Exact Synchronization for Finite-State Sources

We analyze how an observer synchronizes to the internal state of a finite-state information source, using the epsilon-machine causal representation. Here, we treat the case of exact synchronization, when it is possible for the observer to synchronize completely after a finite number of observations. The more difficult case of strictly asymptotic synchronization is treated in a sequel. In both cases, we find that an observer, on average, will synchronize to the source state exponentially fast and that, as a result, the average accuracy in an observer’s predictions of the source output approaches its optimal level exponentially fast as well. Additionally, we show here how to analytically calculate the synchronization rate for exact epsilon-machines and provide an efficient polynomial-time algorithm to test epsilon-machines for exactness.

💡 Research Summary

The paper investigates how an observer can synchronize to the internal state of a finite‑state information source when the source is represented by its ε‑machine, the minimal causal model that partitions past histories into equivalence classes. The authors focus on the case of exact synchronization, defined as the existence of a finite‑length observation (a synchronizing word) that uniquely determines the current hidden state. In this situation the observer, after observing that word, knows the state with certainty and can thereafter make optimal predictions.

Two central theoretical results are established for exact ε‑machines. First, the expected length required for synchronization decays exponentially with the length of the observation. By constructing a synchronization probability matrix that captures the evolution of the set of possible states under successive symbols, the authors show that its spectral radius λ satisfies 0 < λ < 1. The probability that the observer has not yet synchronized after L symbols is bounded by λ^L, implying an average synchronization time on the order of log(1/ε). Second, the conditional entropy of the next output given the observer’s belief, H