Cycles in Mathematics and Applications

Updated 4 July 2026

Cycles are closed or orbit-like structures that appear in diverse fields, defined variously as simple circuits, homological chains, or dynamic rotations.
They serve as fundamental tools in studying extremal graph theory, cycle decompositions, and efficient coding via interlinked cycles and universal cycles.
Cycles enable analysis across disciplines—from capturing feedback loops in ecological models to characterizing nonconvergent dynamics in game theory and solar activity.

1. Graph-theoretic meanings of cycles

In finite graph theory, a simple cycle is a closed walk with no repeated vertices except the start and end. In a digraph $G=(V,E)$ , a directed cycle is a sequence

$v_0,v_1,\dots,v_{k-1},v_k=v_0$

such that each $(v_i,v_{i+1})\in E$ and the vertices $v_0,\dots,v_{k-1}$ are distinct (Knierim et al., 2019). A second, broader notion is the $1$-cycle modulo $2$: a set $C$ of edges such that every vertex is contained in an even number of edges from $C$ . In this sense, a usual graph-theoretic cycle is a $2$0-cycle, but not conversely; sums of $2$1-cycles under symmetric difference remain $2$2-cycles, so they form a vector space over $2$3 (Alkin et al., 2023).

Several graph-theoretic refinements distinguish cycles by metric or induced-subgraph behavior. A cycle is chordless if it has no chord, equivalently if it is an induced cycle; the circumference $2$4 is the length of a longest cycle, and the induced circumference $2$5 is the length of a longest chordless cycle (Hu et al., 2024). A cycle is convex when every shortest path in the ambient graph between vertices of the cycle stays entirely inside the cycle; convexity is therefore stronger than being induced (Azarija et al., 2012). These distinctions matter because different extremal problems are controlled by different versions of “cycle.”

The same homological broadening persists in higher dimension. For a $2$6-dimensional hypergraph, a $2$7-cycle is a set of $2$8-faces such that every $2$9-element subset is contained in an even number of those faces (Alkin et al., 2023). This places ordinary graph cycles inside a larger simplicial pattern in which parity, rather than simple closedness, is fundamental.

2. Extremal, decomposition, and length phenomena

A central line of work studies how many cycles must exist, how long they must be, and how economically they can decompose a graph. For Eulerian digraphs, Hajós’ undirected conjecture motivated the directed conjecture of Bienia–Meyniel, Dean, and Bollobás–Scott that every Eulerian digraph on $G=(V,E)$ 0 vertices can be decomposed into $G=(V,E)$ 1 directed cycles. The best bound in the cited paper is that every Eulerian digraph on $G=(V,E)$ 2 vertices with maximum degree $G=(V,E)$ 3 can be decomposed into $G=(V,E)$ 4 edge-disjoint cycles, and every edge-weighted digraph with out-weight at least $G=(V,E)$ 5 at each vertex contains a cycle of weight $G=(V,E)$ 6 (Knierim et al., 2019). The proof route is explicit: heavy cycles in inverse-degree or general edge-weightings imply small cycle decompositions.

For convex cycles, the extremal problem is different. If $G=(V,E)$ 7 is a simple graph of order $G=(V,E)$ 8, size $G=(V,E)$ 9, and girth $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 0, then the number $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 1 of convex cycles satisfies

$v_0,v_1,\dots,v_{k-1},v_k=v_0$ 2

with equality if and only if $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 3 is an even cycle or a Moore graph (Azarija et al., 2012). In odd-girth graphs, this becomes a characterization of Moore graphs via the number of shortest cycles.

The comparison between longest cycles and longest chordless cycles produces another extremal gap. Thomassen’s chord conjecture asserts that every longest cycle in a $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 4-connected graph has a chord. Harvey’s stronger conjecture, in the $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 5 case emphasized by the paper, states that every $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 6-connected graph $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 7 with minimum degree at least $v_0,v_1,\dots,v_{k-1},v_k=v_0$ 8 should satisfy

$v_0,v_1,\dots,v_{k-1},v_k=v_0$ 9

The paper proves that wheels are the unique Hamiltonian graphs for which $(v_i,v_{i+1})\in E$ 0 and $(v_i,v_{i+1})\in E$ 1 differ by exactly one, and it proves Harvey’s conjecture for $(v_i,v_{i+1})\in E$ 2-holed graphs and for graphs with small induced circumference (Hu et al., 2024).

Cycle lengths constrained modulo an integer also force extremal structure. If $(v_i,v_{i+1})\in E$ 3 is $(v_i,v_{i+1})\in E$ 4-colorable but $(v_i,v_{i+1})\in E$ 5 is not, then for every $(v_i,v_{i+1})\in E$ 6, the edge $(v_i,v_{i+1})\in E$ 7 lies in at least $(v_i,v_{i+1})\in E$ 8 cycles of length $(v_i,v_{i+1})\in E$ 9, while $v_0,\dots,v_{k-1}$ 0 contains at least $v_0,\dots,v_{k-1}$ 1 cycles of length $v_0,\dots,v_{k-1}$ 2 (Moore et al., 2019). In the circular-coloring setting, analogous congruence classes are indexed by the modular inverse $v_0,\dots,v_{k-1}$ 3 of $v_0,\dots,v_{k-1}$ 4 modulo $v_0,\dots,v_{k-1}$ 5, and at least one cycle of length $v_0,\dots,v_{k-1}$ 6 for some $v_0,\dots,v_{k-1}$ 7 must pass through the critical edge (Moore et al., 2019).

3. Cycles as generators, covers, and compact encodings

In graph homology over $v_0,\dots,v_{k-1}$ 8, a connected graph with $v_0,\dots,v_{k-1}$ 9 vertices and $1$0 edges has exactly $1$1 $1$2-cycles, so $1$3. A spanning tree gives a fundamental cycle basis: each non-tree edge determines one unique simple cycle, and every $1$4-cycle is a unique sum of these basis elements (Alkin et al., 2023). In complete and complete bipartite graphs, special short cycles generate the full cycle space. Any $1$5-cycle in $1$6 is a sum of triangles, while any $1$7-cycle in $1$8 is a sum of $1$9-cycles; the corresponding relations are generated by the standard $2$0-vertex and $2$1 parity identities (Alkin et al., 2023).

Cycles also function as coding primitives. In index coding, directed cycles and cliques had long been used to obtain savings, but the paper on interlinked cycles introduces the interlinked-cycle (IC) structure as a common generalization. A $2$2-IC structure is built around a set $2$3 of inner vertices for which each ordered pair of distinct inner vertices has a unique directed path whose internal vertices are non-inner, and the digraph contains no directed cycle with exactly one inner vertex. The corresponding interlinked-cycle-cover (ICC) scheme assigns one coded symbol to all inner vertices and one coded symbol to each non-inner vertex, producing scalar linear code length

$2$4

for a $2$5-IC structure, with savings $2$6; cycles and cliques appear as special cases, and the scheme is optimal for an infinite class of digraphs (Thapa et al., 2016).

Another use of cycles is as compact enumerative devices. A universal cycle is a cyclic sequence that contains every length-$2$7 object from a family exactly once as a consecutive block; De Bruijn cycles are the standard example. Graph universal cycles transport this idea from words to cyclically ordered graphs whose windows are induced ordered subgraphs. The paper introduces graph universal partial cycles, where “diamond edges” encode “do not know” adjacency, and constructs such objects for labeled graphs, threshold graphs, and permutation graphs; for permutation graphs it also proves the existence of graph $2$8-overlap cycles, and for unlabeled graphs it proves the existence of an $2$9-fold gucycle for some $C$ 0 (Kirsch et al., 2022).

Special graph families can be almost edge-pancyclic in a strong sense. In the enhanced hypercube $C$ 1, every edge lies on a cycle of every even length from $C$ 2 to $C$ 3; if $C$ 4 is even, every edge also lies on a cycle of every odd length from $C$ 5 to $C$ 6, and some special edges lie on a shortest odd cycle of length $C$ 7 (Ma, 2015).

4. Cycles in games and strategic interaction

In repeated game theory, “cycles” need not be graph-theoretic at all. In Rock–Paper–Scissors experiments with $C$ 8 subjects, the social state at round $C$ 9 is the population-frequency vector

$C$ 0

A cycle is a nonconvergent trajectory in this discrete simplex that orbits around the mixed-strategy Nash equilibrium $C$ 1 rather than settling there (Xu et al., 2012). To measure such rotation, the paper uses the Cycle Rotation Index

$C$ 2

where $C$ 3 and $C$ 4 count oriented crossings of a Poincaré section. In a simultaneous-move, pure-strategy, random-matching, low-information RPS environment, the reported mean $C$ 5 is significantly positive, the two-sided $C$ 6-test for $C$ 7 yields $C$ 8, and the paper concludes that cycles do not dissipate under simultaneous choice (Xu et al., 2012).

The Game of Cycles studies a very different object: edge orientations on a planar graph under a sink-source rule. A cycle cell is a bounded face whose boundary is oriented consistently clockwise or counterclockwise; creating the first such cell wins immediately. On cactus graphs, the paper analyzes outcomes by reflection symmetry, self-involutive edges, and a modified mirror-reverse strategy. For the triangle-free cactus graphs satisfying its symmetry hypotheses, the decisive invariant is the parity of the total number of self-involutive edges: if this number is even, Player 2 has a winning strategy; if odd, Player 1 wins by first marking one self-involutive edge and then following the modified mirror-reverse strategy (Adefiyiju et al., 2023).

5. Cycles in physical and ecological dynamics

In solar and dynamo theory, cycles refer to large-scale magnetic reversals and their modulation. The Sun exhibits an approximately $C$ 9-year sunspot cycle and an approximately $2$00-year magnetic polarity cycle. In mean-field magnetohydrodynamics, the mean field obeys

$2$01

with

$2$02

The cited simulations compare scale-separation ratios $2$03 and $2$04: the larger ratio yields a fairly regular quasi-sinusoidal cycle with a sharp spectral peak, while the smaller ratio yields erratic cycles with stronger amplitude and phase fluctuations. The paper interprets this as stochastic modulation of the $2$05-effect due to limited scale separation, and it connects cyclic large-scale fields to near-surface sunspot formation via negative effective magnetic pressure instability (Brandenburg et al., 2011).

In mathematical ecology, cycles appear as algebraic feedback loops in the characteristic polynomial of a community matrix. For a Lotka–Volterra equilibrium, the Jacobian or community matrix $2$06 determines local stability, and a cycle is a product

$2$07

corresponding to a closed feedback loop. The paper defines destabilizing or “tipping” cycle sets $2$08 inside the coefficients of the characteristic polynomial and studies how sign structure changes their abundance in predator–prey, mutualistic, and competitive systems. For competitive systems, exactly half the cycles in each coefficient are destabilizing, so $2$09 for all $2$10; for mutualistic systems, $2$11; for predator–prey systems, $2$12, often with some coefficients having no negative cycles at all (Thorne, 2021).

6. Geometric, analytic, and scattering-theoretic cycles

In Lie sphere geometry, a cycle in $2$13 is a point, an oriented sphere of codimension $2$14, or an oriented hyperplane of codimension $2$15. Such objects are represented as points on the Lie quadric

$2$16

where $2$17 is the Lie product on $2$18 (Zlobec et al., 2013). The Lie product encodes geometry: for proper cycles $2$19, the condition $2$20 means oriented contact; projective intersections $2$21 produce families corresponding to subcycles, cones, and tori; and determinant-based discriminants recover radii, angles, and tangential distances (Zlobec et al., 2013).

Complex geometry uses a different notion. An analytic $2$22-cycle on a complex space $2$23 is a locally finite formal sum

$2$24

with irreducible analytic subsets $2$25 of dimension $2$26. A cycle is of finite type when it has only finitely many irreducible components. The monograph develops the topology of $2$27, defines $2$28-analytic families of finite type cycles via quasi-proper graphs, and uses them to formulate strongly quasi-proper maps, geometric $2$29-flattening, existence theorems for meromorphic quotients, and a generalization of Stein factorization (Barlet et al., 2023). Here “cycle” is not a closed path but a geometric chain object with multiplicities.

In the CHY formalism for scattering amplitudes, cycles return to graph theory but in a highly structured way. Every $2$30-regular graph on $2$31 vertices determines a vector in $2$32, and biadjoint scalar partial amplitudes are inner products of vectors associated with a pair of cycles. To reduce general $2$33-regular CHY integrals to such amplitudes, the paper defines a cycle $2$34 to be compatible with a $2$35-regular graph $2$36 when $2$37 admits a Hamiltonian decomposition. It proves that every $2$38-regular graph on $2$39 vertices has at least $2$40 compatible cycles, and at least $2$41 if all cycles of $2$42 have even length (Cachazo et al., 2019).

7. Open problems and recurring themes

Several cited works leave major conjectures unresolved. In digraph theory, the Bollobás–Scott conjecture asks for an $2$43 cycle decomposition of every Eulerian digraph, while stronger weighted conjectures ask for cycle weights of order $2$44 under balanced out-weight conditions (Knierim et al., 2019). In graph theory, Thomassen’s chord conjecture and Harvey’s conjecture remain open in general despite the progress for $2$45-holed graphs and graphs of small induced circumference (Hu et al., 2024). In the study of convex cycles, the paper suggests generalized Moore graphs as a natural class for near-extremal behavior (Azarija et al., 2012).

Open directions also arise in dynamic and combinatorial settings. The RPS paper asks why some earlier experiments failed to detect significant cycles, why HSNG did not find cycles in “Bad RPS,” and whether mixed-strategy $2$46 games admit systematic cycle detection (Xu et al., 2012). The cactus-game paper asks whether modified mirror-reverse arguments extend beyond triangle-free cacti and to broader symmetry classes (Adefiyiju et al., 2023). In solar dynamo theory, unresolved issues include equatorward migration, the interaction of NEMPI with suppression of turbulent heat transport, and the role of stochastic $2$47 in grand minima (Brandenburg et al., 2011). For graph universal cycles, existence of ordinary gucycles for unlabeled graphs on all $2$48 remains conjectural, even though $2$49-fold gucycles are known to exist (Kirsch et al., 2022). For CHY integrals, the paper isolates the problem of finding enough compatible cycles that also form a basis in $2$50 (Cachazo et al., 2019).

Taken together, these literatures show that “cycles” serve at least four distinct technical roles. They are parity objects in homological graph theory, extremal witnesses in structural combinatorics, orbit-like signatures of nonconvergent dynamics, and chain-like or family-like objects in geometry and analysis. This suggests that the persistence of the term across fields is not accidental: in each setting, a cycle records a closed constraint, whether combinatorial, dynamical, geometric, or algebraic.