Combining persistent homology and invariance groups for shape comparison (Q262311)

Suppose that \(X\) is a metric space that has the structure of a finite simplicial complex with non-trivial homology in dimension \(k\). Give the space of continuous real-valued functions on \(X\) the topology induced by the sup-norm \(\|\cdot\|_\infty\), and let \(\Phi\) be a topological subspace that contains the constant functions. Moreover, let \(G\) be a subgroup of self-homeomorphisms of \(X\) such that \(\Phi\) is closed under action by \(G\): for \(\phi\in\Phi\) and \(g\in G\), we have that \(\phi\circ g\in\Phi\). The \textit{natural pseudo-distance} on \(\Phi\) is defined as \(d_G(\phi_1,\phi_2)=\inf_{g\in G}\|\phi_1-\phi_2\circ g\|_\infty\). It is strongly \(G\)-invariant in the sense that \(d_G(\phi_1\circ g_1,\phi_2\circ g_2)=d_G(\phi_1,\phi_2)\) for all \(g_1,g_2\in G\). Let \({\mathcal F}(\Phi,G)\) denote the set of all operators \(F:\Phi\rightarrow\Phi\) such that \(F(\phi\circ g)=F(\phi)\circ g\) and \(\|F(\phi_1)-F(\phi_2)\|_\infty\leq\|\phi_1-\phi_2\|_\infty\). And let \(r_k(\phi)\) denote \(k\)-th persistent Betti number function for the filtration by sub-level sets \(X_t=\phi^{-1}\bigl((\infty,t]\bigr)\); i.e., for each point \((u,v)\) in the plane with \(u<v\), \(r_k(\phi)(u,v)\) is the rank of the image of the map in homology \(H_k(X_u)\rightarrow H_k(X_v)\) induced by inclusion. From persistent homology theory, the persistent Betti number function can be described by a persistence diagram, and the bottleneck distance function measures the difference between two such diagrams. The bottleneck distance gives rise to a pseudo-metric \(d_{\mathrm{match}}\) on the set of all such functions \(r_k(\phi)\), and hence a pseudo-metric on \(\Phi\) for any nonempty subset \({\mathcal F}\subseteq{\mathcal F}(\Phi,G)\) given by \(D_{\mathrm{match}}^{\mathcal F}(\phi_1,\phi_2)=\sup_{F\in{\mathcal F}}d_{\mathrm{match}}(r_k(F(\phi_1)),r_k(F(\phi_2)))\). The authors show that \(D_{\mathrm{match}}^{\mathcal F}\) is strongly \(G\)-invariant, and that \(D_{\mathrm{match}}^{\mathcal F}\leq d_G\). In the case when \({\mathcal F}\) is equal to all of \({\mathcal F}(\Phi,G)\), it is shown that \(D_{\mathrm{match}}^{\mathcal F}\) is in fact equal to the natural pseudo-distance. Moreover, if \(\Phi\) is compact as a metric space, then one can always find a finite subset \({\mathcal F}^\ast\subseteq{\mathcal F}\) such that \(D_{\mathrm{match}}^{{\mathcal F}^\ast}\) approximates \(D_{\mathrm{match}}^{\mathcal F}\) arbitrarily closely. To end the article, the authors present the results of numerical experiments that demonstrate the feasibility of using \(D_{\mathrm{match}}^{{\mathcal F}^\ast}\) to approximate the natural pseudo-distance.