Weighted Cycle-Based Restrictions in Graphs

• Weighted cycle-based restrictions are defined as constraints where cycles, rather than arbitrary subgraphs, serve as the key objects for ensuring feasibility and optimality in graph problems.
• They underpin problems like edge-/vertex-deletion in cycle covering, lex short cycle basis computations, and resistance-balanced chord screening, leading to improved approximations and algorithm efficiencies.
• These methods integrate algebraic, combinatorial, and spectral techniques through local inequalities and equal-cycle-weight laws, enhancing network analysis and design in various weighted graph models.

Weighted cycle-based restrictions are constraints, objective functions, or representation rules in which cycles are not treated as arbitrary subgraphs but as the primary structured objects on which feasibility, optimality, or algebraic realization depends. In the literature, this includes edge- and vertex-deletion problems that forbid prescribed cycle families, restricted cycle-basis constructions, local inequalities that normalize edge weights by the heaviest cycle through each edge, resistance-based screening of chords in weighted cycles, and algebraic formalisms in which only cycle configurations satisfying specific local weight-balance rules are admissible (Tang et al., 2018, Eppstein et al., 2015, Zheng et al., 15 Mar 2025, Ai et al., 19 Mar 2026, Deng et al., 23 May 2026, Weng, 11 Mar 2025).

1. Edge- and vertex-deletion restrictions on target cycle families

A canonical graph-optimization form of a weighted cycle-based restriction is the covering or transversal problem: one pays for deleting edges or vertices so that the residual graph contains no member of a specified cycle family. For a weighted graph $G(V,E)$ with edge-weight function $\mathbf w:E\to Z^{|E|}_{+}$, a $k$-cycle covering is an edge set $A\subseteq E$ such that $G-A$ has no cycle of length exactly $k$. Its minimum weight is the weighted covering number $\tau_k(G_w)$, and the paper formulates it as

$$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$

where $A$ is the $k$-cycle-edge adjacency matrix. The LP relaxation

$\mathbf w:E\to Z^{|E|}_{+}$0

gives a trivial $\mathbf w:E\to Z^{|E|}_{+}$1-approximation by thresholding at $\mathbf w:E\to Z^{|E|}_{+}$2. For odd $\mathbf w:E\to Z^{|E|}_{+}$3, the paper improves this to a $\mathbf w:E\to Z^{|E|}_{+}$4 approximation by thresholding at $\mathbf w:E\to Z^{|E|}_{+}$5, removing those edges, and then deleting the non-crossing edges of a maximum-weight cut in the residual graph so that the remainder becomes bipartite and therefore contains no odd cycle (Tang et al., 2018).

The odd/even dichotomy is structural rather than merely algorithmic. The odd-$\mathbf w:E\to Z^{|E|}_{+}$6 argument relies on bipartization, while for even $\mathbf w:E\to Z^{|E|}_{+}$7 the same cleanup step fails because bipartite graphs can still contain many $\mathbf w:E\to Z^{|E|}_{+}$8-cycles. The paper makes this precise by invoking the Erdős–Stone theorem: for even $\mathbf w:E\to Z^{|E|}_{+}$9,

$k$0

hence

$k$1

This does not prove complexity-theoretic hardness, but it gives an extremal obstruction to extending the odd-cycle method (Tang et al., 2018).

A vertex-deletion analogue is the Even-Cycle Transversal problem on node-weighted planar graphs. Here the input is a planar graph with vertex costs $k$2, and one seeks a minimum-weight vertex set intersecting every even cycle. The standard covering LP is

$k$3

For this problem, a primal-dual method using additional blended inequalities yields a $k$4 approximation on node-weighted planar graphs, and the same number is an upper bound on the integrality gap of the standard LP in the planar setting (Göke et al., 2021).

These two problems exemplify two distinct restriction regimes. In $k$5-cycle covering, the forbidden family is a fixed cycle length and the decision variable is an edge set. In even-cycle transversal, the forbidden family is parity-defined and the decision variable is a vertex set. In both cases, the restriction is global—every target cycle must be hit—but the methods are driven by cycle-specific structure rather than by arbitrary subgraph covering.

2. Restricted cycle bases and canonical cycle families

A second major meaning of weighted cycle-based restriction is that only certain cycles are permitted to serve as basis elements or optimization candidates. A rooted cycle basis is a cycle basis in which every basis cycle contains a specified root edge $k$6. The existence condition is exact: a rooted graph has a rooted cycle basis if and only if $k$7 belongs to the $k$8-core and the $k$9-core is $A\subseteq E$0-vertex-connected. In positively weighted graphs, the minimum-weight rooted cycle basis can still be computed in polynomial time; the paper gives a randomized algorithm with expected running time

$A\subseteq E$1

and a deterministic algorithm with running time

$A\subseteq E$2

By contrast, deciding whether a rooted graph has a fundamental rooted cycle basis is NP-complete, although the problem is fixed-parameter tractable when parameterized by clique-width (Eppstein et al., 2015).

A different restriction appears in weighted partial $A\subseteq E$3-trees. Let $A\subseteq E$4 denote the lex shortest path under the paper’s lexicographic tie-breaking rule, and let $A\subseteq E$5 be the set of lex short cycles, i.e. cycles $A\subseteq E$6 such that for every $A\subseteq E$7,

$A\subseteq E$8

For weighted partial $A\subseteq E$9-trees, the restriction to lex short cycles loses nothing: the set $G-A$0 has cardinality

$G-A$1

and therefore is a minimum cycle basis. This extends an earlier outerplanar result to the full treewidth-$G-A$2 class (Narayanaswamy et al., 2013).

The algorithmic counterpart is sharper still. For a weighted partial $G-A$3-tree $G-A$4 with nonnegative edge weights, a minimum cycle basis can be computed in linear time in implicit form and linear space: $G-A$5 and expanded explicitly in

$G-A$6

time, where $G-A$7 is the total number of edges in the basis counted with multiplicity. The key reductions are that long edges can be peeled off via shortest paths, and the remaining tight-edge graph can be decomposed into outerplanar pieces whose lex short cycles form the basis (Doerr et al., 2013).

These results show that cycle-basis restrictions are not merely heuristic. In one setting, every basis cycle is forced through a designated edge; in another, only lex-shortest-path-consistent cycles are admitted. Yet both restrictions preserve exact weighted optimization on the relevant graph classes.

3. Local inequalities, optimum cycle means, and equal-cycle-weight laws

A third line of work imposes weighted cycle-based restrictions through inequalities that every edge or every cycle must satisfy. For a connected weighted graph $G-A$8 with positive edge weights, define

$G-A$9

Then

$k$0

This is a weighted local analogue of the Bondy–Fan theorem: each edge is normalized by the heaviest cycle through that edge, and the normalized contributions cannot exceed $k$1. The theorem is sharp on a broad equality class including trees and block graphs whose clique blocks of size at least $k$2 have vertex-induced weightings (Ai et al., 19 Mar 2026).

In weighted directed graphs, optimum cycle means induce another class of algebraic restrictions. For a directed cycle $k$3,

$k$4

These extrema are computable in $k$5 time. Every cycle satisfies

$k$6

The paper derives sign-sensitive restrictions on optimal simple cycles from this wedge constraint. For example, if $k$7 is a maximum-weight simple cycle and $k$8, then

$k$9

where $\tau_k(G_w)$0 is the set of max-critical cycles. If $\tau_k(G_w)$1, the inequality reverses to an upper bound on $\tau_k(G_w)$2. Experimentally, these strict algebraic bounds are often loose—median $\tau_k(G_w)$3–$\tau_k(G_w)$4 below true values—whereas heuristic approximations based on optimum means achieve median errors of only $\tau_k(G_w)$5–$\tau_k(G_w)$6 (Dasdan, 31 Dec 2025).

A different equality law governs weightable digraphs. A digraph $\tau_k(G_w)$7 is weightable if there exists $\tau_k(G_w)$8 such that every directed cycle satisfies

$\tau_k(G_w)$9

This is equivalent to the existence of a $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$0-weighting with the same property, to the exclusion of all weak $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$1-double-cycles, and to the condition that for every triple of vertices, all directed cycles containing that triple meet it in the same cyclic order. Planar weightable digraphs are built from circular digraphs by a change-two cut construction, every weightable digraph is built from planar ones by further composition rules, and the paper derives a polynomial-time recognition algorithm with total running time $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$2 (Berger et al., 19 Jan 2026).

Taken together, these results impose restrictions at three levels: edgewise localization by $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$3, cyclewise normalization by $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$4 and $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$5, and exact equalization of all directed-cycle weights. Each converts global cycle structure into explicit algebraic conditions.

4. Network analysis and spectral design on weighted cycles

In network science, weighted cycle-based restrictions often operate as selection rules rather than hard feasibility constraints. In a weighted network $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$6, the indicator WCycle scores node $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$7 by

$$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$8

where $$\tau_{k}(G_{w})=\min\{\mathbf w^T\mathbf x: A\mathbf x\geq\mathbf 1,\mathbf x\in\mathbf \{0,~1\}\},$$9 is a cycle basis derived from a spanning tree, and the outer sum runs over basis cycles containing $A$0. Nodes not lying on any basic cycle receive score $A$1. The method therefore restricts attention to cycle-embedded nodes and to representative cycles from a cycle basis rather than all simple cycles. On six real weighted networks, WCycle selected node groups with lower overlap, higher individuation, greater spatial dispersion, and stronger WSIR spreading performance than the weighted centrality baselines considered in the paper; its experiments include networks up to $A$2 and $A$3 (Zheng et al., 15 Mar 2025).

A more geometric restriction appears in single-chord augmentation of a connected weighted cycle. If a chord joins $A$4 and $A$5, it splits the cycle into two complementary resistance arcs

$A$6

where $A$7 is the total cycle resistance. The endpoint effective resistance is then

$A$8

This exact dependence on the resistance split governs both endpoint resistance reduction and Kirchhoff-index reduction. Under bounded conductances and small resistance discrepancy, near-antipodal resistance-balanced chords are near-optimal for algebraic-connectivity improvement, and an i.i.d. bounded-conductance model yields the same conclusion with high probability. Because the chord maximizing algebraic-connectivity gain need not maximize coherence improvement, the paper formulates the design as a finite Pareto problem and introduces RBAPS and AW-RBAPS as resistance-balanced screening rules. In the reported experiments, AW-RBAPS approximates the exhaustive Pareto front with mean hypervolume ratio $A$9 while evaluating about $k$0 of admissible chords (Deng et al., 23 May 2026).

These two works use cycles differently. WCycle treats cycle participation as a ranking restriction on candidate spreaders. Weighted chord augmentation treats the cycle itself as the ambient geometry, with admissible chords screened by resistance balance. In both cases, the restriction is not arbitrary sparsification: it is derived from explicit weighted cycle structure.

5. Algebraic, cluster-theoretic, and pattern-theoretic restrictions

Weighted cycle-based restrictions also appear as admissibility conditions in algebraic and combinatorial models. On a weave of general Dynkin type, a weighted chain is an oriented path crossing the weave transversely, with the local rule that if the path crosses an edge colored $k$1, the adjacent weights must be $k$2 and $k$3. A weighted cycle is a collection of non-intersecting weighted chains, and every generic interior endpoint must satisfy the balancing condition that the incident nearby weights sum to $k$4. In the simply-laced case, weighted cycles carry a skew-symmetric intersection pairing; in general Dynkin type, the paper defines a skew-symmetrizable pairing on Y-cycles. The weighted-cycle algebra is a Laurent polynomial algebra

$k$5

and in the simply-laced case it admits a quantization governed by the intersection pairing. Merodromies along weighted cycles become functions on decorated flag moduli spaces, and mutations of weighted cycles are compatible with cluster mutations (Weng, 11 Mar 2025).

A different conversion principle arises in permutation combinatorics. An adjacent $k$6-cycle is a cycle of the form

$k$7

Under Foata’s fundamental transformation, the number of adjacent $k$8-cycles in a permutation maps to the sum of occurrences of two mesh patterns $k$9 and $\mathbf w:E\to Z^{|E|}_{+}$00. Consequently, a cycle restriction stated in terms of adjacent $\mathbf w:E\to Z^{|E|}_{+}$01-cycles becomes a mesh-pattern restriction in one-line notation. In particular, permutations with no adjacent $\mathbf w:E\to Z^{|E|}_{+}$02-cycles correspond to permutations avoiding both $\mathbf w:E\to Z^{|E|}_{+}$03 and $\mathbf w:E\to Z^{|E|}_{+}$04 (Claesson et al., 2023).

In random generation problems for the symmetric group, cycle-type restrictions acquire an explicitly weighted form. If $\mathbf w:E\to Z^{|E|}_{+}$05 and $\mathbf w:E\to Z^{|E|}_{+}$06 denote the numbers of $\mathbf w:E\to Z^{|E|}_{+}$07-cycles in two conjugacy classes, then, under the hypothesis $\mathbf w:E\to Z^{|E|}_{+}$08, the probability that two random representatives generate at least $\mathbf w:E\to Z^{|E|}_{+}$09 tends to $\mathbf w:E\to Z^{|E|}_{+}$10 if and only if

$\mathbf w:E\to Z^{|E|}_{+}$11

and is bounded away from $\mathbf w:E\to Z^{|E|}_{+}$12 if and only if the same product is $\mathbf w:E\to Z^{|E|}_{+}$13 and

$\mathbf w:E\to Z^{|E|}_{+}$14

Here fixed points enter with weight $\mathbf w:E\to Z^{|E|}_{+}$15, while transpositions enter through their square root, so the restriction is expressed by a weighted short-cycle statistic rather than by a hard exclusion of cycle lengths (Eberhard et al., 2019).

These examples show that weighted cycle-based restrictions need not be graph-theoretic in a narrow sense. They can instead specify admissible local weight transport on topological curves, translate cycle data into pattern occurrences, or compress cycle-type information into weighted low-length statistics.

6. Scope, limitations, and recurring structural themes

Across these works, weighted cycle-based restrictions fall into a small number of recurrent forms. One form is forbidden-family covering: remove minimum total weight so that no target cycle remains, as in $\mathbf w:E\to Z^{|E|}_{+}$16-cycle covering and even-cycle transversal. Another is admissible-basis restriction: require every basis cycle to pass through a root edge or require the entire minimum cycle basis to lie inside a canonical family such as lex short cycles. A third is cycle-local normalization, where edge or cycle weights are constrained by quantities such as $\mathbf w:E\to Z^{|E|}_{+}$17, $\mathbf w:E\to Z^{|E|}_{+}$18, and $\mathbf w:E\to Z^{|E|}_{+}$19. A fourth is selection by cycle embedding, exemplified by WCycle and resistance-balanced chord screening. A fifth is algebraic admissibility, where only weighted cycles satisfying Weyl-reflection, balancing, and mutation-compatibility rules are meaningful.

The limitations are equally structural. Odd-$\mathbf w:E\to Z^{|E|}_{+}$20 covering improves because bipartization destroys odd cycles, but the same argument collapses for even $\mathbf w:E\to Z^{|E|}_{+}$21 (Tang et al., 2018). Rooted cycle bases remain polynomial-time optimizable, but the stronger fundamental rooted restriction is NP-complete (Eppstein et al., 2015). Optimum-cycle-mean bounds are exact algebraically yet often loose as numerical bounds on longest simple cycles (Dasdan, 31 Dec 2025). WCycle excludes all nodes outside the chosen cycle basis by construction (Zheng et al., 15 Mar 2025). Quantization of weighted cycles on weaves is developed only in the simply-laced case because that is where the relevant pairing is skew-symmetric rather than merely skew-symmetrizable (Weng, 11 Mar 2025). Weighted chord screening is formally justified under bounded conductances and small discrepancy, although the experiments indicate effectiveness beyond that regime (Deng et al., 23 May 2026).

What persists across settings is that cycles are treated as structured carriers of constraints rather than as undifferentiated subgraphs. The restriction may be combinatorial, algebraic, geometric, probabilistic, or spectral, but in each case the weighted data are meaningful only relative to cycle organization: cycle length, parity, basis representation, local heaviest-cycle context, resistance split, or mutation-compatible weight transport. That shared principle gives the topic its coherence despite the diversity of domains.