An upper bound for reflection length in Coxeter groups

Brian Drake, Evan PetersJournal of Algebraic Combinatorics 2021 · DOI

Introduction

In geometric group theory, Coxeter groups provide a fundamental framework for studying symmetry through reflections. For any element $w$ in a Coxeter group $W$, two primary metrics of complexity are defined: the standard length $l(w)$, representing the minimum number of simple reflections needed to express $w$, and the reflection length $l_R(w)$, the minimum number of any reflections required. While it is well-established that $l_R(w) \le l(w)$, establishing a tight general upper bound has remained an open challenge, particularly for "wild" or non-affine groups where no constant bound exists. This paper establishes the first general, non-trivial upper bound for reflection length across all Coxeter groups.

Theoretical Framework

The derivation of the new upper bound relies on two primary mathematical pillars:

  1. Dyer’s Theorem: This theorem provides a combinatorial bridge, stating that the reflection length $l_R(w)$ is equivalent to the minimum number of letters that must be deleted from a reduced expression of $w$ such that the remaining product equals the identity.
  2. Generalized Pigeonhole Principle: By applying a sophisticated version of this principle to sequences of reflections, the authors demonstrate that in sufficiently long words, specific generators must recur with enough frequency to allow for efficient "cancellation" toward the identity, thereby constraining the reflection length.

The General Upper Bound

The central contribution of this work is the establishment of a non-trivial upper bound that scales with both the standard length and the variety of generators used.

Theorem 1: For any element $w$ in a Coxeter group $W$, the reflection length is bounded by: $$l_R(w) \le l(w) - 2 \left\lceil \frac{l(w)}{nd(w)} \right\rceil + 2$$ where $nd(w)$ denotes the number of distinct simple generators appearing in a reduced expression for $w$.

This formula captures the intuition that the "variety" of reflections ($nd(w)$) significantly influences the potential for reduction. Furthermore, the bound respects the parity preservation property, as $l_R(w)$ and $l(w)$ always share the same parity (both even or both odd).

Computational Results and Sharpness

To investigate the tightness of this bound, the authors focused on universal Coxeter groups $U_n$, where no relations exist between generators other than $s^2=1$. Using the Maple computer algebra system, they enumerated the distribution of reflection lengths for all elements in $U_3$ (a 3-generator universal group) up to standard length 17.

Distribution of reflection lengths for elements in the universal Coxeter group $U_3$ up to length 17.

The empirical data, as shown in the table above, confirms that the theoretical bound is "sharp" for universal groups. Specifically, Theorem 2 proves that for powers of Coxeter elements in $U_n$, the reflection length exactly meets the upper bound defined in Theorem 1. The computational results illustrate how the variety of reflection lengths expands as standard length increases, while remaining strictly within the predicted limits. The authors have made their code available at github.com/brian-drake/Reflection_Length_Maple for verification and extension.

Conclusion and Future Directions

This research provides a significant step forward by providing a scalable ceiling for reflection length. While the bound is proven sharp for universal groups, its tightness in other classes, such as hyperbolic groups, remains a subject for future inquiry. Additionally, the search for a two-variable generating function to describe these distributions and a deeper geometric interpretation of the ceiling function in the bound represent significant open problems in the field.

Summary updated

BibTeX
@article{drake2021upper,
  author = {Brian Drake and Evan Peters},
  title = {{An upper bound for reflection length in Coxeter groups}},
  journal = {Journal of Algebraic Combinatorics},
  year = {2021},
  doi = {10.1007/s10801-021-01016-y},
}
Figure