On the amortized complexity of the odometer

Michel Rigo

Université de Liège, Belgium

Coauthor(s): Valérie Berthé, Christiane Frougny, and Jacques Sakarovitch

As usual, the odometer function $ s$ maps the representation of an integer $ n$ onto the representation of its successor $ n+1$ in a given numeration system. In a general setting, i.e., for abstract numeration systems, the odometer maps a word from a rational language $ L$ onto the next word $ s(w)$ in $ L$ assuming that $ L$ has been genealogically ordered. The number $ k(w)$ of operations for computing $ s(w)$ is defined as $ \vert s(w)\vert-\vert p(w,s(w))\vert$ where $ p(x,y)$ denotes the longuest common prefix of $ x$ and $ y$. We estimate the sum of the $ k(w)$'s for all the words of length $ n$ in a given rational language $ L$ and we compare this quantity with the number of words of length $ n$ in $ L$. If we assume that the automaton accepting $ L$ has a primitive adjacency matrix having $ \beta$ as Perron eigenvalue, this ratio tends to $ \beta/(\beta-1)$ if $ n$ tends to infinity. In particular, this results holds for beta-numeration systems built on a Parry number.