Keller–Segel Model: Chemotaxis & Pattern Formation

• The Keller–Segel model is a system of coupled PDEs that models chemotaxis through diffusion and gradient-driven drift, forming the basis for understanding cell aggregation.
• It explains how the interplay between random dispersal and directed chemical signaling leads to phenomena like pattern formation, steady states, and blow-up under critical mass conditions.
• Advanced extensions incorporate fractional diffusion, flux limitations, and cross-disciplinary adaptations, applying the model to fields as diverse as tissue mechanics and socio-economic clustering.

Searching arXiv for recent and foundational Keller–Segel papers to support the article. The Keller–Segel model, often called the Patlak–Keller–Segel model, is a class of coupled partial differential equations for a population density and a chemical field that models chemotaxis: random dispersal competes with directed migration up chemical gradients, and the resulting balance can produce dispersion, aggregation, pattern formation, or blow-up depending on the regime and the variant under consideration (Hausenblas et al., 2020). In its classical biological interpretation, the unknowns are the cell density and the chemoattractant concentration; in the broader literature, the same structural template has been reinterpreted for reaction networks, fractional diffusion, volume exclusion, traveling pulses, tissue mechanics, and even socio-economic clustering (Chavanis, 2008).

1. Canonical equations and modeling interpretation

A standard parabolic Patlak–Keller–Segel system takes the form

$$\partial_t u = r_u \Delta u - \chi \,\mathrm{div}(u\nabla v), \qquad \partial_t v = r_v \Delta v - \alpha v + \beta u,$$

where $u$ is the cell density and $v$ is the chemoattractant concentration. The coefficients $r_u$ and $r_v$ are diffusivities, $\chi$ is the chemotactic sensitivity, $\alpha$ is the damping rate of the chemical, and $\beta$ is the production rate by the cells (Hausenblas et al., 2020). In the simplified two-dimensional parabolic–elliptic model emphasized in the notes of “La ecuación de Keller-Segel,” the system becomes

$$\partial_t \rho = \Delta \rho - \nabla \cdot (\rho \nabla c), \qquad -\Delta c = \rho,$$

or equivalently

$$\partial_t \rho = \Delta \rho - \nabla \cdot \bigl(\rho \nabla G * \rho\bigr),$$

with $u$0 (Fernández-Jiménez, 2021).

The first equation encodes diffusion plus chemotactic drift. The diffusive term spreads the population, while the drift term concentrates it in regions of high signal. The second equation describes the chemical field through diffusion, decay, and production, or through an elliptic closure when the chemical equilibrates rapidly. In bounded domains, homogeneous Neumann conditions are frequently imposed, expressing no-flux confinement for both cells and chemical (Acosta-Soba et al., 2022).

Several derivations coexist in the literature. The probabilistic account in (Fernández-Jiménez, 2021) derives Keller–Segel formally from a stochastic many-particle system in which Brownian motion yields diffusion and biased drift toward chemical gradients yields aggregation. The flux-limited literature instead starts from kinetic run-and-tumble dynamics and obtains a macroscopic drift–diffusion equation after parabolic scaling and stiff-response asymptotics (Perthame et al., 2018). These derivations show that Keller–Segel is not a single equation but a modeling paradigm: the defining structure is diffusion coupled to gradient-driven transport.

2. Mass conservation, free energy, and gradient-flow structure

A central structural property is conservation of total population mass. For sufficiently regular solutions,

$u$1

obtained by integrating the density equation and using the divergence structure of both diffusion and chemotactic fluxes (Fernández-Jiménez, 2021). This conservation law is not merely formal: it is the parameter controlling several threshold phenomena in the classical two-dimensional theory.

The model also possesses a free-energy structure. For the two-dimensional parabolic–elliptic form,

$u$2

and along smooth solutions

$u$3

This identifies Keller–Segel as a Wasserstein gradient flow of $u$4, a viewpoint that underlies entropy methods and minimizing-movement constructions (Fernández-Jiménez, 2021).

A more general formulation replaces linear diffusion and linear mobility by

$u$5

with entropy

$u$6

and free energy $u$7, $u$8. In this setting the Keller–Segel family is interpreted as a nonlinear mean-field Fokker–Planck system with a Lyapunov functional and an $u$9-theorem (Chavanis, 2008). This generalized thermodynamic viewpoint explains why variants with anomalous diffusion, exclusion effects, or nonlinear mobility remain structurally close to the classical model even when their dynamics differ sharply.

The same variational architecture guides numerical analysis. Both the discontinuous Galerkin scheme of (Acosta-Soba et al., 2022) and the semi-implicit Euler scheme of (Huang et al., 3 Mar 2025) are designed to preserve mass, positivity, and energy dissipation because these are the continuous invariants governing aggregation and collapse.

3. Critical mass, blow-up, and instability mechanisms

The best-known threshold result concerns the slightly simplified two-dimensional parabolic–elliptic system. If $v$0, diffusion dominates; if $v$1 and the initial second moment is finite, finite-time blow-up occurs; and if $v$2 with finite second moment, there is blow-up in infinite time with bubbling asymptotics (Fernández-Jiménez, 2021). The second moment identity

$v$3

organizes this classification.

A common misconception is that the number $v$4 is universal for all Keller–Segel models. It is not. In the stochastic one-dimensional setting of (Hausenblas et al., 2020), the issue is existence of martingale solutions rather than a two-dimensional critical-mass collapse. In the fractional mass-critical model of (Bian et al., 2024), the relevant threshold is

$v$5

derived from a sharp variant of the Hardy–Littlewood–Sobolev inequality; solutions exist globally below $v$6, while finite-time blow-up occurs above $v$7 for suitable initial data. In the particle interpretation of the elliptic two-dimensional model, (Fatkullin, 2010) distinguishes the smooth-data blow-up threshold

$v$8

from the lower threshold

$v$9

for trapping by a pre-existing singularity.

The instability mechanism itself is broader than the classical two-field setting. In “Instability in a generalized Keller-Segel model,” chemotactic motion is embedded in a system with one motile species and $r_u$0 interacting chemicals, only one of which is the chemoattractant. The spectral problem for linearization about a homogeneous steady state reduces to a finite-dimensional matrix problem indexed by Neumann Laplacian modes, and sufficiently strong chemotactic feedback destabilizes the homogeneous state under explicit Metzler/graph-theoretic conditions (Leenheer et al., 2011). This shows that Keller–Segel instability is a network-level feedback mechanism, not merely a peculiarity of the minimal two-variable model.

4. Stationary states and generalized model classes

The classical model admits a large family of structurally distinct generalizations. One major branch replaces linear diffusion by porous-medium or quadratic diffusion. In one dimension with quadratic cellular diffusion and Neumann conditions,

$r_u$1

the phase diagram is organized by thresholds

$r_u$2

Below $r_u$3, the uniform state is the global exponential attractor; above $r_u$4, half-bumps, multi-bumps, and asymmetric compactly supported steady states appear, with a strict energy hierarchy favoring the single boundary spike among symmetric bump states (Carrillo et al., 2019). In the two-dimensional disk with quadratic diffusion, explicit radial stationary solutions produce inner rings, outer rings, Mexican-hat states, Volcano states, and higher concentric-ring patterns; as $r_u$5, these profiles converge to $r_u$6-type spikes (Chen et al., 2019).

Another branch regularizes the chemotactic flux itself. The flux-limited Keller–Segel system

$r_u$7

assumes that $r_u$8 is bounded, so the drift velocity saturates for large gradients. Unlike the classical system, solutions of the flux-limited model do not blow up in finite or infinite time, while still exhibiting nontrivial aggregation and, in two dimensions, a critical-mass condition for radial steady states when $r_u$9 (Perthame et al., 2018). This makes explicit that blow-up is a property of specific chemotactic closures, not of the entire Keller–Segel paradigm.

Modern extensions push the framework further. The one-dimensional stochastic model of (Hausenblas et al., 2020) perturbs the deterministic system by multiplicative time-homogeneous spatial Wiener processes in the Stratonovich sense and proves existence of a martingale solution on $r_v$0. The generalized doubly parabolic fractional system

$r_v$1

introduces distinct fractional exponents for cell and chemical diffusion, with local mild well-posedness and small-data global well-posedness in $r_v$2-type spaces (Bronzi et al., 16 Jan 2025). Trait-structured chemotaxis adds a continuous receptor-occupancy trait $r_v$3, leading to a nonlocal model in $r_v$4 with a chemotaxis–proliferation trade-off and a Hopf–Cole asymptotic regime in which the trait distribution concentrates on a dominant phenotype within traveling waves (Freingruber et al., 26 Feb 2025). A different line, pursued in (Elbar et al., 2021), shows that a generalized Keller–Segel system is equivalent to a relaxed degenerate Cahn–Hilliard model, linking chemotaxis with diffuse-interface tissue mechanics and incompressible limits.

5. Traveling waves, computation, and structure-preserving numerics

Traveling-wave theory occupies a distinct subfield. In the singularly perturbed one-dimensional model studied by (Harley et al., 2014), geometric singular perturbation theory constructs waves connecting $r_v$5 to $r_v$6 for $r_v$7, $r_v$8, $r_v$9, and proves convergence to an explicit zero-diffusion wave in the singular limit. In the generalized run-and-tumble derivation of (Seyrich et al., 2019), bounded chemotactic drift yields traveling bacterial concentration pulses with a finite maximum carrying capacity, and the pulse speed, width, and height are analyzed in terms of microscopic swimming parameters.

Because Keller–Segel dynamics are strongly constrained by positivity, mass, and free-energy dissipation, successful discretizations are typically structure preserving. The upwind discontinuous Galerkin scheme of (Acosta-Soba et al., 2022) is mass-conservative, positive, and unconditionally energy stable, and it is explicitly designed around the gradient-flow form with chemical potential

$\chi$0

Its discrete free-energy law remains valid without any restriction on $\chi$1. The semi-implicit Euler time discretization

$\chi$2

is linear, decoupled, first order, and preserves mass, positivity, and discrete energy dissipation unconditionally at the semi-discrete level, with optimal $\chi$3-error estimates for $\chi$4 (Huang et al., 3 Mar 2025).

Near singularity formation, hybrid methods become especially useful. The composite particle-grid method of (Fatkullin, 2010) evolves the bacterial density by particles governed by an Euler–Maruyama discretization of the associated stochastic differential equations while solving the chemoattractant on a grid. This permits resolution of both blow-up and post-blow-up singularity interaction, and numerically confirms the motion of point-mass aggregates along Green-function-induced trajectories.

6. Biological reach and cross-disciplinary reinterpretations

Although Keller–Segel originated as a model of chemotaxis in myxamoebas and related systems, its later use has been markedly broader. In bacterial chemotaxis, it remains the canonical PDE description of aggregation and collapse, but kinetic derivations now connect its coefficients to run-and-tumble statistics and saturation mechanisms (Perthame et al., 2018). In C. elegans L1 aggregation, a Keller–Segel-type model with a short-range attractant and long-range repellent reproduces finite aggregate size and spacing, with periodic boundary conditions used to avoid boundary artifacts in the simulations (Avery et al., 2020). In plant patterning, a Keller–Segel-type instability coupled to radial growth on a disc generates Fibonacci spirals, alternating patterns, and whorled arrangements, with linear stability analysis tying pattern selection to the balance between instability growth and tissue growth (Staddon, 8 Sep 2025).

The framework has also been transplanted outside biology. “Dynamic Keller-Segel Model of Population Density and Economic Factors” replaces the chemoattractant by money concentration: $\chi$5 Here population density is initially uniform, money concentration is initially random, and the simulation produces clusters around economic hubs together with wealth concentration that stabilizes after about 30 years, described as “one generation” (Montgomery, 2023). The paper itself presents this as a simplified exploratory simulation and explicitly notes omissions of geography, political influences, cultural factors, transportation networks, individual heterogeneity, formal stability analysis, and detailed numerical validation. This suggests that the Keller–Segel mechanism is portable as a phenomenological aggregation template, but its interpretive reach depends strongly on what is retained or suppressed in the coupling structure.

Across these variants, the unifying content of the Keller–Segel model is not a single canonical formula but a mechanism: diffusion competes with self-consistent, gradient-driven aggregation. The specific form of diffusion, mobility, signal production, saturation, stochastic forcing, and domain geometry determines whether the outcome is global regularity, metastable clustering, traveling waves, compactly supported stationary bumps, or singular concentration.