数学物理学报  2017, Vol. 37 Issue (2): 326-341   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
段双双
黄守军
Keller-Segel生物学方程组周期解的爆破
段双双, 黄守军     
安徽师范大学数学计算机科学学院 安徽芜湖 241002
摘要:该文主要研究一类Keller-Segel生物学方程组柯西问题的周期解的爆破.假设初值是周期的,且初值在一个周期上的全变差满足有界性条件,可以证明Keller-Segel方程组的光滑周期解将在有限时间内爆破,同时能给出生命区间的上下界估计.
关键词Keller-Segel模型    趋化性    周期解    爆破    生命区间    
Blow up of Periodic Solutions for a Class of Keller-Segel Equations Arising in Biology
Duan Shuangshuang, Huang Shoujun     
School of Mathematics and Computer Science, Anhui Normal University, Anhui Wuhu 241002
Abstract: In this paper, we consider the blow up phenomenon of periodic solutions to the Cauchy problem for a class of Keller-Segel equations arising in biology. The initial data is assumed to be periodic and has a bounded total variations on a periodic interval. Then we are able to prove that the smooth periodic solutions of the Keller-Segel equations will blow up in finite time. Moreover, the lifespan for the solutions is also given.
Key words: Keller-Segel model     Chemotaxis     Periodic solution     Blow up     Lifespan    
1 引言和主要结果

生物的一个特性是趋化性 (chemotactic), 即微生物或者运动的细胞对某种化学物质的吸引或排斥现象.例如, 微生物偏爱朝营养物 (例如糖) 浓度高的地方运动, 或者倾向于躲避有毒物质 (例如苯酚).趋化性因其存在广泛, 变化多样而在生物领域和数学领域引起越来越多的关注, 因此人们建立许多数学模型来刻画这一生物现象.经典的趋化性模型之一是著名的Keller-Segel模型[1], 它是由Keller和Segel在1970年提出.这个趋化性模型由于在生物学中的重要作用而吸引了许多数学家进行研究. Nagai[2]研究了该模型径向对称解的整体存在性与爆破. Rascle[3]忽略细菌的增长以及所有的扩散现象, 考虑一维Keller-Segel模型的黎曼问题.对于高维情形Winkler[4]证明了该模型解的整体存在性与有界性. Zheng[5]考虑Keller-Segel模型的柯西问题, 证明了其光滑整体解存在的充要条件是, 对于方程组满足条件$\partial_r\lambda_1>0, \;\partial_s\lambda_2<0$的黎曼不变量$\left\{r, s\right\}$, 成立$r_x(x, 0)\geq0, \;s_x(x, 0)\leq0, $其中$\lambda_1, \;\lambda_2$表示方程组的特征速度. Perthame和Dalibard[6]证明了带有密度感应的双曲Keller-Segel模型弱熵解的存在性.最近, Xiang[7]研究一类带有交叉扩散的Keller-Segel趋化性模型, 证明了当空间维数小于等于3时其整体弱解的存在性, 并研究了交叉扩散系数对解的影响. Li和Zhao[8]研究一维具有对数敏感度的排斥趋化性Keller-Segel模型, 证明了具有Dirichlet边界条件且大初值的初边值问题整体经典解的存在唯一性, 并且当时间趋于无穷时, 解会以指数方式收敛到边值.同时, 他们研究了具有Neumann-Dirichlet边界条件的初边值问题的解, 它们收敛到方程组的平衡态.这个结果对任意的化学扩散系数都成立. Shi和Wang[9]研究一类多维吸引-排斥趋化性模型, 利用格林函数的方法, 得到带有小初值的柯西问题解的逐点估计, 并给出了解在$W^{s, p}$空间中的衰减率.另外, Shi和Wang[10]研究一类高维吸引-排斥趋化性模型, 证明了当排斥作用强于吸引作用时, 整体有界解的存在性, 并给出了解的光滑性证明, 以及在$W^{s, p}$空间中的衰减率; 相反地, 当吸引作用强于排斥作用时, 解会在有限时间内爆破. Hittmeir和Jüngel[11]研究带有交叉扩散项的二维Keller-Segel模型, 证明了对于充分小的交叉扩散项, 抛物-抛物型方程组弱解是整体存在的, 以及抛物-椭圆型方程组的弱解是整体存在且有界的; 同时, 他给出了抛物-椭圆型方程组解的长时间衰减估计, 并利用有限元法说明了正则化交叉扩散项的影响. Horstmann[12]在他的综述性文章中总结了关于Keller-Segel模型正向趋化性运动的最一般结果, 并给出这些结果到更普遍模型的可能推广.更进一步, 他给出了自1970年这个方程组第一次出现以来, 有关Keller-Segel模型的所有公开问题以及研究进展.

本文我们研究如下形式的Keller-Segel模型

$ \begin{equation}\label{1}\left\{\begin{array}{ll}u_t-(uv)_x=0, \\ v_t-u_x=0, \end{array}\right. \end{equation} $ (1.1)

这里$(x, t)\in\mathbb{R}\times\mathbb{R}_{+}$, $u\geqslant0$表示细菌的密度, $v$表示化学物质的浓度.我们考虑双曲型方程组 (1.1) 的柯西问题, 其中初始条件是具有周期的小初值.对于双曲型方程组, 初值为周期的柯西问题很有意义, 吸引了很多数学家进行研究. Bressan[13]给出了一维双曲守恒律方程组基本理论的综述, 着重给出一些公开问题, 其中就提到具有周期初值和大初值的柯西问题的适定性问题. Glimm和Lax[14]利用黎曼不变量研究$2\times2 $双曲型方程组的周期解和奇性形成, 并讨论了解的大时间衰减. Klainerman和Majda[15]研究了非线性弦振动方程带有周期初值的柯西问题, 其解的生命区间依赖于平衡态附近的非线性效应. Colombini和Del[16]证明了可化为对角型的非线性双曲方程组, 带有周期初值的柯西问题经典解的两个爆破结果, 第一个结果关于$N\times N$真正非线性方程组, 第二个结果关于2$\times$2弱非线性方程组, 在两种情形下其解均会在有限时间内爆破, 并得到生命区间的估计. LeFloch和Xin[17]考虑带有周期初值的气体动力学方程组, 证明了其柯西问题不存在整体光滑解, 并给出生命区间的估计. Li和Kong[18]证明了一维空气动力学方程组周期解的爆破. Qu和Xin[19]将Glimm和Lax[14]的结果推广到$3\times3$的双曲型方程组, 考虑非等熵Euler方程组的周期初值问题.通过选取合适的黎曼不变量和推广的Glimm泛函, 他们得到了当初值具有小变差$\varepsilon$时, 初值问题熵解的生命区间是$O(\varepsilon^{-2})$.最近, Kong和Wang[20]研究平均曲率流下平面曲线运动的奇性的形成.他们将平均曲率流方程组对角化, 化为一个拟线性双曲型方程组,进而研究这类方程组相应的周期解问题.通过详细研究两族特征线的相互作用,得到在初值具有小变差以及满足一定的结构条件时, 平均曲率流方程组的周期解会发生爆破.

到目前为止, 已有不少研究双曲型方程组周期解的工作. Liu, Shi和Wang[21]研究有界区域上Keller-Segel模型的定性行为, 利用Hopf分岔定理以及局部和整体分岔定理, 证明了时间周期和稳定状态的吸引-排斥Keller-Segel模型解的存在性; 同时, 基于作者已经得到的结果, 通过选取合适的参数, 可以从数值上得到精确的时间周期波结构和尖的稳定状态波结构.但是, 众所周知, 一维拟线性双曲方程组具有周期初值的柯西问题的适定性问题, 仍是一个公开问题.此外, 对于Keller-Segel模型 (1.1), 据作者所知, 它的周期解问题没有被研究.本文将研究带有周期初值的Keller-Segel模型的初值问题.我们将方程组化为$2\times2 $的对角型双曲方程组, 它是非真正非线性的, 通过对波相互作用的估计, 我们得到解的爆破结果并给出解的生命区间的上下界估计.

下面给出本文主要结果.

定理1.1  考虑方程组 (1.1) 具有如下形式的初值

$ \begin{equation}\label{2} t=0: \; u=\tilde{u}_0+\varepsilon u_0(x), \; v=\tilde{v}_0+\varepsilon v_0(x), \end{equation} $ (1.2)

其中$\varepsilon$是一个小的正参数, $\tilde{u}_0\geq0$, $u_0(x), v_0(x)\in C^1$且周期为$L$, 即

$ \begin{equation}\label{3} u_0(x+L)=u_0(x);\; v_0(x+L)=v_0(x). \end{equation} $ (1.3)

$\tilde{u}_0, u_0(x)$满足下列条件之一:

$ \begin{equation}\label{200} (a)\;\tilde{u}_0>0;\; (b)\;\tilde{u}_0=0, u_0(x)\geq0. \end{equation} $ (1.4)

进一步, 假设

$ \begin{equation}\label{4} 4\tilde{u}_0\neq3\tilde{v}_0^2, \;R_0(x), S_0(x)\mbox{不恒为常数 }, \;TV_0^L u_0(x)+TV_0^L v_0(x)>0, \end{equation} $ (1.5)

这里$R_0(x), S_0(x)$由 (2.11) 式定义, $TV_0^L u_0(x), \;TV_0^L v_0(x)$分别表示$u_0(x), v_0(x)$$[0, L]$上的全变差, 则存在$\varepsilon_0>0$使得对任意的$\varepsilon$满足$0<\varepsilon<\varepsilon_0$, 初值问题 (1.1), (1.2) 的$C^1$$(u, v)$一定在有限时间内爆破, 且

(ⅰ) 当$\tilde{u}_0>0$或者$\tilde{u}_0=0, \tilde{v}_0>0$时, 生命区间满足

$ \begin{equation}\label{5} T\approx \varepsilon^{-1}, \end{equation} $ (1.6)

也就是, 存在两个与$\varepsilon$无关的正常数$\overline{M}, \;\underline{M}$使得

$ \begin{equation}\label{6} \underline{M}\varepsilon^{-1}<T< \overline{M}\varepsilon^{-1};\end{equation} $ (1.7)

(ⅱ) 当$\tilde{u}_0=0, \tilde{v}_0<0$时, 生命区间满足

$ \begin{equation}\label{311} T\approx \varepsilon^{-2}, \end{equation} $ (1.8)

也就是, 存在两个与$\varepsilon$无关的正常数$\overline{M}', \;\underline{M}'$使得

$ \begin{equation}\label{312} \underline{M}'\varepsilon^{-2}<T<\overline{M}'\varepsilon^{-2}.\end{equation} $ (1.9)

注1.1  由于$\varepsilon$充分小, 条件 (1.4) 可保证初始时刻, 密度$u$是非负的.

注1.2  假设 (1.5) 的第一个条件是为了保证下文中黎曼不变量与原未知量之间的变换是微分同胚的, 而第二个条件和第三个条件将在证明中具有重要作用.

在黎曼不变量 (2.4) 中, 需假设$4u\neq3v^2$.而对于特殊情形$4u=3v^2$, 我们可得到如下结果:

定理1.2  当$4u=3v^2$时, 方程组 (1.1) 化为Burgers方程

$ \begin{equation}\label{201} v_t-\frac{3}{2}vv_x=0. \end{equation} $ (1.10)

考虑如下形式的初值

$ \begin{equation}\label{202} t=0: \; v=\tilde{v}_0+\varepsilon v_0(x), \end{equation} $ (1.11)

其中$\varepsilon$是一个小的正参数, $v_0(x)\in C^1$且周期为$L$, 即

$ \begin{equation}\label{203} v_0(x+L)=v_0(x). \end{equation} $ (1.12)

进一步, 假设

$ \begin{equation}\label{204}TV_0^L v_0(x)>0, \end{equation} $ (1.13)

这里$TV_0^L v_0(x)$表示$v_0(x)$$[0, L]$上的全变差, 则存在$\varepsilon_0>0$使得对任意的$\varepsilon$满足$0<\varepsilon<\varepsilon_0$, 初值问题 (1.10), (1.11) 的$C^1$$v$一定在有限时间内爆破, 且生命区间满足

$ \begin{equation}\label{205} T=\frac{2}{3\varepsilon}\cdot\frac{1}{\sup\limits_{x\in[0, L]} v_0' (x)}, \end{equation} $ (1.14)

这里$0<\sup\limits_{x\in[0, L]}v_0' (x)<\infty$.

注1.3  定理1.2的证明可以利用Li和Kong[22]的方法, 这里就不再叙述.

本文结构如下:第二节给出一些准备知识和引理; 第三节给出定理的证明.

2 准备知识和引理

我们先做一些准备工作.

$U=\bigg( \begin{array}{c} u \\ v \end{array} \bigg)$, 则方程组 (1.1) 等价于

$ \begin{equation}\label{7} U_t+A(U)U_x=0, \end{equation} $ (2.1)

其中

$ A(U)=\left(\begin{array}{cc}-v~~&-u \\ -1~~&0 \end{array}\right).\nonumber $

其特征方程为

$ \begin{equation}\label{8} \lambda^2+v\lambda-u=0. \end{equation} $ (2.2)

因此, 当$v^2+4u<0$时, 方程组 (1.1) 是椭圆型的; 当$v^2+4u=0$时, 方程组 (1.1) 是抛物型的; 当$v^2+4u>0$时, 方程组 (1.1) 是严格双曲型的.特别地, 如果$u\geq0$$(u, v)\neq(0, 0)$, 那么方程组 (1.1) 一定是严格双曲的.

在双曲区域

$ \Omega=\left\{(u, v)|v^{2}+4u>0\right\} $

上, 特征方程 (2.2) 的特征根为

$ \begin{equation}\label{9} \lambda_-=\frac{-v-\sqrt{v^{2}+4u}}{2}<0<\lambda_+=\frac{-v+\sqrt{v^{2}+4u}}{2}, \end{equation} $ (2.3)

直接计算得$\lambda_+$$\lambda_-$的右特征向量为

$ r_+=\left( \begin{array}{c} -\lambda_+ \\ 1 \end{array} \right), \;r_-=\left( \begin{array}{c} -\lambda_- \\ 1 \end{array} \right). $

经计算

$ \nabla\lambda_+\cdot r_+= \bigg(\frac{\partial \lambda_+}{\partial u}, \frac{\partial \lambda_+}{\partial v} \bigg)\cdot(-\lambda_+, 1)=\frac{v-\sqrt{v^{2}+4u}}{\sqrt{v^{2}+4u}}. $

$u=0$$v>0$时, $\nabla\lambda_+\cdot r_+\equiv0$, $\lambda_+$是Lax[23]意义下线性退化的; 当$u=0$$v<0$时, $\nabla\lambda_+\cdot r_+\neq0$, $\lambda_+$是Lax[23]真正非线性的.

同理可得

$ \nabla\lambda_-\cdot r_-= \bigg(\frac{\partial \lambda_-}{\partial u}, \frac{\partial \lambda_-}{\partial v} \bigg)\cdot(-\lambda_-, 1)=-\frac{v+\sqrt{v^{2}+4u}}{\sqrt{v^{2}+4u}}, $

$u=0$$v<0$时, $\nabla\lambda_-\cdot r_-\equiv0$, $\lambda_-$是Lax[23]意义下线性退化的; 当$u=0$$v>0$时, $\nabla\lambda_-\cdot r_-\neq0$, $\lambda_-$是Lax[23]意义下真正非线性的.

因此, 有如下结论:

命题2.1  当$(u, v)\in\Omega$时, 在Lax[23]意义下双曲型方程组 (1.1) 既不是线性退化的, 又不是真正非线性的.

利用Kong和Wang[20]的方法, 引入黎曼不变量

$ \begin{equation}\label{10} R=(v^2+4u)^{\frac{3}{2}}+12uv-v^3, \; S=(v^2+4u)^{\frac{3}{2}}-12uv+v^3, \end{equation} $ (2.4)

方程组 (1.1) 可化为如下对角形式

$ \begin{equation}\label{11}\left\{\begin{array}{ll}R_t+\lambda_-R_x=0, \\ S_t+\lambda_+S_x=0. \end{array}\right. \end{equation} $ (2.5)

相应地, 初始条件 (1.2) 化为

$ \begin{eqnarray*}\label{100}R(0, x)=(v^2(0, x)+4u(0, x))^{\frac{3}{2}} +12u(0, x)v(0, x)-v^3(0, x), \end{eqnarray*} $
$ \begin{equation}\label{12}S(0, x)=(v^2(0, x)+4u(0, x))^{\frac{3}{2}} -12u(0, x)v(0, x)+v^3(0, x).\end{equation} $ (2.6)

注2.1   对于黎曼不变量 (2.4), 我们需假设$4u\neq3v^2$.否则, 若$4u=3v^2$, 则雅克比行列式$\frac{\partial(R, S)}{\partial(u, v)}=0$.当$4u=3v^2$时, 方程组柯西问题解的讨论由定理1.2给出.

下面我们将讨论柯西问题 (1.1) 和 (1.2) 的解$(R(t, x), S(t, x))$.当$\tilde{u}_0=0$$\tilde{v}_0<0$时, 假设生命区间$T$满足

$ \begin{equation}\label{13} T\leq\overline{M}'\varepsilon^{-2}, \end{equation} $ (2.7)

这里$\overline{M}'$是与$\varepsilon$无关的常数, 将在后面的讨论中确定.假设 (2.7) 式的有效性将在第3节得到说明.而当$\tilde{u}_0>0$时, 可假设生命区间$T$满足

$ \begin{equation}\label{613} T\leq\overline{M}\varepsilon^{-1}. \end{equation} $ (2.8)

再按类似方法证明定理1.1中的$(i)$结果.由于篇幅和方法相同的缘故, 具体过程在此略去.

引入Landau记号$O$, 于是 (2.7) 式可写为

$ \begin{equation}\label{14} T=O(\varepsilon^{-2}). \end{equation} $ (2.9)

由Taylor展式, (2.6) 式可化为

$ \begin{eqnarray*}\label{101}R(0, x)=(\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}} +12\tilde{u}_0\tilde{v}_0 -\tilde{v}_0^3+\varepsilon R_0(x)+O(\varepsilon^2), \end{eqnarray*} $
$ \begin{equation}\label{15}S(0, x)=(\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}} -12\tilde{u}_0\tilde{v}_0 +\tilde{v}_0^3+\varepsilon S_0(x)+O(\varepsilon^2), \end{equation} $ (2.10)

其中

$ \begin{eqnarray*}\label{102}R_0(x)=3(4v_0\tilde{u}_0+4u_0\tilde{v}_0-v_0\tilde{v}_0^2+(2u_0+v_0\tilde{v}_0) \sqrt{\tilde{v}_0^2+4\tilde{u}_0}), \end{eqnarray*} $
$ \begin{equation}\label{16}S_0(x)=3(-4v_0\tilde{u}_0-4u_0\tilde{v}_0+v_0\tilde{v}_0^2+(2u_0+v_0\tilde{v}_0) \sqrt{\tilde{v}_0^2+4\tilde{u}_0}), \end{equation} $ (2.11)

这里$R_0(x), S_0(x)$均为周期函数, 由假设 (1.5), $R_0(x)$, $S_0(x)$均不恒为常数.

于是

$ \begin{equation}\label{17}R_x(0, x)=\varepsilon R_0'(x)+O(\varepsilon^2), \; S_x(0, x)=\varepsilon S_0'(x)+O(\varepsilon^2). \end{equation} $ (2.12)

下面我们证明, 在解的存在区域$0\leq t\leq T$上, 解$(R(t, x), S(t, x))$仍然满足 (2.10) 式.

事实上, 令$x=x_1(t, y)$是过$x$轴上任一点$(0, y)$的第一后向特征线, 则

$ \begin{equation}\label{18}\left\{\begin{array}{ll} \frac{{\rm d}x_1(t, y)}{{\rm d}t}=\lambda_-(R(t, x_1(t, y)), S(t, x_1(t, y))), \\ x_1(0, y)=y. \end{array}\right. \end{equation} $ (2.13)

对 (2.5) 式的第一个方程沿着$x=x_1(t, y)$积分, 有

$ \label{19} R(t, x_1(t, y))=R(0, y), $ (2.13’)

由 (2.10) 式, 我们有, 对$\forall\, t\in[0, T], $

$ \begin{equation}\label{20} R(t, x_1(t, y))=(\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}}+12\tilde{u}_0\tilde{v}_0 -\tilde{v}_0^3+\varepsilon R_0(y)+O(\varepsilon^2). \end{equation} $ (2.14)

类似可得

$ \begin{equation}\label{21}\left\{\begin{array}{ll} \frac{{\rm d}x_2(t, z)}{{\rm d}t}=\lambda_+(R(t, x_2(t, z)), S(t, x_2(t, z))), \\ x_2(0, z)=z. \end{array}\right. \end{equation} $ (2.15)

以及

$ \begin{equation}\label{22} S(t, x_2(t, z))=(\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}}-12\tilde{u}_0\tilde{v}_0 +\tilde{v}_0^3+\varepsilon S_0(z)+O(\varepsilon^2), \end{equation} $ (2.16)

这里$x=x_2(t, z)$是过点$(0, z)$的第二前向特征线.

因此, 我们得到解的如下先验估计:

引理2.1  在解的存在区域$0\leq t\leq T$上,

$ R(t, x)= (\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}}+12\tilde{u}_0\tilde{v}_0 -\tilde{v}_0^3+O(\varepsilon), $
$ S(t, x)= (\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}}-12\tilde{u}_0\tilde{v}_0 +\tilde{v}_0^3+O(\varepsilon), $

且沿着特征线, 我们有

$ \begin{equation}\label{23}\left\{\begin{array}{ll}R(t, x_1(t, y))= (\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}}+12\tilde{u}_0\tilde{v}_0 -\tilde{v}_0^3+\varepsilon R_0(y)+O(\varepsilon^2), \\ S(t, x_2(t, z))=(\tilde{v}_0^2+4\tilde{u}_0)^{\frac{3}{2}}-12\tilde{u}_0\tilde{v}_0 +\tilde{v}_0^3+\varepsilon S_0(z)+O(\varepsilon^2). \end{array}\right. \end{equation} $ (2.17)

注2.2  由于沿着特征线, $\lambda_+-\lambda_-=\sqrt{v^2+4u}=(\frac{R+S}{2})^{\frac{1}{3}}= (\tilde{v}_0^2+4\tilde{u}_0)^{\frac{1}{2}}+\frac{2u_0+\tilde{v}_0v_0} {\sqrt{\tilde{v}_0^2+4\tilde{u}_0}} \varepsilon+O(\varepsilon^2)$, 所以当$\varepsilon$充分小时, $\lambda_+-\lambda_->0$, 即解在所讨论的时间区间内一直处于双曲区域.

对 (2.13) 式关于$y$求导, 有

$ \begin{equation}\label{24}\left\{\begin{array}{ll} \frac{\rm d}{{\rm d}t}(\frac{\partial x_1(t, y)}{\partial y})=\frac{\partial \lambda_-}{\partial R}R_y+\frac{\partial \lambda_-}{\partial S}S_x\frac{\partial x_1}{\partial y}, \\ \frac{\partial x_1}{\partial y}(0, y)=1. \end{array}\right. \end{equation} $ (2.18)

由 (2.5) 式中的第二式, 我们有

$ \begin{equation}\label{25} S_x=\frac{\frac{{\rm d}S}{{\rm d}_1 t}}{\lambda_--\lambda_+}, \end{equation} $ (2.19)

其中

$ \begin{equation}\label{26} \frac{{\rm d}S}{{\rm d}_1 t}=S_t+\lambda_-S_x, \end{equation} $ (2.20)

定义函数$h=h(R, S)$, 使得

$ \begin{equation}\label{27} h_S=\frac{\frac{\partial \lambda_-}{\partial S}}{\lambda_--\lambda_+}. \end{equation} $ (2.21)

由引理2.1, 在解的存在区域内成立

$ \|(h_R, h_S)\|_{C^0}\leq C. $

将 (2.19), (2.21) 式代入 (2.18) 式得

$ \begin{equation}\label{28}\left\{\begin{array}{ll} \frac{\rm d}{{\rm d}t}(\frac{\partial x_1(t, y)}{\partial y})=\frac{\partial \lambda_-}{\partial R}R_y+\frac{{\rm d}h}{{\rm d}_1 t}\frac{\partial x_1}{\partial y}, \\ \frac{\partial x_1}{\partial y}(0, y)=1, \end{array}\right. \end{equation} $ (2.22)

解得

$ \begin{equation}\label{29} \frac{\partial x_1}{\partial y} ={\rm e}^{\int_0^tA(\tau){\rm d}\tau} \bigg(1+\int_0^t{\rm e}^{-\int_0^tA(s){\rm d}s}\frac{\partial \lambda_-}{\partial R}R_y {\rm d}\tau \bigg), \end{equation} $ (2.23)

其中

$ \begin{equation}\label{30} A(\tau)=\frac{{\rm d}h}{{\rm d}_1 t} (\tau, x_1(\tau, y)). \end{equation} $ (2.24)

对 (2.15) 式关于$z$求导, 有

$ \begin{equation}\label{31}\left\{\begin{array}{ll} \frac{\rm d}{{\rm d}t}(\frac{\partial x_2(t, z)}{\partial z})=\frac{\partial \lambda_+}{\partial R}R_x\frac{\partial x_2}{\partial z}+\frac{\partial \lambda_+}{\partial S}S_z, \\ \frac{\partial x_2}{\partial z}(0, z)=1. \end{array}\right. \end{equation} $ (2.25)

由 (2.5) 式第一式, 我们有

$ \begin{equation}\label{32} R_x=\frac{\frac{{\rm d}R}{{\rm d}_2 t}}{\lambda_+-\lambda_-}, \end{equation} $ (2.26)

其中

$ \begin{equation}\label{33} \frac{{\rm d}R}{{\rm d}_2 t}=R_t+\lambda_+R_x. \end{equation} $ (2.27)

定义函数$g=g(R, S)$, 使得

$ \begin{equation}\label{34} g_R=\frac{\frac{\partial \lambda_+}{\partial R}}{\lambda_+-\lambda_-}, \end{equation} $ (2.28)

于是由引理2.1, 在解的存在区域内

$ \|(g_R, g_S)\|_{C^0}\leq C. $

将 (2.26), (2.28) 式代入 (2.25) 式有

$ \begin{equation}\label{35}\left\{\begin{array}{ll} \frac{\rm d}{{\rm d } t}(\frac{\partial x_2(t, z)}{\partial z})=\frac{{\rm d}g}{{\rm d}_2 t}\frac{\partial x_2}{\partial z}+\frac{\partial \lambda_+}{\partial S}S_z, \\ \frac{\partial x_2}{\partial z}(0, z)=1, \end{array}\right. \end{equation} $ (2.29)

解得

$ \begin{equation}\label{36} \frac{\partial x_2}{\partial z}={\rm e}^{\int_0^tB(\tau){\rm d}\tau} \bigg(1+\int_0^t{\rm e}^{-\int_0^tB(s){\rm d}s}\frac{\partial \lambda_+}{\partial S}S_z {\rm d}\tau\bigg), \end{equation} $ (2.30)

其中

$ \begin{equation}\label{37} B(\tau)=\frac{{\rm d}g}{{\rm d}_2 t}(\tau, x_2(\tau, z)). \end{equation} $ (2.31)

引理2.2  (1) $\lambda_-(R, S)=\lambda_-^0+O(\varepsilon)$, $\lambda_+(R, S)=\lambda_+^0+O(\varepsilon)$, 其中$\lambda_-^0=\frac{-\tilde{v}_0-\sqrt{\tilde{v}_0^2+4\tilde{u}_0}}{2}$, $\lambda_+^0=\frac{-\tilde{v}_0+\sqrt{\tilde{v}_0^2+4\tilde{u}_0}}{2}$;

(2) 若$\tilde{u}_0>0$或者$\tilde{u}_0=0, \tilde{v}_0>0$, 则$\frac{\partial \lambda_-}{\partial R}(R, S)=C_1+O(\varepsilon)$, 其中$C_1=\frac{4\tilde{u}_0-\tilde{v}_0^2-\tilde{v}_0\sqrt{\tilde{v}_0^{2}+4\tilde{u}_0}} {6(3\tilde{v}_0^2-4\tilde{u}_0)(\tilde{v}_0^{2}+4\tilde{u}_0)}$ $>0$且为常数;

$\tilde{u}_0=0, \tilde{v}_0<0$, 则$\frac{\partial \lambda_-}{\partial R}(R, S)=C_2S_0(z)\varepsilon+O(\varepsilon^2)$, 其中$C_2=-\frac{1}{54\tilde{v}_0^5}>0$且为常数.

  (1) 由Taylor展式, 有

$ \begin{eqnarray*}\lambda_-(R, S)&=&\lambda_-(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))\\ &&+\sum_{j=1}^m\frac{1 }{j!} \bigg[(R-R(\tilde{u}_0, \tilde{v}_0))\frac{\partial }{\partial R}+(S-S(\tilde{u}_0, \tilde{v}_0))\frac{\partial }{\partial S} \bigg]^j\lambda_-(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))\\ &&+O((|R-R(\tilde{u}_0, \tilde{v}_0)|+|S-S(\tilde{u}_0, \tilde{v}_0)|)^{m+1}), \end{eqnarray*} $

其中

$ \frac{\partial \lambda_-}{\partial R}=\frac{\partial \lambda_-}{\partial u}\cdot\frac{\partial u}{\partial R}+\frac{\partial \lambda_-}{\partial v}\cdot\frac{\partial v}{\partial R}, \\ \frac{\partial \lambda_-}{\partial S}=\frac{\partial \lambda_-}{\partial u}\cdot\frac{\partial u}{\partial S}+\frac{\partial \lambda_-}{\partial v}\cdot\frac{\partial v}{\partial S}. $

由 (2.4) 式可得

$ R+S=2(v^2+4u)^{\frac{3}{2}}, \; R-S=24uv-2v^3. $

将上式两边关于$R$求导, 有

$ 1=6(vv_R+2u_R)\sqrt{v^2+4u}, \qquad 1=24(uv_R+u_Rv)-6v^2v_R. $

解得

$ u_R=\frac{v^2-4u+v\sqrt{v^{2}+4u}}{12(3v^2-4u)\sqrt{v^{2}+4u}}, \\ v_R=\frac{2v-\sqrt{v^{2}+4u}}{6(3v^2-4u)\sqrt{v^{2}+4u}}. $

进而

$ \frac{\partial \lambda_-}{\partial R}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))= \frac{4\tilde{u}_0-\tilde{v}_0^2-\tilde{v}_0\sqrt{\tilde{v}_0^{2}+4\tilde{u}_0}} {6(3\tilde{v}_0^2-4\tilde{u}_0)(\tilde{v}_0^{2}+4\tilde{u}_0)}, $

类似计算可得

$ \frac{\partial \lambda_-}{\partial S}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))= \frac{-2\tilde{v}_0^2-\tilde{v}_0\sqrt{\tilde{v}_0^{2}+4\tilde{u}_0}} {6(3\tilde{v}_0^2-4\tilde{u}_0)(\tilde{v}_0^{2}+4\tilde{u}_0)}. $

因此, 由上述Taylor展式, 我们有

$ \begin{eqnarray*} \lambda_-(R, S)&=&\lambda_-^0+ \frac{\partial \lambda_-}{\partial R}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))(R-R(\tilde{u}_0, \tilde{v}_0)) \\ &&+\frac{\partial \lambda_-}{\partial S}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))(S-S(\tilde{u}_0, \tilde{v}_0)) +O(\varepsilon^2), \end{eqnarray*} $

其中$\lambda_-^0=\frac{-\tilde{v}_0-\sqrt{\tilde{v}_0^2+4\tilde{u}_0}}{2}$.

$\tilde{u}_0>0$或者$\tilde{u}_0=0, \tilde{v}_0>0$时, $\frac{\partial \lambda_-}{\partial R}(R(\tilde{u}_0, \tilde{v}_0), $ $S(\tilde{u}_0, \tilde{v}_0))\neq0$, 且$\frac{\partial \lambda_-} {\partial S}(R(\tilde{u}_0, \tilde{v}_0), $ $S(\tilde{u}_0, \tilde{v}_0))\neq0$, 因此, 我们有

$ \lambda_-(R, S)=\lambda_-^0+O(\varepsilon)+O(\varepsilon^2) =\lambda_-^0+O(\varepsilon). $

$\tilde{u}_0=0, \tilde{v}_0<0$时, 易知$\frac{\partial \lambda_-}{\partial R}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))=0$, 但$\frac{\partial \lambda_-}{\partial S}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))\neq0$, 因此, 我们有

$ \lambda_-(R, S)=\lambda_-^0+O(\varepsilon)+O(\varepsilon^2) =\lambda_-^0+O(\varepsilon). $

综上, 我们有

$ \lambda_-(R, S)=\lambda_-^0+O(\varepsilon). $

同理可得

$ \lambda_+(R, S)=\lambda_+^0+O(\varepsilon), $

这里$\lambda_+^0=\frac{-\tilde{v}_0+\sqrt{\tilde{v}_0^2+4\tilde{u}_0}}{2}$.

(2) 类似地, 直接计算可得

$ \begin{eqnarray*} &&\frac{\partial \lambda_-}{\partial R} (R, S)=\frac{\partial \lambda_-}{\partial R}(R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))\\ &&+\sum_{j=1}^m\frac{1 }{j!} \bigg[(R-R(\tilde{u}_0, \tilde{v}_0))\frac{\partial }{\partial R} +(S-S(\tilde{u}_0, \tilde{v}_0))\frac{\partial }{\partial S} \bigg]^j\frac{\partial \lambda_-}{\partial R} (R(\tilde{u}_0, \tilde{v}_0), S(\tilde{u}_0, \tilde{v}_0))\\ &&+O((|R-R(\tilde{u}_0, \tilde{v}_0)|+|S-S(\tilde{u}_0, \tilde{v}_0)|)^{m+1})\\ &=&\frac{4\tilde{u}_0-\tilde{v}_0^2-\tilde{v}_0\sqrt{\tilde{v}_0^2+4\tilde{u}_0}} {6(3\tilde{v}_0^2-4\tilde{u}_0)(\tilde{v}_0^2+4\tilde{u}_0)}\\ &&+\frac{(17\tilde{v}_0^6-196\tilde{u}_0\tilde{v}_0^4-80\tilde{u}_0^2\tilde{v}_0^2- 192\tilde{u}_0^3) +\tilde{v}_0\sqrt{\tilde{v}_0^2+4\tilde{u}_0}(17\tilde{v}_0^4+40\tilde{u}_0\tilde{v}_0^2 +144\tilde{u}_0^2)} {36(3\tilde{v}_0^2-4\tilde{u}_0)^3(\tilde{v}_0^2+4\tilde{u}_0)^2 \sqrt{\tilde{v}_0^2+4\tilde{u}_0}}\varepsilon R_0(y)\\ &&+\frac{(27\tilde{v}_0^6-108\tilde{u}_0\tilde{v}_0^4+160\tilde{u}_0^2\tilde{v}_0^2- 64\tilde{u}_0^3) +\tilde{v}_0\sqrt{\tilde{v}_0^2+4\tilde{u}_0}(9\tilde{v}_0^4-24\tilde{u}_0\tilde{v}_0^2 +16\tilde{u}_0^2)} {36(3\tilde{v}_0^2-4\tilde{u}_0)^3(\tilde{v}_0^2+4\tilde{u}_0)^2 \sqrt{\tilde{v}_0^2+4\tilde{u}_0}}\varepsilon S_0(z)\\ &&+O(\varepsilon^2), \end{eqnarray*} $

这里用到了引理2.1的结果.

如果$\tilde{u}_0>0$或者$\tilde{u}_0=0, \tilde{v}_0>0$, 那么$\frac{4\tilde{u}_0-\tilde{v}_0^2-\tilde{v}_0\sqrt{\tilde{v}_0^2+4\tilde{u}_0}} {6(3\tilde{v}_0^2-4\tilde{u}_0)(\tilde{v}_0^2+4\tilde{u}_0)}\neq0$, 于是我们有

$ \frac{\partial \lambda_-}{\partial R}(R, S)=C_1+O(\varepsilon), $

其中

$ \begin{equation}\label{46} C_1=\frac{4\tilde{u}_0-\tilde{v}_0^2-\tilde{v}_0\sqrt{\tilde{v}_0^{2}+4\tilde{u}_0}} {6(3\tilde{v}_0^2-4\tilde{u}_0)(\tilde{v}_0^{2}+4\tilde{u}_0)}>0. \end{equation} $ (2.32)

如果$\tilde{u}_0=0, \tilde{v}_0<0$, 那么$C_1=0$, 于是上述Taylor展式可化为

$ \frac{\partial \lambda_-}{\partial R}(R, S)=C_2S_0(z)\varepsilon+O(\varepsilon^2), $

其中

$ \begin{equation}\label{302} C_2=-\frac{1}{54\tilde{v}_0^5}>0. \end{equation} $ (2.33)

证毕.

引理2.3   考虑由 (2.13) 和 (2.15) 式定义的特征线$x_1(t, y)$, $x_2(t, z)$.令

$ d = \mathop {\min }\limits_{R,S} \left\{ {{\lambda _ + },{\lambda _ - }} \right\},\;D = \mathop {\max }\limits_{R,S} \left\{ {{\lambda _ + },{\lambda _ - }} \right\}. $ (2.34)

对给定的$z_1<z_2<y$, 取$t_1$, $t_2$使得

$ x_2(t_1, z_1)=x_1(t_1, y), \qquad x_2(t_2, z_2)=x_1(t_2, y), $

$ \begin{equation}\label{48} |t_1-t_2|\leq\frac{D-d}{2Dd}|y-z_2|+\frac{|z_1-z_2|}{2d}. \end{equation} $ (2.35)

进一步, 固定$t, \;y$, 任取$z(t, y)$使得

$ x_2(t, z(t, y))=x_1(t, y), $

$ \begin{equation}\label{49} 2dt\leq|z(t, y)-y|\leq2Dt. \end{equation} $ (2.36)

  证明参见文献[24].

引理2.4  已知$y$, 对$\forall \, z\leq y$, 我们定义$\tau(z, y)$使得

$ \begin{equation}\label{50} x_1(\tau(z, y), y)=x_2(\tau(z, y), z), \end{equation} $ (2.37)

$ \begin{equation}\label{51} \frac{{\rm d}\tau(z, y)}{{\rm d}z}= \frac{\frac{\partial x_2}{\partial z}}{\lambda_--\lambda_+} \bigg|_{(\tau(z, y), z)}. \end{equation} $ (2.38)

同样地, 已知$z$, 对$\forall \, y\leq z$, 我们定义$\hat{\tau}(\hat{y}, z)$使得

$ \begin{equation}\label{52} x_1(\hat{\tau}(\hat{y}, z), \hat{y}) =x_2(\hat{\tau}(\hat{y}, z), z), \end{equation} $ (2.39)

$ \begin{equation}\label{53} \frac{{\rm d}\hat{\tau}(\hat{y}, z)}{{\rm d}\hat{y}} =\frac{\frac{\partial x_1}{\partial \hat{y}}}{\lambda_+-\lambda_-} \bigg|_{(\hat{\tau}(\hat{y}, z), \hat{y})}. \end{equation} $ (2.40)

  证明可参考文献[20], 此处从略.

引理2.5   在解的存在区域$0\leq t\leq T$内, 我们有

(ⅰ) ${\rm e}^{\int_0^tA(\tau){\rm d}\tau}=1+O(\varepsilon), \, {\rm e}^{\int_0^tB(\tau){\rm d}\tau}=1+O(\varepsilon) ;$

(ⅱ) $R_y=\varepsilon R_0' (y)+O(\varepsilon^2), \, S_z=\varepsilon S_0' (z)+O(\varepsilon^2);$

(ⅲ) $|\frac{\partial x_1}{\partial y}(t, y)|\leq C, \, |\frac{\partial x_2}{\partial z}(t, z)|\leq C$, 其中$C$为常数.

  (ⅰ) 由 (2.24) 式

$ \begin{eqnarray*}\int_0^tA(\tau){\rm d}\tau &=&\int_0^t\frac{{\rm d}h}{{\rm d}_1t}{\rm d}\tau =h(R(t, x_1(t, y)), S(t, x_1(t, y)))-h(R(0, y), S(0, y))\\ &=&h_R(R(0, y)+\theta(R(t, x_1(t, y))-R(0, y)), S(0, y)\\ &&+\theta(S(t, x_1(t, y))-S(0, y))) (R(t, x_1(t, y))-R(0, y))\\ &&+h_S(R(0, y)+\theta(R(t, x_1(t, y))-R(0, y)), S(0, y)\\ &&+\theta(S(t, x_1(t, y))-S(0, y))) (S(t, x_1(t, y))-S(0, y))\\ &=&O(\varepsilon), \end{eqnarray*} $

所以

$ {\rm e}^{\int_0^tA(\tau){\rm d}\tau}=1+O(\varepsilon). $

同理可得

$ {\rm e}^{\int_0^tB(\tau){\rm d}\tau}=1+O(\varepsilon). $

(ⅱ) 对 (2.13)'两边关于$y$求导, 有

$ R_y=R_y(0, y), $

再由 (2.12) 式, 即得

$ R_y=\varepsilon R_0' (y)+O(\varepsilon^2). $

(ⅲ) 由 (2.23) 式, 我们有

$ \frac{\partial x_1}{\partial y}=(1+O(\varepsilon)) \bigg(1+\int_0^t(1+O(\varepsilon))\cdot O(1)\cdot O(\varepsilon){\rm d}\tau \bigg)=O(1), $

所以

$ \bigg|\frac{\partial x_1}{\partial y}(t, y)\bigg|\leq C. $

类似地

$ \bigg|\frac{\partial x_2}{\partial z}(t, z)\bigg|\leq C. $

证毕.

命题2.2  对于解的存在区域内给定的$y$$t$, 令

$ y-z(t, y)=NL+\theta(t), 0\leq\theta(t)\leq L, $

其中$z(t, y)$满足

$ x_2(t, z(t, y))=x_1(t, y). $

$(i)$$\tilde{u}_0>0$或者$\tilde{u}_0=0, \tilde{v}_0>0$, 则成立

$ \begin{equation}\label{54} \frac{\partial x_1}{\partial y}(t, y) =1-\frac{C_1}{2(\lambda_+^0-\lambda_-^0)}NL\varepsilon R_0' (y)+O(\varepsilon), \end{equation} $ (2.41)

$(ii)$$\tilde{u}_0=0, \tilde{v}_0<0$, 则成立

$ \begin{equation}\label{310} \frac{\partial x_1}{\partial y}(t, y) =1-\frac{C_2}{2(\lambda_+^0-\lambda_-^0)}N\varepsilon^2R_0' (y)\int_0^LS_0(z){\rm d}z +O(\varepsilon).\end{equation} $ (2.42)

  我们仅证 (ⅱ) 的情形, 对于 (ⅰ) 的情形, 可类似证明.

由 (2.23) 式, 利用引理2.4中的坐标变换, 我们有

$ \begin{align} & \frac{\partial {{x}_{1}}}{\partial y}(t, y)=(1+O(\varepsilon ))(1+\int_{0}^{t}{(1+O(}\varepsilon ))({{C}_{2}}{{S}_{0}}(z)\varepsilon +O({{\varepsilon }^{2}}))(\varepsilon {{R}_{{{0}'}}}(y)+O({{\varepsilon }^{2}}))\text{d}\tau ) \\ & \ \ \ \ \ \ \ \ \ \ \ \ \ =1+{{C}_{2}}{{\varepsilon }^{2}}{{R}_{{{0}'}}}(y)\int_{0}^{t}{{{S}_{0}}}(z)\text{d}\tau +O(\varepsilon ) \\ \end{align} $ (2.43)
$ \begin{align} & \ \ \ \ \ \ \ \ \ \ \ \ \ =1+{{C}_{2}}{{\varepsilon }^{2}}{{R}_{{{0}'}}}(y)\int_{y}^{z(t, y)}{\frac{\frac{\partial {{x}_{2}}}{\partial z}(\tau (z, y), z)}{({{\lambda }_{-}}-{{\lambda }_{+}})(R, S)(\tau (z, y), z)}}{{S}_{0}}(z)\text{d}z+O(\varepsilon ) \\ & \ \ \ \ \ \ \ \ \ \ \ \ \ =1+\frac{{{C}_{2}}}{2(\lambda _{-}^{0}-\lambda _{+}^{0})}{{\varepsilon }^{2}}{{R}_{{{0}'}}}(y)\int_{y}^{z(t, y)}{\frac{\partial {{x}_{2}}}{\partial z}}(\tau (z, y), z){{S}_{0}}(z)\text{d}z+O(\varepsilon ) \\ & \ \ \ \ \ \ \ \ \ \ \ \ \ =1-\frac{{{C}_{2}}}{2(\lambda _{+}^{0}-\lambda _{-}^{0})}{{\varepsilon }^{2}}{{R}_{{{0}'}}}(y)\sum\limits_{n=0}^{N-1}{\int_{y-nL}^{y-(n+1)L}{\frac{\partial {{x}_{2}}}{\partial z}}}(\tau (z, y), z){{S}_{0}}(z)\text{d}z+O(\varepsilon ). \\ \end{align} $ (2.44)

类似地, 我们有

$ \begin{eqnarray} \frac{\partial x_2}{\partial z}(t, z) &=&1+C_2\varepsilon^2S_0'(z)\int_0^tR_0(\hat{y}){\rm d}\hat{\tau}+O(\varepsilon)\nonumber\\ &=&1+C_2\varepsilon^2S_0'(z)\int_z^{\hat{y}(t, z)}\frac{\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, z), \hat{y})}{(\lambda_+-\lambda_-) (R, S)(\hat{\tau}(\hat{y}, z), \hat{y})}R_0(\hat{y}){\rm d}\hat{y}+O(\varepsilon)\nonumber\\ \label{56}&=&1+\frac{C_2}{2(\lambda_+^0-\lambda_-^0)}\varepsilon^2 S_0'(z)\int_z^{\hat{y}(t, z)}\frac{\partial x_1(\hat{\tau}(\hat{y}, z), \hat{y})}{\partial \hat{y}}R_0(\hat{y}){\rm d}\hat{y}+O(\varepsilon). \end{eqnarray} $ (2.45)

将 (2.45) 式代入 (2.44) 式并整理得到

$ \begin{equation}\label{57} \frac{\partial x_1}{\partial y}(t, y)=1-\frac{C_2}{2(\lambda_+^0-\lambda_-^0)}N\varepsilon^2 R_0' (y)\int_0^LS_0(z){\rm d}z+\sum_{n=0}^{N-1}E_n+O(\varepsilon), \end{equation} $ (2.46)

其中

$ \begin{equation}\label{58} E_n=-\frac{C_2^2}{4(\lambda_+^0-\lambda_-^0)^2}\varepsilon^4\int_{y-nL}^{y-(n+1)L} \bigg(R_0'(y)S_0'(z)S_0(z) \int_z^{\hat{y}}\frac{\partial x_1(\hat{\tau}(\hat{y}, z), \hat{y})}{\partial \hat{y}}R_0(\hat{y}){\rm d}\hat{y} \bigg){\rm d}z. \end{equation} $ (2.47)

下面我们只要证明

$ \sum_{n=0}^{N-1}E_n=O(\varepsilon), $

即得 (2.42) 式.

事实上, 对给定的$n$满足$y-(n+1)L\leq z\leq y-nL$, 我们有

$ \hat{\tau}(\hat{y}, y-nL)\leq\hat{\tau}(\hat{y}, z)\leq\hat{\tau}(\hat{y}, y-(n+1)L). $

由 (2.43) 式, 有

$ \begin{eqnarray*} && \bigg|\frac{\partial x_1}{\partial y}(\hat{\tau}(\hat{y}, z), \hat{y})-\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, y-nL), \hat{y}) \bigg|\\ &\leq&|C_2\varepsilon^2 R_0'(y)\int_{\hat{\tau}(\hat{y}, z)}^{\hat{\tau}(\hat{y}, y-nL)}S_0(s, x_1(s, \hat{y})){\rm d}s| +O(\varepsilon)\\ &\leq&O(\varepsilon^2)|\hat{\tau}(\hat{y}, z)-\hat{\tau}(\hat{y}, y-nL)|+O(\varepsilon)\\ &\leq&O(\varepsilon^2) \bigg(\frac{D-d}{2Dd}|\hat{y}-(y-nL)|+\frac{L}{2d}\bigg)+O(\varepsilon)\\ &\leq&O(\varepsilon^2)\bigg(\frac{(D-d)t}{\rm d}+\frac{L}{2d}\bigg)+O(\varepsilon)\\ &=&O(\varepsilon). \end{eqnarray*} $

易得

$ \int_{y-nL}^z\frac{\partial x_1}{\partial\hat{y}}(\hat{\tau}(\hat{y}, z), \hat{y})R_0(\hat{y}){\rm d}\hat{y}\leq O(1). $

$ \tilde{E}_n=-\frac{C_2^2}{4(\lambda_+^0-\lambda_-^0)^2}\varepsilon^4 \int_{y-nL}^{y-(n+1)L} \bigg(R_0'(y)S_0'(z)S_0(z) \int_{y-nL}^{\hat{y}}\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, y-nL), \hat{y})R_0(\hat{y}){\rm d}\hat{y} \bigg){\rm d}z, $

我们有

$ \begin{eqnarray*} E_n&=&\tilde{E}_n-\frac{C_2^2}{4(\lambda_+^0-\lambda_-^0)^2} \varepsilon^4\int_{y-nL}^{y-(n+1)L} \bigg(R_0'(y)S_0'(z)S_0(z)\\ &&\times \int_z^{y-nL}\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, y-nL), \hat{y})R_0(\hat{y}){\rm d}\hat{y} \bigg){\rm d}z\\ &=&\tilde{E}_n+\frac{C_2^2}{4(\lambda_+^0-\lambda_-^0)^2}\varepsilon^4 \int_{y-nL}^{y-(n+1)L} \bigg(R_0'(y)S_0'(z)S_0(z)\\ &&\times \int_z^{y-nL} \bigg(\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, z), \hat{y}) -\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, y-nL), \hat{y})\bigg) R_0(\hat{y}){\rm d}\hat{y} \bigg){\rm d}z\\ &&-\frac{C_2^2}{4(\lambda_+^0-\lambda_-^0)^2}\varepsilon^4\int_{y-nL}^{y-(n+1)L} \bigg(R_0'(y)S_0'(z)S_0(z)\int_z^{y-nL}\frac{\partial x_1}{\partial \hat{y}} (\hat{\tau}(\hat{y}, z), \hat{y})R_0(\hat{y}){\rm d}\hat{y}\bigg){\rm d}z\\ &=&\tilde{E}_n+O(\varepsilon^{4-2+1})+O(\varepsilon^4). \end{eqnarray*} $

$\tilde{E}_n$中, 我们有

$ \begin{eqnarray*} &&\int_{y-nL}^{y-(n+1)L}(R_0'(y)S_0'(z)S_0(z)\int_{y-nL}^{\hat{y}} \bigg(\frac{\partial x_1}{\partial \hat{y}}(\hat{\tau}(\hat{y}, y-nL), \hat{y})R_0(\hat{y}){\rm d}\hat{y} \bigg){\rm d}z\\ &=&\int_{y-nL}^{\hat{y}}\frac{\partial x_1}{\partial \hat{y}} (\hat{\tau}(\hat{y}, y-nL), \hat{y})R_0(\hat{y}) \bigg(\int_{y-nL}^{y-(n+1)L}R_0'(y)S_0'(z)S_0(z){\rm d}z\bigg){\rm d}\hat{y}\\ &=&\int_{y-nL}^{\hat{y}}\frac{\partial x_1}{\partial \hat{y}} (\hat{\tau}(\hat{y}, y-nL), \hat{y})\bigg(R_0'(y) \frac{S_0^2(z)}{2}\bigg|_{y-nL}^{y-(n+1)L}\bigg) {\rm d}\hat{y}). \end{eqnarray*} $

由于$S_0(z)$是周期为$L$的周期函数, 所以

$ R_0'(y)\frac{S_0^2(z)}{2}\bigg|_{y-nL}^{y-(n+1)L}=0. $

进而, 我们有

$ \tilde{E}_n=0. $

因此

$ \sum_{n=0}^{N-1}E_n=N(O(\varepsilon^3)+O(\varepsilon^4)) =O(\varepsilon), $

这里, 由引理2.3可知

$ N=O(t)=O(\varepsilon^{-2}). $

从而, 由 (2.46) 式立得 (2.42) 式.证毕.

3 主要定理的证明

下面我们将对定理1.1进行证明.

  (1) 我们仅证$(ii)$的情形, 对于$(i)$的情形, 可类似证明.

首先我们证明生命区间的上界为$\overline{M}'\varepsilon^{-2}$.

不失一般性, 由假设 (1.5) 的第三个条件, 我们可以得到

$ TV_0^LR_0(x)>0, $

又由于$R_0' (y)\int_0^LS_0(z){\rm d}z$$y$的周期函数, 因此我们可以选取$y_*$, 使得$R_0' (y)\int_0^LS_0(z){\rm d}z$$y_*$处取得正的最大值, 记作$M_0$.由 (2.42) 式, 我们有

$ \begin{eqnarray*}\frac{\partial x_1}{\partial y}(t, y_*)&=&1-\frac{C_2}{2(\lambda_+^0-\lambda_-^0)}N\varepsilon^2M_0+O(\varepsilon)\\ &\leq&1-\frac{C_2}{L}t\varepsilon^2M_0+O(\varepsilon), \end{eqnarray*} $

其中我们利用了$NL\geq2dt$.

现在, 我们取$\overline{M}'=\frac{L}{C_2M_0}$以及充分小的$\varepsilon_0$使得

$ \frac{\partial x_1}{\partial y}(\overline{M}'\varepsilon^{-2}, y_*)<0. $

于是对任意$\varepsilon<\varepsilon_0$, 一定存在某个$\bar{t}>\overline{M}'\varepsilon^{-2}$, 使得

$ \frac{\partial x_1}{\partial y}(\bar{t}, y_*)=0. $

另一方面, 由引理2.5 (ⅱ), 我们有

$ R_y(\bar{t}, x_1(\bar{t}, y_*))=R_x(\bar{t}, x_1(\bar{t}, y_*))\frac{\partial x_1}{\partial y}=\varepsilon R_0'(y)+O(\varepsilon^2)\neq0. $

由此我们得到$R_x$一定在$\bar{t}$处爆破, 因此生命区间的上界为$T\leq\overline{M}'\varepsilon^{-2}$.由此可知, 假设 (2.7) 是合理的.

(2) 现在我们证明生命区间的下界为$\underline{M}' \varepsilon^{-2}$, 其中$\underline{M}' <\overline{M}' $.

由一阶拟线性双曲型方程组Cauchy问题局部$C^1$解的存在唯一性, 我们要得到生命区间的下界, 必须在区间$0\leq T\leq\underline{M}' \varepsilon^{-2}$上对$(R, S, R_x, S_x)$$C^0$范数作一致先验估计.

类似于$(1)$中证明, 我们可以选取$y^*$使得$R_0' (y)\int_0^LS_0(z){\rm d}z$$y^*$处取得负的最小值, 记作$M_0'$, 则

$ \begin{eqnarray*}\frac{\partial x_1}{\partial y}(t, y^*)&=&1-\frac{C_2}{2(\lambda_+^0-\lambda_-^0)}N\varepsilon^2M_0' +O(\varepsilon)\\ &\geq&1-\frac{C_2}{L}t\varepsilon^2M_0'+O(\varepsilon)\\ &\geq&1-Ct\varepsilon^2+O(\varepsilon), \end{eqnarray*} $

其中$C=\frac{C_2 M_0' }{L}<0$.

因此, 令$\underline{M}' =-\frac{1}{2C}$, 在区间$0\leq t\leq\underline{M}' \varepsilon^{-2}$上, 我们有

$ \frac{\partial x_1}{\partial y}\geq\frac{1}{4}. $

由引理2.5 (ⅱ) 我们有

$ R_x=\frac{R_y}{\frac{\partial x_1}{\partial y}}=O(\varepsilon). $

同理可得$S_x$的估计, 于是我们得到$\|(R, S, R_x, S_x)\|_{C^0}$的一致估计, 因此在区间$0\leq t\leq\underline{M}' \varepsilon^{-2}$上存在$C^1$解.

综上可得本文主要结果定理1.1.证毕.

参考文献
[1] Keller E F, Segel L A. Initiation of slime mold aggregation viewed as an instability. Theoret Biol, 1970, 26: 399–415. DOI:10.1016/0022-5193(70)90092-5
[2] Nagai T. Blow-up of radially symmetric solutions to a chemotaxis system. Adv Math Sci Appl, 1995, 5: 581–601.
[3] Rascle M. The Riemann problem for a nonlinear nonstrictly hyperbolic system arising in biology. Comput Math Appl, 1985, 11: 223–238. DOI:10.1016/0898-1221(85)90148-8
[4] Winkler M. Aggregation vs.global diffusive behavior in the higher-dimensional Keller-Segel model. J Differential Equations, 2010, 12: 2889–2905.
[5] Zheng Y S. Globally smooth solutions to Cauchy problem of a quasilinear hyperbolic system arising in biology. Acta Math Sci, 2001, 21(4): 460–468.
[6] Perthame B, Dalabard A L. Existence of solutions of the hyperbolic Keller-Segel model. Trans Amer Math Soc, 2009, 361: 2319–2335.
[7] Xiang T. On a class of Keller-Segel chemotaxis systems with cross-diffusion. J Differential Equations, 2015, 295: 4273–4326.
[8] Li H C, Zhao K. Initial-boundary value problems for a system of hyperbolic balance laws arising from chemotaxis. J Differential Equations, 2015, 528: 302–338.
[9] Shi R K, Wang W K. The pointwise estimates of solutions to the Cauchy problem of a chemotaxis model. Chin Ann Math, 2016, 37B: 111–124.
[10] Shi R K, Wang W K. Well-posedness for a model derived from an attraction-repulsion chemotaxis system. J Math Anal Appl, 2015, 423: 497–520. DOI:10.1016/j.jmaa.2014.10.006
[11] Hittmeir S, Jüngel A. Cross diffusion preventing blow-up in the two-dimensional Keller-Segel model. SIAM J Math Anal, 2011, 43: 997–1022. DOI:10.1137/100813191
[12] Horstmann D. From 1970 until present:the Keller-Segel model in chemotaxis and its consequences. Jahresber Deutsch Math Verein, 2004, 2: 51–69.
[13] Bressan A. Open questions in the theory of one dimensional hyperbolic conservation laws. The IMA Volumes in Math Appl, 2011, 153: 1–22. DOI:10.1007/978-1-4419-9554-4
[14] Glimm J, Lax P D. Decay of solutions of systems of nonlinear hyperbolic conservation laws. Mem Am Math Soc, 1970, 101: 1–112.
[15] Klainerman S, Majda A. Formation of singularities for wave equations including the nonlinear vibrating. Comm Pure Appl Math, 1980, 247: 1694–1719.
[16] Colombini F, Del Santo D. Development of singularities for nonlinear hyperbolic systems with periodic data. Ann Univ Ferrara Sez Ⅶ(N.S.), 1997, XLI: 15–28.
[17] LeFloch P, Xin Z P.Formation of singularities in periodic solutions to gas dynamics equations.1993, Unpublished
[18] Li T T, Kong D X. Blow up of periodic solution to quasilinear hyperbolic systems. Nonlinear Anal Theory Appl, 1996, 26: 1779–1789. DOI:10.1016/0362-546X(94)00366-P
[19] Qu P, Xin Z P. Long time existence of entropy solutions to the one-dimensional non-isentropic Euler equations with periodic initial data. Arch Rational Mech Anal, 2015, 216: 221–259. DOI:10.1007/s00205-014-0807-0
[20] Kong D X, Wang Z G. Formation of singularities in the motion of plane curves under hyperbolic mean curvature fow. J Differential Equations, 2009, 247: 1694–1719. DOI:10.1016/j.jde.2009.04.016
[21] Liu P, Shi J P, Wang Z A. Pattern formation of the attraction-repulsion Keller-Segel system. Discrete Contin Dyn Syst Ser B, 2013, 18: 2597–2625. DOI:10.3934/dcdsb
[22] Li D Q, Kong D X. Blow up of periodic solutions to quasilinear hyperbolic systems. Nonlinear Anal, 1996, 26: 1779–1789. DOI:10.1016/0362-546X(94)00366-P
[23] Lax P D. Hyperbolic systems of conservation laws Ⅱ. Comm Pure Appl Math, 1957, 10: 537–566. DOI:10.1002/(ISSN)1097-0312
[24] Xiao J J.Some Topics on Hyperbolic Conservation Laws[D].Hong Kong:The Chinese University of Hong Kong, 2008