Scattering Number: Graphs, Walks & Resonances

• Scattering number is a measure that quantifies a graph’s vulnerability by counting the maximum increase in connected components per vertex removed.
• In radiative transfer, it estimates the mean number of photon scatterings before escape, with values depending on optical depth and geometry.
• In spectral theory, scattering number serves as a resonance-counting function that bounds scattering poles using determinant-based arguments.

“Scattering number” denotes distinct counting or vulnerability quantities in several research areas. In graph theory, the term usually refers to Jung’s invariant

$$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\},$$

which measures how effectively vertex deletions fragment a connected graph and is tightly related to toughness (Markenzon et al., 2021). The same phrase is also used for the mean number of photon scatterings before escape in finite random-walk media, where the quantity depends on optical depth and geometry (Seon et al., 2023), and for resonance-counting functions that enumerate scattering poles of nonlocal operators on the logarithmic Riemann surface (Toprak, 2023). This terminological overlap makes the graph-theoretic invariant the dominant meaning in combinatorics, while the physical and analytic usages are domain-specific.

1. Graph-theoretic definition and relation to toughness

Let $G=(V,E)$ be a connected graph and let $\omega(G-S)$ denote the number of connected components of $G-S$. The scattering number of $G$, defined by Jung in 1978, is

$$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$

Any set $S\subseteq V$ attaining the maximum is a scattering set (Markenzon et al., 2021). An equivalent formulation used in later work is

$$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$

where $c(G-S)$ is the number of components of the graph obtained by deleting $S$ and all incident edges (Jia et al., 1 Sep 2025). For complete graphs, one source adopts the convention $G=(V,E)$0 (Broersma et al., 2013), whereas another treats complete graphs separately when defining related toughness parameters (Markenzon et al., 2021). This suggests that conventions outside the disconnected regime are not completely uniform across the literature.

The invariant is interpreted as a measure of vulnerability under vertex removal: it captures, in an additive way, how many components can be created per vertex removed in the most effective cut (Markenzon et al., 2021). This additive character is the standard contrast with toughness. For a non-complete graph,

$G=(V,E)$1

where the minimum is taken over all separators $G=(V,E)$2 with $G=(V,E)$3 (Markenzon et al., 2021). Kratsch, Kloks, and Müller proved the key equivalence

$G=(V,E)$4

so graphs with scattering number at most $G=(V,E)$5 are exactly the graphs that are at least $G=(V,E)$6-tough (Markenzon et al., 2021). A related variation due to Enomoto is

$G=(V,E)$7

and the condition $G=(V,E)$8 is equivalent to $G=(V,E)$9 (Jia et al., 1 Sep 2025).

These equivalences are conceptually strong but not algorithmically complete. In strictly chordal graphs, the knowledge of toughness is explicitly stated to be helpful but not sufficient to provide an immediate result for determining the scattering number (Markenzon et al., 2021). Likewise, the literature emphasizes that tough sets and scattering sets need not coincide (Markenzon et al., 2021).

2. Interval graphs: path covers, Hamiltonicity, and linear-time computation

Interval graphs provide one of the sharpest structural characterizations of scattering number. For a graph $\omega(G-S)$0, let $\omega(G-S)$1 denote the size of a smallest path cover. Hung and Chang showed that for all $\omega(G-S)$2, an interval graph has a path cover of size at most $\omega(G-S)$3 if and only if its scattering number is at most $\omega(G-S)$4, and that an interval graph has a Hamilton cycle if and only if its scattering number is at most $\omega(G-S)$5 (Broersma et al., 2013). The later refinement establishes that for all $\omega(G-S)$6, an interval graph $\omega(G-S)$7 is $\omega(G-S)$8-Hamilton-connected if and only if

$\omega(G-S)$9

thereby completing the characterization for all integer thresholds (Broersma et al., 2013).

The resulting trichotomy is exact: $G-S$0 if and only if

• $G-S$1 has a path cover of size at most $G-S$2, when $G-S$3;
• $G-S$4 has a Hamilton cycle, when $G-S$5;
• $G-S$6 is $G-S$7-Hamilton-connected, when $G-S$8 (Broersma et al., 2013).

This places scattering number at the center of Hamiltonian structure on interval graphs. The same paper observes that the maximum $G-S$9 for which an interval graph is $G$0-Hamilton-connected can therefore be computed from the scattering number in linear time (Broersma et al., 2013).

Algorithmically, the decisive object is a spanning $G$1-stave between the leftmost and rightmost vertices $G$2 and $G$3 of a clique-path representation. A $G$4-stave is a set of $G$5 internally vertex-disjoint paths, each with end-vertices $G$6 and $G$7; it is spanning if the union of these paths contains all vertices of the graph (Broersma et al., 2013). The fundamental equivalence is: $G$8 (Broersma et al., 2013). This converts the computation of scattering number into a linear sweep over a clique path, building an optimal spanning stave. The paper gives an $G$9 time algorithm for computing the scattering number of an interval graph, improving the $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$0 time bound of Kratsch, Kloks and Müller (Broersma et al., 2013).

The broader complexity contrast is stark. For general graphs, even deciding whether $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$1 is NP-complete (Broersma et al., 2013). For interval graphs, by contrast, scattering number, Hamiltonicity, path-cover size, and Hamilton-connectivity all admit linear-time treatment through the clique-path structure and spanning-stave characterization (Broersma et al., 2013).

3. Strictly chordal graphs: separator structure and a linear-time algorithm

Strictly chordal graphs, also called block duplicate graphs, form another class where scattering number admits a complete structural treatment (Markenzon et al., 2021). A chordal graph $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$2 is strictly chordal if and only if any two distinct minimal vertex separators are disjoint: $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$3 (Markenzon et al., 2021). For this class, the clique-bipartite graph $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$4, whose vertices are the minimal separators and maximal cliques and where $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$5 is adjacent to $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$6 iff $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$7, is a tree (Markenzon et al., 2021). This tree structure drives the algorithmic analysis.

If $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$8 is a minimal vertex separator of multiplicity $$sc(G)=\max\{\omega(G-S)-|S|: S\subseteq V,\ \omega(G-S)\neq 1\}.$$9, then

$S\subseteq V$0

(Markenzon et al., 2021). Markenzon and Waga previously determined the toughness of non-complete strictly chordal graphs as

$S\subseteq V$1

(Markenzon et al., 2021). The scattering number, however, requires a finer classification.

Several cases are explicit. If $S\subseteq V$2 and $S\subseteq V$3, then

$S\subseteq V$4

(Markenzon et al., 2021). If $S\subseteq V$5 and $S\subseteq V$6, then

$S\subseteq V$7

and every scattering set is a single minimal vertex separator achieving this maximum (Markenzon et al., 2021). For graphs with $S\subseteq V$8, the paper distinguishes type A and type B strictly chordal graphs. In type A, the scattering number collapses to

$S\subseteq V$9

(Markenzon et al., 2021). In type B, scattering sets may be unions of multiple minimal separators, and the paper analyzes border minimal vertex separators $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$0 satisfying $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$1, where $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$2 denotes the boundary cliques containing $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$3 (Markenzon et al., 2021).

The algorithmic conclusion is that, for strictly chordal graphs, both the scattering number and a scattering set can be determined in linear time $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$4 (Markenzon et al., 2021). The method computes maximal cliques, minimal vertex separators, and their multiplicities, branches on the toughness formula, and, in the type B case, performs a depth-first search on the tree $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$5 using structural rules for border separators (Markenzon et al., 2021). The paper explicitly notes that this extends efficient computation to subclasses such as block graphs, 3-leaf power graphs, strictly interval graphs, and generalized core-satellite graphs (Markenzon et al., 2021).

A recurring misconception is that toughness alone determines the relevant cut structure. The strictly chordal analysis contradicts that simplification: the paper gives an example with $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$6 and $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$7 where the tough set and the scattering set are different (Markenzon et al., 2021).

4. Spectral conditions, $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$8-radius, and $$s(G)=\max\{c(G-S)-|S|:S\subseteq V(G),\ c(G-S)>1\},$$9-toughness

A recent spectral direction relates scattering number and $c(G-S)$0-toughness to the $c(G-S)$1-spectral radius

$c(G-S)$2

introduced by Nikiforov; its largest eigenvalue is denoted $c(G-S)$3 (Jia et al., 1 Sep 2025). This interpolates between the adjacency matrix $c(G-S)$4 at $c(G-S)$5 and $c(G-S)$6 at $c(G-S)$7 (Jia et al., 1 Sep 2025).

The central scattering-number theorem in this setting states that if $c(G-S)$8 is a connected graph of order

$c(G-S)$9

with minimum degree $S$0, and

$S$1

then $S$2 unless

$S$3

(Jia et al., 1 Sep 2025). The extremal join graph has $S$4 (Jia et al., 1 Sep 2025), so the bound is tight. When $S$5, this reduces to an adjacency-spectral result of Chen, Li and Xu (Jia et al., 1 Sep 2025). This theorem therefore gives a spectral sufficient condition for $S$6, and, by the equivalence $S$7, also yields a spectral sufficient condition for $S$8 (Jia et al., 1 Sep 2025).

The same paper proves $S$9-spectral sufficient conditions for general $G=(V,E)$00-toughness. For integer $G=(V,E)$01, if

$G=(V,E)$02

under the stated order and $G=(V,E)$03-range assumptions, then $G=(V,E)$04 is a $G=(V,E)$05-tough graph unless it is the extremal join $G=(V,E)$06 (Jia et al., 1 Sep 2025). For $G=(V,E)$07 a positive integer, an analogous theorem uses the extremal graph

$G=(V,E)$08

(Jia et al., 1 Sep 2025). When $G=(V,E)$09, these reduce to previously known signless-Laplacian results of Chen, Li and Xu (Jia et al., 1 Sep 2025).

Technically, the proofs combine monotonicity of $G=(V,E)$10 under edge addition, equitable quotient matrices, and extremal lemmas for joins of cliques. A key ingredient is that for $G=(V,E)$11,

$G=(V,E)$12

is maximized, under fixed total order and component count, by making one component as large as possible and the others as small as the constraints permit (Jia et al., 1 Sep 2025). The spectral program therefore identifies the most vulnerable graphs, with respect to scattering number or $G=(V,E)$13-toughness, as explicit join constructions.

5. Random walks and radiative transfer: mean number of scatterings before escape

In radiative transfer and random-walk theory, the phrase “number of scatterings” refers to a different object: the mean number of scattering events a photon undergoes before escaping from a finite scattering medium (Seon et al., 2023). The medium is either a sphere of radius $G=(V,E)$14 or a slab of half-height $G=(V,E)$15, the optical thickness is

$G=(V,E)$16

and the photon free path in optical depth is exponentially distributed,

$G=(V,E)$17

(Seon et al., 2023). The quantity of interest is

$G=(V,E)$18

(Seon et al., 2023).

A widely quoted rule of thumb is

$G=(V,E)$19

but the paper “Number of Scatterings in Random Walks” shows that this formula is not accurate when one consistently uses the exponential step-length distribution (Seon et al., 2023). In the optically thick limit,

$G=(V,E)$20

(Seon et al., 2023). In the optically thin limit,

$G=(V,E)$21

whereas for a slab

$G=(V,E)$22

with $G=(V,E)$23 the Euler–Mascheroni constant (Seon et al., 2023). The logarithmic enhancement in the slab arises because nearly horizontal directions correspond to very large optical depths even when the vertical optical depth is small (Seon et al., 2023).

The paper also gives interpolation formulas for intermediate optical depths. For a sphere,

$G=(V,E)$24

which matches Monte Carlo results to within $G=(V,E)$25 over the entire range studied (Seon et al., 2023). For a slab,

$G=(V,E)$26

which reproduces Monte Carlo results for isotropic scattering to within about $G=(V,E)$27 across the range of $G=(V,E)$28 considered (Seon et al., 2023). The formulas apply to scattering processes with forward and backward symmetry, including isotropic and Thomson scattering (Seon et al., 2023).

This usage of “scattering number” is therefore not a graph invariant but a transport observable controlled by geometry, optical depth, and free-path statistics. The paper also identifies a specific literature error: the standard $G=(V,E)$29 argument mixes the exponential free-path distribution in the thin limit with an effectively fixed step length in the thick-limit random-walk estimate (Seon et al., 2023).

6. Scattering theory: counting scattering poles of $G=(V,E)$30

In spectral and scattering theory, “scattering number” may denote a resonance-counting function. For odd dimensions $G=(V,E)$31, the paper “Bounds on the number of scattering poles of half-Laplacian in odd dimensions, $G=(V,E)$32” studies

$G=(V,E)$33

on $G=(V,E)$34, where $G=(V,E)$35 is bounded, compactly supported, and complex-valued (Toprak, 2023). The cut-off resolvent

$G=(V,E)$36

extends meromorphically to the whole Riemann surface $G=(V,E)$37 of $G=(V,E)$38, and its poles are the scattering poles, or resonances, of the operator (Toprak, 2023).

The associated counting function is

$G=(V,E)$39

where the $G=(V,E)$40 are poles of $G=(V,E)$41, counted with multiplicity (Toprak, 2023). The main theorem proves the upper bound

$G=(V,E)$42

(Toprak, 2023). For fixed $G=(V,E)$43, the number of poles within radius $G=(V,E)$44 therefore grows like $G=(V,E)$45, while the dependence on the angular parameter $G=(V,E)$46 reflects how many sheets of the logarithmic Riemann surface are included (Toprak, 2023).

The analytic mechanism is determinant-based. The paper introduces

$G=(V,E)$47

where $G=(V,E)$48, proves growth bounds for $G=(V,E)$49, and applies a zero-counting theorem of Vodev to bound the resonance multiplicities through the zeros of $G=(V,E)$50 (Toprak, 2023). The resulting “scattering number” is thus a pole-counting function in a spectral problem, not a fragmentation index and not a random-walk mean.

A plausible implication is that the shared terminology across combinatorics, radiative transfer, and spectral analysis reflects a common counting theme rather than a common underlying definition. In the graph-theoretic literature, scattering number quantifies additive vulnerability under vertex deletion (Markenzon et al., 2021); in random walks it counts expected collision events before escape (Seon et al., 2023); and in resonance theory it counts poles of a meromorphically continued resolvent in sectors of $G=(V,E)$51 (Toprak, 2023).