---
title: 'Star Graph: Theory, Models, and Applications'
url: https://www.emergentmind.com/topics/star-graph
type: topic
---

# Star Graph: Theory, Models, and Applications

to=arxiv_search.search 《凤凰大参考json
{"query":"all:\"star graph\"","max_results":10,"sort_by":"submittedDate","sort_order":"descending"}
to=arxiv_search.search  天天送json
{"query":"ti:\"star graph\"","max_results":10,"sort_by":"submittedDate","sort_order":"descending"}
to=arxiv_search.search _一本道json
{"query":"all:\"star graph\" AND cat:math.CO","max_results":5,"sort_by":"submittedDate","sort_order":"descending"}
to=arxiv_search.search  彩神争霸邀请码json
{"query":"2109.02219 OR 2605.17051 OR 1912.00884 OR 1903.01468","max_results":10,"sort_by":"relevance","sort_order":"descending"}
A star graph is a graph with a single distinguished center adjacent to every other vertex and no edges among the leaves. In one common convention, for $n \ge 2$, the star graph is $S_n=(V,E)$ with $V=\{c\}\cup L$, $L=\{\ell_1,\ldots,\ell_{n-1}\}$, and $E=\{c\ell:\ell\in L\}$; in other literatures, the same term is also used for several non-equivalent constructions, including quantum graphs with one central vertex and multiple metric edges, the intersection graph of maximal stars of a graph, and the star graph associated with a group presentation [2605.17051, 1911.10515, 2605.28366].

## 1. Combinatorial definition and notation

For the finite simple graph-theoretic object, the center–leaf topology is the defining feature. In the convention used for online embedding on star hosts, the center has degree $n-1$, each leaf has degree $1$, $d_H(c,\ell)=1$ for every leaf, $d_H(\ell_i,\ell_j)=2$ for $i\ne j$, and the diameter is $2$ [2605.17051]. In semigroup-theoretic notation, the same graph is written $S_n=K_{1,n-1}$ with vertex set $\{0,1,\ldots,n-1\}$, center $0$, and leaves $\{1,\ldots,n-1\}$ [2401.11741].

| Source convention | Notation | Meaning |
|---|---|---|
| [2605.17051] | $S_n$ | one center and $n-1$ leaves, so $n$ vertices |
| [2510.18265] | $S_n$ | one central vertex $v_0$ and $n$ leaves $v_1,\ldots,v_n$, so $n+1$ vertices |
| [2401.11741] | $S_n=K_{1,n-1}$ | vertex set $\{0,1,\ldots,n-1\}$ with center $0$ |

This variation in notation is persistent rather than accidental. A common misconception is that formulas written for $S_n$ are directly interchangeable across papers. They are not: some results are indexed by the number of vertices, while others are indexed by the number of leaves.

The same hub-and-spoke geometry also appears in physical models. In quantum many-body dynamics, the star graph $S_N$ is specified by one hub node labeled $0$ and leaves labeled $1,\ldots,N-1$, with adjacency matrix entries $A_{0i}=A_{i0}=1$ and $A_{ij}=0$ otherwise; the degree sequence is highly inhomogeneous, with $\deg(0)=N-1$ and $\deg(i)=1$ for every leaf [1903.01468].

## 2. Structural invariants and graph operators

For coloring theory, the star graph is elementary but nontrivial. With the convention in which $S_n$ has one center and $n$ leaves, the $b$-chromatic number satisfies $\phi(S_n)=1$ if $n=0$ and $\phi(S_n)=2$ if $n>0$; a $2$-coloring is obtained by assigning one color to the center and the other to all leaves [2510.18265].

Star graphs become substantially richer under graph operators. For two star graphs $S_n$ and $S_m$ with $n\ge m\ge 1$, the Cartesian product satisfies
\[
\phi(S_n \square S_m)=m+2,
\]
the line graph satisfies
\[
\phi(L(S_n \square S_m))=m+n,
\]
and for the total graph,
\[
\phi(T(S_n \square S_m))=
\begin{cases}
2m+n+1, & \text{if } n>2(m-1),\\
2n+3, & \text{if } n\le 2(m-1),
\end{cases}
\]
for $n\ge m\ge 3$ [2510.18265]. For powers of the Cartesian product, the same paper shows that $\operatorname{Diam}(S_n \square S_m)=4$, and therefore $(S_n \square S_m)^k$ is complete for $k\ge 4$.

In semigroup-theoretic terms, the rigidity of the center–leaf adjacency permits exact formulas for monoids of partial endomorphisms. For $S_n=K_{1,n-1}$, the group of units consists of the leaf permutations fixing the center, $U\simeq S(\Omega_{n-1})$, and several monoids admit closed-form cardinalities; for example,
\[
|PEnd(S_n)|=(n+1)^{n-1}+n^{n-1}+(n-1)2^{n-1},
\]
while $\operatorname{rank}(PEnd(S_n))=7$ for $n\ge 4$ [2401.11741]. The same work gives exact Green’s relations and shows that $PsEnd(S_n)$ and $PswEnd(S_n)$ are regular for all $n$, whereas $PEnd(S_n)$, $PwEnd(S_n)$, and $IEnd(S_n)$ fail to be regular for $n\ge 3$.

The combinatorial Laplacian of a star graph also yields an explicit quantum-state model. For $K_{1,n-1}$, the Laplacian spectrum is
\[
\operatorname{Spec}(L(K_{1,n-1}))=\{n,1^{[n-2]},0^{[1]}\},
\]
and the normalized density-matrix spectrum is
\[
\operatorname{Spec}(\rho(K_{1,n-1}))=
\left\{\frac{n}{2n-2}^{[1]},\frac{1}{2n-2}^{[n-2]},0^{[1]}\right\}.
\]
For the star-like, star-alike, and star-$m$like augmentations obtained by adding leaf–leaf edges, the Von Neumann entropy increases under the edge-addition operation, and the paper shows that these operations are generally not simulable by LOCC in the corresponding pure-state interpretation [1507.05491].

## 3. Algorithmic and online-optimization roles

A star graph is a particularly sharp test case for online reconfiguration because it has one uniquely advantageous location. In the online graph embedding problem on a star host $H=S_n$, a configuration is a bijection $\phi:X\to V(H)$ from guest nodes to host vertices, requests are unordered pairs of guest nodes, serving a request $\{u,v\}$ costs $d_H(\phi_i(u),\phi_i(v))$, and changing the center costs $1$ [2605.17051]. The total online cost is
\[
C(\sigma)=\sum_i \big( d_\tau[\phi_{i-1},\phi_i] + d_H(\phi_i(u_i),\phi_i(v_i))\big).
\]

This model admits tight competitive analysis. The deterministic PivotTracking algorithm is $1.5$-competitive, and no deterministic online algorithm can do better on star hosts. Randomized PivotTracking is $11/9$-competitive against an oblivious adversary, and no randomized online algorithm can beat $11/9$ for $n\ge 10$ [2605.17051]. These are exact optimal ratios rather than upper bounds lacking matching lower bounds.

The proofs rely on a phase decomposition indexed by the current pivot, together with a cheap-versus-expensive request classification relative to the offline optimum. In practical terms, the star host captures a setting where moving one guest to a hub is the only reconfiguration operation considered, so the model is directly relevant to hub-and-spoke network embeddings, virtual machine placement, and VLSI-style central-bus placement [2605.17051].

The computational profile is correspondingly simple: the online update takes $O(1)$ time per request, the migration operation changes only the center, and the candidate-set representation requires $O(n)$ space in the worst case [2605.17051]. This suggests why star hosts are treated as building blocks for trees and more general sparse topologies.

## 4. Quantum graphs, path integrals, and spectral analysis

In analysis and mathematical physics, a quantum star graph is a metric graph consisting of several edges attached to one central vertex. One formulation takes $\Gamma$ to be $N$ semi-infinite edges, each identified with $\mathbb{R}_+$, with functions represented as vectors $u=(u_1,\ldots,u_N)$ and Hilbert spaces
\[
L^q(\Gamma)=\bigoplus_{j=1}^N L^q(\mathbb{R}_+),\qquad
H^1(\Gamma)=\bigoplus_{j=1}^N H^1(\mathbb{R}_+),
\]
together with the continuity constraint
\[
E(\Gamma):=\{u\in H^1(\Gamma):u_1(0)=\cdots=u_N(0)\}
\]
at the vertex [1912.00884]. The corresponding Laplacian with $\delta$-interaction of strength $\alpha$ has domain
\[
\mathrm{dom}(H_\alpha)=
\left\{
v\in H^2(\Gamma):
v_1(0)=\cdots=v_N(0)=:v(0),\
\sum_{j=1}^N v_j'(0)=\alpha v(0)
\right\},
\]
and this framework supports local well-posedness, conservation laws, explicit standing waves, and orbital stability or instability for a nonlinear Klein–Gordon equation on $\Gamma$ [1912.00884].

A complementary path-integral treatment considers a star graph as the one-point union of $N$ semi-infinite edges and uses the folding trick to rewrite the Hilbert space as
\[
\mathcal H(\Gamma)\cong L^2(\mathbb R^+)\otimes \mathbb C^N.
\]
In this picture, self-adjoint boundary conditions at the junction form a $U(N)$ family,
\[
(\mathbb I-\mathbb U)\,\vec\psi(0)-iL_0(\mathbb I+\mathbb U)\,\vec\psi'(0)=\vec 0,
\qquad \mathbb U\in U(N),
\]
and at scale-invariant fixed points the kernel reduces to a $Z_2$-weighted method-of-images formula with a Hermitian unitary matrix $\mathbb U$ satisfying $\mathbb U^2=\mathbb I$ [1104.5481]. This is one of the cleanest instances in which the star topology turns boundary data into matrix-valued path weights.

Finite-length star graphs admit a parallel spectral theory. For a compact quantum star graph with edge lengths $\ell_1,\ldots,\ell_N$, Dirichlet conditions at the outer vertices, and a possibly complex Robin parameter $\alpha$ at the center,
\[
\sum_{j=1}^N u_j'(\ell_j)+\alpha\,u(v)=0,
\]
the resulting operator $H_\alpha$ is maximal sectorial with compact resolvent, and its high-frequency spectrum remains close to that of the selfadjoint Kirchhoff case [1911.04760]. The same paper proves a Weyl law with constant $|\Gamma|/\pi$, shows that $\operatorname{Im}(\lambda)$ has the same sign as $\operatorname{Im}(\alpha)$ for every eigenvalue, and establishes the high-frequency bound
\[
\limsup_{R\to+\infty}
\sup_{\substack{\lambda\in \mathrm{Sp}(H_\alpha)\\ \mathrm{Re}\,\lambda\ge R}}
|\mathrm{Im}\,\lambda|
\le \frac{2|\mathrm{Im}\,\alpha|}{|\Gamma|}.
\]

Symmetry reduction on star graphs is also unusually explicit. For the three-edge star quantum graph with equal edge lengths and Neumann conditions at the loose ends, Band, Berkolaiko, Joyner, and Liu’s quotient-graph theory decomposes the Hamiltonian into quotient segments associated with irreducible representations of $S_3$ or $C_3$; for standard coupling this yields one Neumann–Neumann quotient and two identical Neumann–Dirichlet quotients, while for $\delta$-coupling the symmetric sector becomes Neumann–Robin with effective parameter $\alpha/3$ [2108.05253].

## 5. Synchronization and transport on star topologies

In synchronization theory, the noisy Kuramoto model on a star graph distinguishes a hub phase $\Phi$ from leaf phases $\theta_j$. With hub frequency $\beta\omega$, identical leaf frequency $\omega$, and white noise on the leaves, the model studied in [2204.05581] is
\[
\dot{\Phi}= \beta\omega + \frac{\lambda\beta}{N}\sum_{j=1}^{N}\sin(\theta_j-\Phi),\qquad
\dot{\theta}_j = \omega + \lambda\sin(\Phi-\theta_j) + \eta_j(t).
\]
Using the relative phase $\phi_j=\theta_j-\Phi$, the stationary Fokker–Planck equation yields the complex self-consistency relation
\[
r=\frac{I_{1-iA}(B)}{I_{-iA}(B)},
\qquad
A=-\frac{(\beta-1)\omega+\lambda\beta\,\mathrm{Im}\,r}{D},\quad
B=\frac{\lambda}{D},
\]
and the paper shows a crossover from abrupt synchronization with hysteresis at small noise to a continuous transition at larger noise [2204.05581]. For the reported numerics with $\beta=10$ and $\omega=1$, hysteresis is prominent for small $D$, becomes negligible around $D\approx 1.0$, is continuous around $D\approx 1.5$, and no sharp transition is observed around $D\approx 2.0$ [2204.05581].

Urban traffic models use a different but related star-graph abstraction. In [1912.03380], a star-like arrangement of uni-directional loops around a central intersection is mapped, via the cell transmission method, to a directed cell-transmission star graph in which a hub node feeds all leaves and each leaf returns only to the hub. With speed-matching flow
\[
J_{i\to j}(\rho_i,\rho_j)=\rho_i(t)\,\frac{u_j(t)}{k_i}\,A_{ij},
\]
the dynamics becomes a nonlinear diffusion on the cell-transmission graph [1912.03380]. In the low-density regime, where $u_i(\rho_i)=v_{\max}$, the steady state on the star graph satisfies
\[
\rho_{1,s}=\frac{N\rho_0}{2},\qquad
\rho_{i,s}=\frac{N\rho_0}{2(N-1)},\quad i=2,\ldots,N,
\]
so the hub carries higher density and current than the leaves [1912.03380]. Closed-form high-density steady states are derived explicitly for $N=3$ and $N=4$.

These two examples illustrate a recurring pattern: the star topology separates a single high-degree coordinator from many degree-one peripherals, so mean-field reduction and hub-bottleneck effects become analytically visible without eliminating nontrivial phase behavior.

## 6. Many-body and impurity physics

For generic $2$-local spin-$1/2$ Hamiltonians restricted to the edges of a star graph, the hub degree scales with system size even though the graph is sparse. The disordered Hamiltonians studied in [1903.01468] take the form
\[
H=\sum_{i=1}^{N-1}\sum_{\alpha,\beta\in\{x,y,z\}}
J_i^{\alpha\beta}\,\sigma_0^\alpha\otimes \sigma_i^\beta
+\vec h_0\cdot\vec\sigma_0
+\sum_{i=1}^{N-1}\vec h_i\cdot\vec\sigma_i,
\]
with interactions only along hub–leaf edges. Exact diagonalization shows Wigner–Dyson level repulsion at infinite temperature, with adjacent gap ratio $r\approx 0.60$, compatible with the Gaussian unitary ensemble [1903.01468]. The same work finds that both unregulated and regulated OTOCs become $O(1)$ on $O(1/\mathcal J)$ time scales independent of $N$, and emphasizes that this does not contradict the MSS chaos bound because the locality assumptions behind Lieb–Robinson-type constraints are not satisfied on the star graph [1903.01468].

The hub is the algebraic source of this behavior. In the integrable Ising star, operators localized at the hub instantaneously accumulate products sourced by all leaves, and in the disordered model the early-time decay of the hub two-point function can be captured by an effective random-field two-level system. Brownian Hamiltonian dynamics and random unitary circuits fail to reproduce key features of the Hamiltonian star graph, including coherent protection effects and the specific form of hub-mediated interference [1903.01468].

A closely related star-graph limit appears in multichannel Kondo physics. In the zero-bandwidth reduction of the $K$-channel Kondo model, each channel contributes one local spin-$1/2$ degree of freedom coupled only to the impurity spin, giving the Hamiltonian
\[
H_{\mathrm{star}}=\sum_{\alpha=1}^{K}J_\alpha\,\mathbf S_d\cdot \mathbf S_\alpha.
\]
This is literally a spin-exchange star with one impurity center and $K$ leaf spins [2205.00790]. In the channel-isotropic case, the ground-state degeneracy is
\[
g(K,S)=|K-2S|+1,
\]
and the low-temperature impurity susceptibility in the star-graph limit diverges as
\[
\chi_{\mathrm{imp}}(T\to 0)=\frac{K-1}{12(K+1)}\,\frac{1}{T},
\]
signalling a critical intermediate-coupling fixed point [2205.00790]. Despite the absence of inter-channel coupling in the star-graph Hamiltonian, the mutual information between two channels is nonzero, and the twist/translation algebra on the degenerate ground-state manifold yields quantized phases interpreted as topological quantum numbers [2205.00790].

## 7. Alternative meanings and star-shaped reasoning networks

The expression “star graph” is overloaded across several subfields. In graph-operator theory, the star graph of a graph $H$ is the intersection graph of its maximal induced stars. If $S(H)$ denotes the family of maximal stars of $H$, then the star graph is the intersection graph on $S(H)$, and a graph is star-critical if its star graph changes after removal of any vertex of the pre-image [1911.10515]. This notion has its own structural theory: star graphs are biconnected, every edge lies in at least one triangle, the recognition problem lies in \textsf{NP}, and for a star-critical pre-image $H$ with $m=|V(S(H))|$,
\[
|V(H)|\le \tfrac12(3m^2-m)
\]
[1911.10515]. This object is unrelated to the simple center–leaf graph $K_{1,n-1}$ except through the use of maximal stars in the pre-image.

In combinatorial group theory, the star graph of a presentation $P=\langle X\mid R\rangle$ has vertex set $X^{\pm1}$, and its edges are determined by two-letter subwords in the symmetrised closure of $R$ [2605.28366]. Within this meaning, for each $n\ge 3$ there is an $n$-generator one-relator presentation whose star graph is $K_{n,n}$; these groups are large and hyperbolic, and concise special presentations with star graph $K_{3,3}$ are completely classified up to Tietze equivalence, yielding exactly twelve torsion-free one-relator cases [2605.28366]. Here again, “star graph” is not the graph $K_{1,n-1}$ but a derived combinatorial object.

Machine learning uses the topology in yet another way. In the Star-shaped Reasoning Graph Network for kinship verification, a pair of facial features is converted into a star-shaped graph
\[
\mathcal G_S=(\mathcal V_S,\mathcal E_S),
\qquad
\mathcal V_S=\{h_c\}\cup\{h_d\mid d=1,\ldots,D\},
\qquad
\mathcal E_S=\{e_{cd}\mid d=1,\ldots,D\},
\]
with one central latent node and $D$ surrounding nodes, one for each feature dimension [2109.02219]. Each leaf is initialized by
\[
h_d^0=[g_d(x_i,\Omega)\,\|\,g_d(y_j,\Omega)],
\]
the center is initialized by pooling across the leaves, and message passing proceeds only through center–leaf interactions [2109.02219]. The topology uses $O(D)$ edges and per-layer complexity linear in $D$, in contrast to a fully connected leaf–leaf graph with $O(D^2)$ edges. On four kinship datasets, the reported mean verification rates for S-RGN are $79.5\%$ on KinFaceW-I, $90.6\%$ on KinFaceW-II, $90.7\%$ on TSKinFace bi-subject, $93.2\%$ on TSKinFace tri-subject, and $87.4\%$ on Cornell KinFace; the same paper identifies the single-center bottleneck as the motivation for the hierarchical extension H-RGN [2109.02219].

Taken together, these usages show that the center–leaf star is less a single object than a recurring architectural motif. In finite graph theory it is the complete bipartite graph $K_{1,n-1}$; in metric and quantum settings it is a one-vertex junction of edges; in online algorithms it is the minimal nontrivial hub-and-spoke host; in physics it isolates high-degree effects; and in several neighboring literatures the phrase “star graph” names derived constructions whose relation to $K_{1,n-1}$ is analogical rather than definitional.

Source: https://www.emergentmind.com/topics/star-graph