Backtracking [in cellular automata]

If one wants to find out which of the 2^^{n} possible initial conditions of width n evolve to yield a specific column of colors in a system like an elementary cellular automaton one can usually do somewhat better than just testing all possibilities. The picture below illustrates a typical approach, applied to 3 steps of rule 30. The idea is successively to look at each numbered cell, and to make a tree of possibilities representing what happens if one tries to fill in each possible color for each cell. A branch in the resulting tree continues only if it corresponds to a configuration of cell colors whose evolution is consistent with the specified column of colors.

The picture below shows trees obtained for the column in various elementary cellular automata. In cases like rules 250 and 254 no initial condition gives the specified column, so all branches eventually die out. In class 2 examples like rule 10 many intermediate configurations are possible. Rules like 90 and to some extent 30 that allow sideways evolution yield comparatively simple trees.

If one wants to find just a single initial condition that works then one can set up a recursive algorithm that in effect does a depth-first traversal of the tree. No doubt in many cases the number of nodes that have to be visited eventually increases like 2^^{t}, but many branches usually die off quickly, greatly reducing the typical effort required in practice.