Computation of gamma-linear projected barcodes for multiparameter persistence

Microlocal methods and methods of sheaf theory and homological algebra applied to PDEs (35A27) Computational methods for problems pertaining to algebraic topology (55-08) Sheaf cohomology in algebraic topology (55N30) Persistent homology and applications, topological data analysis (55N31)

Review text

One-parameter persistence modules capture how homology groups change with respect to a single parameter. These modules have a complete invariant called a persistence barcode. Real data often involve multiple meaningful parameters, where no barcode-like complete invariant exists. Furthermore, computing multi-parameter persistence modules of large data is a challenge in itself. The last few years saw significant effort being devoted in developing substitutes for a complete invariant that are computable and carry meaningful information. The \(\gamma\)-linear projected barcodes are such an invariant.\N\NPreviously, the computation of \(\gamma\)-linear projected barcodes was restricted to persistent cohomology modules of \(\mathbb{R}^n\)-valued functions. This article develops the theory and algorithms for computing \(\gamma\)-linear projected barcodes in the more general and direct algebraic setting of multi-parameter persistence modules. This is achieved by using the derived category of \(\gamma\)-sheaves, for a given cone \(\gamma\) in the dual space of \(\mathbb{R}^n\). In particular, the \(\gamma\)-projected barcodes are the derived pushforwards \(Ru_{*}\) of a given linear form \(u\) on \(\mathbb{R}^n\) with operator norm \(1\). A \(\gamma\)-projected barcode is shown to have a combinatorial structure that can be encoded and used for machine learning on data.\N\NThe main contribution of this manuscript is an algorithm that encodes the combinatorial structure of a \(\gamma\)-linear projected barcode from a finitely presented multi-parameter persistence module. To this end, the authors establish novel results in the homological algebra of \(\gamma\)-sheaves. The relevance of microlocal sheaf theory, and \(\gamma\)-sheaves in particular, to the theory of multi-parameter persistent homology and barcodes was demonstrated in [\textit{M. Kashiwara} and \textit{P. Schapira}, J. Appl. Comput. Topol. 2, No. 1--2, 83--113 (2018; Zbl 1423.55013); \textit{N. Berkouk} and \textit{F. Petit}, Algebr. Geom. Topol. 21, No. 1, 247--277 (2021; Zbl 1472.55003)].\N\NThe central mathematical theory underlying the algorithm is that of conic complexes of \(\gamma\)-sheaves. Under the functorial correspondence between \(\gamma\)-sheaves and persistence modules, free resolutions of persistence modules map to conic complexes; moreover, derived pushforwards along linear forms send conic complexes on \(\mathbb{R}^n\) to conic complexes on \(\mathbb{R}\). Finally, a persistence module admitting a finite free resolution can be fully reconstructed from its associated \(\gamma\)-sheaf.\N\NMost works in topological data analysis do not operate in the derived setting. The sheaf-theoretic perspective adopted here demonstrates that derived methods can play a meaningful role in algorithmic design, thereby expanding the toolkit available for studying multi-parameter persistence. This manuscript should be of interest to both algorithmically minded researchers and algebraists working in applied topology.

Cites work

This page was built for publication: Computation of gamma-linear projected barcodes for multiparameter persistence

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6981550)