An extension of fractal Euler number via persistent homology

Kosuke Nishijima · 2026

Read An extension of fractal Euler number via persistent homology by Kosuke Nishijima (2026) 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-2605-22049.txt.

Introduction

Kosuke Nishijima extends the fractal Euler number using persistent homology and, in part, magnitude theory. Llorente-Winter introduced the average fractal Euler number as an Euler characteristic for fractals in Euclidean space, but its applicable class remains very limited. Nishijima modifies the notion with perspectives from applied topology and category theory, widening its reach. He then computes the new average ph-fractal Euler number for classically well-known fractals, especially Cantor dust and the Menger sponge, both excluded from the Llorente-Winter approach.

Catalog record

CREATORKosuke Nishijima
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE39031
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.22049

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

An extension of fractal Euler number via persistent homology Kosuke Nishijima Abstract . In the context of geometric measure theory, Llorente-Winter introduced the (average) fractal Euler number as a notion of the Euler characteristic for fractals embedded in Euclidean space. However, the class of fractals to which it is applicable remains very limited. In the present paper, we modify this notion by applying perspectives of persistent homology and partly the theory of magnitude, which have recently come from applied topology and category theory. We then demonstrate concrete calculation of our average ph-fractal Euler number for some classically well-known fractals, especially the Cantor dust and Menger sponge which are excluded from Llorente-Winter’s approach. 1. Introduction To extract qualitative information about the shape of a compact subset X of Euclidean space \mathbb{R}^{d} , we would first look at its dimension and the number of connected components, or more generally, the number of ‘holes’. As a typically nice case, if X has a retraction to a finite cell complex, the Euler characteristics is defined as the alternative sum \chi(X)=\sum_{i\geq 0}(-1)^{i}\,b_{i}(X) where b_{i}(X) is the i -th Betti number, that is \dim H_{i}(X;\mathbb{R}) , intuitively understood as the number of ‘ i -dimensional holes’ inside X . However, if X is a much more complicated space, like a fractal, it often happens that the dimension becomes a non-integer real number (in the sense of various fractal dimensions) and some Betti numbers do not exist. In fact, a fractal X often admits self-similarity : there exists a finite number of affine contraction self-maps f_{1},\cdots,f_{n}:X\to X such that X=\bigcup_{k=1}^{n}f_{k}(X). If there is a non-trivial i -dimensional cycle \gamma of X , f_{k}(\gamma) may not be trivial and infinitely many non-trivial cycles may possibly be created by iteration of the maps, then b_{i}(X)=+\infty . As a preceding approach in geometric measure theory, Llorente-Winter [ LW07 ] (also Winter and Zähle [ Win06 , WZ13 ] ) introduced a notion of Euler characteristics for such fractals, under a strong assumption. Let X be a compact subset of \mathbb{R}^{d} . Their first step is to fatten the space – consider closed \epsilon -neighborhoods of X (also called \epsilon -parallel sets) X_{\epsilon}:=\{\;x\in\mathbb{R}^{d}\;|\;\inf_{a\in X}d(x,a)\leq\epsilon\;\} where d(\cdot,\cdot) means the Euclidean metric. Now we assume that \chi(X_{\epsilon}) exists for any \epsilon>0 and there is the infimum of s\geq 0 satisfying that \epsilon^{s}|\chi(X_{\epsilon})| is bounded as \epsilon\searrow 0 , called the Euler exponent of X , denoted by \sigma=\sigma(X) . Under this condition, they introduce the fractal Euler number (1.1) \chi_{f}(X):=\lim_{\epsilon\searrow 0}\left(\frac{\epsilon}{d_{\infty}}\right)^{\sigma}\chi(X_{\epsilon}) and the average fractal Euler number (1.2) \chi_{f}^{a}(X):=\lim_{\delta\searrow 0}\frac{1}{|\log\delta|}\int_{\delta}^{1}\left(\frac{\epsilon}{d_{\infty}}\right)^{\sigma}\chi(X_{\epsilon})\frac{d\epsilon}{\epsilon} where d_{\infty} means the diameter of X . Namely, \sigma is the growth rate of \chi(X_{\delta}) as \delta\searrow 0 and \chi_{f}(X) is the proportional constant to (\delta/d_{\infty})^{-\sigma} in asymptotic expansion; if the quantity exists, then \chi_{f}^{a}(X) automatically converges to the same value. However, this is quite rare, that is discussed in detail in [ LW07 ] . Notice that those two notions of ‘Euler characteristics’ represent a relative metric feature of the embedding of X in Euclidean space \mathbb{R}^{d} ; they are obviously invariant under Euclidean motions, but not topological nor fractal intrinsic feature as a metric space X itself, and even not invariant under affine transformations of \mathbb{R}^{d} . In the present paper, we focus on the latter averaged quantity, which measures the limit behavior of proper Riemannian integral in the right hand side of ( 1.2 ) as \delta\searrow 0 . The existence of \chi_{f}^{a}(X) is still a subtle issue [ LW07 ] . Moreover, suppose that X is self-similar and there is a non-zero ‘bad radius’ \epsilon>0 so that |\chi(X_{\epsilon})|=+\infty . Indeed, those are often satisfied even by classically well-known fractals. Then there are infinite many such bad radii and the integral in ( 1.2 ) does no longer make sense for any \sigma . To overcome this trouble, our basic idea is to modify the meaning of the integral, inspired by new concepts from applied topology and category theory. To make it precise, we employ the persistent homology PH_{*}(X_{\bullet}) of the filtered family X_{\bullet}=\{X_{\epsilon}\} indexed by \{\,0\leq\epsilon\leq d_{\infty}\,\mid\,\mbox{$b_{i}(X_{\epsilon})$ exists for $i=0,1,\dots,d$}\,\} . The i -th Betti function b_{i}(X_{\epsilon}) is approximated by a finite step function summing up the characteristic functions over barcodes, that is \sum_{|e|\geq\delta}1\hskip-3.5pt1_{\epsilon} , where the sum runs over all barcodes e\in PH_{i}(X_{\bullet}) with the lifetime |e|=d_{e}-b_{e}\geq\delta ( b_{e} and d_{e} are the birth and death time of e ). Note that, by definition, 1\hskip-3.5pt1_{e}(\epsilon)=1 if \epsilon\in e , 0 otherwise, so \int_{0}^{d_{\infty}}\epsilon^{\sigma-1}1\hskip-3.5pt1_{e}d\epsilon=\frac{1}{\sigma}\left(d_{e}^{\sigma}-b_{e}^{\sigma}\right) (for any \sigma>0 ). In case that there is some non zero bad radius, the Euler exponent \sigma does not exist, so we instead adopt a new kind of fractal dimension, the i -th PH-complexity \sigma_{i}=\sigma_{i}(X) , defined by MacPherson-Schweinhart [ MS12 ] and Schweinhart [ Sch21 ] ; roughly, \sigma_{i} is the growth rate of the number of barcodes in PH_{i}(X_{\bullet}) with |e|\geq\delta as \delta\searrow 0 . Then we replace the integral in the right hand side of ( 1.2 ) by the finite sum (1.3) S_{\delta}(X):=\sum_{i=0}^{d}\frac{(-1)^{i}}{\sigma_{i}}\sum_{|e|\geq\delta}\;\left(\left(\frac{d_{e}}{d_{\infty}}\right)^{\sigma_{i}}-\left(\frac{b_{e}}{d_{\infty}}\right)^{\sigma_{i}}\right) and define a new quantity, the average ph-fractal Euler number , by (1.4) \chi_{a}^{\rm phf}(X):=\lim_{\delta\searrow 0}\frac{1}{|\log\delta|}S_{\delta}(X). Notice that the role of the parameter \delta has been changed to control the lifetime, not just the end point of the region of integration. We also emphasize that this sum ( 1.3 ) can be read off as (an alternative sum of) the magnitude of persistent modules in the sense of Govc-Hepworth [ GH21 ] after taking a suitable change of variables. See §2 and §3 for our precise definition and detail description. Apart from the technical matter of using persistent homology, our Euler number \chi_{a}^{\rm phf}(X) essentially retains the same meaning as Llorente-Winter’s \chi_{f}^{a}(X) , but it is applicable to a wider class of fractals. In fact, we show that \chi_{a}^{\rm phf}(X) coincides with \chi_{f}^{a}(X) under suitable conditions (Proposition 3.5 ). Furthermore, we demonstrate several precise computation of our Euler number \chi_{a}^{\rm phf}(X) on classically famous fractals, where all arguments in computation are elementary. In §4, we treat the 3-adic Cantor set on the line and the Sierpiński carpet in the plane, which are also found in Llerente-Winter [ LW07 ] , and in §5, we take the Cantor dust on the plane and the Menger sponge in 3 -space (Figure 1), which are excluded from Llerente-Winter’s approach because there are non-zero bad radii in each case. In the latter two examples, we observe a new phenomenon, called little barcode dust , which seemingly has not been recognized so far, and we accurately compute the contribution to our Euler number. Finally in §6, we give some conclusion together with supplementary notes. In particular, instead of considering the parallel sets X_{\epsilon} , it would be very interesting to take finite samples {\bf x}\subset X and the persistent homology PH_{*}({\bf x}_{\bullet}) of Ĉech complexes. In fact, Schweinhart [ Sch21 ] has introduced the PH_{i} -dimension through finite sampling as a different fractal dimension from the i -th PH-complexity. It is possible to improve the definition of our average ph-fractal Betti numbers by taking finite samples and the suprimum, that would be expected to relate to magnitude homology. We will discuss it elsewhere. This paper is part of the author’s master

Read the complete text