next up previous
Next: References Up: Careful chaos: taming random Previous: Filter foibles


Other sequences

There is nothing about the desired output of the chaos game that requires a random sequence. The property of ``filling out'' a fractal $ \mathcal{F}$ to some arbitrary resolution $ N\times N$ simply means that all addresses in $ \mathcal{F}$ to some depth $ k$ ($ k$ is a function of $ N$ and the contractions defining $ \mathcal{F}$) must be present in the sequence. A random sequence (with appropriate probabilities for each digit) may do a reasonable job, but will other sequences work?

It turns out that for ``nice'' fractals (all addresses are present) there is a best sequence: the shortest sequence that will include all substrings of length $ n$, considerably shorter than the random sequence that is guaranteed to contain all such substrings. For addresses of length $ k$ over $ j$ digits, this sequence has length $ j^k+k-1$. In A Probabilist Looks at the Chaos Game Goodman mentions an algorithm (Rees and Good) that will generate a best sequence over all digits.

What happens for our fractal-like object in Section 3? For example, how long is the best sequence to generate all $ 15$ strings of length $ 2$ that do not contain ``12''? Certainly you can do no better than length $ 16$ (the $ 15$th $ 2$-string ends at position $ 16$), but perhaps you cannot do that well. What about for longer strings?

Further work might extend the Fractal Sequence applet to use a best sequence driver, perhaps composed with substring filters. Is it possible to create optimal sequences that contain all substrings of length $ k$ but not ``12''?


next up previous
Next: References Up: Careful chaos: taming random Previous: Filter foibles
Danny Heap 2001-05-18