This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Periodic Traveling Waves in an Integro-Difference Equation With Non-Monotonic Growth and Strong Allee Effect

Michael Nestor and Bingtuan Li M. Nestor’s email is mdnest01@gmail.com. B. Li was partially supported by the National Science Foundation under Grant DMS-1515875 and Grant DMS-1951482. Department of Mathematics, University of LouisvilleLouisville, KY 40292
Abstract

We derive sufficient conditions for the existence of a periodic traveling wave solution to an integro-difference equation with a piecewise constant growth function exhibiting a stable period-2 cycle and strong Allee effect. The mean traveling wave speed is shown to be the asymptotic spreading speed of solutions with compactly supported initial data under appropriate conditions. We then conduct case studies for the Laplace kernel and uniform kernel.

Key words: Integro-difference equation, period two cycle, Allee effect, periodic traveling wave.

AMS Subject Classification: 92D40, 92D25.

1 Introduction

Integro-difference equations in the form

un+1(x)=Q[un](x):=k(xy)g(un(y))dy\displaystyle u_{n+1}(x)\;=\;Q[u_{n}](x)\;:=\,\int^{\infty}_{-\infty}k(x-y)\,g\left(u_{n}(y)\right)\,\mathrm{d}y (1.1)

are of great interest in the studies of invasions of populations with discrete generations and separate growth and dispersal stages. They have been used to predict changes in gene frequency [8, 9, 10, 14, 17], and applied to ecological problems [2, 3, 4, 5, 7, 11, 12, 13]. Previous rigorous studies on integro-difference equations have assumed that the growth function is nondecreasing [17, 18], or is non-monotone without Allee effect [10, 16]. The results show existence of constant spreading speeds and travelng waves with fixed shapes and speeds. Sullivan et al.  [15] demonstrated numerically that an integro-difference equation with a non-monotone growth function exhibiting a strong Allee effect can generate traveling waves with fluctuating speeds, and Otto [13] showed that such an equation can have robust non-spreading solutions. In this paper we give conditions for the existence of a periodic traveling wave with two intermediate speeds for an integro-difference equation with a piecewise constant growth function that exhibits a strong Allee effect and a period-2 cycle.

Piecewise constant growth functions have been used in the studies of integro-difference equations; see for example [6, 11, 13, 15]. Such an equation is analytically tractable and it can provide specific insights into the dynamics of solutions. For the piecewise constant growth function

g(u)={0,if u<a,1,if ua,g(u)=\begin{cases}0,&\text{if }u<a,\\ 1,&\text{if }u\geq a,\end{cases} (1.2)

with 0<a<10<a<1, which exhibits a strong Allee effect and monotonicity, Kot et al. [6] and Lutscher [11] (Section 6.1) investigated the range expansion of the population and the spreading speed, and Sullivan et al.  [15] studied oscillations in spreading speeds when the dispersal kernel is density dependent. More discussions about integro-difference equations with piecewise growth functions can be found in Lutscher [11] (Chapter 15).

In this paper, we consider the integro-difference equation (1.1) with

g(u)={0,if u<a,1,if aub,m,if u>b,g(u)=\begin{cases}0,&\text{if }u<a,\\ 1,&\text{if }a\leq u\leq b,\\ m,&\text{if }u>b,\end{cases} (1.3)

with 0<a<m<b<10<a<m<b<1. g(u)g(u) is a piecewise constant non-monotone growth function exhibiting a strong Allee effect [1]. Specifically, it has a stable fixed point at zero and a stable period two cycle (1,m)(1,m) with aa the Allee threshold value. This is the function considered in Otto [13] where non-spreading solitions are studied. It may be viewed as an extension of (1.2). A graphical demonstration of gg defined by (1.3) is given by Figure 1.

Figure 1: The growth of g(u)g(u) defined by (1.3) with 0<a<m<b<10<a<m<b<1.
Refer to caption

We rigorously construct periodic traveling waves with two intermediate speed for (1.1) with gg given by (1.3) and a proper dispersal kernel kk. To the best of our knowledge, this is the first time that traveling waves with oscillating speeds have been analytically established for scalar spatiotemporal equations with constant parameters. We also show that the mean traveling wave speed is the spreading speed of solutions with compactly supported initial data under appropriate conditions. We finally conduct case studies for the Laplace kernel and uniform kernel.

2 Results

In this section, we present our main results about our model. Throughout, we assume all population density functions are bounded elements of C()C(\mathbb{R}), the space of real-valued continuous functions. It is straightforward to verify this space is closed under iteration of QQ. We denote the norm of a function uC()u\in C(\mathbb{R}) by u||u||_{\infty}.

We make the following assumptions about the dispersal kernel.

Hypothesis 2.1.

k(x)k(x) is a piecewise differentiable function satisfying

  1. (H1)

    k(x)0k(x)\geq 0 for all xx, and k(x)dx=1\int_{-\infty}^{\infty}k(x)\,\mathrm{d}x=1,

  2. (H2)

    k(x)=k(x)k(x)=k(-x) for all xx,

  3. (H3)

    k(x)>0k(x)>0 for all x(σ,σ)x\in(-\sigma,\sigma) for some 0<σ0<\sigma\leq\infty,

  4. (H4)

    for all λ(0,1)\lambda\in(0,1), for all A,BA,B\in\mathbb{R} with A>BA>B, the expression

    k(xA)λk(xB)k(x-A)-\lambda k(x-B)

    has <2<2 sign changes, and

  5. (H5)

    for all λ(0,1)\lambda\in(0,1), for all A,BA,B\in\mathbb{R} with A>BA>B, the expression

    k(xrA)λk(xrB)k(x+r+A)+λk(x+r+B)k(x-r-A)-\lambda k(x-r-B)-k(x+r+A)+\lambda k(x+r+B)

    has <4<4 sign changes for sufficiently large r>0r>0.

2.1 Periodic traveling waves

Let w1,w2C()w_{1},w_{2}\in C(\mathbb{R}) be bounded, continuous functions defined by

w1(x):=xk(y)dyw_{1}(x):=\int_{x}^{\infty}k(y)\,\mathrm{d}y (2.1)

and

w2(x):=Q[w1](x+c)=k(x+cy)g(w1(y))dyw_{2}(x):=Q[w_{1}](x+c^{*})=\int_{-\infty}^{\infty}k(x+c^{*}-y)g(w_{1}(y))\,\mathrm{d}y (2.2)

where cc^{*} is a constant defined by

c:=12sup{xQ[w1](x)=a}.c^{*}:=\frac{1}{2}\sup\{x\in\mathbb{R}\mid Q[w_{1}](x)=a\}. (2.3)

The following theorem shows that w1w_{1} and w2w_{2} generate a traveling wave-like solution to (1.1). We term this solution a periodic traveling wave since it alternates between two distinct wave profiles (with different limits at -\infty) while traveling with mean speed cc^{*}.

Theorem 2.2.

Let (Wn)n=0(W_{n})_{n=0}^{\infty} be a sequence of bounded, continuous functions defined by

Wn(x):={w1(xnc),n even,w2(xnc),n odd.W_{n}(x):=\begin{cases}w_{1}(x-nc^{*}),&n\text{ even},\\ w_{2}(x-nc^{*}),&n\text{ odd}.\end{cases} (2.4)

If w2<b||w_{2}||_{\infty}<b, then Wn+1=Q[Wn]W_{n+1}=Q[W_{n}] for all n0n\geq 0.

Proof.

For i=1,2i=1,2, define the right inverse of wiw_{i} by

Φi():=sup{xwi(x)=},i=1,2.\Phi_{i}(\ell):=\sup\{x\in\mathbb{R}\mid w_{i}(x)=\ell\},\quad i=1,2. (2.5)

Note that Φi()\Phi_{i}(\ell) is finite for 0<<wi0<\ell<||w_{i}||_{\infty}, where w1=1||w_{1}||_{\infty}=1, and a<w2<ba<||w_{2}||_{\infty}<b by the assumption of this theorem.

By the translation invariance property of QQ, it suffices to show

Q[w1](x)=w2(xc)Q[w_{1}](x)=w_{2}(x-c^{*}) (2.6)

and

Q[w2](x)=w1(xc)Q[w_{2}](x)=w_{1}(x-c^{*}) (2.7)

for all xx\in\mathbb{R}. The first equation follows immediately from definition (2.2). To prove the second, observe that w1w_{1} is monotonically decreasing and satisfies w1()=0w_{1}(\infty)=0 and w1()=1w_{1}(-\infty)=1 (see for example Figure 2). Thus,

g(w1(x))={m,x<Φ1(b),1,Φ1(b)xΦ1(a),0,x>Φ1(a).g(w_{1}(x))=\begin{cases}m,&x<\Phi_{1}(b),\\ 1,&\Phi_{1}(b)\leq x\leq\Phi_{1}(a),\\ 0,&x>\Phi_{1}(a).\end{cases}

Applying the integro-difference operator yields

w2(xc)\displaystyle w_{2}(x-c^{*}) =Q[w1](x)\displaystyle=Q[w_{1}](x)
=mΦ1(b)k(xy)dy+Φ1(b)Φ1(a)k(xy)dy\displaystyle=m\int_{-\infty}^{\Phi_{1}(b)}k(x-y)\,\mathrm{d}y+\int_{\Phi_{1}(b)}^{\Phi_{1}(a)}k(x-y)\,\mathrm{d}y
=w1(xΦ1(a))(1m)w1(xΦ1(b)).\displaystyle=w_{1}(x-\Phi_{1}(a))-(1-m)\,w_{1}(x-\Phi_{1}(b)).

We can then differentiate with respect to xx:

dw2dx|xc=k(xΦ1(a))+(1m)k(xΦ1(b)).\frac{\,\mathrm{d}w_{2}}{\,\mathrm{d}x}\Big{|}_{x-c^{*}}=-k(x-\Phi_{1}(a))+(1-m)\,k(x-\Phi_{1}(b)).

From (H4), dw2dx\frac{\,\mathrm{d}w_{2}}{\,\mathrm{d}x} has at most 1 sign change; thus, w2w_{2} has at most 1 turning point. We also have w2(x)<bw_{2}(x)<b for all xx, hence g(w2(x)){0,1}g(w_{2}(x))\in\{0,1\}.

We claim

g(w2(x))={1,xc,0,x>c,g(w_{2}(x))=\begin{cases}1,&x\leq c^{*},\\ 0,&x>c^{*},\end{cases} (2.8)

where cc^{*} is defined by (2.3). This can be shown by arguing in cases depending on the number of turning points of w2w_{2}.

  1. Case 1.

    If w2w_{2} has no turning points, then w2w_{2} is monotone decreasing with 0w2(x)m0\leq w_{2}(x)\leq m for all xx. It follows by the intermediate value theorem (IVT) that c=Φ2(a)c^{*}=\Phi_{2}(a) satisfies (2.8).

  2. Case 2.

    If w2w_{2} has one turning point, say x0x_{0}\in\mathbb{R}, then w2w_{2} must be increasing for x<x0x<x_{0} and decreasing for x>x0x>x_{0}. It follows that x0x_{0} is a global maximum, and w2w_{2} is increasing on (,x0)(-\infty,x_{0}) and decreasing on (x0,)(x_{0},\infty). We have g(w2(x))=1g(w_{2}(x))=1 for x<x0x<x_{0}. By the IVT, c=Φ2(a)c^{*}=\Phi_{2}(a) is the unique solution to (2.8).

Taking the convolution of equation (2.8) with k(x)k(x) yields

Q[w2](x)=ck(xy)dy=w1(xc).Q[w_{2}](x)\,=\,\int_{-\infty}^{c^{*}}k(x-y)\,\mathrm{d}y\,=\,w_{1}(x-c^{*}).

This completes the proof. ∎

Refer to caption
(a) w1(x)w_{1}(x)
Refer to caption
(b) w2(x)w_{2}(x)
Figure 2: Profiles of the intermediate traveling waves w1w_{1} and w2w_{2}, with k(x)k(x) equal to the Laplace kernel (see Section 3). Blue points mark the intersection of the curves with the level sets =a\ell=a and =b\ell=b.

Theorem 2.2 describes a rightward periodic traveling wave, but we can define the corresponding leftward periodic traveling wave via xWn(x)x\mapsto W_{n}(-x). This is due to the fact that QQ is symmetric and translation invariant.

We investigated the intermediate wave speed generated by the periodic traveling wave solution (2.4). To do this, we define a sequence (ξn)n=0(\xi_{n})_{n=0}^{\infty} of continuous functions (0,m)(0,m)\to\mathbb{R} by

ξn():=sup{x:Wn(x)=},(0,m).\xi_{n}(\ell):=\sup\{x\in\mathbb{R}:W_{n}(x)=\ell\},\quad\ell\in(0,m). (2.9)

This expression yields the rightmost intersection of the curve u=Wn(x)u=W_{n}(x) with the horizontal line u=u=\ell. By computing the first difference, we obtain a new function sequence (cn)n=0(c_{n})_{n=0}^{\infty} (with the same domain and range) defined by

cn():=ξn+1()ξn(),(0,m).c_{n}(\ell):=\xi_{n+1}(\ell)-\xi_{n}(\ell),\quad\ell\in(0,m). (2.10)

From Theorem 2.2, we have that cn()c_{n}(\ell) converges to a period-22 cycle for every (0,m)\ell\in(0,m).

Our results about the intermediate wave speed sequence are summarized in the following corollary.

Corollary 2.3.

Assume the hypothesis of Theorem 2.2 holds. Then for all (0,m)\ell\in(0,m), the sequence cn()c_{n}(\ell) satisfies

cn()=c+{Φ1()Φ2(),n even,Φ2()Φ1(),n odd.c_{n}(\ell)=c^{*}+\begin{cases}\Phi_{1}(\ell)-\Phi_{2}(\ell),&n\text{ even},\\ \Phi_{2}(\ell)-\Phi_{1}(\ell),&n\text{ odd}.\end{cases} (2.11)
Proof.

By Theorem 2.2, we have

ξn()\displaystyle\xi_{n}(\ell) ={sup{xw1(xnc)=},n even,sup{xw2(xnc)=},n odd,\displaystyle=\begin{cases}\sup\{x\in\mathbb{R}\mid w_{1}(x-nc^{*})=\ell\},&n\text{ even},\\ \sup\{x\in\mathbb{R}\mid w_{2}(x-nc^{*})=\ell\},&n\text{ odd},\end{cases}
=nc+{Φ1(),n even,Φ2(),n odd.\displaystyle=nc^{*}+\begin{cases}\Phi_{1}(\ell),&n\text{ even},\\ \Phi_{2}(\ell),&n\text{ odd}.\end{cases}

for every (0,m)\ell\in(0,m). Taking the first difference yields

cn()\displaystyle c_{n}(\ell) =ξn()ξn1()\displaystyle=\xi_{n}(\ell)-\xi_{n-1}(\ell)
=nc+{Φ1(),n even,Φ2(),n odd,(n1)c+{Φ2(),n even,Φ1(),n odd,\displaystyle=nc^{*}+\begin{cases}\Phi_{1}(\ell),&n\text{ even},\\ \Phi_{2}(\ell),&n\text{ odd},\end{cases}-(n-1)\,c^{*}+\begin{cases}\Phi_{2}(\ell),&n\text{ even},\\ \Phi_{1}(\ell),&n\text{ odd},\end{cases}
=c+{Φ1()Φ2(),n even,Φ2()Φ1(),n odd.\displaystyle=c^{*}+\begin{cases}\Phi_{1}(\ell)-\Phi_{2}(\ell),&n\text{ even},\\ \Phi_{2}(\ell)-\Phi_{1}(\ell),&n\text{ odd}.\end{cases}

Thus, for each fixed (0,m)\ell\in(0,m), we can define intermediate traveling wave speeds c1c_{1}^{*} and c2c_{2}^{*} by

c1():=c+Φ1()Φ2()c_{1}^{*}(\ell):=c^{*}+\Phi_{1}(\ell)-\Phi_{2}(\ell) (2.12)

and

c2():=c+Φ2()Φ1().c_{2}^{*}(\ell):=c^{*}+\Phi_{2}(\ell)-\Phi_{1}(\ell). (2.13)

It follows that (1.1)\eqref{q} has a periodic traveling wave with wave profiles w1(x)w_{1}(x) and w2(x)w_{2}(x), intermediate wave speeds c1()c_{1}^{*}(\ell) and c2()c_{2}^{*}(\ell), and mean wave speed 12(c1()+c2())=c\frac{1}{2}\left(c_{1}^{*}(\ell)+c_{2}^{*}(\ell)\right)=c^{*}. Note that the intermediate wave speeds depend on \ell, but their sum is identically equal to 2c2c^{*}.

2.2 Periodic spreading solutions

By applying the results of Theorem 2.2, we were able to prove the asymptotic spreading speed of solutions with compact initial data. In particular, we showed that cc^{*} is the asymptotic spreading speed of compactly supported initial data with sufficient weight above the Allee threshold but below the overcompensation threshold.

Theorem 2.4.

Assume the hypothesis of Theorem 2.2 holds. Suppose uC()u\in C(\mathbb{R}) is bounded, non-negative, and has compact support. If

  1. i.

    u<b||u||_{\infty}<b,

  2. ii.

    the set A:={x:u(x)a}A:=\{x\in\mathbb{R}:u(x)\geq a\} is connected and has sufficiently large length, and

  3. iii.

    c>0c^{*}>0,

then the sequence (un)n=0(u_{n})_{n=0}^{\infty} defined by u0=uu_{0}=u and un+1=Q[un]u_{n+1}=Q[u_{n}] for n0n\geq 0 spreads asymptotically with mean speed cc^{*}.

Proof.

We define two continuous functions w1~,w2~:×[0,)\tilde{w_{1}},\tilde{w_{2}}:\mathbb{R}\times[0,\infty)\to\mathbb{R} by

w1~(x,r):=w1(x)w1(x+2r)=2r0k(xy)dy\tilde{w_{1}}(x,r):=w_{1}(x)-w_{1}(x+2r)=\int_{-2r}^{0}k(x-y)\,\mathrm{d}y (2.14)

and

w2~(x,r):=Q[w1~(,r)](x+c).\tilde{w_{2}}(x,r):=Q[\tilde{w_{1}}(\,\cdot\,,r)](x+c^{*}). (2.15)

Observe that w1~\tilde{w_{1}} is symmetric with respect to x=rx=-r, and w2~\tilde{w_{2}} is symmetric with respect to x=rcx=-r-c^{*}. This notation can be justified by observing that w1~(x,r)w1(x)\tilde{w_{1}}(x,r)\to w_{1}(x) and w2~(x,r)w2(x)\tilde{w_{2}}(x,r)\to w_{2}(x) as rr\to\infty for all xx. We also set

ϕi(r,):=sup{xw~i(x,r)=},i=1,2.\phi_{i}(r,\ell):=\sup\{x\in\mathbb{R}\mid\tilde{w}_{i}(x,r)=\ell\},\quad i=1,2. (2.16)

ϕ1\phi_{1} and ϕ2\phi_{2} are the right-inverse of w1~\tilde{w_{1}} and w2~\tilde{w_{2}}, respectively. For a fixed r>0r>0, the value of ϕi(r,)\phi_{i}(r,\ell) is finite if 0<<w~i(,r)0<\ell<||\tilde{w}_{i}(\,\cdot\,,r)||_{\infty}, with the latter bound converging to wi||w_{i}||_{\infty} as rr\to\infty. They satisfy

limrϕi(r,)=Φi(),0<<wi,i=1,2.\lim_{r\to\infty}\phi_{i}(r,\ell)=\Phi_{i}(\ell),\quad 0<\ell<||w_{i}||_{\infty},\quad i=1,2.

We now apply the growth function to w1~\tilde{w_{1}}, assuming rr is sufficiently large enough so that a<w1~(,r)<ba<||\tilde{w_{1}}(\,\cdot\,,r)||_{\infty}<b; this is guaranteed by taking r>Φ1(1a2)r>\Phi_{1}\left(\frac{1-a}{2}\right). We have

g(w1~(x,r))={0,x<2rϕ1(r,a),1,2rϕ1(r,a)<x<2rϕ1(r,b),m,2rϕ1(r,b)<x<ϕ1(r,b),1,ϕ1(r,b)<x<ϕ1(r,a),0,ϕ1(r,a)<x.g\left(\tilde{w_{1}}(x,r)\right)=\begin{cases}0,&x<-2r-\phi_{1}(r,a),\\ 1,&-2r-\phi_{1}(r,a)<x<-2r-\phi_{1}(r,b),\\ m,&-2r-\phi_{1}(r,b)<x<\phi_{1}(r,b),\\ 1,&\phi_{1}(r,b)<x<\phi_{1}(r,a),\\ 0,&\phi_{1}(r,a)<x.\end{cases}

Note that the above expression converges to g(w1(x))g(w_{1}(x)) as rr\to\infty for all xx. We then apply the convolution operator:

w2~(xc,r)=Q[w1~(,r)](x)\displaystyle\tilde{w_{2}}(x-c^{*},r)=Q[\tilde{w_{1}}(\,\cdot\,,r)](x) =2rϕ1(r,a)ϕ1(r,a)k(xy)dy(1m)2rϕ1(r,b)ϕ1(r,b)k(xy)dy\displaystyle=\int_{-2r-\phi_{1}(r,a)}^{\phi_{1}(r,a)}k(x-y)\,\mathrm{d}y-(1-m)\,\int_{-2r-\phi_{1}(r,b)}^{\phi_{1}(r,b)}k(x-y)\,\mathrm{d}y
=w1(xϕ1(r,a))(1m)w1(xϕ1(r,b))\displaystyle=w_{1}(x-\phi_{1}(r,a))-(1-m)\,w_{1}(x-\phi_{1}(r,b))
+(1m)w1(x+2r+ϕ1(r,b))w1(x+2r+ϕ1(r,a)).\displaystyle+(1-m)\,w_{1}(x+2r+\phi_{1}(r,b))-w_{1}(x+2r+\phi_{1}(r,a)).

Computing the derivative of this expression yields

dw2~dx|xc\displaystyle\frac{\,\mathrm{d}\tilde{w_{2}}}{\,\mathrm{d}x}\Big{|}_{x-c^{*}} =k(xϕ1(r,a))+(1m)k(xϕ1(r,b))\displaystyle=-k(x-\phi_{1}(r,a))+(1-m)\,k(x-\phi_{1}(r,b))
(1m)k(x+2r+ϕ1(r,b))+k(x+2r+ϕ1(r,a)).\displaystyle-(1-m)\,k(x+2r+\phi_{1}(r,b))+k(x+2r+\phi_{1}(r,a)).

By (H5), the number of turning points of w2~\tilde{w_{2}} is at most 3 (assuming rr is sufficiently large). Furthermore, w2~\tilde{w_{2}} is symmetric around x=rcx=-r-c^{*}; thus the sign changes must come in pairs or occur exactly at x=rcx=-r-c^{*}. We can immediately exclude the cases of 0 and 2 turning points because w2~\tilde{w_{2}} is non-negative, vanishes at ±\pm\infty, and is not identically zero. As in the proof of Theorem 2.2, this leaves two cases:

  1. Case 1.

    If w2~\tilde{w_{2}} has 1 turning point, then it must occur at x=rcx=-r-c^{*}. We have w2~(rc,r)=w2(rc)w2(r+c)m\tilde{w_{2}}(-r-c^{*},r)=w_{2}(-r-c^{*})-w_{2}(r+c^{*})\to m as rr\to\infty; hence, it is sufficient to take rr large enough so that |w2~(rc,r)m|<min{|am|,|bm|}|\tilde{w_{2}}(-r-c^{*},r)-m|<\min\{|a-m|,|b-m|\}. It follows by the triangle inequality that w2~(x,r)w2~(rc,r)<b\tilde{w_{2}}(x,r)\leq\tilde{w_{2}}(-r-c^{*},r)<b for all xx.

  2. Case 2.

    If w2~\tilde{w_{2}} has 3 turning points, then again by symmetry, they must be given by {rct,rc,rc+t}\{-r-c^{*}-t,-r-c^{*},-r-c^{*}+t\}, for some t>0t>0. We can label them in increasing order by t1<t2<t3t_{1}<t_{2}<t_{3}. It follows that t1t_{1} and t3t_{3} are maxima with w2~(t1,r)=w2~(t3,r)w2\tilde{w_{2}}(t_{1},r)=\tilde{w_{2}}(t_{3},r)\to||w_{2}||_{\infty} as rr\to\infty, and t2t_{2} is a local minima with w2~(t2,r)m\tilde{w_{2}}(t_{2},r)\to m as rr\to\infty. Thus w2~\tilde{w_{2}} is monotone increasing on (,t1)(t2,t3)(-\infty,t_{1})\cup(t_{2},t_{3}) and monotone decreasing on (t1,t2)(t3,)(t_{1},t_{2})\cup(t_{3},\infty). By taking rr sufficiently large, we can make |w2~(t1,r)w2|=|w2~(t3,r)w2|<|bw2|\left|\tilde{w_{2}}(t_{1},r)-||w_{2}||_{\infty}\right|=\left|\tilde{w_{2}}(t_{3},r)-||w_{2}||_{\infty}\right|<\left|b-||w_{2}||_{\infty}\right| and |w2~(t2,r)m|<|am|\left|\tilde{w_{2}}(t_{2},r)-m\right|<\left|a-m\right|.

In both cases, we have aw2~(x,r)ba\leq\tilde{w_{2}}(x,r)\leq b on a closed interval of radius r+c+ϕ2(r,a)r+c^{*}+\phi_{2}(r,a) and w2~(x,r)<a\tilde{w_{2}}(x,r)<a elsewhere. Taking composition with gg yields

g(w2~(xc,r))={0,x<2rcϕ2(r,a),1,2rcϕ2(r,a)xc+ϕ2(r,a),0,x>c+ϕ2(r,a).g(\tilde{w_{2}}(x-c^{*},r))=\begin{cases}0,&x<-2r-c^{*}-\phi_{2}(r,a),\\ 1,&-2r-c^{*}-\phi_{2}(r,a)\leq x\leq c^{*}+\phi_{2}(r,a),\\ 0,&x>c^{*}+\phi_{2}(r,a).\end{cases}

Hence,

Q2[w1~(,r)](x)\displaystyle Q^{2}[\tilde{w_{1}}(\,\cdot\,,r)](x) =w1(xcϕ2(r,a))w1(x+2r+c+ϕ2(r,a)).\displaystyle=w_{1}(x-c^{*}-\phi_{2}(r,a))-w_{1}(x+2r+c^{*}+\phi_{2}(r,a)).

Shifting the expression left by c+ϕ2(r,a)c^{*}+\phi_{2}(r,a) units yields

Q2[w1~(,r)](x+c+ϕ2(r,a))\displaystyle Q^{2}[\tilde{w_{1}}(\,\cdot\,,r)](x+c^{*}+\phi_{2}(r,a)) =w1~(x,r+c+ϕ2(r,a)).\displaystyle=\tilde{w_{1}}(x,r+c^{*}+\phi_{2}(r,a)). (2.17)

We are now prepared to prove the theorem using an inductive argument. Let AA be the level set defined in the statement of the theorem. Without loss of generality, we write A=[r0,r0]A=[-r_{0},r_{0}], for some r0>0r_{0}>0. We have

g(u0(x))={1,r0xr0,0,else.g(u_{0}(x))=\begin{cases}1,&-r_{0}\leq x\leq r_{0},\\ 0,&\text{else}.\end{cases}

for all xx. Applying the convolution operator yields

u1(x)\displaystyle u_{1}(x) =w1(xr0)w1(x+r0)\displaystyle=w_{1}(x-r_{0})-w_{1}(x+r_{0})
=w1~(xr0,r0).\displaystyle=\tilde{w_{1}}(x-r_{0},r_{0}).

Let (rn)n=0(r_{n})_{n=0}^{\infty} be a sequence of real numbers satisfying

rn+1=rn+c+ϕ2(rn,a).r_{n+1}=r_{n}+c^{*}+\phi_{2}(r_{n},a). (2.18)

If r0r_{0} is sufficiently large, we can show that rnr_{n} is strictly increasing and unbounded. This is due to our assumption that c>0c^{*}>0. Since ϕ2(r,a)c\phi_{2}(r,a)\to c^{*}, it suffices to assume |ϕ2(r,a)c|<c|\phi_{2}(r,a)-c^{*}|<c^{*}, i.e. r0r_{0} is sufficiently large so that ϕ2(r,a)>0\phi_{2}(r,a)>0 for all r>r0r>r_{0}. Plugging this into the recurrence (2.18) yields rn+1rn=c+ϕ2(r,a)>c>0r_{n+1}-r_{n}=c^{*}+\phi_{2}(r,a)>c^{*}>0.

This argument shows that, assuming r0r_{0} is sufficiently large, equation (2.17) may be applied recursively. From the principle of induction, it follows that u2n+1(x)=w1~(xrn,rn)u_{2n+1}(x)=\tilde{w_{1}}(x-r_{n},r_{n}) for all n0n\geq 0. Knowing that (rn)(r_{n}) is unbounded allows us to use equation (2.18) to compute the asymptotic spreading speed as the following limit.

limn(rn+1rn)=limn(c+ϕ2(rn,a))=c+limrϕ2(r,a)=2c.\lim_{n\to\infty}(r_{n+1}-r_{n})=\lim_{n\to\infty}(c^{*}+\phi_{2}(r_{n},a))=c^{*}+\lim_{r\to\infty}\phi_{2}(r,a)=2c^{*}.

This completes the proof. ∎

3 Examples

In this section, we construct the periodic traveling wave solution for uniform and Laplace kernels and derive formulas for the mean spreading speed cc^{*} in terms of the model parameters.

3.1 Laplace kernel

We applied our main results to the Laplace dispersal kernel, defined

k(x)=12e|x|.k(x)=\frac{1}{2}e^{-|x|}. (3.1)

k(x)k(x) is symmetric and has connected support, satisfying (H1), (H2), and (H3). We will now prove that k(x)k(x) satisfies (H5). Let A,BA,B\in\mathbb{R} with A>BA>B, let λ(0,1)\lambda\in(0,1), and define f(x)=k(xA)λk(xB)f(x)=k(x-A)-\lambda k(x-B). For r>0r>0 sufficiently large, we have

f(xr)f(xr)=k(xrA)λk(xrB)k(x+r+A)+λk(x+r+B)f(x-r)-f(-x-r)=k(x-r-A)-\lambda k(x-r-B)-k(x+r+A)+\lambda k(x+r+B)
={12ex(er+A+mer+BmerB+erA),x<rA,12exrA+12ex(erAmerB+mer+B),rAxrB,12ex(erA+merB)+12ex(merB+erA),rB<x<r+B,12ex(erA+merBmer+B)+12exrA,r+Bxr+Aex(erA+merBmer+B+er+A),x>r+A.=\begin{cases}\frac{1}{2}e^{x}\left(-e^{r+A}+me^{r+B}-me^{-r-B}+e^{-r-A}\right),&x<-r-A,\\ \frac{1}{2}e^{-x-r-A}+\frac{1}{2}e^{x}\left(e^{-r-A}-me^{-r-B}+me^{r+B}\right),&-r-A\leq x\leq-r-B,\\ \frac{1}{2}e^{-x}\left(-e^{-r-A}+me^{-r-B}\right)+\frac{1}{2}e^{x}\left(-me^{-r-B}+e^{-r-A}\right),&-r-B<x<r+B,\\ \frac{1}{2}e^{-x}\left(-e^{-r-A}+me^{-r-B}-me^{r+B}\right)+\frac{1}{2}e^{x-r-A},&r+B\leq x\leq r+A\\ e^{-x}\left(-e^{-r-A}+me^{-r-B}-me^{r+B}+e^{r+A}\right),&x>r+A.\end{cases}

ff is monotone on (,rA)(r+A,)(-\infty,-r-A)\cup(r+A,\infty) with f(±)=0f(\pm\infty)=0; thus it cannot have any sign changes there. On each of the intervals [rA,rB][-r-A,-r-B], (rB,r+B)(-r-B,r+B), and [r+B,r+A][r+B,r+A], ff has the form Cex+DexCe^{x}+De^{-x}. Depending on the sign of CC and DD, ff can have either one or zero sign changes on each interval. Summing over each interval, it follows that ff has at most 33 sign changes. This proves (H5); the proof of (H4) follows a similar argument.

Assuming w2<b||w_{2}||_{\infty}<b, the periodic traveling wave profiles are given by

w1(x)={112ex,if x0,12ex,if x>0,w_{1}(x)=\begin{cases}1-\frac{1}{2}e^{x},&\text{if }x\leq 0,\\ \frac{1}{2}e^{-x},&\text{if }x>0,\end{cases} (3.2)

and

w2(xc)={m+(12(1m)eΦ1(b)12eΦ1(a))ex,x<Φ1(b),112eΦ1(a)ex12(1m)eΦ1(b)ex,Φ1(b)xΦ1(a),(12eΦ1(a)12(1m)eΦ1(b))ex,Φ1(a)<x,w_{2}(x-c^{*})=\begin{cases}m+\left(\frac{1}{2}(1-m)e^{-\Phi_{1}(b)}-\frac{1}{2}e^{-\Phi_{1}(a)}\right)e^{x},&x<\Phi_{1}(b),\\ 1-\frac{1}{2}e^{-\Phi_{1}(a)}e^{x}-\frac{1}{2}(1-m)e^{\Phi_{1}(b)}e^{-x},&\Phi_{1}(b)\leq x\leq\Phi_{1}(a),\\ \left(\frac{1}{2}e^{\Phi_{1}(a)}-\frac{1}{2}(1-m)e^{\Phi_{1}(b)}\right)e^{-x},&\Phi_{1}(a)<x,\end{cases} (3.3)

with

c=12log{12a(eΦ1(a)(1m)eΦ1(b)),aa,eΦ1(a)(1a+(1a)2(1m)eΦ1(b)Φ1(a)),a<a<b,2(ma)[eΦ1(a)(1m)eΦ1(b)]1,ab.c^{*}=\frac{1}{2}\log\begin{cases}\frac{1}{2a}\left(e^{\Phi_{1}(a)}-(1-m)e^{\Phi_{1}(b)}\right),&a\leq\ell_{a},\\ e^{\Phi_{1}(a)}\left(1-a+\sqrt{(1-a)^{2}-(1-m)e^{\Phi_{1}(b)-\Phi_{1}(a)}}\right),&\ell_{a}<a<\ell_{b},\\ 2\,(m-a)\left[e^{-\Phi_{1}(a)}-(1-m)e^{-\Phi_{1}(b)}\right]^{-1},&a\geq\ell_{b}.\end{cases} (3.4)

a\ell_{a} and b\ell_{b} are constants given by

a=12(1(1m)eΦ1(b)Φ1(a)),b=12(1+meΦ1(b)Φ1(a)).\ell_{a}=\frac{1}{2}\left(1-(1-m)e^{\Phi_{1}(b)-\Phi_{1}(a)}\right),\quad\ell_{b}=\frac{1}{2}\left(1+m-e^{\Phi_{1}(b)-\Phi_{1}(a)}\right). (3.5)

The formulas for Φ1\Phi_{1} and Φ2\Phi_{2} are given by

Φ1()={log(2),12,log(22),>12,\Phi_{1}(\ell)=\begin{cases}-\log(2\ell),&\ell\leq\frac{1}{2},\\ \log(2-2\ell),&\ell>\frac{1}{2},\end{cases} (3.6)

and

Φ2()=c+log{12(eΦ1(a)(1m)eΦ1(b)),w2(Φ1(a)),eΦ1(a)(1+(1)2(1m)eΦ1(b)Φ1(a)),w2(Φ1(a))<<w2(Φ1(b)),2(m)[eΦ1(a)(1m)eΦ1(b)]1,w2(Φ1(b)).\Phi_{2}(\ell)=-c^{*}+\log\begin{cases}\frac{1}{2\ell}\left(e^{\Phi_{1}(a)}-(1-m)e^{\Phi_{1}(b)}\right),&\ell\leq w_{2}(\Phi_{1}(a)),\\ e^{\Phi_{1}(a)}\left(1-\ell+\sqrt{(1-\ell)^{2}-(1-m)e^{\Phi_{1}(b)-\Phi_{1}(a)}}\right),&w_{2}(\Phi_{1}(a))<\ell<w_{2}(\Phi_{1}(b)),\\ 2\,(m-\ell)\left[e^{-\Phi_{1}(a)}-(1-m)e^{-\Phi_{1}(b)}\right]^{-1},&\ell\geq w_{2}(\Phi_{1}(b)).\end{cases} (3.7)

The critical spreading speed cc^{*} for the Laplace kernel thus falls into one of three cases:

  1. Case 1.

    a12a\leq\frac{1}{2}, b12b\leq\frac{1}{2}:

    c=12log{ba(1m)4a2b,a<ba(1m)2b,1a+b(1a)2a(1m)2ab,ba(1m)2bab(1+m)a2b,amb(1m)a,a>b(1+m)a2b.c^{*}=\frac{1}{2}\log\begin{cases}\frac{b-a(1-m)}{4a^{2}b},&a<\frac{b-a(1-m)}{2b},\\ \frac{1-a+\sqrt{b(1-a)^{2}-a(1-m)}}{2a\sqrt{b}},&\frac{b-a(1-m)}{2b}\leq a\leq\frac{b(1+m)-a}{2b},\\ \frac{a-m}{b(1-m)-a},&a>\frac{b(1+m)-a}{2b}.\end{cases}
  2. Case 2.

    a12a\leq\frac{1}{2}, b>12b>\frac{1}{2}:

    c=12log{14a(1m)(1b)4a2,a<14a(1m)(1b)2,1a+(1a)24a(1m)(1b)2a,14a(1m)(1b)2a1+m4a(1b)2,4(am)(1b)1m4a(1b),a>1+m4a(1b)2.c^{*}=\frac{1}{2}\log\begin{cases}\frac{1-4a(1-m)(1-b)}{4a^{2}},&a<\frac{1-4a(1-m)(1-b)}{2},\\ \frac{1-a+\sqrt{(1-a)^{2}-4a(1-m)(1-b)}}{2a},&\frac{1-4a(1-m)(1-b)}{2}\leq a\leq\frac{1+m-4a(1-b)}{2},\\ \frac{4(a-m)(1-b)}{1-m-4a(1-b)},&a>\frac{1+m-4a(1-b)}{2}.\end{cases}
  3. Case 3.

    a>12a>\frac{1}{2}, b>12b>\frac{1}{2}:

    c=12log{1a(1m)(1b)a,a<1a(1m)(1b)2(1a),2(1a)(1a+(1a)3(1m)(1b)1a),1a(1m)(1b)2(1a)ab1+m(1a)2(1a),4(ma)(1a)(1b)1b+m(1a),a>b1+m(1a)2(1a).c^{*}=\frac{1}{2}\log\begin{cases}\frac{1-a-(1-m)(1-b)}{a},&a<\frac{1-a-(1-m)(1-b)}{2(1-a)},\\ 2\,(1-a)\left(1-a+\sqrt{\frac{(1-a)^{3}-(1-m)(1-b)}{1-a}}\right),&\frac{1-a-(1-m)(1-b)}{2(1-a)}\leq a\leq\frac{b-1+m(1-a)}{2(1-a)},\\ \frac{4(m-a)(1-a)(1-b)}{1-b+m(1-a)},&a>\frac{b-1+m(1-a)}{2(1-a)}.\end{cases}

Figure 3 shows the results of a numerical simulation with growth parameters were a=0.3a=0.3, m=0.5m=0.5, and b=0.8b=0.8 with the Laplace dispersal kernel, for an average spreading speed of c=12ln229c^{*}=\frac{1}{2}\ln\frac{22}{9}. Figures 3(a) and 3(b) show heatmaps of the periodic traveling wave solution and a spreading solution with compactly supported initial condition. Figure 3(c) shows a numerical simulation of the periodic traveling wave solution, while Figure 3(d) shows the asymptotic convergence of the intermediate wave-speed sequence cn()c_{n}(\ell), evaluated at the Allee threshold =a\ell=a.

Refer to caption
(a) Heatmap of the periodic traveling wave.
Refer to caption
(b) Heatmap of a periodic spreading solution.
Refer to caption
(c) Population density curve of the periodic traveling wave.
Refer to caption
(d) Time series plot of intermediate spreading speed.
Figure 3: Heatmaps of (a) the periodic traveling wave, and (b) the corresponding spreading solution with compactly supported initial data, and Laplace dispersal kernel. In figure (d), the horizontal blue lines correspond to the theoretical values derived in Corollary 2.3.

3.2 Uniform kernel

The uniform dispersal kernel is given by

k(x)={12,if |x|1,0,if |x|>1.k(x)=\begin{cases}\frac{1}{2},&\text{if }|x|\leq 1,\\ 0,&\text{if }|x|>1.\end{cases} (3.8)

Like the previous example, k(x)k(x) is symmetric around zero. We can verify (H4) by letting A>BA>B and λ(0,1)\lambda\in(0,1) and defining f(x)=k(xA)λk(xB)f(x)=k(x-A)-\lambda k(x-B). There are two cases:

  1. Case 1.

    If |AB|<2|A-B|<2, then

    f(x)={0,xB1,λ2,B1<xA1,1λ2,A1<x<B+1,12,B+1x<A+10,xA+1.f(x)=\begin{cases}0,&x\leq B-1,\\ -\frac{\lambda}{2},&B-1<x\leq A-1,\\ \frac{1-\lambda}{2},&A-1<x<B+1,\\ \frac{1}{2},&B+1\leq x<A+1\\ 0,&x\geq A+1.\end{cases}
  2. Case 2.

    If |AB|2|A-B|\geq 2, then

    f(x)={0,xB1,λ2,B1<x<B+1,0,B+1xA1,12,A1<x<A+10,xA+1.f(x)=\begin{cases}0,&x\leq B-1,\\ -\frac{\lambda}{2},&B-1<x<B+1,\\ 0,&B+1\leq x\leq A-1,\\ \frac{1}{2},&A-1<x<A+1\\ 0,&x\geq A+1.\end{cases}

In both cases, the number of sign changes is exactly 1. To extend this argument to (H5), observe that since k(x)k(x) has compact support, we have

f(xr)f(rx)\displaystyle f(x-r)-f(r-x) =k(xrA)λk(xrB)k(x+r+A)+λk(x+r+B)\displaystyle=k(x-r-A)-\lambda k(x-r-B)-k(x+r+A)+\lambda k(x+r+B)
={f(rx),x<r+1,0,r+1xr1,f(xr),x>r1,\displaystyle=\begin{cases}-f(r-x),&x<-r+1,\\ 0,&-r+1\leq x\leq r-1,\\ f(x-r),&x>r-1,\end{cases}

for sufficiently large rr.

The alternating wave profiles are given by

w1(x)={1,x(,1),1212x,x[1,1],0,x(1,),\displaystyle w_{1}(x)=\begin{cases}1,&x\in(-\infty,-1),\\ \frac{1}{2}-\frac{1}{2}x,&x\in[-1,1],\\ 0,&x\in(1,\infty),\end{cases} (3.9)

and

w2(x+c)={m,x(,2b),1m2x+m+bmb,x[2b,2a),m2x+m+bmba,x[2a,22b),12xa+1,x[22b,22a],0,x(2+2a,),\displaystyle w_{2}(x+c^{*})=\begin{cases}m,&x\in(-\infty,-2b),\\ \frac{1-m}{2}x+m+b-mb,&x\in[-2b,-2a),\\ -\frac{m}{2}x+m+b-mb-a,&x\in[-2a,2-2b),\\ -\frac{1}{2}x-a+1,&x\in[2-2b,2-2a],\\ 0,&x\in(2+2a,\infty),\end{cases} (3.10)

with

c={12a,if ab/2,1b+b2am,if a>b/2.c^{*}=\begin{cases}1-2a,&\text{if }a\leq b/2,\\ 1-b+\frac{b-2a}{m},&\text{if }a>b/2.\end{cases} (3.11)

w2w_{2} has a global maximum at x=2ax=-2a so that w2=m+(ba)(1m)||w_{2}||_{\infty}=m+(b-a)(1-m). Thus, a sufficient condition for the existence of the periodic traveling wave is w2(2a)=m+(ba)(1m)<bw_{2}(-2a)=m+(b-a)(1-m)<b.

The intermediate wave speeds are given by

c1()\displaystyle c_{1}^{*}(\ell) =12\displaystyle=1-2\ell (3.12)

and

c2()\displaystyle c_{2}^{*}(\ell) ={14a+2,ab/2,12b+2b4am+2,a>b/2,\displaystyle=\begin{cases}1-4a+2\ell,&a\leq b/2,\\ 1-2b+\frac{2b-4a}{m}+2\ell,&a>b/2,\end{cases} (3.13)

for 0<<m+(ba)(1m)0<\ell<m+(b-a)(1-m). By evaluating the intermediate wave speeds at =a\ell=a, we obtain c1(a)=c2(a)c_{1}^{*}(a)=c_{2}^{*}(a) if ab/2a\leq b/2, and |c1(a)c2(a)|=2(2ab)(1mm)>0|c_{1}^{*}(a)-c_{2}^{*}(a)|=2(2a-b)(\frac{1-m}{m})>0 if a>b/2a>b/2. So for a>b/2a>b/2, the traveling wave is periodic with two different intermediate wave speeds. Furthermore, the difference between these two intermediate speeds is increasing with respect to aa, the Allee threshold, and decreasing with respect to bb, the overcompensation threshold.

Figure 4(a) depicts the periodic traveling wave solution (2.4) with the uniform kernel with growth parameters are a=0.325a=0.325, b=0.6b=0.6, and m=0.45m=0.45. The mean spreading speed can be analytically calculated as c=1345c^{*}=\frac{13}{45}. The intermediate speeds are c1(a)=720c_{1}^{*}(a)=\frac{7}{20} and c2(a)=41180c_{2}^{*}(a)=\frac{41}{180}. Figure 4(b) demonstrates the spreading phenomena with initial domain size r0=1r_{0}=1. Since our uniform kernel is compact with support [1,1][-1,1], a sufficient lower bound on the initial domain size is r03r_{0}\geq 3.

Refer to caption
(a) Heaviside initial data.
Refer to caption
(b) Compact initial data (r0=1r_{0}=1).
Figure 4: Population density plots of (a) the periodic traveling wave, and (b) a spreading solution with uniform dispersal.

References

  • [1] W. C. Allee. 1931. Animal Aggregations. A Study on General Sociology. University of Chicago Press, Chicago, IL.
  • [2] A. Hastings and K. Higgins. 1994. Persistence of transients in spatially structured ecological models. Science 263: 1133-1136.
  • [3] M. Kot and W. M. Schaffer. 1986. Discrete-time growth-dispersal models. Math. Biosci. 80: 109-136.
  • [4] M. Kot. 1989. Diffusion-driven period doubling bifurcations. Biosystems 22: 279-287.
  • [5] M. Kot. 1992. Discrete-time traveling waves: Ecological examples. J. Math. Biol. 30: 413-436.
  • [6] M. Kot, M. A. Lewis, and P. van den Driessche. 1996. Dispersal data and the spread of invading organisms. Ecology 77: 2027-2042.
  • [7] M. Kot. 2001. Elements of Mathematical Ecology. Cambridge University Press. Cambridge, United Kingdom.
  • [8] R. Lui. 1982. A nonlinear integral operator arising from a model in population genetics. I. Monotone initial data. SIAM. J. Math. Anal. 13: 913-937.
  • [9] R. Lui. 1982. A nonlinear integral operator arising from a model in population genetics. II. Initial data with compact support. SIAM. J. Math. Anal. 13: 938-953.
  • [10] R. Lui. 1983. Existence and stability of traveling wave solutions of a nonlinear integral operator. J. Math. Biol. 16:199-220.
  • [11] F. Lutscher. 2019. Integrodifference Equations in Spatial Ecology. Springer.
  • [12] M. Neubert, M. Kot, and M. A. Lewis. 1995. Dispersal and pattern formation in a discrete-time predator-prey model. Theor. Pop. Biol. 48: 7-43.
  • [13] G. Otto. 2017. Non-spreading Solutiona in a Integro-Difference Model Incorporating Allee and Overcompensation Effects. Ph. D thesis, University of Louisville.
  • [14] M. Slatkin. 1973. Gene flow and selection in a cline. Genetice 75: 733-756.
  • [15] L. L. Sullivan, B. Li, T. E. X. Miller, M. G. Neubert, and A. K. Shaw. 2017. Density dependence in demography and dispersal generates fluctuating invasion speeds. Proc. Natl. Acad. Sci. USA 114: 5053-5058.
  • [16] M. H. Wang, M. Kot, and M. G. Neubert. 2002. Integrodifference equations, Allee effects, and invasions. J. Math. Biol. 44: 150-168.
  • [17] H. F. Weinberger. 1978. Asymptotic behavior of a model in population genetics. Nonlinear Partial Differential Equations and Applications, ed. J. M. Chadam. Lecture Notes in Mathematics 648: 47-96. Springer-Verlag, Berlin.
  • [18] H. F. Weinberger. 1982. Long-time beahvior of a class of biological models. SIAM. J. Math. Anal. 13: 353-396.