A graph discretization of vector Laplacian (Q6885567)

This paper formalizes the graph discretization of the vector Laplacian (or Helmholtz operator) by introducing the Helmholtzian matrix \(\mathcal{H}(G)\). While the scalar Laplacian \(-\operatorname{div}\operatorname{grad}\) corresponds to the standard vertex-indexed Laplacian matrix \(L(G)\), the vector Laplacian \(\nabla^{2}F = \operatorname{grad}\operatorname{div} F - \operatorname{curl}\operatorname{curl} F\) requires a higher-order structure to capture edge-triangle interactions. The study is motivated by the need to clarify the distinction between scalar and vector Laplacians in discrete settings, a theoretical gap that persisted in continuous space until the mid-20th century, credit to the work of \textit{P. Moon} and \textit{D. E. Spencer} [J. Franklin Inst. 256, No. 6, 551--558 (1953; \url{doi:10.1016/0016-0032(53)91160-0})]. By utilizing the clique complex of a graph as a structural bridge, the authors define \(\mathcal{H}(G)\) as a square matrix indexed by the edge set, effectively realizing the 1-Laplacian operator within the broader framework of Hodge Laplacians on graphs as introduced by \textit{L.-H. Lim} [SIAM Rev. 62, No. 3, 685--715 (2020; Zbl 1453.05061)].\N\NThe main combinatorial contribution is the derivation of the matrix presentation \(\mathcal{H}(G) = \mathcal{B}(G)\mathcal{B}(G)^{T} + \mathcal{C}(G)^{T}\mathcal{C}(G)\), where \(\mathcal{B}(G)\) and \(\mathcal{C}(G)\) are the edge-vertex and triangle-edge incidence matrices. The authors apply the Hoffman program to the \(\mathcal{H}\)-spectral radius \(\lambda(G)\), identifying limit points below approximately \(4.38\). This characterization builds upon established results for Laplacian limit points of trees [\textit{J.-M. Guo}, Linear Algebra Appl. 429, No. 7, 1705--1718 (2008; Zbl 1144.05317)], as the non-zero eigenvalues of \(L(G)\) and \(\mathcal{H}(G)\) coincide for triangle-free graphs. Theorem 3.18 provides a complete structural classification of connected graphs whose \(\mathcal{H}\)-spectral radii fall into distinct intervals within the range \((0, 4.38)\), identifying families such as paths, even cycles, and specific star-like variants.\N\NThe article further investigates how structural perturbations, specifically the subdivision of edges, influence the \(\mathcal{H}\)-spectrum. Proposition 4.2 proves that for a connected graph where an internal path forms an even cycle, subdividing an edge twice strictly decreases the spectral radius. The authors conclude by highlighting the utility of the Helmholtzian matrix in modern network science. Unlike the scalar Laplacian, \(\mathcal{H}(G)\) explicitly incorporates triangle degrees, making it uniquely sensitive to the clustering features of small-world networks [\textit{D. J. Watts} and \textit{S. H. Strogatz}, Nature, London 393, No. 6684, 440--442 (1998; Zbl 1368.05139)]. Furthermore, the discrete vector Laplacian provides a rigorous basis for analyzing diffusion processes beyond the node-space, such as random walks on edges in simplicial complexes.