
We study upper and lower bounds for the pull-in voltage and the pull-in distance for the one-dimensional prescribed mean curvature problem arising in MEMS
{−(u′(x)√1+(u′(x))2)′=λ(1−u)p, u<1, −L<x<L,u(−L)=u(L)=0,
where λ>0 is a bifurcation parameter, and p,L>0 are two evolution parameters. We further study monotonicity properties and asymptotic behaviors for the pull-in voltage and pull-in distance with respect to positive parameters p and L.
Citation: Yan-Hsiou Cheng, Kuo-Chih Hung, Shin-Hwa Wang, Jhih-Jyun Zeng. Upper and lower bounds for the pull-in voltage and the pull-in distance for a generalized MEMS problem[J]. Mathematical Biosciences and Engineering, 2022, 19(7): 6814-6840. doi: 10.3934/mbe.2022321
[1] | Zhizhou Ma, Yujun Qi, Weiqing Liu . A comparative study of pull-out performance of bolted joints in pultruded FRP with drilled holes or punched holes. Mathematical Biosciences and Engineering, 2019, 16(5): 4213-4228. doi: 10.3934/mbe.2019210 |
[2] | Peng Jing, Weichao Wang, Chengxi Jiang, Ye Zha, Baixu Ming . Determinants of switching behavior to wear helmets when riding e-bikes, a two-step SEM-ANFIS approach. Mathematical Biosciences and Engineering, 2023, 20(5): 9135-9158. doi: 10.3934/mbe.2023401 |
[3] | Linxuan Zhou, Jingwei Gao, Qiao Li, Cheng Hu . Simulation study on tractive performance of off-road tire based on discrete element method. Mathematical Biosciences and Engineering, 2020, 17(4): 3869-3893. doi: 10.3934/mbe.2020215 |
[4] | Yuriy Stoyan, Georgiy Yaskov, Tatiana Romanova, Igor Litvinchev, Sergey Yakovlev, José Manuel Velarde Cantú . Optimized packing multidimensional hyperspheres: a unified approach. Mathematical Biosciences and Engineering, 2020, 17(6): 6601-6630. doi: 10.3934/mbe.2020344 |
[5] | Guirong Jiang, Qishao Lu, Linping Peng . Impulsive Ecological Control Of A Stage-Structured Pest Management System. Mathematical Biosciences and Engineering, 2005, 2(2): 329-344. doi: 10.3934/mbe.2005.2.329 |
[6] | Jiaxin Nan, Wanbiao Ma . Stability and persistence analysis of a microorganism flocculation model with infinite delay. Mathematical Biosciences and Engineering, 2023, 20(6): 10815-10827. doi: 10.3934/mbe.2023480 |
[7] | Hongying Shu, Wanxiao Xu, Zenghui Hao . Global dynamics of an immunosuppressive infection model with stage structure. Mathematical Biosciences and Engineering, 2020, 17(3): 2082-2102. doi: 10.3934/mbe.2020111 |
[8] | Dongmei Zhang . Unveiling dynamics of urbanization, rural logistics, and carbon emissions: A study based on China's empirical data. Mathematical Biosciences and Engineering, 2024, 21(2): 2731-2752. doi: 10.3934/mbe.2024121 |
[9] | Hui Li, Mengyao Zhang, Chenbo Zeng . Circular Jaccard distance based multi-solution optimization for traveling salesman problems. Mathematical Biosciences and Engineering, 2022, 19(5): 4458-4480. doi: 10.3934/mbe.2022206 |
[10] | Guangxun Sun, Binxiang Dai . Stability and bifurcation of a delayed diffusive predator-prey system with food-limited and nonlinear harvesting. Mathematical Biosciences and Engineering, 2020, 17(4): 3520-3552. doi: 10.3934/mbe.2020199 |
We study upper and lower bounds for the pull-in voltage and the pull-in distance for the one-dimensional prescribed mean curvature problem arising in MEMS
{−(u′(x)√1+(u′(x))2)′=λ(1−u)p, u<1, −L<x<L,u(−L)=u(L)=0,
where λ>0 is a bifurcation parameter, and p,L>0 are two evolution parameters. We further study monotonicity properties and asymptotic behaviors for the pull-in voltage and pull-in distance with respect to positive parameters p and L.
This paper is a continuation of Cheng, Hung and Wang [1]. In this paper, we study some structures of bifurcation diagrams of positive solutions u∈C2(−L,L)∩C[−L,L] for the one-dimensional prescribed mean curvature problem arising in electrostatic MEMS (Micro-Electro-Mechanical Systems)
{−(u′(x)√1+(u′(x))2)′=λ(1−u)p, u<1, −L<x<L,u(−L)=u(L)=0, | (1.1) |
where λ>0 is a bifurcation parameter, and p,L>0 are two evolution parameters. In particular, we study upper and lower bounds, monotonicity properties and asymptotic behaviors for the pull-in voltage and the pull-in distance. The singular nonlinearity in (1.1)
f(u)≡1(1−u)p, p>0 |
satisfies
f(0)=1, limu→1−f(u)=∞, and f′(u),f′′(u)>0 on [0,1). | (1.2) |
Notice that the improper integral of f over [0,1) satisfies
∫10f(u)du={∞if p≥1,11−p<∞if 0<p<1. |
The one-dimensional prescribed mean curvature problem
{−(u′(x)√1+(u′(x))2)′=λ˜f(u), −L<x<L,u(−L)=u(L)=0, | (1.3) |
and n-dimensional problem of it, with general nonlinearity ˜f(u) or with many different types nonlinearities, like up (p>0), up+uq (0≤p<q<∞), (1+u)p (p>0), exp(u), exp(u)−1, exp(aua+u) (a>0), exp(aua+u)−1 (a>0), au (a>0), u−u3, and (1−u)−p (p>0) have been investigated intensively since 1990, see, e.g., [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15].
A solution u∈C2(−L,L)∩C[−L,L] of (1.3) with u′∈C([−L,L],[−∞,∞]) is called classical if |u′(±L)|<∞, and it is called non-classical if u′(−L)=∞ or u′(L)=−∞, see [8]. Notice that it can be shown that (see [2,8]), for (1.3),
(i) Any non-trivial positive solution u∈C2(−L,L)∩C[−L,L] is concave on (−L,L) if ˜f(u)>0 for u>0, since the equation in (1.3) can be written in the equivalent form
u′′(x)=−λ(1+u′2)3/2˜f(u)<0 on (−L,L). | (1.4) |
(ii) A positive solution u∈C2(−L,L)∩C[−L,L] must be symmetric on [−L,L]. Thus u′(−L)=−u′(+L).
In this paper for prescribed mean curvature problem (1.1), we simply consider classical positive solutions u. For any fixed p,L>0, we define the bifurcation diagram Cp,L of (1.1) by
Cp,L≡{(λ,‖ |
We say the bifurcation diagram C_{p, L} is \supset -shaped (see e.g., Figure 1(i) depicted below) on the (\lambda, \left \Vert u\right \Vert _{\infty }) -plane if there exists \lambda ^{\ast } > 0 such that C_{p, L} consists of a continuous curve with exactly one turning point at some point (\lambda ^{\ast }, \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }) where the bifurcation diagram C_{p, L} turns to the left.
Brubaker and Pelesko [3] studied existence and multiplicity of positive solutions of the n -dimensional prescribed mean curvature problem
\begin{equation} \left \{ \begin{array}{l} -\text{div}\dfrac{\nabla u(\mathbf{x})}{\sqrt{1+\left \vert \nabla u(\mathbf{x })\right \vert ^{2}}} = \dfrac{\lambda }{(1-u)^{2}}, \ \ u < 1, \ \ \mathbf{x}\in \Omega _{L}, \\ u = 0, \ \ \mathbf{x}\in \partial \Omega _{L}, \end{array} \right. \end{equation} | (1.5) |
where \lambda > 0 is a bifurcation parameter and \Omega _{L}\subset \mathbb{R}^{n} ( n\geq 1 ) is a smooth bounded domain depending on some parameter L > 0 . Problem (1.5) with an inverse square type nonlinearity f(u) = (1-u)^{-p}, p = 2 is a derived variant of a canonical model used in the modeling of electrostatic Micro-Electro Mechanical Systems (MEMS) device obeying the electrostatic Coulomb law with the Coulomb force satisfying the inverse square law with respect to the distance of the two charged objects, which is a function of the deformation variable (cf. [16,p. 1324].) The modeling of electrostatic MEMS device consists of a thin dielectric elastic membrane with boundary supported at 0 below a rigid plate located at +1 . In (1.5), u is the unknown profile of the deflecting MEMS membrane, \lambda is the drop voltage between the ground plate and the deflecting membrane, and the term \left \vert \nabla u\right \vert ^{2} is called a fringing field (cf. [3]). When a voltage \lambda is applied, the membrane deflects towards the ceiling plate and a snap-through may occur when it exceeds a certain critical value \lambda ^{\ast }, referred to as the "pull-in voltage". (So if voltage \lambda exceeds pull-in voltage \lambda ^{\ast } , an equilibrium defection is no longer attainable and the lower surface will touch up on the upper plate.) This creates a so-called "pull-in instability" which greatly affects the design and manufacture of MEMS devices. Also, in the actual design of a MEMS device, typically, one of the primary device design goals is to achieve the maximum possible stable steady-state deflection (that is, \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } ( < 1 ), cf. Theorems 1.1–1.2 and Figures 1 and 2 below), referred to as the "pull-in distance", with a relatively small applied voltage. We refer to [3,17] for detailed discussions on MEMS devices modeling. Notice that the physically relevant dimensions are n = 1 and n = 2 . In the case for n = 1 , \Omega _{L} is a rectangular strip with two opposite edges at x = \pm L fixed ( 2L is the length of the strip) and the remaining two edges free, the deflection u = u(x, y) may be assumed a function of x only. In the case for n = 2 , \Omega _{L} is a planar bounded domain with smooth boundary, and L is the characteristic length (diameter) of the domain. In particular, \Omega _{L} could be a circular disk of radius L .
With general p > 0 , (1.1) is a generalized MEMS problem under the assumption that the Coulomb force satisfies the inverse p -th power law with respect to the distance of the two charged objects, where p > 0 characterizes the force strength. See [18].
Brubaker and Pelesko [4] and Pan and Xing [11] studied global bifurcation diagrams and exact multiplicity of positive (classical) solutions for the one dimensional problem of (1.5),
\begin{equation} \left \{ \begin{array}{l} -\left( \dfrac{u^{\prime }(x)}{\sqrt{1+\left( {u^{\prime }(x)}\right) ^{2}}} \right) ^{\prime } = \dfrac{\lambda }{(1-u)^{2}}, \ \ u < 1, \ \ -L < x < L, \\ u(-L) = u(L) = 0. \end{array} \right. \end{equation} | (1.6) |
Brubaker and Pelesko [4,Theorem 1.1] and Pan and Xing [11,Theorem 1.1] independently proved that, for (1.6), which corresponds to the case p = 2 in (1.1) with L > 0 , there exists a positive number L^{\ast }\approx 0.34997 such that, on the (\lambda, \left \Vert u\right \Vert _{\infty }) -plane, the bifurcation diagram C_{2, L} consists of a (continuous) \supset -shaped curve which emanates from the origin and has exactly one (left) turning point at some point (\lambda ^{\ast }, \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }) when L\geq L^{\ast } , and as L transitions from greater than or equal to L^{\ast } to less than L^{\ast } the upper branch of the bifurcation diagram C_{2, L} splits into two parts. See Figure 1 depicted below and see [4,Theorem 1.1] and [11,Theorem 1.1] for details. In addition, for (1.6), Brubaker and Pelesko [4,Theorem 1.1] proved that the pull-in voltage \lambda ^{\ast } = \lambda ^{\ast }(L) satisfies
\begin{equation} \lambda ^{\ast } < \min \left \{ L^{-1}, \tfrac{\pi ^{2}}{27}L^{-2}\right \} , \end{equation} | (1.7) |
and Pan and Xing [11,Theorem 1.1,part (4)] proved that the pull-in voltage \lambda ^{\ast }(L) is a strictly decreasing function of L > 0. Brubaker and Pelesko [4,Figure 1.2(b)] also gave numerical simulation of the pull-in distance \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } for 0.1\leq L\leq 0.8, which show that \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } is a strictly increasing function of L\in \lbrack 0.1, 0.8] .
In following Theorems 1.1–1.2, Cheng, Hung and Wang [1,Theorems 2.1-2.3] extended and improved the results of Brubaker and Pelesko [4,Theorem 1.1] and Pan and Xing [11,Theorem 1.1] by generalizing the nonlinearity f(u) = (1-u)^{-2} in (1.6) to f(u) = (1-u)^{-p} with general p\in (0, \infty) . Theorems 1.1–1.2 show that p is a bifurcation parameter to prescribed mean curvature problem (1.1). Note that this result remains hold for the standard MEMS problem
\begin{equation} \left \{ \begin{array}{l} -u^{\prime \prime }(x) = \dfrac{\lambda }{(1-u)^{p}}, \ \ u < 1, \ \ -L < x < L, \\ u(-L) = u(L) = 0; \end{array} \right. \end{equation} | (1.8) |
see [19,Section 2]. The standard MEMS problem (1.8) and n -dimensional (generalized) problems of it have been studied by numerous authors, see e.g., [20,21,22]. For semilinear problem (1.8) with any p > 0 and L > 0 , by applying (1.2) and Laetsch [23,Theorems 2.5,2.9 and 3.2], we obtain that, on the (\lambda, \left \Vert u\right \Vert _{\infty }) -plane, the bifurcation diagram of positive solutions consists of a (continuous) \supset -shaped curve which emanates from the origin, initially continues to the right, and making a left turn at some point (\lambda ^{\ast }, \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }) which is a bifurcation point with neutral stability, then continues to the left, and ends at (0, 1) , cf. Figure 1(i). For semilinear problem (1.8) with p = 2 and L = 1, Ghoussoub and Guoand [22,Theorem 1.1,part 4] proved that the pull-in voltage \lambda ^{\ast }\leq 9/8 = 1.125. For semilinear problem (1.8) with p = 2 and L = 1, the value of the pull-in voltage was numerically determined as \lambda ^{\ast }\approx 1.400016469/4\approx 0.350004 in [21,Theorem 2.2] (see also [24,FIGURE 2(a)]), and the value of the pull-in distance was numerically determined as \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }\approx 0.38 in [24,FIGURE 2(a)]. Cowan and Ghoussoub [20,Corollary 2.1,part 1] proved that, for Laplacian problem (1.8) with p > 0 and L = 1, the pull-in distance
\begin{equation*} (1 > ) \ \ \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }\geq \frac{1 }{p+1}. \end{equation*} |
This implies that
\begin{equation} \lim\limits_{p\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = 1 \ \text{for (1.8) with }L = 1. \end{equation} | (1.9) |
Theorem 1.1 (See Figure 1). Consider classical positive solutions u of (1.1) with p\geq 1 . There exists L^{\ast } = L^{\ast }(p) > 0 such that the following assertions (i)–(iii) hold:
(i) (See Figure 1(i).) If L > L^{\ast } , then there exists \lambda ^{\ast } > 0 such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for 0 < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for \lambda = \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
(ii) (See Figure 1(ii).) If L = L^{\ast } , then there exist 0 < \bar{\lambda} ( = \bar{\lambda}(p)) < \lambda ^{\ast } such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for 0 < \lambda < \bar{\lambda} and \bar{\lambda} < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for \lambda = \bar{\lambda}, \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
(iii) (See Figure 1(iii).) If 0 < L < L^{\ast } , then there exist 0 < \hat{\lambda} < \check{\lambda} < \lambda ^{\ast } such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for 0 < \lambda < \hat{\lambda} and \check{\lambda} < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for \hat{\lambda}\leq \lambda \leq \check{\lambda} and \lambda = \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
Theorem 1.2 (See Figure 2). Consider classical positive solutions u of (1.1) with 0 < p < 1 . There exist 0 < L_{\ast } ( = L_{\ast }(p)) < L^{\ast } ( = L^{\ast }(p)) such that the following assertions (i)–(iv) hold:
(i) (See Figure 2(i)–(ii).) If L > L^{\ast }, then there exist 0 < \lambda _{\ast } < \lambda ^{\ast } such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for \lambda _{\ast } < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for 0 < \lambda \leq \lambda _{\ast } and \lambda = \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
(ii) (See Figure 2(iii).) If L = L^{\ast }, then there exist 0 < \lambda _{\ast } < \bar{\lambda} ( = \bar{\lambda}(p)) < \lambda ^{\ast } satisfying \lambda _{\ast } < 1-p < \bar{\lambda} such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for \lambda _{\ast } < \lambda < \bar{\lambda} and \bar{\lambda} < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for 0 < \lambda \leq \lambda _{\ast } and \lambda = \bar{ \lambda}, \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
(iii) (See Figure 2(iv).) If L_{\ast } < L < L^{\ast } , then there exist 0 < \lambda _{\ast } < \hat{\lambda} < \check{\lambda} < \lambda ^{\ast } satisfying \lambda _{\ast } < 1-p < \hat{\lambda} such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for \lambda _{\ast } < \lambda < \hat{\lambda} and \check{\lambda} < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for 0 < \lambda \leq \lambda _{\ast } , \hat{\lambda}\leq \lambda \leq \check{\lambda} and \lambda = \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
(iv) (See Figure 2(v).) If 0 < L\leq L_{\ast } , then there exist 0 < \check{\lambda} < \lambda ^{\ast } satisfying 1-p < \check{\lambda} such that (1.1) has exactly two positive solutions u_{\lambda } , v_{\lambda } with \left \Vert u_{\lambda }\right \Vert _{\infty } < \left \Vert v_{\lambda }\right \Vert _{\infty } for \check{\lambda} < \lambda < \lambda ^{\ast } , exactly one positive solution u_{\lambda } for 0 < \lambda \leq \check{\lambda} and \lambda = \lambda ^{\ast } , and no positive solution for \lambda > \lambda ^{\ast } .
The paper is organized as follows. Section 2 contains statements of the main results (Theorems 2.1–2.3). Section 3 contains several lemmas needed to prove Theorem 2.3. Section 4 contains the proofs of the main results.
The main results in this paper are next Theorems 2.1–2.3 for the generalized MEMS problem (1.1), in which we first study upper and lower bounds for the pull-in voltage \lambda ^{\ast } and the pull-in distance \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } . We further study monotonicity properties and asymptotic behaviors of the pull-in voltage \lambda ^{\ast } and the pull-in distance \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } with respect to positive parameters p and L .
Theorem 2.1 (See Theorems 1.1–1.2 and Figures 1 and 2). Consider (1.1) with p > 0 and L > 0 . Then the pull-in voltage \lambda ^{\ast } = \lambda ^{\ast }(p, L) satisfies the following assertions (i)–(iii):
(i) For p > 0 and L > 0 ,
\begin{eqnarray} \dfrac{2p^{p}}{(p+1)^{p+1}L^{2}\left[ 1+\frac{4}{(p+1)^{2}L^{2}}\right] ^{3/2}} &\leq &\lambda ^{\ast }(p, L) < \min \left \{ L^{-1}, \dfrac{p^{p}}{ 4(p+1)^{p+1}}\pi ^{2}L^{-2}\right \} \\ &\leq &\min \left \{ L^{-1}, \tfrac{1}{4}\pi ^{2}L^{-2}\right \} . \end{eqnarray} | (2.1) |
(ii) For any fixed L > 0, \lambda ^{\ast }(p, L) is a strictly decreasing function of p > 0 , and
\begin{equation*} \dfrac{2}{L^{2}\left[ 1+\frac{4}{L^{2}}\right] ^{3/2}}\leq \lim\limits_{p\rightarrow 0^{+}}\lambda ^{\ast }(p, L)\leq \min \left \{ L^{-1}, \tfrac{1}{4}\pi ^{2}L^{-2}\right \} , \end{equation*} |
\begin{equation*} \lim\limits_{p\rightarrow \infty }\lambda ^{\ast }(p, L) = 0. \end{equation*} |
(iii) For any fixed p > 0 , \lambda ^{\ast }(p, L) is a strictly decreasing function of L > 0 , \lim_{L\rightarrow 0^{+}}\lambda ^{\ast }(p, L) = \infty, and \lim_{L\rightarrow \infty }\lambda ^{\ast }(p, L) = 0 .
Remark 2.2. In (2.1) for (1.1) with p > 0 , the upper bound \min \left \{ L^{-1}, \tfrac{p^{p}}{4(p+1)^{p+1}}\pi ^{2}L^{-2}\right \} for \lambda ^{\ast }(p, L) is reduced to \min \left \{ L^{-1}, \tfrac{\pi ^{2}}{27} L^{-2}\right \} when p = 2; that is, our result for the upper bound for \lambda ^{\ast }(p, L) generalizes (1.7). In particular, when p = 2 and L = 1 , then (2.1) is reduced to
\begin{equation*} 0.171\approx \frac{8}{13\sqrt{13}}\leq \lambda ^{\ast }(2, 1) < \frac{\pi ^{2}}{ 27}\approx 0.366. \end{equation*} |
This suggests that (2.1) give suitable upper and lower bounds for the pull-in voltage \lambda ^{\ast }(p, L) with general p, L > 0.
Theorem 2.3 (See Theorems 1.1–1.2 and Figures 1–5). Consider (1.1) with p > 0 and L > 0 . Then the pull-in distance \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }(p, L) satisfies the following assertions (i)–(iv):
(i) For p > 1 and L > 0 ,
\begin{equation} \bar{L}(p, L) < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < \bar{U} (p), \end{equation} | (2.2) |
where
\begin{eqnarray*} \bar{L}(p, L) &\equiv &\min \left \{ 1-\left[ 1+\frac{1}{3}\left( p-1\right) L \right] ^{\frac{1}{1-p}}, \frac{1}{9p+1}\right \} \\ & = &\left \{ \begin{array}{ll} \dfrac{1}{9p+1} & \ if \ L\geq L_{1}(p)\equiv \frac{3^{3-2p}p\left( \frac{ 1+9p}{p}\right) ^{p}-27p-3}{(p-1)(1+9p)}, \\ 1-\left[ 1+\frac{1}{3}\left( p-1\right) L\right] ^{\frac{1}{1-p}} & \ if \ 0 < L < L_{1}(p), \end{array} \right. \end{eqnarray*} |
and (p, \bar{U}(p)) is the unique positive solution pair of the equation
\begin{equation*} \Gamma (p, r)\equiv (p+1)r+2(1-r)^{p}-2 = 0\text{, }\ p > 1, \ \ 0 < r < 1, \end{equation*} |
and \bar{U}(p) satisfies
\begin{equation} \bar{U}(p)\leq U(p)\equiv \left \{ \begin{array}{ll} 1-2e^{-2} \ \ (\approx 0.729) & \ for \ 1 < p\leq \frac{1}{1-2e^{-2}} \approx 1.371, \\ 1/p & \ for \ \frac{1}{1-2e^{-2}} < p\leq 2, \\ 2/(p+2) & \ for \ p > 2; \end{array} \right. \end{equation} | (2.3) |
see Figure 4.
(ii) For p = 1 and L > 0 ,
\begin{equation*} \min \left \{ 1-e^{-\frac{L}{3}}, \frac{1}{10}\right \} < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < 1-2e^{-2}\approx 0.729, \end{equation*} |
where
\begin{equation*} \min \left \{ 1-e^{-\frac{L}{3}}, \frac{1}{10}\right \} = \left \{ \begin{array}{ll} \frac{1}{10} & \ if \ L\geq L_{2}\equiv 3\ln (\frac{10}{9})\approx 0.316, \\ 1-e^{-\frac{L}{3}} & \ if \ 0 < L < L_{2}. \end{array} \right. \end{equation*} |
(iii) For 0 < p < 1 and L > 0 ,
\begin{equation*} \hat{L}(p, L) < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < 1, \end{equation*} |
where
\begin{eqnarray} \hat{L}(p, L) &\equiv &\min \left \{ 1-\left[ 1-\frac{1}{k}\left( 1-p\right) L \right] ^{\frac{1}{1-p}}, \frac{1}{k^{2}p+1}\right \} \text{, } \ k = \max \left \{ 3, (1-p)L\right \} \\ & = &\left \{ \begin{array}{ll} \frac{1}{p(1-p)^{2}L^{2}+1} & \ if \ (1-p)L\geq 3, \\ \min \left \{ 1-\left[ 1-\frac{1}{3}\left( 1-p\right) L\right] ^{\frac{1}{1-p }}, \frac{1}{9p+1}\right \} & \ if \ 0 < (1-p)L < 3. \end{array} \right. \end{eqnarray} | (2.4) |
(iv) (a) For fixed L > 0 ,
\begin{equation} \lim\limits_{p\rightarrow \infty }\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = 0. \end{equation} | (2.5) |
In addition, for fixed L\geq 3 ,
\begin{equation} \lim\limits_{p\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = 1, \end{equation} | (2.6) |
and for fixed positive L < 3,
\begin{equation} \lim\limits_{p\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }\geq \frac{L}{3}. \end{equation} | (2.7) |
(b) For fixed p > 0 ,
\begin{equation} \lim\limits_{L\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = 0. \end{equation} | (2.8) |
Remark 2.4. In particular, when p = 2 , then in (2.2) we have that \bar{U} (p = 2) = 1/2 and
\begin{equation*} \bar{L}(p = 2, L) = \left \{ \begin{array}{ll} \frac{1}{19}\approx 0.0526 & \ if \ L\geq L_{1}(p = 2) = \frac{1}{6}\approx 0.167, \\ \frac{L}{L+3} & \ if \ 0 < L < L_{1}(p = 2) = \frac{1}{6}. \end{array} \right. \end{equation*} |
Hence, for L\geq 1/6 , the estimates give
\begin{equation*} 0.0526\approx \frac{1}{19} < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }(p = 2, L) < \frac{1}{2} = 0.5, \end{equation*} |
and, for 0 < L < 1/6 , the estimates give
\begin{equation*} (0, \frac{1}{19})\ni \frac{L}{L+3} < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }(p = 2, L) < \frac{1}{2} = 0.5. \end{equation*} |
Our analytic result in (2.2) also agrees with some numerical simulations obtained by Brubaker and Pelesko [4,Figure 1.2(b)–(d)] for p = 2 and 0.1\leq L\leq 0.8 .
Remark 2.5. We conjecture that, for any fixed L > 0, \lim_{p\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = 1 ; cf. (2.6)–(2.7) and (1.9). In addition, for any fixed p > 0 and \lambda ^{\ast } = \lambda ^{\ast }(L), \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } is a strictly increasing function of L > 0 . Note that, when p = 2 , our numerical simulation shows that \lim_{L\rightarrow \infty }\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = \zeta \approx 0.388 for some \zeta , cf. [4,Figure 1.2(b)] and (2.8). Further investigations are needed.
To prove the upper and lower bounds for \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } in Theorem 2.3(i)–(iii) for (1.1) with p > 0 , we need the following Lemmas 3.1–3.6. We first establish sufficient conditions on r and p such that \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < r for all L > 0 . To this purpose, we recall the time map formula T_{p, \lambda }(r) for (1.1) as follows:
\begin{equation} T_{p, \lambda }(r) = \int_{0}^{r}\frac{1+\lambda F(u)-\lambda F(r)}{\sqrt{1- \left[ 1+\lambda F(u)-\lambda F(r)\right] ^{2}}}du, \quad r = \left \Vert u\right \Vert _{\infty }\in I; \end{equation} | (3.1) |
where
\begin{equation} F(u)\equiv \int \nolimits_{0}^{u}f(t)dt = \int \nolimits_{0}^{u}\dfrac{1}{ (1-t)^{p}}dt = \left \{ \begin{array}{ll} \dfrac{1-(1-u)^{1-p}}{1-p} & \text{if }p > 0\text{ and }p\neq 1, \\ -\ln (1-u) & \text{if }p = 1, \end{array} \right. \end{equation} | (3.2) |
and I is the domain of T_{p, \lambda }(r) . Notice that the domain I of T_{p, \lambda }(r) depends on the value p . We have that:
\rm(I) If p\geq 1 , F:[0, 1)\rightarrow \lbrack 0, \infty) is strictly increasing, and hence F^{-1} is well defined on [0, \infty) . Then for any \lambda > 0 , the domain I of T_{p, \lambda }(r) is
\begin{equation*} (0, F^{-1}(\frac{1}{\lambda })]. \end{equation*} |
\rm(II) If 0 < p < 1 , F:[0, 1]\rightarrow \lbrack 0, \frac{1}{1-p}] is strictly increasing, and hence F^{-1} is only defined on [0, \frac{1}{1-p}] . Then for any \lambda > 0 , the domain I of T_{p, \lambda }(r) is
\begin{equation*} \left \{ \begin{array}{ll} (0, F^{-1}(\frac{1}{\lambda })] & \text{if }\lambda > 1-p, \\ (0, 1) & \text{if }0 < \lambda \leq 1-p\text{.} \end{array} \right. \end{equation*} |
See [1,p. 286].
Observe that positive solutions u_{\lambda } for (1.1) correspond to
\begin{equation} \left \Vert u_{\lambda }\right \Vert _{\infty } = r\ \ \text{and }\ T_{p, \lambda }(r) = L. \end{equation} | (3.3) |
Thus, studying of the exact number of positive solutions of (1.1) for any fixed \lambda > 0 is equivalent to studying the shape of the time map T_{p, \lambda }(r) on its domain I . Moreover, we observe that
\begin{eqnarray} &&T_{p, \lambda }^{\prime }(r) \\ & = &\int_{0}^{1}\frac{\left[ 1+\lambda F(rs)-\lambda F(r)\right] \left \{ 1- \left[ 1+\lambda F(rs)-\lambda F(r)\right] ^{2}\right \} +\lambda \left[ rsf(rs)-rf(r)\right] }{\left \{ 1-\left[ 1+\lambda F(rs)-\lambda F(r)\right] ^{2}\right \} ^{3/2}}ds \\ & = &\int_{0}^{1}\frac{\Psi (\lambda , r, s)}{\left \{ 1-\left[ 1+\lambda F(rs)-\lambda F(r)\right] ^{2}\right \} ^{3/2}}ds, \end{eqnarray} | (3.4) |
where
\begin{eqnarray} &&\Psi (\lambda , r, s) \\ &\equiv &\left[ 1+\lambda F(rs)-\lambda F(r)\right] \left \{ 1-\left[ 1+\lambda F(rs)-\lambda F(r)\right] ^{2}\right \} +\lambda \left[ rsf(rs)-rf(r)\right] . \end{eqnarray} | (3.5) |
See [1,(3.2)].
First, we have the next lemma which shows that T_{p, \lambda }(r) for (1.1) has exactly one critical point, a local maximum, on its domain.
Lemma 3.1 ([1,Lemma 3.2]). Consider T_{p, \lambda }(r) for (1.1). The followingassertions (i)–(iii) hold:
(i) For fixed p\geq 1 , T_{p, \lambda }(r) has exactly one criticalpoint, a local maximum, on (0, F^{-1}(1/\lambda)) for any \lambda > 0 .
(ii) For fixed p\in (0, 1) , T_{p, \lambda }(r) has exactly onecritical point, a local maximum, on (0, F^{-1}(1/\lambda)) for any \lambda > 1-p .
(iii) For fixed p\in (0, 1) , T_{p, \lambda }(r) has exactly onecritical point, a local maximum, on (0, 1) for any 0 < \lambda \leq 1-p .
In the following Lemma 3.2 with p > 1 , and Lemma 3.3 with p = 1 , we prove that T_{p, \lambda }^{\prime }(r) < 0 for \lambda \in (0, 1/F(r)) , where r satisfies some conditions stated below. Thus \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < r for L > 0 by applying Theorem 1.1.
Lemma 3.2. Suppose that 0 < r < 1\ and p > 1 . Then T_{p, \lambda}^{\prime }(r) < 0 for \lambda \in (0, 1/F(r)) if \left(p, r\right) satisfies \Gamma (p, r) = (p+1)r+2(1-r)^{p}-2 = 0 .
Lemma 3.3. Suppose that p = 1 . Then T_{1, \lambda }^{\prime}(1-2e^{-2}) < 0 for \lambda \in (0, 1/F(1-2e^{-2})) .
Proof of Lemma 3.2. For p > 1 , we have that f(u) = (1-u)^{-p} and F(u) = \frac{1-(1-u)^{1-p}}{1-p} . Hence, by (3.5), we compute that
\begin{equation} \Psi (\lambda , r, s) = \lambda \Phi (\lambda , r, s) \end{equation} | (3.6) |
where
\begin{eqnarray} &&\Phi (\lambda , r, s) \\ &\equiv &\frac{1}{\left( p-1\right) ^{3}}\left[ (1-r)^{1-p}-(1-rs)^{1-p} \right] ^{3}\lambda ^{2}-\frac{3}{\left( p-1\right) ^{2}}\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] ^{2}\lambda \\ &&+\frac{2}{p-1}\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] +\left[ rs(1-rs)^{-p}-r(1-r)^{-p}\right] . \end{eqnarray} | (3.7) |
For any fixed positive r, s < 1 , \tilde{\Phi}(\lambda)\equiv \Phi (\lambda, r, s) is a quadratic polynomial in \lambda . To prove this lemma, by (3.1)–(3.6), for p > 1 , it suffice to prove that \tilde{\Phi} (\lambda) < 0 on [0, 1/F(r)] which follows by proving that \tilde{\Phi} (\lambda) is convex on [0, 1/F(r)] , \tilde{\Phi}(0) < 0 and \tilde{\Phi} (1/F(r)) < 0 .
\rm(I) First, we consider the leading coefficient of the quadratic polynomial \tilde{\Phi}(\lambda) . Let
\begin{equation*} a_{1}(s)\equiv (1-r)^{1-p}-(1-rs)^{1-p}. \end{equation*} |
We have that a_{1}(1) = 0 and
\begin{equation*} a_{1}^{\prime }(s) = \left( 1-p\right) r\left( 1-rs\right) ^{-p} < 0 \ \text{ for }p > 1\text{ and }0 < r, s < 1\text{.} \end{equation*} |
Hence, a_{1}(s) > 0 for 0 < r, s < 1 . So \tilde{\Phi}(\lambda) is convex on [0, 1/F(r)] .
\rm(II) Secondly, we let
\begin{eqnarray*} a_{2}(s) &\equiv &\left( 1-p\right) (1-r)^{p}(1-rs)^{p}\tilde{\Phi}(0) \\ & = &-\left( 2-r-rp\right) (1-rs)^{p}+\left( 2-rs-rsp\right) (1-r)^{p}\text{.} \end{eqnarray*} |
We find that a_{2}(1) = 0 , and
\begin{equation*} a_{2}(0) = (p+1)r+2(1-r)^{p}-2 = 0 \end{equation*} |
by the assumption. On the other hand, since
\begin{equation*} a_{2}^{\prime }(s) = rp\left( 2-r-rp\right) (1-rs)^{p-1}-\left( r+rp\right) (1-r)^{p} \end{equation*} |
and
\begin{equation*} a_{2}^{\prime \prime }(s) = -p\left( p-1\right) r^{2}\left( 2-r-rp\right) (1-rs)^{p-2}, \end{equation*} |
we find that
\begin{eqnarray*} (1-rs)a_{2}^{\prime \prime }(s)+r\left( p-1\right) a_{2}^{\prime }(s) & = &-\left( p^{2}-1\right) r^{2}(1-r)^{p} \\ & < &0 \ \text{ for }p > 1\text{ and }0 < r, s < 1. \end{eqnarray*} |
That is, for s\in (0, 1) , we find that a_{2}^{\prime \prime }(s) < 0 whenever a_{2}^{\prime }(s) = 0 . Hence a_{2}(s) > 0 for s\in (0, 1) . This implies that \tilde{\Phi}(0) < 0 .
\rm(III) Thirdly, we obtain that
\begin{eqnarray*} a_{3}(s) &\equiv &\left( p-1\right) \tilde{\Phi}(1/F(r)) \\ & = &\frac{1}{\left[ (1-r)^{1-p}-1\right] ^{2}}\left[ (1-r)^{1-p}-(1-rs)^{1-p} \right] ^{3} \\ &&-\frac{3}{(1-r)^{1-p}-1}\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] ^{2} \\ &&-\left( 2-rs-rsp\right) (1-rs)^{-p}+\left( 2-r-rp\right) (1-r)^{-p}. \end{eqnarray*} |
We have that a_{3}(1) = 0 and a_{3}(0) = -r\left(p-1\right) (1-r)^{-p} < 0 for p > 1 and 0 < r < 1 . Moreover, we compute that
\begin{eqnarray*} \phi _{1}(s) &\equiv &\frac{(1-rs)^{p}}{r}a_{3}^{\prime }(s) \\ & = &\frac{3(1-p)}{\left[ (1-r)^{1-p}-1\right] ^{2}}(1-rs)^{2-2p}-\frac{6(1-p) }{\left[ (1-r)^{1-p}-1\right] ^{2}}(1-rs)^{1-p} \\ &&-\frac{3(1-p)\left[ (1-r)^{2-2p}-2(1-r)^{1-p}\right] }{\left[ (1-r)^{1-p}-1 \right] ^{2}}+p\left( rsp+rs-2\right) (1-rs)^{-1}+p+1\text{.} \end{eqnarray*} |
Then we compute that \phi _{1}(0) = 2(p-1) > 0 and
\begin{equation} \phi _{1}(1) = \frac{\left( p-1\right) \left[ (p+1)r-1\right] }{1-r}. \end{equation} | (3.8) |
We claim that \phi _{1}(1) > 0 if \left(p, r\right) satisfies \Gamma (p, r) = \left(p+1\right) r+2\left(1-r\right) ^{p}-2 = 0. We next give a proof of this claim. First, it is easy to see that
\begin{eqnarray} \hat{\Phi} &\equiv &\left \{ \left( p, r\right) :p > 1\text{, }0 < r < 1\text{, } \left( p+1\right) r+2\left( 1-r\right) ^{p}-2\geq 0\right \} \\ &&\left. \cap \left \{ \left( p, r\right) :p > 1\text{, }0 < r < 1\text{, }\left( p+1\right) r-1\geq 0\right \} \neq \emptyset \right. \end{eqnarray} | (3.9) |
since \left(p, r\right) = (2, 1/2)\in \hat{\Phi} . Now, suppose \left(p, r\right) satisfies \left(p+1\right) r-1 = 0 . We find that
\begin{equation*} \left( p+1\right) r+2\left( 1-r\right) ^{p}-2 = 2\left( 1-r\right) ^{p}-1 = 2\left( \frac{p}{p+1}\right) ^{p}-1 < 0 \end{equation*} |
since 2\left(\frac{p}{p+1}\right) ^{p}-1 = 0 when p = 1 and it is a strictly decreasing function of p\geq 1. So, in addition to (3.9), for p > 1 , 0 < r < 1 , we obtain that \left(p+1\right) r-1 > 0 if \left(p+1\right) r+2\left(1-r\right) ^{p}-2\geq 0 ; i.e.,
\begin{eqnarray} &&\left \{ \left( p, r\right) :p > 1\text{, }0 < r < 1\text{, }\left( p+1\right) r+2\left( 1-r\right) ^{p}-2\geq 0\right \} \\ &\subseteq &\left \{ \left( p, r\right) :p > 1\text{, }0 < r < 1\text{, }\left( p+1\right) r-1 > 0\right \} . \end{eqnarray} | (3.10) |
So, by (3.8) and (3.10), \phi _{1}(1) > 0 if \left(p, r\right) satisfies \left(p+1\right) r+2\left(1-r\right) ^{p}-2 = 0 .
We also compute that, if p\geq \frac{3}{2} ,
\begin{eqnarray*} &&\frac{(1-rs)}{r}\phi _{1}^{\prime \prime }(s)+(1-2p)\phi _{1}^{\prime }(s) \\ & = &-\frac{6\left( p-1\right) ^{3}r}{\left[ \left( 1-r\right) ^{1-p}-1\right] ^{2}}(1-rs)^{-p}-rp\left( p-1\right) \left( 2p-3\right) (1-rs)^{-2} < 0 \ \text{ for }s\in (0, 1). \end{eqnarray*} |
So we find that \phi _{1}^{\prime \prime }(s) < 0 whenever \phi _{1}^{\prime }(s) = 0 . This implies that \phi _{1}(s) = \frac{(1-rs)^{p}}{r} a_{3}^{\prime }(s) > 0 for s\in (0, 1) (Observe \phi _{1}(0) > 0 and \phi _{1}(1) > 0 ). Therefore, a_{3}(s) < 0 for s\in (0, 1) , and hence
\begin{equation*} \tilde{\Phi}(1/F(r)) < 0 \ \text{ if }p\geq \frac{3}{2}. \end{equation*} |
\rm(IV) If 1 < p < \frac{3}{2} , we claim that
\begin{equation*} \frac{(1-rs)}{r}\phi _{1}^{\prime \prime }(s)+(1-2p)\phi _{1}^{\prime }(s) < 0 \ \text{ for }s\in (0, 1). \end{equation*} |
We next give a proof of this claim. We compute that
\begin{eqnarray} \phi _{2}(s) &\equiv &\frac{(1-rs)^{2}}{r\left( p-1\right) }\left[ \frac{ (1-rs)}{r}\phi _{1}^{\prime \prime }(s)+(1-2p)\phi _{1}^{\prime }(s)\right] \\ & = &p\left( 3-2p\right) -\frac{6\left( p-1\right) ^{2}}{\left[ (1-r)^{1-p}-1 \right] ^{2}}\left( 1-sr\right) ^{2-p}. \end{eqnarray} | (3.11) |
Then
\begin{equation} \phi _{2}(1) = p\left( 3-2p\right) -\frac{6\left( p-1\right) ^{2}}{\left[ (1-r)^{1-p}-1\right] ^{2}}\left( 1-r\right) ^{2-p} \end{equation} | (3.12) |
and
\begin{equation} \phi _{2}^{\prime }(s) = \frac{6r\left( p-1\right) ^{2}(2-p)}{\left[ (1-r)^{1-p}-1\right] ^{2}}\left( 1-sr\right) ^{1-p} > 0. \end{equation} | (3.13) |
So, if \left(p, r\right) satisfies
\begin{equation*} p\left( 3-2p\right) -\frac{6\left( p-1\right) ^{2}}{\left[ (1-r)^{1-p}-1 \right] ^{2}}\left( 1-r\right) ^{2-p}\leq 0, \end{equation*} |
then
\begin{equation*} \frac{(1-rs)}{r}\phi _{1}^{\prime \prime }(s)+(1-2p)\phi _{1}^{\prime }(s)\leq 0 \ \text{ for }s\in (0, 1). \end{equation*} |
Next, we have that, for 1 < p < \frac{3}{2} ,
\begin{eqnarray} &&\left \{ \left( p, r\right) : \ \ p\in (1, 3/2)\text{, }r\in (0, 1)\text{, }\left( p+1\right) r+2\left( 1-r\right) ^{p}-2 = 0\right \} \\ &\subsetneq &\left \{ \left( p, r\right) :p\in (1, 3/2)\text{, }r\in (0, 1) \text{, }p\left( 3-2p\right) -\frac{6\left( p-1\right) ^{2}}{\left[ (1-r)^{1-p}-1\right] ^{2}}\left( 1-r\right) ^{2-p}\leq 0\right \} ; \\ && \end{eqnarray} | (3.14) |
see Figure 6. (Note that we can provide an analytic proof for (3.14) but we omit it here since it is too tedious.) If p\in (1, 3/2) , then
\begin{equation*} \frac{(1-rs)^{2}}{r\left( p-1\right) }\left \{ \frac{(1-rs)}{r}\phi _{1}^{\prime \prime }(s)+(1-2p)\phi _{1}^{\prime }(s)\right \} < 0 \ \text{ for }s\in (0, 1) \end{equation*} |
by (3.11)–(3.14). So we find that \phi _{1}^{\prime \prime }(s) < 0 whenever \phi _{1}^{\prime }(s) = 0 . This implies that \phi _{1}(s) = \frac{(1-rs)^{p}}{r}a_{3}^{\prime }(s) > 0 for s\in (0, 1) (Observe \phi _{1}(0) > 0 and \phi _{1}(1) > 0 ). Therefore, a_{3}(s) < 0 for s\in (0, 1) , and hence
\begin{equation*} \tilde{\Phi}(1/F(r)) < 0 \ \text{ if }1 < p < \frac{3}{2}. \end{equation*} |
We conclude that, by above parts (I)–(IV), \tilde{\Phi}(\lambda) is convex on [0, 1/F(r)] , \tilde{\Phi}(0) < 0 and \tilde{\Phi}(1/F(r)) < 0 . So \tilde{\Phi}(\lambda) < 0 on [0, 1/F(r)] for p > 1 . By (3.4)–(3.7), T_{p, \lambda }^{\prime }(r) < 0 for \lambda \in (0, 1/F(r)) if p > 1 , r\in (0, 1) , and \left(p, r\right) satisfies \Gamma (p, r) = (p+1)r+2(1-r)^{p}-2 = 0 .
The proof of Lemma 3.2 is complete.
Proof of Lemma 3.3. For p = 1 , we have that f(u) = \frac{1}{1-u} and F(u) = -\ln (1-u) , and (3.5) can be reduced to
\begin{equation} \Psi (\lambda , r, s) = \lambda \Theta (\lambda , r, s), \end{equation} | (3.15) |
where
\begin{eqnarray} \Theta (\lambda , r, s) &\equiv &\left[ \ln (1-rs)-\ln (1-r)\right] ^{3}\lambda ^{2}-3\left[ \ln (1-rs)-\ln (1-r)\right] ^{2}\lambda \\ &&+2\left[ \ln (1-rs)-\ln (1-r)\right] -\frac{r\left( 1-s\right) }{\left( 1-r\right) \left( 1-rs\right) }. \end{eqnarray} | (3.16) |
Let r_{0}\equiv 1-2e^{-2} ( \approx 0.729 ), since \ln (1-r_{0}s)-\ln (1-r_{0}) > 0 for s\in (0, 1) , \tilde{\Theta}(\lambda)\equiv \Theta (\lambda, r_{0}, s) is a convex quadratic polynomial in \lambda . In the following, we will prove that
\begin{equation*} \tilde{\Theta}(0) = 2\left[ \ln (1-r_{0}s)-\ln (1-r_{0})\right] -\frac{ r_{0}\left( 1-s\right) }{\left( 1-r_{0}\right) \left( 1-r_{0}s\right) } < 0 \ \text{ for }s\in (0, 1) \end{equation*} |
and
\begin{equation*} \tilde{\Theta}(1/F(r_{0})) = \frac{1}{\left[ \ln (1-r_{0})\right] ^{2}}\left[ \ln (1-r_{0}s)\right] ^{3}-\ln \left( 1-r_{0}s\right) -\frac{r_{0}\left( 1-s\right) }{\left( 1-r_{0}\right) \left( 1-r_{0}s\right) } < 0 \ \text{ for } s\in (0, 1). \end{equation*} |
Let
\begin{equation*} \theta _{1}(s)\equiv 2\left[ \ln (1-r_{0}s)-\ln (1-r_{0})\right] -\frac{ r_{0}\left( 1-s\right) }{\left( 1-r_{0}\right) \left( 1-r_{0}s\right) }. \end{equation*} |
Then \theta _{1}(0) = -2\ln (1-r_{0})-\frac{r_{0}}{1-r_{0}}\approx -0.0808 < 0 , \theta _{1}(1) = 0 ,
\begin{equation*} \theta _{1}^{\prime }(s) = \frac{-2r_{0}}{1-r_{0}s}+\frac{r_{0}}{\left( 1-r_{0}s\right) ^{2}} \end{equation*} |
and
\begin{equation*} \theta _{1}^{\prime \prime }(s) = \frac{-2r_{0}^{2}}{\left( 1-r_{0}s\right) ^{2}}+\frac{2r_{0}^{2}}{\left( 1-r_{0}s\right) ^{3}}. \end{equation*} |
We compute that
\begin{equation*} sr_{0}^{2}\theta _{1}^{\prime }(s)+\left( 1-r_{0}s\right) ^{2}\theta _{1}^{\prime \prime }(s) = \frac{r_{0}^{3}s}{\left( 1-r_{0}s\right) ^{2}} > 0. \end{equation*} |
This implies that \theta _{1}^{\prime \prime }(s) > 0 whenever \theta _{1}^{\prime }(s) = 0 . Thus, \theta _{1}(s) < 0 for all s\in (0, 1) and then \tilde{\Theta}(0) < 0 for s\in (0, 1) .
On the other hand, let
\begin{equation*} \theta _{2}(s)\equiv \frac{1}{\left[ \ln (1-r_{0})\right] ^{2}}\left[ \ln (1-r_{0}s)\right] ^{3}-\ln \left( 1-r_{0}s\right) -\frac{r_{0}\left( 1-s\right) }{\left( 1-r_{0}\right) \left( 1-r_{0}s\right) }. \end{equation*} |
Then \theta _{2}(0) = \frac{-r_{0}}{1-r_{0}} (\approx -2.695) < 0 , \theta _{2}(1) = 0 ,
\begin{equation*} \theta _{2}^{\prime }(s) = -\frac{3r_{0}}{\left[ \ln (1-r_{0})\right] ^{2}\left( 1-r_{0}s\right) }\left[ \ln (1-r_{0}s)\right] ^{2}+\frac{ r_{0}\left( 2-r_{0}s\right) }{\left( 1-r_{0}s\right) ^{2}} \end{equation*} |
and
\begin{eqnarray*} \theta _{2}^{\prime \prime }(s) & = &-\frac{3r_{0}^{2}}{\left[ \ln (1-r_{0}) \right] ^{2}\left( 1-r_{0}s\right) ^{2}}\left[ \ln (1-r_{0}s)\right] ^{2} \\ &&+\frac{6r_{0}^{2}}{\left[ \ln (1-r_{0})\right] ^{2}\left( 1-r_{0}s\right) ^{2}}\left[ \ln (1-r_{0}s)\right] +\frac{r_{0}^{2}\left( 3-r_{0}s\right) }{ \left( 1-r_{0}s\right) ^{3}}\text{.} \end{eqnarray*} |
We compute that \theta _{2}^{\prime }(1) = \frac{r_{0}}{\left(1-r_{0}\right) ^{2}}\left(2r_{0}-1\right) (\approx 4.566) > 0 , and
\begin{equation} r_{0}\theta _{2}^{\prime }(s)-\left( 1-r_{0}s\right) \theta _{2}^{\prime \prime }(s) = -\frac{r_{0}^{2}}{\left( 1-r_{0}s\right) ^{2}}\left \{ \frac{ 6\left( 1-r_{0}s\right) }{\left[ \ln (1-r_{0})\right] ^{2}}\left[ \ln (1-r_{0}s)\right] +1\right \} . \end{equation} | (3.17) |
Furthermore, let
\begin{equation} \theta _{3}(s)\equiv \frac{6\left( 1-r_{0}s\right) }{\left[ \ln (1-r_{0}) \right] ^{2}}\left[ \ln (1-r_{0}s)\right] +1. \end{equation} | (3.18) |
Then \theta _{3}(0) = 1 > 0 , \theta _{3}(1) = \frac{6\left(1-r_{0}\right) }{ \ln (1-r_{0})}-1 (\approx -0.243) < 0 , and
\begin{equation*} r_{0}\theta _{3}^{\prime }(s)+\left[ \ln (1-r_{0}s)\right] (1-r_{0}s)\theta _{3}^{\prime \prime }(s) = -\frac{6r_{0}^{2}}{\left[ \ln (1-r_{0})\right] ^{2}} < 0\text{.} \end{equation*} |
This implies that \theta _{3}^{\prime \prime }(s) > 0 whenever \theta _{3}^{\prime }(s) = 0 . Then there exists s_{0}\in (0, 1) such that
\begin{equation} \theta _{3}(s)\left \{ \begin{array}{cl} > 0 & \text{on }(0, s_{0}), \\ = 0 & \text{when }s = s_{0}, \\ < 0 & \text{on }(s_{0}, 1). \end{array} \right. \end{equation} | (3.19) |
By (3.17)–(3.19), we obtain that \theta _{2}^{\prime \prime }(s) > 0 (resp. \theta _{2}^{\prime \prime }(s) = 0 , \theta _{2}^{\prime \prime }(s) < 0 ) whenever \theta _{2}^{\prime }(s) = 0 and s\in (0, s_{0}) (resp. s = s_{0} , s\in (s_{0}, 1) ). We next show that \theta _{2}(s) < 0 for all s\in (0, 1) . Observe \theta _{2}(0) < 0 , \theta _{2}(1) = 0 and \theta _{2}^{\prime }(1) > 0 . Assume \theta _{2}(s)\geq 0 for some s\in (0, 1) , then there exist 0 < s_{1} < s_{2} < 1 such that \theta _{2}(s_{1})\geq 0 is a local maximum of \theta _{2}(s) and \theta _{2}(s_{2}) < 0 is a local minimum of \theta _{2}(s) . Thus \theta _{2}^{\prime \prime }(s_{1})\leq 0 and \theta _{2}(s_{2})\geq 0 , which contradicts to (3.17)–(3.19). So \theta _{2}(s) < 0 for all s\in (0, 1) .
Finally, by the above analyses with r_{0} = 1-2e^{-2} , \tilde{\Theta} (\lambda) is convex on [0, 1/F(r_{0})] , \tilde{\Theta}(0) < 0 and \tilde{ \Theta}(1/F(r_{0})) < 0 . So \tilde{\Theta}(\lambda) < 0 on [0, 1/F(r_{0})] . By (3.4), (3.5), (3.15) and (3.16), T_{1, \lambda }^{\prime }(r_{0}) < 0 for \lambda \in (0, 1/F(r_{0})) .
The proof of Lemma 3.3 is complete.
In the following Lemma 3.4 with p > 1 , Lemma 3.5 with p = 1 , and Lemma 3.6 with 0 < p < 1 , we prove that T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] , where r_{p} is defined below. Thus \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } > r_{p} for L > 0 by applying Theorems 1.1–1.2.
Lemma 3.4. Consider p > 1 . Then T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] , where
\begin{equation*} r_{p}\equiv \min \left \{ 1-\left[ 1+\frac{1}{3}\left( p-1\right) L\right] ^{ \frac{1}{1-p}}, \frac{1}{9p+1}\right \} . \end{equation*} |
Lemma 3.5. Consider p = 1 . Then T_{1, \lambda }^{\prime }(r_{1}) > 0 for \lambda \in (0, 1/L] , where r_{1}\equiv \min \left \{ 1-e^{-\frac{L}{3}}, \frac{1}{10}\right \} .
Lemma 3.6. Consider 0 < p < 1 . Then T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] , where
\begin{equation*} r_{p}\equiv \hat{L}(p, L) = \min \left \{ 1-\left[ 1-\frac{1}{k}\left( 1-p\right) L\right] ^{\frac{1}{1-p}}, \frac{1}{k^{2}p+1}\right \} , \ k = \max \left \{ 3, (1-p)L\right \} . \end{equation*} |
Proof of Lemma 3.4. For p > 1 , we have that f(u) = (1-u)^{-p} and F(u) = \frac{(1-u)^{1-p}-1}{p-1} . Let \lambda = \frac{q }{L} with q\in (0, 1] . Hence by (3.5), we compute that
\begin{eqnarray} G(q, r, s) &\equiv &\left( p-1\right) ^{3}L^{3}\Psi (\frac{q}{L}, r, s) \\ & = &q^{3}\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] ^{3}-3q^{2}\left( p-1\right) L \left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] ^{2} \\ &&+2q\left( p-1\right) ^{2}L^{2}\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] \\ &&-q\left( p-1\right) ^{3}L^{2}\left[ r(1-r)^{-p}-rs(1-rs)^{-p}\right] . \end{eqnarray} | (3.20) |
Assume that M > 1 is a given number. Then, for 0 < r, s < 1 satisfying \frac{1 }{1-r}\leq M , by applying Cauchy's Mean Value Theorem, it is easy to check that
\begin{equation*} 0 < r(1-r)^{-p}-rs(1-rs)^{-p}\leq \left( \frac{pM}{p-1}-1\right) \left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] . \end{equation*} |
Therefore,
\begin{equation} G(q, r, s)\geq q\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] \tilde{G}(q, r, s), \end{equation} | (3.21) |
where
\begin{eqnarray} &&\tilde{G}(q, r, s) \\ &\equiv &q^{2}\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] ^{2} \\ &&-3q\left( p-1\right) L\left[ (1-r)^{1-p}-(1-rs)^{1-p}\right] +\left( p-1\right) ^{2}L^{2}\left( p+1-pM\right) . \end{eqnarray} | (3.22) |
Note that 0 < (1-r)^{1-p}-(1-rs)^{1-p} < (1-r)^{1-p}-1\leq M^{p-1}-1 . Let
\begin{equation} g_{M}(z)\equiv z^{2}-3\left( p-1\right) Lz+\left( p-1\right) ^{2}L^{2}\left( p+1-pM\right) \text{.} \end{equation} | (3.23) |
We aim to find a number M_{0} > 1 such that g_{M_{0}}(z) > 0 for 0\leq z\leq M_{0}^{p-1}-1 . This implies that
\begin{equation*} \tilde{G}(q, r, s) > 0 \ \text{ for }0 < q\leq 1, \text{ }0 < r, s < 1 \end{equation*} |
with \frac{1}{1-r}\leq M_{0} . Since g_{M}^{\prime }(0) = -3\left(p-1\right) L < 0 and g_{M} is convex on (0, \infty) , we only need to prove that g_{M_{0}}^{\prime }(M_{0}^{p-1}-1)\leq 0 and g_{M_{0}}(M_{0}^{p-1}-1)\geq 0 for some M_{0} > 1 . We have that
\begin{equation*} g_{M}^{\prime }(z) = 2z-3\left( p-1\right) L \end{equation*} |
and
\begin{equation} g_{M_{0}}^{\prime }(M_{0}^{p-1}-1) = 2M_{0}^{p-1}-2-3\left( p-1\right) L\leq 0 \ \text{ if and only if } \ M_{0}\leq \left[ 1+\frac{3}{2}\left( p-1\right) L \right] ^{\frac{1}{p-1}}. \end{equation} | (3.24) |
Thus we choose M_{0}\equiv \min \left \{ \left[1+\frac{1}{3}\left(p-1\right) L\right] ^{\frac{1}{p-1}}, 1+\frac{1}{9p}\right \} . Then
\begin{eqnarray} g_{M_{0}}(M_{0}^{p-1}-1) & = &(M_{0}^{p-1}-1)^{2}-3\left( p-1\right) L(M_{0}^{p-1}-1)+\left( p-1\right) ^{2}L^{2}\left( p+1-pM_{0}\right) \\ & = &\left[ \frac{8}{3}\left( p-1\right) L-(M_{0}^{p-1}-1)\right] \left[ \frac{ 1}{3}\left( p-1\right) L-(M_{0}^{p-1}-1)\right] \\ &&+\left( p-1\right) ^{2}L^{2}\left( p+\frac{1}{9}-pM_{0}\right) \\ &\geq &\left( p-1\right) ^{2}L^{2}\left( p+\frac{1}{9}-pM_{0}\right) \geq 0 \text{.} \end{eqnarray} | (3.25) |
So g_{M_{0}}^{\prime }(M_{0}^{p-1}-1)\leq 0 and g_{M_{0}}(M_{0}^{p-1}-1) \geq 0 by (3.24) and (3.25).
Finally, we choose
\begin{equation*} r_{p}\equiv 1-\frac{1}{M_{0}} = \min \left \{ 1-\left[ 1+\frac{1}{3}\left( p-1\right) L\right] ^{\frac{1}{1-p}}, \frac{1}{9p+1}\right \} . \end{equation*} |
Then we obtain that T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] by (3.4), (3.5), and (3.20)–(3.25). Observe that r_{p}\in (0, F^{-1}(\frac{1}{\lambda })) for \lambda \in (0, 1/L] .
The proof of Lemma 3.4 is complete.
Proof of Lemma 3.5. For p = 1 , we have that f(u) = (1-u)^{-1} and F(u) = -\ln (1-u) . Let \lambda = \frac{q}{L} with q\in (0, 1] . Hence by (3.5), we compute that
\begin{eqnarray} J(q, r, s) &\equiv &L^{3}\Psi (\frac{q}{L}, r, s) \\ & = &q^{3}\left[ \ln (1-rs)-\ln (1-r)\right] ^{3}-3q^{2}L\left[ \ln (1-rs)-\ln (1-r)\right] ^{2} \\ &&+2qL^{2}\left[ \ln (1-rs)-\ln (1-r)\right] -qL^{2}\left[ \frac{r}{1-r}- \frac{rs}{1-rs}\right] . \end{eqnarray} | (3.26) |
Assume that M > 1 is a given number. Then, for 0 < r, s < 1 satisfying \frac{1 }{1-r}\leq M , by applying Cauchy's Mean Value Theorem, it is easy to check that
\begin{equation*} 0 < \frac{r}{1-r}-\frac{rs}{1-rs}\leq M\left[ \ln (1-rs)-\ln (1-r)\right] . \end{equation*} |
Therefore,
\begin{equation} J(q, r, s)\geq q\left[ \ln (1-rs)-\ln (1-r)\right] \tilde{J}(q, r, s), \end{equation} | (3.27) |
where
\begin{equation} \tilde{J}(q, r, s)\equiv q^{2}\left[ \ln (1-rs)-\ln (1-r)\right] ^{2}-3qL\left[ \ln (1-rs)-\ln (1-r)\right] +L^{2}\left( 2-M\right) . \end{equation} | (3.28) |
Note that 0 < \ln (1-rs)-\ln (1-r) < \ln \frac{1}{1-r}\leq \ln M . Let
\begin{equation} j_{M}(z)\equiv z^{2}-3Lz+L^{2}\left( 2-M\right) \text{.} \end{equation} | (3.29) |
We aim to find a number M_{0} > 1 such that j_{M_{0}}(z) > 0 for 0\leq z\leq \ln M_{0} . This implies that
\begin{equation*} \tilde{J}(q, r, s) > 0 \ \text{ for }0 < q\leq 1, \text{ }0 < r, s < 1 \end{equation*} |
with \frac{1}{1-r}\leq M_{0} . Since j_{M}^{\prime }(0) = -3L < 0 and j_{M} is convex on (0, \infty) , we only need to prove that j_{M_{0}}^{\prime }(\ln M_{0})\leq 0 and j_{M_{0}}(\ln M_{0})\geq 0 for some M_{0} > 1 . We have that
\begin{equation*} j_{M}^{\prime }(z) = 2z-3L \end{equation*} |
and
\begin{equation} j_{M_{0}}^{\prime }(\ln M_{0}) = 2\ln M_{0}-3L\leq 0 \ \text{ if and only if } \ M_{0}\leq e^{\frac{3}{2}L}. \end{equation} | (3.30) |
Thus we choose M_{0}\equiv \min \left \{ e^{\frac{L}{3}}, \frac{10}{9} \right \} . Then
\begin{eqnarray} j_{M_{0}}(\ln M_{0}) & = &(\ln M_{0})^{2}-3L(\ln M_{0})+L^{2}\left( 2-M_{0}\right) \\ & = &\left( \frac{8}{3}L-\ln M_{0}\right) \left( \frac{1}{3}L-\ln M_{0}\right) +L^{2}\left( \frac{10}{9}-M_{0}\right) \\ &\geq &L^{2}\left( \frac{10}{9}-M_{0}\right) \geq 0\text{.} \end{eqnarray} | (3.31) |
So j_{M_{0}}^{\prime }(\ln M_{0})\leq 0 and j_{M_{0}}(\ln M_{0})\geq 0 by (3.30) and (3.31).
Finally, we choose
\begin{equation*} r_{1}\equiv 1-\frac{1}{M_{0}} = \min \left \{ 1-e^{-\frac{L}{3}}, \frac{1}{10} \right \} . \end{equation*} |
Then we obtain that T_{1, \lambda }^{\prime }(r_{1}) > 0 for \lambda \in (0, 1/L] by (3.4), (3.5), and (3.26)–(3.31). Observe that r_{1}\in (0, F^{-1}(\frac{1}{\lambda })) for \lambda \in (0, 1/L] .
The proof of Lemma 3.5 is complete.
Proof of Lemma 3.6. For 0 < p < 1 , we have that f(u) = (1-u)^{-p} and F(u) = \frac{1-(1-u)^{1-p}}{1-p} . Let \lambda = \frac{q }{L} with q\in (0, 1] . Hence by (3.5), we compute that
\begin{eqnarray} H(q, r, s) &\equiv &\left( 1-p\right) ^{3}L^{3}\Psi (\frac{q}{L}, r, s) \\ & = &q^{3}\left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] ^{3}-3q^{2}\left( 1-p\right) L \left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] ^{2} \\ &&+2q\left( 1-p\right) ^{2}L^{2}\left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] \\ &&-q\left( 1-p\right) ^{3}L^{2}\left[ r(1-r)^{-p}-rs(1-rs)^{-p}\right] . \end{eqnarray} | (3.32) |
Assume that M > 1 is a given number. Then, for 0 < r, s < 1 satisfying \frac{1 }{1-r}\leq M , by applying Cauchy's Mean Value Theorem, it is easy to check that
\begin{equation*} 0 < r(1-r)^{-p}-rs(1-rs)^{-p}\leq \left( \frac{pM}{1-p}+1\right) \left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] . \end{equation*} |
Therefore,
\begin{equation} H(q, r, s)\geq q\left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] \tilde{H}(q, r, s), \end{equation} | (3.33) |
where
\begin{eqnarray} &&\tilde{H}(q, r, s) \\ &\equiv &q^{2}\left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] ^{2} \\ &&-3q\left( 1-p\right) L\left[ (1-rs)^{1-p}-(1-r)^{1-p}\right] +\left( 1-p\right) ^{2}L^{2}\left( p+1-pM\right) . \end{eqnarray} | (3.34) |
Note that 0 < (1-rs)^{1-p}-(1-r)^{1-p} < 1-(1-r)^{1-p}\leq 1-M^{p-1} . Let
\begin{equation} h_{M}(z)\equiv z^{2}-3\left( 1-p\right) Lz+\left( 1-p\right) ^{2}L^{2}\left( p+1-pM\right) \text{.} \end{equation} | (3.35) |
We aim to find a number M_{0} > 1 such that h_{M_{0}}(z) > 0 for 0\leq z\leq 1-M_{0}^{p-1} . This implies that
\begin{equation*} \tilde{H}(q, r, s) > 0 \ \text{ for }0 < q\leq 1, \text{ }0 < r, s < 1 \end{equation*} |
with \frac{1}{1-r}\leq M_{0} . Since h_{M}^{\prime }(0) = -3\left(1-p\right) L < 0 and h_{M} is convex on (0, \infty) , we only need to prove that h_{M_{0}}^{\prime }(1-M_{0}^{p-1})\leq 0 and h_{M_{0}}(1-M_{0}^{p-1})\geq 0 for some M_{0} > 1 . We have that
\begin{equation*} h_{M}^{\prime }(z) = 2z-3\left( 1-p\right) L \end{equation*} |
and
\begin{equation*} h_{M_{0}}^{\prime }\left( 1-M_{0}^{p-1}\right) = 2-2M_{0}^{p-1}-3\left( 1-p\right) L. \end{equation*} |
Thus we choose
\begin{equation*} M_{0}\equiv \min \left \{ \left[ 1-\frac{1}{k}\left( 1-p\right) L\right] ^{ \frac{1}{p-1}}, 1+\frac{1}{k^{2}p}\right \} \ \text{ with }k = \max \left \{ 3, (1-p)L\right \} . \end{equation*} |
Then
\begin{eqnarray} h_{M_{0}}\left( 1-M_{0}^{p-1}\right) & = &\left( 1-M_{0}^{p-1}\right) ^{2}-3\left( 1-p\right) L\left( 1-M_{0}^{p-1}\right) +\left( 1-p\right) ^{2}L^{2}\left( p+1-pM_{0}\right) \\ & = &\left[ \frac{1}{k}\left( 1-p\right) L-\left( 1-M_{0}^{p-1}\right) \right] \left[ (3-\frac{1}{k})\left( 1-p\right) L-\left( 1-M_{0}^{p-1}\right) \right] \\ &&+\left( 1-p\right) ^{2}L^{2}\left( \frac{1}{k^{2}}-\frac{3}{k} +p+1-pM_{0}\right) \\ &\geq &\left( 1-p\right) ^{2}L^{2}\left( \frac{1}{k^{2}}-\frac{3}{k} +p+1-pM_{0}\right) \\ &\geq &\left( 1-p\right) ^{2}L^{2}\left( \frac{1}{k^{2}}+p-pM_{0}\right) \geq 0 \end{eqnarray} | (3.36) |
and
\begin{equation} h_{M_{0}}^{\prime }(1-M_{0}^{p-1}) = 2-2M_{0}^{p-1}-3\left( 1-p\right) L\leq \left( \frac{2}{k}-3\right) \left( 1-p\right) L < 0. \end{equation} | (3.37) |
In (3.36) and (3.37), notice p-1 < 0 and hence M_{0}^{p-1}\geq 1-\frac{1}{k}\left(1-p\right) L .
Finally, we choose
\begin{equation*} r_{p}\equiv 1-\frac{1}{M_{0}} = \min \left \{ 1-\left[ 1-\frac{1}{k}\left( p-1\right) L\right] ^{\frac{1}{1-p}}, \frac{1}{k^{2}p+1}\right \} \ \text{ with }k = \max \left \{ 3, (1-p)L\right \} . \end{equation*} |
Then we obtain that T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] by (3.4), (3.5), and (3.32)–(3.37). Observe that r_{p}\in (0, F^{-1}(\frac{1}{\lambda })) for \lambda \in (0, 1/L] .
The proof of Lemma 3.6 is complete.
Proof of Theorem 2.1.
\rm(I) We prove Theorem 2.1(i). First, for p > 0, the upper bound \min \left \{ L^{-1}, \tfrac{p^{p}}{4(p+1)^{p+1}}\pi ^{2}L^{-2}\right \} for \lambda ^{\ast }(p, L) in (2.1) can be obtained by slightly modifying the proof of the upper bound for \lambda ^{\ast } for p = 2 in (1.7) in [4,Theorem 1.1]; we omit the proof. Also, it is easy to see that
\begin{equation*} \min \left \{ L^{-1}, \tfrac{p^{p}}{4(p+1)^{p+1}}\pi ^{2}L^{-2}\right \} \leq \min \left \{ L^{-1}, \tfrac{1}{4}\pi ^{2}L^{-2}\right \} \end{equation*} |
since \tfrac{p^{p}}{(p+1)^{p+1}} is a strictly decreasing function of p > 0 and \lim_{p\rightarrow 0^{+}}\tfrac{p^{p}}{(p+1)^{p+1}} = 1 .
We then prove the lower bound for \lambda ^{\ast } in (2.1) by modifying the proof of Wang and Ruan [25,Ineq. (2.10)] and by Pan and Xing [12,Theorem 3.1]. We first take the function
\begin{equation*} w(x)\equiv \frac{1}{p+1}\left( 1-\frac{x^{2}}{L^{2}}\right) < 1 \ \text{ for } x\in (-L, L), \end{equation*} |
which satisfies w(x) > 0 on (-L, L) and w(\pm L) = 0 . We then compute that, for x\in (-L, L),
\begin{eqnarray*} -w^{\prime \prime }(x) & = &\frac{2}{(p+1)L^{2}} = \frac{2\left[ 1-\frac{1}{(p+1) }\right] ^{p}}{(p+1)L^{2}}\frac{1}{\left[ 1-\frac{1}{(p+1)}\right] ^{p}} \\ &\geq &\frac{2p^{p}}{(p+1)^{p+1}L^{2}}\frac{1}{\left[ 1-\frac{1}{(p+1)} \left( 1-\frac{x^{2}}{L^{2}}\right) \right] ^{p}} \\ & = &\frac{2p^{p}}{(p+1)^{p+1}L^{2}}\frac{1}{\left[ 1-w(x)\right] ^{p}} \\ & = &\frac{2p^{p}}{(p+1)^{p+1}L^{2}}\frac{\left[ 1+(w^{\prime }(x))^{2}\right] ^{3/2}}{\left[ 1-w(x)\right] ^{p}}\frac{1}{\left[ 1+(w^{\prime }(x))^{2} \right] ^{3/2}} \\ & = &\frac{2p^{p}}{(p+1)^{p+1}L^{2}}\frac{\left[ 1+(w^{\prime }(x))^{2}\right] ^{3/2}}{\left[ 1-w(x)\right] ^{p}}\frac{1}{\left[ 1+\frac{4x^{2}}{ (p+1)^{2}L^{4}}\right] ^{3/2}} \\ &\geq &\frac{2p^{p}}{(p+1)^{p+1}L^{2}}\frac{\left[ 1+(w^{\prime }(x))^{2} \right] ^{3/2}}{\left[ 1-w(x)\right] ^{p}}\frac{1}{\left[ 1+\frac{4L^{2}}{ (p+1)^{2}L^{4}}\right] ^{3/2}} \\ & = &\frac{2p^{p}}{(p+1)^{p+1}L^{2}\left[ 1+\frac{4}{(p+1)^{2}L^{2}}\right] ^{3/2}}\frac{\left[ 1+(w^{\prime }(x))^{2}\right] ^{3/2}}{\left[ 1-w(x) \right] ^{p}}. \end{eqnarray*} |
So for
\begin{equation*} \lambda = \frac{2p^{p}}{(p+1)^{p+1}L^{2}\left[ 1+\frac{4}{(p+1)^{2}L^{2}} \right] ^{3/2}}, \end{equation*} |
w(x) is a supersolution of (1.1) on (-L, L) as (1.1) can be written in the equivalent form
\begin{equation*} -u^{\prime \prime }(x) = \lambda \frac{\left[ 1+(u^{\prime }(x))^{2}\right] ^{3/2}}{\left[ 1-u(x)\right] ^{p}} \ \text{ on }(-L, L); \end{equation*} |
see (1.4) with \tilde{f}(u) = \frac{1}{(1-u)^{p}} . Since w_{0}(x)\equiv 0 is a subsolution of (1.1) on (-L, L) and f(0) = 1 > w(x) > w_{0}(x) = 0 on (-L, L) , by applying Pan and Xing [12,Theorem 3.1] obtained by the lower and upper solution method, there exists a (classical) solution \tilde{w}(x)\in C^{2}[-L, L] of (1.1) satisfying 0 < \tilde{w}(x)\leq w(x) on (-L, L) . This proves that
\begin{equation*} \frac{2p^{p}}{(p+1)^{p+1}L^{2}\left[ 1+\frac{4}{(p+1)^{2}L^{2}}\right] ^{3/2} }\leq \lambda ^{\ast }(p, L). \end{equation*} |
\rm(II) We prove Theorem 2.1(ii). Consider L > 0 be fixed. For any fixed \lambda > 0 , r\in I and 0 < u < r , in T_{p, \lambda }(r) in (3.1), the integrand \frac{1+\lambda F(u)-\lambda F(r)}{\sqrt{1-\left[1+\lambda F(u)-\lambda F(r)\right] ^{2}}} is strictly decreasing in p > 0 since
\begin{equation*} F(r)-F(u) = \int_{u}^{r}\frac{1}{(1-t)^{p}}dt \end{equation*} |
is increasing in p > 0 . So T_{p, \lambda }(r) is a strictly decreasing function of p > 0 . Hence \lambda ^{\ast }(p, L) is a strictly decreasing function of p > 0 . The rest of part (ii) follow from part (i) and by simple calculus with the fact that \lim_{p\rightarrow 0^{+}}\tfrac{p^{p}}{ (p+1)^{p+1}} = 1 and \lim_{p\rightarrow \infty }\tfrac{p^{p}}{(p+1)^{p+1}} = 0 .
\rm(III) We prove Theorem 2.1(iii). Consider fixed p > 0 . Let
\begin{equation*} h_{p}(\lambda )\equiv \left \{ \begin{array}{ll} \sup \left \{ T_{p, \lambda }(r):r\in (0, F^{-1}(\frac{1}{\lambda })]\right \} , & \text{if }(p\geq 1\text{, }\lambda > 0)\text{ or }(0 < p < 1\text{, }\lambda > 1-p)\text{, } \\ \sup \left \{ T_{p, \lambda }(r):r\in (0, 1)\right \} & \text{if }0 < p < 1\text{, }0 < \lambda \leq 1-p\text{.} \end{array} \right. \end{equation*} |
See [1,(1.9) and (1.11)]. By [1,Lemma 3.3(ii)], h_{p}(\lambda) is a continuous, strictly decreasing function of \lambda > 0 , \lim_{\lambda \rightarrow 0^{+}}h_{p}(\lambda) = \infty and \lim_{\lambda \rightarrow \infty }h_{p}(\lambda) = 0 . Thus we obtain that
\begin{equation} h_{p}(\lambda ^{\ast }(p, L)) = L \ \text{ for all }L > 0; \end{equation} | (4.1) |
see [1,Proofs of Theorems 2.1 and 2.2]. Moreover, we obtain that \lambda ^{\ast }(p, L) is a strictly decreasing function of L > 0 , \lim_{L\rightarrow 0^{+}}\lambda ^{\ast }(p, L) = \infty and \lim_{L\rightarrow \infty }\lambda ^{\ast }(p, L) = 0 .
The proof of Theorem 2.1 is complete.
Proof of Theorem 2.3.
In (3.3), positive solutions u_{\lambda } for (1.1) correspond to
\begin{equation*} \left \Vert u_{\lambda }\right \Vert _{\infty } = r\ \ \text{and }\ T_{p, \lambda }(r) = L. \end{equation*} |
Thus, studying of the exact number of positive solutions of (1.1) for any fixed \lambda > 0 is equivalent to studying the shape of the time map T_{p, \lambda }(r) on its domain I . Moreover, T_{p, \lambda }(r) has exactly one critical point, a local maximum, on its domain I by Lemma 3.1. Hence we have that:
\rm(i) If there exists r^{\ast } > 0 such that T_{p, \lambda }^{\prime }(r^{\ast }) < 0 for \lambda \in (0, 1/F(r^{\ast })) , then \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < r^{\ast } and hence r^{\ast } is an upper bound of \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } .
\rm(ii) If there exists r_{\ast } > 0 such that T_{p, \lambda }^{\prime }(r_{\ast }) > 0 for \lambda \in (0, 1/L] , then \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } > r_{\ast } and hence r_{\ast } is a lower bound of \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } . Observe \lambda ^{\ast }(p, L) < 1/L by Theorem 2.1(i).
We are now in a position to prove Theorem 2.3(i)–(iii).
\rm(I) We prove the upper bounds for \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } in Theorem 2.3(i)–(iii).
(A) For p > 1 and r = \bar{U}(p) , by Lemma 3.2, we have that T_{p, \lambda }^{\prime }(r) < 0 for \lambda \in (0, 1/F(r)) , and hence
\begin{equation*} \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < r = \bar{U}(p)\text{ (} \leq U(p)\text{)} \end{equation*} |
for L > 0 by applying (3.3) and Lemma 3.1. See also Theorem 1.1. It is easy to check that \bar{U}(p) is bounded above by U(p) in (2.3) for p > 1 , we omit the proof.
(B) For p = 1 , similarly, we have that \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < 1-2e^{-2}\approx 0.729 for L > 0 by applying (3.3), Lemmas 3.1 and 3.3. See also Theorem 1.1.
(C) For 0 < p < 1, it is trivial that \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } < 1 for L > 0 .
By above (A)–(C), we obtain the upper bounds for \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } in Theorem 2.3(i)–(iii).
\rm(II) We prove the lower bounds for \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } in Theorem 2.3(i)–(iii).
(A) For p > 1 and
\begin{equation*} r_{p}\equiv \bar{L}(p, L)\equiv \min \left \{ 1-\left[ 1+\frac{1}{3}\left( p-1\right) L\right] ^{\frac{1}{1-p}}, \frac{1}{9p+1}\right \} , \end{equation*} |
by Lemma 3.4, we have that T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] . Hence
\begin{equation*} \bar{L}(p, L) < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } \end{equation*} |
for L > 0 by applying (3.3) and Lemma 3.1. See also Theorem 1.1.
(B) For p = 1 and
\begin{equation*} r_{1}\equiv \min \left \{ 1-e^{-\frac{L}{3}}, \frac{1}{10}\right \} , \end{equation*} |
by Lemma 3.5, we have that T_{1, \lambda }^{\prime }(r_{1}) > 0 for \lambda \in (0, 1/L] . Hence
\begin{equation*} \min \left \{ 1-e^{-\frac{L}{3}}, \frac{1}{10}\right \} < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } \end{equation*} |
for L > 0 by applying (3.3) and Lemma 3.1. See also Theorem 1.1.
(C) For 0 < p < 1 and
\begin{equation*} r_{p}\equiv \hat{L}(p, L) = \min \left \{ 1-\left[ 1-\frac{1}{k}\left( 1-p\right) L\right] ^{\frac{1}{1-p}}, \frac{1}{k^{2}p+1}\right \} , \text{ } \ k = \max \left \{ 3, (1-p)L\right \} , \end{equation*} |
by Lemma 3.6, we have that T_{p, \lambda }^{\prime }(r_{p}) > 0 for \lambda \in (0, 1/L] . Hence
\begin{equation*} \min \left \{ 1-\left[ 1-\frac{1}{k}\left( 1-p\right) L\right] ^{\frac{1}{1-p }}, \frac{1}{k^{2}p+1}\right \} < \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } \end{equation*} |
for L > 0 by applying (3.3) and Lemma 3.1. See also Theorem 1.2.
By above (A)–(C) in this part (II), we obtain the lower bounds for \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } in Theorem 2.3(i)–(iii).
\rm(III) We prove Theorem 2.3(iv). Assertion (2.5) follows immediately by (2.2)–(2.3) and since
\begin{equation*} \lim\limits_{p\rightarrow \infty }\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }\leq \lim\limits_{p\rightarrow \infty }U(p) = \lim\limits_{p\rightarrow \infty } \frac{2}{p+2} = 0. \end{equation*} |
In addition, (2.6) and (2.7) follow easily by (2.4).
We then prove (2.8). For any fixed L > 0 , by (4.1), we obtain that h_{p}(\lambda ^{\ast }) = h_{p}(\lambda ^{\ast }(L)) = L . Since \left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }\in (0, F^{-1}(\frac{1 }{\lambda ^{\ast }})) = (0, F^{-1}(\frac{1}{\lambda ^{\ast }(L)})) , we have that
\begin{eqnarray*} 0 &\leq &\lim\limits_{L\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty }\leq \lim\limits_{L\rightarrow 0^{+}}F^{-1}(\frac{1}{\lambda ^{\ast }(L)}) \\ & = &F^{-1}\left( \lim\limits_{L\rightarrow 0^{+}}\frac{1}{\lambda ^{\ast }(L)}\right) \\ & = &F^{-1}(0) \ \text{ (since }\lim \nolimits_{L\rightarrow 0^{+}}\lambda ^{\ast }(L) = \infty \text{ by Theorem 2.1(iii))} \\ & = &0\text{.} \end{eqnarray*} |
Thus \lim_{L\rightarrow 0^{+}}\left \Vert u_{\lambda ^{\ast }}\right \Vert _{\infty } = 0 . So (2.8) holds.
The proof of Theorem 2.3 is complete.
The authors thank the referees for constructive comments and recommendations which will help to improve the readability and quality of the paper. Most of the computation in this paper has been checked using the symbolic manipulator Mathematica 11.0. This work was partially supported by the Ministry of Science and Technology of the Republic of China under grant No. MOST 110-2115-M-167-001-MY2.
The authors declare there is no conflict of interest.
[1] |
Y. H. Cheng, K. C. Hung, S. H. Wang, Global bifurcation diagrams and exact multiplicity of positive solutions for a one-dimensional prescribed mean curvature problem arising in MEMS, Nonlinear Anal. Theory Methods Appl., 89 (2013), 284–298. https://doi.org/10.1016/j.na.2012.04.025 doi: 10.1016/j.na.2012.04.025
![]() |
[2] |
D. Bonheure, P. Habets, F. Obersnel, P. Omari, Classical and non-classical solutions of a prescribed curvature equation, J. Differ. Equ., 243 (2007), 208–237. https://doi.org/10.1016/j.jde.2007.05.031 doi: 10.1016/j.jde.2007.05.031
![]() |
[3] |
N. D. Brubaker, J. A. Pelesko, Non-linear effects on canonical MEMS models, Eur. J. Appl. Math., 22 (2011), 455–470. https://doi.org/10.1017/S0956792511000180 doi: 10.1017/S0956792511000180
![]() |
[4] |
N. D. Brubaker, J. A. Pelesko, Analysis of a one-dimensional prescribed mean curvature equation with singular nonlinearity, Nonlinear Anal. Theory Methods Appl., 75 (2012), 5086–5102. https://doi.org/10.1016/j.na.2012.04.025 doi: 10.1016/j.na.2012.04.025
![]() |
[5] |
F. Obersnel, P. Omari, Positive solutions of the Dirichlet problem for the prescribed mean curvature equation, J. Differ. Equ., 249 (2010), 1674–1725. https://doi.org/10.1016/j.jde.2010.07.001 doi: 10.1016/j.jde.2010.07.001
![]() |
[6] |
H. Pan, One-dimensional prescribed mean curvature equation with exponential nonlinearity, Nonlinear Anal. Theory Methods Appl., 70 (2009), 999–1010. https://doi.org/10.1016/j.na.2008.01.027 doi: 10.1016/j.na.2008.01.027
![]() |
[7] |
H. Pan, R. Xing, Time maps and exact multiplicity results for one-dimensional prescribed mean curvature equations, Nonlinear Anal. Theory Methods Appl., 74 (2011), 1234–1260. https://doi.org/10.1016/j.na.2010.09.063 doi: 10.1016/j.na.2010.09.063
![]() |
[8] |
H. Pan, R. Xing, Time maps and exact multiplicity results for one-dimensional prescribed mean curvature equations, II, Nonlinear Anal. Theory Methods Appl., 74 (2011), 3751–3768. https://doi.org/10.1016/j.na.2011.03.020 doi: 10.1016/j.na.2011.03.020
![]() |
[9] |
H. Pan, R. Xing, A note on the nonexistence of solutions for prescribed mean curvature equations on a ball, Nonlinear Anal. Theory Methods Appl., 74 (2011), 7437–7445. https://doi.org/10.1016/j.na.2011.07.063 doi: 10.1016/j.na.2011.07.063
![]() |
[10] |
H. Pan, R. Xing, Radial solutions for a prescribed mean curvature equation with exponential nonlinearity, Nonlinear Anal. Theory Methods Appl., 75 (2012), 103–116. https://doi.org/10.1016/j.na.2011.08.010 doi: 10.1016/j.na.2011.08.010
![]() |
[11] |
H. Pan, R. Xing, Exact multiplicity results for a one-dimensional prescribed mean curvature problem related to a MEMS model, Nonlinear Anal. Real World Appl., 13 (2012), 2432–2445. https://doi.org/10.1016/j.nonrwa.2012.02.012 doi: 10.1016/j.nonrwa.2012.02.012
![]() |
[12] |
H. Pan, R. Xing, Sub- and supersolution methods for prescribed mean curvature equations with Dirichlet boundary conditions, J. Differ. Equ., 254 (2013), 1464–1499. https://doi.org/10.1016/j.jde.2012.10.025 doi: 10.1016/j.jde.2012.10.025
![]() |
[13] |
H. Pan, R. Xing, On the existence of positive solutions for some nonlinear boundary value problems and applications to MEMS models, Discrete Contin. Dyn. Syst., 35 (2015), 3627–3682. https://doi.org/10.3934/dcds.2015.35.3627 doi: 10.3934/dcds.2015.35.3627
![]() |
[14] |
H. Pan, R. Xing, Applications of total positivity theory to 1D prescribed curvature problems, J. Math. Anal. Appl., 428 (2015), 113–144. https://doi.org/10.1016/j.jmaa.2015.03.002 doi: 10.1016/j.jmaa.2015.03.002
![]() |
[15] |
H. Pan, R. Xing, Bifurcation results for a class of prescribed mean curvature equations in bounded domains, Nonlinear Anal. Theory Methods Appl., 171 (2018), 21–31. https://doi.org/10.1016/j.na.2018.01.010 doi: 10.1016/j.na.2018.01.010
![]() |
[16] |
F. Lin, Y. Yang, Nonlinear non-local elliptic equation modelling electrostatic actuation, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci., 463 (2007), 1323–1337. https://doi.org/10.1098/rspa.2007.1816 doi: 10.1098/rspa.2007.1816
![]() |
[17] |
N. D. Brubaker, J. I. Siddique, E. Sabo, R. Deaton, J. A. Pelesko, Refinements to the study of electrostatic deflections: theory and experiment, Eur. J. Appl. Math., 24 (2013), 343–370. https://doi.org/10.1017/S0956792512000435 doi: 10.1017/S0956792512000435
![]() |
[18] |
M. Mazars, Ewald methods for inverse power-law interactions in tridimensional and quasi-two-dimensional systems, J. Phys. A: Math. Theor., 43 (2010), 425002 (16 pp). https://doi.org/10.1088/1751-8113/43/42/425002 doi: 10.1088/1751-8113/43/42/425002
![]() |
[19] |
H. A. Levine, Quenching, nonquenching, and beyond quenching for solution of some parabolic equations, Ann. Mat. Pura Appl., 155 (1989), 243–260. https://doi.org/10.1007/BF01765943 doi: 10.1007/BF01765943
![]() |
[20] |
C. Cowan, N. Ghoussoub, Estimates on pull-in distances in microelectromechanical systems models and other nonlinear eigenvalue problems, SIAM J. Math. Anal., 42 (2010), 1949–1966. https://doi.org/10.1137/090752857 doi: 10.1137/090752857
![]() |
[21] |
G. Flores, G. Mercado, J. A. Pelesko, N. Smyth, Analysis of the dynamics and touchdown in a model of electrostatic MEMS, SIAM J. Appl. Math., 67 (2006), 434–446. https://doi.org/10.1137/060648866 doi: 10.1137/060648866
![]() |
[22] |
N. Ghoussoub, Y. Guo, On the partial differential equations of electrostatic MEMS devices: stationary case, SIAM J. Math. Anal., 5 (2007), 1423–1449. https://doi.org/10.1137/050647803 doi: 10.1137/050647803
![]() |
[23] |
T. Laetsch, The number of solutions of a nonlinear two point boundary value problem, Indiana Univ. Math. J., 20 (1970), 1–13. https://doi.org/10.1512/iumj.1971.20.20001 doi: 10.1512/iumj.1971.20.20001
![]() |
[24] |
D. Bonheure, A. E. Lindsay, The onset of multi-valued solutions of a prescribed mean curvature equation with singular non-linearity, Eur. J. Appl. Math., 24 (2013), 631–656. https://doi.org/10.1017/S0956792513000077 doi: 10.1017/S0956792513000077
![]() |
[25] |
Z. Wang, L. Ruan, On a class of semilinear elliptic problems with singular nonlinearities, Appl. Math. Comput., 193 (2007), 89–105. https://doi.org/10.1016/j.amc.2007.03.056 doi: 10.1016/j.amc.2007.03.056
![]() |
1. | JI-HUAN HE, QIAN YANG, CHUN-HUI HE, HAI-BIN LI, EERDUN BUHE, PULL-IN STABILITY OF A FRACTAL MEMS SYSTEM AND ITS PULL-IN PLATEAU, 2022, 30, 0218-348X, 10.1142/S0218348X22501857 |