Emergent Mind

Rauzy dimension and finite-state dimension

(2406.18383)
Published Jun 26, 2024 in cs.IT , cs.FL , and math.IT

Abstract

In a paper of 1976, Rauzy studied two complexity notions, $\underline{\beta}$ and $\overline{\beta}$, for infinite sequences over a finite alphabet. The function $\underline{\beta}$ is maximum exactly in the Borel normal sequences and $\overline{\beta}$ is minimum exactly in the sequences that, when added to any Borel normal sequence, the result is also Borel normal. Although the definition of $\underline{\beta}$ and $\overline{\beta}$ do not involve finite-state automata, we establish some connections between them and the lower $\underline{\rm dim}$ and upper $\overline{\rm dim}$ finite-state dimension (or other equivalent notions like finite-state compression ratio, aligned-entropy or cumulative log-loss of finite-state predictors). We show tight lower and upper bounds on $\underline{\rm dim}$ and $\overline{\rm dim}$ as functions of $\underline{\beta}$ and $\overline{\beta}$, respectively. In particular this implies that sequences with $\overline{\rm dim}$ zero are exactly the ones that that, when added to any Borel normal sequence, the result is also Borel normal. We also show that the finite-state dimensions $\underline{\rm dim}$ and $\overline{\rm dim}$ are essentially subadditive. We need two technical tools that are of independent interest. One is the family of local finite-state automata, which are automata whose memory consists of the last $k$ read symbols for some fixed integer $k$. We show that compressors based on local finite-state automata are as good as standard finite-state compressors. The other one is a notion of finite-state relational (non-deterministic) compressor, which can compress an input in several ways provided the input can always be recovered from any of its outputs. We show that such compressors cannot compress more than standard (deterministic) finite-state compressors.

We're not able to analyze this paper right now due to high demand.

Please check back later (sorry!).

Generate a summary of this paper on our Pro plan:

We ran into a problem analyzing this paper.

Newsletter

Get summaries of trending comp sci papers delivered straight to your inbox:

Unsubscribe anytime.