Scramble Number: Invariants & Mixing Dynamics

• Scramble number is a measure that quantifies a system’s resistance to resolution, defined via connected vertex sets in graphs or maximum cycle lengths in permutation scramblings.
• It bridges combinatorial invariants with operational measures, linking treewidth and divisorial gonality in graph theory with mixing times in puzzles like the Rubik’s cube.
• Computing the scramble number is NP-hard in graphs, yet its related concepts offer practical insights for analyzing randomization and convergence in both deterministic and stochastic systems.

“Scramble number” appears in contemporary research in more than one technical sense. Its primary established meaning is a graph invariant defined from collections of connected vertex sets called scrambles, introduced to study chip-firing and divisorial gonality. In other settings, the same expression is naturally interpreted as a quantitative measure of scrambling complexity, such as the longest cycle of a permutation induced by a hardware scrambling map or the mixing time of a random walk intended to randomize a Rubik’s cube (Harp et al., 2020, Rangineni, 2011, Qu et al., 2024).

1. Terminological scope

In graph theory, scramble number is a formal invariant of a connected finite multigraph. It was introduced as a refinement of bramble-based lower bounds and satisfies

$\operatorname{tw}(G)\le \operatorname{sn}(G)\le \gon(G),$

placing it between treewidth and divisorial gonality (Harp et al., 2020).

The same phrase is not used uniformly outside graph theory. In Kak’s mesh-array work on matrix-position scrambling, the paper defines the “period” of the scrambling transformation as the maximum cycle length of the associated permutation; this maximum cycle length is the natural scramble number in that setting, although the phrase itself is not explicitly used in the paper (Rangineni, 2011). In the Rubik’s-cube literature, a plausible operational interpretation is the mixing time of a Markov chain modeling random face turns, that is, the number of random moves needed before the distribution of cube states is close to uniform in total variation distance (Qu et al., 2024).

This plurality of usage matters because the underlying mathematical objects differ sharply. In the graph-theoretic sense, scramble number is a max–min invariant over connected subsets and cuts. In permutation-based or stochastic settings, it is naturally attached to orbit length or convergence time. The common theme is not a shared definition but a shared role: each quantity measures how resistant a system is to being “resolved” by a small obstruction, whether that obstruction is a hitting set, a cut, a short cycle, or insufficient randomization time.

2. Graph-theoretic definition

Let $G=(V,E)$ be a connected finite multigraph. A scramble $\mathcal{S}$ on $G$ is a finite collection of subsets

$$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$

called eggs, such that each induced subgraph $G[E_i]$ is connected. Two auxiliary parameters are attached to $\mathcal{S}$ (Cenek et al., 2022).

A hitting set is a vertex set $H\subseteq V(G)$ meeting every egg, and the minimum size of such a set is

$h(\mathcal{S}).$

An egg-cut is an edge set $T\subseteq E(G)$ such that $G=(V,E)$0 has at least two connected components that each contain an egg; the minimum size of an egg-cut is

$G=(V,E)$1

The order of the scramble is then

$G=(V,E)$2

and the scramble number of $G=(V,E)$3 is

$G=(V,E)$4

This generalizes the bramble number. If the eggs are required to pairwise touch, one recovers brambles and treewidth; in that sense, scramble number extends the bramble formalism while incorporating edge-separation data directly into the invariant (Cenek et al., 2022).

The connection with chip-firing is structural rather than superficial. A divisor on $G=(V,E)$5 is an integer assignment to vertices, chip-firing moves preserve degree, and divisorial gonality $G=(V,E)$6 is the minimum degree of a divisor of positive rank. Scramble number furnishes a combinatorial obstruction to low-degree positive-rank divisors, yielding the inequality

$G=(V,E)$7

while bramble theory gives $G=(V,E)$8 (Harp et al., 2020).

3. Structure, bounds, and complexity

The central tree-cut analogue of scramble number is screewidth, defined from a variant of tree-cut decompositions whose width is

$G=(V,E)$9

For every graph,

$\mathcal{S}$0

so screewidth provides a systematic upper-bound framework for scramble number (Cenek et al., 2022).

Several basic structural properties are now standard. Trees are exactly the graphs of scramble number $\mathcal{S}$1. Scramble number is monotone under subgraphs, invariant under subdividing edges and smoothing 2-valent vertices, and therefore monotone under topological minors. It is not minor monotone in general (Eagleton et al., 2022, Cenek et al., 2022). The non-minor-monotonicity is one of the key points distinguishing scramble number from treewidth.

The low-end structure is unusually rigid. A graph has scramble number at most $\mathcal{S}$2 if and only if it has none of $\mathcal{S}$3, $\mathcal{S}$4, $\mathcal{S}$5, and $\mathcal{S}$6 as a topological minor (Eagleton et al., 2022). By contrast, no finite list of forbidden topological minors characterizes graphs of any fixed scramble number greater than $\mathcal{S}$7 (Eagleton et al., 2022). This failure of finite obstruction theory at higher values reflects the existence of infinite families of topological-minor-minimal graphs with prescribed scramble number.

The computational picture is correspondingly difficult. Computing scramble number is NP-hard (Echavarria et al., 2021). The carton number, defined as the minimum size of a maximum-order scramble, can be exponential in the size of the graph, which shows that scrambles are not valid NP certificates in general (Connor et al., 2023). This removes the most obvious route to placing the decision problem “$\mathcal{S}$8?” in NP. A restricted variant behaves better: the disjoint scramble number is fixed-parameter tractable when parameterized by $\mathcal{S}$9 and $G$0, via an MS$G$1 formulation and Courcelle’s theorem (Connor et al., 2023).

The invariant is also tightly linked to other width and congestion parameters. Vertex congestion is an upper bound on screewidth and hence on scramble number, leading to

$G$2

for maximum degree $G$3, and consequently to $G$4 upper bounds for bounded-degree planar graphs and, more generally, bounded-degree graphs in minor-closed families excluding a fixed complete graph (Connor et al., 2023).

4. Uniform scrambles and computed families

A particularly tractable family of scrambles is the $G$5-uniform scramble $G$6, whose eggs are all connected induced subgraphs on exactly $G$7 vertices. For a connected graph $G$8 on $G$9 vertices, the order of $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$0 is

$$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$1

where $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$2 is the $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$3-restricted edge-connectivity and $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$4 is the $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$5-component independence number (Cenek et al., 2021).

This formula yields a broad girth-based theorem. If $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$6 has girth at least $$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$7, then

$$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$8

and if in addition

$$\mathcal{S}=\{E_1,\dots,E_\ell\},\qquad E_i\subseteq V(G),$$9

then

$G[E_i]$0

This converts restricted edge-connectivity and generalized independence information directly into exact gonality and exact scramble number (Cenek et al., 2021).

A number of graph families admit exact formulas. For cycles,

$G[E_i]$1

and the same formula also holds for $G[E_i]$2. For complete multipartite graphs $G[E_i]$3, if

$G[E_i]$4

then

$G[E_i]$5

again matching scramble number. For grid graphs $G[E_i]$6,

$G[E_i]$7

for stacked prism graphs $G[E_i]$8,

$G[E_i]$9

and for toroidal grids $\mathcal{S}$0,

$\mathcal{S}$1

(Cenek et al., 2022).

Product formulas obtained by lower-bounding scramble number and matching Aidun–Morrison upper bounds for gonality are especially strong. If $\mathcal{S}$2 and $\mathcal{S}$3 are trees, then

$\mathcal{S}$4

If $\mathcal{S}$5 is a cycle and $\mathcal{S}$6 a tree, then

$\mathcal{S}$7

and for $\mathcal{S}$8,

$\mathcal{S}$9

(Harp et al., 2020).

Specific examples illustrate both sharpness and limitations. In the cube graph, scramble number closes the gap between treewidth and gonality: $H\subseteq V(G)$0 (Beougher et al., 2024). In the dodecahedron graph, the $H\subseteq V(G)$1-uniform scramble yields

$H\subseteq V(G)$2

(Beougher et al., 2024). In the icosahedron graph, however,

$H\subseteq V(G)$3

so scramble number is not always sharp (Beougher et al., 2024).

Hypercubes provide a particularly instructive sequence. The paper on uniform scrambles proves

$H\subseteq V(G)$4

and also establishes that

$H\subseteq V(G)$5

This shows that scramble number alone cannot prove the conjectural formula $H\subseteq V(G)$6 for all higher-dimensional hypercubes (Cenek et al., 2021).

5. Permutation-based scrambling in Kak’s mesh array

In Kak’s mesh-array literature, the relevant object is not a graph invariant but a permutation induced on matrix positions. Kak’s mesh array is a two-dimensional systolic architecture for multiplying two $H\subseteq V(G)$7 matrices in $H\subseteq V(G)$8 time steps. When its output placement is interpreted as a permutation of matrix positions, repeated application yields a deterministic scrambling transformation (Rangineni, 2011).

For $H\subseteq V(G)$9, the standard row-major order

$h(\mathcal{S}).$0

is sent to the mesh order

$h(\mathcal{S}).$1

The induced permutation decomposes as

$h(\mathcal{S}).$2

with cycle lengths $h(\mathcal{S}).$3 (Rangineni, 2011).

The paper defines the period of the scrambling transformation, for practical purposes, as the maximum cycle length, even though the full return time of the permutation is the least common multiple of all cycle lengths. In that sense, the natural scramble number for order $h(\mathcal{S}).$4 is

$h(\mathcal{S}).$5

For the $h(\mathcal{S}).$6 case,

$h(\mathcal{S}).$7

(Rangineni, 2011).

The paper computes longest cycle lengths for orders $h(\mathcal{S}).$8 through $h(\mathcal{S}).$9, then forms a binary sequence by recording parity: even longest-cycle length gives bit $T\subseteq E(G)$0, odd longest-cycle length gives bit $T\subseteq E(G)$1. After converting $T\subseteq E(G)$2 to $T\subseteq E(G)$3, it studies the autocorrelation

$T\subseteq E(G)$4

The reported outcome is that the autocorrelation is “effectively two valued,” with a dominant value at $T\subseteq E(G)$5 and nearly flat small values for $T\subseteq E(G)$6, which the paper interprets as evidence that the parity sequence is random-like (Rangineni, 2011).

This use of scramble number is therefore orbit-theoretic rather than graph-theoretic. It quantifies the depth of positional mixing produced by iterating a fixed permutation, not the obstruction structure of a graph.

6. Operational and adjacent usages

In the Rubik’s-cube literature, a natural operational meaning of scramble number is the mixing time of a random walk on cube configurations. The standard model chooses one of the six faces uniformly and then a rotation by $T\subseteq E(G)$7, $T\subseteq E(G)$8, or $T\subseteq E(G)$9 uniformly, producing a Markov chain on the approximately $G=(V,E)$00 reachable states of the $G=(V,E)$01 cube. The mixing time is

$G=(V,E)$02

where $G=(V,E)$03 is the total variation distance to the uniform distribution (Qu et al., 2024).

The first rigorous lower bound in this model is

$G=(V,E)$04

equivalently, 25 random moves are not enough to bring the distribution within total variation distance $G=(V,E)$05 of uniform (Qu et al., 2024). A later paper introduces a stopping time $G=(V,E)$06: the first time every pair of small cubes is “unlinked” along all three coordinate axes. This is a natural candidate for a strong uniform time, but the paper shows that it is not strongly uniform. It nevertheless gives informative heuristics: for the Pocket Cube, the exact mixing time is $G=(V,E)$07, while the expected time to unlink all pairs of corners is approximately $G=(V,E)$08; for the full $G=(V,E)$09 cube, simulations suggest $G=(V,E)$10, and the corresponding non-rigorous heuristic scale for the mixing time is at most $G=(V,E)$11 (Fernique, 15 Sep 2025).

Several adjacent literatures use “scrambling” intensively without fixing “scramble number” as a formal term. In number-conserving local random quantum circuits, the relevant quantitative pair is the maximum effective design order $G=(V,E)$12, with

$G=(V,E)$13

and the depth $G=(V,E)$14 required to approach an approximate $G=(V,E)$15-design, for which the paper gives

$G=(V,E)$16

and conjectures

$G=(V,E)$17

(Hearth et al., 2023). In randomized quasi-Monte Carlo, coarse scrambling of Sobol’ and Niederreiter sequences preserves the $G=(V,E)$18-sequence property, achieves variance

$G=(V,E)$19

and has maximal gain coefficient

$G=(V,E)$20

(Suzuki, 2 Oct 2025). In scrambled $G=(V,E)$21-linear pseudorandom number generators, scrambling denotes nonlinear maps such as $G=(V,E)$22, $G=(V,E)$23, $G=(V,E)$24, and $G=(V,E)$25 applied to a linear state engine to suppress linear artifacts and pass strong statistical tests (Blackman et al., 2018).

These adjacent usages do not define a single invariant called scramble number. They do, however, preserve the underlying conceptual pattern: a system has an intrinsic scrambling capability, and the corresponding quantitative parameter measures how much structure survives under iteration, randomization, or nonlinear post-processing.