K-Process Model Overview

• The K-process Model is a multi-faceted label that encompasses distinct stochastic and dynamical processes across probability theory, network science, and statistical physics.
• Depending on context, the symbol k or K can denote pruning thresholds, capacity limits, latent-factor counts, or physical constants such as dielectric values or kaon channels.
• Methodologies include scaling limits, recursive pruning algorithms, and hierarchical latent-factor models, each revealing critical dynamics and phase transitions in complex systems.

In arXiv literature, “K-process Model” is not a single standardized construction. The label is used for several mathematically and physically distinct objects: a class of Markov processes with a single unstable state in trap-model theory, a family of $k$-core pruning dynamics on networks, the $k$-cap or $k$-winners-take-all process on geometric random graphs, a one-dimensional kinetic contact/replication process with parallel update, the $(q,t)$ $K$-ASEP, the data-driven K-Process Model (KPM) for stellar abundances, an in situ $k$-restore process for ultra-low-$k$ dielectrics, and kaon-production reaction models in hadronic phenomenology (0808.3494, Fontes et al., 2012, Baxter et al., 2015, Shi et al., 2018, Wu et al., 2018, Reid et al., 2022, Ovchinnikov et al., 2024, Ayyer et al., 2023, Griffith et al., 2023, Förster et al., 2017, Volkov et al., 2018, Ahn et al., 2018, Guo et al., 3 Jul 2025). This suggests that the term functions primarily as a context-dependent family name rather than a universal formalism.

1. Scope and nomenclature

The semantic range of “K-process” is unusually broad. In probability theory and disordered systems, it denotes a specific Markov-process construction on a countable compactification. In network science, it denotes recursive pruning dynamics leading to the $k$-core. In geometric graph dynamics, it denotes exact-cardinality winner selection. In nonequilibrium statistical physics, it denotes either a synchronous contact/replication process or a bounded-occupancy exclusion process. In Galactic chemical evolution, it denotes a hierarchical latent-process model with $K$ nucleosynthetic channels. In materials science, $k$ denotes the dielectric constant. In hadronic physics, $k$0 denotes kaons rather than a counting parameter.

Usage	Defining feature	Representative paper
Trap-model K-process	Single unstable state $k$1, uniform entrance into finite sets	(0808.3494)
K-process on a tree	Infinite-volume limit of $k$2-level trap models	(Fontes et al., 2012)
$k$3-core pruning	Recursive deletion of nodes with degree $k$4	(Baxter et al., 2015)
NBEB / analytical $k$5-core model	Single-variable or nonbacktracking formulation of pruning	(Shi et al., 2018, Wu et al., 2018)
$k$6-cap process	Exact top-$k$7 winner update on geometric random graphs	(Reid et al., 2022)
Kinetic contact/replication K-process	Parallel-update PCA with hidden percolative backbones	(Ovchinnikov et al., 2024)
$k$8 $k$9-ASEP	Ring exclusion process with at most $k$0 particles per site	(Ayyer et al., 2023)
KPM for nucleosynthesis	Linear combination of $k$1 metallicity-dependent processes	(Griffith et al., 2023)
$k$2-restore process	Plasma-fragment repair of ultra-low-$k$3 dielectrics	(Förster et al., 2017)

A common source of confusion is the assumption that all of these models share a common stochastic architecture. The literature does not support that interpretation. The recurring symbol $k$4 or $k$5 denotes, depending on context, pruning threshold, assembly size, site capacity, number of latent processes, hierarchical depth, dielectric constant, or kaons.

2. Probabilistic K-processes in trap-model theory

In the probabilistic line initiated for Bouchaud-type trap models, a K-process is a Markov process on the one-point compactification $k$6, where all finite states are stable and $k$7 is the unique unstable state. The construction assumes a weight function $k$8 with $k$9, and the basic Dirichlet form is

$(q,t)$0

The associated process has exponential holding time of mean $(q,t)$1 at state $(q,t)$2, explicit hitting-time transforms such as

$(q,t)$3

and the characteristic entrance law that, starting from $(q,t)$4, the process enters any finite set $(q,t)$5 with uniform distribution. The probabilistic construction via independent Poisson clocks and exponential marks yields the $(q,t)$6-process, with $(q,t)$7 controlling the time-change at $(q,t)$8. The characterization theorem states that a càdlàg strong Markov process with exponential holding times at stable states and uniform entrance from $(q,t)$9 into finite sets must be a $K$0-process (0808.3494).

This framework is used as the scaling limit of Bouchaud’s trap model on the complete graph at low temperature. After ordering trap depths and rescaling time by

$K$1

the rescaled dynamics converge to a $K$2-process in a random environment $K$3 generated by a stable Lévy process. In this limit, macroscopic small-time aging is governed almost surely by the arcsine law: $K$4 This identifies the K-process as both a scaling-limit object and a mechanism for rigorous aging asymptotics (0808.3494).

Fontes, Gava, and Gayrard extended this construction from the complete graph to a hierarchical $K$5-level tree. The finite-volume model is a Markov jump process on the leaves $K$6 of a rooted tree $K$7, with leaf waiting times $K$8 and level-dependent coin-toss parameters $K$9. Its infinite-volume limit is the K-process $k$0, built by recursive clock-and-mark processes on the infinite tree $k$1. The key integrability hypothesis is

$k$2

Under pointwise convergence, product-summability, and negligibility of extra marks, finite-volume trap models converge weakly in Skorokhod space to the K-process on the infinite tree. With Sasaki–Nemoto’s GREM-like trap environment and the fine tuning

$k$3

the K-process is also the scaling limit on extreme time scales (Fontes et al., 2012).

3. Network pruning and $k$4-core dynamics

In network science, the K-process model denotes recursive $k$5-core pruning: at each step all vertices of degree $k$6 are removed together with incident edges, and the procedure repeats until no such vertices remain. For an uncorrelated random network with degree distribution $k$7, Baxter, Dorogovtsev, Lee, Mendes, and Goltsev formulated exact dynamical equations in terms of the time-dependent degree distribution $k$8 and the probability

$k$9

that a random edge leads to a vertex that will be pruned at the next step. The exact update for $k$0 is

$k$1

with accompanying formulas for the branching distribution $k$2 and mean branching number $k$3. Their theory shows three dynamical regimes: above threshold, exponential relaxation to the $k$4-core; at threshold, critical power-law decay

$k$5

and below threshold, a long plateau followed by collapse, with duration diverging as the mean degree approaches the critical value. For Erdős–Rényi graphs, the threshold is obtained from the tangency conditions of the self-consistency map, with reported values $k$6 and $k$7 (Baxter et al., 2015).

A later analytical simplification reduces the full degree-distribution evolution to a single auxiliary series $k$8. For a general degree distribution with generating functions $k$9 and $k$0,

$k$1

and the remaining fraction after the $k$2-th pruning is

$k$3

At stationarity,

$k$4

For Poisson degree $k$5, these become

$k$6

This single-variable formulation was presented as resolving the analytical difficulty of tracking the entire pruning trajectory (Shi et al., 2018).

The nonbacktracking expansion branch (NBEB) method gives a complementary message-passing formulation. Let $k$7 denote the probability that a nonbacktracking branch belongs to the survival set $k$8; then for uncorrelated networks

$k$9

and the fraction of edges in the $K$0-core is $K$1. For correlated networks with joint excess-degree distribution $K$2, the method generalizes to a vector recursion in the degree class $K$3, making it one of the few analytical approaches in the data block that directly addresses correlations (Wu et al., 2018).

4. Geometric winner-take-all dynamics: the $K$4-cap process

In the paper “The $K$5-Cap Process on Geometric Random Graphs” (Reid et al., 2022), the K-process model is a discrete-time winner-take-all dynamic on a directed geometric random graph. The graph has $K$6 vertices with hidden positions $K$7, and directed edges are present independently with Gaussian kernel

$K$8

At time $K$9, the active set $k$0 has $k$1, and each vertex is scored by

$k$2

The next winner set is the top-$k$3 set under this score, with threshold $k$4 chosen so that

$k$5

where $k$6 is sampled uniformly among ties at $k$7.

The continuous one-dimensional $k$8-cap analogue replaces the finite graph by $k$9 and defines

$k$00

with $k$01. Under an even, nonnegative, integrable, differentiable kernel with $k$02 for $k$03, the fixed points are exactly the single intervals of length $k$04, and any finite union of intervals converges to such an interval in

$k$05

steps. The discrete model behaves differently. In the regime

$k$06

the process does not converge to a fixed winner set. Instead, it localizes spatially. The first update $k$07 can be covered by $k$08 balls of radius $k$09, separated by at least $k$10. These balls then shrink multiplicatively,

$k$11

until, after polylogarithmic time, the winners lie in a single ball of radius

$k$12

For all $k$13, and any fixed $k$14, there is with high probability a ball of radius $k$15 containing at least $k$16 winners. The paper emphasizes that geometry rather than plasticity drives this localization, in contrast to earlier Erdős–Rényi assembly models associated with Papadimitriou and collaborators (Reid et al., 2022).

5. Percolation and interacting-particle meanings

One statistical-physics meaning of “K-process” is the one-dimensional kinetic contact/replication process with parallel update. The model is defined on a periodic lattice of length $k$17, with binary occupation $k$18. The local update probabilities depend on the triplet $k$19, with control parameters $k$20 and $k$21, including the two-neighbor seeding probability

$k$22

The empty configuration is absorbing, and the standard density order parameter is

$k$23

The central result is that the active phase is not monolithic: it contains a hierarchy of hidden directed-percolation backbones, specifically dipole backbones $k$24 and $k$25, quadrupole backbones $k$26 and $k$27, and a plaquette backbone $k$28. At fixed $k$29, the paper reports the sequence

$k$30

for decreasing $k$31, and

$k$32

for increasing $k$33. All absorbing-to-active and backbone transitions are reported to fall in the $k$34D directed percolation universality class, with measured exponents $k$35–0.1595, $k$36, $k$37–1.58, $k$38, and $k$39–1.11. The authors interpret this as motivating an extension of the Janssen–Grassberger conjecture from a unique global absorbing state to multiple backbone-specific absorbing vacua (Ovchinnikov et al., 2024).

A different interacting-particle usage appears in the exactly solvable $k$40 $k$41-ASEP on a ring. Here the state space is

$k$42

with at most $k$43 particles per site and periodic boundary conditions. The clockwise and counterclockwise nearest-neighbor hopping rates are

$k$44

where $k$45 is the $k$46-deformed integer. As a special case of the misanthrope process, the model has a product-form steady state independent of $k$47. The canonical measure is

$k$48

The stationary weights are palindromic polynomials in $k$49, and although the dynamics lacks particle-hole symmetry for $k$50, the steady state satisfies

$k$51

For $k$52 the model reduces to the ordinary ASEP on the ring; for $k$53 it reduces to the Schütz–Sandow model with weights $k$54 (Ayyer et al., 2023).

6. Data-driven and materials-science formulations

In Galactic chemical evolution, the K-Process Model is a hierarchical latent-factor model for stellar abundances. For stars $k$55, elements $k$56, and processes $k$57, the expected abundance is

$k$58

where $k$59 are star-specific process amplitudes and $k$60 are metallicity-dependent process yields. In the fiducial $k$61 model, the two processes are interpreted as prompt and delayed. Identifiability is fixed by anchoring Mg and Fe: $k$62

$k$63

The amplitudes and process vectors are fitted by alternating Gauss–Newton optimization of a robust objective

$k$64

with inverse-variance softening parameter $k$65. Applied to 48,659 APOGEE DR17 red-giant stars and 15 elements, the model recovers the disk’s prompt–delayed abundance bimodality and, in the $k$66 extension, adds Ce-anchored and Mn-anchored processes to capture s-process AGB-like and SNIa-yield-family residual structure. The paper stresses that some inferred fractional contributions depend on the detailed anchor choices, especially the prompt Fe fraction (Griffith et al., 2023).

In materials science, by contrast, the $k$67-process model concerns the dielectric constant of porous ultra-low-$k$68 (ULK) materials. Damage replaces nonpolar Si–CH$k$69 environments with polar Si–OH and Si–H groups, thereby increasing $k$70. The proposed $k$71-restore process uses plasma-enhanced fragmentation of silylation precursors to generate small reactive fragments that repair damaged sites by restoring Si–CH$k$72-rich environments. The DFT workflow uses DMol3, the PBE functional, a DNP(3.5) basis, and Grimme DFT-D, with reaction energy

$k$73

A representative fragment-formation reaction is reported as strongly exothermic, with $k$74 kcal/mol for a transformation replacing one O by two CH$k$75 groups. The central design rules are explicit: oxygen-containing fragments are required to repair Si–H damage, and fragments with at least one dangling Si bond are most effective for Si–OH repair. The paper therefore uses “K-process” in a sense completely different from the probabilistic and graph-dynamical usages: here $k$76 is the relative dielectric constant, not a graph parameter or number of latent processes (Förster et al., 2017).

7. Hadronic reaction-process usage and the limits of the term

A further set of papers uses “K-process” descriptively for kaon-involved reaction processes. In the extended Nambu–Jona-Lasinio treatment of $k$77, the amplitude combines a contact term with intermediate $k$78, $k$79, and $k$80 mesons in ground and first radially excited states. The full amplitude is written

$k$81

and the resulting kaon form factor leads to

$k$82

The paper emphasizes satisfactory agreement with Novosibirsk and Stanford data from $k$83 to approximately $k$84 GeV and notes that the omission of higher resonances becomes important beyond that range (Volkov et al., 2018).

In effective-Lagrangian studies of $k$85 scattering, the same terminological shift occurs. For $k$86 and $k$87, the model includes nine $k$88 hyperons and resonances in $k$89- and $k$90-channel exchange. Two fitting strategies are reported because of inconsistent charged-channel data near $k$91 GeV. Model A assigns all data equal weight and gives $k$92 with $k$93; Model B reweights the problematic window and worsens the global fit. The dominant contribution is the $k$94 intermediate process, and differential cross sections are predicted for both charged and neutral channels as future tests at J-PARC (Guo et al., 3 Jul 2025).

A related tree-level Born analysis of $k$95 and $k$96 near threshold includes $k$97-pole and $k$98-pole mechanisms with covariant form factors

$k$99

Using parameters fitted to the three-body reaction, the paper predicts that the two-body process has a total cross section peaking at approximately $k$00 around $k$01 GeV/$k$02, together with a strong enhancement at backward $k$03 angles due to dominant $k$04-channel contributions (Ahn et al., 2018).

A common misconception is therefore especially visible in hadronic phenomenology: the presence of “K-process” in a title or summary does not imply any relation to the probabilistic K-process of trap models or to the $k$05-core process on graphs. In these papers, $k$06 names kaons and kaon-induced channels. More generally, the surveyed literature indicates that “K-process Model” is a polysemous technical label whose meaning must be inferred from the surrounding mathematical structure, not from the phrase alone.