数学物理学报  2017, Vol. 37 Issue (2): 287-298   PDF    
扩展功能
加入收藏夹
复制引文信息
加入引用管理器
Email Alert
RSS
本文作者相关文章
郝新安
刘立山
含参数四阶微分方程非局部边值问题的正解
郝新安, 刘立山     
曲阜师范大学数学科学学院 山东曲阜 273165
摘要:通过构造特殊的锥并利用锥中的Krasnosel'skii-Zabreiko不动点定理,该文研究了含有两个参数的四阶微分方程广义Sturm-Liouville边值问题正解的存在性,推广和改进了一些已知的结果.
关键词正解    积分边界条件    不动点        
Positive Solutions for Nonlocal Boundary Value Problems of Fourth Order Differential Equation with Parameters
Hao Xinan, Liu Lishan     
School of Mathematical Sciences, Qufu Normal University, Shandong Qufu 273165
Abstract: This paper deals with the existence of positive solutions to fourth order nonlocal boundary value problems with two parameters. The proofs are based on a specially constructed cone and a fixed point theorem in a cone for a completely continuous operator, due to Krasnosel'skii and Zabreiko. The results extend and improve some known results.
Key words: Positive solution     Integral boundary conditions     Fixed point     Cone    
1 引言

本文研究含有两个参数的四阶微分方程

$ {{x}^{(4)}}+\beta {x}''-\alpha x=f(t, x), \quad 0<t<1 $ (1.1)

在广义的Sturm-Liouville积分边界条件

$ \begin{equation}\label{eq:12}\left\{ \begin{array} {ll} \displaystyle ax(0)-bx'(0)=\int_0^1g(s)x(s){\rm d}s, & cx(1)+dx'(1)=\int_0^1h(s)x(s){\rm d}s, \\ \displaystyle ax''(0)-bx'''(0)=\int_0^1g(s)x''(s){\rm d}s, \ & cx''(1)+dx'''(1)=\int_0^1h(s)x''(s){\rm d}s \end{array}\right.\end{equation} $ (1.2)

下正解的存在性, 其中常数$ a, b, c, d\geq 0, \ ac+ad+bc>0$, 函数$g, h\in L^1[0, 1]$且非负. $\alpha, \ \beta, \ f$满足下列条件:

$(H_1)$ $f: [0, 1]\times [0, +\infty)\rightarrow [0, +\infty)$连续;

$(H_2)$ $\beta \leq 0, -\frac{\beta^2}{4}\leq \alpha\leq 0$.

四阶微分方程 (1.1) 来源于应用数学、工程学和物理学的许多分支[1-5], 它主要用来描述弹性梁的形变.例如, 在Lidstone边界条件$x(0)=x(1)=x''(0)=x''(1)=0$下, 方程 (1.1) 用来模拟两端简单支撑的弹性梁的形变现象[1, 3-4].在focal边界条件下, 方程 (1.1) 用来描述弹性梁两端固定或一段简单支撑另一端由滑动夹具夹住的形变现象.对Lidstone边界条件, 文献[6]建立了方程 (1.1) 正解的存在性.对于多点边界条件

$ x^{(2i)}(0)=\sum_{i=1}^{m-2}a_ix^{(2i)}(\xi_i), \quad x^{(2i)}(1)=\sum_{i=1}^{m-2}b_ix^{(2i)}(\xi_i), \quad i=0, 1, $

文献[7]利用锥中的不动点定理得到了方程 (1.1) 正解的存在性, 并对一类特殊的非线性项作者获得了唯一性结果.文献[8]利用锥拉伸压缩不动点定理得到了四阶微分方程四点边值问题正解存在的一些充分条件.

带有积分边界条件的微分方程边值问题起源于各种物理、生物和化学现象[9-10], 比如热传导、化学工程和等离子体物理学.积分边值问题是近年来数学中一类重要而有趣的问题, 它包含两点、三点、多点和非局部边值问题.近二十年来, 非线性微分方程积分边值问题正解的存在性受到了广泛关注, 见文献[11-19]及其参考文献.文献[11]应用Krasnosel'skii不动点定理研究了边值问题 (1.3) 一个和两个正解的存在性

$ \begin{equation}\label{eq:13}\left\{ \begin{array} {ll} \displaystyle x^{(4)}(t)+\beta x''(t)=\lambda f(t, x(t), x''(t)), \quad 0<t<1, \\[2mm] \displaystyle x(0)= x(1)=\int_0^1p(s)x(s){\rm d}s, \quad x''(0)= x''(1)=\int_0^1q(s)x''(s){\rm d}s, \end{array}\right.\end{equation} $ (1.3)

其中$p, q\in L^1 [0, 1], \ \lambda>0, \ f\in C([0, 1]\times [0, +\infty)\times (-\infty, 0], [0, +\infty))$.当$\lambda=1$, 非线性项$f(t, x, y)$关于$x, y$满足一定的单调性条件时, 文献[12]应用单调迭代方法获得了四阶非局部边值问题 (1.3) 对称正解的存在性.文献[19]应用变换技巧和不动点定理研究了四阶微分方程积分边值问题 (1.4) 正解的存在性和多重性

$ \begin{equation}\label{eq:14}\left\{ \begin{array} {lll} \displaystyle x^{(4)}(t)=\omega(t) f(t, x(t), x''(t)), \quad 0<t<1, \\[2mm] \displaystyle x(0)= x(1)=\int_0^1h(s)x(s){\rm d}s, \\[3mm] \displaystyle ax''(0)-bx'''(0)=\int_0^1g(s)x''(s){\rm d}s, \quad ax''(1)+bx'''(1)=\int_0^1g(s)x''(s){\rm d}s.\end{array}\right.\end{equation} $ (1.4)

文献[14]研究了下列四阶微分方程积分边值问题

$ \left\{ \begin{array} {ll} \displaystyle x^{(4)}(t)=h(t)f(t, x), \quad 0<t<1, \\[2mm] \displaystyle x(0)= x(1)=\int_0^1p(s)x(s){\rm d}s, \quad x''(0)= x''(1)=\int_0^1q(s)x(s){\rm d}s, \end{array}\right. $

当非线性项满足条件

$ \mathop {\lim \inf }\limits_{x \to {0^ + }} \mathop {\min }\limits_{t \in [0, 1]} \frac{{f(t, x)}}{x} > {\lambda _*}, \quad \mathop {\lim \sup }\limits_{x \to + \infty } \mathop {\max }\limits_{t \in [0, 1]} \frac{{f(t, x)}}{x} < {\lambda _*}, $

应用相应线性算子的第一特征值和不动点指数理论得到了至少一个对称正解的存在性结果, 其中$\lambda_*$是相应线性算子的第一特征值.但对于另一种情况

$ \mathop {\lim \inf }\limits_{x \to {0^ + }} \mathop {\max }\limits_{t \in [0, 1]} \frac{{f(t, x)}}{x} < {\lambda _*}, \quad \mathop {\lim \inf }\limits_{x \to + \infty } \mathop {\min }\limits_{t \in [0, 1]} \frac{{f(t, x)}}{x} > {\lambda _*}, $

作者没有得到类似的结论.

受以上论文的启发, 我们研究四阶微分方程广义Sturm-Liouville积分边值问题 (1.1) 和 (1.2) 正解的存在性.我们在更广泛的积分边界条件下研究四阶微分方程, 使用的条件是较弱的相应线性算子的第一特征值, 而且我们的方法不同于文献[6-19]中的方法, 所得主要结果在一定程度上改进和包含了一些已知的结论.此外, 本文边界条件 (1.2) 中的Riemann积分可以替换为Riemann-Stieltjes积分.

本文的主要工具是锥中的Krasnosel'skii-Zabreiko不动点定理[20].

引理1.1   设$E$是实Banach空间, $W$$E$中的锥.算子$A:(\overline{B}_R\backslash B_r)\cap W \rightarrow W$全连续, 其中$0<r<R, \ B_\rho=\{x\in E:\|x\|<\rho\}$.若下列两个条件之一满足:

(1) $\forall x\in \partial B_r\cap W, \ Ax\not\leq x;\ \forall x\in \partial B_R\cap W, \ Ax\not\geq x$.

(2) $\forall x\in \partial B_r\cap W, \ Ax\not\geq x;\ \forall x\in \partial B_R\cap W, \ Ax\not\leq x$.

$A$$(B_R\backslash \overline{B}_r)\cap W$中至少有一个不动点.

2 预备知识及引理

$ E = C[0, 1], \quad \left\| x \right\| = \mathop {\max }\limits_{t \in [0, 1]} |x(t)|, \quad P = \{ x \in E:x(t) \ge 0, \;\forall t \in [0, 1]\}, $

$(E, \|\cdot\|)$是Banach空间, $P$$E$中的锥.边值问题 (1.1) 和 (1.2) 的正解$x$是指函数$x$$(0, 1)$上是正的且满足微分方程 (1.1) 和边界条件 (1.2).

引理2.1  假设$(H_2)$成立, 则存在唯一的$\varphi_i, \ \psi_i \ (i=1, 2)$分别满足下列边值问题

$ \left\{ \begin{array} {ll} \displaystyle-\varphi_i''+\lambda_i\varphi_i=0, \\ \displaystyle \varphi_i(0)=b, \quad \varphi'_i(0)=a, \end{array}\right. \quad\quad \left\{ \begin{array} {ll} \displaystyle-\psi_i''+\lambda_i\psi_i=0, \\ \displaystyle \psi_i(1)=d, \quad \psi'_i(1)=-c, \end{array}\right. $

$\varphi_i(t)\geq 0, \ \psi_i(t)\geq 0, \ i=1, 2, \ t\in [0, 1]$, 其中$\lambda_1, \ \lambda_2$是多项式$\lambda^2+\beta\lambda-\alpha=0$的根, 即

$ \lambda_1=\frac{-\beta+\sqrt{\beta^2+4\alpha}}{2}, \quad \lambda_2=\frac{-\beta-\sqrt{\beta^2+4\alpha}}{2}. $

此外, $\varphi_1, \ \varphi_2$$[0, 1]$上非减, $\psi_1, \ \psi_2$$[0, 1]$上非增.

  由$(H_2)$$\lambda_1\geq \lambda_2\geq 0$.记$w_i=\sqrt{\lambda_i}, \ i=1, 2$.通过计算可知, 若$\lambda_i>0$, 则

$ \varphi_i(t)=b \cosh w_it+\frac{a}{w_i}\sinh w_i t, \quad \psi_i(t)=d \cosh w_i(1-t)+\frac{c}{w_i}\sinh w_i (1-t); $

$\lambda_i=0$, 则

$ \varphi_i(t)=b+at, \quad \psi_i(t)=d +c-ct. $

显然, 在$[0, 1]$$\varphi_i(t)\geq 0, \ \psi_i(t)\geq 0, \ (i=1, 2), \ \varphi_1, \ \varphi_2$非减, $\psi_1, \ \psi_2$非增.

$i=1, 2$, 记

$ \begin{matrix} {{\rho }_{i}}=\left| \begin{matrix} {{\psi }_{i}}(0)\quad {{\varphi }_{i}}(0) \\ {{\psi }_{{{i}'}}}(0)\quad {{\varphi }_{{{i}'}}}(0) \\ \end{matrix} \right|, \quad {{\Delta }_{i}}=\left| \begin{matrix} \int_{0}^{1}{g}(s){{\varphi }_{i}}(s)\text{d}s~~ & \int_{0}^{1}{g}(s){{\psi }_{i}}(s)\text{d}s-{{\rho }_{i}} \\ \int_{0}^{1}{h}(s){{\varphi }_{i}}(s)\text{d}s-{{\rho }_{i}}~~ & \int_{0}^{1}{h}(s){{\psi }_{i}}(s)\text{d}s \\ \end{matrix} \right|, \\ {{G}_{i}}(t, s)=\frac{1}{{{\rho }_{i}}}\left\{ \begin{array}{*{35}{l}} {{\varphi }_{i}}(t){{\psi }_{i}}(s), \quad 0\le t\le s\le 1, \\ {{\varphi }_{i}}(s){{\psi }_{i}}(t), \quad 0\le s\le t\le 1, \\ \end{array} \right. \\ \end{matrix} $ (2.1)

$\rho_i>0, \ G_i(t, s)$是下列线性微分方程边值问题的Green函数

$ \left\{ \begin{array} {ll} \displaystyle-x''(t)+\lambda_ix(t)=0, \\ \displaystyle ax(0)-bx'(0)=cx(1)+dx'(1)=0. \end{array}\right. $

引理2.2  $G_i(t, s), \ \varphi_i(t)$$\psi_i(t)(i=1, 2)$有如下性质:

(ⅰ) $G_i(t, s)=G_i(s, t)>0, \ t, s\in (0, 1)$;

(ⅱ) $q_i(t)G_i(s, s)\leq G_i(t, s)\leq G_i(s, s), \quad t, s\in [0, 1], $其中

$ q_i(t)=\min\left\{\frac{\varphi_i(t)}{\|\varphi_i\|}, \ \frac{\psi_i(t)}{\|\psi_i\|}\right\}, \quad t\in [0, 1]. $

给出本文假设条件:

$(A_i)\quad \int_0^1g(s)\psi_i(s){\rm d}s<\rho_i, \ \int_0^1h(s)\varphi_i(s){\rm d}s<\rho_i, \ i=1, 2$.

$(H_{3i})\quad \Delta_i<0, \ i=1, 2$.

引理2.3  假设条件$(A_i), \ (H_2)$$(H_{3i})$成立, 则对任意的$y\in C[0, 1]$, 边值问题

$ \begin{equation}\label{eq:22}\left\{ \begin{array} {ll} \displaystyle -x''+\lambda_i x=y(t), \quad 0<t<1, \\[2mm] \displaystyle ax(0)-bx'(0)=\int_0^1g(s)x(s){\rm d}s, \ \ cx(1)+dx'(1)=\int_0^1h(s)x(s){\rm d}s \end{array}\right.\end{equation} $ (2.2)

有唯一解

$ x_i(t)=\int_0^1G_i(t, s)y(s){\rm d}s+A_i(y)\varphi_i(t)+B_i(y)\psi_i(t), \quad t\in [0, 1], $

其中

$ \begin{equation}\label{eq:23}A_i(y)=-\frac{1}{\Delta_i}\left|\begin{array}{cc} \int_0^1g(s)\int_0^1G_i(s, \tau)y(\tau){\rm d}\tau{\rm d}s~~ &\int_0^1g(s)\psi_i(s){\rm d}s-\rho_i \\[3mm] \int_0^1h(s)\int_0^1G_i(s, \tau)y(\tau){\rm d}\tau{\rm d}s~~ & \int_0^1h(s)\psi_i(s){\rm d}s \\ \end{array}\right|, \end{equation} $ (2.3)
$ \begin{equation}\label{eq:24}B_i(y)=-\frac{1}{\Delta_i}\left|\begin{array}{cc} \int_0^1g(s)\varphi_i(s){\rm d}s ~~& \int_0^1g(s)\int_0^1G_i(s, \tau)y(\tau){\rm d}\tau{\rm d}s \\[3mm] \int_0^1h(s)\varphi_i(s){\rm d}s-\rho_i~~& \int_0^1h(s)\int_0^1G_i(s, \tau)y(\tau){\rm d}\tau{\rm d}s \\ \end{array}\right|.\end{equation} $ (2.4)

此外, 若$y\geq 0$, 则边值问题 (2.2) 的唯一解$x_i$满足$x_i(t)\geq 0, \ t\in [0, 1], \ i=1, 2$.

  由常规计算直接可证此引理成立.

注意到

$ x^{(4)}(t)+ \beta x''(t)-\alpha x(t)=\left(-\frac{{\rm d}^2}{{\rm d}t^2}+\lambda_1\right)\left(-\frac{{\rm d}^2}{{\rm d}t^2}+\lambda_2\right)x(t), $

易知下面的引理2.4成立.

引理2.4  假设$(H_2), \ (H_{31})$$(H_{32})$成立, 则对任意的$y\in C[0, 1]$, 边值问题

$ \left\{ \begin{array} {lll} \displaystyle x^{(4)}+\beta x''-\alpha x=y(t), \quad 0<t<1, \\[2mm] \displaystyle ax(0)-bx'(0)=\int_0^1g(s)x(s){\rm d}s, \ \ cx(1)+dx'(1)=\int_0^1h(s)x(s){\rm d}s, \\[3mm] \displaystyle ax''(0)-bx'''(0)=\int_0^1g(s)x''(s){\rm d}s, \ cx''(1)+dx'''(1)=\int_0^1h(s)x''(s){\rm d}s \end{array}\right. $

有唯一解

$ \begin{eqnarray*} x(t)&=&\int_0^1G_2(t, s)F_2(s){\rm d}s+A_2(F_2)\varphi_2(t)+B_2(F_2)\psi_2(t)\nonumber\\ &=&\int_0^1G_1(t, s)F_1(s){\rm d}s+A_1(F_1)\varphi_1(t)+B_1(F_1)\psi_1(t), \end{eqnarray*} $

其中$G_i, \ A_i, \ B_i\ (i=1, 2)$由 (2.1) 式, (2.3) 式和 (2.4) 式定义,

$ \begin{align} & {{F}_{2}}(t)=\int_{0}^{1}{{{G}_{1}}}(t, s)y(s)\text{d}s+{{A}_{1}}(y){{\varphi }_{1}}(t)+{{B}_{1}}(y){{\psi }_{1}}(t), \\ & {{F}_{1}}(t)=\int_{0}^{1}{{{G}_{2}}}(t, s)y(s)\text{d}s+{{A}_{2}}(y){{\varphi }_{2}}(t)+{{B}_{2}}(y){{\psi }_{2}}(t). \\ \end{align} $

此外, 若$(A_1), \ (A_2)$成立且$y\geq 0$, 则$x(t)\geq 0, \ t\in [0, 1]$.

注2.1  对任意的$y\in C[0, 1]$且非负, $A_i(y), \ B_i(y)\ (i=1, 2)$是线性泛函且关于$y$非减.

定义算子$T: E\rightarrow E$如下

$ \begin{equation}\label{eq:25}(Tx)(t)=\int_0^1G_2(t, s)e(f)(s){\rm d}s+A_2(e(f))\varphi_2(t)+B_2(e(f))\psi_2(t), \ t\in [0, 1], \end{equation} $ (2.5)

其中

$ (ef)(t)=\int_0^1G_1(t, s)f(s, x(s)){\rm d}s+A_1(f)\varphi_1(t)+B_1(f)\psi_1(t), \ t\in [0, 1]. $

由引理2.4知, $x$是边值问题 (1.1) 和 (1.2) 的解当且仅当$x$是算子$T$的不动点.由$(A_1)$, $(A_2)$, $(H_1)$, $(H_2)$, $(H_{31})$$(H_{32})$知算子$T: P\rightarrow P$全连续.定义算子$L, \ L^*:E\rightarrow E$如下

$ \begin{equation}\label{eq:26}(Lx)(t)=\int_0^1G_2(t, s)(l_2x)(s){\rm d}s+A_2(l_2(x))\varphi_2(t)+B_2(l_2(x))\psi_2(t), \end{equation} $ (2.6)
$ (L^*x)(t)=\int_0^1G_1(t, s)(l_1x)(s){\rm d}s+A_1(l_1(x))\varphi_1(t)+B_1(l_1(x))\psi_1(t), $

其中

$ (l_2x)(t)=\int_0^1G_1(t, s)x(s){\rm d}s+A_1(x)\varphi_1(t)+B_1(x)\psi_1(t), $
$ (l_1x)(t)=\int_0^1G_2(t, s)x(s){\rm d}s+A_2(x)\varphi_2(t)+B_2(x)\psi_2(t). $

根据条件$(A_1), \ (A_2), \ (H_2), \ (H_{31})$$(H_{32})$易知$L, \ L^*:P\rightarrow P$为全连续线性算子.

引理2.5  假设$(A_1), \ (A_2), \ (H_2), \ (H_{31})$$(H_{32})$成立, 则由 (2.6) 式定义的算子$L$的谱半径$r(L)\not=0$, 对于其第一特征值$\lambda^*=(r(L))^{-1}, \ L$存在对应的正特征函数$\phi^*$, 即$\phi^*=\lambda^*L\phi^*$.

  易知存在$t_0\in (0, 1)$使得$G_1(t_0, t_0)G_2(t_0, t_0)>0$.从而存在$[t_1, t_2]\subset [0, 1]$使得$t_0\in (t_1, t_2)$

$ G_2(t, \tau)G_1(\tau, s)>0, \quad t, \tau, s\in [t_1, t_2]. $

$x\in E$使得$x(t)\geq 0, \ \forall t\in [0, 1], \ x(t_0)>0$$x(t)=0, \ \forall t\not\in [t_1, t_2]$, 则对$t\in [t_1, t_2]$, 有

$ \begin{eqnarray*} (Lx)(t)&=&\int_0^1G_2(t, s)(l_2x)(s){\rm d}s+A_2(l_2(x))\varphi_2(t)+B_2(l_2(x))\psi_2(t)\\ &=&\int_0^1\int_0^1G_2(t, s)G_1(s, \tau)x(\tau){\rm d}\tau{\rm d}s+\int_0^1G_2(t, s)A_1(x)\varphi_1(s){\rm d}s\\ &&+\int_0^1G_2(t, s)B_1(x)\psi_1(s){\rm d}s+A_2(l_2(x))\varphi_2(t)+B_2(l_2(x))\psi_2(t)\\ &\geq&\int_{t_1}^{t_2}\int_{t_1}^{t_2}G_2(t, s)G_1(s, \tau)x(\tau){\rm d}\tau{\rm d}s+\int_{t_1}^{t_2}G_2(t, s)A_1(x)\varphi_1(s){\rm d}s\\ &&+\int_{t_1}^{t_2}G_2(t, s)B_1(x)\psi_1(s){\rm d}s+A_2(l_2(x))\varphi_2(t)+B_2(l_2(x))\psi_2(t)>0.\end{eqnarray*} $

于是存在常数$c > 0$, 使得对任意的$t\in [0, 1], \ c(Lx)(t)\geq x(t)$.根据Krein-Rutman定理[21], 谱半径$r(L)\not=0$, 且对于第一特征值$\lambda^*=(r(L))^{-1}, \ L$存在对应的正特征函数.

注 2.2  根据引理2.4和引理2.5, $L=L^*$, 谱半径$r(L)=r(L^*)$, 且对于第一特征值$\lambda^*=(r(L))^{-1}=(r(L^*))^{-1}$, $L, \ L^*$有相同的正特征函数$\phi^*$, 即$\phi^*=\lambda^* L\phi^*, \ \phi^*=\lambda^* L^*\phi^*$.

$ \begin{matrix} {{C}_{0}}=\int_{0}^{1}{{{G}_{2}}}(s, s){{q}_{1}}(s)\text{d}s+{{A}_{2}}({{q}_{1}}(s))\|{{\varphi }_{2}}\|+{{B}_{2}}({{q}_{1}}(s))\|{{\psi }_{2}}\|, \\ M=\int_{0}^{1}{{{G}_{2}}}(s, s)\text{d}s+{{A}_{2}}(1)\|{{\varphi }_{2}}\|+{{B}_{2}}(1)\|{{\psi }_{2}}\|, \quad \kappa =\frac{{{C}_{0}}}{M}\int_{0}^{1}{{{q}_{2}}}(s){{\phi }^{*}}(s)\text{d}s, \\ K=\left\{ x\in P:\int_{0}^{1}{{{\phi }^{*}}}(t)x(t)\text{d}t\ge \kappa \|x\| \right\}. \\ \end{matrix} $

显然, $C_0>0, \ M>0, \ \kappa>0, \ K$$E$中的锥.

引理2.6  假设$(A_1), \ (A_2), \ (H_2), \ (H_{31})$$(H_{32})$成立, 则$L(P)\subset K$ (从而$L(K)\subset K$).

  根据 (2.6) 式, 引理2.2和引理2.3, 对任意的$x\in P, \ t\in [0, 1]$, 有$Lx(t)\geq 0$, 且

$ \begin{eqnarray}\label{eq:27} (Lx)(t)&\leq&\left[\int_0^1G_2(s, s){\rm d}s+A_2(1)\|\varphi_2\|+B_2(1)\|\psi_2\|\right] \nonumber\\ &&\cdot \left[\int_0^1G_1(s, s)x(s){\rm d}s+A_1(x)\|\varphi_1\|+B_1(x)\|\psi_1\|\right], \end{eqnarray} $ (2.7)

从而

$ \|Lx\|\leq M\left[\int_0^1G_1(s, s)x(s){\rm d}s+A_1(x)\|\varphi_1\|+B_1(x)\|\psi_1\|\right]. $

另一方面, 对任意的$x\in P, \ t\in [0, 1]$, 由引理2.2可推得

$ (l_2x)(t)\geq q_1(t)\left[\int_0^1G_1(s, s)x(s){\rm d}s+A_1(x)\|\varphi_1\|+B_1(x)\|\psi_1\|\right], $
$ \begin{eqnarray}\label{eq:28}(Lx)(t)&\geq & q_2(t)\left[\int_0^1G_2(s, s)(l_2x)(s){\rm d}s+A_2(l_2x)\|\varphi_2\|+B_2(l_2x)\|\psi_2\|\right]\nonumber\\ &\geq & q_2(t)\left[\int_0^1G_1(s, s)x(s){\rm d}s+A_1(x)\|\varphi_1\|+B_1(x)\|\psi_1\|\right]\nonumber\\ &&\cdot\left[\int_0^1G_2(s, s)q_1(s){\rm d}s+A_2(q_1(s))\|\varphi_2\|+B_2(q_1(s))\|\psi_2\|\right]\nonumber\\ &=&C_0q_2(t)\left[\int_0^1G_1(s, s)x(s){\rm d}s+A_1(x)\|\varphi_1\|+B_1(x)\|\psi_1\|\right].\end{eqnarray} $ (2.8)

根据 (2.7) 式和 (2.8) 式得到

$ (Lx)(t)\geq \frac{C_0}{M}q_2(t)\|Lx\|, \quad t\in [0, 1], $

所以

$ \int_0^1\phi^*(t)(Lx)(t){\rm d}t\geq \kappa \|Lx\|. $

于是, $L(P)\subset K$.证毕.

注2.3  类似地, $T(P)\subset K$, 进而有$T(K)\subset K$.

注2.4  若$x$是算子$T$$P$中的不动点, 则

$ x(t)\geq \frac{C_0 }{M}q_2(t)\|x\|, \quad t\in [0, 1], $

于是$x(t)>0, \ \forall t\in (0, 1)$, 因此$x$是边值问题 (1.1) 和 (1.2) 的正解.

3 主要结果

定理3.1  假设$(A_1), \ (A_2), \ (H_1), \ (H_2), \ (H_{31})$$(H_{32})$成立, 且

$ {f_\infty } = \mathop {\lim \inf }\limits_{u \to + \infty } \mathop {\min }\limits_{t \in [0,1]} \frac{{f(t,u)}}{u} > {\lambda ^*},\quad {f^0} = \mathop {\lim \sup }\limits_{u \to {0^ + }} \mathop {\max }\limits_{t \in [0,1]} \frac{{f(t,u)}}{u} < {\lambda ^*}, $

其中$\lambda^*$是由 (2.6) 式定义的算子$L$的第一特征值, 则边值问题 (1.1), (1.2) 至少有一个正解.

  由$f_\infty>\lambda^*$知, 存在$\varepsilon_1>0$使得当$u$充分大时有

$ f(t, u) \geq (\lambda^*+\varepsilon_1)u, \quad t \in [0, 1]. $

根据$f$的连续性, 存在$R_1>0$使得

$ f(t, u) \geq (\lambda^*+\varepsilon_1)u-R_1, \quad t \in [0, 1], \ u\in [0, +\infty). $

对任意的$x\in P$$t\in [0, 1]$, 有

$ \begin{eqnarray*} (ef)(t)&=&\int_0^1G_1(t, s)f(s, x(s)){\rm d}s+A_1(f)\varphi_1(t)+B_1(f)\psi_1(t)\\ &\geq & (\lambda^*+\varepsilon_1)(l_2x)(t)-R_1(l_2{\bf 1})(t), \end{eqnarray*} $

从而由 (2.5) 式可知

$ \begin{equation}\label{eq:31} (Tx)(t) \geq (\lambda^*+\varepsilon_1)(Lx)(t)-R_1(L{\bf 1})(t), \end{equation} $ (3.1)

这里1表示常值函数${\bf 1}(t)\equiv 1$.令

$ \Omega_1=\{x\in K: x\geq Tx\}, $

$\Omega_1$$K$中的有界集.事实上, 若$x\in \Omega_1$, 由 (3.1) 式知

$ x(t)\geq (\lambda^*+\varepsilon_1)(Lx)(t)-R_1(L{\bf 1})(t), \quad t\in [0, 1]. $

在上式两边同乘以$\phi^*(t)$并在$[0, 1]$上积分, 应用分部积分并交换积分顺序得

$ \begin{eqnarray*} \int_0^1x(t)\phi^*(t){\rm d}t&\geq &(\lambda^*+\varepsilon_1)\int_0^1(Lx)(t)\phi^*(t){\rm d}t-R_1\int_0^1(L{\bf 1})(t)\phi^*(t){\rm d}t\\ &=&(\lambda^*+\varepsilon_1)\int_0^1x(t)(L^*\phi^*)(t){\rm d}t-R_1\int_0^1(L\phi^*)(t){\rm d}t\\ &=&(\lambda^*+\varepsilon_1)\int_0^1r(L)x(t)\phi^*(t){\rm d}t-R_1\int_0^1r(L)\phi^*(t){\rm d}t, \end{eqnarray*} $

所以

$ \int_0^1x(t)\phi^*(t){\rm d}t\leq \frac{R_1\int_0^1\phi^*(t){\rm d}t}{\varepsilon_1}. $

因此

$ \|x\|\leq \frac{R_1\int_0^1\phi^*(t){\rm d}t}{\kappa \varepsilon_1}, \quad \forall x\in \Omega_1. $

这就证明了$\Omega_1$的有界性.取$r_1>\sup\limits_{x\in \Omega_1}\|x\|$, 则有

$ \begin{equation}\label{eq:32} Tx\not\leq x, \quad x\in \partial B_{r_1}\cap K.\end{equation} $ (3.2)

$f^0<\lambda^*$知, 存在$0<\varepsilon_2<\lambda^*$$0<r_2<r_1$, 使得

$ f(t, u) \leq (\lambda^*-\varepsilon_2)u, \quad t \in [0, 1], \ u\in [0, r_2], $

从而对任意的$x\in \overline{B}_{r_2}\cap P$$t\in [0, 1]$, 有

$ (ef)(t)\leq (\lambda^*-\varepsilon_2)(l_2x)(t), $
$ \begin{equation}\label{eq:33} (Tx)(t)\leq(\lambda^*-\varepsilon_2)(Lx)(t). \end{equation} $ (3.3)

$ \Omega_2=\{x\in \overline{B}_{r_2}\cap K: x\leq Tx\}, $

$\Omega_2=\{0\}$.事实上, 若$x\in \Omega_2$, 根据 (3.3) 式, 有

$ x(t)\leq (\lambda^*-\varepsilon_2)(Lx)(t), \quad t\in [0, 1]. $

在上式两边同乘以$\phi^*(t)$并在[0, 1]上积分, 应用分部积分并交换积分顺序得

$ \begin{eqnarray*}\int_0^1x(t)\phi^*(t){\rm d}t&\leq &(\lambda^*-\varepsilon_2)\int_0^1(Lx)(t)\phi^*(t){\rm d}t\\ &=&(\lambda^*-\varepsilon_2)\int_0^1x(t)(L^*\phi^*)(t){\rm d}t\\ &=&(\lambda^*-\varepsilon_2)\int_0^1r(L)x(t)\phi^*(t){\rm d}t, \end{eqnarray*} $

于是$\int_0^1x(t)\phi^*(t){\rm d}t=0$, 再结合$x\in K$可得$x=0$.这就证明了$\Omega_2=\{0\}$.因此,

$ \begin{equation}\label{eq:34} Tx\not\geq x, \quad x\in \partial B_{r_2}\cap K.\end{equation} $ (3.4)

根据 (3.2) 式、(3.4) 式和引理1.1可得$T$至少有一个不动点$x^*\in (B_{r_1}\setminus \overline{B}_{r_2} )\cap K$, 再根据注2.4知$x^*$是边值问题 (1.1), (1.2) 的正解.

定理3.2  假设$(A_1), \ (A_2), \ (H_1), \ (H_2), \ (H_{31})$$(H_{32})$成立, 且

$ {f_0} = \mathop {\lim \inf }\limits_{u \to {0^ + }} \mathop {\min }\limits_{t \in [0,1]} \frac{{f(t,u)}}{u} > {\lambda ^*},\quad {f^\infty } = \mathop {\lim \sup }\limits_{u \to + \infty } \mathop {\max }\limits_{t \in [0,1]} \frac{{f(t,u)}}{u} < {\lambda ^*}, $

其中$\lambda^*$是由 (2.6) 式定义的算子$L$的第一特征值, 则边值问题 (1.1), (1.2) 至少有一个正解.

  由$f_0>\lambda^*$知, 存在$\varepsilon_3>0$$r_3>0$使得

$ f(t, u) \geq (\lambda^*+\varepsilon_3)u, \quad t \in [0, 1], \ u \in [0, r_3]. $

从而对任意$x\in \overline{B}_{r_3}\cap P$$t\in [0, 1]$, 有

$ (ef)(t)\geq (\lambda^*+\varepsilon_3)(l_2x)(t), $
$ \begin{equation} \label{eq:35}(Tx)(t)\geq (\lambda^*+\varepsilon_3)(Lx)(t). \end{equation} $ (3.5)

$ \Omega_3=\{x\in \overline{B}_{r_3}\cap K: x\geq Tx\}. $

下证$\Omega_3=\{0\}$.事实上, 若$x\in \Omega_3$, 根据 (3.5) 式得

$ x(t)\geq (\lambda^*+\varepsilon_3)(Lx)(t). $

在上式两边同乘以$\phi^*(t)$并在$[0, 1]$上积分, 应用分部积分并交换积分顺序得

$ \begin{eqnarray*} \int_0^1x(t)\phi^*(t){\rm d}t&\geq &(\lambda^*+\varepsilon_3)\int_0^1(Lx)(t)\phi^*(t){\rm d}t\\ &=&(\lambda^*+\varepsilon_3)\int_0^1x(t)(L^*\phi^*)(t){\rm d}t\\ &=&(\lambda^*+\varepsilon_3)\int_0^1r(L)x(t)\phi^*(t){\rm d}t, \end{eqnarray*} $

于是$\int_0^1x(t)\phi^*(t){\rm d}t=0$.结合$x\in K$可推出$x=0$.这就证明了$\Omega_3=\{0\}$.因此

$ \begin{equation} \label{eq:36} Tx\not\leq x, \quad x\in \partial B_{r_3}\cap K.\end{equation} $ (3.6)

另一方面, 根据$f^\infty<\lambda^*$知, 存在$0<\varepsilon_4<\lambda^*$使得当$u$充分大时

$ f(t, u) \leq (\lambda^*-\varepsilon_4)u, \quad t \in [0, 1]. $

根据$f$的连续性, 存在$R_2>0$使得

$ f(t, u) \leq (\lambda^*-\varepsilon_4)u+R_2, \quad t \in [0, 1], \ u\in [0, +\infty). $

从而对任意的$x\in P$$t\in [0, 1]$,

$ (ef)(t)\leq (\lambda^*-\varepsilon_4)(l_2x)(t)+R_2(l_2{\bf 1})(t), $
$ \begin{equation}\label{eq:37} (Tx)(t) \leq (\lambda^*-\varepsilon_4)(Lx)(t)+R_2(L{\bf 1})(t). \end{equation} $ (3.7)

$ \Omega_4=\{x\in K: x\leq Tx\}. $

下证$\Omega_4$$K$的有界子集.事实上, 若$x\in \Omega_4$, 由 (3.7) 式可知

$ x(t)\leq (\lambda^*-\varepsilon_4)(Lx)(t)+R_2(L{\bf 1})(t). $

在上式两边同乘以$\phi^*(t)$并在$[0, 1]$上积分, 应用分部积分并交换积分顺序得

$ \begin{eqnarray*}\int_0^1x(t)\phi^*(t){\rm d}t&\leq &(\lambda^*-\varepsilon_4)\int_0^1(Lx)(t)\phi^*(t){\rm d}t+R_2\int_0^1(L{\bf 1})(t)\phi^*(t){\rm d}t\\ &=&(\lambda^*-\varepsilon_4)\int_0^1x(t)(L^*\phi^*)(t){\rm d}t+R_2\int_0^1(L\phi^*)(t){\rm d}t\\ &=&(\lambda^*-\varepsilon_4)\int_0^1r(L)x(t)\phi^*(t){\rm d}t+R_2\int_0^1r(L)\phi^*(t){\rm d}t, \end{eqnarray*} $

所以

$ \int_0^1x(t)\phi^*(t){\rm d}t\leq \frac{R_2\int_0^1\phi^*(t){\rm d}t}{\varepsilon_4}. $

结合$x\in K$

$ \|x\|\leq \frac{R_2\int_0^1\phi^*(t){\rm d}t}{\kappa \varepsilon_4}, \quad \forall x\in \Omega_4. $

这就证明了$\Omega_4$的有界性.取$ {r_4} > \max \{ \mathop {\sup }\limits_{x \in {\Omega _4}} \left\| x \right\|,\;{r_3}\} $, 则有

$ \begin{equation}\label{eq:38} Tx\not\geq x, \quad x\in \partial B_{r_4}\cap K.\end{equation} $ (3.8)

因此, 根据 (3.6) 式, (3.8) 式和引理1.1可知$T$$(B_{r_4}\setminus \overline{B}_{r_3} )\cap K$中至少有一个不动点, 此不动点为边值问题 (1.1), (1.2) 的正解.

注3.1  当$\alpha=\beta=0, \ a=c=1, \ b=d=0$时, 文献[14]在$f_0>\lambda^*, \ f^\infty<\lambda^*$情况下得到了正解的存在性.但对于另一种情况: $f^0<\lambda^*, \ f_\infty>\lambda^*$, 文献[14]没有得到类似的结果.因此, 本文的定理3.1是新的, 定理3.2也在一定程度上推广了文献[14]的结果.

注3.2  当$a=c=1, \ b=d=0, \ g(t)= h(t)=0$$\lambda^*=\pi^4- \beta\pi^2-\alpha$, 在这种特殊情况下本文定理3.1和定理3.2变为文献[6]中的定理1.1.

注3.3  对于更一般的边界条件

$ \left\{ \begin{array} {ll} a_1x(0)-b_1x'(0)=\int_0^1g_1(s)x(s){\rm d}s, & c_1x(1)+d_1x'(1)=\int_0^1h_1(s)x(s){\rm d}s, \\[3mm] a_2x''(0)-b_2x'''(0)=\int_0^1g_2(s)x''(s){\rm d}s, ~ & c_2x''(1)+d_2x'''(1)=\int_0^1h_2(s)x''(s){\rm d}s, \end{array}\right. $

这里$g_1\not=g_2, \ h_1\not=h_2$, 本文的方法不再适用.

4 例子

例4.1  取$\beta=-1, \ \alpha=0, \ a=b=c=d=1, \ g(t)=1, \ h(t)=t$$f(t, u)=\mu t \ln(1+u)+ u^2$, 其中$\mu> 0$充分小, 则边值问题 (1.1) 和 (1.2) 变为如下非局部边值问题

$ \begin{equation}\label{eq:41}\left\{ \begin{array} {lll} \displaystyle x^{(4)}(t)- x''(t)=\mu t \ln(1+x(t))+ x^2(t), \quad 0<t<1, \\[2mm] \displaystyle x(0)-x'(0)=\int_0^1x(s){\rm d}s, \ \ x(1)+x'(1)=\int_0^1sx(s){\rm d}s, \\[3mm] \displaystyle x''(0)-x'''(0)=\int_0^1x''(s){\rm d}s, \ x''(1)+x'''(1)=\int_0^1sx''(s){\rm d}s. \end{array}\right.\end{equation} $ (4.1)

在这种情况下, $\varphi_1(t)={\rm e}^t, \ \psi_1(t)={\rm e}^{1-t}, \ \varphi_2(t)=1+t, $$\psi_2(t)=2-t, \ \rho_1=2{\rm e}, \ \rho_2=3, \ \int_0^1g(s)\psi_1(s){\rm d}s=\int_0^1{\rm e}^{1-s}{\rm d}s={\rm e}-1, $$\int_0^1h(s)\varphi_1(s){\rm d}s=\int_0^1s{\rm e}^s{\rm d}s=1, \ \int_0^1g(s)\psi_2(s){\rm d}s=\int_0^1(2-s){\rm d}s=\frac{3}{2}, $$\int_0^1h(s)\varphi_2(s){\rm d}s=\int_0^1s(1+s){\rm d}s=\frac{5}{6}, \ \Delta_1=-{\rm e}^2-4{\rm e}+3, \ \Delta_2=-\frac{9}{4}$.通过直接计算可得$f_\infty=+\infty, \ f^0=\mu$.于是定理3.1的条件成立, 边值问题 (4.1) 至少有一个正解.

例4.2 $\alpha, \beta, a, b, c, d, \ g(t), \ h(t)$取法同例4.1, 取$f(t, u)=u^2 {\rm e}^{-u}+\mu \sin u$, 其中$\mu> 0$充分大, 则边值问题 (1.1) 和 (1.2) 变为如下非局部边值问题

$ \begin{equation}\label{eq:42}\left\{ \begin{array} {lll} \displaystyle x^{(4)}(t)- x''(t)=x^2(t) {\rm e}^{-x(t)}+\mu \sin x(t), \quad 0<t<1, \\[2mm] \displaystyle x(0)-x'(0)=\int_0^1x(s){\rm d}s, \ \ x(1)+x'(1)=\int_0^1sx(s){\rm d}s, \\[3mm] \displaystyle x''(0)-x'''(0)=\int_0^1x''(s){\rm d}s, \ x''(1)+x'''(1)=\int_0^1sx''(s){\rm d}s. \end{array}\right.\end{equation} $ (4.2)

在这种情况下, $f_0=\mu, \ f^\infty=0$, 定理3.2条件成立, 因此边值问题 (4.2) 至少有一个正解.

参考文献
[1] Aftabizadeh A R. Existence and uniqueness theorems for fourth-order boundary value problems. J Math Anal Appl, 1986, 116(2): 415–426. DOI:10.1016/S0022-247X(86)80006-3
[2] Agarwal R P. On fourth order boundary value problems arising in beam analysis. Differential Integral Equations, 1989, 2(1): 91–110.
[3] Gupta C P. Existence and uniqueness theorems for the bending of an elastic beam equation. Appl Anal, 1988, 26(4): 289–304. DOI:10.1080/00036818808839715
[4] Ma R Y, Wang H Y. On the existence of positive solutions of fourth-order ordinary differential equations. Appl Anal, 1995, 59: 225–231. DOI:10.1080/00036819508840401
[5] Timoshenko S P.Theory of Elastic Stability.New York:McGraw-Hill, 1961 http://www.scirp.org/reference/ReferencesPapers.aspx?ReferenceID=1251212
[6] Li Y X. Positive solutions of fourth-order boundary value problems with two parameters. J Math Anal Appl, 2003, 281(2): 477–484. DOI:10.1016/S0022-247X(03)00131-8
[7] Ma H L. Positive solution for m-point boundary value problems of fourth order. J Math Anal Appl, 2006, 321(1): 37–49. DOI:10.1016/j.jmaa.2005.08.015
[8] Yu C C, Chen S H, Austin F, Lu J H. Positive solutions of four-point boundary value problem for fourth order ordinary differential equation. Math Comput Modelling, 2010, 52(1/2): 200–206.
[9] Gallardo J M. Second-order differential operators with integral boundary conditions and generation of analytic semigroups. Rocky Mountain J Math, 2000, 30(4): 1265–1291. DOI:10.1216/rmjm/1021477351
[10] Lomtatidze A, Malaguti L. On a nonlocal boundary value problem for second order nonlinear singular differential equations. Georgian Math J, 2000, 7(1): 133–154.
[11] Bai Z B. Positive solutions of some nonlocal fourth-order boundary value problem. Appl Math Comput, 2010, 215(12): 4191–4197.
[12] Pang H H, Cai C.Monotone iterative technique and symmetric positive solutions to fourth-order boundary value problem with integral boundary conditions.Discrete Dyn Nat Soc, 2014, Art ID:583875 http://ftp.math.utah.edu/pub/tex/bib/fortran3.html
[13] Hao X A, Liu L S, Wu Y H, Sun Q. Positive solutions for nonlinear nth-order singular eigenvalue problem with nonlocal conditions. Nonlinear Anal, 2010, 73(6): 1653–1662. DOI:10.1016/j.na.2010.04.074
[14] Ma H L. Symmetric positive solutions for nonlocal boundary value problems of fourth order. Nonlinear Anal, 2008, 68(3): 645–651. DOI:10.1016/j.na.2006.11.026
[15] Lv X Z, Wang L B, Pei M H. Monotone positive solution of a fourth-order BVP with integral boundary conditions. Bound Value Probl, 2015, 2015: 172. DOI:10.1186/s13661-015-0441-2
[16] Webb J R L, Infante G. Positive solutions of nonlocal boundary value problems:a unified approach. J London Math Soc, 2006, 74(3): 673–693. DOI:10.1112/S0024610706023179
[17] Webb J R L, Infante G. Positive solutions of nonlocal boundary value problems involving integral conditions. NoDEA Nonlinear Differential Equations Appl, 2008, 15(1/2): 45–67.
[18] Infante G, Pietramala P, Tenuta M. Existence and localization of positive solutions for a nonlocal BVP arising in chemical reactor theory. Commun Nonlinear Sci Numer Simul, 2014, 19(7): 2245–2251. DOI:10.1016/j.cnsns.2013.11.009
[19] Zhang X M, Feng M Q. Positive solutions of singular beam equations with the bending term. Bound Value Probl, 2015, 2015: 84. DOI:10.1186/s13661-015-0348-y
[20] Krasnosel'skii M A, Zabreiko P P. Geometrical Methods of Nonlinear Analysis. Berlin: Springer-Verlag, 1984.
[21] Deimling K. Nonlinear Functional Analysis. Berlin: Springer-Verlag, 1985.