Quantum Bayesian networks with application to games displaying Parrondo's paradox

TL;DR

This paper introduces a quantum extension of Bayesian networks, reformulating them with linear maps, and applies this framework to analyze Parrondo's paradox in classical and quantum game scenarios, demonstrating bounds and quantum walk implementations.

Contribution

It presents a novel quantum formulation of Bayesian networks and applies it to analyze and bound Parrondo's paradox in classical and quantum games.

Findings

Quantum Bayesian networks can model quantum and mixed classical-quantum systems.

Bounds for Parrondo's paradox discrepancies are established.

Quantum walks can realize the bounds of the paradox.

Abstract

Bayesian networks and their accompanying graphical models are widely used for prediction and analysis across many disciplines. We will reformulate these in terms of linear maps. This reformulation will suggest a natural extension, which we will show is equivalent to standard textbook quantum mechanics. Therefore, this extension will be termed "quantum". However, the term "quantum" should not be taken to imply this extension is necessarily only of utility in situations traditionally thought of as in the domain of quantum mechanics. In principle, it may be employed in any modeling situation, say forecasting the weather or the stock market--it is up to experiment to determine if this extension is useful in practice. Even restricting to the domain of quantum mechanics, with this new formulation the advantages of Bayesian networks can be maintained for models incorporating quantum and mixed classical-quantum behavior. The use of these will be illustrated by various basic examples. Parrondo's paradox refers to the situation where two, multi-round games with a fixed winning criteria, both with probability greater than one-half for one player to win, are combined. Using a possibly biased coin to determine the rule to employ for each round, paradoxically, the previously losing player now wins the combined game with probability greater than one-half. Using the extended Bayesian networks, we will formulate and analyze classical observed, classical hidden, and quantum versions of a game that displays this paradox, finding bounds for the discrepancy from naive expectations for the occurrence of the paradox. A quantum paradox inspired by Parrondo's paradox will also be analyzed. We will prove a bound for the discrepancy from naive expectations for this paradox as well. Games involving quantum walks that achieve this bound will be presented.