Temporal sequences [in cellular automata]

So far we have considered possible sequences of cells that can occur at a particular step in the evolution of a cellular automaton. But one can also consider sequences formed from the color of a particular cell on a succession of steps. For class 1 and 2 cellular automata, there are typically only a limited number of possible sequences of any length allowed. And when the length is large, the sequences are almost always either just uniform or repetitive. For class 3 cellular automata, however, the number of sequences of length n typically grows rapidly with n. For additive rules such as 60 and 90, and for partially additive rules such as 30 and 45, any possible sequence can occur if an appropriate initial condition is given. For rule 18, it appears that any sequence can occur that never contains more than one adjacent black cell. I know of no general characterization of temporal sequences analogous to the finite automaton one used for spatial sequences above. However, if one defines the entropy or dimension h_{t} for temporal sequences by analogy with the definition for spatial sequences above, then it follows for example that h_{t} ≤ 2 λ h_{x}, where λ is the maximum rate at which changes grow in the cellular automaton. The origin of this inequality is indicated in the picture below. The basic idea is that the size of the region that can affect a given cell in the course of t steps is 2 λ t. But for large sizes x the total number of possible configurations of this region is k^{hx x}. (Inequalities between entropies and Lyapunov exponents are also common in dynamical systems based on numbers, but are more difficult to derive.) Note that in effect, h_{x} gives the information content of spatial sequences in units of bits per unit distance, while h_{t} gives the corresponding quantity for temporal sequences in units of bits per unit time. (One can also define directional entropies based on sequences at different slopes; the values of such entropies tend to change discontinuously when the slope crosses λ.)

Different classes of cellular automata show characteristically different entropy values. Class 1 has h_{x} = 0 and h_{t} = 0. Class 2 has h_{x} ≠ 0 but h_{t} = 0. Class 3 has h_{x} ≠ 0 and h_{t} ≠ 0. Class 4 tends to show fluctuations which prevent definite values of h_{x} and h_{t} from being found.