Alpha magnitude (Q6103327)

Magnitude was introduced by \textit{T. Leinster} [Doc. Math. 18, 857--905 (2013; Zbl 1284.51011)] to study cardinality-like invariants of a fixed space. This idea was later translated to a filtration of a space by \textit{D. Govc} and \textit{R. Hepworth} [J. Pure Appl. Algebra 225, No. 3, Article ID 106517, 40 p. (2021; Zbl 1457.55004)], i.e. a parameterized family of spaces \(\{X_a\}_{a\in \mathbb{R}}\) where \(X_{a}\subseteq X_{b}\) for \(a \leq b\). Specifically, given a filtration, we can compute a persistence module which is a collection of vector spaces \(M_p = \{H_p(X_a)\}_{a \in \mathbb{R}}\) with linear maps \(H_p(X_{a})\to H_p(X_{b})\), and then decompose this into a barcode for degree \(p\) denoted \(\{[a_{p,i},b_{p,i})\}_{i \in [1,\dots,m_p]}\) which is a finite collection of intervals. Then the persistent magnitude of the full module is defined to be \[ |M| = \sum_{k=0}^{\infty} \sum_{i=0}^{m_{p}}(-1)^{p}(e^{-a_{k,i}}-e^{-b_{k,i}}). \] Also the persistent magnitude function \(t \mapsto |tM|\) was studied where \(tM\) denotes the construction computed after scaling the input metric space. This paper focuses on common simplicial filtrations arising from a finite metric space \((X,d)\); namely the Rips, alpha, and Čech filtrations. The alpha magnitude, in particular, is better suited to real world applications than the previously studied Rips magnitude since the dimension of homology is bounded by the ambient space dimension rather than the number of points in \(X\). They further extend this finite point cloud idea to work for compact subspaces in Euclidean space, provide methods for estimating the value from finite samples, and study the construction as related to fractal dimension.