Di-graphs with tightly connected clusters: Effective Graph Laplacians and resolvent convergence
Christian Koke · 2026
Read Di-graphs with tightly connected clusters: Effective Graph Laplacians and resolvent convergence by Christian Koke (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-18057.txt.
Introduction
Christian Koke studies Laplacians on graphs whose clusters grow ever more tightly linked. His 2026 preprint follows graphs on a shared node set where edge weights inside set clusters tend to infinity. As the links inside a cluster grow, the graph Laplacians converge, in the resolvent sense, to an effective graph Laplacian. The note gives a clean limit picture for graphs with tight groups.
Catalog record
| CREATOR | Christian Koke |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 146 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-18057
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Di-graphs with tightly connected clusters: Effective Graph Laplacians and resolvent convergence Christian Koke Abstract . In this note, we study Laplacians on graphs for which connectivity within certain sub-graphs tends to infinity. Our main focus are graphs sharing a common node set on which edge weights within certain clusters grow to infinity. As intra-cluster connectivity increases, we show that the corresponding graph Laplacians converge—in the resolvent sense—to an effective graph Laplacian. This effective limit Laplacian is defined on a coarsened graph, where each highly connected cluster is collapsed into a single node. In the undirected setting, the effective Laplacian arises naturally from agregating over tightly connected clusters. In the directed case, the limiting graph structure depends on the precise manner in which connectivity increases; with the corresponding effects mediated by the left and right kernel structure of the Laplacian restricted to high-connectivity clusters. Our results shed light on the emergence of coarse-grained dynamics in large-scale networks and contribute to spectral graph theory of directed graphs. 1. Introduction Graphs are a fundamental tool for modeling pairwise interactions in systems across a wide range of disciplines, including biology [ 23 ] , physics [ 55 ] , computer science [ 14 ] , as well as social sciences [ 54 ] . Fundamentally graphs may be distinguished by whether they are undirected or directed . While directed graphs are mathematically somewhat more subtle and less well understood than their undirected counterparts [ 50 , 49 ] , they are essential for modeling a diverse array of systems. Applications include the structure of the internet [ 9 ] , social networks [ 10 ] , food webs [ 39 ] , epidemic spread [ 24 ] , chemical reaction networks [ 44 ] , graph-structured databases [ 2 ] , communication networks [ 1 ] , and coordination in multi-agent systems [ 16 ] . In many such settings, tightly connected sub graphs – often referred to as communities or clusters – emerge and play a central role in shaping the system’s global behavior. Detecting such communities has been shown to be essential for uncovering functional subunits and understanding how local organization drives global dynamics [ 17 , 41 , 18 , 40 , 42 ] . . Given the prevalence of such hierarchical structure, it is natural to investigate representations of networks in which tightly connected subgraphs are abstracted as single nodes. This process, commonly referred to as coarse-graining [ 38 , 45 ] , not only reduces the complexity of the system but also reveals its large-scale organization [ 53 , 43 ] , facilitates scalable computation [ 37 , 22 ] and provides insight into the mechanisms by which local interactions influence global behavior [ 52 , 51 ] . In this present paper we then rigorously address the question: What is the appropriate effective representation of a network containing such strongly connected clusters? To introduce our approach, consider as a motivating example a graph modeling an ensemble of coupled heat reservoirs. From the perspective of heat flow, edge weights then correspond to thermal conductivities. In the limit where the conductivity between two nodes tends to infinity, heat exchange becomes instantaneous, and the nodes equilibrate immediately—effectively behaving as a single unit. Heat flow over a graph is governed by the diffusion equation \partial_{t}\psi(t)=L\psi(t) , with graph Laplacian L and global-in-time solution \psi(t)=e^{-tL}\psi_{0} . Heuristically, we expect this fast-equilibration behavior to be reflected in the evolution operator (i.e. the heat kernel): If two nodes are tightly coupled by a very large edge weight ’ \beta ’, then we expect e^{-tL_{\beta}}\rightarrow J_{\uparrow}e^{-t\underline{L}}J_{\downarrow} as \beta\rightarrow\infty . Here L_{\beta} is the gaph Laplacian on the original graph. \underline{L} is the Laplacian of a coarsened graph in which the tightly connected nodes are collapsed into a single “supernode”. J_{\downarrow} and J_{\uparrow} interpolate between original and reduced graph. In this work, we formalize this intuition by studying the asymptotic behavior of graph Laplacians associated with a sequence of weighted graphs G_{\beta} in which connectivity within specified clusters increases without bound. We show that, under suitable conditions, this sequence of Laplacians converges to a limiting operator: the effective Laplacian on a coarsened graph. The structure of this coarsified graph \underline{G} is determined by the original graph G and the spectral projection onto the kernel of the restriction of the original graph Laplacian to the tightly knit clusters. Instead of establishing convergence of heat-kernels directly, we find it more convenient to instead establish (1.1) \displaystyle(L_{\beta}-zI_{\ell^{2}(G)})^{-1}\rightarrow J^{\uparrow}(\underline{L}-zI_{\ell^{2}(\underline{G})})^{-1}J^{\downarrow},\quad\text{as $\beta\rightarrow\infty$}; i.e. convergence of resolvents . This then also implies convergence of heat kernels. In the reduced graph \underline{G} , each high-connectivity cluster is represented as a single node. Inter-cluster connections on the original G are then aggregated to determine the graph structure of \underline{G} for which the convergence ( 1.1 ) holds. The way this aggregation proceeds – and hence also the resulting graph structure on \underline{G} – differs between the undirected and directed settings: In the undirected setting , we find that the limit graph structure is insensitive to the specific way in which connectivity within subclusters diverges. In the directed case , however, the situation is subtler: The structure of the effective graph \underline{G} depends not only on the clusters themselves, but on how exactly connectivity is taken to infinity. This dependence is mediated by the left and right kernel structures of the Laplacian restricted to each cluster. These kernel components encode asymmetries and directional flow patterns, introducing a rich variety of possible limiting graphs even for the same cluster partitioning. Taken together, these results provide a rigorous framework for understanding coarse-graining and effective descriptions for potentially directed networks, and contribute to the broader development of spectral graph theory in potentially directed and weighted settings. Our results also have direct implications for spectral methods in clustering, community detection and graph based machine learning [ 27 , 33 , 26 ] . 1 1 1 Additional in-progress work using methods developed in this current paper have presented at non-archival venues [ 29 , 28 , 30 , 31 , 32 ] Our paper is organized as follows: Section 2 discusses the general parlance in which our results are formulated. In particular it recalls the notion of in- and out-degree Laplacians and characterizes the kernels of these operators in terms of the connectivity structure of the corresponding directed graphs. Section 3 then covers convergence results for undirected graphs. Section 4 discusses convergence in the technically more involved directed setting. 2. Graphs and Graph Laplacians 2.1. Graphs and the notion of Connectedness The principal objects of our study are weighted and generically directed graphs [ 3 , 6 , 7 , 21 , 35 ] ): Definition 2.1 . A (weighted and directed) graph G is a tuple G=(V,m,E,a) where (i) V is an at most countably-infinite set of vertices, (ii) m:V\rightarrow(0,\infty) is a function assigning a positive mass m(v)>0 to each v\in V , (iii) E\subseteq V\times V is the (locally finite) set of edges in the graph G , (iv) a:V\times V\rightarrow[0,\infty) is a weight function satisfying a(v,w)>0 iff (v,w)\in E . The transpose G^{\intercal} of an original graph G is the tuple G^{\intercal}=(V,m,E^{\intercal},a^{\intercal}) , with E^{\intercal}=\{(i,j)\in V\times V:(j,i)\in E\}