数学物理学报  2016, Vol. 36 Issue (4): 601-622   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
MarkoKostić
李成刚
李淼
抽象多项Riemann-Liouville分数阶微分方程
MarkoKostić1, 李成刚2, 李淼3     
1. University of Novi Sad, Faculty of Technical Sciences, Trg Dositeja Obradovica 6, 21125 Novi Sad, Serbia ;
2. 西南交通大学峨眉校区基础课部 四川峨眉山 614202 ;
3. 四川大学数学学院 成都 610065
摘要:该文研究如下抽象多项分数阶微分方程 $ D_t^{\alpha_{n}} u(t) + \sum \limits_{j=1}^{n-1}A_{j}D_t^{\alpha_{j}} u(t)= AD_t^{\alpha} u(t)+f(t), \quad t \in (0,\tau),\\ $ 其中$n\in {\mathbb N}\setminus \{1\}, $算子$A, A_{1}, \cdot \cdot \cdot, A_{n-1}$为复Banach空间$E$上的闭线性算子, $0 \leq \alpha_{1} < \cdots < \alpha_{n}, $ $0\leq \alpha < \alpha_{n}, $ $0 < \tau \leq \infty, $ $f(t)$为 $E$ -值函数, $D_{t}^{\alpha}$表示$\alpha$阶Riemann-Liouville分数阶导数[5].延续着作者先前在文献[22, 24-25]和[34]中的研究工作, 该文引入并系统分析了方程(0.1)的若干类新的 $k$ -正则 $(C_{1}, C_{2})$ -存在和唯一(生成)族, 并对抽象的理论性结果给出了丰富的例子来阐明.
关键词抽象多项分数阶微分方程     Riemann-Liouville分数阶导数     (a, k)-正则 C-豫解族     适定性    
Abstract Multi-Term Fractional Differential Equations with Riemann-Liouville Derivatives
Marko Kostić1, Li Chenggang2, Li Miao3     
1. University of Novi Sad, Faculty of Technical Sciences, Trg Dositeja Obradovića 6, 21125 Novi Sad, Serbia ;
2. Department of Foundational Courses, Southwest Jiaotong University-Emei, Sichuan Emeishan 614202 ;
3. Department of Mathematics, Sichuan University, Chengdu 610065
Abstract: In this paper, we investigate the following abstract multi-term fractional differential equation $ D_t^{\alpha_{n}} u(t) + \sum \limits_{j=1}^{n-1}A_{j}D_t^{\alpha_{j}} u(t) = AD_t^{\alpha} u(t)+f(t), \quad t \in (0,\tau), $ where $n\in {\mathbb N}\setminus \{1\}, $ $A$ and $A_{1}, \cdot \cdot \cdot, A_{n-1}$ are closed linear operators on a complex Banach space $E, $ $0 \leq \alpha_{1} < \cdots < \alpha_{n}, $ $0\leq \alpha < \alpha_{n}, $ $0 < \tau \leq \infty, $ $f(t)$ is an $E$-valued function, and $D_{t}^{\alpha}$ denotes the Riemann-Liouville fractional derivative of order $\alpha$ ([5]). We introduce and systematically analyze several new types of $k$-regularized $(C_{1}, C_{2})$-existence and uniqueness (propagation) families for (0.1), continuing in such a way our previous researches raised in [22, 24-25] and [34]. Plenty of various examples illustrates our abstract results.
Key words: Abstract multi-term fractional differential equations     Riemann-Liouville fractional derivatives     (a, k)-Regularized C-resolvent families     Well-posedness    
1 引言及预备知识

工程、数学物理、化学等领域中的许多抽象时间分数阶方程可以统一表述成形如(0.1)式的抽象Cauchy问题的形式.现有的研究抽象多项时间分数阶微分方程的文献中仅限于分数阶导数是Caputo型导数的形式, 故本文将对具有Riemann-Liouville分数阶导数的方程(0.1)发展一类算子解法.

与参考文献[22, 24-25]和[34]相比较, 本文对问题(0.1)的研究将会面临新的困难, 即不得不对初值的设置加以细致考察.如后文所述, 我们把此问题分成三个重要的情形: (SC1), (SC2)和(SC3).一般地, 对(SC3)的分析非常繁复, 所以我们着重于对(SC1)和(SC2)的分析研究.我们在定义2.1和定义2.2中引入了方程(0.1)的mild解和强解的概念, 类似于文献[22]的方式, 在定义2.3中引入了方程(0.1)的$k$ -正则$C_{1}$ -存在生成族, $k$ -正则$C_{2}$ -存在生成族, $k$ -正则$C$ -豫解生成族.在定理2.1和定理2.2中, 我们研究了前述生成族的结构性质.由肖体俊和梁进在文献[49, p54-55, 引理2.2.1]中证明的命题, 最近在文献[24, 定理3.1, 注3.1]中已经被推广到具有Caputo导数的抽象时间分数阶微分方程的情形.利用几乎相同的想法, 我们在定理2.3和注记2.1中证明了问题(0.1)的强解的存在性的若干充分性条件(定理3.1(ⅲ)中证明了唯一性).在2.1节中, 我们延续论文献[25] (亦可见文献[21, 2.10.4节])中的研究, 考虑具有Riemann-Liouville分数阶导数的某些特定类型的抽象多项分数阶方程的研究中$(a, k)$ -正则$C$-豫解族的可能应用(在此处略提一下, 关于抽象二阶微分方程的适定性, 文献[25, 定理2.1]中运用$C$ -正则半群的方法比文献[49, 定理6.3.1]中运用积分半群的方法能提供更好的结果).

在第3章中, 我们考察在问题(0.1)的研究中各类$k$ -正则$(C_{1}, C_{2})$ -存在和唯一族的可能应用(更多的细节参见文献[24, 第3-4章]).在对带阻尼的Klein-Gordon方程的Riemann-Liouville分数阶导数的类比(例3.3)的分析中, 我们着重运用扰动型结果(参见文献[24, 定理2.3]).在例2.1, 例3.1和例3.2中阐明了对本文的抽象的结论的应用, 其中例2.1研究了来自于神经细胞中的反常电扩散(参见文献[30-33]和[50])的分数阶电缆方程模型.本文第一作者已在文献[28-29]中研究了Caputo导数的退化多项分数阶微分方程及其超循环和拓扑混合性质(更多的有关抽象时间分数阶方程的性质, 可参见文献[1-2, 4-9, 11-14, 16-17, 19-27, 30-40, 42-47, 50]).我们将在后续研究中考察Riemann-Liouville导数的退化多项分数阶微分方程.

如同Caputo导数的抽象多项方程的情形, 在对$L^{p}$空间上的Riemann-Liouville分数阶导数的抽象多项分数阶微分方程的研究中是很难应用Fourier乘子定理的.甚至是Riemann-Liouville分数阶导数的标量型多项分数阶微分方程的解的形式都有可能非常复杂.如下例所示, 考虑时间分数阶电报方程

$ D_{t}^{2\alpha}u(t, x) + cD_{t}^{\alpha}u(t, x) =a^2\frac{\partial^2}{\partial x^2}u(t, x), \quad t >0, \\ (g_{2-\alpha}*u)'_t(0, x)=f(x), \quad (g_{2-\alpha}*u)(0, x)=0, $ (1.1)

其中$f(\cdot)$属于${\mathbb R}$上的速降函数, $c\in {\mathbb R} \setminus \{0\}.$先对时间变量$t$作Laplace变换

$ ({\cal L} u)(s, x)= \int_0^ \infty {\rm e}^{-st} u(t, x){\rm d}t. $

应用如下的$\alpha$-阶Riemann-Liouville分数阶导数的Laplace变换公式[5, (1.22)式]

$ \bigl({\cal L}D_t^\alpha u\bigr)(s, x)=s^\alpha ({\cal L} u)(s, x)- \sum_{k=0}^{\lceil \alpha \rceil-1} (g_{\lceil \alpha \rceil-\alpha}*u)_t^{(k)}(0)s^{\lceil \alpha \rceil-1-k}, $

易得

$ s^{2\alpha} ({\cal L} u)(s, x) + c s^\alpha ({\cal L} u)(s, x) - a^2 \frac{\partial^2}{\partial x^2}({\cal L} u)(s, x)= f(x). $

接着对上述方程两边关于空间变量$x$作Fourier变换

$ s^{2\alpha} \widehat{({\cal L} u)}(s, \xi) + c s^\alpha \widehat{({\cal L} u)}(s, \xi) +4\pi^2\xi^2 a^2 \widehat{({\cal L} u)}(s, \xi)= \hat{f}(\xi), $

故得

$ \widehat{({\cal L} u)}(s, \xi) = \frac{\hat{f}(\xi)}{s^{2\alpha}+ c s^\alpha+4\pi^2 \xi^2a^2}. $

作Laplace逆变换, 可得

$ \hat{u}(t, \xi)= \hat{f}(\xi){\mathcal L}^{-1}\Big(\frac{1}{s^{2\alpha}+ c s^\alpha+4\pi^2\xi^2 a^2}\Big)(t). $

再应用文献[45, 公式(30)]

$ \begin{eqnarray*} &&{\mathcal L}^{-1}\Big(\frac{1}{s^{2\alpha}+ c s^\alpha+4\pi^2\xi^2 a^2}\Big)(t)\\ &=& \frac{t^{\alpha -1}}{\sqrt{c^2 - 16\pi^2 \xi^2 a^2}}\Big[E_{\alpha, \alpha} \Big(\frac{1}{2}(-c +\sqrt{c^2-16\pi^2 \xi^2 a^2})t^\alpha\Big) \\ &&-E_{\alpha, \alpha} \Big(\frac{1}{2}(-c-\sqrt{c^2 - 16\pi^2 \xi^2 a^2})t^\alpha\Big)\Big] \end{eqnarray*} $

易得方程(1.1)的解为

$ \begin{eqnarray*} u(t, x) &=& \int_{-\infty}^\infty f(x-y) \int_{-\infty}^{\infty} {\rm e}^{2\pi{\rm i} y \xi} \frac{t^{\alpha -1}}{\sqrt{c^2 - 16\pi^2 \xi^2 a^2}} \\ && \times \Big[E_{\alpha, \alpha} \Big(\frac{1}{2}(-c +\sqrt{c^2-16\pi^2 \xi^2 a^2})t^\alpha\Big) \\ &&-E_{\alpha, \alpha} \Big(\frac{1}{2}(-c-\sqrt{c^2 - 16\pi^2 \xi^2 a^2})t^\alpha\Big)\Big]{\rm d}\xi {\rm d}y. \end{eqnarray*} $

简便起见, 本文的研究仅限于在Banach空间的框架内(注意当$E$为一般的序列完备局部凸空间时, 定义2.3和定义3.1中的概念, 以及例2.1仍然是有意义的).本文将会使用如下标准的记号. $E$$L(E)$分别表示非平凡Banach和从$E$$E$的连续线性映射组成的空间. $I$表示$E$上的恒等算子. $\|x\|$表示元素$x\in E$的范数. $A$表示作用于$E$的闭线性算子. $C\in L(E)$表示单射.卷积映射$\ast$表示$f\ast g(t):=\int_{0}^{t}f(t-s)g(s){\rm d}s.$定义域, 豫解集和$A$的值域分别为$D(A), $ $\rho(A)$$R(A).$本文亦用$A$表示它的图, 在本文中不会引起混淆.定义$A$$C$ -豫解集$\rho_{C}(A)$

$ \rho_{C}(A):=\Bigl \{\lambda \in {\mathbb C} : \lambda -A~ \mbox{ 是 单射, 且 }~ (\lambda-A)^{-1}C\in L(E)\Bigr \}. $

$F$$E$的线性子空间, 算子$A$$F$中的“部分” $A_{|F}$为线性算子$D(A_{|F}):=\{x\in D(A)\cap F : Ax\in F\}$, $A_{|F}x:=Ax, $ $x\in D(A_{|F}).$

$s\in {\mathbb R}$, 令$\lfloor s \rfloor:=\sup \{ l\in {\mathbb Z} : s\geq l \}$, $\lceil s \rceil:=\inf \{ l\in {\mathbb Z} : s\leq l \}.$设记号${\mathbb N}_{l}:=\{1, \cdots, l\}, $ ${\mathbb N}_{l}^{0}:=\{0, 1, \cdots, l\}, $ $0^{\zeta}:=0, $ $g_{\zeta}(t):=t^{\zeta-1}/\Gamma(\zeta)$ ($\zeta>0, $ $t>0$), $g_{0}(t):=$表示$\delta$ -分布. $\chi_{S}(\cdot)$表示集合$S$的特征函数.若$\delta \in (0, \pi], $定义$\Sigma_{\delta}:=\{\lambda \in {\mathbb C} : \lambda \neq 0, \ |\arg \lambda| < \delta \}.$$0 < \tau \leq \infty, $ $I=(0, \tau), $ Sobolev空间$W^{m, 1}(I : E)$[5, p7]

$ \begin{eqnarray*} W^{m, 1}(I :E):=\biggl \{ f & | & \exists \varphi \in L^{1}(I : E), \ \exists c_{k}\in {\mathbb C} \ (0\leq k\leq m-1), \\ && f(t)=\sum_{k=0}^{m-1}c_{k}g_{k+1}(t)+\bigl(g_{m}\ast \varphi \bigr)(t) \mbox{ 对几乎所有}~ t\in (0, \tau)\biggr\}, \end{eqnarray*} $

其中$\varphi(t)=f^{(m)}(t)$为分布意义下的导数, 且$c_{k}=f^{(k)}(0)$ ($0\leq k\leq m-1$).设$0 < \tau \leq \infty$, 且$k\in L_{\rm loc}^{1}([0, \tau)), $称函数$k(t)$$[0, \tau)$上的核, 当且仅当对任意$f\in C([0, \tau))$都有$\int^{t}_{0}k(t-s)f(s){\rm d}s=0, $ $t\in [0, \tau)$蕴含$f(t)=0, $ $t\in [0, \tau).$

在部分章节中如下条件将会被用到:

(P1) $h(t) :[0, \infty) \rightarrow E$是可Laplace变换的, 即$h\in L_{\rm loc}^{1}( [0, \infty) : E)$且存在$\beta \in {\mathbb R}$使得$\tilde{h}(\lambda):={\mathcal L}(h)(\lambda):=\lim\limits_{b \rightarrow \infty}\int_{0}^{b}{\rm e}^{-\lambda t}h(t){\rm d}t:=\int ^{\infty}_{0}{\rm e}^{-\lambda t}h(t){\rm d}t$对所有$\lambda \in {\mathbb C}$ ($\Re\lambda>\beta$)成立.记${abs}(h):=inf\{ \Re\lambda : \tilde{h}(\lambda) \mbox{ 存在} \}, $且以${\mathcal L}^{-1}$表示Laplace逆变换.

包含关系$H(\lambda) \in LT_{E}$表示存在函数$h(t) :[0, \infty) \rightarrow E$满足条件(P1), 及$a>{abs}(h)$使得$\tilde{h}(\lambda)=H(\lambda)$, $\lambda>a.$更多的有关向量值Laplace变换的知识见参考文献[3]和[49].文献[21]包含了关于$(a, k)$ -正则$C$-豫解族的最近的结果的综述.

定义Mittag-Leffler函数$E_{\alpha, \beta}(z)$

$ E_{\alpha, \beta}(z) : = \sum_{n=0}^\infty \frac{z^n}{\Gamma(\alpha n + \beta)} = \frac{1}{2\pi {\rm i}} \int_L \frac{ \mu^{\alpha - \beta} e^\mu}{\mu^\alpha -z} \, {\rm d}\mu, \quad \alpha, \ \beta>0, \ z \in \mathbb{C}, $ (1.2)

其中$L$表示起点和终点均为$ -\infty$, 环绕圆盘$|t| \le |z|^{1/\alpha}$的逆时针方向的Hankel围道.记$E_{\alpha}(z) := E_{\alpha, 1}(z)$ ($z\in {\mathbb C}$).对$\gamma \in (0, 1), $则Wright函数(参见文献[5, 21, 39]) $\Phi_{\gamma}(\cdot)$定义为

$ \Phi_{\gamma}(t):={{\mathcal L}^{-1}}\bigl(E_{\gamma}(-\lambda)\bigr)(t), \quad t\geq 0. $

$\alpha>0$, 且$m=\lceil \alpha \rceil.$则函数$f\in L^{1}(I : E)$$\alpha>0$阶Riemann-Liouville分数阶积分定义为

$ J_t^\alpha f(t):= \bigl(g_\alpha*f\bigr)(t), \quad t>0. $

对任意满足条件$g_{m -\alpha}\ast f\in W^{m, 1}(I: E)$的函数$f\in L^{1}(I : E)$, 定义$\alpha>0$阶Riemann-Liouville分数阶导数为

$ D_t^\alpha f(t):=\frac{{\rm d}^m}{{\rm d}t^m} \bigl(g_{m-\alpha}*f\bigr)(t) = D_t^m J_t^{m-\alpha}f(t), \quad t>0. $

由文献[5, 定理1.5], 对$f \in L^1(I:E)$, Riemann-Liouville分数阶积分和导数满足如下等式

$ J_t^\alpha J_t^\beta f(t) = J_t^{\alpha + \beta} f(t), \quad D_t^\alpha J_t^\alpha f(t)=f(t), $ (1.3)

且对满足$g_{m -\alpha}\ast f\in W^{m, 1}(I : E)$$f\in L^{1}(I : E)$, 有

$ J_t^\alpha D_t^\alpha f(t)= f(t) - \sum_{k=0}^{m-1} (g_{m-\alpha}*f)^{(k)}(0) g_{\alpha +k +1-m}(t). $ (1.4)

本文以记号${\mathbf D}_{t}^{\alpha}u(t)$表示函数$u(t)$$\alpha$-阶Caputo导数[5].

最后, 考虑如下非齐次方程

$ u(t)+\sum \limits^{n-1}_{j=1}\bigl(g_{\alpha_{n}-\alpha_{j}}\ast A_{j} u\bigr)(t)=f(t)+\bigl(g_{\alpha_{n}-\alpha}\ast Au\bigr)(t), \quad t\in [0, T], $ (1.5)

其中$T>0$, $f\in C([0, T]:E).$按照文献[22, 定义2.7]的方式, 定义函数$u\in C([0, T]: E)$

(ⅰ) 方程(1.5)的强解, 当且仅当$A_{j}u\in C([0, T]: E), $ $j\in {{\mathbb N}_{n-1}^{0}}$且(1.5)式对每一$t\in [0, T]$都成立;

(ⅱ) 方程(1.5)的mild解, 当且仅当$(g_{\alpha_{n}-\alpha_{j}}\ast u)(t)\in D(A_{j}), $ $t\in [0, T], $ $j\in {{\mathbb N}_{n-1}^{0}}$, 且

$ u(t)+\sum \limits^{n-1}_{j=1}A_{j}\bigl(g_{\alpha_{n}-\alpha_{j}}\ast u\bigr)(t)=f(t)+A\bigl(g_{\alpha_{n}-\alpha}\ast u\bigr)(t), \quad t\in [0, T]. $
2 方程(0.1)的$k$ -正则$(C_{1}, C_{2})$ -存在和唯一生成族的基本结构性质

如摘要中所述, 假设$E$为完备Banach空间, $n\in {\mathbb N}\setminus \{1\}, $ $A$, $A_{1}, \cdots, A_{n-1}$$E$上的闭线性算子, $0 \leq \alpha_{1} < \cdots < \alpha_{n}$, $0\leq \alpha < \alpha_{n}.$$\alpha_{0}:=\alpha, $ $m:=\lceil \alpha \rceil $, 且$A_{0}:=A.$

定义 2.1 设$0 < \tau \leq \infty$, 且$f\in L^{1}((0, \tau): E).$若函数$u\in L^{1}((0, \tau): E)$, $g_{m_{j}-\alpha_{j}}\ast u \in W^{m_{j}, 1}((0, \tau):E)$ ($0\leq j\leq n$), $A_{j}D_t^{\alpha_{j}} u(t)\in L^{1}((0, \tau) :E)$ ($0\leq j\leq n-1$), 且方程(0.1)对几乎所有$t\in (0, \tau)$成立, 则称函数$u$为方程(0.1)的强解.

以下引入方程(0.1)的mild解的概念, 并对方程(0.1)赋予相应的初值条件.设$u(t)$为方程(0.1)的强解, 对方程(0.1)两边积分$\alpha_{n}$ -次, 应用公式(1.4)以及算子$A_{j}$ ($j\in {\mathbb N}_{n-1}^{0}$)的闭性, 可得对所有$t\in [0, \tau)$都有

$ \begin{eqnarray*} &&\bigg[u(t)-\sum \limits_{i=0}^{m_{n}-1}\bigl( g_{m_{n}-\alpha_{n}} \ast u\bigr)^{(i)}(0)g_{\alpha_{n}+i+1-m_{n}}(t)\bigg] \\\nonumber && +\sum \limits_{j=1}^{n-1} g_{\alpha_{n} - \alpha_{j}} \ast A_{j} \bigg[u(t)- \sum \limits_{i=0}^{m_{j}-1}\bigl( g_{m_{j}-\alpha_{j}} \ast u\bigr)^{(i)}(0)g_{\alpha_{j}+i+1-m_{j}}(t)\bigg] \\\label{tm} & =& g_{\alpha_{n} - \alpha} \ast A \bigg[u(t)-\sum \limits_{i=0}^{m-1}\bigl( g_{m-\alpha} \ast u\bigr)^{(i)}(0)g_{\alpha+i+1-m}(t)\bigg]+\bigl(g_{\alpha_{n}}\ast f\bigr)(t), \end{eqnarray*} $ (2.1)

此式亦蕴含着方程(0.1)的任意强解都满足条件$u\in C([0, \tau) : E).$

定义 2.2 设$0 < \tau \leq \infty, $ $f\in L^{1}((0, \tau): E).$若函数满足条件$u\in C([0, \tau) : E) \cap L^{1}((0, \tau): E)$, $g_{m_{j}-\alpha_{j}}\ast u \in W^{m_{j}, 1}((0, \tau):E)$ ($0\leq j\leq n$), $A_{j}(g_{\alpha_{n}}\ast D_t^{\alpha_{j}} u(\cdot))(\cdot)\in C([0, \tau) :E)$ ($0\leq j\leq n-1$), 且

$ \begin{eqnarray*} &&\bigg[u(t)-\sum \limits_{i=0}^{m_{n}-1}\bigl( g_{m_{n}-\alpha_{n}} \ast u\bigr)^{(i)}(0)g_{\alpha_{n}+i+1-m_{n}}(t)\bigg] +\sum \limits_{j=1}^{n-1}A_{j}\Bigl(g_{\alpha_{n}}\ast D_t^{\alpha_{j}} u(\cdot)\Bigr)(t) \\ \label{tm1} & =& A\Bigl(g_{\alpha_{n}}\ast D_t^{\alpha}u(\cdot)\Bigr)(t)+\bigl(g_{\alpha_{n}}\ast f\bigr)(t), \quad t\in [0, \tau), \end{eqnarray*} $ (2.2)

则称函数$u(t)$是方程(0.1)的mild解.

由前述的计算, 方程(0.1)的任意强解都是此方程的mild解; 一般地, 反之不一定成立.注意到对任意$t\in [0, \tau), $方程(2.2)可写为如下形式

$ \begin{eqnarray} \nonumber &&\bigg[u(t)-\sum \limits_{i=0}^{m_{n}-1}\bigl( g_{m_{n}-\alpha_{n}} \ast u\bigr)^{(i)}(0)g_{\alpha_{n}+i+1-m_{n}}(t)\bigg] \\\nonumber && +\sum \limits_{j=1}^{n-1} A_{j} \bigg(g_{\alpha_{n} - \alpha_{j}} \ast \bigg[u(\cdot)- \sum \limits_{i=0}^{m_{j}-1}\bigl( g_{m_{j}-\alpha_{j}} \ast u\bigr)^{(i)}(0)g_{\alpha_{j}+i+1-m_{j}}(\cdot)\bigg]\bigg)(t) \\\label{tm-prim} & =& A \bigg( g_{\alpha_{n} - \alpha} \ast \bigg[u(\cdot)-\sum \limits_{i=0}^{m-1}\bigl( g_{m-\alpha} \ast u\bigr)^{(i)}(0)g_{\alpha+i+1-m}(\cdot)\bigg]\bigg)(t)+\bigl(g_{\alpha_{n}}\ast f\bigr)(t), \end{eqnarray} $ (2.3)

$$ {\mathcal T}_{(0.1)}:=\left\{ \begin{array}{l} 1, \quad \mbox{ 若存在 }~ j\in {\mathbb N}_{n}^{0}~ \mbox{ 使得 }~ \alpha_{j}\in {\mathbb N}, \\ 0, \quad \mbox{ 其它, } \end{array} \right. $$

$S:=\{j\in {\mathbb N}_{n}^{0} : \alpha_{j}\in {\mathbb N}\}.$方程(0.1)可分为如下三类:

(SC1) $\alpha_{n}>1:$对每一$i\in {\mathbb N}_{m_{n}-1}$定义集合${\mathcal D}_{i}$${\mathcal D}_{i}:=\{j\in {\mathbb N}_{n}^{0} : m_{j}-1\geq i\}.$注意到$n\in {\mathcal D}_{i}$ ($i\in {\mathbb N}_{m_{n}-1}$), 且${\mathcal D}_{m_{n}-1}\subseteq \cdots \subseteq {\mathcal D}_{1}.$$S_{i}:=\{m_{j}-\alpha_{j} : j\in {\mathcal D}_{i}\}$, $s_{i}:=\mbox{card}(S_{i}).$则有$S_{i}\subseteq [0, 1)$$S_{i}$写为如下形式

$ S_{i}=\bigl\{ a_{i, 1}, \cdots, a_{i, s_{i}}\bigr\}, $

其中$0\leq a_{i, 1} < \cdots < a_{i, s_{i}}\leq 1$ ($i\in {\mathbb N}_{m_{n}-1}$).定义${\mathcal D}_{i}^{l}:=\{j\in {\mathcal D}_{i} : m_{j}-\alpha_{j}=a_{i, l}\}$ ($i\in {\mathbb N}_{m_{n}-1}, $ $1\leq l\leq s_{i}$).则对$i\in {\mathbb N}_{m_{n}-1}$, 对$(g_{m_{j}-\alpha_{j}}\ast u)^{(i)}(0)$项可引入$s_{i}$个初值$x_{i, 1}, \cdots, x_{i, s_{i}}$, $j\in {\mathcal D}_{i}.$进一步地, 若存在$j\in {\mathbb N}_{n}^{0}$使得$\alpha_{j}\in {\mathbb N}, $即若$S\neq \emptyset, $则必须对$(g_{0}\ast u)(0)\equiv u(0)$项引入新的初值$x_{0}.$

(SC2) $\alpha_{n}=1:$仅需对$(g_{0}\ast u)(0)\equiv u(0)$引入一个初值.

(SC3) $\alpha_{n} < 1:$则对方程(0.1)不考虑初始条件.

定义

$ {\mathcal B}_{(0.1)}:=\left\{ \begin{array}{ll} s_{1}+\cdots +s_{m_{n}-1}+{\mathcal T}_{(0.1)}, \quad & \mbox{ 若 }~ \alpha_{n}>1, \\ 1, & \mbox{ 若 }~ \alpha_{n}=1, \\ 0, & \mbox{ 若 }~ \alpha_{n} < 1. \end{array} \right. $

对求和“再求和”, 可得方程(0.1)有${\mathcal B}_{(0.1)}$个初始条件.

为了更好地理解, 现考虑子类(SC1).在方程(2.1)中取$x_{i, l}=(g_{m_{j}-\alpha_{j}}\ast u)^{(i)}(0)$, $j\in {\mathcal D}_{i}^{l}, $且取其它初值为零, 可得对所有$ t\in [0, \tau)$

$ \Bigl[u(t)-\chi_{{\mathcal D}_{i}^{l}}(n) g_{\alpha_{n}+i+1-m_{n}}(t)x_{i, l}\Bigr] +\sum \limits_{j=1}^{n-1} g_{\alpha_{n}-\alpha_{j}} \ast A_{j}\Bigl[u(t)-\chi_{{\mathcal D}_{i}^{l}}(j)g_{\alpha_{j}+i+1-m_{j}}(t)x_{i, l}\Bigr] \\ \label{tm2} =g_{\alpha_{n}-\alpha} \ast A\Bigl[u(t)-\chi_{{\mathcal D}_{i}^{l}}(0)g_{\alpha+i+1-m}(t)x_{i, l}\Bigr] . $ (2.4)

$S\neq \emptyset, $则在方程(2.1)中取$u(0)=x_{0}$, 且对$i\in {\mathbb N}_{m_{n}-1}$$1\leq l\leq s_{i}, $选择$x_{i, l}$为零.可类似地得到对所有$ t\in [0, \tau)$

$ \begin{eqnarray} \nonumber & & \Bigl[u(t)-\chi_{S}(n)x_{0}\Bigr] +\sum \limits_{j=1}^{n-1} g_{\alpha_{n}-\alpha_{j}} \ast A_{j}\Bigl[u(t)-\chi_{S}(j)x_{0}\Bigr] \\ \label{tm3} & =&g_{\alpha_{n}-\alpha} \ast A\Bigl[u(t)-\chi_{S}(0)x_{0}\Bigr]. \end{eqnarray} $ (2.5)

仅为进一步地分析, 现在假设$0 < \tau \leq \infty, $ $K(t)\neq 0$, $K(t)\in L_{\rm loc}^{1}([0, \tau))$, 且$k(t)=\int^{t}_{0}K(s){\rm d}s, $ $t\in [0, \tau).$在上面的方程中以$K(t)$作卷积, 应用类似于抽象多项Caputo分数阶方程的研究中的方法, 给出如下定义.

定义 2.3 设$0 < \tau\leq \infty, $ $k\in C([0, \tau)), $ $C, \ C_{1}, \ C_{2}\in L(E), $且设$C$$C_{2}$是单射.

(ⅰ) (SC1) 设对所有$i\in {\mathbb N}_{m_{n}-1}$$l\in {\mathbb N}_{s_{i}}, $都有$(R_{i, l}(t))_{t\in [0, \tau)}\subseteq L(E)$强连续.且对所有$t\in [0, \tau), $ $x\in E, $ $i\in {\mathbb N}_{m_{n}-1}$$l\in {\mathbb N}_{s_{i}}, $下述泛函方程

$ \begin{eqnarray} \nonumber && \Bigl[R_{i, l}(t)x-\chi_{{\mathcal D}_{i}^{l}}(n) \bigl(k\ast g_{\alpha_{n}+i-m_{n}}\bigr)(t)C_{1}x\Bigr] \\ \nonumber & & +\sum \limits_{j=1}^{n-1} A_{j}\Bigl[g_{\alpha_{n}-\alpha_{j}} \ast \Bigl( R_{i, l}(\cdot)x- \chi_{{\mathcal D}_{i}^{l}}(j)\bigl(k\ast g_{\alpha_{j}+i-m_{j}}\bigr)(\cdot)C_{1}x\Bigr)\Bigr](t) \\ \label{jednakost} & =& A\Bigl[g_{\alpha_{n}-\alpha} \ast \Bigl( R_{i, l}(\cdot)x-\chi_{{\mathcal D}_{i}^{l}}(0) \bigl(k\ast g_{\alpha+i-m}\bigr)(\cdot)C_{1}x\Bigr)\Bigr](t) \end{eqnarray} $ (2.6)

成立.若$S\neq \emptyset, $则也可引入强连续算子族$(R_{0, 1}(t))_{t\in [0, \tau)}\subseteq L(E)$对所有$t\in [0, \tau)$$x\in E, $满足

$ \begin{eqnarray*} && \Bigl[R_{0, 1}(t)x-\chi_{S}(n) k(t)C_{1}x\Bigr] +\sum \limits_{j=1}^{n-1} A_{j}\Bigl[g_{\alpha_{n}-\alpha_{j}} \ast \Bigl(R_{0, 1}(\cdot)x- \chi_{S}(j)k(\cdot)C_{1}x\Bigr)\Bigr](t) \\& =&A\Bigl[g_{\alpha_{n}-\alpha} \ast \Bigl(R_{0, 1}(\cdot)x-\chi_{S}(0) k(\cdot)C_{1}x\Bigr)\Bigr](t), \end{eqnarray*} $

则称序列$((R_{i, l}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$ ($S=\emptyset$)或序列$((R_{i, l}(t))_{t\in [0, \tau)}, $ $ (R_{0, 1}(t))_{t\in [0, \tau)})$ ${}_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$ ($S\neq \emptyset$)为方程(0.1)的(当$\tau < \infty$时, 是局部的) $k$-正则$C_{1}$-存在生成族.

(SC2) 若强连续算子族$(R(t))_{t\in [0, \tau)}\subseteq L(E)$对所有$t\in [0, \tau)$$x\in E$满足

$ \begin{eqnarray}\label{fgh} \Bigl[R(t)x-k(t)C_{1}x\Bigr]+\sum \limits_{j=1}^{n-1} A_{j}\bigl(g_{\alpha_{n}-\alpha_{j}} \ast R(\cdot)x\bigr)(t)=A\bigl(g_{\alpha_{n}-\alpha} \ast R(\cdot)x\bigr)(t), \end{eqnarray} $ (2.7)

则称$(R(t))_{t\in [0, \tau)}$为方程(0.1)的(当$\tau < \infty$时, 是局部的)$k$-正则$C_{1}$-唯一生成族.

(SC3) 若强连续算子族$(R(t))_{t\in [0, \tau)}\subseteq L(E)$对所有$t\in [0, \tau)$$x\in E$满足

$ \begin{eqnarray*} R(t)x+\sum \limits_{j=1}^{n-1} A_{j}\bigl(g_{\alpha_{n}-\alpha_{j}} \ast R(\cdot)x\bigr)(t)=A\bigl(g_{\alpha_{n}-\alpha} \ast R(\cdot)x\bigr)(t), \end{eqnarray*} $

则称$(R(t))_{t\in [0, \tau)}$为方程(0.1)的(当$\tau < \infty$时, 是局部的)存在生成族.

(ⅱ) (SC1) 设对所有$i\in {\mathbb N}_{m_{n}-1}$, $l\in {\mathbb N}_{s_{i}}, $ $(W_{i, l}(t))_{t\in [0, \tau)}\subseteq L(E)$是强连续的, 且

$ \begin{eqnarray*} && \Bigl[W_{i, l}(t)x-\chi_{{\mathcal D}_{i}^{l}}(n) \bigl(k\ast g_{\alpha_{n}+i-m_{n}}\bigr)(t)C_{2}x\Bigr] \\ & & +\sum \limits_{j=1}^{n-1} g_{\alpha_{n}-\alpha_{j}} \ast \Bigl[W_{i, l}(t)A_{j}x- \chi_{{\mathcal D}_{i}^{l}}(j)\bigl(k\ast g_{\alpha_{j}+i-m_{j}}\bigr)(t)C_{2}A_{j}x\Bigr] \\& =& g_{\alpha_{n}-\alpha} \ast \Bigl[W_{i, l}(t)Ax-\chi_{{\mathcal D}_{i}^{l}}(0) \bigl(k\ast g_{\alpha+i-m}\bigr)(t)C_{2}Ax\Bigr], \end{eqnarray*} $

对所有$i\in {\mathbb N}_{m_{n}-1}, $ $l\in {\mathbb N}_{s_{i}}, $ $t\in [0, \tau)$$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j})$成立.若$S\neq \emptyset, $则也可以引入强连续算子族$(W_{0, 1}(t))_{t\in [0, \tau)}\subseteq L(E)$对所有$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j}), $满足

$ \begin{eqnarray*} && \Bigl[W_{0, 1}(t)x-\chi_{S}(n) k(t)C_{2}x\Bigr] +\sum \limits_{j=1}^{n-1} g_{\alpha_{n}-\alpha_{j}} \ast \Bigl[W_{0, 1}(t)A_{j}x- \chi_{S}(j)k(t)C_{2}A_{j}x\Bigr] \\& =&g_{\alpha_{n}-\alpha} \ast \Bigl[W_{0, 1}(t)Ax-\chi_{S}(0) k(t)C_{2}Ax\Bigr], \quad t\in [0, \tau). \end{eqnarray*} $

则称序列$((W_{i, l}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$ ($S=\emptyset$时)或序列$((W_{i, l}(t))_{t\in [0, \tau)}, $ $ (W_{0, 1}(t))_{t\in [0, \tau)})$ ${}_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$ ($S\neq \emptyset$时)为方程(0.1)的(当$\tau < \infty$时, 是局部的) $k$-正则$C_{2}$-唯一生成族.

(SC2) 若强连续算子族$(W(t))_{t\in [0, \tau)}\subseteq L(E)$对所有$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j})$$t\in [0, \tau), $满足

$ \Bigl[W(t)x-k(t)C_{2}x\Bigr]+\sum \limits_{j=1}^{n-1} \bigl(g_{\alpha_{n}-\alpha_{j}} \ast W(\cdot)C_{2}A_{j}x\bigr)(t)= \bigl(g_{\alpha_{n}-\alpha} \ast W(\cdot)C_{2}Ax\bigr)(t), $

则称$(W(t))_{t\in [0, \tau)}$为方程(0.1)的(当$\tau < \infty$时, 是局部的) $k$-正则$C_{2}$-唯一生成族.

(SC3) 若强连续算子族$(W(t))_{t\in [0, \tau)}\subseteq L(E)$对所有$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j})$$t\in [0, \tau), $满足

$ W(t)x+\sum \limits_{j=1}^{n-1} \bigl(g_{\alpha_{n}-\alpha_{j}} \ast W(\cdot)A_{j}x\bigr)(t)= \bigl(g_{\alpha_{n}-\alpha} \ast W(\cdot)Ax\bigr)(t), $

则称$(W(t))_{t\in [0, \tau)}$为方程(0.1)的(当$\tau < \infty$时, 是局部的)唯一生成族.

(ⅲ) 对方程(0.1)的$k$-正则$C$-唯一生成族组成的$L(E)$中的强连续算子族序列, 若此序列中的任一算子族都与算子$A_{j}$ ($j\in {\mathbb N}_{n-1}^{0}$)和$C$可交换, 且有$CA_{j}\subseteq A_{j}C$ ($j\in {\mathbb N}_{n-1}^{0}$), 则称此算子族序列为方程(0.1)的$k$-正则$C$ -豫解生成族, 简称为方程(0.1)的$k$ -正则$C$ -生成族.

子类(SC3)过于特殊, 本论文将不再考虑此子类.以上各种类型的生成族也可以用纯代数方程的方法定义(见文献[8, 27, 38]), 本论文不对此方面的细节进行研究.

接下来仅考虑方程(0.1)的非退化的算子族.当$k(t)=g_{\zeta+1}(t)$ (其中$\zeta \geq 0)$时, 称$k$ -正则$C$ -豫解生成族为方程(0.1)的$\zeta$ -次积分$C$ -豫解生成族; 方程(0.1)的$0$ -次积分$C$ -豫解生成族简称为方程(0.1)的$C$ -豫解生成族; 相似的术语也将会被应用于方程(0.1)的$C_{1}$ -存在生成族和$C_{2}$ -唯一生成族.单个算子族的诸如局部有界性(指数有界性, 解析性, $\cdots $)等概念可通过直接的方式转化为本文研究的方程(0.1)的$C$ -生成族的情形.当$C=I$ ($C_{2}=I$)时, $k$ -正则$C$ -豫解生成族($C_{2}$ -唯一生成族)也被称为$k$ -正则豫解生成族(唯一生成族).注意到依照文献[22, 例5.2]中的分析, 可以构造出全局(局部)非指数有界的$k$ -正则豫解生成族的例子.

若依照文献[22, 定义2.7]中的同样的方式定义问题(2.4)-(2.5)的mild解和强解, 则有如下成立(简洁起见, 仅考虑子类(SC1)):

(A) 若$S= \emptyset$$((R_{i, l}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$C_{1}$-存在生成族, 则函数$u_{i, l}(t):=R_{i, l}(t)C_{1}x, $ $t\in [0, \tau)$是方程(2.4)的mild解, 其中$x_{i, l}=C_{1}x, $ $1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}.$$S\neq \emptyset$$((R_{i, l}(t))_{t\in [0, \tau)}, \ (R_{0, 1}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$C_{1}$-存在生成族, 则函数$u_{0, 1}(t):=R_{0, 1}(t)C_{1}x, $ $t\in [0, \tau)$是方程(2.5)的mild解, 其中$x_{0}=C_{1}x.$

(B) 若$S=\emptyset, $ $((W_{i, l}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$C_{2}$-唯一生成族, 且还有$A_{j}W_{i, l}(t)x=W_{i, l}(t)A_{j}x, $ $t\in [0, \tau), $ $x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j}), $ $C_{2}A_{j}\subseteq A_{j}C_{2}, $ $j\in {\mathbb N}_{n-1}^{0}, $$u_{i, l}(t):=W_{i, l}(t)C^{-1}_{2}x_{i, l}, $ $t\in [0, \tau)$是方程(2.4)的强解, 其中$1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}, $ $x_{i, l}\in C_{2}(\bigcap\limits_{0\leq j\leq n-1}D(A_{j})).$$S\neq \emptyset, $ $((W_{i, l}(t))_{t\in [0, \tau)}, \ (W_{0, 1}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$C_{2}$-唯一生成族, 且还有(在$S=\emptyset$的情形的假设中再附加上以下假设) $A_{j}W_{0, 1}(t)x=W_{0, 1}(t)A_{j}x, $ $t\in [0, \tau), $ $x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j}), $则函数$u_{0}(t):=W_{0, 1}(t)C^{-1}_{2}x_{0}, $ $t\in [0, \tau)$是方程(2.4)的强解, 其中$1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}, $ $x_{0}\in C_{2}(\bigcap\limits_{0\leq j\leq n-1}D(A_{j})).$

本文的进一步的分析将主要集中于对$A_{j}=c_{j}I$的情形的研究, 其中$c_{j}$ ($1\leq j\leq n-1$)是复常数; 更多的信息可参阅文献[1, 11, 21-24, 34]和[43].我们也称算子$A$是方程(0.1)的$k$ -正则$C$ -豫解生成族(方程(0.1)的$k$ -正则$C_{1}$ -存在生成族, 方程(0.1)的$k$ -正则$C_{2}$ -唯一生成族)的次生成元.这些生成族的所有次生成元组成的集合不必是有限的, 我们也可引入方程(0.1)的$k$ -正则$C$ -豫解生成族的积分生成元的概念(如同Caputo导数的方程的情形一样).各种类型的算子族的次生成元的重要性质已被应用于对不适定Volterra积微分方程的研究, 感兴趣的读者可参阅文献[20-22].为了能够聚焦于Riemann-Liouville分数阶导数的多项方程的本质, 我们顺带提及文献[22, 性质2.3, 2.5, 2.6]的结论可以转化到本文的框架之下, 文献[22, 定理2.8]的结论同样如此, 正如如下的定理所述:

定理 2.1 (SC1) 设$S=\emptyset, $ $((R_{i, l}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$k$-正则$C$-豫解生成族, 满足泛函方程(2.6), 其中$C_{1}=C.$$0 < T < \tau, $ $f\in C([0, T]:E).$若函数$u(t)$是问题(1.5)的强解, 则在$[0, T]$上有

$ \begin{eqnarray*} &&\Bigl[R_{i, l}(\cdot)x-\chi_{{\mathcal D}_{i}^{l}}(n) \bigl(k\ast g_{\alpha_{n}+i-m_{n}}\bigr)(\cdot)C\Bigr]\ast f \\ & =&\sum \limits_{j=1}^{n-1}\Bigl[\chi_{{\mathcal D}_{i}^{l}}(j) \bigl(g_{\alpha_{n}+i-m_{j}}\ast kC\ast A_{j}u\bigr)(\cdot) -\chi_{{\mathcal D}_{i}^{l}}(n)\bigl(g_{2\alpha_{n}-\alpha_{j}+i-m_{n}}\ast kC\ast A_{j}u\bigr)(\cdot)\Bigr] \\ && -\chi_{{\mathcal D}_{i}^{l}}(0)\bigl(g_{\alpha_{n}+i-m}\ast kC\ast Au\bigr)(\cdot) \ \ \bigl( 1\leq i \leq m_{n}-1, 1\leq l\leq s_{i} \bigr). \end{eqnarray*} $

$S\neq \emptyset, $ $((R_{i, l}(t))_{t\in [0, \tau)}, \ (R_{0, 1}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$k$-正则$C$-豫解生成族, 则在$[0, T]$上有如下等式成立

$ \begin{eqnarray*} &&\Bigl[R_{0, 1}(\cdot)x-\chi_{S}(n) k(\cdot)C\Bigr]\ast f \\ & =&-\chi_{S}(0)\bigl(g_{\alpha_{n}-\alpha}\ast kC\ast Au\bigr)(\cdot)+\sum \limits_{j=1}^{n-1} \bigl(\chi_{S}(j)-\chi_{S}(n)\bigr) \bigl(g_{\alpha_{n}-\alpha_{j}}\ast kC\ast A_{j}u\bigr)(\cdot). \end{eqnarray*} $

(SC2) 设$(R(t))_{t\in [0, \tau)}$是方程(0.1)的$k$-正则$C$-豫解生成族, 满足泛函方程(2.7), 其中$C_{1}=C.$$0 < T < \tau, $ $f\in C([0, T]:E).$若函数$u(t)$是问题(1.5)的强解, 则在$[0, T]$上有如下等式成立

$ \Bigl[R(\cdot)x-k(\cdot)C\Bigr]\ast f =\bigl(g_{\alpha_{n}-\alpha}\ast kC\ast Au\bigr)(\cdot)-\sum \limits_{j=1}^{n-1} \bigl(g_{\alpha_{n}-\alpha_{j}}\ast kC\ast A_{j}u\bigr)(\cdot). $

应用Laplace变换可证明如下的基本结果.

定理 2.2 设$k(t)$满足(P1), 且$\omega \geq \max(0, \mbox{abs}(k)).$ $A$$E$上的闭线性算子, $C_{1}, \ C_{2}\in L(E), $且设$C_{2}$是单射.记$P_{\lambda}:= \lambda^{\alpha_{n}-\alpha}+\sum\limits _{j=1}^{n-1}\lambda^{\alpha_{j}-\alpha}A_{j}-A, $ $\lambda \in {\mathbb C} \setminus \{0\}.$

(i) 设$A_{j}\in L(E), $ $j\in {{\mathbb N}_{n-1}}.$

(SC1) 设$S=\emptyset, $ $((R_{i, l}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$$L(E)$中的强连续算子族序列, 集合$\{{\rm e}^{-\omega t}R_{i, l}(t) : t\geq 0\}$$L(E)$中有界, 其中$1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}.$

$ ((R_{i, l}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}} $

是方程(0.1)的$k$ -正则$C_{1}$ -存在生成族的充要条件是如下等式成立

$ \begin{eqnarray} \nonumber && \widetilde{R_{i, l}}(\lambda)x +\sum \limits_{j=1}^{n-1}\lambda^{\alpha_{j}-\alpha_{n}}A_{j}\widetilde{R_{i, l}}(\lambda)x\\ \nonumber &=& \chi_{ {\mathcal D}_{i}^{l}}(n)\tilde{k}(\lambda)\lambda^{m_{n}-\alpha_{n}-i}C_{1}x +\sum \limits_{j=1}^{n-1}\chi_{ {\mathcal D}_{i}^{l}}(j)\tilde{k}(\lambda)\lambda^{m_{j}-\alpha_{n}-i}A_{j}C_{1}x \\ \label{cvb} & & +\lambda^{\alpha-\alpha_{n}}A\Bigl[\widetilde{R_{i, l}}(\lambda)x -\chi_{ {\cal D}_{i}^{l}}(0)\tilde{k}(\lambda)\lambda^{m-\alpha-i}C_{1}x\Bigr], \quad x\in E, \ \Re \lambda>\omega. \end{eqnarray} $ (2.8)

而当$S\neq \emptyset $时, 设$((R_{i, l}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$满足上面的的假设, 且设$(R_{0, 1}(t))_{t\geq 0}$是强连续的且集合$\{{\rm e}^{-\omega t}R_{0, 1}(t) : t\geq 0\}$$L(E)$中有界, 则$((R_{i, l}(t))_{t\geq 0}, (R_{0, 1}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$k$ -正则$C_{1}$ -存在生成族的充要条件是(2.8)式和如下的等式成立

$ \begin{eqnarray*} && \widetilde{R_{0, 1}}(\lambda)x +\sum \limits_{j=1}^{n-1}\lambda^{\alpha_{j}-\alpha_{n}}A_{j}\widetilde{R_{0, 1}}(\lambda)x\\ \nonumber &=& \chi_{S}(n)\tilde{k}(\lambda)C_{1}x +\sum \limits_{j=1}^{n-1}\chi_{S}(j)\tilde{k}(\lambda)A_{j}C_{1}x \\ \nonumber && +\lambda^{\alpha-\alpha_{n}}A\Bigl[\widetilde{R_{0, 1}}(\lambda)x -\chi_{S}(0)\tilde{k}(\lambda)C_{1}x\Bigr], \quad x\in E, \ \Re \lambda>\omega. \end{eqnarray*} $

(SC2) 设$(R(t))_{t\geq 0}$是强连续的且集合$\{{\rm e}^{-\omega t}R(t) : t\geq 0\}$$L(E)$中有界.则$(R(t))_{t\geq 0}$是方程(0.1)的$k$ -正则$C_{1}$ -存在生成族的充要条件是如下等式成立

$ P_{\lambda}\tilde{R}(\lambda)x=\tilde{k}(\lambda)\lambda^{\alpha_{n}-\alpha}C_{1}x, \quad x\in E, \ \Re \lambda>\omega . $

(ⅱ) (SC1) 设$S=\emptyset, $ $((W_{i, l}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$$L(E)$中的强连续算子族序列, 集合$\{{\rm e}^{-\omega t}W_{i, l}(t) : t\geq 0\}$$L(E)$中有界, 其中$1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}.$$((W_{i, l}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$k$ -正则$C_{2}$ -唯一生成族的充要条件是如下等式对任意$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j})$$\Re \lambda>\omega $成立

$ \begin{eqnarray} \nonumber && \widetilde{W_{i, l}}(\lambda)x +\sum \limits_{j=1}^{n-1}\lambda^{\alpha_{j}-\alpha_{n}}\widetilde{W_{i, l}}(\lambda)A_{j}x\\ \nonumber &=& \chi_{{\mathcal D}_{i}^{l}}(n)\tilde{k}(\lambda)\lambda^{m_{n}-\alpha_{n}-i}C_{2}x +\sum \limits_{j=1}^{n-1}\chi_{{\mathcal D}_{i}^{l}}(j)\tilde{k}(\lambda)\lambda^{m_{j}-\alpha_{j}-i}C_{2}A_{j}x \\ \label{cvba} & & +\lambda^{\alpha-\alpha_{n}}\Bigl[\widetilde{W_{i, l}}(\lambda)Ax -\chi_{{\mathcal D}_{i}^{l}}(0)\tilde{k}(\lambda)\lambda^{m-\alpha-i}C_{2}Ax\Bigr]. \end{eqnarray} $ (2.9)

而当$S\neq \emptyset $时, 设$((W_{i, l}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$满足上面的的假设, 且设$(W_{0, 1}(t))_{t\geq 0}$是强连续的且集合$\{{\rm e}^{-\omega t}W_{0, 1}(t) : t\geq 0\}$$L(E)$中有界, 则$((W_{i, l}(t))_{t\geq 0}, $ $(W_{0, 1}(t))_{t\geq 0})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的$k$ -正则$C_{2}$ -唯一生成族的充要条件是(2.9)式和如下等式对任意$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j})$$\Re \lambda>\omega $成立

$ \begin{eqnarray*} && \widetilde{W_{0, 1}}(\lambda)x +\sum \limits_{j=1}^{n-1}\lambda^{\alpha_{j}-\alpha_{n}}\widetilde{W_{0, 1}}(\lambda)A_{j}x\\ \nonumber &= & \chi_{S}(n)\tilde{k}(\lambda)C_{2}x +\sum \limits_{j=1}^{n-1}\chi_{S}(j)\tilde{k}(\lambda)C_{2}A_{j}x +\lambda^{\alpha-\alpha_{n}}\Bigl[\widetilde{W_{0, 1}}(\lambda)Ax -\chi_{S}(0)\tilde{k}(\lambda)C_{2}Ax\Bigr]. \end{eqnarray*} $

(SC2) 设$(W(t))_{t\geq 0}$是强连续的且集合$\{{\rm e}^{-\omega t}W(t) : t\geq 0\}$$L(E)$中有界.则$(W(t))_{t\geq 0}$是方程(0.1)的$k$ -正则$C_{2}$ -唯一生成族的充要条件是如下等式对任意$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j})$$\Re \lambda>\omega $成立

$ \tilde{W}(\lambda)x-\tilde{k}(\lambda)C_{2}x+\sum \limits_{j=1}^{n-1}\lambda^{\alpha_{j}-\alpha_{n}}\tilde{W}(\lambda)C_{2}A_{j}x=\lambda^{\alpha-\alpha_{n}}\tilde{W} (\lambda)C_{2}Ax. $

注意到文献[22, 定理2.10-2.12]中的结论可以转述为Riemann-Liouville分数阶导数的抽象多项方程的情形.故而文献[22, 例5.1(ⅰ)]可以在本文的新的研究中重新加以考察.

例 2.1 基于CTRW方法, Langlands, Henry和Wearne在文献[30]中运用Riemann-Liouville分数阶导数来逼近spiny dendrites的passive electrotonic性质.他们提出的方程可写为如下抽象的形式[30, (1.18)式]

$ D_{t}^{1}u(t, x)+\mu^{2}D_{t}^{1-\kappa}u(t, x)=D_{t}^{1-\gamma}\Delta u(t, x), $ (2.10)

其中$0 < \gamma < \kappa < 1, $ $\mu \in {\mathbb R} \setminus \{0\}.$方程(2.10)是属于(SC2)类型的, 类似于文献[22, (5.4)式]中的分析讨论(亦可见定理2.2)易证在一大类Banach空间(由作用在有限或者无限区域上的函数组成)上, (2.10)式存在指数等度连续, 角度为$\theta=\pi/2$的解析豫解生成族(注意此处并不能确定解析辐角$\theta$是否可以取到值$\min(\pi, \pi/\gamma-\pi/2))$.

以下通过更多的例子来解释说明文献[22, 例5.1(ⅰ)]中的思想.

例 2.2 设$n\in {\mathbb N}\setminus \{1\}, $ $c_{1}, \cdots, c_{n-1} \in {\mathbb C}, $ $0 \leq \alpha_{1} < \cdot \cdot \cdot < \alpha_{n}\leq 2, $ $0 < \beta < 1, $ $1 < \gamma \leq 2, $ $k_{\beta}, \ k_{\gamma}>0, $ $L>0.$以下的标量型多项时—空Caputo-Riesz分数阶对流扩散方程(简写为MT-TSCR-FADE)已经由Jiang, Liu, Turner和Burrage在文献[17]中加以研究

$ {\mathbf D}_{t}^{\alpha_{n}}u(t, x)+ c_{n-1}{\mathbf D}_{t}^{\alpha_{n-1}}u(t, x)+\cdots + c_{1}{\mathbf D}_{t}^{\alpha_{1}}u(t, x)=k_{\beta}\frac{\partial^{\beta}u(t, x)}{\partial |x|^{\beta}}+k_{\gamma}\frac{\partial^{\gamma}u(t, x)}{\partial |x|^{\gamma}}, \\ \Bigr(\frac{\partial^{k}}{\partial t^{k}}u(t, x)\Bigr)_{t=0}=u_{k}(x), \quad k=0, \cdots, \lceil \alpha_{n} \rceil -1, \ 0\leq x\leq L, $ (2.11)

其中$\frac{\partial^{\beta}u(t, x)}{\partial |x|^{\beta}}$表示$\beta$阶Riesz分数阶算子.在文献[21, 例2.9.53]中, 我们已经考察了问题(2.11)的两种不同的发展模型.在第一种模型中, 方程(2.11)被改写成如下多项分数阶方程的形式

$ \begin{eqnarray*} & {\mathbf D}_{t}^{\alpha_{n}}u(t)+ c_{n-1}{\mathbf D}_{t}^{\alpha_{n-1}}u(t)+\cdots + c_{1}{\mathbf D}_{t}^{\alpha_{1}}u(t)=-k_{\beta}A_{\beta/2}-k_{\gamma}A_{\gamma/2}, \\ & u^{(k)}(0)=u_{k}, \quad k=0, \cdots, \lceil \alpha_{n} \rceil -1, \end{eqnarray*} $ (2.12)

其中算子$A$属于${\mathcal M}_{C, m}$类, $m\in {\mathbb R}, $且作用在$[0, L]$的适当的函数空间上(关于几乎$C$ -扇形算子的更多的细节可参看文献[21, 2.9节]).如同文献[21], 当$C\neq I$或者$m\neq -1$时, 对方程(2.12)的分析是非常复杂的; 为了简洁, 设$A$是辐角为$\omega \in [0, \pi)$的扇形算子, $D(A)$$R(A)$$E$中稠密.假设“抛物性”条件$2\pi >(\beta +\gamma)\omega$成立, $\alpha_{n}^{-1}(\pi-(\gamma \omega/2))-(\pi/2)>0, $$1 < \alpha_{n} < 2.$则借助于Da Prato-Grisvard定理[15, 定理9.3.1, 推论9.3.2]不难证明算子$k_{\beta}A_{\beta/2}+ k_{\gamma}A_{\gamma/2}$是辐角为$\gamma \omega /2$的扇形算子.由此易证对任一$x_{1, 1}\in E, $相应的积分方程(2.4)存在唯一的mild解$t\mapsto u(t), $ $t>0, $且此解可解析延拓到扇形区域$\Sigma_{\delta}, $其中$\delta = \min(\pi/2, \alpha_{n}^{-1}(\pi-(\gamma \omega/2))-(\pi/2)).$在第二个模型中, 我们研究了向后方程(MT-TSCR-FADE)

$ \begin{array}{ll} & {\mathbf D}_{t}^{\alpha_{n}}u(t)+ c_{n-1}{\mathbf D}_{t}^{\alpha_{n-1}}u(t)+\cdots + c_{1}{\mathbf D}_{t}^{\alpha_{1}}u(t)=-k_{\beta}A_{\beta}-k_{\gamma}A_{\gamma}, \\ & u(0)=u_{0} \end{array} $ (2.13)

$C$ -适定性, 其中$0 < \beta < 1, $ $1 < \gamma < 2, $ $k_{\beta}, \ k_{\gamma}>0, $ $0 \leq \alpha_{1} < \cdots < \alpha_{n} < 1, $ $\alpha_{n}^{-1}(\pi-(\gamma \pi/2))>\pi/2, $ $ E:=\{f\in C^{\infty}[0, 1] \ ; \ \|f\| :=\sup\limits_{p\in {{\mathbb N}_{0}}}\frac{\|f^{(p)}\|_{\infty}}{p!^{\zeta}} < \infty\} $ ($\zeta>1$), 且$A:=-{\rm d}/{\rm d}s, $ $D(A):=\{f\in E : f'\in E, \;f(0)=0\}.$不过在这种形式中, (2.13)式是修正的抽象时间松弛方程, 其相应的Riemann-Liouville分数阶导数的方程几乎没有与本文相关的有意义的内容可供研究(子类(SC3)).

下面的定理刻画了方程(0.1)的强解的存在性(相应的Caputo分数阶导数的方程的结论, 参见文献[24, 定理3.1]或[21, 定理2.10.42]).

定理 2.3$A, $ $A_{1}, \cdots, $ $A_{n-1}$$E$上的闭线性算子, $\omega>0, $ $0 < \tau < \infty, $ $C\in L(E)$是单射, $f(t)\equiv 0, $$\lambda>\omega $算子$P_{\lambda}$是单射, $\lambda>\omega, $$D(P_{\lambda}^{-1}C)=E, $ $\lambda>\omega.$

(SC1) 设$\lambda>\omega, $$j\in {\mathbb N}_{n-1}^{0} \cap {\mathcal D}_{i}^{l}$$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}, $ $n\in {\mathcal D}_{i}^{l}, $ $Cx_{i, l}\in D(P_{\lambda}^{-1}A_{j}), $又设$j\in {\mathbb N}_{n-1}^{0} \setminus {\mathcal D}_{i}^{l}$$\alpha_{j}-\alpha_{n}+m_{n}-1-i < 0, $

$ \begin{eqnarray} \nonumber &&\lambda^{\alpha_{n}}P_{\lambda}^{-1} \bigg[\lambda^{m_{n}-i-1-\alpha}Cx_{i, l} +\sum \limits_{j=1}^{n-1}A_{j}\Bigl( \chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1-\alpha}Cx_{i, l}\Bigr) \\\label{prcq} & &-A\Bigl( \chi_{{\mathcal D}_{i}^{l}}(0)\lambda^{m-i-1-\alpha}Cx_{i, l}\Bigr) \bigg]-\lambda^{m_{n}-1-i}Cx_{i, l}\in LT_{E} \end{eqnarray} $ (2.14)

$ \begin{eqnarray} \nonumber && A_{j} \bigg \{ \lambda^{\alpha_{j}}P_{\lambda}^{-1}\bigg[\lambda^{m_{n}-i-1-\alpha}Cx_{i, l} +\sum \limits_{j=1}^{n-1}A_{j}\Bigl( \chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1-\alpha}Cx_{i, l}\Bigr) \\\label{prcq1} &&-A\Bigl( \chi_{{\mathcal D}_{i}^{l}}(0)\lambda^{m-i-1-\alpha}Cx_{i, l}\Bigr) \bigg]-\chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1}Cx_{i, l}\bigg\}\in LT_{E} \end{eqnarray} $ (2.15)

成立.初值$x_{i, l}$$Cx_{i, l}$代替, 且其它初值为零时, 则方程(0.1)在$(0, \tau)$上存在强解.若$S\neq \emptyset, $且上面的关于初值$x_{0, 1}=u(0)$的条件成立, 其中集合${\mathcal D}_{i}^{l}$$S$ ($i=0, $ $l=1$)代替, 则方程(0.1)在$(0, \tau)$上存在强解, 其中初值$x_{0, 1}$$Cx_{0, 1}$代替, 其它初值为零.

(SC2) 设$\lambda^{1-\alpha}P_{\lambda}^{-1}Cx_{0, 1}-Cx_{0, 1}\in LT_{E}, $$\lambda^{\alpha_{j}-\alpha}A_{j}P_{\lambda}^{-1}Cx_{0, 1}\in LT_{E}.$则方程(0.1)在$(0, \tau)$上存在强解, 其中初值$x_{0, 1}$$Cx_{0, 1}$代替, 其它初值为零.

 我们仅对定理在(SC1)的情形给出证明, 其中$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}.$$u_{i, l} \in L_{\rm loc}^{1}([0, \infty) : E)$$F_{i, l, n} \in L_{\rm loc}^{1}([0, \infty) : E)$满足

$ \begin{eqnarray} \nonumber \int_{0}^{\infty} {\rm e}^{-\lambda t}u_{i, l}(t){\rm d}t &=& P_{\lambda}^{-1}\bigg[\lambda^{m_{n}-i-1-\alpha}Cx_{i, l} +\sum \limits_{j=1}^{n-1}A_{j}\Bigl( \chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1-\alpha}Cx_{i, l}\Bigr) \\\nonumber & &-A\Bigl( \chi_{{\mathcal D}_{i}^{l}}(0)\lambda^{m-i-1-\alpha}Cx_{i, l}\Bigr) \bigg] \end{eqnarray} $

$ \begin{eqnarray} \nonumber \int_{0}^{\infty} {\rm e}^{-\lambda t}F_{i, l, n}(t){\rm d}t &=& \lambda^{\alpha_{n}}P_{\lambda}^{-1}\bigg[\lambda^{m_{n}-i-1-\alpha}Cx_{i, l}-A\Bigl( \chi_{{\mathcal D}_{i}^{l}}(0)\lambda^{m-i-1-\alpha}Cx_{i, l}\Bigr) \\\nonumber & & +\sum \limits_{j=1}^{n-1}A_{j}\Bigl( \chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1-\alpha}Cx_{i, l}\Bigr) \bigg]-\lambda^{m_{n}-1-i}Cx_{i, l}, \end{eqnarray} $

其中$\lambda>\omega$足够大, 见(2.14)式.利用Laplace变换, 容易验证

$ \bigl( g_{m_{n}}\ast F_{i, l, n}\bigr)(t)=\bigl( g_{m_{n}-\alpha_{n}}\ast u_{i, l}\bigr)(t)-g_{i+1}(t)Cu_{i, l}, \quad t> 0. $ (2.16)

这蕴含着$D_{t}^{\alpha_{n}}u_{i, l}(t)$$t>0$是良定义的(更精确地说, 在$(0, \infty)$的任意有限子区间上是良定义的)且$F_{i, l, n}(t)=D_{t}^{\alpha_{n}}u_{i, l}(t), $ $t>0.$因为对$j\in {\mathbb N}_{n-1}^{0} \setminus {\mathcal D}_{i}^{l}, $$n\in {\mathcal D}_{i}^{l}$$\alpha_{j}-\alpha_{n}+m_{n}-1-i < 0, $由(2.14)式可得对所有$j\in {\mathbb N}_{n-1}^{0}, $都有$\lambda^{\alpha_{j}}\widetilde{u_{i, l}}(\lambda)-\chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1}Cx_{i, l}\in LT_{E}, $且对所有$j\in {\mathbb N}_{n-1}^{0}$$t>0, $满足如下条件的$D_{t}^{\alpha_{j}}u_{i, l}(t)$是良定义的

$ \begin{eqnarray} \nonumber \int_{0}^{\infty}{\rm e}^{-\lambda t}D_{t}^{\alpha_{j}}u_{i, l}(t){\rm d}t= \lambda^{\alpha_{j}}\widetilde{u_{i, l}}(\lambda)-\chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1}Cx_{i, l}\in LT_{E}. \end{eqnarray} $

利用(2.15)式和文献[3, 性质1.7.6], 易得$A_{j}D_{t}^{\alpha_{j}}u_{i, l}(t)$$t>0$是良定义的, 且对所有$j\in {\mathbb N}_{n-1}^{0}$

$ \int_{0}^{\infty} {\rm e}^{-\lambda t}A_{j}D_{t}^{\alpha_{j}}u_{i, l}(t){\rm d}t=A_{j}\bigg[ \lambda^{\alpha_{j}}\widetilde{u_{i, l}}(\lambda)-\chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1}Cx_{i, l}\bigg]\in LT_{E}. $

最后, 简单的计算表明

$ \int_{0}^{\infty}{\rm e}^{-\lambda t}\Bigl[ D_{t}^{\alpha_{n}}u_{i, l}(t)+ A_{n-1} D_{t}^{\alpha_{n-1}}u_{i, l}(t)+\cdots + A_{1} D_{t}^{\alpha_{1}}u_{i, l}(t)-A D_{t}^{\alpha}u_{i, l}(t)\Bigr]\, {\rm d}t=0, $

由Laplace变换的唯一性定理, 这蕴含着$u_{i, l}(\cdot)$是问题(0.1)的强解, 其中初值$x_{i, l}$$Cx_{i, l}$代替, 其余初值为零.

注 2.1 考虑子类(SC1), 设$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}, $ $n\in {\mathcal D}_{i}^{l}, $以及$\alpha_{j}-\alpha_{n}+m_{n}-1-i < 0, $其中$j\in {\mathbb N}_{n-1}^{0} \setminus {\mathcal D}_{i}^{l}.$对以下这一项

$ P_{\lambda}^{-1}\bigg[\lambda^{m_{n}-i-1-\alpha}Cx_{i, l} +\sum \limits_{j=1}^{n-1}A_{j}\Bigl( \chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1-\alpha}Cx_{i, l}\Bigr) -A\Bigl( \chi_{{\mathcal D}_{i}^{l}}(0)\lambda^{m-i-1-\alpha}Cx_{i, l}\Bigr) \bigg], $ (2.17)

即解$t\mapsto u_{i, l}(t), $ $t>0 $的Laplace变换, 用下面的项

$ \begin{eqnarray} \nonumber &&P_{\lambda}^{-1} \bigg[\lambda^{m_{n}-i-1-\alpha_{n}}P_{\lambda}Cx_{i, l}-\sum \limits_{j\in {{\mathbb N}_{n-1}} \setminus {\mathcal D}_{i}^{l} }A_{j}\Bigl(\lambda^{m_{n}-i-1-\alpha_{n}+\alpha_{j}-\alpha}Cx_{i, l}\Bigr) \\\label{prcqqww} &&+A\Bigl( \chi_{{{\mathbb N}_{n-1}} \setminus {\mathcal D}_{i}^{l}}(0)\lambda^{m_{n}-i-1-\alpha_{n}}Cx_{i, l}\Bigr) \bigg] \end{eqnarray} $ (i)

代替, 且以条件$Cx_{i, l}\in D(P_{\lambda}^{-1}A_{j}), $ $\lambda>\omega, $ $j\in {{\mathbb N}_{n-1}^{0}}\setminus {\mathcal D}_{i}^{l}$代替条件$Cx_{i, l}\in D(P_{\lambda}^{-1}A_{j}), $ $\lambda>\omega, $ $j\in {{\mathbb N}_{n-1}^{0}}\cap {\mathcal D}_{i}^{l}, $则直接的计算表明定理(2.3)的结论仍然成立.若$S\neq \emptyset, $ $i=-0, $ $l=1, $则以$S$代替集合${\mathcal D}_{i}^{l}.$对子类(SC2)的相应的分析留给有兴趣的读者.

2.1 $(a, k)$ -正则$C$-豫解族在对方程(0.1)的研究中的应用

在本节中, 总是假设$A=0, $ $n\geq 3, $ $0=\alpha_{1} < \cdots < \alpha_{n};$注意这些假设并不需要如此严格, 因为如果需要的话, 总是可以在对各项进行显而易见的重排之后, 得到方程(0.1)的左端的$A{\mathbf D}_{t}^{\alpha}u(t)$项, 以及在(0.1)的左端加上$0u(t)$这项.为了简便起见, 我们仅考虑子类(SC1).

定理 2.4 设$n\in {\mathbb N}\setminus \{1, 2\}, $ $\sigma \in [1, 2], $ $r\geq 0, $ $\alpha_{n}-\alpha_{n-1}=\sigma, $ $\alpha_{n-1}-\alpha_{n-2}\geq \sigma, $ $\omega \geq 0, $ $D(A_{n-1})\subseteq \bigcap\limits_{i=0}^{n-2}D(A_{i}), $$(\omega^{\sigma}, \infty) \subseteq \rho(-A_{n-1}).$$\breve{A_{i}}(\lambda):=\lambda^{\alpha_{i}-\alpha_{n-1}} A_{i}(\lambda^{\sigma} + A_{n-1})^{-1}, $ $\lambda >\omega, $ $i\in {\mathbb N}_{n-2}$且假设以下条件成立:

(i) $A_{i}A_{j}x=A_{j}A_{i}x, \ 1\leq i, \ j\leq n-1, \ x\in D(A_{n-1}^{2})$$CA_{j}\subseteq A_{j}C, $ $j\in {\mathbb N}_{n-2}.$

(ⅱ) 存在$\omega_{0} \geq \omega$$c\in (0, 1/(n-2))$满足

$ \bigl \| \breve{A_{i}}(\lambda)x\bigr\|\leq c\|x\|, \ \lambda>\omega_{0}, \ x\in E, \ i\in {\mathbb N}_{n-2}. $ (2.19)

设对所有$j\in{{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}, $$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}, $ $n\in {\mathcal D}_{i}^{l}, $ $x_{i, l}\in D(A_{j})$; 对所有$j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}, $$\alpha_{j}-\alpha_{n}+m_{n}-i < 1$.

(a) 当$j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}$时, 算子$ -A_{n-1}$$E$$(g_{\sigma}, g_{\sigma r+1})$ -正则豫解族$(S_{\sigma, r}(t))_{t\geq 0}$的积分生成元, $\{{\rm e}^{-\omega t}S_{\sigma, r}(t) : t\geq 0\}$$L(E)$中有界, 且$A_{j}x_{i, l}\in D\bigl(A_{n-1}^{\max([\frac{1}{\sigma}(\sigma r+m_{n}-i+\alpha_{j}-\alpha_{n} )], 0)}\bigr), $或者

(b) 当$j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}$时, 算子$ -A_{n-1}$$E$$(g_{\sigma}, C)$ -正则豫解族$(T_{\sigma}(t))_{t\geq 0}$的积分生成元, $\{{\rm e}^{-\omega t}T_{\sigma}(t) : t\geq 0\}$$L(E)$中有界, 且$C^{-1}A_{j}x_{i, l}\in D\bigl(A_{n-1}^{\max( [\frac{1}{\sigma}(\sigma r+m_{n}-i+\alpha_{j}-\alpha_{n} )], 0)}\bigr), $则抽象Cauchy问题(0.1)有唯一的强解, 其中$f(t)\equiv 0, $ $x_{i, l}=(g_{a_{i, l}}\ast u)^{(i)}(0), $其余的初值为零.若$S\neq \emptyset, $ $i=0, $ $l=1, $且以集合$S$代替集合${\mathcal D}_{i}^{l}, $则上述结论仍然成立.

 我们仅简述(a)成立时的证明的主要细节, 其中$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}.$$\mu_{0} < -\omega_{0}^{\sigma}.$由文献[25, 定理2.1]的证明, 级数

$ B_{\lambda}x\equiv \sum \limits_{k=0}^{\infty}\bigl(\lambda^{\sigma} +A_{n-1}\bigr)^{-1}\bigg[-\sum \limits_{i=1}^{n-2} \breve{A_{i}}(\lambda)\bigg]^{k}x $

对任意$\lambda>\omega_{0}$是收敛的, 且$ P_{\lambda}B_{\lambda}x=\lambda^{\alpha_{n-1}}x, $ $x\in E, $ $\lambda>\omega_{0}, $ $P_{\lambda}$$\lambda>\omega_{0}$时是单射, $B_{\lambda}x=\lambda^{\alpha_{n-1}}P_{\lambda}^{-1}x, $ $x\in E, $ $\lambda>\omega_{0}.$由定理2.3和注2.1, 要证明强解的存在性, 只需证明

$ \lambda^{\alpha_{n}}\widetilde{u_{i, l}}(\lambda)-\lambda^{m_{n}-i-1}x_{i, l}\in LT_{E} $

$ A_{j}\bigg[\lambda^{\alpha_{j}}\widetilde{u_{i, l}}(\lambda)-\chi_{{\mathcal D}_{i}^{l}}(j)\lambda^{m_{j}-i-1}x_{i, l} \bigg]\in LT_{E} $

即可, 也就是证明

$ P_{\lambda}^{-1}\sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}} \lambda^{m_{n}-i-1+\alpha_{j}}A_{j}x_{i, l}\in LT_{E} $ (2.20)

$ \lambda^{\alpha_{j}}A_{j}P_{\lambda}^{-1}\sum \limits_{v\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{v}-\alpha_{n}}A_{v}x_{i, l}\in LT_{E}, \quad j\in {\mathbb N}_{n-1} $ (2.21)

即可(见(2.17)-(2.18)式).这可由如下事实得到$\lambda^{\alpha_{n}-1}P_{\lambda}^{-1}x\in LT_{E}$ ($x\in E$), $\lambda^{-r_{1}}\bigl(\mu_{0}- A_{n-1} \bigr)^{-r_{0}}B_{\lambda}x\in LT_{E} $ ($r_{0}=-1, 0, 1, \cdots ;$ $r_{1}\in {\mathbb R}, $ $r_{1}+r_{0}\sigma \geq \sigma r+1-\sigma, $ $x\in E$), 分解$A_{j}x=A_{j}(\mu_{0}-A_{n-1})^{-1}(\mu_{0}-A_{n-1})x, $ $x\in D(A_{n-1}), $ $j\in {\mathbb N}_{n-1}, $其余的讨论可参阅文献[25, 定理2.1]中的证明.

当方程含有超过两项控制项时, 定理2.4有一些缺陷, 在应用于Riemann-Liouville分数导数的阶严格大于2的实际问题时会遇到障碍; 也就是说, 必然不符合下面的定理2.5的情形.值得注意的是数学物理中存在一些重要的微分方程关于时间的阶严格大于2, 如在高强超声和弹性结构的线性振动的分析中出现的三阶Moore-Gibson-Thompson方程[2, 10, 18]. $(a, k)$ -正则$C$-豫解族[2, 21]的理论提供了研究这种方程的分数阶类比的有效的算子理论方法; 不过注意定理2.4和定理2.5在此无法应用.

为引入定理2.5, 设$s_{i, l, j, r}:= \max( [\frac{1}{\sigma}(\sigma r+m_{n}-i+\alpha_{j}-\alpha_{n} )], 0), $其中$r\geq 0, $ $1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}, $ $j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}$ (当$S\neq \emptyset$时, 为$r\geq 0, $ $i=0, $ $l=1, $ $j\in {\mathbb N}_{n-1} \setminus S$).

定理 2.5 设$n\in {\mathbb N}\setminus \{1, 2\}, $ $\sigma \in (0, 2], $ $M\geq 1, $ $r\geq 0, $ $\alpha_{n}>1, $ $\alpha_{n}-\alpha_{n-1}=\sigma, $ $D(A_{n-1})\subseteq \bigcap\limits_{i=0}^{n-2}D(A_{i}), $ $(\omega^{\sigma}, \infty) \subseteq \rho(-A_{n-1}).$$\check{A_{i}}(\lambda)x:=\lambda^{\alpha_{i}-\alpha_{n-1}} (\lambda^{\sigma} + A_{n-1})^{-1}A_{i}x, $ $b_{i}:=\max([\sigma^{-1}(\alpha_{i}-\alpha_{n-1}+\sigma r+1)], 0), $ $v_{i}:=\max([\sigma^{-1}(\alpha_{i}-\alpha_{n-1}+1)], 0), $其中$x\in D(A_{n-1}), $ $\lambda >\omega, $ $i\in {\mathbb N}_{n-2}.$$\mu_{0} < -\omega^{\sigma}.$$n\in {\mathcal D}_{i}^{l}, $ $x_{i, l}\in D(A_{j}), $ $\alpha_{j}-\alpha_{n}+m_{n}-i < 1, $ $j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}, $且下述成立:

(a) 算子$ -A_{n-1}$是指数有界$(g_{\sigma}, g_{\sigma r+1})$ -正则豫解族$(S_{\sigma, r}(t))_{t\geq 0}$的积分生成元, 且对任意$x\in D\bigl(A_{n-1}\bigr), $ $i\in {\mathbb N}_{n-2}, $

$ \Bigl \|\bigl( \mu_{0}-A_{n-1}\bigr)^{b_{i}}A_{i}x\Bigr \| \leq M \Bigl[\|x\|+\bigl\|A_{n-1}x\bigr \|\Bigr], $ (2.22)

$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}$, $j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}, $$A_{j}x_{i, l}\in D(A_{n-1}^{s_{i, l, j, r}}), $

(b) 算子$ -A_{n-1}$是指数有界$(g_{\sigma}, C)$ -正则豫解族$(T_{\sigma}(t))_{t\geq 0}$的积分生成元, 且对任意$x\in D\bigl(A_{n-1}\bigr), $ $i\in {\mathbb N}_{n-2}, $

$ \Bigl\|\bigl( \mu_{0}-A_{n-1}\bigr)^{v_{i}}C^{-1}A_{i}x\Bigr\|\leq M \Bigl[\|x\|+\bigl\|A_{n-1}x\bigr \|\Bigr], $

$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}$, $j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}$时, $C^{-1}A_{l}u_{k}\in D(A_{n-1}^{s_{i, l, j, 0}}), $

(c) 算子$ -A_{n-1}$是指数有界$(g_{\sigma}, C)$ -正则豫解族$(T_{\sigma}(t))_{t\geq 0}$的积分生成元, (a)成立, 且对$1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}$$j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}, $$C^{-1}A_{l}u_{k}\in D(A_{n-1}^{s_{i, l, j, 0}}), $则抽象Cauchy问题(0.1)在任意有限区间$(0, \tau)$上存在唯一的强解, 其中$x_{i, l}=(g_{a_{i, l}}\ast u)^{(i)}(0), $其余的初值为零.若$S\neq \emptyset, $ $i=0, $ $l=1, $以集合$S$代替${\mathcal D}_{i}^{l}, $则上述结论仍然成立.

 我们仅就(a)成立的情形证明强解的存在性.由文献[25, 定理2.4]的证明, 对任意$x\in D(A_{n-1})$$\lambda >\omega', $级数

$ {\mathcal B}_{\lambda}x:=\sum \limits_{k=0}^{\infty} \bigg[-\sum \limits_{i=1}^{n-2}\check{A_{i}}(\lambda)\bigg]^{k}x $

按照$[D(A_{n-1})]$的拓扑是收敛的.由等式$\alpha_{n}-\alpha_{n-1}=\sigma, $易知算子$P_{\lambda}$$\lambda>\omega'$是单射, 且还有

$ \begin{eqnarray}\label{plf} {\mathcal B}_{\lambda}\bigl( \lambda^{\sigma} +A_{n-1}\bigr)^{-1}x=\lambda^{\alpha_{n-1}}P_{\lambda}^{-1}x, \quad x\in X, \ \lambda>\omega'. \end{eqnarray} $ (2.23)

进一步地, 存在强连续指数有界算子族$(H(t))_{t\geq 0}\subseteq L([D(A_{n-1})])$使得

$ {\mathcal B}_{\lambda}x-x=\int^{\infty}_{0} {\rm e}^{-\lambda t}H(t)x{\rm d}t, \quad x\in D\bigl(A_{n-1} \bigr), \ \lambda>\omega'. $ (2.24)

由于对任意$j\in {\mathbb N}_{n-1} \setminus {\mathcal D}_{i}^{l}, $$A_{j}x_{i, l}\in D(A_{n-1}^{s_{i, l, j, r}}), $ $\alpha_{j}-\alpha_{n}+m_{n}-i < 1, $应用推广的预解方程(参见文献[21, 式(6)])易得存在连续函数$t\mapsto G(t)\in [D(A_{n-1})], $ $t\geq 0$使得$\|G(t)\|+\|A_{n-1}G(t)\|\leq M'{\rm e}^{\omega ' t}, $ $t\geq 0, $以及

$ \sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l}=\int ^{\infty}_{0}{\rm e}^{-\lambda t}G(t){\rm d}t, \quad \lambda>\omega'. $ (2.25)

定义$v(t):=G(t)+(H \ast G)(t), $ $t\geq 0.$由(2.23)-(2.25)式, 可得映射$t\mapsto v(t)\in [D(A_{n-1})], $ $t\geq 0$是连续且指数有界的, 且

$ \tilde{v}(\lambda)=-\lambda^{\alpha_{n-1}}P_{\lambda}^{-1} \sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l}, \quad \lambda>\omega'. $ (2.26)

从而

$ \lambda^{\alpha_{n-1}}\bigl(\mu_{0}-A_{n-1})P_{\lambda}^{-1}\sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l} \in LT_{E} $ (2.27)

且(2.21)式成立.由定理2.3和注2.1 (亦可见定理2.4的证明), 现只需证明(2.20)式成立即可.这可由(2.27)式, 等式

$ \begin{eqnarray*} && \lambda^{\alpha_{n-1}} A_{n-1}P_{\lambda}^{-1}\sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l} \\ &=&\sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l} -\lambda^{\alpha_{n}}P_{\lambda}^{-1}\sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}}\lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l} \\ && -\sum \limits^{n-2}_{j=1}\lambda^{\alpha_{j}}A_{j}P_{\lambda}^{-1} \sum \limits_{j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}} \lambda^{m_{n}-i-1+\alpha_{j}-\alpha_{n}}A_{j}x_{i, l}, \quad \lambda>\omega' \ \mbox{充分大, } \end{eqnarray*} $

以及对所有$j\in {{\mathbb N}_{n-1}}\setminus {\mathcal D}_{i}^{l}$都有$\alpha_{j}-\alpha_{n}+m_{n}-i < 1$这些条件容易的得出.

有兴趣的的读者可以考虑把文献[25, 注2.2, 2.5]中的结论转移到Riemann-Liouville分数阶导数的抽象多项方程的可能性; 文献[26]中考虑的抽象微分算子可供作为定理2.4和定理2.5的重要应用.

在文献[25, 第4章]中, 我们已经研究了非齐次多项分数阶微分方程

$ {\mathbf D}_{t}^{\alpha_{n}}u(t)+ \sum \limits_{i=1}^{n-1}A_{i}{\mathbf D}_{t}^{\alpha_{i}}u(t)= f(t), \quad t > 0, \\ u^{(k)}(0)=u_k, \quad k=0, \cdots, m_{n} -1, $ (2.28)

其中$f\in C([0, \infty) : E)$, 且文献[25, 定理2.1]或[25, 定理2.4]的假设成立.其方法的基础是寻找方程(2.28)在$u_{k}=0$ (对所有$k\in {\mathbb N}_{m_{n}-1}^{0}$)时的解.若$t\mapsto u(t), $ $t>0$是此方程的强解, 则显然等式$D_{t}^{\alpha_{i}}u(t)={\mathbf D}_{t}^{\alpha_{i}}u(t), $ $t>0$ ($i\in {\mathbb N}_{n}^{0}$)成立, 故当定理2.4或定理2.5的假设成立时, 文献[25, 第4章] (亦可见文献[21, 注2.10.46(ⅵ), 2.10.50])的结果可以应用于对Riemann-Liouville分数阶导数的非齐次抽象多项方程的研究.例如, 设定理2.4(a)的假设成立, $1\leq i\leq m_{n}-1$, $1\leq l\leq s_{i}.$设映射$t\mapsto (\mu_{0}-A_{n-1})^{[\sigma^{-1}(\sigma r+1)] }f(t), $ $t\geq 0$连续且存在有限常数$M\geq 1$使得

$ \Bigl\|\bigl(\mu_{0}-A_{n-1}\bigr)^{ [\sigma^{-1}(\sigma r+1) ] }f(t)\Bigr\|\leq M{\rm e}^{\omega t}, \quad t\geq 0. $

则存在连续函数$t\mapsto u_{f}(t), $ $t\geq 0$使得$u(t):=u_{f}(t)+u_{i, l}(t), $ $t\geq 0$是方程(0.1)在有限区间$(0, \tau)$上的唯一强解, 其中$x_{i, l}=(g_{a_{i, l}}\ast u)^{(i)}(0)$且其余初值为零.

3 方程(0.1)的$k$ -正则$(C_{1}, C_{2})$ -存在和唯一族

本章中, 依赖方程(0.1)的$k$ -正则$(C_{1}, C_{2})$ -存在和唯一族的基本的结构性质, 我们继续对Riemann-Liouville分数阶导数的抽象多项问题进行研究.设$X$$Y$为复Banach空间; 元素$y\in Y$的范数记为$\|y\|_{Y}, $ $L(Y, X)$表示从$Y$$X$的所有有界线性算子组成的空间.依照肖体俊和梁进在文献[48]中的方式, 我们最近在文献[22]中引入了如下的定义.

定义 3.1[22, 定义3.1]$0 < \tau \leq \infty, $ $k\in C([0, \tau)), $ $C_{1}\in L(Y, X), $$C_{2}\in L(X)$是单射.

(ⅰ) 强连续算子族$(E(t))_{t\in [0, \tau)}\subseteq L(Y, X)$称为(若$\tau < \infty$, 则为局部的) $k$-正则$C_{1}$-存在族, 当且仅当对任意$y\in Y, $下述成立:对任意$i\in {{\mathbb N}_{0}}, $ $i < m_{n}-1, $$E(\cdot)y\in C^{m_{n}-1}([0, \tau) : X), $ $E^{(i)}(0)y=0$, 对$0\leq j\leq n-1, $$A_{j}(g_{\alpha_{n}-\alpha_{j}}\ast E^{(m_{n}-1)})(\cdot)y\in C([0, \tau) : X), $且对任意$t\in [0, \tau)$

$ E^{(m_{n}-1)}(t)y+\sum \limits^{n-1}_{j=1} A_{j}\bigl(g_{\alpha_{n}-\alpha_{j}}\ast E^{(m_{n}-1)}\bigr)(t)y -A\bigl(g_{\alpha_{n}-\alpha}\ast E^{(m_{n}-1)}\bigr)(t)y=k(t)C_{1}y. $ (3.1)

(ⅱ) 强连续算子族$(U(t))_{t\in [0, \tau)}\subseteq L(X)$称为(若$\tau < \infty$, 则为局部的) $k$-正则$C_{2}$-唯一族, 当且仅当对任意$\tau \in [0, \tau)$$x\in \bigcap\limits_{0\leq j\leq n-1}D(A_{j}), $下述成立

$ U(t)x+\sum \limits^{n-1}_{j=1} \bigl(g_{\alpha_{n}-\alpha_{j}}\ast U(\cdot)A_{j}x\bigr)(t)-\bigl(g_{\alpha_{n}-\alpha}\ast U(\cdot)Ax\bigr)(t)y=\bigl(k\ast g_{m_{n}-1}\bigr)(t)C_{2}x. $

(ⅲ) 强连续算子族$((E(t))_{t\in [0, \tau)}, (U(t))_{t\in [0, \tau)}) \subseteq L(Y, X) \times L(X)$称为(若$\tau < \infty$, 则为局部的) $k$-正则$(C_{1}, C_{2})$-存在和唯一族, 当且仅当$(E(t))_{t\in [0, \tau)}$$k$-正则$C_{1}$-存在族, 且$(U(t))_{t\in [0, \tau)}$$k$-正则$C_{2}$-唯一族.

(ⅳ) 设$Y=X, $ $C=C_{1}=C_{2}.$则强连续算子族$(R(t))_{t\in [0, \tau)}\subseteq L(X)$称为(若$\tau < \infty$, 则为局部的) $k$-正则$C$-豫解族, 当且仅当$(R(t))_{t\in [0, \tau)}$$k$-正则$C$-唯一族, $R(t)A_{j}\subseteq A_{j}R(t), $ $R(t)C=CR(t), $ $CA_{j}\subseteq A_{j}C, $其中$0\leq j\leq n-1, $ $t\in [0, \tau).$

$k(t)=g_{\zeta+1}(t), $ $\zeta \geq 0$的情形, 也称$(E(t))_{t\in [0, \tau)}$$\zeta$-次积分$C_{1}$-存在族; $0$-次积分$C_{1}$ -存在族也称为$C_{1}$ -存在族.还可以以自然的方式定义$C_{1}$ -存在族的解析性; 以上的术语约定同样适用于定义3.1中的其他类型的唯一和豫解族.

对任意$t\in [0, \tau), $ $y\in Y$$l\in {{\mathbb N}_{m_{n}-1}^{0}}, $

$ E^{(l)}(t)y+\sum \limits^{n-1}_{j=1}A_{j}\bigl(g_{\alpha_{n}-\alpha_{j}}\ast E^{(l)}\bigr)(t)y-A\bigl(g_{\alpha_{n}-\alpha}\ast E^{(l)}\bigr)(t)y=\bigl(k\ast g_{m_{n}-1-l}\bigr)(t)C_{1}y. $ (3.2)

在存在$C_{1}$-存在族$(E(t))_{t\in [0, \tau)} \subseteq L(Y, X)$$k$-正则$C_{2}$-唯一族$(U(t))_{t\in [0, \tau)}\subseteq L(X)$的条件下, 我们已经在文献[22, 定理3.4]中研究了Caputo分数阶导数的抽象多项齐次问题的强解的存在性和唯一性.下文中我们将会简要叙述对Riemann-Liouville分数阶导数的抽象多项方程, 非标量型抽象Volterra方程的结论[23, 44]如何帮助我们证明类似于上述的结果; 简便起见, 我们仅考虑子类(SC1).设$(E(t))_{t\in [0, \tau)}$是(局部的) $C_{1}$-存在族, $1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}.$则积分方程(2.4)可改写为如下的形式

$ u(t)=f(t)+\int^{t}_{0}A(t-s)u(s){\rm d}s, \ \quad t\in (0, \tau), $

其中, 形式地, $A(t)=-\sum\limits^{n-1}_{j=1}g_{\alpha_{n}-\alpha_{j}}A_{j}+g_{\alpha_{n}-\alpha}(t)A, $ $t\in (0, \tau)$

$ \begin{eqnarray*} f(t)&=&\chi_{{\mathcal D}_{i}^{l}}(n)g_{\alpha_{n}+i+1-m_{n}}(t)x_{i, l}+ \sum^{n-1}_{j=1}A_{j}\bigl(\chi_{{\mathcal D}_{i}^{l}}(j)g_{\alpha_{n}+i+1-m_{j}}(t)x_{i, l}\bigr) \\ && -A\bigl(\chi_{{\mathcal D}_{i}^{l}}(0)g_{\alpha_{n}+i+1-m}(t)x_{i, l}\bigr)\quad (t\in (0, \tau)). \end{eqnarray*} $

对上述方程积分$m_{n}$次, 利用文献[23, 性质2(ⅱ)]易得当算子$C_{1}$是单射时, 解$u(t)$应具有如下形式

$ u(t)=\frac{\rm d}{{\rm d}t}\int^{t}_{0}E^{(m_{n}-1)}(t-s)C_{1}^{-1}f(s){\rm d}s, \ \quad t\in (0, \tau), $

当以下条件

$x_{i, l}\in R\bigl(C_{1}\bigr), ~\mbox{ 若 }~n\in {\mathcal D}_{i}^{l}; ~ \mbox{ 且 }A_{j}x_{i, l}=C_{1}y_{i, l, j}~\mbox{ 对所有 }~j\in {\mathbb N}_{n-1}^{0} \cap {\mathcal D}_{i}^{l}~\mbox{ 成立 }$

成立时, 前面的分析以及直接的计算表明积分方程(2.4)的mild解存在.在叙述下个定理之前, 我们举一个例子来说明问题(0.1)的强解的存在性可以平凡地证明得到, 并且说明存在一大类多项问题, 其具有与算子$A_{j}$ ($j\in{\mathbb N}_{0}$的选择无关的强解; 关于各种类型的$k$ -正则$(C_{1}, C_{2})$ -存在和唯一族的信息, 可见于文献[22, 第3-4章].

例 3.1 考虑子类(SC1), 其中$1\leq i\leq m_{n}-1, $ $l=1, $ $m_{j}\geq i+1, $ $j\in {\mathbb N}_{0}, $ $m_{j}-\alpha_{j}=a_{i, 1}, $ $j\in {\mathbb N}_{0}.$$0 < \tau < \infty$$f(t)\equiv 0.$$D_{t}^{\alpha_{j}}g_{i+1-a_{i, 1}}(t)=0, $ $j\in {\mathbb N}_{0}$故函数$u(t)=g_{i+1-a_{i, 1}}(t)x_{i, 1}, $ $t\in [0, \tau)$是问题(0.1)的强解, 其中$x_{i, 1}=(g_{m_{j}-\alpha_{j}}\ast u)^{(i)}(0)\in X$且其余的初值为零.

定理 3.1 设$0 < \tau \leq \infty, $ $C_{1}\in L(Y, X), $$C_{2}\in L(X)$是单射.令$P_{\lambda}:=\lambda^{\alpha_{n}-\alpha}+\sum\limits_{j=1}^{n-1}\lambda^{\alpha_{j}- \alpha}A_{j}-A, $ $\lambda \in {\mathbb C} \setminus \{0\}.$

(ⅰ) 设$(E(t))_{t\in [0, \tau)}$是(局部) $C_{1}$-存在族, $1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}.$若当$x_{i, l}\in R\bigl(C_{1}\bigr)\mbox{, 且 }n\in {\mathcal D}_{i}^{l}$时, 对任意$j\in {\mathbb N}_{n-1}^{0}\cap {\mathcal D}_{i}^{l}, $都存在元素$y_{i, l, j}\in Y$使得$A_{j}x_{i, l}=C_{1}y_{i, l, j}, $则问题(2.4)存在mild解.若$S\neq \emptyset, $ $i=0, $ $l=1, $且以$S$代替集合${\mathcal D}_{i}^{l}, $则上述结论仍然成立.

(ⅱ) 设$(U(t))_{t\in [0, \tau)}$$k$-正则$C_{2}$-唯一族, 且设$k(t)$是核.则方程(0.1)的任意两个具有相同初值(见(2.1)和(2.3)式)的强(mild)解在$[0, \tau)$上都恒等.

(ⅲ) 令$\lambda>0, $ $L(X)\ni C$是单射, $D(P_{n\lambda}^{-1}C)=X, $ $n\in {\mathbb N}.$$CA_{j}\subseteq A_{j}C, $ $j\in {\mathbb N}_{n-1}^{0}, $且设对任意正实数$\sigma>0$$X$中收敛于$0$的序列$(x_{n})_{n\in {\mathbb N}}, $都有

$ \lim \limits_{n\rightarrow \infty}{\rm e}^{-n\lambda \sigma}P_{n\lambda}^{-1}Cx_{n}=0. $ (3.3)

则方程(0.1)的任意两个具有相同初值的强(mild)解在$[0, \tau)$上都恒等.

 我们已经叙述了如何证明(ⅰ).应用方程(0.1)的线性性及文献[22, 定理3.4(ⅱ)]可直接得到(ⅱ); 相应地, 应用文献[21, 定理2.10.44]可得到(ⅲ) (参见文献[24, 定理3.2]).

与Caputo分数阶导数的多项方程的情形不同, 在存在(局部)$C_{1}$ -存在族的条件下要陈述方程(0.1)的强解的存在性的令人满意的结果是非常困难的.在文献[22, 定理3.7, 注3.8]中, 我们已经研究了Caputo分数阶导数的非齐次多项方程.注意到只要有$\alpha>0, $ ${\mathbf D}_{t}^{\alpha}u(t)$有定义, 且对所有$i\in {\mathbb N}_{m-1}^{0}$都有$u^{(i)}(0)=0$, 就成立$D_{t}^{\alpha}u(t)={\mathbf D}_{t}^{\alpha}u(t), $ $t>0, $故上述的结果经过适当的修正对Riemann-Liouville分数阶导数的多项方程的非齐次问题仍然是成立的.进一步地, 想要陈述清楚$k$ -正则$C_{1}$ -存在生成族($C_{2}$ -唯一生成族, $C$ -正则生成族)与$k$ -正则$C_{1}$ -存在族($C_{2}$ -唯一族, $C$ -正则族)之间的关系也是非常困难的; 关于Caputo分数阶导数的方程的上述方面的结果, 有兴趣的读者可参考文献[21, 定义2.10.35, 定理2.10.36-2.10.37, 注2.10.38].举例来说, 设在子类(SC1)中有$S=\emptyset, $ $1\leq i\leq m_{n}-1, $ $1\leq l\leq s_{i}, $ $\chi_{{\mathcal D}_{i}^{l}}(n)=1, $$\chi_{{\mathcal D}_{i}^{l}}(j)=0$对所有$j\in {\mathbb N}_{n-1}^{0}$成立.则对$E=X=Y$有下述成立(对$C_{2}$ -唯一族和$C$ -正则族也有类似的结论成立):

(ⅰ) 设$((R_{i, l}(t))_{t\in [0, \tau)})_{1\leq i \leq m_{n}-1, 1\leq l\leq s_{i}}$是方程(0.1)的(局部) $k$-正则$C_{1}$-存在生成族, 且$k_{1}(t)=(k\ast g_{\alpha_{n}+i-m_{n}})(t), $ $t\in [0, \tau).$定义$E(t)x:=(g_{m_{n}-1}\ast R_{i, l}(\cdot)x)(t), $ $t\in [0, \tau), $ $x\in X.$$(E(t))_{t\in [0, \tau)}$是(局部) $k_{1}$-正则$C_{1}$-存在族.

(ⅱ) 设$(E(t))_{t\in [0, \tau)}$是(局部) $k_{1}$-正则$C_{1}$-存在族, 且存在$l_{0}\in {\mathbb N}_{m_{n}-1}^{0}$使得$(k_{1}\ast g_{m_{n}-1-l_{0}}) (t)=(k\ast g_{\alpha_{n}+i-m_{n}})(t), $ $t\in [0, \tau).$$(E^{(l_{0})}(t))_{t\in [0, \tau)}\subseteq L(X)$满足泛函方程(2.6)的强连续算子族, 其中$(R_{i, l}(t))_{t\in [0, \tau)}$$(E^{(l_{0})}(t))_{t\in [0, \tau)}$代替.

现在通过一个例子来说明结论(ⅱ).

例 3.2 设$H=L^2(\Omega, \mu), $ $(Af)(x) := m(x)f(x), $其中$D(A) := \{f \in H: mf \in H\}$, $m: \Omega \to {\mathbb R}$是可测函数.再设存在常数$\omega \ge 0$使得对$x \in D(A)$$(Ax, x) \le-\omega \|x\|^2, $故而$m(x) \le-\omega$对几乎所有$x \in \Omega$成立.设常数$c$满足$c^2 \le \omega, $ $1/2 < \alpha < 1, $ $c\in {\mathbb R}, $且设$f \in H$.考虑分数阶电报方程

$ \begin{eqnarray*} && D_{t}^{2\alpha}u(t, x) + 2c D_{t}^{\alpha}u(t, x) = Au(t, x), \quad t >0, \ x\in \Omega, \\ &&(g_{2-2\alpha}*u)^{\prime}_t(0, x)=f(x), \quad (g_{2-\alpha}*u)(0, x)=0. \end{eqnarray*} $

由(A)-(B)以及来自于文献[34, 例8.2]的分析, 易得对$x_{1, 1}=f(\cdot)$$x_{0, 1}=0, $相应的积分方程(2.4)的唯一的mild解为

$ \begin{eqnarray*} u(t, x)&=& \frac{1}{2}\Big[\Big( 1- {\rm i}\frac{c}{h(x)}\Big)t^{\alpha-1} E_{\alpha, \alpha}\Big(t^\alpha({\rm i}h(x)-c)\Big) \\ && + \Big(1+ {\rm i}\frac{c}{h(x)} \Big)t^{\alpha-1} E_{\alpha, \alpha}\Big(t^\alpha(-{\rm i}h(x)-c)\Big)\Big]f(x), \end{eqnarray*} $

其中应用了等式

$ \int_0^t g_{\alpha-1}(t-s) E_\alpha \bigl(as^\alpha \bigr){\rm d}s= t^{\alpha-1} E_{\alpha, \alpha}\bigl(at^\alpha \bigr), \quad t>0 $

$ h(x) = \sqrt{-(c^2 + m(x))}. $

例 3.3 设$1\leq p\leq \infty, $ $X:=L^{p}({\mathbb R}), $ $a\in {\mathbb R}, $ $r>0, $ $1/2 < \gamma\leq 1, $ $T>0, $ $\vartheta \in L^{\infty}({\mathbb R})$, $f\in C([0, T] : X).$$A_{2}:=a{\rm d}/{\rm d}x$$A_{1}u:=r\Delta u-\vartheta(\cdot)u$, 并且赋予分布意义下的最大的定义域.延续着肖体俊和梁进在文献[48, 例4.1]中的分析方法, 我们最近在文献[22, 例5.3]和文献[21, 例2.10.33]中对带阻尼的Klein-Gordon方程的分数阶类比

$ \begin{array}{ll} {\mathbf D}_{t}^{2\gamma}u(t, x)+ a\frac{\partial}{\partial x}{\mathbf D}_{t}^{\gamma}u(t, x)-r\Delta_{x}u(t, x)+\vartheta(x)u(t, x)=f(t, x), \quad t > 0, \ x\in {\mathbb R}, \\ u(0, x)=\phi(x), \ u_{t}(0, x)=\psi(x), \quad x\in {\mathbb R}, \end{array} $ (3.4)

证明了部分关于$C$ -适定性的结果.现在把方程(3.4)中的Caputo导数${\mathbf D}_{t}^{2\gamma}u(t, x)$${\mathbf D}_{t}^{\gamma}u(t, x)$换成相应的Riemann-Liouville导数$D_{t}^{2\gamma}u(t, x)$$ D_{t}^{\gamma}u(t, x);$简便起见, 我们仅考虑(3.4)式的齐次问题, 即假设$f(t, x)\equiv 0, $并把方程简记为(KG-RL).首先注意到方程(KG-RL)是(SC1)类型的, 其中$n=3, $ $\alpha_{3}=2\gamma, $ $m_{3}=2, $ $\alpha_{2}=\gamma, $ $m_{2}=1, $ $\alpha_{0}=\alpha_{1}=m_{0}=m_{1}=0$, $A_{0}=0.$与之前在第2章中的研究相适应, 对方程(KG-RL), 赋予两个初值$x_{1, 1}=(g_{2-2\gamma}\ast u)^{\prime}(0)$$u(0)=x_{0, 1}.$

先设函数$\vartheta(\cdot)$是正常数: $\vartheta(x)\equiv \vartheta>0 .$$1 < p < \infty$时, 取$\kappa := |1/2-1/p|, $$p\in \{1, \infty\}$时, 取$\kappa>1/2.$$C:=(1-\Delta)^{-\frac{1}{2}\kappa}.$则对$\gamma=1$问题(3.4)存在强$C$ -生成族$\{(S_{0}(t))_{t\geq 0}, (S_{1}(t))_{t\geq 0}\}$ (见文献[49, p116, 定义5.1; p130, 例5.8];值得注意的是, 一般来说, 关于是否能对形如方程(0.1)的方程建立强$C$ -生成族的概念的问题是很难解决的; 然而对一些特定的情形或者一些特定的抽象多项分数阶微分方程则有可能解决).定义

$ S_{1, \gamma}(t)x:=\int ^{\infty}_{0}t^{-\gamma}\Phi_{\gamma}\bigl(st^{-\gamma}\bigr)S_{1}^{\prime}(s)x {\rm d}s, \ x\in X, \ t>0, S_{1, \gamma}(0):=S_{1}^{\prime}(0), $ (3.5)

且在(3.5)式中分别用$(S_{1}(t))_{t\geq 0}$$(\int^{t}_{0}S_{1}(s) {\rm d}s)_{t\geq 0}$代替$(S_{1}^{\prime}(t))_{t\geq 0}, $可分别定义出$(S_{1}^{\gamma}(t))_{t\geq 0}$$(S_{1, \gamma, -1}(t))_{t\geq 0}$.当$C^{-1}x_{1, 1}\in W^{2, p}({\mathbb R})$时(若$1 < p < \infty, $这意味着$x_{1, 1}\in {\mathrm S}^{2+\kappa, p}({\mathbb R}), $即文献[41, p297, 定义12.3.1]中的$2+\kappa$阶分数次Sobolev空间), 利用Laplace变换以及文献[5, 定理3.1]中的讨论, 不难证明函数$u(t;x_{1, 1}):=(g_{2\gamma-1}\ast S_{1, \gamma}(t)C^{-1}x_{1, 1})(t), $ $t\geq 0$是相应的积分方程(2.4)的唯一的强解, 其中$x_{0, 1}=0.$进一步地, 若$C^{-1}A_{1}x_{0, 1}$是良定义的(若$1 < p < \infty, $这意味着$x_{0, 1}\in {\mathrm S}^{2+\kappa, p}({\mathbb R})$), 则函数$u(t;x_{0, 1})=(g_{\gamma}\ast S_{1}^{\gamma}(\cdot)C^{-1}A_{1}x_{1, 1})(t)=S_{1, \gamma, -1}(t)C^{-1}A_{1}x, $ $t\geq 0$是相应的积分方程(2.4)的唯一的强解(见文献[49, p116, 定义5.1, l.2]), 其中$x_{1, 1}=0.$$W_{1}(t)x:=S_{1}^{\prime}(t)x, $ $W_{1}(t)x :=A_{1}\int^{t}_{0}S_{1}(s)x{\rm d}s, $$W_{2}(t):=A_{2}S_{1}^{\prime}(t)x$ ($t\geq 0, $ $x\in X$).则$(W_{i}(t))_{t\geq 0}\subseteq L(X)$是指数有界强连续族($i=1, 2, 3$)且文献[22, 定理3.5(ⅰ)-(b)]的假设成立, 故$\gamma=1$的方程(KG-RL)存在指数有界的$C$ -存在族$(S_{1}(t))_{t\geq 0}.$

以下考虑一般的情形.假设从$L^{p}({\mathbb R})$$L^{p}({\mathbb R})$的映射

$ \varphi \mapsto \bigl(1-\Delta \bigr)^{\frac{1}{2}\kappa} \bigl[ (\vartheta(\cdot)-\vartheta)\varphi(\cdot) \bigr] $ (3.6)

是连续的.应用文献[21, 定理2.10.40] (亦可参考文献[24, 定理2.3]), 可得存在指数有界的$C$ -存在族$(S_{1, \vartheta}(t))_{t\geq 0}$使得$(S_{1, \vartheta}^{\prime}(t))_{t\geq 0}$ ($\gamma=1$)是指数有界的.利用上面提到的定理的证明(例如文献[21, (437)式], $K_{0}(\cdot)$)的定义, 以及$(W_{i}(t))_{t\geq 0}$$i=1, 2, 3$的指数有界性, 不难证明算子族$(A_{j}(g_{2-j}\ast S_{1, \vartheta}^{\prime}(\cdot)\cdot)(t))_{t\geq 0}$$j=1, 2$ ($\gamma=1$)是强连续且指数有界的.也就是说, 扰动$C$ -存在族$(S_{1, \vartheta}(t))_{t\geq 0}$也满足文献[21, 定理3.5(ⅰ)-(b)]的假设.故由从属原理[22, 定理4.1]可得(KG-RL)存在指数有界的$C$ -存在族$(S_{\gamma, \vartheta}(t))_{t\geq 0}$.进一步地, 算子族$(S_{\gamma, \vartheta}^{\prime}(t))_{t\geq 0}$是强连续且指数有界的.若$C^{-1}A_{1}x_{1, 1}$是良定义的, 由方程(3.2)和文献[21, 定理3.5(i)], 易得函数$u(t;x_{0, 1})=(g_{2\gamma}\ast S_{\gamma, \vartheta}^{\prime}(\cdot)C^{-1}A_{1}x_{0, 1})(t), $ $t\geq 0$相应的积分方程(2.4)的唯一的强解, 其中$x_{1, 1}=0.$$x_{0, 1}=0$的情形, 则有对任一$x_{1, 1}\in X$, 函数$u(t;x_{1, 1})=(g_{2\gamma-1}\ast S_{\gamma, \vartheta}^{\prime}(\cdot)x_{1, 1})(t), $ $t\geq 0$是相应的积分方程(2.4)的唯一的mild解, 其中$x_{0, 1}=0$ (就作者所知, 可保证解是强解的关于$x_{1, 1}$的条件是并非显见的).

最后, 我们可以注意到, 当赋予适当的初始条件时, 可以类似地考虑如下的多项方程

$ D_{t}^{2\gamma}u(t, x)+ \biggl(\rho_{1}\frac{\partial^{3}}{\partial x^{3}}-\rho_{2}\frac{\partial^{2}}{\partial x^{2}} \biggr) D_{t}^{\gamma}u(t, x)+\biggl( c\frac{\partial^{2}}{\partial x^{2}}+a(x)\biggr)u(t, x)=f(t, x) $

$C$ -适定性.

参考文献
[1] Alvarez-Pardo E, Lizama C. Mild solutions for multi-term time-fractional differential equations with nonlocal initial conditions. Electron J Diff Equ , 2014, 39 : 1–10.
[2] Araya D, Lizama C. Existence of asymptotically almost automorphic solutions for a third order differential equation. Electron J Diff Equ , 2012, 53 : 1–20.
[3] Arendt W, Batty C J K, Hieber M, Neubrander F. Vector-valued Laplace Transforms and Cauchy Problems . Basel: Birkhäuser , 2001 .
[4] Ashyralyev A, Dal F, Pinar Z. A note on the fractional hyperbolic differential and difference equations. Applied Math Comp , 2011, 217 : 4642–4664.
[5] Bazhlekova E. Fractional Evolution Equations in Banach Spaces[D]. Eindhoven: Eindhoven University of Technology, 2001 http://www.oalib.com/references/13074216
[6] Bogacheva Yu V. Resolution's Problems of Initial Problems for Abstract Differential Equations with Fractional Derivatives[D]. Belgorod: Belgorod Univ, 2006
[7] Bogacheva Yu V, Glushak A V. About one abstract problem such as Cauchy with fractional derivatives//The International Conference on Differential Equations and Dynamical Systems. Suzdal , 2006 : 224.
[8] Chen C, Li M. On fractional resolvent operator functions. Semigroup Forum , 2010, 80 : 121–142. DOI:10.1007/s00233-009-9184-7
[9] Chen C, Li M, Li F B. On boundary values of fractional resolvent families. J Math Anal Appl , 2011, 384 : 453–467. DOI:10.1016/j.jmaa.2011.05.074
[10] Conejero J A, Lizama C, Rodenas F. Chaotic behaviour of the solutions of the Moore-Gibson-Thompson equation. Preprint .
[11] Diethelm K. The Analysis of Fractional Differential Equations . Berlin: Springer-Verlag , 2010 .
[12] Fujita Y. Integrodifferential equation which interpolates the heat equation and the wave equation (Ⅰ). Osaka J Math , 1990, 27 : 309–321.
[13] Fujita Y. Integrodifferential equation which interpolates the heat equation and the wave equation (Ⅱ). Osaka J Math , 1990(27) : 797–804.
[14] Guidetti D, Karasozen B, Piskarev S. Approximation of abstract differential equations. J Math Sci , 2004, 122 : 3013–3054. DOI:10.1023/B:JOTH.0000029696.94590.94
[15] Haase M. The Functional Calculi for Sectorial Operators . Basel: Birkhäuser Verlag , 2006 .
[16] Hilfer R. Applications of Fractional Calculus in Physics. Singapore: World Scientific Publ Co , 2000 .
[17] Jiang H, Liu F, Turner I, Burrage K. Analytical solutions for the multi-term time-space Caputo-Riesz fractional advection-diffusion equations on a finite domain. J Math Anal Appl , 2012, 389 : 1117–1127. DOI:10.1016/j.jmaa.2011.12.055
[18] Kaltenbacher B, Lasiecka I. Exponential decay for low and higher energies in the third order linear Moore-Gibson-Thompson equation with variable viscosity. Palestine J Math , 2012, 1 : 1–10.
[19] Kilbas A A, Srivastava H M, Trujillo J J. Theory and Applications of Fractional Differential Equations. Amsterdam: Elsevier , 2006.
[20] Kostić M. Generalized Semigroups and Cosine Functions. Belgrade: Mathematical Institute SANU , 2011.
[21] Kostić M. Abstract Volterra Integro-Differential Equations. Boca Raton: CRC Press , 2015 .
[22] Kostić M, Li C G, Li M. On a class of abstract time-fractional equations on locally convex spaces. Abstr Appl Anal, 2012, Article ID: 131652
[23] Kostić M. Generalized well-posedness of hyperbolic Volterra equations of non-scalar type. Ann Acad Rom Sci Ser Math Appl , 2014, 6 : 19–45.
[24] Kostić M, Li C G, Li M. Abstract multi-term fractional differential equations. Krag J Math , 2014, 38 : 51–71. DOI:10.5937/KgJMath1401051L
[25] Kostić M. On the existence and uniqueness of solutions of certain classes of abstract multi-term fractional differential equations. Funct Anal Appr Comp , 2014, 6 : 13–33.
[26] Kostić M. Abstract differential operators generating fractional resolvent families. Acta Math Sin (English Ser) , 2014, 30 : 1989–1998. DOI:10.1007/s10114-014-1279-8
[27] Kostić M. (a, k)-Regularized (C1, C2)-existence and uniqueness families. Bull Cl Sci Math Nat Sci Math , 2013, 38 : 9–26.
[28] Kostić M. Degenerate multi-term fractional differential equations in locally convex spaces. Preprint .
[29] Kostić M. Hypercyclic and topologically mixing properties of degenerate multi-term fractional differential equations. Diff Eqns Dyn Sys, 2015, 3:[DOI:10.1007/s12591-015-0238-x]
[30] Langlands T A M, Henry B I, Wearne S L. Fractional cable equation models for anomalous electrodiffusion in nerve cells: finite domain solutions. SIAM J Appl Math , 2011, 71 : 1168–1203. DOI:10.1137/090775920
[31] Langlands T A M, Henry B I, Wearne S L. Fractional cable equation models for anomalous electrodiffusion in nerve cells: infinite domain solutions. J Math Biol , 2009, 59 : 761–808. DOI:10.1007/s00285-009-0251-1
[32] Langlands T A M, Henry B I, Wearne S L. Anomalous subdiffusion with multispecies linear reaction dynamics. Phys Rev E , 2008, 77(3) : 021111.
[33] Langlands T A M, Henry B I, Wearne S L. Anomalous diffusion with linear reaction dynamics: from continuous time random walks to fractional reaction-diffusion equations. Phys Rev E , 2006, 74(3) : 031116. DOI:10.1103/PhysRevE.74.031116
[34] Li C G, Kostić M, Li M, Piskarev S. On a class of time-fractional differential equations. Fract Calc Appl Anal , 2012, 15 : 639–668.
[35] Li F B, Li M, Zheng Q. Fractional evolution equations governed by coercive differential operators. Abstr Appl Anal, 2009, Article ID: 438690
[36] Li K X, Peng J G, Jia J X. Cauchy problems for fractional differential equations with Riemann-Liouville fractional derivatives. J Funct Anal , 2012, 263 : 476–510. DOI:10.1016/j.jfa.2012.04.011
[37] Li M, Chen C, Li F B. On the fractional powers of generators of fractional resolvent families. J Funct Anal , 2010, 259 : 2702–2726. DOI:10.1016/j.jfa.2010.07.007
[38] Lizama C, Poblete F. On a functional equation associated with (a, k)-regularized resolvent families. Abstr Appl Anal, 2012, Article ID: 495487
[39] Mainardi F. Fractional Calculus and Waves in Linear Viscoelasticity. London: Imperial College Press , 2010 .
[40] Mainardi F, Gorenflo R. On Mittag-Leffler-type functions in fractional evolution processes. J Comp Appl Math , 2000, 118 : 283–299. DOI:10.1016/S0377-0427(00)00294-6
[41] Martinez C, Sanz M. The Theory of Fractional Powers of Operators. Amsterdam: Elsevier , 2001.
[42] Orsingher E, Beghin L. Time-fractional telegraph equations and telegraph processes with brownian time. Probab Theory Relat Fields , 2004, 128 : 141–160. DOI:10.1007/s00440-003-0309-8
[43] Podlubny I. Fractional Differential Equations. New York: Academic Press , 1999 .
[44] Prüss J. Evolutionary Integral Equations and Applications . Basel: Birkhaüser Verlag , 1993 .
[45] Saxena R K, Mathai A M, Haubold H J. Reaction-diffusion systems and nonlinear waves. Astrophys Space Sci , 2006, 305 : 297–303. DOI:10.1007/s10509-006-9190-0
[46] Schneider W R, Wyss W. Fractional diffusion and wave equations. J Math Phys , 1989, 30 : 134–144. DOI:10.1063/1.528578
[47] Wyss W. Fractional diffusion equation. J Math Phys , 1986, 27 : 2782–2785. DOI:10.1063/1.527251
[48] Xiao T J, Liang J. Higher order abstract Cauchy problems: their existence and uniqueness families. J London Math Soc , 2003, 67(2) : 149–164.
[49] Xiao T J, Liang J. The Cauchy Problem for Higher-Order Abstract Differential Equations . Berlin: Springer-Verlag , 1998 .
[50] Yang Q. Novel Analytical and Numerical Methods for Solving Fractional Dynamical Systems[D]. Brisbane: Queensland University of Technology, 2010