Looping-Edge Graphs: Theory and Applications

• Looping-edge graphs are structures where edges return to their originating vertex, affecting adjacency, Laplacians, and overall graph dynamics in both discrete and quantum contexts.
• In discrete settings, self-loops modify local diagonal structure and enable spectral analysis via Laplacian adjustments, incidence matrices, and lifted graph methodologies.
• In metric graphs, looping edges form compact loops with attached half-lines, supporting standing waves, self-adjoint extension theory, and multi-pulse localization phenomena.

Searching arXiv for recent and foundational papers on looping-edge graphs and related self-loop/looped graph notions. Looping-edge graphs are graph-theoretic objects in which an edge returns to its incident vertex, but the term is used in two distinct technical senses. In discrete graph theory, a looping edge is a self-loop $(i,i)$ on a finite undirected graph, and its principal effect is to modify adjacency, degree, Laplacian, factorization, and degree-preserving dynamics by adding local diagonal structure (Acikmese, 2015). In quantum-graph and nonlinear-wave literature, a looping-edge graph usually denotes a metric graph consisting of a circular edge together with a finite number of half-lines attached to a common vertex; in that setting, the loop is a compact component supporting periodic or localized profiles, while the attached rays provide non-compact scattering channels (Pava et al., 14 Jul 2025). The subject therefore spans combinatorial spectral theory, Cartesian product factorization, Markov-chain connectivity of loopy graph spaces, self-adjoint extension theory, and the existence and stability of standing waves for nonlinear Schrödinger equations.

1. Terminological scope and canonical models

In the discrete setting, the basic object is a finite undirected graph $G=(V,E)$ in which edges may include self-loops $(i,i)$. One formalism uses a vertex-edge incidence matrix $E$ in which an ordinary edge contributes a row with a $+1$ and a $-1$, while a self-loop contributes a row with a single nonzero entry $+1$. The associated Laplacian satisfies

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$

where $G^o$ is obtained by removing the self-loops. This exhibits the defining spectral effect of looping edges: they contribute positive diagonal terms to the Laplacian (Acikmese, 2015).

A second discrete usage arises in the theory of Cartesian products of graphs with loops. There the factors are finite undirected graphs without multiple edges, but loops are allowed, and the product is defined as in the simple-graph case: two vertices are adjacent precisely when they differ in exactly one coordinate and that coordinate pair forms an edge in the corresponding factor. The loop structure in the product is determined by an OR-condition: the product graph has a loop at $(v_1,\dots,v_k)$ if and only if at least one coordinate $G=(V,E)$0 has a loop in $G=(V,E)$1 (Boiko et al., 2014).

In the metric-graph setting, a looping-edge graph is a non-compact quantum graph with one compact loop and $G=(V,E)$2 half-lines attached to a single vertex. One standard model writes

$G=(V,E)$3

where $G=(V,E)$4 is the circular edge and the $G=(V,E)$5 for $G=(V,E)$6 are half-lines identified with $G=(V,E)$7, all meeting at the common vertex $G=(V,E)$8. A state on the graph is represented by

$G=(V,E)$9

with $(i,i)$0 defined on the loop and $(i,i)$1 on the $(i,i)$2-th half-line (Pava et al., 1 Jan 2026).

This terminological bifurcation is substantive rather than merely linguistic. In discrete graphs, looping edges alter combinatorial invariants and finite-dimensional operators. In metric graphs, the loop is a compact geometric component that interacts with differential operators, vertex couplings, and nonlinear bound states. A plausible implication is that the shared term reflects a common topological motif—a local return of an edge to a vertex—but the analytical consequences differ sharply across the two settings.

2. Discrete looping edges: Laplacians, positivity, and lifted graphs

For finite undirected graphs with self-loops, the Laplacian decomposition

$(i,i)$3

makes explicit that self-loops add positive diagonal mass. This is the key reason why looped graphs can have Laplacians that are positive definite even when the loop-free part would possess the usual zero mode associated with the vector $(i,i)$4 (Acikmese, 2015).

To formalize when the nullspace disappears, the notion of a pseudo-connected graph is introduced. An undirected graph $(i,i)$5 without multiple edges is pseudo-connected if every vertex is connected to itself and/or to another vertex, and every connected subgraph has at least one vertex with a self-loop. For such graphs, the Laplacian is positive definite: $(i,i)$6 hence every Laplacian eigenvalue is strictly positive. This extends the familiar positive-semidefinite theory of loop-free connected graphs by showing that the self-loop terms eliminate the standard one-dimensional kernel (Acikmese, 2015).

The central technical device is a lifted graph $(i,i)$7 without self-loops. If $(i,i)$8 has $(i,i)$9 vertices, then $E$0 has $E$1 vertices: one copy of each original vertex, a second copy of each original vertex, and a middle vertex $E$2. Each ordinary edge is duplicated across the two copies, while each self-loop $E$3 is replaced by the two-edge path

$E$4

This construction embeds the original looped spectrum into the loop-free lifted spectrum: $E$5 Moreover,

$E$6

If $E$7 is pseudo-connected, the embedded eigenvalues lie in the positive spectrum of the lifted graph: $E$8 Thus self-loops can be analyzed by passing to a larger loop-free graph while retaining sharp spectral information (Acikmese, 2015).

The principal significance of this theory is that it converts an ostensibly nonstandard Laplacian problem into a standard loop-free spectral problem. This suggests a general methodological pattern: looping edges can often be traded for auxiliary structure, after which familiar operator-theoretic tools become available.

3. Product structure, factorization, and swap dynamics in loopy discrete graphs

The Cartesian product theory for graphs with loops extends the Sabidussi–Vizing paradigm from simple graphs to connected finite graphs with at least one unlooped vertex. The product remains commutative, associative, and distributive over disjoint union, and $E$9 remains the multiplicative unit. However, entirely looped graphs are exceptional: if $+1$0 is entirely looped, then $+1$1 is entirely looped for every $+1$2, and product structure becomes insensitive to the detailed loop pattern of the factors. For this reason, the unique prime factorization theorem is stated for connected graphs with at least one unlooped vertex (Boiko et al., 2014).

Within this nondegenerate class, every nontrivial connected graph with at least one unlooped vertex admits a factorization into irreducible factors with respect to the Cartesian product, and the factorization is unique up to isomorphism and permutation of factors. Layers remain fundamental, but loops create a new dichotomy: a factor layer is either an actual copy of the factor or a version in which every vertex has a loop, depending on whether the fixed complementary coordinates are unlooped. The same paper also proves that the prime factorization can be computed in $+1$3 time, where $+1$4 is the number of edges of the graph (Boiko et al., 2014).

A distinct structural issue appears in state spaces of loopy graphs with fixed degree sequence. For loopy graphs—self-loops allowed, multiedges forbidden—the usual degree-preserving double edge swap does not always connect the full realization space. The obstruction is exact: the graph of realizations is disconnected if and only if some realization lies in one of two special structural classes, $+1$5 or $+1$6. These classes generalize, respectively, clique-type and cycle-type disconnected families (Nishimura, 2017).

To recover irreducibility, one augments double swaps by triangle-loop swaps,

$+1$7

and the reverse move. With this additional move, the state space $+1$8 becomes connected for every degree sequence. The resulting Markov chain is connected, aperiodic, and satisfies detailed balance for the symmetric proposal mechanism, yielding the uniform distribution over loopy graphs with the given degree sequence as the unique stationary distribution (Nishimura, 2017).

These results show that looping edges are not merely a local decoration. They can change global decomposition theory, alter the geometry of realization spaces, and invalidate standard connectivity heuristics that are correct for simple graphs, multigraphs, or pseudographs.

4. Looping-edge graphs as metric graphs and self-adjoint operator models

In the quantum-graph literature, a looping-edge graph is a metric graph formed by one loop $+1$9 and $-1$0 half-lines $-1$1 meeting at a common vertex. The differential operators act edgewise, while the graph topology is encoded entirely in the vertex conditions. For the Schrödinger operator, the minimal symmetric operator is

$-1$2

with minimal domain

$-1$3

and adjoint domain

$-1$4

Its deficiency indices are

$-1$5

so self-adjoint extensions exist in abundance (Pava et al., 14 Jul 2025).

The extension theory is formulated through boundary data at the loop endpoints and half-line origins. One introduces vectors

$-1$6

and organizes the Green boundary form in Krein spaces. Self-adjoint extensions correspond to $-1$7-self-orthogonal subspaces of the boundary space, equivalently to graphs of unitary operators between boundary Krein spaces. By Stone’s theorem, every such self-adjoint extension generates a unitary Schrödinger dynamics (Pava et al., 14 Jul 2025). A closely related operator-theoretic treatment also covers Airy and Schrödinger-type operators on looping-edge graphs and characterizes unitary or contractive dynamics through linear operators acting on the associated boundary Krein spaces (Angulo et al., 2024).

Several distinguished couplings occur repeatedly. A $-1$8-type family imposes continuity of the function at the vertex together with a derivative jump: $-1$9

$+1$0

A $+1$1-type family instead enforces continuity of derivatives but not of function values. In one formulation,

$+1$2

with $+1$3. The operator $+1$4 is self-adjoint for all real $+1$5 (Pava et al., 1 Jan 2026).

These formulations clarify the analytical meaning of a looping edge in the metric setting. The loop contributes two boundary traces at the same geometric vertex, and that duplication is precisely what distinguishes the extension theory from that of a pendant or internal edge. In the Airy setting, the same geometry yields richer possibilities because third-order operators can generate either unitary groups or contraction semigroups depending on the balance of deficiency structure and boundary coupling (Angulo et al., 2024).

5. Standing waves and edge-localized states on looping-edge graphs

The principal nonlinear model is the focusing nonlinear Schrödinger equation on a quantum graph,

$+1$6

or, in another convention,

$+1$7

Standing waves of the form $+1$8 or $+1$9 satisfy stationary equations on the loop and tails, coupled by the chosen vertex conditions (Kairzhan et al., 2021).

For cubic NLS on the looping-edge graph with $$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$0-type interaction, one seeks states

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$1

where the loop component solves

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$2

and each half-line component solves the corresponding decaying ODE on $$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$3. In the cubic case, the tail profiles are translated sech-solitary waves,

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$4

with the shifts determined by the vertex relations (Pava et al., 1 Jan 2026).

In the periodic case $$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$5, $$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$6, the loop profile is a translated dnoidal Jacobi elliptic function,

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$7

with parameters satisfying

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$8

The common tail profile is

$$\mathcal{L}(G)=\mathcal{L}(G^o)+\sum_{(i,i)\in E} e_i e_i^T,$$9

valid for

$G^o$0

A smooth family of standing waves

$G^o$1

exists on an admissible interval

$G^o$2

where $G^o$3 are roots of an explicit quadratic $G^o$4 (Pava et al., 1 Jan 2026).

The non-periodic $G^o$5-coupled case with $G^o$6 is obtained perturbatively by the Implicit Function Theorem. Fixing $G^o$7 and a base frequency $G^o$8, one constructs a smooth local branch

$G^o$9

such that

$(v_1,\dots,v_k)$0

The branch converges in $(v_1,\dots,v_k)$1 to the periodic dnoidal-plus-tail state as $(v_1,\dots,v_k)$2 (Pava et al., 1 Jan 2026).

The corresponding stability theory is governed by the linearized operators

$(v_1,\dots,v_k)$3

For the periodic branch, if $(v_1,\dots,v_k)$4, then

$(v_1,\dots,v_k)$5

and

$(v_1,\dots,v_k)$6

For the perturbed branch, the same picture persists near the base point, and the positivity of

$(v_1,\dots,v_k)$7

or its perturbed analogue yields orbital stability by the Grillakis–Shatah–Strauss framework. In the symmetric even-$(v_1,\dots,v_k)$8 branch, the Morse index increases to $(v_1,\dots,v_k)$9 in the high-frequency regime, producing orbital instability. The transition occurs at

$G=(V,E)$00

(Pava et al., 1 Jan 2026). Closely related results for $G=(V,E)$01-type interactions on the same looping-edge geometry are developed through self-adjoint extension theory and dnoidal-plus-tail constructions in (Pava et al., 14 Jul 2025).

6. Multi-pulse localization and broader relations to looped graphs

Looping edges also appear in the large-mass theory of edge-localized states on general quantum graphs. In that setting, bounded edges are classified as pendant, looping, or internal, and a looping edge of length $G=(V,E)$02 is represented as $G=(V,E)$03 with both endpoints attached to the same vertex. For the rescaled stationary equation

$G=(V,E)$04

the large-mass limit supports positive localized states concentrated on prescribed bounded edges, including arbitrary finite collections of looping edges (Kairzhan et al., 2021).

A central existence theorem states that, for sufficiently large $G=(V,E)$05, there exists a positive $G=(V,E)$06-pulse edge-localized state concentrated on a selected set of bounded edges $G=(V,E)$07, with exponentially small leakage onto the complement: $G=(V,E)$08 When a selected edge is a loop, the localized pulse is centered at its midpoint $G=(V,E)$09 and is symmetric and monotone toward the shared vertex at both ends (Kairzhan et al., 2021).

The technical novelty of this construction is a Dirichlet-to-Neumann decomposition of the graph into the localized part and its complement. A looping edge contributes two endpoint fluxes at the same vertex, and this is reflected in the boundary matching formula by a factor $G=(V,E)$10 for each loop and a factor $G=(V,E)$11 in the exponentially small flux term associated with looping edges. This is one of the clearest analytical manifestations of the loop’s double attachment (Kairzhan et al., 2021).

For states localized only on pendant and looping edges, the Morse index theorem is exact: $G=(V,E)$12 where $G=(V,E)$13. Thus each localized loop pulse contributes exactly one negative eigenvalue. The proof uses a homotopy between Neumann–Kirchhoff and Dirichlet boundary conditions and then a Sturm-theoretic count on each selected edge (Kairzhan et al., 2021).

The earlier large-mass analysis of single-edge localization yields a complementary energetic classification. On a loop of total length $G=(V,E)$14 attached at a vertex of degree $G=(V,E)$15,

$G=(V,E)$16

$G=(V,E)$17

$G=(V,E)$18

Compared to pendant-edge states, loop-centered states carry roughly double leading mass and double leading energy magnitude, but they remain less favorable than pendants at the same mass. The energetic ordering for compact graphs at sufficiently large mass is: a pendant edge if one exists; otherwise certain loops; then internal edges, with edge length and attachment degree entering through explicit asymptotic coefficients (Berkolaiko et al., 2019).

Beyond quantum graphs, looping edges retain independent significance in other areas of graph theory and combinatorics. They enter extremal homomorphism problems through loop-threshold target graphs (Cutler et al., 2016), exact enumeration of tree-like multigraphs with prescribed numbers of self-loops and multiple edges (Azam et al., 25 Oct 2025), and the commutative algebra of monomial edge ideals in graphs with loops, where loop generators appear as squares $G=(V,E)$19 and affect linear quotients, linear resolutions, vertex cover ideals, and linear-type properties (Imbesi et al., 2011). Taken together, these results indicate that looping-edge graphs form not a single theory but a family of closely related theories in which a local loop structure systematically reshapes spectral, algebraic, combinatorial, and nonlinear-analytic behavior.