Persistent Topological Laplacians -- a Survey

Xiaoqi Wei,Guo-Wei Wei

DOI: https://doi.org/10.26434/chemrxiv-2023-8c9vh

2023-12-14

Abstract:Persistent topological Laplacians constitute a new class of tools in topological data analysis (TDA), motivated by the necessity to address challenges encountered in persistent homology when handling complex data. These Laplacians combines multiscale analysis with topological techniques to characterize the topological and geometrical features of functions and data. Their kernels fully retrieve the topological invariants of persistent homology, while their nonharmonic spectra provide supplementary information, such as the homotopic shape evolution of data. Persistent topological Laplacians have demonstrated superior performance over persistent homology in addressing large-scale protein engineering datasets. In this survey, we offer a pedagogical review of persistent topological Laplacians formulated on various mathematical objects, including simplicial complexes, path complexes, flag complexes, diraphs, hypergraphs, hyperdigraphs, cellular sheaves, as well as $N$-chain complexes. Alongside fundamental mathematical concepts, we emphasize the theoretical formulations associated with various persistent topological Laplacians and illustrate their applications through numerous simple geometric shapes.

Chemistry

What problem does this paper attempt to address?

The paper aims to introduce and review the concept of Persistent Topological Laplacians and their applications in Topological Data Analysis (TDA). Specifically, the paper attempts to address the following issues:

1. **Overcoming the limitations of traditional topological homology**: Traditional topological homology theory can only describe topological spaces up to homeomorphism or homotopy equivalence, which is insufficient when dealing with complex data. For example, adding an edge in a simple graph keeps the zeroth Betti number unchanged, even if the connectivity of the graph changes. Additionally, it cannot capture important information such as the number and composition of points in loops.

2. **Introducing multi-scale analysis**: Persistent Topological Laplacians combine multi-scale analysis with combinatorial Laplacians, not only preserving the information of persistent Betti numbers but also capturing additional information on shape evolution through non-harmonic spectra.

3. **Expanding the toolkit of topological data analysis**: Persistent Topological Laplacians can be applied to various mathematical objects, including simplicial complexes, path complexes, flag complexes, directed graphs, hypergraphs, hyperdirected graphs, cellular sheaves, and N-chain complexes. These extensions help in better understanding and analyzing the topological and geometric features of complex datasets.

4. **Improving machine learning model performance**: The paper demonstrates the superior performance of Persistent Topological Laplacians on large-scale protein engineering datasets, and utilizing non-harmonic spectral information can significantly enhance the performance of machine learning models based on persistent homology.

In summary, this review paper aims to introduce the emerging tool of Persistent Topological Laplacians to a broader audience and promote further development in this field.