Loading [MathJax]/jax/element/mml/optable/MathOperators.js
  数学物理学报  2015, Vol. 35 Issue (6): 1059-1070   PDF (360 KB)    
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
冯育强
王蔚敏
李寿贵
带有奇异非线性项的分数微分方程周期解的存在性与唯一性
冯育强1, 王蔚敏2, 李寿贵2    
1 武汉科技大学理学院 武汉 430065;
2 冶金工业过程系统科学湖北省重点实验室 武汉 430081
摘要: 该文目的在于给出如下分数阶微分方程解的存在唯一性结论
{Dαx(t)=f(t,x(t)),tJ:=(0,1], 0<α<1,limt0+t1αx(t)=x(1), (PBVP)
其中ft=0可以是奇异的.主要的工具是上下解方法、最大值原理和单调迭代技术.最后举例说明所获结论的应用.
关键词: 分数微分方程     周期边值问题     存在性     唯一性     奇异性    
Existence and Uniqueness Results for the Periodic Boundary Value Problems of Fractional Differential Equations with Singular Nonlinearities
Feng Yuqiang1, Wang Weimin2, Li Shougui2    
1 School of Science, Wuhan University of Science and Technology, Wuhan 430065;
2 Hubei Province Key Laboratory of Systems Science in Metallurgical Process, Wuhan 430081
Abstract: The purpose of this paper is to give some sufficient conditions for the existence and uniqueness of solutions to the fractional differential equation as follows
{Dαx(t)=f(t,x(t)),tJ:=(0,1], 0<α<1,limt0+t1αx(t)=x(1), (PBVP)
where Dα denotes the Riemann-Liouville fractional derivative, f may be singular at t=0. Lower and upper solutions method, maximum principle together with iterative technique are employed. An example is presented to illustrate the application of results obtained.
Key words: Fractional differential equation     Periodic boundary value problem     Existence     Uniqueness     Singularity    
1 引 言

由于分数微积分具有刻画记忆和遗传性质的特性,所以在描述具有记忆和遗传性质的各种物质、 流变学、材料和力学系统、信号处理和系统辨识、ANN (神经网络)、分形和混沌时, 分数微积分比整数阶微积分更加准确. 分数阶模型涵盖了流体力学,流变学,粘弹性力学, 分数控制系统与分数控制器,各种电子回路,电分析化学,生物系统的电传导, 神经的分数模型以及分数回归模型等领域(见文献[1]及其参考文献). 正是由于其广泛而重要的应用,分数微分方程成为了目前理论研究的热点, 涌现出大量的研究成果,如文献[2,3,4,5,6,7]及其参考文献.

最近,分数阶微分方程的周期边值问题引起了诸多关注. 文献[8]首次考察了如 下分数阶微分方程边值问题的可解性

{Dαx(t)λx(t)=f(t,x(t)),tJ:=(0,1], 0<α<1,limt0+t1αx(t)=x(1). 作者计算了该问题的Green函数,并利用Schaeffer不动点定理和Bananch不动点定理证 明了一些解的存在性和唯一性结论. 随后,在文献[9]和[11]中, 作者给出了以下周期边值问题的一些比较原理

{Dαx(t)λx(t)=σ(t),tJ:=(0,1], 0<α<1,limt0+t1αx(t)=x(1). 文献[10]利用单调迭代方法研究了分数微分方程周期解的存在性与唯一性. 文献[12]则将该项研究推广到时滞微分方程的情形. 在这些文献中,大都要求非线 性项是Lipschitz 连续的或者单增且连续.

受以上工作以及减算子不动点研究[13]的启发,本文探讨了当非线性项为奇异的且满足 一定的递减性质时,如下的分数阶微分方程接的存在性与唯一性

{Dαx(t)=f(t,x(t)),tJ:=(0,1], 0<α<1,limt0+t1αx(t)=x(1). (PBVP)
主要工具是上下解方法、极大值原理和单调迭代技术.

本文的结构为: 第2节给出一些必要的记号与预备知识; 解的存在性与唯一性结论在第3节给出; 第4节举例说明所获结论的应用.

2 预备知识

设C[0, 1]表示[0,1]所有连续函数的集合. 若在该线性空间赋予范数

u=max0t1|u(t)|, 则(C[0,1],||.||)是一个Banach空间. 对于 r>0,u C(0,1],定义

tru(t)|t=0=limt0+tru(t), Cr[0,1]={uC(0,1]|tru(t)C[0,1]}.Cr[0,1]在范数 ur=max0t1|tru(t)| 下也成为一个Banach空间.

0<r<1,Cr[0,1]C(0,1]L1[0,1],这里 L1[0,1] 表示[0,1] 上所有Lebesgue可积的函数集合.

P={uCr[0,1]u(t)0,t(0,1]}.

2.1 P 是一个正规锥.

(1) PCr[0,1]中的一个锥.

这可以用P的定义来验证.

(2) P 是正规锥.

事实上,设 u,vP 满足 0u(t)v(t),t(0,1],则有 0tru(t)trv(t),t[0,1],即 urvr.

因此,P 是一个正规锥.

一个连续函数u:(0,1]Rα(0<α<1) 阶 Riemann-Liouville导数定义为

Iαu(t)=1Γ(α)t0(tτ)α1u(τ)dτ, 其中 Γ(.) 表示Γ函数.

引理2.20<α<1,如果 uCα[0,1],IαuC[0,1].

uCα[0,1],则

Iαu(t)=1Γ(α)t0(tτ)α1u(τ)dτ=1Γ(α)t0(tτ)α1(τ)αU(τ)dτ  (U(τ)=ταu(τ))=1Γ(α)10(1s)α1(s)αU(st)ds  (s=τt).

注意到 U(t)[0,1] 上一致连续,即对 ϵ>0, 存在 δ>0,当 t1,t2[0,1]|t1t2|<δ就有

|U(t1)U(t2)|<ϵ. 所以

|Iαu(t1)Iαu(t2)|=|1Γ(α)10(1s)α1(s)αU(st1)ds1Γ(α)10(1s)α1(s)αU(st2)ds|=|1Γ(α)10(1s)α1(s)α[U(st1)U(st2)]ds|1Γ(α)10(1s)α1(s)αU(st1)U(st2)dsϵ1Γ(α)10(1s)α1(s)αds  ( 其中 B(.,.)表示Beta函数.

因此有 I^{\alpha}u\in C[0,1].

一个连续函数u\in C(0,1]\alpha(0<\alpha<1)阶Riemann-Liouville导数定义为

D^{\alpha}u(t)=\frac{1}{\Gamma(1-\alpha)}\frac{\rm d}{{\rm d}t}\int_0^t(t-\tau)^{-\alpha}u(\tau){\rm d}\tau=\frac{\rm d}{{\rm d}t}I^{1-\alpha}u(t). 作为该定义的直接结论,分数阶导数具有以下性质

(1)当 \beta>-1时,

D^{\alpha}(t^{\beta})=\frac{1}{\Gamma(1-\alpha)}\frac{\rm d}{{\rm d}t}\int_0^t(t-\tau)^{-\alpha}{\tau}^{\beta}{\rm d}\tau=\frac{\Gamma(\beta+1)}{\Gamma(\beta+2-\alpha)}\frac{\rm d}{{\rm d}t}(t^{1+\beta-\alpha}).

(2)对于任意的 u\in C(0,1]\bigcap L^1[0,1],D^{\alpha}I^{\alpha}u=u.

(3)当 u\in C(0,1]\bigcap L^1[0,1]时,分数微分方程 D^{\alpha}u(t)=0 有解为 u(t)=ct^{\alpha-1},c\in R.

(4)当 u\in C(0,1]\bigcap L^1(0,1]D^{\alpha}u\in C(0,1]\bigcap L^1(0,1] (0<\alpha<1)时,下式成立

I^{\alpha}D^{\alpha}u(t)=u(t)+ct^{\alpha-1},\ c\in R.

对于 \alpha>0,\beta>0,二元的 Mittag-Leffler 函数定义如下

E_{\alpha,\beta}(z)=\sum_{k=0}^{\infty}\frac{z^k}{\Gamma(\alpha k+\beta)},\ \ z\in {\Bbb C}. 据文献[10]可知,当 x,y\in {\Bbb R},\ \ x<0<y

0<E_{\alpha,\alpha}(x)<E_{\alpha,\alpha}(0)=\frac{1}{\Gamma(\alpha)}<E_{\alpha,\alpha}(y), \lim\limits_{x\rightarrow +\infty}E_{\alpha,\alpha}(x)=+\infty,\ \ \lim\limits_{x\rightarrow -\infty}E_{\alpha,\alpha}(x)=0.

为证明主要结论,需要以下的引理.

{\bf 引理2.3}[9]h\in C[0,1],\alpha\in (0,1),\lambda\in {\Bbb R}E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}. 如果 u(t)\in C_{1-\alpha}[0,1] 满足

(i) D^{\alpha}u(t)-\lambda u(t)=h(t)(0<t<1);

(ii) \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}u(t)=u(1).

则有

u(t)=\int_0^1G_{\lambda,\alpha}(t,s)h(s){\rm d}s, 其中 G_{\lambda,\alpha}(t,s) 的定义为: 当 0\leq s\leq t\leq1,

\frac{\Gamma(\alpha)E_{\alpha,\alpha}(\lambda t^{\alpha})E_{\alpha,\alpha}(\lambda (1-s)^{\alpha}) t^{\alpha-1}(1-s)^{\alpha-1}}{1-\Gamma(\alpha)E_{\alpha,\alpha}(\lambda)}+(t-s)^{\alpha-1}E_{\alpha,\alpha}(\lambda (t-s)^{\alpha}).

0\leq t\leq s\leq1,

\frac{\Gamma(\alpha)E_{\alpha,\alpha}(\lambda t^{\alpha})E_{\alpha,\alpha}(\lambda (1-s)^{\alpha}) t^{\alpha-1}(1-s)^{\alpha-1}}{1-\Gamma(\alpha)E_{\alpha,\alpha}(\lambda)}.

{\bf 注2.1}\quad 据文献[10,引理2.2],当 \alpha\in (0,1),\ \lambda\in R,\ \lambda<0时, 0<E_{\alpha,\alpha}(\lambda)< \frac{1}{\Gamma(\alpha)}. 如果E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}, 容易验证 G_{\lambda,\alpha}(t,s)>0 for (t,s)\in (0,1)\times (0,1).

{\bf 引理2.4}[9]\quad\alpha\in (0,1),\lambda\in {\Bbb R}E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}. 如果 u(t)\in C_{1-\alpha}[0,1] 满足

(i) D^{\alpha}u(t)-\lambda u(t)\geq 0(0<t<1);

(ii) \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}u(t)=u(1).

u(t)\geq 0,\forall t\in (0,1].

3 主要结论

在本节中,始终假定以下条件成立.

(H_1) f:(0,1]\times R\longrightarrow R 是连续函数;

(H_2) 对于 \forall x\in C_{1-\alpha}[0,1],\ f(t,x(t))\in C_{1-\alpha}[0,1], 其中

t^{1-\alpha}f(t,x(t))|_{t=0}:=\lim\limits_{t\rightarrow 0^+}t^{1-\alpha}f(t,x(t)). 值得注意的是: 当 (H_1)-(H_2) 成立时,f(t,x) 可以在t=0 是奇异的.

易见 (PBVP) 与如下问题等价

\left \{\begin{array}{ll} D^{\alpha}x(t)-\lambda x(t)=f(t,x(t))-\lambda x(t),t\in J:= (0,1],\ 0<\alpha<1,\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x(t)=x(1), \end{array} \right. 其中参数 \lambda\in{\Bbb R} 且满足 E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}.

对于 x\in C_{1-\alpha}[0,1],定义映射 T 如下

(Tx)(t)=\int_0^1G_{\lambda,\ \alpha}(t,s)[f(s,x(s))-\lambda x(s)]{\rm d}s, 其中 G_{\lambda,\ \alpha}(t,s) 的定义见引理2.3. 于是 x\in C_{1-\alpha}(PBVP) 的一个解当且仅当 x\in C_{1-\alpha}T的一个不动点.

{\bf 引理3.1}\quad 当条件 (H_1)-(H_2)成立时, 算子 TC_{1-\alpha}[0,1]C_{1-\alpha}[0,1]T 是连续的.

{\bf 证}\quad 结论分两步证明.

(1) 若 x\in C_{1-\alpha}[0,1],则Tx\in C_{1-\alpha}[0,1].

显然当 x\in C_{1-\alpha}[0,1] 时,[f(t,x(t))-\lambda x(t)]\in C_{1-\alpha}[0,1].

\sigma(t)=f(t,x(t))-\lambda x(t),u(t)=(Tx)(t),则有

\left \{\begin{array}{ll} D^{\alpha}u(t)-\lambda u(t)=\sigma(t),t\in J:= (0,1],\ 0<\alpha<1,\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}u(t)=u(1). \end{array} \right.

据文献[9],当 \sigma\in C_{1-\alpha}[0,1]E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)} 时,下述线性分数微分方程边值问题

\left \{\begin{array}{ll} D^{\alpha}u(t)-\lambda u(t)=\sigma(t),t\in J:= (0,1],\ 0<\alpha<1,\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x(t)=x(1) \end{array} \right. 的解是

u(t)=u_0\Gamma(\alpha)t^{1-\alpha}E_{\alpha,\alpha}(\lambda t^{\alpha})+\int_0^t {(t-s)}^{\alpha-1}E_{\alpha,\alpha}(\lambda {(t-s)^{\alpha})}\sigma(s){\rm d}s, 其中 u_0为一常数.

以下来证明 u\in C_{1-\alpha}[0,1].

事实上,

t^{1-\alpha}u(t)=u_0\Gamma(\alpha)t^{2(1-\alpha)}E_{\alpha,\alpha}(\lambda t^{\alpha})+t^{1-\alpha}\int_0^t {(t-s)}^{\alpha-1}E_{\alpha,\alpha}(\lambda {(t-s)^{\alpha})}\sigma(s){\rm d}s

只需说明 t^{1-\alpha}\int_0^t {(t-s)}^{\alpha-1}E_{\alpha,\alpha}(\lambda {(t-s)^{\alpha})}\sigma(s){\rm d}st的连续函数. 注意到

\begin{eqnarray*} &&t^{1-\alpha}\int_0^t {(t-s)}^{\alpha-1}E_{\alpha,\alpha}(\lambda {(t-s)^{\alpha})}\sigma(s){\rm d}s\\ & =&\sum_{k=0}^{\infty}\frac{t^{1-\alpha}\lambda^{k\alpha}}{\Gamma(\alpha k+\alpha)}\int_0^t(t-s)^{(k+1)\alpha-1}\sigma(s){\rm d}s\\ &=&\sum_{k=0}^{\infty}\frac{t^{1-\alpha}\lambda^{k\alpha}}{\Gamma(\alpha k+\alpha)}\int_0^t(t-s)^{(k+1)\alpha-1}s^{\alpha-1}\phi(s){\rm d}s\\ &=&\sum_{k=0}^{\infty}\frac{t^{(k+1)\alpha}\lambda^{k\alpha}}{\Gamma(\alpha k+\alpha)}\int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}\phi(ut){\rm d}u, \end{eqnarray*} 其中\phi(t)=t^{1-\alpha}\sigma(t)\in C[0,1]. 由于 \phi(t)[0,1] 上一致连续,即 \forall \epsilon >0,存在\delta>0,当 t_1,t_2\in [0,1]|t_1- t_2|<\delta时,

|\phi(t_1)-\phi(t_2)|<\epsilon. 此时,

\begin{eqnarray*} &&\bigg| \int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}\phi(ut_1){\rm d}u-\int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}\phi(ut_2){\rm d}u\bigg|\\ & =&\bigg| \int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}(\phi(ut_1)-\phi(ut_2)){\rm d}u\bigg|\\ &\leq& \int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}| \phi(ut_1)-\phi(ut_2)|{\rm d}u\\ & \leq& \epsilon \int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}{\rm d}u\\ &=&\epsilon B((k+1)\alpha,\alpha). \end{eqnarray*} 这说明对于任意的k\frac{t^{(k+1)\alpha}}{\Gamma(\alpha k+\alpha)} \int_0^1(1-u)^{(k+1)\alpha-1}u^{\alpha-1}\phi(ut){\rm d}ut 的连续函数.

因此,t^{1-\alpha}u(t)\in C[0,1].

(2) 在假设 (H_1)-(H_2)下,利用Lebesgue控制收敛定理可以证明 T 是连续的.

u,v\in C_{1-\alpha}[0,1] 满足 u(t)\leq v(t),t\in [0,1],记

[u,v]=\{x\in C_{1-\alpha}[0,1]| u(t)\leq x(t)\leq v(t),t\in (0,1]\}.

{\bf 定理3.1}\quad 如果存在 u,v \in C_{1-\alpha}[0,1],u\leq v 和常数 \lambda (\lambda 满足E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}), 使得以下条件成立

(A_1) 对于任意的t\in [0,1],x,y\in [u,v],x\leq y,

f(t,y(t))-f(t,x(t))\leq \lambda (y(t)-x(t));

(A_2) 对于任意的t\in [0,1],x,y\in [u,v],

f(t,lx(t)+(1-l)y(t)\leq lf(t,x(t))+(1-l)f(t,y(t));

(A_3) 对于任意的 t\in (0,1),

D^{\alpha}(v+u)(t)+\lambda (v-u)(t)\leq 2f(t,v(t)), f(t,u(t))\leq D^{\alpha}v(t)-\lambda(v-u)(t);

(A_4) \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}u(t)=u(1),\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}v(t)=v(1).

那么(PBVP) 存在唯一解 x^*\in C_{1-\alpha}[0,1]\cap [u,v].

{\bf证}\quad 分数阶微分方程周期边值问题 (PBVP) 等价于如下边值问题

\left \{\begin{array}{ll} D^{\alpha}x(t)-\lambda x(t)=f(t,x(t))-\lambda x(t),0< t< 1,\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x(t)=x(1). \end{array} \right.

x_0 为下述边值问题的解

\left \{\begin{array}{ll} D^{\alpha}x(t)-\lambda x(t)=f(t,v(t))-\lambda v(t),\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x(t)=x(1), \end{array} \right.

x_0=\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,v(s))-\lambda v(s)]{\rm d}s, 这里 G_{\lambda,\alpha}(t,s) 的定义见引理 2.2. 构造迭代列 \{x_n\} 如下

\left\{ \begin{array}{ll} D^{\alpha}x_{n+1}(t)-\lambda x_{n+1}(t)=f(t,x_{n}(t))-\lambda x_{n}(t),\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x_{n+1}(t)=x_{n+1}(1), \\ n=0,1,2,\cdots . \end{array} \right.

由于 f 连续,所以点列 \{x_n\} 有意义且有 \{x_n\}\subset C_{1-\alpha}[0,1].

以下来证明 \{x_n\} 收敛到问题 (PBVP) 在序区间 [u,v] 中的唯一解.

证明过程分为5步.

第1步) v(t)\geq x_0(t)\geq (\frac{u+v}{2})(t)\geq u(t).

由定理条件 (A1),(A3),(A4),可知

\begin{eqnarray*} D^{\alpha} \Big(\frac{u+v}{2}\Big)(t)-\lambda\Big(\frac{u+v}{2}\Big)(t) &\leq & f(t,v(t))-\lambda v(t)\leq f(t,u(t))-\lambda u(t) \\ & \leq& D^{\alpha}v(t)-\lambda v(t). \end{eqnarray*} 于是,根据 x_0 的定义易得

D^{\alpha}v(t)-\lambda v(t)\geq D^{\alpha}x_0(t)-\lambda x_0(t)\geq D^{\alpha} \Big(\frac{u+v}{2}\Big)(t)-\lambda\Big(\frac{u+v}{2}\Big)(t). 注意到

\lim\limits_{t\rightarrow 0^+}t^{1-\alpha}u(t)=u(1), \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}v(t)=v(1), \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x_0(t)=x_0(1). 据引理 2.5,有

v(t)\geq x_0(t)\geq \Big(\frac{u+v}{2}\Big)(t)\geq u(t).

第2步)对于n=0,1,2,\cdots , 成立着

v(t)\geq x_{2n+1}(t)\geq x_{2n}(t)\geq \Big(\frac{u+v}{2}\Big)(t)\geq u(t).

利用数学归纳法来证明结论.

事实上,由于

D^{\alpha}x_{1}(t)-\lambda x_{1}(t)=f(t,x_{0}(t))-\lambda x_{0}(t).

根据第1步证明及条件(A1)

f(t,u(t))-\lambda u(t)\geq f(t,x_{0}(t))-\lambda x_{0}(t)\geq f(t,v(t))-\lambda v(t). 注意到

f(t,u(t))-\lambda u(t) \leq D^{\alpha}v(t)-\lambda v(t), 所以有

D^{\alpha}v(t)-\lambda v(t)\geq D^{\alpha}x_{1}(t)-\lambda x_{1}(t)\geq D^{\alpha}x_{0}(t)-\lambda x_{0}(t).

根据引理 2.4,

v(t)\geq x_{1}(t)\geq x_{0}(t). 所以当 n=0结论成立.

设若 n=k 时结论也成立,即

v(t)\geq x_{2k+1}(t)\geq x_{2k}(t)\geq \Big(\frac{u+v}{2}\Big)(t)\geq u(t).

则由假设 (A1),对于任意的 t\in [0,1]

\begin{eqnarray*} f(t,u(t))-\lambda u(t)&\geq& f(t,x_{2k}(t))-\lambda x_{2k}(t)\geq f(t,x_{2k+1}(t))-\lambda x_{2k+1}(t)\\ &\geq& f(t,v(t))-\lambda v(t). \end{eqnarray*} 这说明

\begin{eqnarray*} D^{\alpha}v(t)-\lambda v(t)&\geq& D^{\alpha}x_{2k+1}(t)- \lambda x_{2k+1}(t)\geq D^{\alpha}x_{2k+2}(t)-\lambda x_{2k+2}(t)\\ &\geq & D^{\alpha}x_{0}(t)-\lambda x_{0}(t). \end{eqnarray*} 再根据引理 2.3,可知

v(t)\geq x_{2k+1}(t)\geq x_{2k+2}(t)\geq x_{0}(t)\geq u(t).

重复此过程,可得

\left\{ \begin {array} {ll} f(t,u(t))-\lambda u(t)\geq f(t,x_{2k+2}(t))-\lambda x_{2k+2}(t)\\ \geq f(t,x_{2k+1}(t))-\lambda x_{2k+1}(t)\geq f(t,v(t))-\lambda v(t),\\ D^{\alpha}v(t)-\lambda v(t)\geq D^{\alpha}x_{2k+3}(t)-\lambda x_{2k+3}(t)\\ \geq D^{\alpha}x_{2k+2}(t)-\lambda x_{2k+2}(t)\geq D^{\alpha}x_{0}(t)-\lambda x_{0}(t),\\ v(t)\geq x_{2k+3}(t)\geq x_{2k+2}(t)\geq x_{0}(t)\geq u(t), \end {array} \right. 即当 n=k+1 结论也成立.

所以对所有的 n\in \{0\}\bigcup N,结论都成立.

第3步) \{x_{2n}(t)\} 递增,\{x_{2n+1}(t)\} 递减.

由第2步证明过程可知

v(t)\geq x_1(t)\geq u(t). 于是根据条件 (A1)

f(t,u(t))-\lambda u(t)\geq f(t,x_{1}(t))-\lambda x_{1}(t)\geq f(t,v(t))-\lambda v(t). 这表明

D^{\alpha}v(t)-\lambda v(t)\geq D^{\alpha}x_{2}(t)-\lambda x_{2}(t)\geq D^{\alpha}x_{0}(t)-\lambda x_{0}(t). v(t)\geq x_{2}(t)\geq x_0(t) .

类似于步骤2,利用数学归纳法可以证明 \{x_{2n}(t)\} 是递增的. 同理可证\{x_{2n+1}(t)\} 是递减的.

第4步) \{x_{n}(t)\} 收敛到问题 (PBVP)的一个解.

由步骤1--3,对于 n=0,1,2,\cdots,

u(t)\leq \Big(\frac{u+v}{2}\Big)(t)\leq x_{2n}(t) \leq x_{2n+2}(t)\leq x_{2n+3}(t)\leq x_{2n+1}(t) \leq v(t).

y_n(t)=x_n(t)-u(t),则对 n=0,1,2,\cdots,

0\leq \frac{v(t)-u(t)}{2}\leq y_{2n}(t) \leq y_{2n+2}(t)\leq y_{2n+3}(t)\leq y_{2n+1}(t) \leq v(t)-u(t).

r_n=\sup \{r\in R| y_{2n}(t)\geq r y_{2n+1}(t)\}, 则数列 \{r_n\}是良定的. 易验证 \frac{1}{2}\leq r_n\leq 1\{r_n\} 递增,所以 \{r_n\} 收敛.

由假设 (A1),(A2),可得

\begin{eqnarray*} \displaystyle &&y_{2n+3}(t) \leq y_{2n+1}(t) =x_{2n+1}(t)-u(t) \\&=& \int_0^1 G_{\lambda,\alpha}(t,s)[f(s,x_{2n}(s))-\lambda x_{2n}(s)]{\rm d}s-u(t) \\&=&\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,(y_{2n}+u)(s))-\lambda (y_{2n}+u)(s)]{\rm d}s-u(t) \\&\leq&\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,(r_ny_{2n+1}+u)(s))-\lambda (r_ny_{2n+1}+u)(s)]{\rm d}s-u(t) \\&=& \int_0^1 G_{\lambda,\alpha}(t,s)[f(s,r_nx_{2n+1}+(1-r_n)u(s))-\lambda (r_nx_{2n+1}(s)+(1-r_n)u(s))]{\rm d}s-u(t) \\ &\leq&\int_0^1 G_{\lambda,\alpha}(t,s)[r_nf(s,x_{2n+1}(s))+(1-r_n)f(s,u(s)) \\ &&-\lambda (r_nx_{2n+1}(s)+(1-r_n)u(s))]{\rm d}s-u(t) \\&=&r_n\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,x_{2n+1}(s))-\lambda x_{2n+1}(s)]{\rm d}s \\ &&+(1-r_n)\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,u(s))-\lambda u(s)]{\rm d}s-u(t) \\&\leq&r_nx_{2n+2}(t)+(1-r_n)v(t)-u(t) \\&=&r_ny_{2n+2}(t)+(1-r_n)(v(t)-u(t)) \\&=&r_ny_{2n+2}(t)+2(1-r_n)\frac{(v(t)-u(t))}{2} \\&\leq& r_ny_{2n+2}(t)+2(1-r_n)y_{2n+2}(t) \\&\leq&(2-r_n)y_{2n+2}(t), \end{eqnarray*} 从而

r_{n+1}=\sup \{r\in R| y_{2n+3}(t)\geq r y_{2n+2}(t)\}\geq \frac{1}{2-r_n}.

\lim\limits_{n\rightarrow \infty}r_n=r,上述不等式表明 r\geq \frac{1}{2-r},所以 r=1.

对任意的偶数 p>0,

0\leq y_{2n+p}-y_{2n}\leq y_{2n+1}-y_{2n}\leq (1-r_n)y_{2n+1}\leq (1-r_n)(v-u)(t).r_n\rightarrow 1 及锥P 的正规性可知,\{y_{2n}\}C_{1-\alpha}[0,1] 中的Cauchy列. 类似地,可以证明 \{y_{2n+1}\} 也是C_{1-\alpha}[0,1] 中的Cauchy列. 于是

\lim\limits_{n\rightarrow \infty}y_{2n}=\lim\limits_{n\rightarrow \infty}y_{2n+1}, 所以 \{y_n\} 收敛.

y^*=\lim\limits_{n\rightarrow \infty}y_{n},x^*=y^*(t)+u(t), 则有

\lim\limits_{n\rightarrow \infty}x_{n}(t)={x^*}(t). 以及

\begin{eqnarray*} {x^*}(t)&=&\lim\limits_{n\rightarrow \infty}\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,x_{n}(s))-\lambda x_{n}(s)]{\rm d}s \\ &=&\int_0^1 G_{\lambda,\alpha}(t,s)[f(s,x^*(s))-\lambda x^*(s)]{\rm d}s, \end{eqnarray*} 这表明 x^* 是问题 (PBVP) 的一个解.

第5步) x^* 是问题 (PBVP) 在序区间 [u,v]中的唯一解.

\overline{x}\in [u,v](PBVP) 的一个解,则

f(t,v(t))-\lambda v(t) \leq f(t,\overline{x}(t))-\lambda \overline{x}(t) \leq f(t,u(t))-\lambda u(t). 从而

D^{\alpha}x_0(t)-\lambda x_0(t) \leq D^{\alpha}\overline{x}(t)-\lambda \overline{x}(t) \leq D^{\alpha}v(t)-\lambda v(t).

据引理 2.4,有 x_0(t)\leq \overline{x}(t)\leq v(t).

因而

f(t,v(t))-\lambda v(t) \leq f(t,\overline{x}(t))-\lambda \overline{x}(t) \leq f(t,x_0(t))-\lambda x_0(t). 这表明

D^{\alpha}x_0(t)-\lambda x_0(t) \leq D^{\alpha}\overline{x}(t)-\lambda \overline{x}(t) \leq D^{\alpha}x_1(t)-\lambda x_1(t). 由引理2.4可知 x_0(t)\leq \overline{x}(t)\leq x_1(t).

继续这一过程,容易验证

x_{2n}(t)\leq \overline{x}(t)\leq x_{2n+1}(t),n=0,1,2,\cdots . 因此

\overline{x}(t)=\lim\limits_{n\rightarrow \infty}x_{n}(t)={x^*}(t). 证毕.

{\bf 注3.1}\quad 定理3.1是一个局部性结论, 即在一个局部其解是存在且唯一的; 但从全局来看,边值问题可能会有多个解.

{\bf 推论3.1}\quad 如果存在常数 c>0,\lambda<0 满足

1) \frac{\lambda c}{2}\leq t^{1-\alpha}f(t,c t^{1-\alpha})\leq t^{1-\alpha}f(t,0)\leq -\lambda c;

2) 对于 \forall t\in [0,1] ,f(t,.)为凸函数;

3) 对于 \forall t\in [0,1],x,y\in R,x\leq y,成立着f(t,y)-f(t,x)\leq \lambda (y-x);

4) E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}.

(PBVP) 有唯一解 u^* 满足 0 \geq {u^*}(t)\geq ct^{\alpha-1},\forall t\in [0,1]. 进一步, 如果存在 t_0\in (0,1) 使得 f(t_0,0)\neq 0,那么 u^*(t)>0,t\in(0,1).

{\bf 证}\quad 在推论3.1的假设下,设 u(t)\equiv 0,v(t)=ct^{\alpha-1},可以验证定理3.1的条件(A1)--(A5)都满足. 于是根据定理3.1可知周期边值问题存在唯一解 u^* 满足 0 \geq {u^*}(t)\geq ct^{\alpha-1},t\in [0,1].

进一步,如果存在 t_0\in(0,1] 使得 f(t_0,0)\neq 0,那么 u^*(t)(0,1)内不会恒为0,从而根据引理2.3得 u^*(t)>0,\ t\in(0,1).

4 例子

考察如下的边值问题

\left \{\begin{array}{ll} D^{\alpha}x(t)={\rm e}^{- x(t)}+\lambda x(t)+bt^{\alpha-1},0<t< 1,\\ \lim\limits_{t\rightarrow 0^+}t^{1-\alpha}x(t)=x(1), \end{array} \right. 其中 f(t,x)={\rm e}^{-x}+\lambda x+bt^{\alpha-1},显然ft=0 是奇异的. 如果 E_{\alpha,\alpha}(\lambda)<\frac{1}{\Gamma(\alpha)}, -\frac{\lambda}{2}\leq b\leq -(\lambda+1),那么据推论3.1可知该问题有唯一解 u^* 满足

0\leq u^*(t)\leq ct^{\alpha-1},~~t\in (0,1]. 进一步,由f(t,0)\neq 0 可知 u^*(t)>0,t\in(0,1).

参考文献
[1] Podlubny I. Fractional Differential Equations. San Diego:Academic Press, 1999
[2] 胡雷, 张淑琴, 侍爱玲.分数阶微分方程耦合系统共振边值问题解的存在性. 数学物理学报, 2014, 34(5):1313-1326
[3] Chen P, Li Y, Li Q. Existence of mild solutions for fractional evolution equations with nonlocal initial conditions. Ann Polon Math, 2014, 110(1):13-24
[4] Lakshmikantham V. Theory of fractional functional differential equations. Nonlinear Anal, 2008, 69(10):3337-3343
[5] Bai Z, Qiu T. Existence of positive solution for singular fractional differential equation. Appl Math Comput, 2009, 215(1):2761-2767
[6] Zhang S. Monotone iterative nethod for initial value problem involving Riemann-Liouville fractional derivatives. Nonlinear Anal, 2009, 71(5/6):2087-2093
[7] Benchohra M, Hellar M. Global uniqueness results for fractional partial hyperbolic differential equations with state-dependent delay. Ann Polon Math, 2014, 110(3):259-281
[8] Belmekki M, Nieto Juan J, Rodriguez-Lopez R. Existence of periodic solution for a nonlinear fractional differential equations. Boundary Value Problem, 2009, Article ID 324561, 18 pages
[9] Nieto Juan J. Maximum principles for fractional differential equations derived from Mittag-Leffler functions. Appl Math Lett, 2010, 23(10):1248-1251
[10] Wei Z, Dong W, Che J. Periodic boundary value problems for fractional differential equations involving a Riemann-Liouville fractional derivative. Nonlinear Anal, 2010, 73(10):3232-3238
[11] Nieto Juan J. Comparision results for periodic boundary value problem of fractional differential equations. Fractional Differential Calculus, 2011, 1(1):99-104
[12] Karami H, Babakhani A, Baleanu D. Existence results for a class of fractional differential equations with periodic boundary conditions and with delay. Abstract and Appllied Analysis, 2013, Article ID 176180, 8 pages
[13] 李福义, 冯锦峰, 沈沛龙. 一类减算子的不动点定理及其应用. 数学学报, 1999, 42(2):193-196
带有奇异非线性项的分数微分方程周期解的存在性与唯一性
冯育强, 王蔚敏, 李寿贵