Unimodal category and the monotonicity conjecture (Q2063200)

The notion of \textit{unimodal category} of a continuous non-negative function on a topological space is motivated by statistics and was introduced by \textit{Y. Barishnikov} and \textit{R. Ghrist} in [``Unimodal category and topological statistics'', in: Proceedings of NOLTA (2011), \url{https://www2.math.upenn.edu/~ghrist/preprints/unimodalsnote.pdf}]. A continuous non-negative function is called unimodal if all of its positive superlevel sets (or excursion sets) are contractible, which generalizes the notion that the function has a unique local maximum. Many important probability distributions like binomial and Poisson distributions are unimodal. The unimodal category \(\mathrm{ucat}(f)\) of a continuous function \(f:X \to [0,+\infty)\), where \(X\) is a topological space, denotes the minimal number \(n \in \mathbb{N}\), such that \(f= \sum_{i=1}^n u_i\) holds pointwise for some unimodal functions \(u_1,\dots,u_n:X \to [0,+\infty)\). Similarly, one defines numbers \(\mathrm{ucat}^p(f)\) for \(p>0\) by replacing the previous equation by demanding that \(f^p =\sum_{i=1}^n u_i^p\) holds pointwise. Loosely speaking, the unimodal category might thus be used to study decompositions of arbitrary probability distributions into elementary ones. Barishnikov and Ghrist further conjectured a monotonicity property of unimodal category, namely that with \(f\) given as above, it should hold that \(\mathrm{ucat}^{p_1}(f) \leq \mathrm{ucat}^{p_2}(f)\) whenever \(p_1 \leq p_2\). In the present article, the author carries out further details and results on unimodal category and shows that this monotonicity conjecture is false in general by providing explicit counterexamples in various cases. While the seminal article of Baryshnikov and Ghrist is sketchy in nature, the present article further adds details to the seminal constructions. It might be read without prior knowledge of unimodal category and provides a well-written introduction to the topic. After an introductory section, the author uses Section 2 to recall unimodal category and to carry out relations between unimodal category and the total variation of a function, thereby generalizing observations made by Baryshnikov and Ghrist. In Section 3 the author uses total variations to generalize a method of computing the unimodal category of compactly supported functions on \(\mathbb{R}\) from the seminal article, a so-called sweeping algorithm, to arbitrary continuous functions on \(\mathbb{R}\). Afterwards, he further transfers the construction to functions on \(S^1\). The abovementioned monotonicity conjecture of Baryshnikov and Ghrist is studied in Section 4 of the article. Here and in the appendix, the author provides detailed explicit examples of functions that violate the conjecture. He first constructs counterexamples that are piecewise linear functions on certain graphs containing cycles. The author then studies functions on \(\mathbb{R}^2\), first of all constructing an explicit examples out of some piecewise linear functions. Afterwards, the author elaborates on an idea of \textit{L. Hickok}, \textit{J. Villatoro} and \textit{X. Wang} [``Unimodal category of 2-dimensional distributions'', unpublished note (undated), \url{https://faculty.math.illinois.edu/~xwang105/unimodal.pdf}] to show that for a certain class of Morse functions on \(\mathbb{R}^2\) the monotonicity conjecture indeed holds true. Towards the end, the author further lines out some ideas for future research, e.g. a generalization of unimodal category to so-called multimodal functions, whose positive superlevel sets are not necessarily contractible, but instead homotopy-equivalent to finite sets.