CIF: Small: Fundamental limits in ambiguous communication

Noise, either arising from the environment or due to inherent randomness, is an unavoidable feature of information channels, which are used to transmit data between two or more parties. So-called zero-error capacities of such channels are the optimal rates at which information can be transmitted via the channel with a zero probability of error when multiple uses of the channel are allowed.

While in practice the effects of noise change with time, the capacity theories available until now focus mostly on the simplified picture where the successive uses of the channel are independent. This project aims at establishing the theoretical foundations that would allow the estimation of zero-error capacities and closely related parameters in the state-dependent regime, where each individual use of the channel depends on its previous uses and the current state of the surrounding noise.

It is using the developed framework to study a special case of information channels, which are viewed as strategies for games -- called two-prover games -- in which two players are tested by a verifier, through a question-answer process, for the joint possession of a certain piece of knowledge. Such games have until now been considered only in the memoryless situation where the successive rounds of the game are played independently from each other.

The project is studying the winning rates of two-prover games, where the players and the verifier possess memory and the rules of the game -- the tested knowledge -- may evolve with time.

In the setting of information channels, the project is defining and studying the zero-error capacity of a channel with memory, which covers as a special case the usual zero-error capacity of a memoryless channel as a one-shot parameter. This is being achieved through the introduction of the confusability graph of a state-dependent channel, while estimations and bounds for this parameter are obtained by developing a measurable version of the well-known Lovász number of a finite graph.

In the setting of two prover games, the project is developing no-signaling correlations between players with memory, including useful subclasses such as the class of quantum correlations, as a far-reaching generalization of no-signaling correlations between players of games with finite question-answer sets. This approach allows the consideration of value-separation questions and the identification of the players' asymptotic behavior in a general framework, allowing the application of functional analytic and operator theoretic tools.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.