Does ignorance of the whole imply ignorance of the parts? Large violations of noncontextuality in quantum theory

A central question in our understanding of the physical world is how our knowledge of the whole relates to our knowledge of the individual parts. One aspect of this question is the following: to what extent does ignorance about a whole preclude knowledge of at least one of its parts? Relying purely on classical intuition, one would certainly be inclined to conjecture that a strong ignorance of the whole cannot come without significant ignorance of at least one of its parts. Indeed, we show that this reasoning holds in any noncontextual (NC) hidden-variable model (HV). Curiously, however, such a conjecture is false in quantum theory: we provide an explicit example where a large ignorance about the whole can coexist with an almost perfect knowledge of each of its parts. More specifically, we provide a simple information-theoretic inequality satisfied in any NC HV, but which can be arbitrarily violated by quantum mechanics.

Errata:
We first describe an error in our Letter. To derive Equation (22) in the Supplemental Material of the Letter, as well as the expression at the top of page 5, which are key in proving Theorem II.2, the sum over all ontic states is interchanged with the maximum over the dits or, respectively. This requires a restriction to deterministic ontological models, and can be justified by coarse-graining ontic states that assign the same outcome. As we will show, minor modifications to the proof of Theorem II.2 extend its validity to general, nondeterministic ontological models.