Cycle Walk: Models, Dynamics, and Applications

• Cycle Walk is a family of models that use cyclic structures—from finite cycle graphs in ARW and quantum walks to random k-cycle processes in symmetric groups and MCMC in redistricting.
• The models employ varied techniques such as cyclotomic polynomial factorization, character ratio analysis, and coupling arguments to reveal phase transitions and mixing behaviors.
• Key findings include fixation time bounds (linear to exponential regimes), unusual periodicity in Hadamard quantum walks, and cutoff phenomena in permutation-based cycles.

Searching arXiv for recent and relevant uses of “cycle walk” and closely related cycle-based walk models. The literature surveyed here suggests that “cycle walk” is not a single standardized process. In arXiv usage, the expression is associated with several distinct constructions: stochastic particle dynamics on the finite cycle $\mathbb{Z}/n\mathbb{Z}$, coined and Szegedy-type quantum walks on cycle graphs $C_N$, random walks on $S_n$ generated by $k$-cycles, and a recent Markov chain on spanning forests that creates and breaks cycles to sample balanced graph partitions (Basu et al., 2017, Konno et al., 2015, Hough, 2016, DeFord et al., 10 Sep 2025). This suggests a unifying motif—dynamics organized by cyclic geometry or by operations on cycles—while the underlying state spaces, observables, and asymptotic regimes remain field-specific.

1. Terminological scope

In graph-based probability and quantum information, a cycle walk typically means a walk whose position space is a finite cycle graph. The cycle may be written as $C_N$, $\mathbb{Z}_n$, or $\mathbb{Z}/n\mathbb{Z}$, with nearest-neighbor motion and periodic boundary conditions. This is the setting for Activated Random Walk on a cycle, stochastic sandpile on $\mathbb{Z}_n$, discrete-time quantum walks on $C_N$, and communication protocols built from $k$-cycle recurrence (Basu et al., 2017, Melchionna, 2021, Konno et al., 2015, Panda et al., 2022).

In algebraic probability, a cycle walk means something different: a random walk on the symmetric group $C_N$0 whose steps are sampled from the conjugacy class of $C_N$1-cycles. Here the relevant geometry is not a graph cycle but permutation cycle structure, and the principal observables are total variation distance, character ratios, and cycle-count statistics such as $C_N$2, the number of $C_N$3-cycles of $C_N$4 (Hough, 2016, Arcona, 16 Dec 2025).

A newer redistricting usage is more literal: the Cycle Walk is a Markov chain on spanning forests in which one adds edges to create a cycle and then removes edges from that cycle so as to return to a forest while respecting population balance. In this context, “cycle” refers to an intermediate combinatorial object used to generate proposals, not to the ambient state space itself (DeFord et al., 10 Sep 2025).

2. Stochastic particle systems on the finite cycle

For Activated Random Walk (ARW) on $C_N$5, the initial condition is $C_N$6, all particles are initially active, each active particle performs a simple symmetric random walk at rate $C_N$7, and a solitary active particle falls asleep at rate $C_N$8. Sleepy particles are reactivated when an active particle lands on their site. The paper analyzes the fixation time $C_N$9, defined as the total number of instructions used until stabilization in the Diaconis–Fulton toppling picture. Two asymptotic regimes are proved. If

$S_n$0

then

$S_n$1

for some constants $S_n$2. For every fixed $S_n$3, there exists $S_n$4 such that if $S_n$5, then

$S_n$6

Thus the finite cycle exhibits a linear-up-to-polylogarithmic fixation regime and an exponentially slow fixation regime, reflecting the fixation/non-fixation phase transition of the corresponding infinite system on $S_n$7 (Basu et al., 2017).

The same cycle geometry appears in the stochastic sandpile on $S_n$8 with exactly $S_n$9 particles. A site is unstable when $k$0, and a toppling sends two particles through independent lazy symmetric random walk steps: stay put with probability $k$1, move clockwise with probability $k$2, or move counterclockwise with probability $k$3. The stabilization time is

$k$4

where $k$5 is the sandpile odometer. A formal coupling to ARW is established under the parameter matching

$k$6

with

$k$7

If

$k$8

then there exists $k$9 independent of $C_N$0 such that

$C_N$1

This transfers a cycle-ARW stabilization estimate to the sandpile model via a quotient/coupling construction (Melchionna, 2021).

A recurring misconception is that finite-cycle ARW can exhibit true non-fixation. On the finite graph $C_N$2, unless there are more than $C_N$3 particles, the process fixates almost surely in finite time; what survives from the infinite-volume phase transition is the asymptotic scale of the fixation time rather than genuine persistent activity (Basu et al., 2017).

3. Quantum walks on cycle graphs: periodicity, memory, and exceptional behavior

For the Hadamard discrete-time quantum walk on the cycle $C_N$4, periodicity is defined by

$C_N$5

with $C_N$6 if no such $C_N$7 exists. The complete classification is

$C_N$8

The proof combines a path-counting parity argument, reduction modulo $C_N$9 via powers of the adjacency matrix, and cyclotomic polynomial factorization of the characteristic polynomial. Exact global revival is therefore highly exceptional for the Hadamard walk on a cycle (Konno et al., 2015).

The history-dependent quantum walk on a cycle with recycled coins introduces two coin registers and a memory parameter $\mathbb{Z}_n$0. The one-step evolution is

$\mathbb{Z}_n$1

and the main long-time observable is the time-averaged distribution

$\mathbb{Z}_n$2

A key symmetry theorem states that if

$\mathbb{Z}_n$3

then

$\mathbb{Z}_n$4

and likewise for $\mathbb{Z}_n$5. The numerical analysis further reports that for $\mathbb{Z}_n$6 not divisible by $\mathbb{Z}_n$7, non-uniformity was observed only for $\mathbb{Z}_n$8, while for $\mathbb{Z}_n$9 divisible by $\mathbb{Z}/n\mathbb{Z}$0, non-uniformity was observed only for $\mathbb{Z}/n\mathbb{Z}$1. The same paper proves that the separate quantum walk with memory on cycles is exactly equivalent to the recycled-coin walk with $\mathbb{Z}/n\mathbb{Z}$2, up to a permutation of the initial coin basis (Krawec, 2014).

A different phenomenon occurs in Szegedy-type quantum walk search on the cycle. For the one-dimensional periodic lattice or cycle with any arrangement of marked vertices, the evolution reduces to sign flips only, leaving the vertex-measurement distribution uniform for all time. If $\mathbb{Z}/n\mathbb{Z}$3 vertices are marked, the success probability remains $\mathbb{Z}/n\mathbb{Z}$4, so the expected number of repetitions is $\mathbb{Z}/n\mathbb{Z}$5. The paper argues that comparing this directly with classical hitting time is misleading, and that mixing time is the more meaningful benchmark in this exceptional configuration (Wong et al., 2016).

These results jointly distinguish three regimes on cycle graphs: rare exact periodicity, parameter-sensitive long-time averaging in memoryful models, and search dynamics that fail to amplify marked vertices.

4. Experimental and communication realizations of cycle quantum walks

Cycle quantum walks have also been used as experimental primitives. On IBM superconducting hardware, an 8-node, 8-step discrete-time quantum walk on ibmq_quito uses 3 position qubits and 1 coin qubit, while a 4-node, 4-step walk uses 2 position qubits and 1 coin qubit. The larger circuit contains Toffoli and C4-X gates, which expand heavily under transpilation to IBM’s native basis $\mathbb{Z}/n\mathbb{Z}$6: a Toffoli becomes about 18 gates and a C4-NOT about 34 gates. This produces a marked fidelity gap. For the 8-node walk, average Hellinger fidelity is about $\mathbb{Z}/n\mathbb{Z}$7 from step 1 onward, whereas for the 4-node walk fidelity remains above $\mathbb{Z}/n\mathbb{Z}$8 for all steps. A custom noise model calibrated to ibmq_santiago further suggests that a 16-node, 16-step cycle DTQW would require noise to be reduced by approximately $\mathbb{Z}/n\mathbb{Z}$9 to achieve consistent high fidelity (Wadhia et al., 2023).

On a four-qubit NMR processor, a two-step coined quantum random walk on a 4-cycle was used to transfer an arbitrary single-qubit state from Alice to Bob. The 4-cycle is encoded by two arena qubits with vertex map

$\mathbb{Z}_n$0

After the walk and Bob’s controlled recovery operations, process fidelities were reported as

$\mathbb{Z}_n$1

Using the entanglement witness

$\mathbb{Z}_n$2

the experimentally reconstructed states for $\mathbb{Z}_n$3 and $\mathbb{Z}_n$4 yielded negative witness values, confirming genuine quadripartite entanglement (Singh et al., 2023).

A further communication-oriented construction uses recurrence in $\mathbb{Z}_n$5-cycle DTQW with a photon whose polarization is the coin and orbital angular momentum (OAM) is the position space. For specific pairs $\mathbb{Z}_n$6, the walk recurs after a fixed $\mathbb{Z}_n$7; the paper lists $\mathbb{Z}_n$8, $\mathbb{Z}_n$9, $C_N$0, $C_N$1, $C_N$2, and $C_N$3. The protocol uses OAM shifts as message encoding, exploits the commutation of the encoding with the walk, and analyzes intercept-resend security and the effects of amplitude damping and depolarizing noise on recurrence and mutual information between polarization and OAM (Panda et al., 2022).

A plausible implication is that the cycle graph serves not only as a mathematically tractable configuration space but also as a hardware-efficient testbed for conditional shift operators, recurrence, and walk-generated entanglement.

5. Cycle-generated random walks on the symmetric group

In permutation-group probability, the random $C_N$4-cycle walk on $C_N$5 starts at the identity and multiplies by independent uniformly random $C_N$6-cycles. Because the step distribution lies in a conjugacy class, the walk is class-invariant, and because every $C_N$7-cycle has fixed parity, the equilibrium measure is the uniform measure on the appropriate parity class, denoted $C_N$8. The main theorem states that for $C_N$9, the walk exhibits cutoff in total variation at

$k$0

More precisely, if

$k$1

then

$k$2

whereas if

$k$3

then

$k$4

The proof relies on asymptotic estimates of symmetric-group characters evaluated at cycles via a Frobenius contour integral formula (Hough, 2016).

A related chain starts with one random $k$5-cycle and then applies random transpositions. For fixed $k$6, the convergence scale is linear in $k$7. The paper shows that after

$k$8

steps, the law is close to the stationary parity-coset distribution $k$9, and the upper bound uses estimates for normalized characters of transpositions. For $C_N$00, the defining representation character $C_N$01 is analyzed in detail to obtain lower bounds on total variation distance (Özdemir, 2017).

Representation-theoretic analysis has since been extended from mixing in total variation to cycle statistics themselves. For the random $C_N$02-cycle walk, with

$C_N$03

later work derives a stable character decomposition of $C_N$04 and proves Poisson limits by the method of moments. For fixed $C_N$05, after $C_N$06 random $C_N$07-cycles,

$C_N$08

while fixed points require the longer scale $C_N$09 and converge to $C_N$10 (Arcona, 16 Dec 2025).

One common misunderstanding is to treat all “cycle walks” on $C_N$11 as variants of random transpositions. The cited results show a sharper taxonomy: sparse random $C_N$12-cycles mix at $C_N$13, an initial large cycle followed by transpositions mixes on order $C_N$14, and specific cycle-count observables can equilibrate on $C_N$15 even when fixed points still require $C_N$16.

6. Cycle Walk as a spanning-forest Markov chain for redistricting

In redistricting, Cycle Walk denotes a Markov chain on spanning forests

$C_N$17

whose connected components encode districts. A partition $C_N$18 is given probability

$C_N$19

and each forest induces a partition $C_N$20. The paper introduces two lifted measures,

$C_N$21

and a one-parameter family

$C_N$22

The chain itself is a mixture

$C_N$23

Its defining operation is to add an edge or pair of edges to create a cycle and then remove an edge or pair of edges from that cycle so as to recover a forest while respecting population balance (DeFord et al., 10 Sep 2025).

The 1-tree Cycle Walk modifies one tree internally and preserves the partition: $C_N$24 The 2-tree Cycle Walk chooses two adjacent trees, adds two boundary edges to create a unique cycle in the merged graph, and removes a pair of edges from that cycle so that the graph again splits into two trees satisfying the population constraints. The Metropolis–Hastings acceptance probability has the general form

$C_N$25

and the paper gives explicit reverse-probability formulas for both the 1-tree and 2-tree moves (DeFord et al., 10 Sep 2025).

Numerically, the method is validated on a $C_N$26 grid, compared with Metropolized Forest RECOM on the North Carolina precinct graph, and studied as $C_N$27 varies in $C_N$28. The experiments report that adding internal 1-tree mixing improves convergence and that a ratio around $C_N$29 internal-to-2-tree proposals appears effective. The same study is explicit about limitations: the stationary distribution of the un-Metropolized Cycle Walk is not theoretically understood in a simple way, and no mixing proofs are provided (DeFord et al., 10 Sep 2025).

This is the most specialized current use of the capitalized name Cycle Walk. Unlike cycle-graph walks or permutation walks, it is a forest-based proposal family whose “walk” occurs in the space of balanced connected partitions and whose “cycle” is an intermediate combinatorial device for reversible proposal construction.

7. Recurring themes and major distinctions

Across these literatures, several recurrent analytical themes appear. Abelian or order-independence principles govern stabilization in ARW and stochastic sandpile on the cycle (Basu et al., 2017, Melchionna, 2021). Spectral and algebraic obstructions determine periodicity and mixing in quantum and permutation settings, through cyclotomic factors, Fourier decomposition, and character ratios (Konno et al., 2015, Hough, 2016). Recurrence is sometimes a resource, as in communication protocols on $C_N$30-cycle DTQW, and sometimes a rare exception, as in exact Hadamard periodicity (Panda et al., 2022, Konno et al., 2015). Local cycle surgery serves as a reversible proposal mechanism in redistricting rather than as a physical trajectory (DeFord et al., 10 Sep 2025).

The comparison also clarifies what should not be conflated. A cycle walk on $C_N$31 is a walk on a graph with periodic boundary conditions; a random $C_N$32-cycle walk on $C_N$33 is a walk on a nonabelian group generated by a conjugacy class; and the redistricting Cycle Walk is neither of these, but a lifted MCMC on spanning forests. This suggests that the expression is best treated as a family resemblance term rather than a universal definition.

In that broader sense, “cycle walk” names a cluster of models in which cyclic structure is operationally central: it may govern particle fixation times, exact or approximate recurrence, representation-theoretic mixing, or Metropolizable proposal design. The technical content, however, is specific to the ambient category—interacting particle systems, quantum information, algebraic random walks, or combinatorial sampling—and the relevant invariants must be read accordingly.