--- title: 'Central Graph: Theory & Applications' url: https://www.emergentmind.com/topics/central-graph type: topic --- # Central Graph: Theory & Applications Searching arXiv for recent and relevant papers on “central graph” and closely related graph-theoretic usages. The term **central graph** has multiple technical meanings in contemporary arXiv literature. In classical graph theory, the standard meaning is an operation on a simple graph \(G\): subdivide every edge exactly once and then join every pair of original vertices that were nonadjacent in \(G\) [2105.05368]. This construction, usually denoted \(C(G)\), is the dominant usage in recent spectral, resistance-distance, and coloring work [2404.06317]. A distinct but related line of research uses **central** to denote graph-theoretic centrality structures such as centers, centered peripheries, central vertices, or central parts [1705.07982], [2110.00738], [2501.18932]. In applied machine learning and information retrieval, the phrase is also used more loosely for graph representations organized around central entities or centrality-aware operators [1901.05392], [2601.21733], [2111.06995]. The graph-theoretic operation \(C(G)\) remains the most precise and standardized sense; however, the broader literature shows that “central graph” also functions as a family resemblance term for graph constructions in which metric center, centrality hierarchy, or title-centered semantic organization is the governing principle. ## 1. Standard graph-theoretic construction For a simple graph \(G\) with \(n\) vertices and \(m\) edges, the **central graph** \(C(G)\) is obtained by subdividing every edge of \(G\) exactly once and then joining all pairs of vertices that were nonadjacent in \(G\) [2105.05368]. Equivalently, if \(G\) has vertex set \(V(G)=\{v_1,\dots,v_n\}\) and edge set \(E(G)=\{e_1,\dots,e_m\}\), then \(C(G)\) contains the original vertices together with one new vertex for each edge \(e_i\in E(G)\); these inserted vertices are often described as the vertices corresponding to the edges of \(G\) [2404.06317]. The construction has \(m+n\) vertices and \(m+\frac{n(n-1)}{2}\) edges [2105.05368]. This definition induces a characteristic two-layer structure. Among the original vertices, adjacency in \(C(G)\) is governed by nonadjacency in \(G\); between original and inserted vertices, incidence is inherited from the edge subdivision; and the inserted vertices themselves form the subdivision layer. In matrix terms, if \(I(G)\) denotes the incidence matrix and \(A(\bar G)=J_n-I_n-A(G)\) is the adjacency matrix of the complement, then with a suitable labeling \[ A(C(G))= \begin{pmatrix} A(\bar G) & I(G)\\ I(G)^T & O_{m\times m} \end{pmatrix}, \] and for an \(r\)-regular graph, \[ D(C(G))= \begin{pmatrix} (n-1)I_n & O_{n\times m}\\ O_{m\times n} & 2I_m \end{pmatrix} \] [2105.05368]. The same block structure underlies Laplacian and resistance-distance analyses, where the Laplacian is written as \[ L(C(G))= \begin{bmatrix} nI-J+A(G) & -Q(G)\\ -Q(G)^T & 2I \end{bmatrix}, \] with \(Q(G)\) the incidence matrix [2404.06317]. A useful interpretive point is that \(C(G)\) is neither merely a subdivision graph nor merely a complement-based augmentation. It combines both operations. This suggests why many invariants of \(C(G)\) split into one part controlled by the original spectrum of \(G\) and another part controlled by the inserted edge-vertices. ## 2. Structural and metric centrality notions distinct from \(C(G)\) A second major usage of “central graph” concerns **central vertices** and **centers** of graphs, rather than the operation \(C(G)\). For a connected graph \(G\), the center is the set of vertices of minimum eccentricity, \[ C(G)=\{v\in V(G): e(v)=\min_{x\in V(G)} e(x)\}, \] where \[ e(v)=\max_{u\in V(G)} d(v,u) \] [2110.00738]. The same paper treats the median, security center, characteristic center, subgraph core, and core vertices as six distinct “central parts,” and proves that for a connected vertex-transitive graph each of these six central parts is the whole vertex set [2110.00738]. This is a different notion from the central graph operation, although the overlap in notation is substantial. Within ring-theoretic graph constructions, the term **central vertices** appears in the zero-divisor graph \(\Gamma(\mathbb{Z}_n)\). There the vertices are the nonzero zero-divisors of \(\mathbb{Z}_n\), and the paper determines the center completely: \[ \operatorname{Center}(\Gamma(\mathbb{Z}_n)) = \bigcup_{p\mid n}\bigl(\operatorname{ann}(p)\setminus\{0\}\bigr) \] [2501.18932]. Concretely, the central vertices are exactly the nonzero multiples of \(n/p\) for primes \(p\mid n\) [2501.18932]. The same work also characterizes cut-edges in \(\Gamma(\mathbb{Z}_n)\), showing that for odd \(n\) there are no cut-edges [2501.18932]. A further extension is the theory of **uniform central graphs (UCGs)**. For a graph \(G\), the centered periphery is \[ CP(G)=\bigcup_{z\in Z(G)} \mathrm{EC}(z), \] where \(Z(G)\) is the center and \(\mathrm{EC}(v)\) is the set of eccentric vertices of \(v\) [1705.07982]. A graph is a uniform central graph if all central vertices have exactly the same set of eccentric vertices: \[ \mathrm{EC}(c)=CP(G)\quad\text{for every }c\in Z(G) \] [1705.07982]. The central-peripheral appendage number \(A_{ucg}(C,P)\) then measures the minimum number of intermediate vertices needed to realize a prescribed center \(C\) and centered periphery \(P\) within a UCG [1705.07982]. These lines of work should not be conflated. In one case, “central graph” means the graph operation \(C(G)\). In the other, “central” refers to vertices or substructures defined by metric or combinatorial centrality. The shared vocabulary reflects a common concern with graph-theoretic centrality, but the objects are different. ## 3. Spectral theory of the central graph operation The spectral theory of \(C(G)\) is particularly tractable when \(G\) is regular. For an \(r\)-regular graph with adjacency eigenvalues \(\lambda_1=r,\lambda_2,\dots,\lambda_n\), the normalized Laplacian characteristic polynomial of \(C(G)\) is \[ f(\mathcal{L}(C(G)),x) = (x-1)^{m-n} \prod_{i=1}^{n} \left( (x-1)\left((x-1)-\frac{P(\lambda_i)-1-\lambda_i}{n-1}\right) -\frac{\lambda_i+r}{2(n-1)} \right), \] where \(P\) is the polynomial associated with the adjacency spectrum of \(G\) [2105.05368]. This yields a complete normalized Laplacian description: the eigenvalue \(1\) appears with multiplicity \(m-n\), while the remaining eigenvalues come from quadratic factors indexed by the adjacency eigenvalues of \(G\) [2105.05368]. The same spectral decomposition leads directly to closed formulas for Kemeny’s constant and the degree Kirchhoff index. If \(G\) is \(r\)-regular of order \(n\) and size \(m\), then \[ K(C(G)) = m-n+\frac{n-1}{n-1+r} + \sum_{i=2}^{n}\frac{2(2n+\lambda_i-1)}{2n+\lambda_i-r}, \] and \[ Kf^*(C(G)) = (2m+n(n-1)) \left[ m-n+\frac{n-1}{n-1+r} + \sum_{i=2}^{n}\frac{2(2n+\lambda_i-1)}{2n+\lambda_i-r} \right] \] [2105.05368]. The structure of these formulas reflects a recurring theme: subdivision vertices generate a large repeated spectral block, while the complement-induced interactions among original vertices inject the dependence on the adjacency spectrum of \(G\). A different spectral program studies the **eccentricity matrix** \(\epsilon(G)\) for central graphs of regular graphs. For a triangle-free \(r\)-regular \((p,q)\)-graph \(G\), the eccentricity matrix of \(C[G]\) is written in block form as \[ \epsilon(C[G])= \begin{pmatrix} 2A(G) & 2(J-R(G))\\ 2(J-R(G))^T & 3(J-I-B(G)) \end{pmatrix}, \] where \(R(G)\) is the incidence matrix and \(B(G)\) is the adjacency matrix of the line graph \(L(G)\) [2411.12599]. From this, the paper derives the \(\epsilon\)-spectrum, \(\epsilon\)-energy, inertia, and irreducibility of \(C[G]\) [2411.12599]. In particular, for triangle-free regular graphs, \[ In(\epsilon(C[G]))=(q,p,0) \] and \[ E_\epsilon(C[G]) =3(q-p) +\sqrt{(3q-8r+3)^2+16(p-2)(q-r)} +\sum_{i=2}^{p}\sqrt{(5\lambda_i+3(r-1))^2+16(\lambda_i+r)} \] [2411.12599]. This suggests that the central graph operation is unusually amenable to block-matrix methods. Because \(C(G)\) retains a clean separation between original vertices and edge-vertices, spectral quantities that are difficult for generic graph products remain explicit for large regular families. ## 4. Resistance distance, Kirchhoff index, and random-walk quantities The resistance-distance theory of central graphs is built from the Laplacian block form and Schur complement identities. For a connected graph \(G\) of order \(n\), the resistance distance between two original vertices \(v_i,v_j\in V(G)\) in \(C(G)\) is \[ r_{ij}(C(G)) = \left(\!-\frac{1}{2}L(G)+nI\right)^{-1}_{ii} + \left(\!-\frac{1}{2}L(G)+nI\right)^{-1}_{jj} - 2\left(\!-\frac{1}{2}L(G)+nI\right)^{-1}_{ij} \] [2404.06317]. For an inserted vertex corresponding to \(e_i=v_pv_q\) and an original vertex \(v_j\), \[ r_{ij}(C(G)) = \frac{1}{2}\left( 1+r_{pj}(C(G))+r_{qj}(C(G))-\frac{1}{2}r_{pq}(C(G)) \right), \] and for two inserted vertices corresponding to \(e_i=v_pv_q\) and \(e_j=v_sv_t\), \[ r_{ij}(C(G)) = 1+\frac{1}{4}\Bigl( r_{ps}(C(G))+r_{pt}(C(G))+r_{qs}(C(G))+r_{qt}(C(G)) -r_{pq}(C(G))-r_{st}(C(G)) \Bigr) \] [2404.06317]. These formulas yield a general Kirchhoff-index expression for \(C(G)\). For a connected graph \(G\) of order \(n\) and size \(m\), \[ \begin{aligned} Kf(C(G)) &= (m+n)\,\operatorname{tr}\!\left(\!-\frac12L(G)+nI\right)^{-1} -\frac14\,\alpha^T \operatorname{tr}\!\left(\!-\frac12L(G)+nI\right)^{-1}\alpha \\ &\quad +\frac{m(m+n-1)}{2} -\frac{2m+n}{n} +\frac{(m+n)(n-1)}{4}\\ &\quad +\frac{m+n}{4}\Biggl( \sum_{i