A generalized Goulden-Jackson cluster method and lattice path enumeration

The Goulden-Jackson cluster method is a powerful tool for obtaining generating functions for counting words in a free monoid by occurrences of a set of subwords. We introduce a generalization of the cluster method for monoid networks, which generalize the combinatorial framework of free monoids. As a sample application of the generalized cluster method, we compute bivariate and multivariate generating functions counting Motzkin paths—both with height bounded and unbounded—by statistics corresponding to the number of occurrences of various subwords, yielding both closed-form and continued fraction formulae.

💡 Research Summary

The paper presents a substantial extension of the classic Goulden–Jackson cluster method, which originally provides generating functions for words in a free monoid counted by occurrences of a prescribed set of subwords. The authors introduce the notion of a “monoid network,” a combinatorial structure that generalizes free monoids by encoding words as walks on a directed graph whose arcs are labelled with subsets of an alphabet. In this setting each walk is a sequence of pairs (letter, arc), and a projection map sends the walk to the underlying word while preserving its start and end vertices. The key requirement for a monoid network is that two distinct walks cannot produce the same word with the same endpoints; this guarantees a one‑to‑one correspondence between walks and words in the network.

Using a matrix representation, each labelled arc‑letter pair is associated with an elementary matrix whose non‑zero entry carries the letter. A homomorphism λ maps concatenations of such pairs to matrix products, thereby translating combinatorial enumeration into linear‑algebraic expressions. The authors prove that the generating matrix for all walks in the network is simply ((I_m-\sum_{p\in P}M_p)^{-1}), where (M_p) are the elementary matrices and (I_m) is the identity. This result mirrors the classic formula (F(t)=\bigl(1-\sum_{a\in A}a-L(t-1)\bigr)^{-1}) for free monoids, but now holds in a matrix‑valued context.

The central contribution is a generalized cluster theorem for monoid networks. Given a (possibly infinite) set (B={\beta_1,\beta_2,\dots}) of forbidden subwords, the authors define for each (\beta_k) the collection of walks whose projected word equals (\beta_k). Clusters are minimal marked words that cannot be decomposed into two non‑empty marked words, exactly as in the original theory. The cluster generating matrix (\mathbf{L}_G(t_1,t_2,\dots)) records, for each walk, the multivariate weight (\prod_k t_k^{\mathrm{mk}_k(c)}) where (\mathrm{mk}_k(c)) counts marked occurrences of (\beta_k) inside the cluster. The generalized theorem then states: \