A smooth word over an alphabet {a, b} is an infinite sequence where applying a “derivative” operation — replacing each maximal run of identical letters by its length — produces another word over the same alphabet. The Oldenburger-Kolakoski word (the sequence that describes its own run lengths) is the most famous example. How fast does the subword complexity — the number of distinct factors of length n — grow?
Note (2026-11-12; read against the paper's full text): the complexity in the conjecture is that of the language of all smooth words, not of each smooth sequence. The paper says that over alphabets where a plus b is even some smooth sequences have linear factor complexity and do not contain all smooth words, and that over alphabets where a plus b is odd it seems (as a conjecture) that every smooth sequence contains all smooth words. The essay's statement that any smooth word over {a,b} has the same complexity growth, and that the growth rate is forced, is therefore not what the paper shows. The paper proves the lower bound for any alphabet and the conjectured upper bound over even alphabets (both letters even), improves the known upper bounds over odd and mixed alphabets, and gives a dichotomy linking the language, the morphicity and the factor complexity of the Oldenburger-Kolakoski sequence.
Sing's conjecture predicts Θ(n^(log(a+b)/log((a+b)/2))), a polynomial growth rate determined entirely by the alphabet (arXiv:2603.10733). The conjecture is now proved for even alphabets, the lower bound holds for all binary alphabets, and the upper bound is improved for odd alphabets. The key insight: f-smooth words (the finite variant) are exactly the factors of smooth words — the finite and infinite versions share the same combinatorial structure.
The structural observation: the complexity growth rate is encoded in the alphabet parameters, not the specific word. Any smooth word over {a, b} has the same complexity growth, determined by the ratio log(a+b)/log((a+b)/2). The derivative operation — which destroys local information — preserves enough global structure that the complexity is an invariant of the operation itself, not of its fixed points. The self-describing property constrains the combinatorics so tightly that the growth rate is forced.