Generalized persistence diagrams for persistence modules over posets (Q2063197)

The field of Topological Data Analysis (TDA) can be argued to have started with the introduction of persistence diagrams. The input data is that of a persistence module \(\mathcal{V} = (\{V_a\},\{\phi_{a,b}\})\) consisting of vector spaces \(\{V_a \mid a \in \mathbb{R}\}\) with morphisms \(\varphi_{a,b}:V_a \to V_b\) satisfying composition laws \(\varphi_{a,c} = \varphi_{b, c} \circ \varphi_{a,b}\). Under certain finiteness assumptions (see e.g.~[\textit{S. Y. Oudot}, Persistence theory. From quiver representations to data analysis. Providence, RI: American Mathematical Society (AMS) (2015; Zbl 1335.55001)]), these modules always decompose into so-called interval modules, which are defined as follows. Fixing a connected interval \(I \subseteq \mathbb{R}\), the interval module \(\mathcal{I}\) is one for which every vector space is either the one-dimensional vector space \(V_a = k\) if \(a \in I\); or is 0 otherwise. The maps are defined to be isomorphisms if possible, and the 0-map otherwise. The decomposition \(\mathcal{V} = \bigoplus_{I \in B} \mathcal{I}_I \) is unique up to isomorphism; so we call the collection \(B\) the barcode of \(\mathcal{V}\). These are often visualized as a collection of points \((a,b)\) for each interval \(I=(a,b)\) in \(B\), called a persistence diagram. Following [\textit{P. Bubenik} and \textit{J. A. Scott}, Discrete Comput. Geom. 51, No. 3, 600--627 (2014; Zbl 1295.55005)], we can view the persistence module as a functor, \(\mathcal{V}: (\mathbb{R}, \leq) \to \mathbf{vec}\). With this vantage point, it is immediate to generalize the study of persistence modules to functors \(F: \mathbf{P} \to \mathcal{C}\) from a poset to an arbitrary category. Pairs of choices for \(\mathbf{P}\) and \(\mathcal{C}\) result in many objects of study in the TDA literature: persistence modules of course, but also multiparameter persistence, Reeb graphs, mapper graphs, zigzag persistence, and merge trees to name a few. This paper provides a category theoretic version of the rank invariant (Definition 3.5), which generalizes the standard rank invariant \(rk(\phi_{a,b})\). This paper also extends the work of \textit{A. Patel} [J. Appl. Comput. Topol. 1, No. 3--4, 397--419 (2018; Zbl 1398.18015)] in by providing a combinatorial computation for Patel's generalized persistence diagram (Definition 3.13). By way of application, the authors further show that this can be used to compute 0-dimensional levelset persistence in the special case of Reeb graphs (realized when \(\mathcal{C} = \mathbf{Set}\)) without needing to pass to vector space representations.