs underpin modern foundational research in quantum mechanics, information processing, contextuality, and experimental tests of nonclassicality, forming an indispensable bridge between abstract probabilistic structure and laboratory observables.

1. Operational and Mathematical Framework

GPTs articulate physical theories in terms of convex geometry and operational constructs. A GPT is specified by a triple $(S, E, u)$, where $S$ is the set of normalized states (a compact convex subset of a real vector space $V$), $E$ is the set of effects (a convex subset of the dual space $V^*$), and $u \in E$ is the unit effect satisfying $u(s)=1$ for all $s \in S$ (Janotta et al., 2014, Plávala, 2021). States arise as equivalence classes of preparation procedures, effects as equivalence classes of one-bit measurements. The essential probabilistic law in GPTs is

$$p(e|s) = e(s),\quad\text{with}\quad 0 \leq e(s) \leq 1.$$

States and effects, when extended to subnormalized entities, define convex cones $V_+$ and $V_+^*$ respectively. Measurements are collections of effects summing to the unit effect: $\sum_{i} e_i = u$. Transformations are positive, normalization-preserving linear maps. Under the no-restriction hypothesis, all probability-valued functionals on $S$ constitute physically allowed effects; relaxing this postulate yields broader families of theories (Janotta et al., 2013).

2. Composition, Tomography, and Tensor Structures

System composition in GPTs is governed by cone tensor products, subject to operational constraints such as local tomography and no-signaling. The minimal tensor product, $V_+^A \otimes_{\min} V_+^B$, comprises all classical mixtures of product states; the maximal tensor product, $V_+^A \otimes_{\max} V_+^B$, is defined via positivity on all product effects (Janotta et al., 2014). Local tomography requires that joint states are determined by statistics of local measurements and demands $V^{AB} \simeq V^A \otimes V^B$ as vector spaces.

Some GPTs, notably quantum theory, satisfy local tomography; others, such as various higher-dimensional or boxworld constructions, may violate it (Kleinmann et al., 2012). Interpretation and analysis of experimental data within GPTs may necessitate theory-agnostic tomography, extracting accessible fragments and their "shadows" to bound the underlying state and effect spaces (Mazurek et al., 2017, Schmid et al., 2024, Selby et al., 2021).

3. Axiomatic Reconstructions, Self-Duality, and Quantum Uniqueness

Quantum theory emerges as a distinguished GPT under operational and information-theoretic constraints (Janotta et al., 2014). Canonical axioms include:

• Purification: Every mixed state admits a minimal pure-state extension in a composite system, unique up to reversible transformations.
• Local tomography: Joint states are fully characterized by local statistics.
• Continuous reversibility: Pure states form a transitive manifold under reversible transformations.
• Absence of higher-order interference: Multi-slit interference terms vanish beyond the pairwise level.

When these are imposed, the only finite-dimensional GPT satisfying them is complex Hilbert-space quantum theory (Janotta et al., 2014, Alegre, 14 Dec 2025). Quantum theory is further characterized by strong self-duality—state and effect cones coincide up to an inner product—spectrality, purification, and continuous reversible dynamics. In self-dual GPTs, techniques such as convex optimization fully characterize operational tasks (e.g., optimal state discrimination) in terms of the geometry of state and effect spaces (Bae et al., 2017).

4. Nonclassicality, Contextuality, and Resource Structures

Nonclassical features in GPTs—including contextuality, entanglement, and nonlocality—derive from the structure of state and effect spaces and the interplay between them. Classical (simplicial) GPTs, where pure states and effects form dual bases, admit noncontextual ontological models iff the no-restriction hypothesis is satisfied and only a single nonrefinable measurement exists (Shahandeh, 2019). Adding a single "resourceful" effect or state can render a GPT contextual (Shahandeh, 2019).

Generalized contextuality is systematically organized via resource theories, with monotones such as classical excess ($E_\mathrm{classical}$) quantifying the minimal error in embedding a GPT into infinite classical systems, and parity-oblivious multiplexing success probability ($P_\mathrm{success}$) bounding operational advantages (Catani et al., 2024).

In experimental practice, accessible GPT fragments are extracted from self-consistent tomography, and cone-equivalent transformations preserve classical or nonclassical explanations, regardless of detector inefficiencies or incompatibility (Selby et al., 2021). Relative tomographic completeness ensures robustness of contextuality proofs, even with coarse-grained or emergent experimental subsystems (Schmid et al., 2024).

5. Incompatibility, Uncertainty, and Entropic Relations

Measurement incompatibility—failure of joint measurability—is prevalent in non-simplicial GPTs and quantified using tensor norms, generalized spectrahedra, and crossnorm ratios on Banach spaces (Bluhm et al., 2020). The compatibility degree $\gamma(g;V,V^+)$ can be captured by ratios of injective and projective tensor norms and bounded by 1-summing constants; for qubits $(d=2)$, $\gamma(g)=1/\sqrt{g}$ for $g\leq 3$ (Bluhm et al., 2020, Plávala, 2021).

Universal entropic uncertainty relations manifest in transitive, self-dual GPTs. Preparation and measurement uncertainty bounds are linked quantitatively, often generalized from Landau–Pollak-type relations. If preparation uncertainty exists for two observables, measurement uncertainty follows, with corresponding inequalities extending quantum results to polygonal and other nonquantum GPTs (Takakura et al., 2020, Takakura et al., 2020).

6. Quantum vs. Supra-Quantum Nonlocality, Limitations, and Experimental Tests

GPTs encompass both quantumly achievable and supra-quantum nonlocal correlations. Most notably, boxworld admits maximal CHSH violations $S=4$, surpassing Tsirelson's bound $2\sqrt{2}$ (Janotta et al., 2014, Dmello et al., 2024), while quantum theory is limited by its self-dual structure and measurement completeness. GPTs featuring restricted effect spaces, self-dualization, or composite system designs can interpolate between classical, quantum, and superquantum behaviors (Janotta et al., 2013). Entanglement swapping and repeated CHSH games expose the limits and the possibility of unbounded nonlocality in GPTs, with explicit constructions (e.g., oblate-stabilizer theory) showcasing CHSH=$4$ after arbitrary rounds (Dmello et al., 2024).

Experimental schemes perform self-consistent tomography without presupposing Hilbert space structure, bounding the set of compatible GPTs with physical data. Deviations from quantum predictions are constrained at the percent level for polarization experiments, both for nonlocality (CHSH tests) and for universal noncontextuality (Mazurek et al., 2017).

7. Foundational Extensions and Recent Directions

GPT research penetrates foundational questions such as the emergence and necessity of complex Hilbert spaces (Shaiia, 2021), reconstructions from operational axioms (Alegre, 14 Dec 2025), and the role of causality and process structures in multipartite settings (Chen et al., 2024). The GPT formalism encompasses the convex-geometric core of quantum theory and its probabilistic laws, solving classical “measurement problems” while providing formats for extensions, such as the study of morphophoric measurements and Protourgleichung-type affine deformations of the Born rule (Weiss, 2023).

The universality of the GPT formalism and its flexibility in modelling both standard and exotic information-processing tasks (nonlocality distillation, discrimination, resource quantification) affects diverse areas of quantum foundations, quantum information, and experimental testability. Open problems include the classification and operational role of pure morphophoric designs beyond complex quantum theory, infinite-dimensional extensions, and the search for normative principles uniquely selecting the Born rule and Hilbert-space structure among all GPTs (Weiss, 2023, Catani et al., 2024, Shaiia, 2021).