betaFeedback

WorksAgricultural and Biological Sciences

Diversification under Global Competition: A Cutoff-Induced Localization Mechanism

Abstract

Niche differentiation promotes biological diversity but does not necessarily eliminate competition for shared limiting resources. What determines the extent of diversity when evolutionary diversification takes place under such constraints? We investigate this question using a simple, analytically tractable model combining mutations in continuous trait space, local competition, and global competition for shared resources. We show that, without an establishment threshold, the population spreads indefinitely in trait space even as its total abundance approaches a finite limit. In contrast, an arbitrarily small establishment threshold allows a stable localized distribution. Together with mortality induced by global competition, the threshold creates bistability, and global feedback drives the system exponentially close to the Maxwell point, where an isolated invasion front is stationary. The resulting soliton has height H∝ε2/3H\propto\varepsilon^{2/3} and width W∝ε−2/3W\propto\varepsilon^{-2/3}. Simulations with discrete individuals reveal an archipelago of fluctuating clusters whose spatial extent eventually reaches statistical saturation, unlike the continuous deterministic soliton. Nevertheless, an effective cutoff εeff≃0.8/(Nμ)\varepsilon_{\mathrm{eff}}\simeq0.8/(N\sqrtμ) reproduces the observed scaling of the population mass deficit. Remarkably, both occupied trait-site richness and effective species richness exhibit approximately the same scaling as the deterministic soliton width. Thus, deterministic cutoff theory captures global diversity scaling despite failing to reproduce the community's spatial organization.

Introduction

The coexistence of what Darwin called “endless forms most beautiful and most wonderful” is a source of wonder, but also poses a scientific puzzle: how can such an extraordinary diversity be sustained in a world of limited resources? One of the central answers offered by evolutionary ecology is niche differentiation: species with different traits may specialize in exploiting different resources, thereby reducing competition among themselves [1, 2]. Yet in many ecological systems, niche differentiation does not eliminate competition for shared resources. Two tree species specializing in different nutrients still compete for water, sunlight, and space; in general, species occupying distinct niches may remain dependent on common limiting resources. What, then, determines the species richness that can be sustained under such conditions? Under what circumstances can evolutionary diversification continue despite competition for shared resources?

Ecological limits to diversification need not arise from a shortage of available niches. Storch and Okie argued that, although the addition of species may increase total community abundance by allowing more complete utilization of resources, this increase is ultimately constrained by finite resource and energy availability. Consequently, as species richness grows, individual populations tend to become smaller and more vulnerable to extinction, potentially limiting diversity even when niche space remains unsaturated [3]. A complementary perspective was provided by Tilman, who showed that the establishment of new species depends on the resources left available by resident species within a habitat, as well as on the stochastic survival of initially rare colonists [4].

Theoretical studies have also shown how trait-dependent competition can lead to differentiation and clustering in niche space [2, 5]. Yet the emergence of distinct niches does not, by itself, determine how far diversification can proceed when populations occupying different niches remain constrained by shared resources. This raises a more specific question: how do global competition for shared resources and the ability of initially small populations to establish themselves in previously unoccupied niches jointly determine the extent of diversification in trait space?

Here we introduce a simple, analytically tractable model of evolutionary diversification. Mutations allow a population to spread along a continuous axis of ecological traits; local competition limits the density of individuals with similar traits, while global competition for shared resources constrains the total population size. We examine how these ingredients determine the extent of ecological diversity, represented in the model by the width of the trait distribution. In the simplest version of the dynamics, the population continues to spread indefinitely in trait space even though its total mass approaches a finite limit. Consequently, an ever wider range of niches is occupied at progressively lower average density.

The unbounded spreading predicted by the deterministic continuum model relies on the assumption that reproduction remains possible at arbitrarily low local densities. In real populations, reproduction at low densities may be suppressed by an ecological Allee effect [6]; moreover, even in the absence of such an effect, demographic discreteness limits the validity of the continuum description and generates an effective cutoff [7, 8]. We therefore begin by examining the deterministic consequences of a sharp establishment threshold, below which reproduction is suppressed, and subsequently investigate what changes when demographic fluctuations are treated explicitly. We show that even an arbitrarily small threshold qualitatively changes the dynamics, allowing a stable localized trait distribution. The analytical tractability of the model enables us to construct the family of stationary localized solutions, determine their conditions of existence, and examine which branch is dynamically selected.

Physically, the establishment cutoff, together with the mortality imposed by global competition, creates bistability between empty and populated states. Global competition then provides feedback that nearly balances the invasion of empty niches against the retreat of populated ones, driving the system exceptionally close to the Maxwell point at which an isolated front is stationary. We show that the distance from this point is exponentially small in the inverse global competition strength (with a cutoff-dependent scale), whereas the width of the localized profile diverges logarithmically as the Maxwell point is approached. These results lead to nontrivial scaling laws: as the threshold ε\varepsilon tends to zero, the profile height scales as H∝ε2/3H \propto\varepsilon^{2/3}, while its width scales as W∝ε−2/3W \propto\varepsilon^{-2/3}.

Finally, we return to the demographic interpretation of the establishment threshold and ask how faithfully a deterministic cutoff can describe a population composed of discrete individuals. Previous studies have shown that demographic fluctuations can generate clusters in trait space even when the corresponding deterministic dynamics does not predict amplitude oscillations [9, 10]. We make a quantitative comparison between the individual-based model and the deterministic soliton, whose properties have been characterized analytically. We find that whereas the deterministic cutoff produces a continuous, localized profile, the stochastic dynamics generates an ever-changing archipelago of separated clusters whose spatial extent eventually reaches statistical saturation. Despite this fundamental difference in spatial organization, the analytical soliton prediction, with the cutoff identified as an effective discreteness scale, approximately captures the scaling of the population mass deficit in the individual-based model. This deficit, g=1−κ⟨M⟩g = 1 - \kappa\langle M\rangle, measures the fractional shortfall of the mean total population mass ⟨M⟩\langle M\rangle from its limiting value 1/κ1/\kappa in the absence of a cutoff. Moreover, the number of occupied trait niches and the effective species richness exhibit approximately the same scaling as the deterministic soliton width, although the spatial extent of the stochastic population is substantially larger. These results demonstrate that a simple deterministic description can capture the scaling laws of global ecological properties even when it fails to reproduce the spatial organization of the community.

A SOLVABLE ONE-DIMENSIONAL CONTINUUM MODEL OF LINEAGE LOCALIZATION

In this section, we introduce a minimal continuum model of evolution in trait space. We assume that xx represents a phenotypic trait and that ϕ(x,t)\phi(x,t) denotes the population density along this trait axis at time tt. During reproduction, offspring inherit their parents’ traits with small random modifications, so that, in the continuum limit, mutations are represented by diffusion in trait space with diffusion coefficient DD. We assume that competition has two components: local competition between individuals with similar traits, represented by a quadratic density-dependent term, and global competition for a shared limiting resource, which persists even as organisms diversify into distinct ecological niches. The strength of global competition is proportional to the total population, M(t)=∫ϕ(x,t) dxM(t) = \int\phi(x,t)\,dx. Finally, we introduce a local establishment threshold ε\varepsilon, below which reproduction is suppressed while competition-induced mortality remains operative. This threshold admits two distinct interpretations. It may represent a genuine ecological constraint, such as an idealized Allee effect, in which small local populations cannot reproduce effectively. Alternatively, it may serve as a deterministic approximation to demographic discreteness, following the cutoff approach of Brunet and Derrida [7]. These ingredients lead to a nonlinear reaction–diffusion equation, a nonlocal form of the cut-off Fisher-KPP equation [11] which we use to investigate how establishment thresholds and global competition for shared resources jointly regulate the spread of an evolving population in trait space.

Under these assumptions, the dynamics of the population density in trait space is governed by

∂ϕ(x,t)∂t=D∂2ϕ(x,t)∂x2+ϕ(x,t)Θ(ϕ(x,t)−ε)−ϕ2(x,t)−κM(t)ϕ(x,t),(1)\frac{\partial\phi(x,t)}{\partial t}=D\frac{\partial^2\phi(x,t)}{\partial x^2}+\phi(x,t)\Theta\left(\phi(x,t)-\varepsilon\right)-\phi^2(x,t)-\kappa M(t)\phi(x,t), \tag*{(1)}

where

M(t)=∫−∞∞ϕ(x,t) dx.(2)M(t)=\int_{-\infty}^{\infty}\phi(x,t)\,\mathrm{d}x. \tag*{(2)}

Here, Θ\Theta is the Heaviside step function, and κ\kappa controls the strength of global competition. We have chosen units in which both the reproduction rate and the coefficient of local competition are unity. Reproduction therefore occurs only where the population density exceeds the threshold ε\varepsilon, whereas mortality due to local and global competition operates at all densities. Crucially, the global competition term depends on the integral over the entire trait space, so that populations occupying widely separated regions of this space continue to compete for the same shared resource.

We wish to determine whether the dynamics can support a stationary state in which the population remains localized within an effectively bounded region of trait space, despite the continuous spreading induced by mutations. Such a state does not imply that reproduction and mutation have ceased, but rather that a dynamical balance has been established in which the trait distribution remains constant in time. We therefore seek a stationary profile ϕ(x,t)=ϕ(x)\phi(x,t)=\phi(x) with finite total mass, satisfying ϕ(x)≥0\phi(x)\ge0 and ϕ(x)→0\phi(x)\to0 as ∣x∣→∞|x|\to\infty. Since the model is translationally invariant, we may center the profile at x=0x=0 and seek a symmetric solution. Importantly, the strength of global competition is determined by the total population in addition to κ\kappa. Defining r≡κMr\equiv\kappa M, we first seek localized stationary solutions for a given rr, and then impose the self-consistency condition

r=κ∫−∞∞ϕr(x) dx,(3)r=\kappa\int_{-\infty}^{\infty}\phi_r(x)\,\mathrm{d}x, \tag*{(3)}

where ϕr(x)\phi_r(x) denotes the stationary profile obtained for that value of rr.

In the stationary state, the equation of motion then becomes

Dd2ϕdx2+ϕΘ(ϕ−ε)−ϕ2−rϕ=0.(4)D\frac{\mathrm{d}^2\phi}{\mathrm{d}x^2}+\phi\Theta(\phi-\varepsilon)-\phi^2-r\phi=0. \tag*{(4)}

To find localized solutions, we employ the standard technique of a mechanical analogy in which the trait coordinate xx plays the role of time, while ϕ(x)\phi(x) represents the position of a particle moving, with no friction, in an effective potential U(ϕ)U(\phi). The stationary equation can then be written as

Dd2ϕdx2=−dUdϕ.(5)D \frac{d^2\phi}{dx^2}=-\frac{dU}{d\phi}. \tag*{(5)}

The effective potential is obtained by integrating Eq. (5). The integration constants are fixed by choosing U(0)=0U(0)=0 and requiring the potential to be continuous at ϕ=ε\phi=\varepsilon, where the reproduction term switches on. This gives

U(ϕ)={−r2ϕ2−13ϕ3,0≤ϕ<ε,1−r2ϕ2−13ϕ3−ε22,ϕ≥ε.(6)U(\phi)= \begin{cases} -\frac{r}{2}\phi^2-\frac{1}{3}\phi^3, & 0\leq\phi<\varepsilon,\\ \frac{1-r}{2}\phi^2-\frac{1}{3}\phi^3-\frac{\varepsilon^2}{2}, & \phi\geq\varepsilon. \end{cases} \tag*{(6)}

The mechanical analogy provides a simple way to identify localized stationary solutions. Since we seek a profile for which both ϕ\phi and its derivative vanish as x→±∞x\to\pm\infty, and since U(0)=0U(0)=0, conservation of mechanical energy requires

D2(dϕdx)2+U(ϕ)=0.(7)\frac{D}{2}\left(\frac{d\phi}{dx}\right)^2+U(\phi)=0. \tag*{(7)}

A localized solution corresponds to a homoclinic orbit: the particle starts at ϕ=0\phi=0 as x→−∞x\to-\infty, moves towards positive values of ϕ\phi, reaches a turning point ϕ=A\phi=A, and returns to ϕ=0\phi=0 as x→+∞x\to+\infty. Choosing the turning point at x=0x=0, the corresponding population profile satisfies

ϕ(−∞)=ϕ(+∞)=0,ϕ(0)=A,ϕ′(0)=0.(8)\phi(-\infty)=\phi(+\infty)=0,\qquad\phi(0)=A,\qquad\phi'(0)=0. \tag*{(8)}

The turning point, at which the kinetic energy vanishes, thus satisfies

U(A)=0.(9)U(A)=0. \tag*{(9)}

Eq. (7) implies

(dϕdx)2=−2U(ϕ)D.(10)\left(\frac{d\phi}{dx}\right)^2=-\frac{2U(\phi)}{D}. \tag*{(10)}

Thus, a localized solution requires the potential to be negative between the origin and the turning point,

U(ϕ)<0,0<ϕ<A.(11)U(\phi)<0,\qquad0<\phi<A. \tag*{(11)}

As ϕ\phi increases from 0, the potential initially decreases till ϕ=ε\phi=\varepsilon, then starts rising and eventually reaches a maximum beyond which it decreases (Figure 1). Thus the maximum value of UU has to be positive for there to be a turning point AA. The local maximum of the potential occurs at ϕm=1−r\phi_m = 1-r, where

Mechanical potential of the one-dimensional continuum model

Figure 1. Mechanical potential of the one-dimensional continuum model. A representative example for ε=0.4\varepsilon= 0.4. For r<rcr < r_c (solid), the zero-energy orbit leaves ϕ=0\phi= 0, reaches the first positive zero AA of UU, and returns, giving a localized stationary pulse. At r=rcr = r_c (dashed), the right-hand maximum touches U=0U = 0 at ϕc=(3ε2)1/3\phi_c = (3\varepsilon^2)^{1/3}; the turning point becomes marginal and the pulse width diverges. For r>rcr > r_c, the maximum lies below zero and no zero-energy localized pulse exists.

U(ϕm)=(1−r)36−ε22.(12)U(\phi_m) = \frac{(1-r)^3}{6} - \frac{\varepsilon^2}{2}. \tag*{(12)}

Requiring U(ϕm)>0U(\phi_m) > 0 yields

0<r<rc(ε)≡1−31/3ε2/3,(13)0 < r < r_c(\varepsilon) \equiv1 - 3^{1/3}\varepsilon^{2/3}, \tag*{(13)}

provided the cutoff is sufficiently small, ε<1/3\varepsilon< 1/\sqrt{3}, so that rc>0r_c > 0.

For every 0<r<rc0 < r < r_c, we have, therefore, a localized stationary solution ϕr(x)\phi_r(x). However, in the original model, rr is not an independent parameter but must satisfy the self-consistency condition (Eq. (3)) r=κMr = \kappa M. Thus, each soliton in this family can be associated with a global competition strength for which it is a stationary solution:

κ(r)=rM(r),M(r)=∫−∞∞ϕr(x) dx.(14)\kappa(r) = \frac{r}{M(r)}, \qquad M(r) = \int_{-\infty}^{\infty} \phi_r(x)\,\mathrm{d}x. \tag*{(14)}

Figure 2 shows κ(r)\kappa(r) for ε=0.4\varepsilon= 0.4 and D=1D = 1. The function is nonmonotonic and exhibits a maximum. Consequently, a given global competition strength may admit either zero or two distinct stationary solutions. For example, the horizontal line corresponding to κ=0.04\kappa= 0.04 intersects the curve at two points, r1r_1 and r2r_2, representing solitons with different heights and widths, as illustrated by the profiles shown in the insets. The existence of these stationary solutions does not, by itself, establish their dynamical stability.

Self-consistency along the soliton family

Figure 2. Self-consistency along the family of localized stationary solutions. For ε=0.4\varepsilon= 0.4 and D=1D = 1, each 0<r<rc0 < r < r_c admits a localized solution ϕr(x)\phi_r(x) with total mass M(r)M(r). The solid curve shows the corresponding global competition strength, κ(r)=r/M(r)\kappa(r) = r/M(r). For the chosen value κ0=0.04\kappa_0 = 0.04 (horizontal dashed line), two intersections are obtained, at r1≃0.12r_1 \simeq0.12 and r2≃0.21r_2 \simeq0.21, representing two distinct stationary solutions of the same model. The maximum κmax⁡≃0.05\kappa_{\max} \simeq0.05 and the critical value rc≃0.22r_c \simeq0.22 are indicated by dotted lines. The insets show the corresponding density profiles, with dotted horizontal lines marking the reproduction cutoff ε\varepsilon. As r→rc−r \to r_c^{-}, the mass diverges logarithmically and κ(r)\kappa(r) approaches zero; this limiting behavior occurs on a scale too narrow to be resolved in the main plot.

To understand the behavior of the localized solutions, we first consider the limit in which the reproduction cutoff ε\varepsilon approaches zero. In the absence of this cutoff, our model reduces to a Fisher-type reaction–diffusion equation with global competition between individuals, regardless of their separation in trait space. While global competition limits the total population, it cannot prevent its spreading in trait space. In particular, as we demonstrate in Appendix § 6, even in the presence of infinite-range interactions, the population distribution continues to spread indefinitely when ε=0\varepsilon= 0. Since the total population remains bounded, this spreading is accompanied by a progressive decrease in population density. Thus, the reproduction cutoff ε\varepsilon provides the mechanism that allows a localized stationary profile to exist, and we expect its width to diverge as ε→0\varepsilon\to0. We therefore focus on the broad branch of stationary solutions, which turns out to be dynamically stable, whereas the narrow branch is unstable, as discussed in Appendix § 7.

From the mechanical analogy, we know that this divergence occurs as rr approaches its critical value rcr_c from below. In this limit, the particle spends an increasingly long time near the maximum of the effective potential, producing a broad, nearly flat population profile whose height approaches the position of that maximum.

Using the critical value rcr_c obtained in Eq. (13), the soliton height satisfies

H≡max⁡xϕ(x)≃1−rc=(3ε2)1/3.(15)H \equiv\max_{x}\phi(x) \simeq1-r_c=(3\varepsilon^2)^{1/3}. \tag*{(15)}

On the other hand, the self-consistency condition requires

M=rκ≃rcκ⟶1κ(ε→0).(16)M=\frac{r}{\kappa}\simeq\frac{r_c}{\kappa}\longrightarrow\frac{1}{\kappa}\qquad(\varepsilon\to0). \tag*{(16)}

Thus, although the soliton height vanishes, its total mass remains finite. Defining the effective soliton width as the ratio of its total mass to its height, we obtain

W≡MH≃1κ(3ε2)1/3.(17)W\equiv\frac{M}{H}\simeq\frac{1}{\kappa(3\varepsilon^2)^{1/3}}. \tag*{(17)}

We therefore arrive at the scaling laws

H∝ε2/3,W∝ε−2/3,M⟶1κ.(18)\begin{aligned} H &\propto\varepsilon^{2/3},\\ W &\propto\varepsilon^{-2/3},\\ M &\longrightarrow\frac{1}{\kappa}. \tag*{(18)} \end{aligned}

Interestingly, the diffusion coefficient DD does not appear in the leading-order expressions for the soliton height and width. In this limit, the height is determined by the reproduction cutoff, the total mass by global competition, and the width follows from their ratio. These scaling laws are confirmed numerically in Figure 3(B)–(D).

Numerical plots of soliton profiles, peak height, width, total mass, and distance from criticality

Figure 3. Numerical verification of soliton scaling laws. Results for fixed global competition strength κ=0.04\kappa= 0.04 and diffusion coefficient D=1D = 1. (A) Stationary density profiles ϕ(x)\phi(x) on the broad soliton branch for decreasing values of the reproduction cutoff ε\varepsilon. As the cutoff decreases, the soliton becomes broader and lower, developing a nearly flat plateau. The horizontal dotted lines indicate the corresponding cutoff values. (B) Peak height HH as a function of ε\varepsilon. The dashed line shows the asymptotic prediction H=(3ε2)1/3H = (3\varepsilon^2)^{1/3}. (C) Effective width W=M/HW = M/H and full width at half maximum (FWHM). Both approach the parameter-free asymptotic prediction W=1/[κ(3ε2)1/3]W = 1/[\kappa(3\varepsilon^2)^{1/3}] (dashed line). (D) Total mass MM as a function of ε\varepsilon. The dashed horizontal line marks the limiting value 1/κ1/\kappa, while the dotted curve shows the approximation M≃rc/κM \simeq r_c/\kappa, where rc=1−(3ε2)1/3r_c = 1 - (3\varepsilon^2)^{1/3}. (E) Exponential approach to criticality. Writing a=(3ε2)1/3a = (3\varepsilon^2)^{1/3} and δ=rc−r\delta= r_c - r, we plot ln⁡(δ/a)\ln(\delta/a) against rc/(κDa)r_c/(\kappa\sqrt{Da}). The dashed line has the theoretically predicted slope −1-1, with a fitted additive constant. Blue circles represent self-consistent stationary solutions calculated from the full mechanical energy integral. Red squares show the long-time states obtained by direct numerical integration of the time-dependent equation, confirming convergence to the broad stationary branch. The results in panel (E) were obtained independently of the asymptotic approximation, by solving the full self-consistency condition using arbitrary-precision arithmetic with δ\delta as the numerical parameter.

We can further characterize how the broad soliton approaches the critical point. Let us define

a=1−rc=(3ε2)1/3,δ=rc−r.(19)a=1-r_c=(3\varepsilon^2)^{1/3},\qquad\delta=r_c-r. \tag*{(19)}

As δ→0+\delta\to0^{+}, the local maximum of the effective potential approaches zero energy. In the mechanical analogy, the particle spends an increasingly long time near a quadratic maximum, causing the soliton width to diverge logarithmically:

W∼Daln⁡(10.34aδ).(20)W \sim\sqrt{\frac{D}{a}} \ln\left(\frac{10.34a}{\delta}\right). \tag*{(20)}

Since the profile becomes broad and nearly flat, with a height approaching aa, its total mass also diverges logarithmically:

M∼Daln⁡(0.8187aδ).(21)M \sim\sqrt{Da}\ln\left(\frac{0.8187a}{\delta}\right). \tag*{(21)}

However, the self-consistency condition requires

M=rκ≃1κ.(22)M = \frac{r}{\kappa} \simeq\frac{1}{\kappa}. \tag*{(22)}

Consequently,

ln⁡(0.8187aδ)∼1κDa.(23)\ln\left(\frac{0.8187a}{\delta}\right) \sim\frac{1}{\kappa\sqrt{Da}}. \tag*{(23)}

Substituting this result into Eq. (20) gives W∼1/(κa)∝ε−2/3W \sim1/(\kappa a) \propto\varepsilon^{-2/3}, in agreement with Eq. (18). This result fixes the leading exponential dependence of δ\delta on 1/κ1/\kappa,

δ≃0.8187 aexp⁡{−1/(κDa)}.\delta\simeq0.8187\,a\exp\{-1/(\kappa\sqrt{Da})\}.

Thus, self-consistency drives the stable soliton exponentially close to the critical point as ε→0\varepsilon\to0, as confirmed numerically in Figure 3(E). This proximity allows the soliton to become extremely broad even as its height vanishes.

For completeness, we briefly examine the asymptotic decay of the soliton tails. In the region where ϕ<ε\phi< \varepsilon, reproduction is suppressed, and at sufficiently low densities the stationary equation reduces to

Dϕ′′−rϕ≃0.(24)D\phi'' - r\phi\simeq0. \tag*{(24)}

For r>0r > 0, the tails therefore decay exponentially,

ϕ(x)∝exp⁡(−rD∣x∣),(25)\phi(x) \propto\exp\left(-\sqrt{\frac{r}{D}}|x|\right), \tag*{(25)}

with a characteristic decay length

ℓtail=Dr.(26)\ell_{\mathrm{tail}} = \sqrt{\frac{D}{r}}. \tag*{(26)}

Formally, when r=0r = 0, the local competition term must be retained, yielding an algebraic decay,

ϕ(x)∼6Dx2.(27)\phi(x) \sim\frac{6D}{x^2}. \tag*{(27)}

However, this limit is not relevant to the broad self-consistent soliton considered here. As ε→0\varepsilon\to0 at fixed κ\kappa, we have r→rc→1r \to r_{\mathrm{c}} \to1, so that ℓtail→D\ell_{\mathrm{tail}} \to\sqrt{D} remains finite. The divergence of the soliton width, W∝ε−2/3W \propto\varepsilon^{-2/3}, therefore originates from the expansion of its nearly flat, populated interior rather than from the dilute tails.

Demographic discreteness and deterministic establishment thresholds

We have so far represented the failure of small populations to establish themselves by a deterministic reproductive cutoff. This approximation has a natural demographic interpretation, but it is not equivalent to an explicit population of discrete individuals. A single individual arriving in an unoccupied niche can reproduce, although its descendants may subsequently become extinct through demographic fluctuations. In contrast, the deterministic cutoff suppresses reproduction altogether below a prescribed density. The question is therefore which predictions of the localized state survive when the cutoff is replaced by stochastic birth, death, and mutation. Previous work has shown that deterministic cutoffs can capture aspects of front propagation while missing spreading driven by stochastic cluster formation [10]. Our analytically characterized localized solution provides an opportunity to test this correspondence quantitatively.

We perform this comparison on a one-dimensional lattice of trait values, keeping the mutation and competition rules fixed while changing the description of the population. In the individual-based model, site ii contains an integer number nin_i of individuals, with density ϕi=ni/N\phi_i = n_i/N and total rescaled population M=∑iϕiM = \sum_i \phi_i. Each individual reproduces at rate one, hops to either neighboring site at rate μ\mu per direction, and dies at a rate

ni−1N+κM.(28)\frac{n_i - 1}{N} + \kappa M. \tag*{(28)}

The first term represents competition with other individuals at the same site, and the second represents competition for shared global resources. There is no reproductive cutoff or Allee effect in this stochastic model: the per-capita reproduction rate is independent of local density. As a complementary approximation, we simulate demographic noise in a continuous-density lattice model using the scheme developed by [12], which builds on the split-step algorithm introduced by [13], with a leading-order noise variance of 2ϕi/N2\phi_i/N per unit time. Finally, we evolve the deterministic lattice equation with a reproductive cutoff,

ϕ˙i=μ(ϕi+1+ϕi−1−2ϕi)+ϕiΘ(ϕi−ε)−ϕi2−κMϕi.(29)\dot{\phi}_i = \mu(\phi_{i+1} + \phi_{i-1} - 2\phi_i) + \phi_i\Theta(\phi_i - \varepsilon) - \phi_i^2 - \kappa M\phi_i. \tag*{(29)}

The use of the same trait lattice separates the effects of discreteness in trait space from those of demographic fluctuations.

Demographic fluctuations are known to generate spontaneous spatial clustering even in the absence of attractive interactions between individuals [14, 15, 16]. More directly relevant to evolutionary diversification, Rogers, McKane, and Rossberg demonstrated that demographic noise can produce clusters in trait space even when the corresponding deterministic dynamics does not predict pattern formation [9]. Figure 4(A) illustrates this effect in our model: the continuous deterministic soliton is replaced by an archipelago of spatially separated population clusters, each comprising individuals with similar traits. In an evolutionary interpretation, these clusters may be viewed as groups of closely related species, analogous to genera separated in trait space. Following an initial period of expansion, the stochastic population reaches a statistically stationary spatial extent, as shown in the inset, although its saturated width is substantially larger than that of the deterministic soliton. Individual clusters continue to fluctuate, migrate, appear, and disappear, so that the stochastic state is not a stationary density profile but a dynamically maintained, fragmented population. This pronounced difference in spatial organization raises a natural question: to what extent do the analytical predictions obtained for the deterministic soliton remain applicable to the stochastic population?

Comparison between the deterministic soliton and the discrete-individual population

Figure 4. Comparison between the deterministic soliton and the discrete-individual population. (A) Population density in trait space for N=100N = 100, κ=0.04\kappa= 0.04, and μ=1\mu= 1. The dark-blue curve shows the deterministic soliton with εeff=0.80/N\varepsilon_{\mathrm{eff}} = 0.80/N; the red profile is a Gillespie realization at t=106t = 10^{6}. The dotted line marks the single-individual density 1/N1/N. The inset shows the evolution of W95W_{95} for three realizations (thin lines), their eight-realization mean (thick line), and the deterministic soliton width. (B) Stationary mass deficit g=1−κ⟨M⟩g = 1 - \kappa\langle M\rangle versus NμN\sqrt{\mu}. Orange and blue triangles show Gillespie results for κ=0.04\kappa= 0.04 and 0.40.4, respectively; teal-filled and open grey symbols show split-step and discrete-time simulations. Marker orientation distinguishes mutation rates. For κ=0.04\kappa= 0.04, averages are taken over the stationary regime of 8 realizations (16 at N=1000N = 1000). The solid line is the analytical prediction g=(3ε2)1/3g = (3\varepsilon^{2})^{1/3} with ε=1/(Nμ)\varepsilon= 1/(N\sqrt{\mu}); the dashed line uses εeff=0.80/(Nμ)\varepsilon_{\mathrm{eff}} = 0.80/(N\sqrt{\mu}), with a single fitted coefficient. (C) Spatial extent W95W_{95} (triangles), occupied trait sites SS (circles), and effective species richness SeffS_{\mathrm{eff}} (squares), for μ=1\mu= 1. Filled and open symbols correspond to κ=0.4\kappa= 0.4 and 0.040.04, respectively. Solid and dashed lines show the corresponding deterministic soliton widths WsolW_{\mathrm{sol}} with εeff=0.80/N\varepsilon_{\mathrm{eff}} = 0.80/N. The inset shows the ratios to WsolW_{\mathrm{sol}}. The indicated exponents are power-law fits over the investigated range.

We begin with the total population mass. To compare the analytical prediction for the deterministic soliton with the individual-based model, we must identify the reproductive cutoff ε\varepsilon with an appropriate discreteness scale in the stochastic system. Since a single individual at a lattice site corresponds to a density 1/N1/N, this would appear to be the natural scale. However, diffusion reduces local demographic fluctuations by mixing individuals between neighboring sites. When distances are measured in units of the diffusive length scale μ\sqrt{\mu}, the demographic noise strength is controlled by the combination NμN\sqrt{\mu} rather than by NN alone. This suggests an effective cutoff of the form εeff=c/(Nμ)\varepsilon_{\mathrm{eff}} = c/(N\sqrt{\mu}), where the numerical prefactor cc is not determined by this scaling argument. For the deterministic soliton, we found that the mass deficit obeys g=1−κM≃(3ε2)1/3g = 1 - \kappa M \simeq(3\varepsilon^2)^{1/3}. Figure 4(B) shows that the mean mass deficit in the stochastic model is well described by the same expression, with εeff≃0.8/(Nμ)\varepsilon_{\mathrm{eff}} \simeq0.8/(N\sqrt{\mu}). This agreement holds across different values of the population-size parameter NN, the mutation rate μ\mu, and the global competition strength κ\kappa.

Next we examine whether a similar correspondence holds for the extent and diversity of the stochastic population in trait space. Figure 4(C) compares the width of the deterministic soliton with three measures of the stochastic population: the spatial extent W95W_{95} containing 95% of all individuals, the number of occupied trait sites S=∑iΘ(ni)S = \sum_i \Theta(n_i), which may be interpreted as species richness, and the effective species richness Seff=(∑ini)2/∑ini2S_{\mathrm{eff}} = (\sum_i n_i)^2/\sum_i n_i^2, which accounts for differences in abundance among species. Despite the fragmented structure of the stochastic population, both measures of richness exhibit approximately the same N2/3N^{2/3} scaling as the deterministic soliton width. In particular, the ratio of occupied sites to soliton width approaches a value of approximately 1.6, while the ratio S/SeffS/S_{\mathrm{eff}} remains nearly constant. The spatial extent W95W_{95} also increases with NN, but is substantially larger and grows somewhat faster over the investigated range, reflecting the empty gaps separating the clusters. Thus, although the deterministic theory does not reproduce the spatial organization of the stochastic population, it approximately captures the scaling of the number of occupied trait niches.

In summary, the comparison between the deterministic and individual-based models shows that an effective reproductive cutoff can approximately reproduce the scaling laws of total population mass and species richness, even though it fails to capture the spatial organization of the population. Whereas the deterministic cutoff produces a continuous, localized soliton, demographic fluctuations generate a dynamic archipelago of separated clusters spanning a substantially broader range of trait space. This distinction highlights two different interpretations of the reproductive cutoff. A genuine ecological establishment threshold, such as an Allee effect, may arrest diversification through the front-pinning mechanism analyzed here. Demographic discreteness, in contrast, does not completely prevent small populations from establishing, but instead gives rise to stochastic dynamics in which clusters continually emerge and disappear. Despite these differences, our results suggest that a simple deterministic description can correctly predict the scaling of global ecological properties even when it fails to reproduce the spatial organization of the community.

Discussion

A central question in evolutionary ecology is what determines the extent of biological diversity when differentiation into new niches reduces competition among species but does not eliminate competition for shared resources. Our results suggest that resource limitation, even when it regulates the total population size, is not sufficient by itself to constrain the spread of a population in trait space: without an establishment threshold, the population occupies an ever-widening range of traits at ever lower density, whereas an arbitrarily small threshold allows a stable localized distribution. Thus, the extent of diversity depends not only on the availability of potential niches or the total amount of resources, but also on the ability of initially small populations to establish themselves in new niches. The analytical tractability of the model makes this dependence quantitative, W∝ε−2/3W \propto\varepsilon^{-2/3}. Although the width of a trait distribution is not equivalent to species richness, it provides a quantitative measure of the extent of ecological differentiation and of the constraints acting upon it.

The localization mechanism identified here differs from the pattern-forming instabilities found in models with nonlocal competition, where trait-dependent interactions can destabilize homogeneous states and generate spatial patterns [5, 17, 18]. Our model has no such finite-wavelength instability. Instead, the establishment threshold, together with mortality imposed by global competition, creates bistability between empty and populated states, allowing localization through the arrest of invasion fronts. As the population expands, global competition increases and suppresses further invasion of unoccupied niches; as it contracts, competition weakens. In the small-cutoff regime, this feedback drives the system exponentially close to the Maxwell point, where invasion and retreat of an isolated front are balanced. The logarithmic divergence of the profile width near this point, combined with the global self-consistency condition, explains how a finite extent is selected without an externally imposed niche-scale length. Front arrest and localized colonies have previously been studied in systems with Allee effects and nonlocal interactions [6, 19, 20, 21]. The distinctive contribution of the present model is an analytically tractable description of how a local establishment threshold and global resource competition jointly select the extent of evolutionary diversification.

The establishment threshold admits two biological interpretations, and distinguishing between them is essential. When it represents an Allee effect, it describes a genuine ecological limitation on the ability of small populations to establish themselves, and the front-arrest mechanism analyzed here provides a possible explanation for the limitation of diversification into new niches. When the cutoff serves instead as a deterministic proxy for demographic discreteness, a more complex picture emerges. As established in previous studies, demographic fluctuations can generate separated clusters in trait space even when the corresponding deterministic dynamics does not predict such patterns [9, 15, 16]. Nevertheless, our quantitative comparison shows that, despite this fundamental difference in spatial organization, the deterministic cutoff approximately reproduces the scaling of the population mass deficit. when identified with an effective discreteness scale, εeff∼1/(Nμ)\varepsilon_{\mathrm{eff}} \sim1/(N\sqrt{\mu}). Moreover, both the number of occupied trait niches and the effective species richness exhibit approximately the same N2/3N^{2/3} scaling as the deterministic soliton width. Thus, the deterministic theory captures the scaling of global community properties even though it fails to describe the spatial organization of the population.

The difference between the two descriptions is particularly evident in the role of rare events at the edges of the population. In the deterministic cutoff model, reproduction is completely suppressed below the threshold, preventing invasion fronts from advancing beyond the populated region. In the stochastic model, however, even a single individual arriving in an unoccupied niche may reproduce and establish a new population. Such events, together with local extinctions, generate a dynamic archipelago of clusters that appear, disappear, and migrate through trait space. Our simulations show that, following an initial period of expansion, the spatial extent of this archipelago also reaches statistical saturation, although its saturated width is substantially larger than that of the deterministic soliton. Furthermore, whereas the number of occupied trait niches approximately follows the soliton scaling law, the spatial extent grows somewhat faster over the investigated parameter range, reflecting the empty gaps between clusters. Thus, the mechanism determining the scale of ecological diversity does not necessarily determine its spatial extent. Deriving the observed scaling laws directly from the microscopic stochastic dynamics, and understanding the factors controlling the separation between clusters, remain open theoretical questions.

Our results also have implications for eco-evolutionary models based on generalized Lotka–Volterra dynamics, in which new species inherit interaction coefficients similar to those of their ancestors [22, 23, 24]. In such models, evolutionary differentiation may progressively weaken competition between species, potentially driving the community toward nearly noninteracting populations. Similarly, models with finite-range competition in niche space explicitly allow sufficiently dissimilar species to avoid competition altogether [25]. When such models are intended to describe communities constrained by shared limiting resources, this complete decoupling should be prevented. One simple implementation is to retain a positive baseline of competition, irrespective of the evolutionary distance between species. Our analysis shows, however, that this global competitive constraint limits total population size without necessarily limiting the extent of diversification in trait space. A mechanism suppressing the establishment of sufficiently small populations is also needed to obtain localization.

Finally, our model describes an idealized ecological system: a one-dimensional trait space, local competition between individuals with identical traits, and uniform global competition for shared resources. In natural communities, competition typically extends over a finite range of trait differences, while shared resources may themselves be heterogeneous or fluctuate in time. An important question is therefore how robust the front-arrest mechanism and the resulting scaling laws are to relaxing these assumptions. In particular, it would be interesting to consider smooth establishment thresholds and finite-range competition kernels, and to investigate under what conditions the clusters emerging from stochastic dynamics may develop into distinct evolutionary lineages.

The role of the dimensionality of trait space is particularly interesting. In one dimension a mutant almost always lands in a niche that is already occupied, so that mutation acts as diffusion and neighboring populations continually rescue one another. In the opposite limit of a high-dimensional trait space a mutant never lands in an occupied niche; every mutation founds a new type, and the individual-based model of § 3 reduces to an infinite-allele community in which each type is limited by its own density while all types share a common competitive load. This is the niche-stabilized neutral model solved by Pigolotti and Cencini [26], with their stabilization parameter corresponding to 1/N1/N. In that limit there are no fronts and no soliton: each type fluctuates around its own carrying capacity until it is lost, and the mass deficit is set by the balance between origination and extinction of types. The scaling laws found here need not survive in that limit.

More broadly, our results emphasize that the emergence of new ecological opportunities does not guarantee that they will be occupied. The fate of initially rare populations attempting to establish themselves in new niches may play a decisive role in determining the extent of biological diversity.

Appendices

Absence of localization without a cutoff

Here we show that, in the absence of a reproduction cutoff, global competition can limit the total population but cannot prevent its spreading in trait space. In particular, any initially localized population must broaden indefinitely.

Setting ε=0\varepsilon= 0, the evolution equation becomes

∂tϕ=D∂x2ϕ+ϕ−ϕ2−κMϕ,M(t)=∫−∞∞ϕ(x,t) dx.(30)\partial_t \phi= D\partial_x^2\phi+ \phi- \phi^2 - \kappa M\phi,\qquad M(t) = \int_{-\infty}^{\infty} \phi(x,t)\,dx. \tag*{(30)}

We consider nonnegative, nonzero initial conditions with finite total mass and sufficient regularity and decay for the following integrations to be valid.

To separate the total population from its spatial distribution, we introduce the normalized density

p(x,t)=ϕ(x,t)M(t),∫−∞∞p(x,t) dx=1.(31)p(x,t) = \frac{\phi(x,t)}{M(t)},\qquad\int_{-\infty}^{\infty} p(x,t)\,dx = 1. \tag*{(31)}

Integrating Eq. (30) over space gives

dMdt=M−κM2−M2Q,Q(t)=∫−∞∞p2(x,t) dx.(32)\frac{dM}{dt} = M - \kappa M^2 - M^2 Q,\qquad Q(t) = \int_{-\infty}^{\infty} p^2(x,t)\,dx. \tag*{(32)}

Substituting ϕ=Mp\phi= Mp into Eq. (30), and using Eq. (32), we obtain

∂tp=D∂x2p+Mp(Q−p).(33)\partial_t p = D\partial_x^2p + Mp(Q-p). \tag*{(33)}

Notice that the global competition term −κMϕ-\kappa M\phi has disappeared explicitly from the equation for the normalized distribution. Global competition regulates the total population, but does not directly favor one position in trait space over another.

The quantity Q=∫p2 dxQ = \int p^2\,dx provides a convenient measure of localization. For a distribution that is approximately uniform over a region of width WW, normalization gives p∼1/Wp \sim1/W, and hence Q∼1/WQ \sim1/W. We may therefore define an effective width

Weff(t)=1Q(t).(34)W_{\mathrm{eff}}(t) = \frac{1}{Q(t)}. \tag*{(34)}

To establish unlimited spreading, it is sufficient to show that Q(t)Q(t) tends to zero.

Differentiating QQ and using Eq. (33), we find

dQdt=−2D∫−∞∞(∂xp)2 dx+2M[Q2−∫−∞∞p3 dx].(35)\frac{dQ}{dt} = -2D\int_{-\infty}^{\infty}(\partial_x p)^2\,dx + 2M\left[Q^2-\int_{-\infty}^{\infty}p^3\,dx\right]. \tag*{(35)}

where we have integrated the diffusion term by parts. By the Cauchy–Schwarz inequality,

Q2=(∫p2 dx)2≤(∫p3 dx)(∫p dx)=∫p3 dx.(36)Q^2 = \left(\int p^2\,dx\right)^2 \le\left(\int p^3\,dx\right)\left(\int p\,dx\right) = \int p^3\,dx. \tag*{(36)}

Consequently, both terms on the right-hand side of Eq. (35) are nonpositive, and

dQdt≤−2D∫−∞∞(∂xp)2 dx≤0.(37)\frac{dQ}{dt} \le-2D \int_{-\infty}^{\infty}(\partial_xp)^2\,dx \le0. \tag*{(37)}

Thus diffusion and local competition both act to reduce localization.

To show that the spreading continues indefinitely, rather than approaching a finite width, we use the one-dimensional Nash inequality [27]. For a normalized nonnegative density pp, it states that

∫−∞∞(∂xp)2 dx≥CN(∫−∞∞p2 dx)3=CNQ3,(38)\int_{-\infty}^{\infty}(\partial_xp)^2\,dx \ge C_N\left(\int_{-\infty}^{\infty}p^2\,dx\right)^3 = C_N Q^3, \tag*{(38)}

where CN>0C_N > 0 is a universal constant. Combining Eqs. (37) and (38) gives

dQdt≤−2DCNQ3.(39)\frac{dQ}{dt} \le-2DC_NQ^3. \tag*{(39)}

Integrating this differential inequality yields

Q(t)≤[Q(0)−2+4DCNt]−1/2.(40)Q(t) \le\left[Q(0)^{-2} + 4DC_Nt\right]^{-1/2}. \tag*{(40)}

It follows that

Weff(t)≥[Q(0)−2+4DCNt]1/2∼4DCNt.(41)W_{\mathrm{eff}}(t) \ge\left[Q(0)^{-2} + 4DC_Nt\right]^{1/2} \sim\sqrt{4DC_Nt}. \tag*{(41)}

Therefore, the effective width grows without bound, at least as fast as t1/2t^{1/2} at long times. In particular, no nonzero localized stationary solution can exist when ε=0\varepsilon= 0.

Finally, we show that this unlimited spreading is compatible with a finite total population. Introducing Y=1/MY = 1/M, Eq. (32) becomes

dYdt=−Y+κ+Q(t).(42)\frac{dY}{dt} = -Y + \kappa+ Q(t). \tag*{(42)}

Since Eq. (40) establishes Q(t)→0Q(t) \to0, the solution of this linear equation satisfies Y(t)→κY(t) \to\kappa. Hence

M(t)⟶1κ,Weff(t)⟶∞.(43)M(t) \longrightarrow\frac{1}{\kappa}, \qquad W_{\mathrm{eff}}(t) \longrightarrow\infty. \tag*{(43)}

Thus, without a cutoff, global competition determines the asymptotic population mass but does not confine its distribution in trait space. The population becomes increasingly dilute as it spreads over an ever-growing range of traits. A reproduction cutoff is therefore essential for the localized stationary solutions studied in the main text.

Stability of the localized solutions

We analyze the stability of localized stationary solutions of the full nonlocal evolution equation introduced in the main text, which we reproduce here for completeness:

∂ϕ(x,t)∂t=D∂2ϕ(x,t)∂x2+ϕ(x,t)Θ(ϕ(x,t)−ε)−ϕ2(x,t)−κM(t)ϕ(x,t),M(t)=∫−∞∞ϕ(x,t) dx.(44)\begin{aligned} \frac{\partial\phi(x,t)}{\partial t} &= D\frac{\partial^2\phi(x,t)}{\partial x^2}+\phi(x,t)\Theta(\phi(x,t)-\varepsilon)-\phi^2(x,t)-\kappa M(t)\phi(x,t),\\ M(t) &= \int_{-\infty}^{\infty}\phi(x,t)\,\mathrm{d}x. \tag*{(44)} \end{aligned}

As explained in the main text, stationary localized profiles can be constructed by temporarily treating r=κMr=\kappa M as an independent parameter and solving

Dϕr′′+ϕrΘ(ϕr−ε)−ϕr2−rϕr=0(45)D\phi_r^{\prime\prime}+\phi_r\Theta(\phi_r-\varepsilon)-\phi_r^2-r\phi_r=0 \tag*{(45)}

for ϕr(x)\phi_r(x). The corresponding global competition strength is then fixed by the self-consistency condition

κ(r)=rM(r),M(r)=∫−∞∞ϕr(x) dx.(46)\kappa(r)=\frac{r}{M(r)},\qquad M(r)=\int_{-\infty}^{\infty}\phi_r(x)\,\mathrm{d}x. \tag*{(46)}

As shown in Figure 2 of the main text, κ(r)\kappa(r) has a maximum κmax⁡\kappa_{\max}. For any fixed 0<κ0<κmax⁡0<\kappa_0<\kappa_{\max}, the condition κ(r)=κ0\kappa(r)=\kappa_0 admits two stationary localized solutions, a narrow pulse and a broad pulse. We now examine their dynamical stability.

For either stationary solution, let ϕ(x,t)=ϕr(x)+u(x,t)\phi(x,t)=\phi_r(x)+u(x,t), where uu is a small perturbation and κ0=κ(r)\kappa_0=\kappa(r) is held fixed. Linearizing the full evolution equation gives

∂u∂t=Aru≡Lru−κ0ϕr(x)∫−∞∞u(y,t) dy,(47)\frac{\partial u}{\partial t}=\mathcal{A}_r u\equiv\mathcal{L}_r u-\kappa_0\phi_r(x)\int_{-\infty}^{\infty}u(y,t)\,\mathrm{d}y, \tag*{(47)}

where

Lr=D∂x2+dfεdϕ∣ϕ=ϕr−r,fε(ϕ)=ϕΘ(ϕ−ε)−ϕ2.(48)\mathcal{L}_r=D\partial_x^2+\left.\frac{\mathrm{d}f_\varepsilon}{\mathrm{d}\phi}\right|_{\phi=\phi_r}-r,\qquad f_\varepsilon(\phi)=\phi\Theta(\phi-\varepsilon)-\phi^2. \tag*{(48)}

For the sharp cutoff, dfε/dϕ\mathrm{d}f_\varepsilon/\mathrm{d}\phi is understood in the distributional sense (or as the limit of a smooth cutoff).

To identify the zero mode associated with the fold, we differentiate the stationary profile equation (B2) with respect to the parameter rr, obtaining

Lr∂rϕr=ϕr.(49)\mathcal{L}_r\partial_r\phi_r=\phi_r. \tag*{(49)}

Applying the full linearized operator and using Eq. (49), we obtain

Ar∂rϕr=ϕr−κ(r)ϕr∫∂rϕr dx=[1−κ(r)dMdr]ϕr=M(r)dκdrϕr.(50)\mathcal{A}_r \partial_r\phi_r = \phi_r-\kappa(r)\phi_r\int\partial_r\phi_r\,dx = \left[1-\kappa(r)\frac{dM}{dr}\right]\phi_r = M(r)\frac{d\kappa}{dr}\phi_r. \tag*{(50)}

Here we used ∫∂rϕr dx=dM/dr\int\partial_r\phi_r\,dx=dM/dr and M(r) dκ/dr=1−κ(r) dM/drM(r)\,d\kappa/dr=1-\kappa(r)\,dM/dr, the latter following directly from κ(r)=r/M(r)\kappa(r)=r/M(r). At the maximum of the self-consistency curve, dκ/dr=0d\kappa/dr=0, and hence

Ar∂rϕr=0.(51)\mathcal{A}_r\partial_r\phi_r=0. \tag*{(51)}

Thus the fold is accompanied by a neutral mode corresponding to a change in the stationary pulse profile. This mode is distinct from the neutral translation mode, which exists for every rr.

To examine which side of the fold is stable, we choose a fixed κ0\kappa_0 just below κmax⁡\kappa_{\max}. As illustrated in Figure 5, the line κ=κ0\kappa=\kappa_0 intersects the self-consistency curve at r1<r∗r_1<r_\ast (the narrow pulse) and r2>r∗r_2>r_\ast (the broad pulse). To test the stability of each stationary solution ϕri\phi_{r_i}, with i=1,2i=1,2, at fixed κ0\kappa_0, we replace it by the nearby stationary profile ϕri+δr\phi_{r_i+\delta r}, where δr>0\delta r>0. This new profile corresponds to κ(ri+δr)=κ0+Δκi\kappa(r_i+\delta r)=\kappa_0+\Delta\kappa_i. We then use it as the initial condition for the dynamics at the original competition strength κ0\kappa_0. On the rising branch Δκ1>0\Delta\kappa_1>0, whereas on the falling branch Δκ2<0\Delta\kappa_2<0.

Schematic enlargement of the fold showing narrow and broad stationary pulses

Figure 5. Schematic enlargement of the fold in Figure 2 of the main text. At κ0<κmax⁡\kappa_0 < \kappa_{\max}, the two filled points denote the narrow and broad stationary pulses. Rightward shifts δr>0\delta r > 0 lead to nearby auxiliary stationary profiles (open points), requiring κ0+Δκi\kappa_0 + \Delta\kappa_i, with Δκ1>0\Delta\kappa_1 > 0 and Δκ2<0\Delta\kappa_2 < 0. In the argument, κ0\kappa_0 is very close to κmax⁡\kappa_{\max}; their separation and the shifts are exaggerated for visibility.

Write ϕi+=ϕri+δr\phi_i^+=\phi_{r_i+\delta r} and Mi+=M(ri+δr)M_i^+=M(r_i+\delta r). The shifted profile is stationary at κ0+Δκi\kappa_0+\Delta\kappa_i, and therefore satisfies

0=D(ϕi+)′′+fε(ϕi+)−(κ0+Δκi)Mi+ϕi+.0=D(\phi_i^+)''+f_\varepsilon(\phi_i^+)-(\kappa_0+\Delta\kappa_i)M_i^+\phi_i^+.

The evolution of the same profile at the original competition strength κ0\kappa_0 instead obeys, at t=0t=0,

∂ϕ∂t∣t=0=D(ϕi+)′′+fε(ϕi+)−κ0Mi+ϕi+.\left.\frac{\partial\phi}{\partial t}\right|_{t=0} =D(\phi_i^+)''+f_\varepsilon(\phi_i^+)-\kappa_0M_i^+\phi_i^+.

Subtracting the stationary equation cancels the diffusion and local growth terms, leaving

∂ϕ∂t∣t=0=ΔκiMi+ϕi+.\left.\frac{\partial\phi}{\partial t}\right|_{t=0} =\Delta\kappa_i M_i^+\phi_i^+.

Integrating over space, we obtain the exact initial mass response:

dMdt∣t=0=Δκi[M(ri+δr)]2.(52)\left.\frac{dM}{dt}\right|_{t=0} =\Delta\kappa_i\left[M(r_i+\delta r)\right]^2. \tag*{(52)}

To interpret the sign of this response, note that κ(r)=r/M(r)\kappa(r) = r/M(r) implies dM/dr∣r∗=M(r∗)/r∗>0\left.dM/dr\right|_{r_*} = M(r_*)/r_* > 0 at the fold. Thus the positive shift δr\delta r increases the initial mass on both branches. For the narrow pulse, Δκ1>0\Delta\kappa_1 > 0, so the mass initially grows still further; for the broad pulse, Δκ2<0\Delta\kappa_2 < 0, so it initially decreases. Thus the initial mass response reinforces the perturbation for the narrow pulse and opposes it for the broad pulse. Although this is not a complete stability analysis, it provides a plausible physical explanation for the instability of the narrow pulse and the stability of the broad one, consistent with the numerical time evolution. This picture is also supported by Appendix § 6, which shows that without a reproduction cutoff (ε=0\varepsilon= 0) a localized population spreads indefinitely. The resulting expectation that the stable localized solution becomes arbitrarily broad as ε→0\varepsilon\to0 is consistent with the behavior of the broad branch established in the main text.

References

  1. [1]U. Dieckmann and M. Doebeli, On the origin of species by sympatric speciation, Nature 400, 354 (1999).
  2. [2]M. Scheffer and E. H. van Nes, Self-organized similarity, the evolutionary emergence of groups of similar species, Proceedings of the National Academy of Sciences 103, 6230 (2006).pubmed.ncbi.nlm.nih.gov/16585519
  3. [3]D. Storch and J. G. Okie, The carrying capacity for species richness, Global Ecology and Biogeography 28, 1519 (2019).DOI
  4. [4]D. Tilman, Niche tradeoffs, neutrality, and community structure: A stochastic theory of resource competition, invasion, and community assembly, Proceedings of the National Academy of Sciences 101, 10854 (2004).pubmed.ncbi.nlm.nih.gov/15243158
  5. [5]S. Pigolotti, C. López, and E. Hernández-García, Species clustering in competitive Lotka–Volterra models, Physical Review Letters 98, 258101 (2007).DOI
  6. [6]T. H. Keitt, M. A. Lewis, and R. D. Holt, Allee effects, invasion pinning, and species’ borders, The American Naturalist 157, 203 (2001).DOI
  7. [7]E. Brunet and B. Derrida, Shift in the velocity of a front due to a cutoff, Physical Review E 56, 2597 (1997).DOI
  8. [8]B. Perthame and M. Gauduchon, Survival thresholds and mortality rates in adaptive dynamics: Conciliating deterministic and stochastic simulations, Mathematical Medicine and Biology 27, 195 (2010).DOI
  9. [9]T. Rogers, A. J. McKane, and A. G. Rossberg, Demographic noise can lead to the spontaneous formation of species, EPL (Europhysics Letters) 97, 40008 (2012).DOI
  10. [10]Y. A. Ganan and D. A. Kessler, Front propagation and clustering in the stochastic nonlocal Fisher equation, Physical Review E 97, 042213 (2018).DOI
  11. [11]H. Berestycki, G. Nadin, B. Perthame, and L. Ryzhik, The non-local Fisher–KPP equation: travelling waves and steady states, Nonlinearity 22, 2813 (2009).DOI
  12. [12]H. Weissmann, N. M. Shnerb, and D. A. Kessler, Simulation of spatial systems with demographic noise, Physical Review E 98, 022131 (2018).DOI
  13. [13]I. Dornic, H. Chaté, and M. A. Munoz, Integration of Langevin equations with multiplicative noise and the viability of field theories for absorbing phase transitions, Physical Review Letters 94, 100601 (2005).pubmed.ncbi.nlm.nih.gov/15783467
  14. [14]M. Meyer, S. Havlin, and A. Bunde, Clustering of independently diffusing individuals by birth and death processes, Physical Review E 54, 5567 (1996).pubmed.ncbi.nlm.nih.gov/9965742
  15. [15]W. R. Young, A. J. Roberts, and G. Stuhne, Reproductive pair correlations and the clustering of organisms, Nature 412, 328 (2001).pubmed.ncbi.nlm.nih.gov/11460162
  16. [16]B. Houchmandzadeh, Clustering of diffusing organisms, Physical Review E 66, 052902 (2002).pubmed.ncbi.nlm.nih.gov/12513541
  17. [17]Y. E. Maruvka and N. M. Shnerb, Nonlocal competition and logistic growth: patterns, defects, and fronts, Physical Review E 73, 011903 (2006).pubmed.ncbi.nlm.nih.gov/16486181
  18. [18]M. Fuentes, M. Kuperman, and V. Kenkre, Nonlocal interaction effects on pattern formation in population dynamics, Physical Review Letters 91, 158104 (2003).pubmed.ncbi.nlm.nih.gov/14611503
  19. [19]M. G. Clerc, D. Escaff, and V. M. Kenkre, Analytical studies of fronts, colonies, and patterns: Combination of the Allee effect and nonlocal competition interactions, Physical Review E 82, 036210 (2010).pubmed.ncbi.nlm.nih.gov/21230164
  20. [20]Y. E. Maruvka and N. M. Shnerb, Nonlocal competition and front propagation in branching-coalescence systems, Physical Review E—Statistical, Nonlinear, and Soft Matter Physics 75, 042901 (2007).pubmed.ncbi.nlm.nih.gov/17500943
  21. [21]D. A. Kessler and H. Levine, Pulse instabilities can shape virus-immune coevolution, Physical Review Research 6, 033300 (2024).DOI
  22. [22]L. R. Ginzburg, H. R. Akçakaya, and J. Kim, Evolution of community structure: Competition, Journal of Theoretical Biology 133, 513 (1988).DOI
  23. [23]E. Shtilerman, D. A. Kessler, and N. M. Shnerb, Emergence of structured communities through evolutionary dynamics, Journal of Theoretical Biology 383, 138 (2015).DOI
  24. [24]F. Farahpour, M. Saeedghalati, V. S. Brauer, and D. Hoffmann, Trade-off shapes diversity in eco-evolutionary dynamics, eLife 7, e36273 (2018).DOI
  25. [25]S. Y. Li, M. Kardar, Z. Feng, and W. Taylor, Minimal model of self-organized clusters with phase transitions in ecological communities, Physical Review E 113, 044306 (2026).DOI
  26. [26]S. Pigolotti and M. Cencini, Species abundances and lifetimes: From neutral to niche-stabilized communities, Journal of Theoretical Biology 338, 1 (2013).DOI
  27. [27]J. Nash, Continuity of solutions of parabolic and elliptic equations, American Journal of Mathematics 80, 931 (1958).DOI

Paper details

vlnc:W_ugmp8wqa
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Agricultural and Biological Sciences

Not recorded for this paper

View statement
Looking for related papers…
Contents