Weak Fragmentation in Complex Systems

• Weak fragmentation is a regime where microscopic splitting occurs but the overall system retains a dominant, bounded structure.
• It is characterized by limited macroscopic impact, with metrics such as bounded complexity, finite block multiplicity, or asymptotically slow breakup.
• Applications span dynamic spectrum access, quantum models, and kinetic equations, providing insights into system stability amid fragmentation.

Weak fragmentation is not a single universal concept. In different research literatures it denotes fragmentation that is present but limited in a precise sense: a fragmented allocation state whose average structural complexity remains bounded over time, a Hilbert space that splits into exponentially many invariant sectors while one giant sector still dominates, an effective or prethermal fragmentation that is exact only in a reduced description, or a breakup process whose rate becomes asymptotically slow for large objects. Closely related literatures also use “weak” in different ways—most commonly for weak solutions or weak convergence of fragmentation equations and fragmentation-valued processes—so the meaning of the phrase is inseparable from context (Coffman et al., 2010, Liu, 18 May 2026, Yoshinaga et al., 2021, Barik, 2019, Zhou et al., 6 Apr 2026, Ojeda et al., 7 Jul 2025).

1. Terminological scope and recurrent structure

In dynamic spectrum access, the exact phrase is not introduced, but the closest formal notion is a uniform exponential-moment bound on the total number of fragments plus gaps, even when fragments may be arbitrarily small (Coffman et al., 2010). In Hilbert-space fragmentation, “weak” is defined by block-size statistics: exponentially many Krylov sectors still exist, but one giant connected component contains an overwhelming majority of states (Liu, 18 May 2026). In quantum fragmentation with entangled frozen states, the weak case is defined by

$\frac{D_{\max}}{D_q}\sim \mathcal O(1),$

and, in the models studied there, equivalently by

$N_{\rm irr}(L)=\mathcal O(1),$

after removing the entangled frozen subspace (Zhou et al., 6 Apr 2026). In coagulation–fragmentation theory, by contrast, weak fragmentation is an explicit large-mass asymptotic condition on the selection rate,

$$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$

so that “the rate of splitting of large size particles are very slow” (Barik, 2019).

This suggests a common schematic interpretation: fragmentation is allowed, and may even be exponentially prolific at the microscopic level, but its macroscopic effect is limited by dominance of a giant sector, bounded average complexity, finite block multiplicity, or asymptotically slow breakup. The same word therefore organizes several non-equivalent but structurally related ideas.

2. Bounded average fragmentation in dynamic allocation systems

A mathematically strong version of weak fragmentation appears in the continuous-spectrum allocation model of “Channel Fragmentation in Dynamic Spectrum Access Systems - a Theoretical Study” (Coffman et al., 2010). The system lives on the interval $[0,1]$, serves requests at capacity, and permits each request to be allocated as a collection of disjoint subintervals with no lower bound on fragment size. The formal state is

$x=(L_1,\ldots,L_r;u),$

where $r$ is the number of active channels, $L_i$ is the list of fragments occupied by channel $i$, and admissibility requires

$u > 1-\sum_i s_i.$

Request sizes are i.i.d. uniform on $(0,\alpha]$, residence times are i.i.d. exponential with mean $N_{\rm irr}(L)=\mathcal O(1),$0, and the embedded process at departure epochs is Markov.

The basic fragmentation observable is

$N_{\rm irr}(L)=\mathcal O(1),$1

with $N_{\rm irr}(L)=\mathcal O(1),$2 the total number of fragments and $N_{\rm irr}(L)=\mathcal O(1),$3 the number of gaps. The central boundedness theorem states that there exists $N_{\rm irr}(L)=\mathcal O(1),$4 such that

$N_{\rm irr}(L)=\mathcal O(1),$5

This is stronger than bounded first moments and implies, in particular,

$N_{\rm irr}(L)=\mathcal O(1),$6

The point is not that fragmentation is combinatorially forbidden. The model allows arbitrarily many arbitrarily small fragments, and the paper explicitly does not prove a deterministic cap on fragment counts, a minimum fragment size, pathwise boundedness, or full ergodicity for all $N_{\rm irr}(L)=\mathcal O(1),$7. What is proved is that fragmentation does not drift upward without bound in expectation. In the paper’s own formulation, “even if fragments can be arbitrarily small, the system does not degrade with time.”

The proof uses a Foster–Lyapunov argument for an embedded chain over four departure epochs. A key combinatorial identity is

$N_{\rm irr}(L)=\mathcal O(1),$8

where $N_{\rm irr}(L)=\mathcal O(1),$9 counts type-$$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$0 fragments and $$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$1 indicates whether a gap starts at the origin. At departure times,

$$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$2

so admissions increase $$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$3, while departures of low-adjacency fragments decrease it. Because $$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$4 alone is insufficient, the paper uses

$$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$5

with a drift inequality of the form

$$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$6

whenever $$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$7 is large.

The same paper also proves that Knuth’s 50% rule survives in asymptotic form despite fragmentation:

$$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$8

Algorithmically, Linear Scan, Circular Scan, and Largest-First Scan all preserve this asymptotic gap-to-channel ratio, but their fragmentation performance differs substantially. Largest-First Scan reduces the average number of fragments per channel by a factor greater than 3 relative to Linear Scan and Circular Scan for moderately small $$S_R(m)\le k_3 \varphi(m)m^{1+\gamma},\qquad \lim_{m\to\infty}\varphi(m)=0,$$9, and the introduction summarizes some regimes as showing “almost an order of magnitude” reduction in total fragmentation measures. At the same time, the paper distinguishes temporal stability from parameter blow-up: for fixed $[0,1]$0, expected fragmentation is bounded in time, but as $[0,1]$1 the average number of channels scales like $[0,1]$2, the average number of gaps like $[0,1]$3, the average number of fragments per channel linearly in $[0,1]$4, and the average total number of fragments on the order of $[0,1]$5. Weak fragmentation here is therefore a stability statement, not a uniform smallness statement.

3. Weak Hilbert-space fragmentation in constrained quantum models

In the dipole-conserving Bose-Hubbard chain of “Weak Fragmentation and Thermalization in a Dipole-Conserving Bose-Hubbard Chain,” weak fragmentation has an explicit graph-theoretic meaning (Liu, 18 May 2026). The Hamiltonian is

$[0,1]$6

with exact conservation of

$[0,1]$7

Inside a fixed $[0,1]$8 sector, basis states form a connectivity graph under nonzero off-diagonal Hamiltonian matrix elements, and connected components are Krylov sectors. The model fragments because many states with the same $[0,1]$9 are dynamically disconnected.

The weak–strong distinction is defined by the distribution of Krylov-sector sizes. In the weak case, exponentially many sectors still exist, but one giant block occupies an overwhelming fraction of the Hilbert-space dimension. In the studied $x=(L_1,\ldots,L_r;u),$0 sector, the number of disconnected blocks grows as

$x=(L_1,\ldots,L_r;u),$1

yet the largest block contains more than $x=(L_1,\ldots,L_r;u),$2 of the total dimension for the accessible sizes, and its fraction appears to approach $x=(L_1,\ldots,L_r;u),$3 as $x=(L_1,\ldots,L_r;u),$4. This is the paper’s precise definition of weak Hilbert-space fragmentation.

The model also possesses exponentially many frozen product states, defined by

$x=(L_1,\ldots,L_r;u),$5

These are one-dimensional Krylov sectors. Their exact number is difficult to compute, but the paper derives Fibonacci-type upper and lower bounds and finds numerically

$x=(L_1,\ldots,L_r;u),$6

Frozen states therefore provide strong evidence of fragmentation, but they do not dominate the Hilbert space; the giant connected component does.

The thermalization diagnostics show why the adjective “weak” matters physically. At large $x=(L_1,\ldots,L_r;u),$7, most eigenstates in the middle of the spectrum have half-chain entanglement close to the Page value, $x=(L_1,\ldots,L_r;u),$8 decreases with system size in the density-relaxation test, and level statistics are GOE-like. At small $x=(L_1,\ldots,L_r;u),$9, entanglement is suppressed, $r$0 remains finite, and level statistics become Poisson-like. The paper places the crossover around

$r$1

while emphasizing that accessible sizes are too small for reliable finite-size scaling. The key conclusion is that weak Hilbert-space fragmentation does not preclude quantum chaos or thermalization: rare nonthermal sectors persist, but they remain thermodynamically subdominant.

4. Emergent and quantum weak fragmentation beyond a giant classical block

A distinct but related notion appears in “Emergence of Hilbert Space Fragmentation in Ising Models with a Weak Transverse Field” (Yoshinaga et al., 2021). Here the microscopic Hamiltonian is

$r$2

with $r$3 in $r$4. Fragmentation is not exact microscopically. Instead, first-order degenerate perturbation theory yields

$r$5

and the effective model exactly conserves domain-wall number

$r$6

Thus fragmentation is exact in the effective model but only approximate and prethermal in the full TFIM. Frozen regions and melting regions partition the dynamics, generating exponentially many disconnected Krylov sectors inside a fixed domain-wall sector. The model therefore realizes a weak or emergent fragmentation in the sense that it is generated by a weak-field regime and melts at sufficiently long times under the full Hamiltonian. The paper also shows that at least some connected fragments are internally nonintegrable: after resolving symmetries, a subspace without frozen regions exhibits GOE level statistics.

“Quantum Hilbert Space Fragmentation and Entangled Frozen States” refines the distinction further by locating the mechanism in rank deficiency of local Hamiltonian terms (Zhou et al., 6 Apr 2026). If the local coupling matrix has null directions, these can be propagated into an entangled frozen subspace

$r$7

inside a classically mobile sector, giving the decomposition

$r$8

Weak quantum fragmentation is then defined by

$r$9

and, in all models studied there, equivalently by $L_i$0, where $L_i$1 is the mobile quantum sector after removing entangled frozen states.

This definition supports a sharp weak–strong taxonomy. In the asymmetric qubit projector model, $L_i$2; in the $L_i$3-symmetric GHZ projector model, the symmetry-stable all-mobile sector splits into $L_i$4; in the $L_i$5-symmetric cyclic qutrit projector model, symmetry-stable all-mobile sectors split into $L_i$6. In each weakly fragmented case, the irreducible blocks are individually ergodic with GOE level statistics, while the unresolved spectrum follows an $L_i$7GOE distribution. By contrast, the Temperley–Lieb model is strongly fragmented: the number of irreducible blocks grows with system size,

$L_i$8

and the unresolved gap-ratio distribution approaches Poisson. Weak fragmentation here does not mean absence of frozen states; it means that after removing them, the mobile quantum sector still contains only finitely many macroscopic blocks.

5. Weak fragmentation in kinetic equations, and its distinction from weak-solution theory

In the continuous coagulation–multiple fragmentation equation studied in “Gelation in coagulation and multiple fragmentation equation with a class of singular rates,” weak fragmentation is an explicit asymptotic condition on the fragmentation selection rate for large masses (Barik, 2019). The equation is

$L_i$9

and the weak-fragmentation hypothesis is

$i$0

with $i$1. The paper states explicitly that “the rate of splitting of large size particles are very slow” and that this condition “is also known as the weak fragmentation.” A related assumption,

$i$2

is used in the gelation theorem. The point is that fragmentation of large particles is asymptotically too weak to offset coagulation strongly. Global existence of gelling weak solutions is proved under these assumptions, and gelation persists even in the presence of fragmentation. Weak fragmentation here therefore means slower-than-coagulation effective action on large masses, not bounded complexity or dominant-block structure.

This meaning should be separated from the much broader PDE literature in which “weak” modifies the solution concept rather than the fragmentation regime. “Weak solutions to the continuous coagulation equation with multiple fragmentation” constructs global weak solutions for unbounded coagulation kernels and multiple fragmentation kernels that may have a singularity at the origin (Giri et al., 2011). “Existence of mass-conserving weak solutions to the singular coagulation equation with multiple fragmentation” proves global mass-conserving weak solutions by weak $i$3 compactness for conservative and non-conservative truncations (Barik, 2018). “On the uniqueness for coagulation and multiple fragmentation equation” proves uniqueness of weak solutions under growth assumptions linking coagulation and fragmentation (Giri, 2012). In a size-structured growth–coagulation–fragmentation model, the fragmentation operator is represented weakly through

$i$4

which is a weak-form device rather than a weak-fragmentation regime (Si et al., 7 Apr 2025). The same distinction holds for weak-solution theories with mass transfer in nonlinear fragmentation (Jaiswal et al., 2024) and for fragmentation-controlled global classical solvability under domination of coagulation by a power of the fragmentation rate (Banasiak, 2019). In this part of the literature, “weak fragmentation” is a specialized asymptotic-rate condition; “weak solutions” and “weak convergence” are different notions.

6. Observational analogues, adjacent usages, and common disambiguations

Several neighboring literatures use related language without defining weak fragmentation as a formal term. In massive dense cores, Palau et al. report a “weak (inverse) trend” between fragmentation level and density power-law index: steeper density profiles tend to show lower fragmentation, while fragmentation increases with density inside $i$5 pc and is described as “consistent with Jeans fragmentation” (Palau et al., 2014). The least fragmented cores tend to have concentrated envelopes, often with $i$6, and the preferred physical interpretation is strong magnetic regulation of the density structure. This is an observational low-fragmentation regime, not a generic formal definition.

In social-network dynamics, “Positive algorithmic bias cannot stop fragmentation in homophilic networks” studies whether weak ties or rewiring can prevent segregation (Blex et al., 2020). There, fragmentation is the asymptotic extinction of cross-type ties, quantified by $i$7 or $i$8. Theorem 2 states that this still occurs “even in the presence of secondary ties,” so weak ties do not hinder fragmentation. The phrase weak fragmentation is not introduced; the result concerns weak ties failing to prevent fragmentation.

In gravo-turbulent fragmentation, suppression is expressed through the competition between the scale-dependent variance $i$9 and barrier $u > 1-\sum_i s_i.$0, and below an effective sonic scale the cascade is sharply weakened (Hopkins, 2012). Gas becomes stable below that scale for polytropic $u > 1-\sum_i s_i.$1, while fragmentation still occurs on larger scales. This is a continuum analogue of a weakened or truncated fragmentation cascade rather than a block-dominance notion.

Other papers use “weak” in entirely different ways. “A tightness criterion for fragmentations” is about weak functional convergence of fragmentation processes and reduces tightness to control of

$u > 1-\sum_i s_i.$2

not about a weak-fragmentation regime (Ojeda et al., 7 Jul 2025). “Isospin Symmetry of Fragmentation Functions” studies weak-decay-induced violations of isospin symmetry in observed fragmentation functions; under its assumptions,

$u > 1-\sum_i s_i.$3

and weak decays generate only tiny violations for other hadrons (Chen et al., 2021). “Predictions for Boson-Jet Observables and Fragmentation Function Ratios from a Hybrid Strong/Weak Coupling Model for Jet Quenching” treats showering and fragmentation as weakly coupled while medium energy loss is modeled by strong-coupling input; the weak/strong distinction there refers to coupling regimes, not to fragmentation being limited (Casalderrey-Solana et al., 2015). “De-Fragmenting the Cloud” introduces “relative resource fragmentation,”

$u > 1-\sum_i s_i.$4

meaning free capacity of one resource becomes unusable because another required resource is unavailable (Mishra et al., 2015). These usages are conceptually adjacent but terminologically distinct.

Across these domains, the safest encyclopedic conclusion is therefore a qualified one. Weak fragmentation is a family of context-dependent notions for regimes in which fragmentation exists but remains limited in dominance, rate, scale range, or long-run consequence. In some literatures the limitation is probabilistic and dynamical, in others combinatorial or spectral, and in still others purely asymptotic. The term is informative only when anchored to the precise state space, observable, and theorem that define what is “weak.”