Probabilistic Foundations of Contextuality

Abstract: Contextuality is usually defined as absence of a joint distribution for a set of measurements (random variables) with known joint distributions of some of its subsets. However, if these subsets of measurements are not disjoint, contextuality is mathematically impossible even if one generally allows (as one must) for random variables not to be jointly distributed. To avoid contradictions one has to adopt the Contextuality-by-Default approach: measurements made in different contexts are always distinct and stochastically unrelated to each other. Contextuality is reformulated then in terms of the (im)possibility of imposing on all the measurements in a system a joint distribution of a particular kind: such that any measurements of one and the same property made in different contexts satisfy a specified property, $\mathcal{C}$. In the traditional analysis of contextuality $\mathcal{C}$ means "are equal to each other with probability 1". However, if the system of measurements violates the "no-disturbance principle", due to signaling or experimental biases, then the meaning of $\mathcal{C}$ has to be generalized, and the proposed generalization is "are equal to each other with maximal possible probability" (applied to any set of measurements of one and the same property). This approach is illustrated on arbitrary systems of binary measurements, including most of quantum systems of traditional interest in contextuality studies (irrespective of whether "no-disturbance" principle holds in them).