Shannon Weights for binary dynamical recurrent sources of zero entropy

A probabilistic source is defined as the set of infinite words (over a given denumerable alphabet) endowed with a probability $\mu$. The paper deals with general binary sources where the distribution of any symbol (0 or 1) may depend on an unbounded part of the previous history. The paper studies Shannon weights: whereas the classical Shannon entropy ${\cal E}_{\mu}$ is the average amount of information brought by one symbol of the emitted word, the Shannon weight sequence deals with the average amount of information $m_{\mu}(n)$ that is brought by the emitted prefix of length $n$. For a source with a non zero entropy, the estimate $m_{\mu}(n)\sim{\cal E}_{\mu} \cdot n$ thus holds. The paper considers the model of dynamical sources, where a source word isemitted as an encoded trajectory of a dynamical system of the unit interval, when endowed with probability $\mu$. It focus on sources with zero entropy and gives explicit constructions for sources whose Shannon weight sequence satisfies $m_{\mu}(n)=o(n)$, with a prescribed behaviour. In this case, sources with zero entropy lead to dynamical systems built on maps with an indifferent fixed point. This class notably contains the celebrated Farey source, which presents well-known intermittency phenomena. Methods are based on analytic combinatorics and generating functions, and they are enlarged, in the present dynamical case, with dynamical systems tools (mainly transfer operators).