betaFeedback

WorksMathematics

Exact multiplicity for a Minkowski-curvature Neumann problem with a step weight

Abstract

The DEG1 challenge asks whether positive solutions of a super-exponential Minkowski-curvature equation are unique for step weights with a single sign change and negative mean. In this paper, we give a Neumann counterexample by studying (u′1−(u′)2)′+λaμ,h(t)(eu2−1)=0\left(\frac{u'}{\sqrt{1-(u')^2}}\right)' + λa_{μ,h}(t)(e^{u^2}-1)=0 with u′(0)=u′(2h)=0u'(0)=u'(2h)=0, where aμ,h=1a_{μ,h}=1 on (0,h)(0,h) and aμ,h=−μa_{μ,h}=-μ on (h,2h)(h,2h), with μ>1μ>1. Using the shooting method, we prove that the positive solution curve has exactly two nondegenerate folds for (λ,h)(λ,h) in a nonempty open region. As μμ increases through the corresponding critical values, the number of positive solutions is successively 11, 22, 33, 22, and 11.

Introduction and main result

The DEG1 challenge [1] asks whether the equation

(u′1−(u′)2)′+a(t)(eup−1)=0,p>1,\left(\frac{u'}{\sqrt{1-(u')^2}}\right)' + a(t)(e^{u^p}-1)=0,\qquad p>1,

has a unique positive solution under Neumann or periodic boundary conditions when aa is a step function with one sign change and negative mean. Boscaggin, Feltrin, and Zanolin [2 Corollary 2.1] proved existence for both boundary conditions under hypotheses that cover these weights. They also noted that numerical experiments suggested uniqueness in the super-exponential case [3 Section 4.2].

We give a Neumann counterexample with p=2p=2 and two steps of equal length. Consider

{(u′1−(u′)2)′+λaμ,h(t)(eu2−1)=0,0<t<2h,u′(0)=u′(2h)=0,(1)\left\{ \begin{aligned} \left(\frac{u'}{\sqrt{1-(u')^2}}\right)' + \lambda a_{\mu,h}(t)(e^{u^2}-1)&=0,\qquad0<t<2h,\\ u'(0)=u'(2h)&=0, \end{aligned} \right. \tag*{(1)}

where λ,h>0\lambda,h>0, μ>1\mu>1, and

aμ,h(t)={1,0<t<h,−μ,h<t<2h.(2)a_{\mu,h}(t)= \begin{cases} 1, & 0<t<h,\\ -\mu, & h<t<2h. \end{cases} \tag*{(2)}

A positive solution is a function u∈C1([0,2h])u\in C^{1}([0,2h]) with u>0u>0 and ∣u′∣<1|u'|<1 on [0,2h][0,2h], whose flux u′/1−(u′)2u'/\sqrt{1-(u')^{2}} is absolutely continuous and which satisfies (1) almost everywhere. We denote the number of distinct positive solutions by N(λ,h,μ)N(\lambda,h,\mu).

Multiplicity results for indefinite Minkowski-curvature equations include pairs of positive radial Neumann solutions for power nonlinearities [4] and positive periodic and subharmonic solutions [5]. Shooting methods have also been used to study multiplicity for radial Neumann problems [6].

Ma, Su, and Zhao [7] studied the Neumann problem

−(u′1−(u′)2)′=λa(t)f(u),0<t<T,u′(0)=u′(T)=0,-\left(\frac{u'}{\sqrt{1-(u')^{2}}}\right)'=\lambda a(t)f(u),\qquad0<t<T,\qquad u'(0)=u'(T)=0,

with an integrable weight aa allowed to change sign. Under suitable growth conditions on ff, they obtained at least two positive solutions for sufficiently large λ\lambda. For the same form of equation on (0,1)(0,1), Su, Ma, and Yang [8] fixed a continuous sign-changing weight aa and used bifurcation methods to establish the existence of SS-shaped connected components of positive solutions as the overall multiplier λ\lambda varies. In (1) we fix the positive coefficient λ\lambda and vary the negative one through the ratio μ\mu. This lets us study how the number of positive solutions changes with the balance between the two steps.

In the semilinear setting, Feltrin and Sovrano [9 Theorem 1.1] used shooting to obtain at least three positive Neumann solutions for −u′′=(λa+−μa−)g(u)-u''=(\lambda a^{+}-\mu a^{-})g(u) under their assumptions. Feltrin, Sovrano, and Tellini [10 Section 1] studied solution counts for −u′′=aλ,μ(t)u2(1−u)-u''=a_{\lambda,\mu}(t)u^{2}(1-u) with Neumann conditions and a step weight having two positive intervals separated by a negative one. Exact multiplicity and bifurcation curves for autonomous Minkowski-curvature Dirichlet problems were studied in [11, 12]. In particular, Huang and Wang [13 Theorem 4] studied the autonomous perturbed Gelfand problem

−(u′1−(u′)2)′=λexp⁡(aua+u),−d<t<d,u(−d)=u(d)=0.-\left(\frac{u'}{\sqrt{1-(u')^{2}}}\right)'=\lambda\exp\left(\frac{au}{a+u}\right),\qquad-d<t<d,\qquad u(-d)=u(d)=0.

They used a time map relating the maximum of a positive solution to the half-length dd to classify the shape of the (λ,∥u∥∞)(\lambda,\|u\|_{\infty}) curve as aa and dd vary. In particular, the curve has exactly two turning points when aa is above a threshold and dd is sufficiently large. For nonautonomous Dirichlet problems, SS-shaped connected components and more complicated bifurcation curves have been obtained in [14, 15].

For (λ,h)(\lambda,h) in the open region of Theorem 1.1, matching the trajectories across the jump in the weight gives exactly one μ>1\mu>1 for each initial value s=u(0)>0s=u(0)>0. The resulting graph parametrizes all positive solutions of (1.1). A uniform C2C^2 expansion on bounded amplitude intervals gives two nondegenerate folds. A variational estimate proves strict monotonicity at large amplitudes, excluding further folds on the unbounded part of the graph. Counting intersections with the lines of constant μ\mu then gives the exact number of positive solutions for every μ>1\mu> 1 in this regime.

Theorem 1.1. There exist constants L∗>0L_* > 0 and h∗>0h_* > 0 such that, for every

(λ,h)∈D:={(λ,h)∈(0,∞)2:0<h<h∗,L∗<λh<2L∗},(3)(\lambda,h) \in\mathcal{D} := \{(\lambda,h) \in(0,\infty)^2 : 0 < h < h_*, \quad L_* < \lambda h < 2L_*\}, \tag*{(3)}

there are unique critical values 1<μ−(λ,h)<μ+(λ,h)<∞1 < \mu_-(\lambda,h) < \mu_+(\lambda,h) < \infty such that, for every μ>1\mu> 1,

N(λ,h,μ)={1,1<μ<μ−(λ,h),2,μ=μ−(λ,h),3,μ−(λ,h)<μ<μ+(λ,h),2,μ=μ+(λ,h),1,μ>μ+(λ,h).(4)N(\lambda,h,\mu) = \begin{cases} 1, & 1 < \mu< \mu_-(\lambda,h),\\ 2, & \mu= \mu_-(\lambda,h),\\ 3, & \mu_-(\lambda,h) < \mu< \mu_+(\lambda,h),\\ 2, & \mu= \mu_+(\lambda,h),\\ 1, & \mu> \mu_+(\lambda,h). \end{cases} \tag*{(4)}

The functions μ−\mu_- and μ+\mu_+ are smooth on D\mathcal{D}. For each fixed (λ,h)∈D(\lambda,h) \in\mathcal{D}, the set of positive solutions is a single smooth curve parametrized by s=u(0)>0s = u(0) > 0. Its projection onto the μ\mu-axis has exactly two critical points, both nondegenerate. At each critical value precisely one solution is degenerate, with a one-dimensional kernel for the linearized Neumann problem. Every other positive solution is nondegenerate.

For any (λ,h)∈D(\lambda,h) \in\mathcal{D} and μ∈(μ−(λ,h),μ+(λ,h))\mu\in(\mu_-(\lambda,h),\mu_+(\lambda,h)), the weight λaμ,h\lambda a_{\mu,h} has one sign change and integral λh(1−μ)<0\lambda h(1-\mu) < 0. The three positive solutions therefore disprove uniqueness in the Neumann part of the DEG1 challenge. Uniqueness is recovered for μ>μ+(λ,h)\mu> \mu_+(\lambda,h).

Section 2 constructs the global shooting curve. Section 3 analyzes its shape and proves the main theorem. Section 4 derives the critical-value asymptotics and illustrates the curve numerically.

The global shooting curve

Throughout the paper, set

g(u)=eu2−1,G(u)=∫0ug(r) dr,g(u) = e^{u^2} - 1,\qquad G(u) = \int_0^u g(r)\,\mathrm{d}r,
ϕ(p)=p1−p2,ψ(v)=v1+v2,K(v)=1+v2−1.(5)\phi(p) = \frac{p}{\sqrt{1-p^2}},\qquad\psi(v) = \frac{v}{\sqrt{1+v^2}},\qquad K(v) = \sqrt{1+v^2} - 1. \tag*{(5)}

Thus ψ=ϕ−1\psi= \phi^{-1} and K′=ψK' = \psi. With v=−ϕ(u′)v = -\phi(u'), problem (1.1) becomes

u′=−ψ(v),v′={λg(u),0<t<h,−λμg(u),h<t<2h,v(0)=v(2h)=0.(6)u' = -\psi(v),\qquad v' = \begin{cases} \lambda g(u), & 0 < t < h,\\ -\lambda\mu g(u), & h < t < 2h, \end{cases} \qquad v(0) = v(2h) = 0. \tag*{(6)}

For each finite initial condition, ∣u′∣<1|u'| < 1 bounds uu on [0,2h][0,2h], and the equation for vv then bounds v′v'. Thus neither state variable can escape to infinity in finite time. Applying the continuation criterion [16 Corollary 2.16] to each autonomous system gives existence throughout [0,2h][0,2h] by concatenation at hh. Both vector fields are smooth, so their flows depend smoothly on the initial data and parameters by [16 Theorem 2.11]. The composition preserves this smooth dependence.

Lemma 2.1. Every positive solution of (1.1) satisfies

v(t)>0(0<t<2h),u′(t)<0(0<t<2h).(7)v(t)>0 \quad(0<t<2h), \qquad u'(t)<0 \quad(0<t<2h). \tag*{(7)}

Moreover,

μ=∫0hg(u(t)) dt∫h2hg(u(t)) dt>1.(8)\mu=\frac{\int_{0}^{h}g(u(t))\,\mathrm{d}t}{\int_{h}^{2h}g(u(t))\,\mathrm{d}t}>1. \tag*{(8)}

Proof. The sign assertions in (7) follow from [3 Section 2.1, Eqs. (2.5)–(2.6)] with (x,y)=(u,−v)(x,y)=(u,-v), (a+,a−)=(λ,λμ)(a_{+},a_{-})=(\lambda,\lambda\mu), and (τ,T)=(h,2h)(\tau,T)=(h,2h). Integrating the equation for vv gives the ratio in (8). It is greater than one because gg is strictly increasing and u(t)>u(t+h)u(t)>u(t+h) for 0≤t≤h0\leq t\leq h. □

We now fix L=λhL=\lambda h. The following condition ensures that every positive initial value can be used for shooting:

Lh2eh2<1.(9)Lh^{2}e^{h^{2}}<1. \tag*{(9)}

Lemma 2.2. Assume (9). For every s>0s>0, the solution of the first-half system in (2.2) with u(0)=su(0)=s and v(0)=0v(0)=0 is positive on [0,h][0,h]. In particular,

c(s):=u(h)>0,q(s):=v(h)>0.(10)c(s):=u(h)>0,\qquad q(s):=v(h)>0. \tag*{(10)}

Proof. If s≥hs\geq h, then u(t)>s−t≥0u(t)>s-t\geq0 for 0<t≤h0<t\leq h, since ∣u′∣<1|u'|<1. Suppose that 0<s<h0<s<h and that uu has a first zero in [0,h][0,h]. Before this zero, 0<u≤s0<u\leq s and 0≤v(t)≤λhg(s)=Lg(s)0\leq v(t)\leq\lambda h g(s)=Lg(s). Consequently,

s−u(t)≤hLg(s)≤hLs2eh2<(Lh2eh2)s<s,s-u(t)\leq hLg(s)\leq hLs^{2}e^{h^{2}}<(Lh^{2}e^{h^{2}})s<s,

which is impossible at the first zero. □

Proposition 2.3. Assume (9). For each s>0s>0, there is exactly one μ=μL,h(s)>0\mu=\mu_{L,h}(s)>0 for which the solution of (2.2) with u(0)=su(0)=s is positive and satisfies v(2h)=0v(2h)=0. The function μL,h:(0,∞)→(1,∞)\mu_{L,h}:(0,\infty)\to(1,\infty) is smooth, and its graph parametrizes all positive solutions.

For fixed L,hL,h, let

F(s,μ)=v(2h;s,μ),(11)F(s,\mu)=v(2h;s,\mu), \tag*{(11)}

where the dependence on L,hL,h is suppressed. At every positive solution,

Fμ(s,μ)<0.(12)F_{\mu}(s,\mu)<0. \tag*{(12)}

Proof. Fix s>0s > 0 and use (10) as initial data for the second half of the trajectory. In this proof only, time on that half is measured from zero. We use the energy identity and time quadrature for the negative step from [3 Section 2.1, Eqs. (2.7)–(2.8)]. In our notation, the conserved quantity is

K(v)−λμG(u)=K(q)−λμG(c).(13)K(v) - \lambda\mu G(u) = K(q) - \lambda\mu G(c). \tag*{(13)}

A positive stopping point (b,0)(b,0) therefore satisfies

G(b)=G(c)−K(q)λμ.(14)G(b) = G(c) - \frac{K(q)}{\lambda\mu}. \tag*{(14)}

Such a point exists precisely when

μ>μ0(s):=K(q)λG(c).(15)\mu> \mu_0(s) := \frac{K(q)}{\lambda G(c)}. \tag*{(15)}

In that case it is unique and belongs to (0,c)(0,c). The trajectory has u′<0u' < 0 and v′<0v' < 0 before reaching (b,0)(b,0). Solving (13) for vv gives its stopping time:

Ts(μ)=∫bc1+λμ(G(u)−G(b))λμ(G(u)−G(b))[2+λμ(G(u)−G(b))] du.(16)T_s(\mu) = \int_b^c \frac{1+\lambda\mu\left(G(u)-G(b)\right)}{\sqrt{\lambda\mu\left(G(u)-G(b)\right)\left[2+\lambda\mu\left(G(u)-G(b)\right)\right]}}\,\mathrm{d}u. \tag*{(16)}

The singularity at u=b>0u=b>0 is of order (u−b)−1/2(u-b)^{-1/2}, so this time is finite. The stopping event is transverse because v′(Ts)=−λμg(b)<0v'(T_s)=-\lambda\mu g(b)<0.

To prove strict monotonicity in μ\mu, write z=−vz=-v on the second half-interval. Then u′=ψ(z)u'=\psi(z) and z′=λμg(u)z'=\lambda\mu g(u), with (u(0),z(0))=(c,−q)(u(0),z(0))=(c,-q). The derivatives X=∂μuX=\partial_\mu u and Y=∂μzY=\partial_\mu z satisfy

X′=ψ′(z)Y,Y′=λg(u)+λμg′(u)X,X(0)=Y(0)=0.(17)X'=\psi'(z)Y,\qquad Y'=\lambda g(u)+\lambda\mu g'(u)X,\qquad X(0)=Y(0)=0. \tag*{(17)}

On the positive trajectory, this is a cooperative linear system with a strictly positive forcing in its second equation. Hence X(t)>0X(t)>0 and Y(t)>0Y(t)>0 for 0<t≤Ts(μ)0<t\le T_s(\mu). Differentiating z(Ts(μ),μ)=0z(T_s(\mu),\mu)=0 yields

Ts′(μ)=−Y(Ts(μ))λμg(b)<0.(18)T_s'(\mu)=-\frac{Y(T_s(\mu))}{\lambda\mu g(b)}<0. \tag*{(18)}

As μ↓μ0(s)\mu\downarrow\mu_0(s), equation (14) gives b↓0b\downarrow0. The integrand in (16), extended by zero to (0,b)(0,b), converges pointwise on (0,c)(0,c) to

1+λμ0G(u)λμ0G(u)(2+λμ0G(u)).\frac{1+\lambda\mu_0G(u)}{\sqrt{\lambda\mu_0G(u)\left(2+\lambda\mu_0G(u)\right)}}.

Since G(u)∼u3/3G(u)\sim u^3/3 at zero, this limit is asymptotic to a positive constant times u−3/2u^{-3/2}. Fatou’s lemma gives Ts(μ)→∞T_s(\mu)\to\infty. As μ→∞\mu\to\infty, we have b→cb\to c, and integration of −v′=λμg(u)≥λμg(b)-v'=\lambda\mu g(u)\ge\lambda\mu g(b) gives Ts(μ)≤q/(λμg(b))→0T_s(\mu)\le q/(\lambda\mu g(b))\to0. Thus there is exactly one μ>μ0(s)\mu>\mu_0(s) with Ts(μ)=hT_s(\mu)=h. The resulting solution is positive on the whole interval, and Lemma 2.1 shows that μ>1\mu>1. Conversely, every positive Neumann solution must stop in this way.

At a matched trajectory, (17), evaluated at the fixed right-hand time hh, gives Fμ=−Y(h)<0F_\mu=-Y(h)<0. The ordinary implicit function theorem therefore gives smooth dependence of μL,h(s)\mu_{L,h}(s) on ss, and also on L,hL,h as long as (9) holds. Uniqueness identifies the local implicit functions with the global curve just constructed.

Lemma 2.4. Under (9),

lim⁡s→∞μL,h(s)=∞.(19)\lim_{s\to\infty}\mu_{L,h}(s)=\infty. \tag*{(19)}

Proof. For s>hs>h, the first half of the trajectory satisfies u(t)>s−hu(t)>s-h and v(t)≥Astv(t)\ge A_s t, where As:=λg(s−h)A_s:=\lambda g(s-h). It follows that

c(s)≤s−∫0hψ(Ast) dt=s−1+As2h2−1As≤s−h+1As.(20)\begin{aligned} c(s)\le s-\int_0^h\psi(A_s t)\,\mathrm{d}t=s-\frac{\sqrt{1+A_s^2h^2}-1}{A_s} \\ &\le s-h+\frac{1}{A_s}. \tag*{(20)} \end{aligned}

For all sufficiently large ss, this gives c(s)≤s−3h/4c(s)\le s-3h/4. On [0,h/2][0,h/2] we have u(t)≥s−h/2u(t)\ge s-h/2, whereas on [h,2h][h,2h] we have u(t)≤c(s)u(t)\le c(s). Applying (8), we obtain

μL,h(s)≥g(s−h/2)2g(s−3h/4)⟶∞.\mu_{L,h}(s)\ge\frac{g(s-h/2)}{2g(s-3h/4)}\longrightarrow\infty.

The last limit follows from (s−h/2)2−(s−3h/4)2=hs/2−5h2/16(s-h/2)^2-(s-3h/4)^2=hs/2-5h^2/16.

Proof of the main theorem

By Proposition 2.3, for fixed L=λhL=\lambda h and hh, counting positive solutions amounts to counting the initial values s>0s>0 for which μL,h(s)=μ\mu_{L,h}(s)=\mu. We first analyze the leading term of the shooting curve on bounded intervals. We then prove that the two critical points persist and that the curve is strictly increasing for s≥2s\ge2, so these are its only turning points. The two endpoint limits and the three monotone branches give the exact count, and the linearized shooting equation identifies the degenerate solutions.

The reduced shooting curve

A uniform expansion on bounded amplitude intervals

Define, for s>0s>0 and k>0k>0,

ℓ(s)=g′(s)g(s)=2s1−e−s2,W(k)=k1+k2−arsinh⁡kk2,(21)\ell(s)=\frac{g'(s)}{g(s)}=\frac{2s}{1-e^{-s^2}},\qquad W(k)=\frac{k\sqrt{1+k^2}-\operatorname{arsinh}k}{k^2}, \tag*{(21)}

and set W(0)=0W(0)=0. An integral representation that will be useful below is

W(k)=2∫01rψ(kr) dr=2k∫01r21+k2r2 dr.(22)W(k)=2\int_0^1r\psi(kr)\,\mathrm{d}r=2k\int_0^1\frac{r^2}{\sqrt{1+k^2r^2}}\,\mathrm{d}r. \tag*{(22)}

The reduced shooting function is

ML(s)=ℓ(s)W(Lg(s)),s>0.(23)M_L(s)=\ell(s)W\bigl(Lg(s)\bigr),\qquad s>0. \tag*{(23)}

Proposition 3.1. Let S>0S>0 and let J⋐(0,∞)J\Subset(0,\infty) be a compact interval. For sufficiently small h>0h>0, uniformly for L∈JL\in J, the shooting function extends smoothly to s=0s=0, with μL,h(0)=1\mu_{L,h}(0)=1. There is a constant CJ,SC_{J,S} such that

sup⁡L∈J∥μL,h−1−hML∥C2([0,S])≤CJ,Sh2.(24)\sup_{L\in J}\left\|\mu_{L,h}-1-hM_L\right\|_{C^2([0,S])}\le C_{J,S}h^2. \tag*{(24)}

Here MLM_L has a smooth extension to s=0s=0, and

ML(0)=0,ML′(0)=4L3.(25)M_L(0)=0,\qquad M_L'(0)=\frac{4L}{3}. \tag*{(25)}

The function (μL,h−1)/h(\mu_{L,h}-1)/h itself extends smoothly to h=0h=0, where its value is MLM_L.

Proof. Put x=t/hx=t/h and write u(hx)=sy(x)u(hx)=sy(x) and v(hx)=Lg(s)z(x)v(hx)=Lg(s)z(x). For s>0s>0, the normalized equations are

yx=−hLg(s)sz1+[Lg(s)z]2,zx=a^μ(x)g(sy)g(s),y(0)=1,z(0)=0.(26)y_x=-h\frac{Lg(s)}{s}\frac{z}{\sqrt{1+[Lg(s)z]^2}},\qquad z_x=\widehat a_\mu(x)\frac{g(sy)}{g(s)},\qquad y(0)=1,\quad z(0)=0. \tag*{(26)}

where a^μ=1\widehat a_\mu=1 on (0,1)(0,1) and a^μ=−μ\widehat a_\mu=-\mu on (1,2)(1,2). Let E(r)=(er−1)/rE(r)=(e^r-1)/r for r≠0r\ne0, with the smooth positive extension E(0)=1E(0)=1. The identities

g(s)s=sE(s2),g(sy)g(s)=y2E(s2y2)E(s2)(27)\frac{g(s)}{s}=sE(s^2),\qquad\frac{g(sy)}{g(s)}=y^2\frac{E(s^2y^2)}{E(s^2)} \tag*{(27)}

remove the apparent singularities at s=0s=0. Thus the two vector fields in (26) are smooth in (s,L,h,μ,y,z)(s,L,h,\mu,y,z) near the compact parameter set under consideration.

At h=0h=0 the solution is

y≡1,z(x)={x,0≤x≤1,1−μ(x−1),1≤x≤2.(28)y\equiv1,\qquad z(x)= \begin{cases} x,&0\le x\le1,\\ 1-\mu(x-1),&1\le x\le2. \end{cases} \tag*{(28)}

Smooth dependence [16 Theorem 2.11], applied successively on the two half-intervals, provides a common neighborhood of [0,S]×J×{0}×{1}[0,S]\times J\times\{0\}\times\{1\} in which the flow exists up to x=2x=2 and depends smoothly on all parameters. For its terminal value F(s,L,h,μ):=z(2;s,L,h,μ)F(s,L,h,\mu):=z(2;s,L,h,\mu), we have

F(s,L,0,μ)=1−μ,∂μF(s,L,0,1)=−1.(29)F(s,L,0,\mu)=1-\mu,\qquad\partial_\mu F(s,L,0,1)=-1. \tag*{(29)}

The implicit function theorem and compactness therefore yield a single smooth function μ=Γ(s,L,h)\mu=\Gamma(s,L,h), defined for s∈[0,S]s\in[0,S], L∈JL\in J, and ∣h∣|h| small, with Γ(s,L,0)=1\Gamma(s,L,0)=1. At s=0s=0, equations (26)–(27) give y≡1y\equiv1 for every such hh, and hence Γ(0,L,h)=1\Gamma(0,L,h)=1. For s>0s>0 and small positive hh, the normalized solution has y>0y>0. After also imposing (2.5), Proposition 2.3 identifies Γ(s,L,h)\Gamma(s,L,h) with μL,h(s)\mu_{L,h}(s).

We compute the derivative with respect to hh at zero. Fix s>0s>0 and put k=Lg(s)k=Lg(s). Along the implicit branch, let Uh(x)=u(hx)U_h(x)=u(hx) and Vh(x)=v(hx)V_h(x)=v(hx), where the subscript denotes the parameter, not differentiation. At h=0h=0,

U0(x)=s,V0(x)={kx,0≤x≤1,k(2−x),1≤x≤2.U_0(x)=s,\qquad V_0(x)= \begin{cases} kx, & 0\le x\le1,\\ k(2-x), & 1\le x\le2. \end{cases}

Since (Uh)x=−hψ(Vh)(U_h)_x=-h\psi(V_h), its parameter derivative satisfies

U˙(x):=∂hUh(x)∣h=0=−∫0xψ(V0(r)) dr.\dot{U}(x):=\left.\partial_h U_h(x)\right|_{h=0}=-\int_0^x\psi(V_0(r))\,\mathrm{d}r.

Rescaling (2.4) to [0,2][0,2] and differentiating gives

∂hΓ(s,L,0)=ℓ(s)(∫01U˙(x) dx−∫12U˙(x) dx)=2ℓ(s)∫01rψ(kr) dr=ML(s).\begin{aligned} \partial_h\Gamma(s,L,0)&=\ell(s)\left(\int_0^1\dot{U}(x)\,\mathrm{d}x-\int_1^2\dot{U}(x)\,\mathrm{d}x\right)\\ &=2\ell(s)\int_0^1r\psi(kr)\,\mathrm{d}r=M_L(s). \end{aligned}

The second equality follows by changing the order of integration on each half-interval.

The quotient W(k)/kW(k)/k extends smoothly to zero and W(k)=2k/3+O(k3)W(k)=2k/3+O(k^3). Writing ML(s)=Lg′(s)W(Lg(s))/(Lg(s))M_L(s)=Lg'(s)W(Lg(s))/(Lg(s)) proves smooth extendibility at s=0s=0 and (3.5). Taylor’s formula now gives

Γ(s,L,h)=1+hML(s)+h2R(s,L,h),(30)\Gamma(s,L,h)=1+hM_L(s)+h^2\mathcal{R}(s,L,h), \tag*{(30)}

where

R(s,L,h)=∫01(1−θ)∂h2Γ(s,L,θh) dθ\mathcal{R}(s,L,h)=\int_0^1(1-\theta)\partial_h^2\Gamma(s,L,\theta h)\,\mathrm{d}\theta

is smooth on a neighborhood of the compact set. Its first two ss-derivatives are uniformly bounded, proving (3.4). Equation (3.10) also proves the final assertion.

Two critical points of the reduced curve

We first record the one-variable identities underlying the shape of MLM_L. Set

Q(k)=kW′(k)W(k)(k>0),R(x)=(1+2x)e−x−12x(x>0).(31)\begin{aligned} Q(k)&=\frac{kW'(k)}{W(k)}\quad(k>0),\qquad R(x)=\frac{(1+2x)e^{-x}-1}{2x}\quad(x>0). \tag*{(31)} \end{aligned}

with Q(0)=1Q(0)=1 and R(0)=1/2R(0)=1/2.

Lemma 3.2. The function QQ is smooth at zero and strictly decreasing on (0,∞)(0,\infty). It satisfies

11+k2<Q(k)<1,lim⁡k→0Q(k)=1,lim⁡k→∞Q(k)=0,(32)\frac{1}{1+k^2}<Q(k)<1,\qquad\lim_{k\to0}Q(k)=1,\qquad\lim_{k\to\infty}Q(k)=0, \tag*{(32)}

and

kQ′(k)=(Q(k)+2)(11+k2−Q(k)).(33)kQ'(k)=(Q(k)+2)\left(\frac{1}{1+k^2}-Q(k)\right). \tag*{(33)}

In particular, there is a unique κ>0\kappa>0 such that Q(κ)=1/2Q(\kappa)=1/2, and Q′(κ)<0Q'(\kappa)<0.

Proof. Differentiating (22) yields

Q(k)=∫01r2(1+k2r2)−3/2 dr∫01r2(1+k2r2)−1/2 dr.Q(k)=\frac{\int_0^1 r^2(1+k^2r^2)^{-3/2}\,\mathrm{d}r}{\int_0^1 r^2(1+k^2r^2)^{-1/2}\,\mathrm{d}r}.

This is a weighted average of (1+k2r2)−1(1+k^2r^2)^{-1}, which proves the strict inequalities in (32). The formula for WW also gives

kW′(k)+2W(k)=2k1+k2.(34)kW'(k)+2W(k)=\frac{2k}{\sqrt{1+k^2}}. \tag*{(34)}

Taking a logarithmic derivative of Q(k)+2=2k/(1+k2W(k))Q(k)+2=2k/(\sqrt{1+k^2}W(k)) proves (33); its right-hand side is negative. The expansion of WW at zero proves smoothness and Q(0)=1Q(0)=1. Finally, dominated convergence in (22) gives W(k)→1W(k)\to1 as k→∞k\to\infty. Equation (34) then implies Q(k)→0Q(k)\to0.

The equation

ex0=1+2x0(35)e^{x_0}=1+2x_0 \tag*{(35)}

has exactly one positive root, and 1<x0<21<x_0<2. Indeed, ex−1−2xe^x-1-2x is strictly convex, vanishes at zero, has negative derivative there, and changes sign between 1 and 2. Consequently,

R(x)>0(0≤x<x0),R(x)≤0(x≥x0),R′(x0)<0.(36)R(x)>0 \quad(0\le x<x_0),\qquad R(x)\le0 \quad(x\ge x_0),\qquad R'(x_0)<0. \tag*{(36)}

A direct differentiation of ℓ\ell gives

−ℓ′(s)ℓ(s)2=R(s2).(37)-\frac{\ell'(s)}{\ell(s)^2}=R(s^2). \tag*{(37)}

Combining this identity with (23) yields

ML′(s)=ℓ(s)2W(Lg(s))[Q(L(es2−1))−R(s2)].(38)M'_L(s)=\ell(s)^2W(Lg(s))\left[Q\left(L(e^{s^2}-1)\right)-R(s^2)\right]. \tag*{(38)}

Proposition 3.3. There exists L0>0L_0>0 such that, for every L>L0L>L_0, the function MLM_L has exactly two critical points

0<αL<βL<x0<2.0<\alpha_L<\beta_L<\sqrt{x_0}<2.

They are nondegenerate, with

ML′(s){>0,0<s<αL,<0,αL<s<βL,>0,s>βL,ML′′(αL)<0<ML′′(βL).(39)M'_L(s) \begin{cases} >0, & 0<s<\alpha_L,\\ <0, & \alpha_L<s<\beta_L,\\ >0, & s>\beta_L, \end{cases} \qquad M''_L(\alpha_L)<0<M''_L(\beta_L). \tag*{(39)}

The functions L↦αLL\mapsto\alpha_L and L↦βLL\mapsto\beta_L are smooth on (L0,∞)(L_0,\infty).

Proof. By (38), the critical points correspond to the positive zeros of

DL(x)=Q(L(ex−1))−R(x),x=s2.(40)D_L(x)=Q\left(L(e^x-1)\right)-R(x),\qquad x=s^2. \tag*{(40)}

There are no zeros for x≥x0x \ge x_0, by (36).

Choose ε>0\varepsilon> 0 so small that x0−ε>0x_0-\varepsilon> 0 and

R′(x)≤−c1<0for x∈[x0−ε,x0].(41)R'(x) \le-c_1 < 0 \quad\text{for } x \in[x_0-\varepsilon,x_0]. \tag*{(41)}

Next choose 0<δ<x0−ε0 < \delta< x_0-\varepsilon, and set c0=min⁡[0,δ]R>0c_0=\min_{[0,\delta]} R>0. By Lemma 3.2, we can choose a fixed K>κK>\kappa with Q(K)<c0Q(K)<c_0.

For the first zero, use the coordinate k=L(ex−1)k=L(e^x-1) and consider

HL(k)=Q(k)−R(log⁡(1+k/L)),0≤k≤K.(42)H_L(k)=Q(k)-R\left(\log(1+k/L)\right), \qquad0\le k\le K. \tag*{(42)}

As L→∞L\to\infty,

HL⟶Q−12in C1([0,K]).H_L\longrightarrow Q-\frac{1}{2}\quad\text{in }C^1([0,K]).

The limit has exactly one zero, κ\kappa, and its derivative there is negative. To see that the same exact count holds for HLH_L, choose a closed interval around κ\kappa on which Q′Q' is bounded above by a negative constant. Outside this interval, ∣Q−1/2∣\lvert Q-1/2\rvert has a positive lower bound. The C1C^1 convergence gives one strictly decreasing zero inside the interval and no zeros outside. For sufficiently large LL, the whole range 0≤k≤K0\le k\le K corresponds to 0≤x<δ0\le x<\delta. If x≤δx\le\delta but k=L(ex−1)≥Kk=L(e^x-1)\ge K, then

DL(x)≤Q(K)−c0<0.(43)D_L(x)\le Q(K)-c_0<0. \tag*{(43)}

Thus there is exactly one zero in [0,δ][0,\delta], and its sign change is from positive to negative.

On [δ,x0−ε][\delta,x_0-\varepsilon], the function RR has a positive minimum, while Q(L(ex−1))→0Q(L(e^x-1))\to0 uniformly. Hence DL<0D_L<0 throughout this interval for large $L.

On [x0−ε,x0][x_0-\varepsilon,x_0], differentiation and (33) give, with k=L(ex−1)k=L(e^x-1),

ddxQ(L(ex−1))=exex−1(Q(k)+2)(11+k2−Q(k)).(44)\frac{\mathrm{d}}{\mathrm{d}x}Q(L(e^x-1)) =\frac{e^x}{e^x-1}(Q(k)+2)\left(\frac{1}{1+k^2}-Q(k)\right). \tag*{(44)}

The right-hand side tends uniformly to zero. Together with (41), this implies DL′(x)≥c1/2>0D_L'(x)\ge c_1/2>0 for all sufficiently large LL. We already know that DL(x0−ε)<0D_L(x_0-\varepsilon)<0, whereas DL(x0)=Q(L(ex0−1))>0D_L(x_0)=Q(L(e^{x_0}-1))>0. There is therefore exactly one zero in this last interval, with sign change from negative to positive.

These intervals cover [0,x0][0,x_0], and both zeros are simple. Since the prefactor in (38) is positive for s>0s>0, the conclusions in (39) follow. Smoothness of the two critical points is a consequence of the implicit function theorem and their uniqueness.

The two folds of the shooting curve

Strict monotonicity at large amplitudes

The preceding expansion controls only bounded intervals of ss. We next establish a uniform estimate on the remaining half-line.

Proposition 3.4. Let J=[L−,L+]⊂(0,∞)J=[L_-,L_+]\subset(0,\infty). There exists hJ>0h_J>0 such that, for L∈JL\in J and 0<h<hJ0<h<h_J, the global shooting curve is defined and

μL,h′(s)>0for every s≥2.(45)\mu'_{L,h}(s)>0 \qquad\text{for every }s\ge2. \tag*{(45)}

Proof. Choose hJ≤1/4h_J\le1/4 small enough that (2.5) holds for every L∈JL\in J. Fix s≥2s\ge2 and write u(t;s),v(t;s)u(t;s),v(t;s) for the positive solution on the shooting curve. Since ∣u′∣<1|u'|<1,

u(t;s)>s−2h≥32(0≤t≤2h).(46)u(t;s)>s-2h\ge\frac{3}{2}\qquad(0\le t\le2h). \tag*{(46)}

The logarithmic derivative satisfies

ℓ′(u)=2[1−(1+2u2)e−u2](1−e−u2)2>0for u≥32,(47)\ell'(u)=\frac{2[1-(1+2u^2)e^{-u^2}]}{(1-e^{-u^2})^2}>0\qquad\text{for }u\ge\frac{3}{2}, \tag*{(47)}

because u2>x0u^2>x_0 in this range.

We differentiate along the shooting curve, keeping L,hL,h fixed:

A(t)=ddsu(t;s),B(t)=ddsv(t;s).(48)A(t)=\frac{d}{ds}u(t;s),\qquad B(t)=\frac{d}{ds}v(t;s). \tag*{(48)}

On the first half-interval these are also the fixed-μ\mu derivatives, since the left equation does not involve μ\mu. They satisfy

A′=−ψ′(v)B,B′=λg′(u)A,A(0)=1,B(0)=0.(49)A'=-\psi'(v)B,\qquad B'=\lambda g'(u)A,\qquad A(0)=1,\qquad B(0)=0. \tag*{(49)}

Before a first possible zero of AA, we have 0<A≤10<A\le1 and B>0B>0. Moreover, u≤su\le s and (47) give B′=v′ℓ(u)A≤ℓ(s)v′B'=v'\ell(u)A\le\ell(s)v', hence B(t)≤ℓ(s)v(t)B(t)\le\ell(s)v(t). On this half-interval v′≥λg(s−h)v'\ge\lambda g(s-h), so

1−A(t)=∫0tψ′(v(τ))B(τ) dτ≤ℓ(s)λg(s−h)∫0v(t)r(1+r2)3/2 dr<ℓ(s)λg(s−h).(50)\begin{aligned} 1-A(t)&=\int_0^t\psi'(v(\tau))B(\tau)\,d\tau\\ &\le\frac{\ell(s)}{\lambda g(s-h)}\int_0^{v(t)}\frac{r}{(1+r^2)^{3/2}}\,dr <\frac{\ell(s)}{\lambda g(s-h)}. \tag*{(50)} \end{aligned}

The constant

C∞:=sup⁡s≥2ℓ(s)g(s−1/4)(51)C_\infty:=\sup_{s\ge2}\frac{\ell(s)}{g(s-1/4)} \tag*{(51)}

is finite: the ratio is continuous and tends to zero at infinity. Since λ=L/h\lambda=L/h and h≤1/4h\le1/4, we have ℓ(s)/(λg(s−h))≤hC∞/L−\ell(s)/(\lambda g(s-h))\le hC_\infty/L_-. Decrease hJh_J so that hC∞/L−<1/2hC_\infty/L_-<1/2. Estimate (50) then excludes a first zero of AA and gives

A(t)>12(0≤t≤h),B(t)>0(0<t≤h).(52)A(t)>\frac{1}{2}\qquad(0\le t\le h),\qquad B(t)>0\qquad(0<t\le h). \tag*{(52)}

Set

C(t)=ℓ(u(t))A(t),Ch=C(h)>0.(53)C(t)=\ell(u(t))A(t),\qquad C_h=C(h)>0. \tag*{(53)}

By (49) and (47),

C′=−ℓ′(u)ψ(v)A−ℓ(u)ψ′(v)B<0(0<t≤h).C'=-\ell'(u)\psi(v)A-\ell(u)\psi'(v)B<0 \qquad(0<t\le h).

Since B′=v′CB'=v'C on the first half-interval,

B(h)=∫0hC(t)v′(t) dt>Chv(h).(54)B(h)=\int_0^h C(t)v'(t)\,dt>C_hv(h). \tag*{(54)}

Suppose, by contradiction, that μL,h′(s)≤0\mu'_{L,h}(s)\le0, and put m=μL,h′(s)/μL,h(s)≤0m=\mu'_{L,h}(s)/\mu_{L,h}(s)\le0. On the second half-interval the total derivatives in (48) satisfy

A′=−ψ′(v)B,B′=v′(m+C),v′<0.(55)A'=-\psi'(v)B,\qquad B'=v'(m+C),\qquad v'<0. \tag*{(55)}

On any interval starting at hh on which B>0B>0, the function AA is decreasing. If A(t)≥0A(t)\ge0, then u(t)≤u(h)u(t)\le u(h) and (47) give C(t)=ℓ(u(t))A(t)≤ℓ(u(h))A(h)=ChC(t)=\ell(u(t))A(t)\le\ell(u(h))A(h)=C_h. If A(t)<0A(t)<0, the same inequality follows from C(t)<0<ChC(t)<0<C_h. Thus m+C(t)≤Chm+C(t)\le C_h as long as BB remains positive. Because v′<0v'<0, (55) gives B′(t)≥Chv′(t)B'(t)\ge C_hv'(t). Integrating yields

B(t)≥B(h)+Ch(v(t)−v(h))=[B(h)−Chv(h)]+Chv(t)>0.(56)B(t)\ge B(h)+C_h\bigl(v(t)-v(h)\bigr)=\bigl[B(h)-C_hv(h)\bigr]+C_hv(t)>0. \tag*{(56)}

Here (54) and v(t)≥0v(t)\ge0 are used. By continuity, this estimate excludes a first zero of BB on [h,2h][h,2h]. At t=2ht=2h, it gives B(2h)>0B(2h)>0. But the identity v(2h;s,μL,h(s))=0v(2h;s,\mu_{L,h}(s))=0 implies B(2h)=0B(2h)=0. The contradiction proves (45). □\square

Persistence of the two critical points

Lemma 3.5. Let L0L_0 be as in Proposition 3.3, and let J⋐(L0,∞)J\Subset(L_0,\infty) be a compact interval. For all sufficiently small h>0h>0, uniformly for L∈JL\in J, the function μL,h\mu_{L,h} has exactly two critical points on (0,∞)(0,\infty), denoted by

0<αL,h<βL,h<2.0<\alpha_{L,h}<\beta_{L,h}<2.

They are nondegenerate, and

μL,h′(s){>0,0<s<αL,h,<0,αL,h<s<βL,h,>0,s>βL,h.(57)\mu'_{L,h}(s)\begin{cases} >0, & 0<s<\alpha_{L,h},\\ <0, & \alpha_{L,h}<s<\beta_{L,h},\\ >0, & s>\beta_{L,h}. \end{cases} \tag*{(57)}

Furthermore,

αL,h=αL+O(h),βL,h=βL+O(h),(58)\alpha_{L,h}=\alpha_L+O(h),\qquad\beta_{L,h}=\beta_L+O(h), \tag*{(58)}

uniformly for L∈JL\in J.

Proof. Define

H(s,L,h)={(μL,h(s)−1)/h,h≠0,ML(s),h=0,\mathcal{H}(s,L,h)= \begin{cases} (\mu_{L,h}(s)-1)/h, & h\ne0,\\ M_L(s), & h=0, \end{cases}

using the local extension from Proposition 3.1 when h<0h<0. This function is smooth on a neighborhood of [0,2]×J×{0}[0,2]\times J\times\{0\}, and

sup⁡L∈J∥H(⋅,L,h)−ML∥C2([0,2])=O(h).(59)\sup_{L\in J}\|\mathcal{H}(\cdot,L,h)-M_L\|_{C^2([0,2])}=O(h). \tag*{(59)}

The functions αL,βL\alpha_L,\beta_L are continuous on JJ, are separated from each other and from the endpoints 0,20,2, and satisfy the strict second-derivative inequalities in (3.19). Compactness therefore gives η,c>0\eta,c>0 such that the intervals

Iα(L)=[αL−η,αL+η],Iβ(L)=[βL−η,βL+η]I_\alpha(L)=[\alpha_L-\eta,\alpha_L+\eta],\qquad I_\beta(L)=[\beta_L-\eta,\beta_L+\eta]

are disjoint and contained in (0,2)(0,2), and

ML′′≤−c on Iα(L),ML′′≥c on Iβ(L),M_L''\le-c\text{ on }I_\alpha(L),\qquad M_L''\ge c\text{ on }I_\beta(L),

for every L∈JL\in J. On the compact complement of the interiors of these two moving intervals, ∣ML′∣|M_L'| has a positive lower bound. By (3.39), the derivative Hs\mathcal{H}_s has the same sign as ML′M_L' on this complement, is strictly decreasing on Iα(L)I_\alpha(L), and is strictly increasing on Iβ(L)I_\beta(L), once hh is small. The endpoint signs on each interval are opposite. Thus Hs\mathcal{H}_s has exactly one zero in each interval and no other zeros in [0,2][0,2]. Both zeros are simple.

Proposition 3.4 excludes every further critical point for s≥2s\ge2, after decreasing the upper bound for hh if necessary. This proves (3.37) and nondegeneracy. The implicit function theorem applied to Hs=0\mathcal{H}_s=0 gives smooth dependence of the critical points on (L,h)(L,h) up to h=0h=0. Their hh-derivatives are uniformly bounded on a smaller compact parameter neighborhood, which proves (3.38). □\square

Exact multiplicity and nondegeneracy

Proof of Theorem 1.1. Choose L∗>L0L_\ast>L_0 and apply Lemma 3.5 with J=[L∗,2L∗]J=[L_\ast,2L_\ast]. Take h∗>0h_\ast>0 small enough for that lemma, Propositions 3.1 and 3.4, and condition (2.5) to hold uniformly on JJ. For (λ,h)∈D(\lambda,h)\in D, set L=λhL=\lambda h and define

μ+(λ,h)=μL,h(αL,h),μ−(λ,h)=μL,h(βL,h).(60)\mu_+(\lambda,h)=\mu_{L,h}(\alpha_{L,h}),\qquad\mu_-(\lambda,h)=\mu_{L,h}(\beta_{L,h}). \tag*{(60)}

The first point is a strict local maximum and the second a strict local minimum. By (2.4) and (3.37),

1<μ−(λ,h)<μ+(λ,h).1<\mu_-(\lambda,h)<\mu_+(\lambda,h).

Proposition 3.1 and Lemma 2.4 give

lim⁡s↓0μL,h(s)=1,lim⁡s→∞μL,h(s)=∞.(61)\lim_{s\downarrow0}\mu_{L,h}(s)=1,\qquad\lim_{s\to\infty}\mu_{L,h}(s)=\infty. \tag*{(61)}

The three strictly monotone pieces of the shooting curve consequently have ranges

(1,μ+),(μ−,μ+),(μ−,∞),(1,\mu_+),\qquad(\mu_-,\mu_+),\qquad(\mu_-,\infty),

when their turning-point endpoints are omitted. Counting intersections with a horizontal line, and counting a turning point only once, yields (4). Proposition 2.3 ensures that these intersections account for all positive solutions. The same description proves uniqueness of the pair (μ−,μ+)(\mu_-,\mu_+). Smoothness of the critical values follows from the smooth critical-point functions and the change of variables L=λhL=\lambda h.

It remains to relate the turning points to the linearized boundary value problem. For fixed (λ,h)(\lambda,h), use the shooting function FF from (11). On the positive solution curve,

Fs+FμμL,h′=0,Fμ<0.(62)F_s+F_\mu\mu'_{L,h}=0,\qquad F_\mu<0. \tag*{(62)}

The linearized equation at a positive solution uu is

(ϕ′(u)w′)′+λaμ,h(t)g′(u)w=0,w′(0)=w′(2h)=0.(63)(\phi'(u)w')'+\lambda a_{\mu,h}(t)g'(u)w=0,\qquad w'(0)=w'(2h)=0. \tag*{(63)}

It is understood with continuous ww and ϕ′(u)w′\phi'(u)w' across hh. Since ϕ′(u)\phi'(u) is continuous and strictly positive, this is equivalent to continuity of w,w′w,w'. Every solution of the linearized equation satisfying the left Neumann condition is a scalar multiple of the fixed-μ\mu initial-value variation w=∂suw=\partial_su with w(0)=1w(0)=1. For this variation,

∂sv=−ϕ′(u)w′,Fs=∂sv(2h)=−w′(2h),\partial_sv=-\phi'(u)w',\qquad F_s=\partial_sv(2h)=-w'(2h),

since u′(2h)=0u'(2h)=0 and ϕ′(0)=1\phi'(0)=1. Thus the linearized Neumann problem has a nonzero solution if and only if Fs=0F_s=0, or equivalently μL,h′=0\mu'_{L,h}=0. At either critical point its kernel is one-dimensional; elsewhere it is trivial.

Let s∗s_* be one of the two critical initial values and let μ∗=μL,h(s∗)\mu_*=\mu_{L,h}(s_*). Differentiating the shooting identity twice at s∗s_* gives

Fss(s∗,μ∗)=−Fμ(s∗,μ∗)μL,h′′(s∗)≠0.(64)F_{ss}(s_*,\mu_*)=-F_\mu(s_*,\mu_*)\mu''_{L,h}(s_*)\ne0. \tag*{(64)}

Equivalently, the local solution curve has the form

μ−μ∗=(s−s∗)2J∗(s),J∗(s)=∫01(1−θ)μL,h′′(s∗+θ(s−s∗)) dθ,\mu-\mu_*=(s-s_*)^2J_*(s),\qquad J_*(s)=\int_0^1(1-\theta)\mu''_{L,h}\bigl(s_*+\theta(s-s_*)\bigr)\,d\theta,

where J∗(s∗)=μL,h′′(s∗)/2≠0J_*(s_*)=\mu''_{L,h}(s_*)/2\ne0. The smooth local coordinate ξ=(s−s∗)∣J∗(s)∣\xi=(s-s_*)\sqrt{\lvert J_*(s)\rvert} therefore puts the projection into the normal form

μ−μ∗=sgn⁡(μL,h′′(s∗))ξ2.(65)\mu-\mu_*=\operatorname{sgn}\bigl(\mu''_{L,h}(s_*)\bigr)\xi^2. \tag*{(65)}

This is a nondegenerate fold. The parameterization by ss is regular as a curve of functions because its derivative has value 11 at t=0t=0. At each critical value, the other solution lies on a strictly monotone part of the curve and is nondegenerate.

Critical-value asymptotics and numerical illustration

The reduced function also gives the first-order locations of the two critical parameter values.

Corollary 4.1. Let J⋐(L0,∞)J \Subset(L_0,\infty) be a compact interval. For L∈JL \in J and sufficiently small h>0h>0, define the critical values by (3.40), using Lemma 3.5. Uniformly for L∈JL \in J, as h↓0h \downarrow0,

μ+(L/h,h)=1+hML(αL)+O(h2),μ−(L/h,h)=1+hML(βL)+O(h2).(66)\begin{aligned} \mu_+(L/h,h) &= 1+hM_L(\alpha_L)+O(h^2),\\ \mu_-(L/h,h) &= 1+hM_L(\beta_L)+O(h^2). \tag*{(66)} \end{aligned}

In particular,

μ+(L/h,h)−μ−(L/h,h)=h[ML(αL)−ML(βL)]+O(h2).(67)\mu_+(L/h,h)-\mu_-(L/h,h) =h\left[M_L(\alpha_L)-M_L(\beta_L)\right]+O(h^2). \tag*{(67)}

Proof. Substitute (3.38) into (3.4). The functions MLM_L and their first derivatives are uniformly bounded on [0,2]×J[0,2]\times J. This gives (4.1), and subtraction gives (4.2).

Proposition 4.2. Let κ\kappa and x0x_0 be defined by Q(κ)=1/2Q(\kappa)=1/2 and (3.15), respectively. The reduced critical points satisfy

L αL⟶κ,βL⟶x0(L→∞).(68)\sqrt{L}\,\alpha_L \longrightarrow\sqrt{\kappa},\qquad \beta_L \longrightarrow\sqrt{x_0}\qquad(L\to\infty). \tag*{(68)}

Their critical values satisfy

ML(αL)L⟶2W(κ)κ,ML(βL)⟶1+2x0x0.(69)\frac{M_L(\alpha_L)}{\sqrt{L}}\longrightarrow\frac{2W(\kappa)}{\sqrt{\kappa}}, \qquad M_L(\beta_L)\longrightarrow\frac{1+2x_0}{\sqrt{x_0}}. \tag*{(69)}

Proof. In the first part of the proof of Proposition 3.3, the zero kL=L(eαL2−1)k_L=L(e^{\alpha_L^2}-1) converges to κ\kappa. Thus LαL2=Llog⁡(1+kL/L)→κL\alpha_L^2=L\log(1+k_L/L)\to\kappa. For the second zero, βL2\beta_L^2 belongs to a fixed neighborhood of x0x_0 and satisfies R(βL2)=Q(L(eβL2−1))→0R(\beta_L^2)=Q(L(e^{\beta_L^2}-1))\to0. Since x0x_0 is the only zero of RR in that neighborhood, βL→x0\beta_L\to\sqrt{x_0}. Now ℓ(s)∼2/s\ell(s)\sim2/s as s↓0s\downarrow0, so

ML(αL)L=ℓ(αL)LW(kL)⟶2W(κ)κ.\frac{M_L(\alpha_L)}{\sqrt{L}} =\frac{\ell(\alpha_L)}{\sqrt{L}}W(k_L) \longrightarrow\frac{2W(\kappa)}{\sqrt{\kappa}}.

For the second value, W(Lg(βL))→1W(Lg(\beta_L))\to1, and

ℓ(x0)=2x01−e−x0=1+2x0x0.\ell(\sqrt{x_0})=\frac{2\sqrt{x_0}}{1-e^{-x_0}}=\frac{1+2x_0}{\sqrt{x_0}}.

This proves (4.4).

Remark 4.3. The two limiting procedures above are distinct. Corollary 4.1 is uniform for LL in a fixed compact interval, whereas Proposition 4.2 concerns the reduced function as L→∞L\to\infty. A joint limit in which LL is unbounded and h→0h\to0 would require additional uniform estimates.

For numerical illustration, Figure 1 shows a portion of the shooting curve for (λ,h)=(400,0.05)(\lambda, h) = (400, 0.05).

Figure 1. Numerical shooting curve for (1.1) with (λ,h)=(400,0.05)(\lambda, h) = (400, 0.05). The markers μ−\mu_{-} and μ+\mu_{+} indicate the local minimum and maximum, respectively, of μ\mu as a function of u(0)u(0).

For each sampled s=u(0)>0s = u(0) > 0, we solve F(s,μ)=0F(s, \mu) = 0 numerically for μ\mu using the shooting function (11). OpenAI Codex assisted in writing the code for this computation and the plot. The curve was computed with SciPy and plotted with Matplotlib. The two turning points of the computed curve lie at μ−≈1.16942\mu_{-} \approx1.16942 and μ+≈1.31644\mu_{+} \approx1.31644.

Declaration of generative AI use

The author used OpenAI Codex to assist with the organization and English-language editing of the manuscript, the review of references and mathematical derivations, and the preparation of the numerical shooting and plotting code for Figure 1. The author reviewed and revised the material prepared with this assistance and takes full responsibility for the content of the article.

References

  1. [1]DEG1, DEG1 challenges, entry on uniqueness for a super-exponential Minkowski-curvature equation. https://deg1.uniud.it/challenges.html.
  2. [2]A. Boscaggin, G. Feltrin, and F. Zanol in, Positive solutions for a Minkowski-curvature equation with indefinite weight and super-exponential nonlinearity, Commun. Contemp. Math. 25 (2023), no. 4, Paper No. 2250005, 20 pp. doi:10.1142/S0219199722500055.DOI
  3. [3]A. Boscaggin, G. Feltrin, and F. Zanol in, Uniqueness of positive solutions for boundary value problems associated with indefinite φ-Laplacian-type equations, Open Math. 19 (2021), no. 1, 163–183. doi:10.1515/math-2021-0003.
  4. [4]A. Boscaggin and G. Feltrin, Pairs of positive radial solutions for a Minkowski-curvature Neumann problem with indefinite weight, Nonlinear Anal. 196 (2020), Paper No. 111807, 14 pp. doi:10.1016/j.na.2020.111807.DOI
  5. [5]A. Boscaggin and G. Feltrin, Positive periodic solutions to an indefinite Minkowski-curvature equation, J. Differential Equations 269 (2020), no. 7, 5595–5645. doi:10.1016/j.jde.2020.04.009.DOI
  6. [6]A. Boscaggin, F. Colasuonno, and B. Noris, Positive radial solutions for the Minkowski-curvature equation with Neumann boundary conditions, Discrete Contin. Dyn. Syst. Ser. S 13 (2020), no. 7, 1921–1933. doi:10.3934/dcdss.2020150.DOI
  7. [7]R. Ma, X. Su, and Z. Zhao, Global structure of positive solutions for a Neumann problem with indefinite weight in Minkowski space, J. Fixed Point Theory Appl. 25 (2023), no. 2, Paper No. 43. doi:10.1007/s11784-023-01047-x.DOI
  8. [8]X. Su, R. Ma, and W. Yang, S-shaped connected components of positive solutions for Minkowski-curvature Neumann problems with indefinite weight, J. Math. Phys. 66 (2025), no. 7, Paper No. 071510. doi:10.1063/5.0239395.DOI
  9. [9]G. Feltrin and E. Sovrano, Three positive solutions to an indefinite Neumann problem: a shooting method, Nonlinear Anal. 166 (2018), 87–101. doi:10.1016/j.na.2017.10.006.DOI
  10. [10]G. Feltrin, E. Sovrano, and A. Tellini, On the number of positive solutions to an indefinite parameter-dependent Neumann problem, Discrete Contin. Dyn. Syst. 42 (2022), no. 1, 21–71. doi:10.3934/dcds.2021107.
  11. [11]S.-Y. Huang, Exact multiplicity and bifurcation curves of positive solutions of a one-dimensional Minkowski-curvature problem and its application, Commun. Pure Appl. Anal. 17 (2018), no. 3, 1271–1294. doi:10.3934/cpaa.2018061.DOI
  12. [12]S.-Y. Huang, Global bifurcation and exact multiplicity of positive solutions for the one-dimensional Minkowski-curvature problem with sign-changing nonlinearity, Commun. Pure Appl. Anal. 18 (2019), no. 6, 3267–3284. doi:10.3934/cpaa.2019147.DOI
  13. [13]S.-Y. Huang and S.-H. Wang, Bifurcation curves for the one-dimensional perturbed Gelfand problem with the Minkowski-curvature operator, J. Differential Equations 416 (2025), 700–726. doi:10.1016/j.jde.2024.10.002.DOI
  14. [14]R. Ma and M. Xu, S-shaped connected component for a nonlinear Dirichlet problem involving mean curvature operator in one-dimension Minkowski space, Bull. Korean Math. Soc. 55 (2018), no. 6, 1891–1908. KCI record.
  15. [15]Y.-H. Lee, I. Sim, and R. Yang, Repeated S-shaped and Σ-shaped bifurcation curves for the one-dimensional non-autonomous Minkowski-curvature problem, J. Math. Anal. Appl. 539 (2024), no. 2, Paper No. 128548. doi:10.1016/j.jmaa.2024.128548.DOI
  16. [16]G. Teschl, Ordinary Differential Equations and Dynamical Systems, Graduate Studies in Mathematics, vol. 140, American Mathematical Society, Providence, RI, 2012. doi:10.1090/gsm/140.DOI

Paper details

vlnc:W_7zah5tdc
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.

  • Mathematics

Codex

View statement
Looking for related papers…
Contents