Information causality for almost-quantum correlations
- Fields
- Topics
Problem
Do all almost-quantum correlations satisfy information causality? A bipartite conditional distribution \(p(a,b\mid x,y)\) with finite input and output sets is almost quantum if there are a Hilbert space \(\mathcal H\), a unit vector \(|\psi\rangle\), and projective measurements \(\{P_a^x\}_a\), \(\{Q_b^y\}_b\) on \(\mathcal H\) such that
Each measurement consists of orthogonal projections summing to the identity. The commutation condition in (1) is imposed only on the shared state, not as an operator identity.
Alice receives \(L\) independent uniform bits \(Z_1,\ldots,Z_L\). Bob receives an independent uniform index \(J\in\{1,\ldots,L\}\). They may use shared randomness and almost-quantum boxes independent of these inputs. Alice sends Bob an \(m\)-bit classical message, with no communication from Bob to Alice, and Bob outputs a binary guess \(G\) for \(Z_J\). Information causality requires
where \(I\) is Shannon mutual information in bits. Must (2) hold for every finite protocol of this form and all integers \(L\geq1\) and \(m\geq0\)?
Source
Navascués, Guryanova, Hoban, and Acín define almost-quantum correlations and leave their compatibility with information causality unresolved (Definition 1 and Sections 9–10 of the arXiv version) [NGHA15]. The information-causality task and inequality originate with Pawłowski et al. [PPKSWZ09].
Progress
Reports do not certify correctness or automatically change the problem's status. Progress policy.
Navascués et al. verify several consequences of information causality for almost-quantum correlations and report supporting numerical evidence, while explicitly leaving the full principle unproved [NGHA15].
Das Bhowmik and Banik give a negative answer in preprint version 1, Corollary 1 [BB26]. Their Lemma 1 supplies a finite-dimensional almost-quantum box with \(N\) settings and two outcomes per party. A single use of this box and one classical bit yield, for \(L=kN\) independent data bits, the information score
\begin{equation} \sum_{j=1}^{kN} I(Z_j:G\mid J=j) \geq \frac{k}{17}\left(1-(1-2^{-k})^N\right). \tag{3} \end{equation}For \(k=18\) and \(N=757693\), the right-hand side of (3) exceeds \(1.00000018\), contradicting (2) with \(m=1\). Taking \(N=k2^k\) instead gives the unbounded lower bound \(k(1-e^{-k})/17\). The family uses increasing numbers of settings and data bits. The claim is not an unbounded violation by one fixed finite box.
Comment
The universal information-causality claim is false by the finite counterexample in [BB26]. The Solved designation records this negative resolution and its preprint source. The score includes every trial. Bob always supplies a binary guess, so the violation does not arise from conditioning on successful certificates.
References
- [NGHA15]
- Miguel Navascués, Yelena Guryanova, Matty J. Hoban, and Antonio Acín, “Almost quantum correlations,” Nature Communications 6, 6288 (2015).DOIarXiv
- [PPKSWZ09]
- Marcin Pawłowski, Tomasz Paterek, Dagomir Kaszlikowski, Valerio Scarani, Andreas Winter, and Marek Żukowski, “Information causality as a physical principle,” Nature 461, 1101–1104 (2009).DOIarXiv