Attraction-Repulsion Chemotaxis Model

• Attraction–repulsion chemotaxis model is a system of PDEs describing cell density movement driven by opposing chemical gradients, where attractants promote aggregation and repellents induce dispersion.
• The model combines diffusion with nonlinear chemotactic fluxes and reaction kinetics, allowing analysis through techniques like Keller–Segel reduction and energy methods.
• Key insights include the balance between aggregation and repulsion, conditions for global boundedness versus finite-time blow-up, and implications of nonlinear production and damping.

Attraction–repulsion chemotaxis models are PDE systems for a cell density $u(x,t)$ coupled to at least two chemical fields, typically an attractant $v(x,t)$ and a repellent $w(x,t)$, in which directed motion toward higher $v$ and away from higher $w$ competes with diffusion, reaction kinetics, and, in many formulations, logistic damping or nonlinear production and consumption. A canonical form is

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$

with $\chi,\xi>0$, where the term $-\chi \nabla\!\cdot(u\nabla v)$ induces aggregation and $+\xi \nabla\!\cdot(u\nabla w)$ counteracts it. Across the literature, the same attraction–repulsion architecture appears in fully parabolic, parabolic–elliptic, quasilinear, nonlocal, logistic, fractional, and fluid-coupled variants, and the central analytical question is whether the competition among attraction, repulsion, diffusion, and damping yields global boundedness, stabilization, or finite-time blow-up (Chiyo et al., 2021).

1. Canonical formulations and modeling conventions

The basic attraction–repulsion chemotaxis system is posed on a domain $\Omega\subset\mathbb{R}^n$, usually bounded and smooth, with homogeneous Neumann boundary conditions

$v(x,t)$0

so that diffusive and chemotactic fluxes are confined within $v(x,t)$1. In the standard local formulation, $v(x,t)$2 denotes cell density, $v(x,t)$3 a chemoattractant, and $v(x,t)$4 a chemorepellent. The simplest linear-diffusion version takes the form

$v(x,t)$5

supplemented by either parabolic signal equations such as

$v(x,t)$6

or elliptic signal equations such as

$v(x,t)$7

These two regimes are commonly described as fully parabolic and parabolic–elliptic, respectively (Chiyo et al., 2021).

A broad class of nonlinear generalizations replaces linear diffusion and linear sensitivities by density-dependent coefficients. One representative quasilinear system is

$v(x,t)$8

with elliptic equations for $v(x,t)$9 and $w(x,t)$0. Here $w(x,t)$1 models density-dependent diffusion, while $w(x,t)$2 and $w(x,t)$3 encode density-dependent sensitivities to attraction and repulsion (Chiyo, 2022). Another fully nonlinear local model includes a source

$w(x,t)$4

together with nonlinear productions $w(x,t)$5 and $w(x,t)$6 for the chemical fields (Li et al., 2024).

The chemical kinetics vary substantially across applications. In some models both signals are produced by the cells; in others both are consumed; and in mixed production–consumption systems the chemoattractant is produced while the chemorepellent is consumed, or conversely (Li et al., 2022). In the “double saturation” setting,

$w(x,t)$7

both chemicals are consumed, and the saturation terminology refers to the fact that the concentrations remain uniformly bounded by the maximum principle, not to sensitivity saturation (Frassu et al., 2021). Nonlocal variants replace local productions by mean-subtracted source terms, so that $w(x,t)$8 and $w(x,t)$9 represent deviations from their spatial averages and satisfy zero-mean constraints (Columbu et al., 2022).

2. Competition between attraction and repulsion

The defining structural feature of the model is the competition between the attractive flux and the repulsive flux. In the most common sign convention, attraction is encoded by

$v$0

which pulls cells up the attractant gradient, whereas repulsion is encoded by

$v$1

which pushes cells down the repellent gradient. This competition is often quantified not by $v$2 and $v$3 alone, but by effective combinations involving production and decay parameters of the chemicals (Lankeit, 2021).

In linear production models with

$v$4

the quantity

$v$5

is the natural measure of attraction versus repulsion. The regime $v$6 is described as attraction-dominated, while $v$7 is repulsion-dominated (Lankeit, 2021). In the parabolic–parabolic and parabolic–elliptic systems studied with measure-valued initial data, the transformed coupling

$v$8

is used instead, so that “strong repulsion” corresponds to $v$9 (Heihoff, 2022).

A recurrent misconception is that repulsion alone precludes singularity formation. The literature does not support that general statement. Finite-time blow-up persists in attraction-dominated fully parabolic systems in three dimensions, even though repulsion is present (Lankeit, 2021). Conversely, sufficiently strong repulsion, or repulsion combined with damping, can ensure global boundedness and stabilization in many regimes (Chiyo, 2022). This suggests that the analytically relevant issue is not the mere presence of a repellent, but the quantitative balance among attraction, repulsion, diffusion, production, consumption, and logistic or gradient-dependent damping.

The same balance appears in one-dimensional half-line models with logistic growth. There the effective attraction and repulsion strengths are

$w$0

and global boundedness is obtained under conditions of the form

$w$1

where $w$2 is a resolvent-based constant measuring the mismatch of the two elliptic signal operators (Bao et al., 2018).

3. Reduction to effective Keller–Segel structures

One of the most important analytical devices in attraction–repulsion chemotaxis is the reduction of the two-signal system to a single effective signal. In the symmetric fully parabolic model

$w$3

the linear transformation

$w$4

yields

$w$5

This is exactly a parabolic–parabolic Keller–Segel-like system with effective coupling $w$6 (Chiyo et al., 2021).

The structural significance of this reduction is twofold. First, it converts the competing chemotactic drifts into the single advection structure $w$7. Second, it imports the classical Keller–Segel energy machinery. The canonical functional becomes

$w$8

and satisfies the dissipation inequality

$w$9

with

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$0

Under $$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$1, this reproduces the higher-dimensional Keller–Segel blow-up mechanism in the attraction–repulsion setting (Chiyo et al., 2021).

A related transformation is used in the three-dimensional attraction-dominated fully parabolic system with unequal damping and production parameters. Writing

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$2

the system becomes

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$3

When $$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$4, this is the classical parabolic–parabolic Keller–Segel structure; when $$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$5, the exact Lyapunov structure is broken, but an energy method remains possible (Lankeit, 2021).

The reduction idea also appears in repulsion-dominant quasilinear parabolic–elliptic–elliptic systems, but there it no longer closes because the nonlinear sensitivity factors cannot be written as a single $$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$6 drift. The stabilization analysis therefore uses a tailored entropy functional rather than a direct Keller–Segel reduction (Chiyo, 2022).

4. Global boundedness mechanisms

The boundedness theory is organized around several distinct mechanisms. One is the weakening of signal growth through consumption. In the double-saturation model

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$7

the maximum principle yields

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$8

and a weighted functional

$$u_t=\Delta u-\chi \nabla\!\cdot(u\nabla v)+\xi \nabla\!\cdot(u\nabla w),$$9

provides uniform-in-time $\chi,\xi>0$0 bounds on $\chi,\xi>0$1 when

$\chi,\xi>0$2

These $\chi,\xi>0$3 bounds with $\chi,\xi>0$4 imply $\chi,\xi>0$5-boundedness by heat-semigroup smoothing and bootstrap (Frassu et al., 2021).

Another mechanism is sublinear or weaker-than-competing production. In the parabolic–elliptic–elliptic system

$\chi,\xi>0$6

with $\chi,\xi>0$7, the sublinear production of the attractant lowers the power of the aggregation term so that it can be absorbed by diffusion and repulsion. The resulting theorem states that for any $\chi,\xi>0$8, any $\chi,\xi>0$9, and any nonnegative nontrivial $-\chi \nabla\!\cdot(u\nabla v)$0, the problem admits a unique global classical solution that is uniformly bounded in time, with no restriction on the sizes of $-\chi \nabla\!\cdot(u\nabla v)$1 and $-\chi \nabla\!\cdot(u\nabla v)$2 (Pintus et al., 2019).

A third mechanism is nonlinear or nonlocal logistic damping. In systems with

$-\chi \nabla\!\cdot(u\nabla v)$3

the nonlocal damping term controls both the mass and higher $-\chi \nabla\!\cdot(u\nabla v)$4-norms of $-\chi \nabla\!\cdot(u\nabla v)$5. For elliptic signals, global boundedness holds if either

$-\chi \nabla\!\cdot(u\nabla v)$6

or

$-\chi \nabla\!\cdot(u\nabla v)$7

For fully parabolic signals, the corresponding thresholds involve $-\chi \nabla\!\cdot(u\nabla v)$8 and reflect the added difficulty of time-dependent signal dynamics (Fuentes et al., 15 Feb 2026).

Further boundedness mechanisms include density-dependent diffusion, repulsion-dominant sensitivity growth, and gradient-dependent damping. In the local and nonlocal fully nonlinear systems with source

$-\chi \nabla\!\cdot(u\nabla v)$9

global boundedness follows under

$+\xi \nabla\!\cdot(u\nabla w)$0

The dissipation coming from

$+\xi \nabla\!\cdot(u\nabla w)$1

precludes concentration and thus prevents what the paper calls $+\xi \nabla\!\cdot(u\nabla w)$2-formations (Li et al., 2024).

5. Finite-time blow-up and singularity formation

Finite-time blow-up remains a central phenomenon in attraction–repulsion chemotaxis, especially in attraction-dominated regimes. In the symmetric fully parabolic model on a ball $+\xi \nabla\!\cdot(u\nabla w)$3, $+\xi \nabla\!\cdot(u\nabla w)$4, with

$+\xi \nabla\!\cdot(u\nabla w)$5

finite-time blow-up occurs for radially symmetric initial data in a dense class whenever $+\xi \nabla\!\cdot(u\nabla w)$6. More precisely, for any $+\xi \nabla\!\cdot(u\nabla w)$7 and $+\xi \nabla\!\cdot(u\nabla w)$8, there exist $+\xi \nabla\!\cdot(u\nabla w)$9 and $\Omega\subset\mathbb{R}^n$0 such that if the initial data belong to the class $\Omega\subset\mathbb{R}^n$1 with negative energy

$\Omega\subset\mathbb{R}^n$2

then the maximal existence time satisfies $\Omega\subset\mathbb{R}^n$3 and

$\Omega\subset\mathbb{R}^n$4

This closes the previously open $\Omega\subset\mathbb{R}^n$5 gap in the symmetric fully parabolic attraction–repulsion setting (Chiyo et al., 2021).

In the three-dimensional fully parabolic attraction-dominated system

$\Omega\subset\mathbb{R}^n$6

posed on a ball with Neumann boundary conditions, finite-time blow-up occurs whenever

$\Omega\subset\mathbb{R}^n$7

The result is stated by approximation: for every radial positive triple $\Omega\subset\mathbb{R}^n$8 and every $\Omega\subset\mathbb{R}^n$9, there exist arbitrarily close radial initial data in $v(x,t)$00 with the same cell mass whose corresponding classical solutions blow up in finite time (Lankeit, 2021).

The blow-up analysis in fully parabolic systems also yields explicit lower bounds on blow-up time. For

$v(x,t)$01

with

$v(x,t)$02

the energy

$v(x,t)$03

satisfies a superlinear ODI, leading to the explicit estimate

$v(x,t)$04

for admissible $v(x,t)$05. This result does not establish blow-up, but quantifies the minimal blow-up time when blow-up occurs (Le et al., 2022).

6. Extensions, regularization, and broader variants

The attraction–repulsion architecture extends well beyond the basic local Keller–Segel system. One notable direction concerns regularization from singular initial data. In the two-dimensional parabolic–parabolic system and the two- and three-dimensional parabolic–elliptic system,

$v(x,t)$06

may be a positive Radon measure, including sums of Dirac masses. Under sufficiently strong repulsion, classical solutions become instantaneously smooth. The precise conditions are

$v(x,t)$07

for the parabolic–parabolic system in $v(x,t)$08,

$v(x,t)$09

for the parabolic–elliptic system in $v(x,t)$10, and

$v(x,t)$11

for the parabolic–elliptic system in $v(x,t)$12. The solution satisfies

$v(x,t)$13

for every $v(x,t)$14 as $v(x,t)$15 (Heihoff, 2022).

Another direction adds fractional diffusion and whole-space propagation. In

$v(x,t)$16

with elliptic signals

$v(x,t)$17

global boundedness follows under conditions involving

$v(x,t)$18

and, under balanced intensities

$v(x,t)$19

the exact spreading speed is

$v(x,t)$20

This reflects the accelerated propagation induced by fractional diffusion (Song et al., 27 Mar 2026).

Attraction–repulsion models also couple to fluid motion. In the chemotaxis–Navier–Stokes system

$v(x,t)$21

with transport equations for the chemical signals and incompressible fluid velocity, global mild solutions exist on bounded domains in two and three dimensions for small initial data in $v(x,t)$22-spaces (Duarte-Rodríguez et al., 2017). At a different scale, a field-theoretic model for run-and-tumble particles uses a density field $v(x,t)$23, polarization field $v(x,t)$24, and chemical field $v(x,t)$25, with chemotaxis entering through a biased tumbling rate

$v(x,t)$26

thereby recovering Keller–Segel-type drift in an appropriate closure while retaining particle-resolution, elasticity, and soft-core repulsion (Chatterjee et al., 2020).

Across these variants, a coherent picture emerges. Repulsion can regularize measure-valued data, enlarge boundedness regimes, or determine spreading behavior, but it does not universally eliminate blow-up. A plausible implication is that attraction–repulsion chemotaxis should be understood not as a single model with one threshold, but as a family of competing-drift systems whose long-time behavior depends on which structural mechanism—effective attraction, repulsion, sublinear production, signal consumption, nonlinear diffusion, logistic damping, nonlocal feedback, or gradient dissipation—dominates in the relevant scaling regime (Heihoff, 2022).