1: \begin{abstract}

2: This work presents a comprehensive theoretical analysis of graph p-Laplacian based framelet network (pL-UFG) to establish a solid understanding of its properties. We begin by conducting a convergence analysis of the p-Laplacian based implicit layer integrated after the framelet convolution, providing insights into the asymptotic behavior of pL-UFG. By exploring the generalized Dirichlet energy of pL-UFG, we demonstrate that the Dirichlet energy remains non-zero, ensuring the avoidance of over-smoothing issues in pL-UFG as it approaches convergence. Furthermore, we elucidate the dynamic energy perspective through which the implicit layer in pL-UFG synergizes with graph framelets, enhancing the model's adaptability to both homophilic and heterophilic data. Remarkably, we establish that the implicit layer can be interpreted as a generalized non-linear diffusion process, enabling training using diverse schemes. These multifaceted analyses lead to unified conclusions that provide novel insights for understanding and implementing pL-UFG, contributing to advancements in the field of graph-based deep learning.

3: % Comprehensive experiments are conducted to verify our claims from the theoretical outcomes, and the outstanding empirical results demonstrate the effectiveness of incorporating additional non-linear (p-Laplacian) processes with multi-scale graph neural networks.  

4:     

5:     % \EScomment{Non-linear diffusion, asymptotic analysis, over-smoothing, generalized Dirichlet energy }

6: \end{abstract}

7: