On the number of t-ary trees with a given path length

Abstract

We show that the number of $t$-ary trees with path length equal to $p$ is $\exp (h(t^{-1})t\log t\frac{p}{\log p}(1+o(1)))$, where $\entropy(x){=}{-}x\log x {-}(1{-}x)\log (1{-}x)$ is the binary entropy function. Besides its intrinsic combinatorial interest, the question recently arose in the context of information theory, where the number of $t$-ary trees with path length $p$ estimates the number of universal types, or, equivalently, the number of different possible Lempel-Ziv'78 dictionaries for sequences of length $p$ over an alphabet of size $t$.

July 2007: added journal reference and DOI, updated references, minor typographical corrections