Regularity Conditions for Critical Point Convergence
Thomas J. Maullin-Sapey, Samuel Davenport · 2025
Read Regularity Conditions for Critical Point Convergence by Thomas J. Maullin-Sapey, Samuel Davenport (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-01854.txt.
Introduction
When a sequence of bumpy landscapes settles toward a limit shape, scientists often assume the count of peaks, pits, and passes settles too. Thomas Maullin-Sapey and Samuel Davenport ask when that assumption is actually safe. Working with functions on a compact surface with an edge, they seek conditions under which the counts of local maxima, minima, and saddle points converge to those of the limit.
In the smoother of their two settings they adapt classic topology results about how features must balance. In the rougher setting they follow an approach modelled on a standard lemma about critical points. They close by restating the main theorems in the language of empirical processes, the form data analysts meet in practice. The work gives a firmer footing for reading features off noisy estimated surfaces.
Catalog record
| CREATOR | Thomas J. Maullin-Sapey, Samuel Davenport |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 120 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-01854
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Regularity Conditions for Critical Point Convergence Thomas J. Maullin-Sapey, Samuel Davenport Abstract We focus on a sequence of functions { f n } \{f_{n}\} , defined on a compact manifold with boundary S S , converging in the C k C^{k} metric to a limit f f . A common assumption implicitly made in the empirical sciences is that when such functions represent random processes derived from data, the topological features of f n f_{n} will eventually resemble those of f f . In this work, we investigate the validity of this claim under various regularity assumptions, with the goal of finding conditions sufficient for the number of local maxima, minima and saddle of such functions to converge. In the C 1 C^{1} setting, we do so by employing lesser-known variants of the Poincaré-Hopf and mountain pass theorems, and in the C 2 C^{2} setting we pursue an approach inspired by the homotopy-based proof of the Morse Lemma. To aid practical use, we end by reformulating our central theorems in the language of the empirical processes. 1 Introduction Practical image analysis settings conventionally assume that a sequence of continuous functions, f_{n} , defined on a compact domain S , converge in some appropriately chosen metric to a limiting process, f (for instance, see the discussions of adler1981geometry , Cheng2017 and davenport2022confidenceregionslocationpeaks ). Although often not acknowledged, many empirical studies implicitly assume that, as more data is collected and n increases, topological features of f_{n} , such as the number of critical points, will converge to those of f . Without appropriate assumptions in place, this is not always the case, and, in fact, there are many settings where assuming such convergences hold would be misleading and inaccurate. For instance, images of the night sky found in the cosmic background explorer dataset will exhibit fractal-like properties, with more stars appearing as resolution increases, limiting the finite resolution interpretation of topological features ( COLEMAN1992 ; Pino1995 ) . By contrast in medical imaging fields such as neuroimaging, great importance is assigned to topological features such as the number of ‘peaks’ and ‘clusters’ in an image of activity in the human brain derived from a finite number of subjects, n Friston1994 ; Chumbley2010 ; maullin2023spatial . However little is provided in the way of theoretical guarantees that these features are representative of the limiting average. In this work, we aim to answer a fundamental question; under what regularity conditions can we assume that the number of local maxima, minima and saddles of f_{n} converge to that of f ? Specifically, we explore the consequences of assuming f_{n}\rightarrow f in the C^{k} metric for various k\in\mathbb{N} , with an underlying C^{k} structure placed on the domain S . It should be noted that in many real-world settings, authors have had to derive such conditions for specific applications ( Adler2009 ; Cheng2017 ; Eldar2024 ) . However, we believe that in such settings the regularity assumptions made can be greatly reduced, and there is a clear need in the literature for a canonical theory, so that authors may bypass such derivations in future. Although seemingly simple, the problem we consider requires much machinery to resolve. A first approach which we consider is to apply the techniques of Morse theory to draw a correspondence between critical points of f_{n} and f . Doing so is complicated by two factors. Firstly, standard Morse theoretic arguments guarantee the existence of neighborhoods around critical points in which f_{n} and f behave nicely, but typically do not quantify the size of such neighborhoods. This is important, as to draw a bijection between the critical points of f_{n} and f , we need to rule out cases where f has a critical point with a Morse neighborhood which contains multiple Morse neighborhoods of critical points of f_{n} . Secondly, such an approach does not tell us about the convergence of degenerate critical points, nor anything about what can be said when f_{n} converges in the C^{1} metric. An alternative technique is to use the tools of homology theory. This is an approach which we also pursue, allowing us to analyze more complex degenerate saddles and settings in which f_{n} and f are only C^{1} . However, this approach again suffers multiple complications. First, to draw a correspondence between critical points, we require a non-conventional notion of homological index for manifolds with boundary, which we draw from jubin2009generalized . Second, the homological index does not distinguish between local maxima, minima and saddles in all dimensions. For instance, in two-dimensions the homological index of a local minimum and a local maximum are both one. In order to state something meaningful about local maxima, minima and saddles in such contexts, we shall instead require a variant of the mountain pass theorem for convex domains (c.f. Jabri_2003 and Appendix D ). This paper is structured as follows. In Section 2 , we detail the notation we shall use, with Section 2.1 introducing homological concepts and Section 2.2 outlining Morse-theoretic definitions. Section 3 then details our theory with illustrative examples given throughout. Specifically, Section 3.1 briefly illustrates why convergence in the C^{0} metric is not sufficient for our purposes. Following this, in Section 3.2 , we use the extended Poincaré-Hopf theorem of jubin2009generalized , and a variant of the mountain pass theorem for convex domains provided by Jabri_2003 to investigate the behavior of the critical points under the assumption of C^{1} convergence. Then, in Section 3.3 , using the variant of Thom’s homotopic proof ( arnold1985singularities ) provided by ioffe1997 , we investigate whether Morse theory can be used to draw a correspondence between critical points under the assumption of C^{2} convergence. We end with a short section, Section 4 , adapting our core results into the framework of empirical random processes outlined by van1996weak , along with a short discussion in Section 5 . Supplemental lemmas are contained in Appendix A , while the versions of the Poincaré-Hopf theorem, mountain pass theorem, and Morse lemma that we employ are provided in Appendices B , D and E , respectively. As we draw on notions from a range of areas, supplemental material is provided acting as a primer on the concepts we reference from empirical random processes, manifold theory and singular homology. For all figures of parameterized surfaces in this document, corresponding interactive Desmos plots are also linked in the supplement. 2 Notation In this section, we list the notation which will be used throughout the document. As we draw from several mathematical domains, a nomenclature is included at the end of the document to help track the notation employed. The following notation is universally employed in both the document, appendices and supplement. For an arbitrary set A , we denote its complement as A^{c} , closure as \overline{A} , topological boundary as \partial A and interior as Int (A) . For an arbitrary point s\in U\subseteq\mathbb{R}^{D} , we denote the closed \epsilon -ball around s as B_{\epsilon}(s):=\{s^{\prime}\in U:|s-s^{\prime}|\leq\epsilon\} , where the ambient space U is context-dependent. For brevity, we also adopt the shorthand B_{\epsilon}:=B_{\epsilon}(0) . We assume that we are working on a D -dimensional compact C^{k} -manifold with boundary, denoted S , where k