Chapter 12: The Principle of Computational Equivalence

Section 8: Undecidability and Intractability

Word problems

The question of whether a particular string can be generated in a given multiway system is an example of a so-called word problem. An original more specialized version of this was posed by Max Dehn in 1911 for groups and by Axel Thue in 1914 for semigroups. As discussed on page 938 a finitely presented group or semigroup can be viewed as a special case of a multiway system, in which the rules of the multiway system are obtained from relations between strings consisting of products of generators. The word problem then asks if a given product of such generators is equal to the identity element. Following work by Alan Turing in the mid-1930s, it was shown in 1947 by Emil Post from the undecidability of PCP that the word problem for semigroups is in general undecidable. Andrei Markov gave a specific example of this for a semigroup with 13 generators and 33 relations, and by 1966 Gennadií Makanin had found the simpler example


Using these relations as rules for a multiway system most initial strings yield behavior that either dies out or becomes repetitive. The shortest initial strings that give unbounded growth are "BBBBABB" and "BBBBBBA"—though both of these still eventually yield just exponentially increasing numbers of distinct strings. In 1967 Yuri Matiyasevich constructed a semigroup with 3 complicated relations that has an undecidable word problem. It is not yet known whether undecidability can occur in a semigroup with a single relation. The word problem is known to be decidable for commutative semigroups.

The word problem for groups was shown to be undecidable in the mid-1950s by Petr Novikov and William Boone. There are however various classes of groups for which it is decidable. Abelian groups are one example. Another are so-called automatic groups, studied particularly in the 1980s, in which equivalence of words can be recognized by a finite automaton. (Such groups turn out to have definite geometrical properties, and are associated with spaces of negative curvature.) Even if a group ultimately has only a finite number of distinct elements, its word problem (with elements specified as products of generators) may still be undecidable. Constructions of groups with undecidable word problems have been based on setting up relations that correspond to the rules in a universal Turing machine. With the simplest such machine known in the past (see page 706) one gets a group with 32 generators and 142 relations. But with the universal Turing machine from page 707 one gets a group with 14 generators and 52 relations. (In general s k+4 generators and 5 s k+2 relations are needed.) From the results in this book it seems likely that there are still much simpler examples—some of which could perhaps be found by setting up groups to emulate rule 110. Note that groups with just one relation were shown always to have decidable word problems by Wilhelm Magnus in 1932.

For ordinary multiway (semi-Thue) systems, an example with an undecidable word problem is known with 2 types of elements and 5 very complicated rules—but I am quite certain that much simpler examples are possible. (1-rule multiway systems always have decidable word problems.)

From Stephen Wolfram: A New Kind of Science [citation]