Discrete Heat Kernels on Simplicial Complexes and Its Application to Functional Brain Networks

Networks constitute fundamental organizational structures across biological systems, although conventional graph-theoretic analyses capture exclusively pairwise interactions, thereby omitting the intricate higher-order relationships that characterize network complexity. This work proposes a unified framework for heat kernel smoothing on simplicial complexes, extending classical signal processing methodologies from vertices and edges to cycles and higher-dimensional structures. Through Hodge Laplacian, a discrete heat kernel on a finite simplicial complex K is constructed to smooth signals on k-simplices via the boundary operator partialk. Computationally efficient sparse algorithms for constructing boundary operators are developed to implement linear diffusion processes on k-simplices. The methodology generalizes heat kernel smoothing to k-simplices, utilizing boundary structure to localize topological features while maintaining homological invariance. Simulation studies demonstrate qualitative signal enhancement across vertex and edge domains following diffusion processes. Application to parcellated human brain functional connectivity networks reveals that simplex-space smoothing attenuates spurious connections while amplifying coherent anatomical architectures, establishing practical significance for computational neuroscience applications.