September 9, 2024 2024 年 9 月 9 日 ^(1){ }^{1} School of Mathematics and Statistics, Beijing Institute of Technology, Beijing, China 北京理工大学数学与统计学院,中国北京
Chapter 1 第一章
Introduction 介绍
1.1 Objectives of these book 1.1 本书的目标
This book is an introduction to the theory of partial differential equations with two main axis 这本书是关于偏微分方程理论的介绍,主要有两个轴心
the computation of simple explicit solutions that can be studied in details in order to understand the properties of the corresponding partial differential equations, 简单显式解的计算可以详细研究,以便理解相应偏微分方程的性质,
the study of the existence, uniqueness and qualitative properties of solutions of simple partial differential equations, relying on methods which do not use explicit solutions. These methods are general and can be applied to many different equations. 对简单偏微分方程解的存在性、唯一性和定性性质的研究,依赖于不使用显式解的方法。这些方法是一般性的,可以应用于许多不同的方程。
The reader should be familiar with calculus, integration theory, implicit function theorem and elementary Fourier analysis. 读者应熟悉微积分、积分理论、隐函数定理和基础傅里叶分析。
1.2 Notations 1.2 符号
1.2.1 Derivatives 1.2.1 导数
In R^(n)\mathbb{R}^{n}, we will use the following notations 在 R^(n)\mathbb{R}^{n} 中,我们将使用以下符号
where j=1,2,dots,nj=1,2, \ldots, n and x_(j)x_{j} is the j^("th ")j^{\text {th }} coordinate. 其中 j=1,2,dots,nj=1,2, \ldots, n 和 x_(j)x_{j} 是 j^("th ")j^{\text {th }} 坐标。
In R\mathbb{R}, we will denote the partial derivative with respect to xx by del_(x)\partial_{x}. 在 R\mathbb{R} 中,我们将用 del_(x)\partial_{x} 表示对 xx 的偏导数。
Moreover, grad\nabla is the operator “nabla” 此外, grad\nabla 是算子“nabla”
is the divergence of VV 是 VV 的散度
If V(t,x)=(V_(1)(t,x),dots,V_(n)(t,x))V(t, x)=\left(V_{1}(t, x), \ldots, V_{n}(t, x)\right) and if phi(t,x)\phi(t, x) is a function, then 如果 V(t,x)=(V_(1)(t,x),dots,V_(n)(t,x))V(t, x)=\left(V_{1}(t, x), \ldots, V_{n}(t, x)\right) ,并且如果 phi(t,x)\phi(t, x) 是一个函数,那么
is called the transport of phi\phi by VV. 被称为 VV 对 phi\phi 的运输。
1.2.2 Functional spaces 1.2.2 功能空间
The measure will always be Lebesgue measure. Let Omega\Omega be R^(n),R^(+)xxR\mathbb{R}^{n}, \mathbb{R}^{+} \times \mathbb{R}, or an open set of R^(n)\mathbb{R}^{n}. The space L^(oo)(Omega)L^{\infty}(\Omega) is the set of measurable functions such that 该度量将始终是勒贝格测度。设 Omega\Omega 为 R^(n),R^(+)xxR\mathbb{R}^{n}, \mathbb{R}^{+} \times \mathbb{R} ,或 R^(n)\mathbb{R}^{n} 的开集。空间 L^(oo)(Omega)L^{\infty}(\Omega) 是可测函数的集合,使得
||f||_(L^(oo))=s u p_(x in Omega)|f(x)| < +oo.\|f\|_{L^{\infty}}=\sup _{x \in \Omega}|f(x)|<+\infty .
For 1 <= p < +oo1 \leq p<+\infty, the space L^(p)(Omega)L^{p}(\Omega) is the set of measurable function ff such that 对于 1 <= p < +oo1 \leq p<+\infty ,空间 L^(p)(Omega)L^{p}(\Omega) 是可测函数 ff 的集合,使得
||f||_(L^(p))=(int_(Omega)|f(x)|^(p)dx)^(1//p) < +oo\|f\|_{L^{p}}=\left(\int_{\Omega}|f(x)|^{p} d x\right)^{1 / p}<+\infty
We recall that, for every p in[1,oo],L^(p)(Omega)p \in[1, \infty], L^{p}(\Omega) is complete for the norm ||*||_(L^(p))\|\cdot\|_{L^{p}} and that L^(2)(Omega)L^{2}(\Omega) is an Hilbert space with the associated scalar product 我们回忆到,对于每个 p in[1,oo],L^(p)(Omega)p \in[1, \infty], L^{p}(\Omega) 在范数 ||*||_(L^(p))\|\cdot\|_{L^{p}} 下是完备的,并且 L^(2)(Omega)L^{2}(\Omega) 是一个具有相关标量积的希尔伯特空间
(:f,g:)_(L^(2))=int_(Omega)f(x)g(x)dx\langle f, g\rangle_{L^{2}}=\int_{\Omega} f(x) g(x) d x
1.2.3 Miscellaneous 1.2.3 杂项
If f(t,x)f(t, x) is a function of two (or more) variables, f(t,*)f(t, \cdot) denotes the function of xx defined by x rarr f(t,x)x \rightarrow f(t, x). 如果 f(t,x)f(t, x) 是一个两个(或更多)变量的函数, f(t,*)f(t, \cdot) 表示由 x rarr f(t,x)x \rightarrow f(t, x) 定义的 xx 的函数。
Let a inRa \in \mathbb{R}, then 1_(x∣ea)1_{x \mid e a} is the function which equals 1 if x <= ax \leq a and 0 if x > ax>a. Similarly, 1_(x >= a)1_{x \geq a} equals 1 if x >= ax \geq a and 0 if x < ax<a. More generally, if II an interval, or a set, 1_(I)(x)1_{I}(x) equals 1 if x in Ix \in I and 0 if x!in Ix \notin I. 设 a inRa \in \mathbb{R} ,则 1_(x∣ea)1_{x \mid e a} 是一个函数,当 x <= ax \leq a 时等于 1,当 x > ax>a 时等于 0。类似地, 1_(x >= a)1_{x \geq a} 在 x >= ax \geq a 时等于 1,在 x < ax<a 时等于 0。更一般地,如果 II 是一个区间或集合, 1_(I)(x)1_{I}(x) 在 x in Ix \in I 时等于 1,在 x!in Ix \notin I 时等于 0。
Chapter 2 第二章
Transport equations 运输方程
2.1 A first example in R\mathbb{R} 2.1 第一个例子在 R\mathbb{R}
2.1.1 Introduction 2.1.1 介绍
We first consider the transport equation in dimension 1 with constant speed c inRc \in \mathbb{R}. Let phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}). We want to “transport” the values of phi_(0)\phi_{0}. 我们首先考虑一维常速 c inRc \in \mathbb{R} 的传输方程。设 phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) 。我们想要“传输” phi_(0)\phi_{0} 的值。
We define the function phi(t,x)\phi(t, x) by: the value of phi\phi at time tt and position x+ctx+c t equals the value of phi_(0)\phi_{0} at position xx, namely 我们通过以下方式定义函数 phi(t,x)\phi(t, x) :在时间 tt 和位置 x+ctx+c t 时, phi\phi 的值等于在位置 xx 时 phi_(0)\phi_{0} 的值,即
phi(t,x+ct)=phi_(0)(x),quad AA t inR^(+),AA x inR\phi(t, x+c t)=\phi_{0}(x), \quad \forall t \in \mathbb{R}^{+}, \forall x \in \mathbb{R}
This defines a function phi\phi by 这通过定义一个函数 phi\phi
phi(t,x)=phi_(0)(x-ct),quad AA t inR^(+),AA x inR\phi(t, x)=\phi_{0}(x-c t), \quad \forall t \in \mathbb{R}^{+}, \forall x \in \mathbb{R}
If phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}), then phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right). We observe that 如果 phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) ,那么 phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 。我们观察到
{:[del_(t)phi+cdel_(x)phi=0","quad AA t inR^(+)","AA x inR],[phi(0","x)=phi_(0)(x)","quad AA x inR]:}\begin{aligned}
\partial_{t} \phi+c \partial_{x} \phi & =0, \quad \forall t \in \mathbb{R}^{+}, \forall x \in \mathbb{R} \\
\phi(0, x) & =\phi_{0}(x), \quad \forall x \in \mathbb{R}
\end{aligned}
The equation (2.3) is called the transport equation. Equation (2.4) provides an initial data. Let us introduce some vocabulary 方程 (2.3) 被称为输运方程。方程 (2.4) 提供了初始数据。让我们引入一些词汇。
in equation (2.1), we “follow” the values of phi\phi. It is called a Lagrangian point of view. 在方程(2.1)中,我们“跟随” phi\phi 的值。这被称为拉格朗日视角。
in equation (2.3), we do not “move”. It is called an Eulerian point of view. 在方程(2.3)中,我们不“移动”。这被称为欧拉视角。
2.1.2 Solution of the transport equation 2.1.2 运输方程的解
Theorem 2.1.1. Let phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}). Then there exists a unique solution phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) to 2.3 [2.4). Moreover, phi\phi is given by 定理 2.1.1. 设 phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) 。则存在唯一解 phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 对于 2.3 [2.4)。此外, phi\phi 由以下给出
phi(t,x)=phi_(0)(x-ct)\phi(t, x)=\phi_{0}(x-c t)
Proof. First we note that phi(t,x)=phi_(0)(x-ct)\phi(t, x)=\phi_{0}(x-c t) satisfies 证明。首先我们注意到 phi(t,x)=phi_(0)(x-ct)\phi(t, x)=\phi_{0}(x-c t) 满足
Thus phi\phi is a solution. It is C^(1)C^{1} since phi_(0)\phi_{0} is C^(1)C^{1}. 因此 phi\phi 是一个解决方案。它是 C^(1)C^{1} ,因为 phi_(0)\phi_{0} 是 C^(1)C^{1} 。
Now, if phi(t,x)\phi(t, x) is a solution, let us define 现在,如果 phi(t,x)\phi(t, x) 是一个解,我们定义为
2.1.3 Well-posedness in the sense of Hadamard 2.1.3 哈达玛德意义下的良好适定性
Definition 2.1.2. A mathematical problem is said to be well-posed in the sense of Hadamard if 定义 2.1.2。如果一个数学问题在哈达玛德的意义上被称为良好提出,则
it has a solution, 它有一个解决方案,
its solution is unique, 它的解决方案是独特的,
its solution continuously depends on the data (parameters, initial data). 其解决方案持续依赖于数据(参数、初始数据)。
The transport equation in 1D with constant velocity is well-posed in the sense of Hadamard. The previous Theorem gives the existence and uniqueness of the solution 一维常速传输方程在哈达玛尔意义上是良构的。前面的定理给出了解的存在性和唯一性。
Let us study the continuity with respect to the initial data. Let us assume that phi_(n,0)rarrphi_(oo,0)\phi_{n, 0} \rightarrow \phi_{\infty, 0} in C^(1)(R)C^{1}(\mathbb{R}) in the sense: 让我们研究关于初始数据的连续性。我们假设 phi_(n,0)rarrphi_(oo,0)\phi_{n, 0} \rightarrow \phi_{\infty, 0} 在 C^(1)(R)C^{1}(\mathbb{R}) 中的意义是:
s u p_(x inR)|phi_(n,0)(x)-phi_(oo,0)(x)|+s u p_(x inR)|del_(x)phi_(n,0)(x)-del_(x)phi_(oo,0)(x)|rarr0\sup _{x \in \mathbb{R}}\left|\phi_{n, 0}(x)-\phi_{\infty, 0}(x)\right|+\sup _{x \in \mathbb{R}}\left|\partial_{x} \phi_{n, 0}(x)-\partial_{x} \phi_{\infty, 0}(x)\right| \rightarrow 0
as n rarr+oon \rightarrow+\infty. Let phi_(n)\phi_{n} be the solutions of the transport equation with initial data phi_(n,0)\phi_{n, 0} and let phi_(oo)\phi_{\infty} be he solution of the transport equation with initial data phi_(oo,0)\phi_{\infty, 0}. We want to prove that phi_(n)(t,x)rarrphi_(oo)(t,x)\phi_{n}(t, x) \rightarrow \phi_{\infty}(t, x) in C^(1)(RxxR)C^{1}(\mathbb{R} \times \mathbb{R}) in the sense 作为 n rarr+oon \rightarrow+\infty 。设 phi_(n)\phi_{n} 为具有初始数据 phi_(n,0)\phi_{n, 0} 的输运方程的解,设 phi_(oo)\phi_{\infty} 为具有初始数据 phi_(oo,0)\phi_{\infty, 0} 的输运方程的解。我们想证明 phi_(n)(t,x)rarrphi_(oo)(t,x)\phi_{n}(t, x) \rightarrow \phi_{\infty}(t, x) 在 C^(1)(RxxR)C^{1}(\mathbb{R} \times \mathbb{R}) 的意义上。
{:[s u p_(x inR,t inR)|phi_(n)(t,x)-phi_(oo)(t,x)|+s u p_(x inR,t inR)|del_(x)phi_(n)(t,x)-del_(x)phi_(oo)(t,x)|],[+s u p_(x inR,t inR)|del_(t)phi_(n)(t,x)-del_(t)phi_(oo)(t,x)|rarr0]:}\begin{aligned}
\sup _{x \in \mathbb{R}, t \in \mathbb{R}}\left|\phi_{n}(t, x)-\phi_{\infty}(t, x)\right| & +\sup _{x \in \mathbb{R}, t \in \mathbb{R}}\left|\partial_{x} \phi_{n}(t, x)-\partial_{x} \phi_{\infty}(t, x)\right| \\
& +\sup _{x \in \mathbb{R}, t \in \mathbb{R}}\left|\partial_{t} \phi_{n}(t, x)-\partial_{t} \phi_{\infty}(t, x)\right| \rightarrow 0
\end{aligned}
as n rarr+oon \rightarrow+\infty. Let 作为 n rarr+oon \rightarrow+\infty 。让
and psi_(n,0)rarr0\psi_{n, 0} \rightarrow 0 in C^(1)(R)C^{1}(\mathbb{R}) as n rarr+oon \rightarrow+\infty. We have 和 psi_(n,0)rarr0\psi_{n, 0} \rightarrow 0 在 C^(1)(R)C^{1}(\mathbb{R}) 作为 n rarr+oon \rightarrow+\infty 。我们有
s u p_(t,x)|psi_(n)(t,x)|≲s u p_(x)|psi_(n,0)(x)|rarr0\sup _{t, x}\left|\psi_{n}(t, x)\right| \lesssim \sup _{x}\left|\psi_{n, 0}(x)\right| \rightarrow 0
s u p_(t,x)|del_(x)psi_(n)(t,x)|≲s u p_(x)|del_(x)psi_(n,0)(x)|rarr0\sup _{t, x}\left|\partial_{x} \psi_{n}(t, x)\right| \lesssim \sup _{x}\left|\partial_{x} \psi_{n, 0}(x)\right| \rightarrow 0
and similarly for del_(t)psi_(n)\partial_{t} \psi_{n}. Thus psi_(n)rarr0\psi_{n} \rightarrow 0 in C^(1)C^{1}, as n rarr+oon \rightarrow+\infty, which ends the study. 并且对于 del_(t)psi_(n)\partial_{t} \psi_{n} 也是如此。因此 psi_(n)rarr0\psi_{n} \rightarrow 0 在 C^(1)C^{1} 中,作为 n rarr+oon \rightarrow+\infty ,这结束了研究。
Remark. The solution is also continuous in cc, but the proof is a little more delicate. 备注。该解在 cc 处也是连续的,但证明稍微复杂一些。
2.1.4 Notion of weak solutions 2.1.4 弱解的概念
We note that 我们注意到
phi(t,x)=phi_(0)(x-ct)\phi(t, x)=\phi_{0}(x-c t)
has a meaning even if phi_(0)\phi_{0} in not C^(1)(R)C^{1}(\mathbb{R}). It may only be L_("loc ")^(1)L_{\text {loc }}^{1} for instance. 即使 phi_(0)\phi_{0} 不在 C^(1)(R)C^{1}(\mathbb{R}) 中,它也有意义。它可能只是 L_("loc ")^(1)L_{\text {loc }}^{1} ,例如。
" Can we say that "phi" is a solution of (2.3) if "phi_(0)inL_("loc ")^(1)" ? "\text { Can we say that } \phi \text { is a solution of (2.3) if } \phi_{0} \in L_{\text {loc }}^{1} \text { ? }
We have to change the notion of solution, since phi\phi is not differentiable. Instead of looking at precise values at a precise point (t,x)(t, x), we look at averages. This idea is at the starting point of the theory of the distributions, discovered by Laurent Schwartz (Fields medal, 1950). 我们必须改变解决方案的概念,因为 phi\phi 是不可微的。我们不再关注在精确点 (t,x)(t, x) 的精确值,而是关注平均值。这个想法是分布理论的起点,该理论由洛朗·施瓦茨(菲尔兹奖,1950 年)发现。
Let us make a preliminary computation. Let psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) be a function with compact support. Let phi_(0)\phi_{0} be C^(1)C^{1} and let phi inC^(1)\phi \in C^{1} be the corresponding solution of (2.3). Then 让我们进行初步计算。设 psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 是一个具有紧支撑的函数。设 phi_(0)\phi_{0} 为 C^(1)C^{1} ,并且设 phi inC^(1)\phi \in C^{1} 为(2.3)的相应解。然后
int_(R^(+))int_(R)psi(t,x)[del_(t)phi(t,x)+cdel_(x)phi(t,x)]dxdt=0\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \psi(t, x)\left[\partial_{t} \phi(t, x)+c \partial_{x} \phi(t, x)\right] d x d t=0
We integrate by parts in xx. There is no boundary term since the support of psi\psi is compact (i.e. psi\psi vanishes for large tt and xx ) 我们在 xx 中分部积分。由于 psi\psi 的支撑是紧致的(即 psi\psi 在大 tt 和 xx 时消失),因此没有边界项。
int_(R^(+))int_(R)psi(t,x)del_(x)phi(t,x)dxdt=-int_(R^(+))int_(R)del_(x)psi(t,x)phi(t,x)dxdt\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \psi(t, x) \partial_{x} \phi(t, x) d x d t=-\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \partial_{x} \psi(t, x) \phi(t, x) d x d t
We integrate by parts in tt. There is a boundary term at t=0t=0, 我们在 tt 处分部积分。在 t=0t=0 处有一个边界项,
{:[int_(R)int_(R^(+))psi(t","x)del_(t)phi(t","x)dtdx=-int_(R)int_(R^(+))del_(t)psi(t","x)phi(t","x)dtdx],[-int_(R)psi(0","x)phi(0","x)dx.]:}\begin{aligned}
\int_{\mathbb{R}} \int_{\mathbb{R}^{+}} \psi(t, x) \partial_{t} \phi(t, x) d t d x= & -\int_{\mathbb{R}} \int_{\mathbb{R}^{+}} \partial_{t} \psi(t, x) \phi(t, x) d t d x \\
& -\int_{\mathbb{R}} \psi(0, x) \phi(0, x) d x .
\end{aligned}
Hence 因此
int_(R^(+))int_(R)phi(t,x)[del_(t)psi(t,x)+cdel_(x)psi(t,x)]dxdt+int_(R)psi(0,x)phi_(0)(x)dx=0.\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \phi(t, x)\left[\partial_{t} \psi(t, x)+c \partial_{x} \psi(t, x)\right] d x d t+\int_{\mathbb{R}} \psi(0, x) \phi_{0}(x) d x=0 .
This formula does not involve derivatives of phi\phi. Thus, the left hand side of 2.5) has a meaning even if phi\phi is only L_(loc)^(1)(R^(+)xxR)L_{l o c}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right). We take this formula as the definition of a new notion of solutions. 这个公式不涉及 phi\phi 的导数。因此,即使 phi\phi 仅为 L_(loc)^(1)(R^(+)xxR)L_{l o c}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) ,2.5)的左侧也有意义。我们将这个公式视为新解的定义。
Definition 2.1.3. Let phi_(0)inL_(loc)^(1)(R)\phi_{0} \in L_{l o c}^{1}(\mathbb{R}). We say that phi inL_("loc ")^(1)(R^(+)xxR)\phi \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) is a weak solution of (2.3 2.4) if, for every function psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) with compact support, 定义 2.1.3. 设 phi_(0)inL_(loc)^(1)(R)\phi_{0} \in L_{l o c}^{1}(\mathbb{R}) 。我们称 phi inL_("loc ")^(1)(R^(+)xxR)\phi \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 为 (2.3 2.4) 的弱解,如果对于每个具有紧支撑的函数 psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) ,
int_(R^(+))int_(R)phi(t,x)[del_(t)psi(t,x)+cdel_(x)psi(t,x)]dxdt+int_(R)psi(0,x)phi_(0)(x)dx=0.\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \phi(t, x)\left[\partial_{t} \psi(t, x)+c \partial_{x} \psi(t, x)\right] d x d t+\int_{\mathbb{R}} \psi(0, x) \phi_{0}(x) d x=0 .
The function psi\psi is called a “test function”. Test functions are C^(1)C^{1} and with compact support. The notion of “weak solution” is linked to the particular equation under study. 函数 psi\psi 被称为“测试函数”。测试函数是 C^(1)C^{1} 并具有紧支撑。 “弱解”的概念与所研究的特定方程相关。
If phi inC^(1)\phi \in C^{1} is a solution in the usual sense, then phi\phi is called a “strong solution”, or a “classical solution”. 如果 phi inC^(1)\phi \in C^{1} 是通常意义上的解,那么 phi\phi 被称为“强解”或“经典解”。
2.1.5 An example of weak solution 2.1.5 弱解的一个例子
Let phi_(0)(x)=1_(x >= 0)\phi_{0}(x)=1_{x \geq 0}. Let us show that 让 phi_(0)(x)=1_(x >= 0)\phi_{0}(x)=1_{x \geq 0} 。让我们证明。
phi(t,x)=1_(x >= ct)\phi(t, x)=1_{x \geq c t}
is a weak solution of (2.32.4). Let psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) with compact support. Then we have 是(2.32.4)的弱解。设 psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 具有紧支撑。然后我们有
{:[int_(R^(+))int_(R)phi(t","x)[del_(t)psi(t,x)+cdel_(x)psi(t,x)]dxdt],[=int_(0)^(+oo)int_(0)^(x//c)del_(t)psi(t","x)dtdx+cint_(0)^(+oo)int_(ct)^(+oo)del_(x)psi(t","x)dxdt],[=int_(0)^(+oo)psi((x)/(c),x)dx-int_(0)^(+oo)psi(0","x)dx-cint_(0)^(+oo)psi(t","ct)dt],[=-int_(0)^(+oo)psi(0","x)dx.]:}\begin{aligned}
& \int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \phi(t, x)\left[\partial_{t} \psi(t, x)+c \partial_{x} \psi(t, x)\right] d x d t \\
= & \int_{0}^{+\infty} \int_{0}^{x / c} \partial_{t} \psi(t, x) d t d x+c \int_{0}^{+\infty} \int_{c t}^{+\infty} \partial_{x} \psi(t, x) d x d t \\
= & \int_{0}^{+\infty} \psi\left(\frac{x}{c}, x\right) d x-\int_{0}^{+\infty} \psi(0, x) d x-c \int_{0}^{+\infty} \psi(t, c t) d t \\
= & -\int_{0}^{+\infty} \psi(0, x) d x .
\end{aligned}
Thus phi\phi is a weak solution. 因此 phi\phi 是一个弱解。
2.1.6 Link between weak and strong solutions 2.1.6 弱解与强解之间的联系
Theorem 2.1.4. If phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) is a strong solution of (2.3 2.4), then it is also a weak solution of (2.3 2.4). 定理 2.1.4。如果 phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 是 (2.3 2.4) 的强解,那么它也是 (2.3 2.4) 的弱解。
If phi inL_("loc ")^(1)(R^(+)xxR)\phi \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) is a weak solution of (2.3.2.4) and if moreover phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right), then phi\phi is a strong solution of (2.3 (2.4). 如果 phi inL_("loc ")^(1)(R^(+)xxR)\phi \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 是(2.3.2.4)的弱解,并且如果 phi inC^(1)(R^(+)xxR)\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) ,那么 phi\phi 是(2.3(2.4))的强解。
Proof. If phi inC^(1)\phi \in C^{1}, then we do again the computations of section 2.1.4 which gives (2.6), namely phi\phi is a weak solution. 证明。如果 phi inC^(1)\phi \in C^{1} ,那么我们再次进行 2.1.4 节的计算,得到(2.6),即 phi\phi 是一个弱解。
If phi\phi is a weak solution and if moreover phi inC^(1)\phi \in C^{1}, then we can integrate by parts (2.6), which gives 如果 phi\phi 是一个弱解,并且如果 phi inC^(1)\phi \in C^{1} ,那么我们可以通过分部积分(2.6)进行积分,这给出
int_(R^(+))int_(R)psi(t,x)[del_(t)phi(t,x)+cdel_(x)phi(t,x)]dxdt=0\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \psi(t, x)\left[\partial_{t} \phi(t, x)+c \partial_{x} \phi(t, x)\right] d x d t=0
We recall that if a function f(t,x)inC^(0)f(t, x) \in C^{0} satisfies 我们回忆起,如果一个函数 f(t,x)inC^(0)f(t, x) \in C^{0} 满足
int_(R^(+))int_(R)psi(t,x)f(t,x)dxdt=0\int_{\mathbb{R}^{+}} \int_{\mathbb{R}} \psi(t, x) f(t, x) d x d t=0
for any psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right), then ff is identically 0 (c.f. lectures on the Lebesgue integral). Thus 对于任何 psi inC^(1)(R^(+)xxR)\psi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) ,则 ff 恒等于 0(参见关于勒贝格积分的讲座)。因此
Weak solution and regularity rarr\rightarrow strong solution. 弱解和正则性 rarr\rightarrow 强解。
2.1.7 Existence and uniqueness of weak solutions 2.1.7 弱解的存在性和唯一性
Let us now prove the existence and uniqueness of a weak solution. 现在让我们证明弱解的存在性和唯一性。
Theorem 2.1.1. Let phi_(0)inL_("loc ")^(1)(R^(+)xxR)\phi_{0} \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right). Then there exists a unique weak solution phi inL_("loc ")^(1)(R^(+)xxR)\phi \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) to 定理 2.1.1. 设 phi_(0)inL_("loc ")^(1)(R^(+)xxR)\phi_{0} \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 。则存在唯一的弱解 phi inL_("loc ")^(1)(R^(+)xxR)\phi \in L_{\text {loc }}^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 使得
with initial data phi_(0)\phi_{0}. 带有初始数据 phi_(0)\phi_{0} 。
Proof. Let us begin with the existence part and let us check that phi(t,x)=\phi(t, x)=phi_(0)(x-ct)\phi_{0}(x-c t) is a weak solution. Let psi\psi be C^(1)(R^(+)xxR)C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) with compact support. We have 证明。我们先从存在性部分开始,检查 phi(t,x)=\phi(t, x)=phi_(0)(x-ct)\phi_{0}(x-c t) 是否是一个弱解。设 psi\psi 为具有紧支撑的 C^(1)(R^(+)xxR)C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 。我们有
int_(R^(+)xxR)phi(t,x)[del_(t)psi+cdel_(x)psi]dxdt=int_(R^(+)xxR)phi_(0)(x-ct)[del_(t)psi+cdel_(x)psi]dxdt\int_{\mathbb{R}^{+} \times \mathbb{R}} \phi(t, x)\left[\partial_{t} \psi+c \partial_{x} \psi\right] d x d t=\int_{\mathbb{R}^{+} \times \mathbb{R}} \phi_{0}(x-c t)\left[\partial_{t} \psi+c \partial_{x} \psi\right] d x d t
We make the change of variables y=x-cty=x-c t in the previous equation, which leads to 我们在前面的方程中进行变量变换 y=x-cty=x-c t ,这导致了
{:[int_(R^(+)xxR)phi_(0)(y)[del_(t)psi(t,y+ct)+cdel_(x)psi(t,y+ct)]dydt],[=int_(R^(+)xxR)phi_(0)(y)(d)/(dt)[psi(t","y+ct)]dydt=-int_(R)phi_(0)(y)psi(0","y)dy]:}\begin{aligned}
& \int_{\mathbb{R}^{+} \times \mathbb{R}} \phi_{0}(y)\left[\partial_{t} \psi(t, y+c t)+c \partial_{x} \psi(t, y+c t)\right] d y d t \\
& =\int_{\mathbb{R}^{+} \times \mathbb{R}} \phi_{0}(y) \frac{d}{d t}[\psi(t, y+c t)] d y d t=-\int_{\mathbb{R}} \phi_{0}(y) \psi(0, y) d y
\end{aligned}
which ends the proof of existence. 这结束了存在性的证明。
Let us turn to the uniqueness. Let phi_(1)\phi_{1} and phi_(2)\phi_{2} be two weak solutions. Then, by linearity, phi=phi_(1)-phi_(2)\phi=\phi_{1}-\phi_{2} is a weak solution with initial data 0 . 让我们转向唯一性。设 phi_(1)\phi_{1} 和 phi_(2)\phi_{2} 是两个弱解。那么,由于线性性质, phi=phi_(1)-phi_(2)\phi=\phi_{1}-\phi_{2} 是一个初始数据为 0 的弱解。
If we prove that 如果我们证明了
int_(R^(+)xxR^(2))phi(t,x)theta(t,x)dxdt=0\int_{\mathbb{R}^{+} \times \mathbb{R}^{2}} \phi(t, x) \theta(t, x) d x d t=0
for every function theta inC^(1)(R^(+)xxR)\theta \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) with compact support, then we obtain that phi=0\phi=0, which will prove the uniqueness of the solution. 对于每个具有紧支撑的函数 theta inC^(1)(R^(+)xxR)\theta \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) ,我们可以得到 phi=0\phi=0 ,这将证明解的唯一性。
Let theta\theta be given. It remains to construct psi inC^(1)\psi \in C^{1} with compact support such that 设 theta\theta 已知。接下来需要构造 psi inC^(1)\psi \in C^{1} 具有紧支撑,使得
f(t,x)=-int_(t)^(+oo)theta(s,x+cs)dsf(t, x)=-\int_{t}^{+\infty} \theta(s, x+c s) d s
We then have 我们接下来有
psi(t,x)=f(t,x-ct)\psi(t, x)=f(t, x-c t)
Then psi(t,x)\psi(t, x) is a solution of (2.7) with compact support. We have 那么 psi(t,x)\psi(t, x) 是具有紧支撑的 (2.7) 的解。我们有
int_(R^(+)xxR^(-))phi(t,x)theta(t,x)dxdt=int_(R^(+)xxR)phi(t,x)[del_(t)psi(t,x)+cdel_(x)psi(t,x)]dxdt=0\int_{\mathbb{R}^{+} \times \mathbb{R}^{-}} \phi(t, x) \theta(t, x) d x d t=\int_{\mathbb{R}^{+} \times \mathbb{R}} \phi(t, x)\left[\partial_{t} \psi(t, x)+c \partial_{x} \psi(t, x)\right] d x d t=0
since phi\phi is a weak solution with phi(0,x)=0\phi(0, x)=0, which ends the proof. 由于 phi\phi 是一个弱解,且满足 phi(0,x)=0\phi(0, x)=0 ,这就结束了证明。
2.1.8 Qualitative properties 2.1.8 质量属性
In PDEs, a “qualitative property” is a property of the equation, like its sign, estimates on its maximum, its behavior in long time, namely, any property of the solution which allows to describe it. 在偏微分方程中,“定性属性”是方程的一个属性,比如它的符号、对其最大值的估计、在长时间内的行为,即任何允许描述解的属性。
For instance, if the support of psi_(0)\psi_{0} is compact, then for every tt, the support of phi(t,x)\phi(t, x) is compact, as stated in the next proposition. 例如,如果 psi_(0)\psi_{0} 的支撑是紧致的,那么对于每个 tt , phi(t,x)\phi(t, x) 的支撑也是紧致的,如下一个命题所述。
Proposition 2.1.5. Let M > 0M>0 and let us assume that phi_(0)inC^(1)\phi_{0} \in C^{1} and vanishes outside [-M,M][-M, M]. Then phi(t,*)\phi(t, \cdot) vanishes outside [-M+ct,+M+ct][-M+c t,+M+c t]. 命题 2.1.5. 设 M > 0M>0 ,并假设 phi_(0)inC^(1)\phi_{0} \in C^{1} 在 [-M,M][-M, M] 之外消失。那么 phi(t,*)\phi(t, \cdot) 在 [-M+ct,+M+ct][-M+c t,+M+c t] 之外消失。
Proof. As phi(t,x)=phi_(0)(x-ct)\phi(t, x)=\phi_{0}(x-c t), we have phi(t,x)=0\phi(t, x)=0 if |x-ct| >= M|x-c t| \geq M, which gives the proposition. 证明。如 phi(t,x)=phi_(0)(x-ct)\phi(t, x)=\phi_{0}(x-c t) ,我们有 phi(t,x)=0\phi(t, x)=0 如果 |x-ct| >= M|x-c t| \geq M ,这给出了命题。
This proof however completely relies on the explicit solution of 2.3 2.4 and can not be generalized to other kinds of equations. It is thus interesting to find another proof, which does not rely on this explicit solution. The proof will be more “robust” and can be applied to many other equations where there is no explicit solution. 然而,这个证明完全依赖于 2.3 和 2.4 的显式解,无法推广到其他类型的方程。因此,寻找另一个不依赖于这个显式解的证明是很有趣的。这个证明将更加“稳健”,可以应用于许多没有显式解的其他方程。
Let us prove the following weaker result, without using the explicit formula: if for any t,phi(t,x)t, \phi(t, x) goes to 0 as |x|rarr+oo|x| \rightarrow+\infty, then phi(t,*)\phi(t, \cdot) is supported in [-M+ct,M+ct][-M+c t, M+c t]. 让我们证明以下较弱的结果,而不使用显式公式:如果对于任何 t,phi(t,x)t, \phi(t, x) ,当 |x|rarr+oo|x| \rightarrow+\infty 趋向于 0 时, phi(t,*)\phi(t, \cdot) 在 [-M+ct,M+ct][-M+c t, M+c t] 上是有支撑的。
Let us introduce 让我们介绍一下
I(t)=int_(x >= M+ct)|phi(t,x)|^(2)dxI(t)=\int_{x \geq M+c t}|\phi(t, x)|^{2} d x
Then I(t)I(t) is C^(1)C^{1} and 然后 I(t)I(t) 是 C^(1)C^{1} 和
Thus, I(t)I(t) is a constant. As I(0)=0I(0)=0 by assumption, I(t)=0I(t)=0 for every tt, which implies that phi(t,x)=0\phi(t, x)=0 for every t >= 0t \geq 0 and x >= M+ctx \geq M+c t. We then repeat the argument on x < -M+ctx<-M+c t. 因此, I(t)I(t) 是一个常数。根据假设, I(0)=0I(0)=0 对于每个 tt ,这意味着 phi(t,x)=0\phi(t, x)=0 对于每个 t >= 0t \geq 0 和 x >= M+ctx \geq M+c t 。然后我们在 x < -M+ctx<-M+c t 上重复这个论证。
This kind of method is called an “energy method”. We define an “energy” I(t)I(t) and try to get informations on it. Such techniques are of constant use in PDEs. 这种方法被称为“能量方法”。我们定义一个“能量” I(t)I(t) ,并试图获取关于它的信息。这种技术在偏微分方程中常常使用。
Similar ideas can be used to prove the uniqueness of compactly supported solutions. Namely, let phi_(1)\phi_{1} and phi_(2)\phi_{2} be two solutions of (2.3) 2.4 which go to 0 at inifinity for each time t >= 0t \geq 0. Then phi=phi_(1)-phi_(2)\phi=\phi_{1}-\phi_{2} goes to 0 as |x|rarr+oo|x| \rightarrow+\infty and satisy 类似的思想可以用来证明紧支撑解的唯一性。即,设 phi_(1)\phi_{1} 和 phi_(2)\phi_{2} 是(2.3) 2.4 的两个解,对于每个时间 t >= 0t \geq 0 ,它们在无穷远处趋向于 0。那么 phi=phi_(1)-phi_(2)\phi=\phi_{1}-\phi_{2} 在 |x|rarr+oo|x| \rightarrow+\infty 时也趋向于 0,并满足
with initial data phi(0,x)=0\phi(0, x)=0, for every x inRx \in \mathbb{R}. We have 初始数据为 phi(0,x)=0\phi(0, x)=0 ,对于每个 x inRx \in \mathbb{R} 。我们有 del_(t)int_(R)|phi(t,x)|^(2)dx=2int_(R)phidel_(t)phi dx=-2cint_(R)phidel_(x)phi dx=-cint_(R)del_(x)(phi^(2))dx=0\partial_{t} \int_{\mathbb{R}}|\phi(t, x)|^{2} d x=2 \int_{\mathbb{R}} \phi \partial_{t} \phi d x=-2 c \int_{\mathbb{R}} \phi \partial_{x} \phi d x=-c \int_{\mathbb{R}} \partial_{x}\left(\phi^{2}\right) d x=0,
thus phi(t,x)=0\phi(t, x)=0 for every t >= 0t \geq 0 and x inRx \in \mathbb{R}, which gives uniqueness. 因此,对于每个 t >= 0t \geq 0 和 x inRx \in \mathbb{R} ,这赋予了唯一性。
Let us now turn to another property: the transport equation satisfies the following “maximum principle” 现在让我们转向另一个性质:传输方程满足以下“极大值原理”
Proposition 2.1.6. Let phi\phi be a classical solution of 2.3.2.4). Then, if phi_(0)(x) <= M\phi_{0}(x) \leq M for every x inRx \in \mathbb{R}, 命题 2.1.6. 设 phi\phi 是 2.3.2.4) 的一个经典解。那么,如果对每个 x inRx \in \mathbb{R} 都有 phi_(0)(x) <= M\phi_{0}(x) \leq M 。
phi(t,x) <= M,quad AA t >= 0,quad AA x inR\phi(t, x) \leq M, \quad \forall t \geq 0, \quad \forall x \in \mathbb{R}
If phi_(0)(x) >= 0\phi_{0}(x) \geq 0 for every x inRx \in \mathbb{R}, then phi(t,x) >= 0\phi(t, x) \geq 0 for every t >= 0t \geq 0 and x inRx \in \mathbb{R}. 如果 phi_(0)(x) >= 0\phi_{0}(x) \geq 0 对于每个 x inRx \in \mathbb{R} ,那么 phi(t,x) >= 0\phi(t, x) \geq 0 对于每个 t >= 0t \geq 0 和 x inRx \in \mathbb{R} 。
Proof. The proof of this proposition is obvious, since, for any tt and xx, 证明。这个命题的证明显而易见,因为对于任何 tt 和 xx ,
phi(t,x)=phi_(0)(x-ct) <= M\phi(t, x)=\phi_{0}(x-c t) \leq M
This proof again completely relies on the explicit solution of 2.3 2.4 and can not be generalized to other kinds of equations. It is thus interesting to find another proof, which does not rely on this explicit solution. 这个证明再次完全依赖于 2.3 和 2.4 的显式解,无法推广到其他类型的方程。因此,寻找一个不依赖于这个显式解的其他证明是很有趣的。
Let f inC^(1)(R)f \in C^{1}(\mathbb{R}). Then f(phi)f(\phi) is C^(1)C^{1} and 让 f inC^(1)(R)f \in C^{1}(\mathbb{R}) 。那么 f(phi)f(\phi) 是 C^(1)C^{1} 和
Assume now that f(y)=0f(y)=0 if y <= My \leq M. Then we have psi(0,x)=0\psi(0, x)=0 for every xx. Using the uniqueness of the solution of the equation, this gives psi(t,x)=0\psi(t, x)=0 for every tt and xx. Let us take 假设现在如果 y <= My \leq M ,则 f(y)=0f(y)=0 。那么我们对每个 xx 都有 psi(0,x)=0\psi(0, x)=0 。利用方程解的唯一性,这为每个 tt 和 xx 提供了 psi(t,x)=0\psi(t, x)=0 。让我们取
f(y)=(s u p(0,y-M))^(2)f(y)=(\sup (0, y-M))^{2}
Then as f(phi(t,x))=0f(\phi(t, x))=0 for every tt and xx, this means that phi(t,x) <= M\phi(t, x) \leq M everywhere which ends the proof. 那么作为 f(phi(t,x))=0f(\phi(t, x))=0 对于每个 tt 和 xx ,这意味着 phi(t,x) <= M\phi(t, x) \leq M 到处都是,这结束了证明。
The proof that phi_(0)(x) >= 0\phi_{0}(x) \geq 0 for every xx implies phi(t,x) >= 0\phi(t, x) \geq 0 for ever tt and xx is similar. 证明 phi_(0)(x) >= 0\phi_{0}(x) \geq 0 对于每个 xx 意味着 phi(t,x) >= 0\phi(t, x) \geq 0 对于每个 tt 和 xx 是类似的。
Remark. The idea to introduce nonlinear functions of the solution, like f(u)f(u) is called “renomarlization”. It is of wide use in many different kinds of PDEs. 备注。引入解的非线性函数的想法,如 f(u)f(u) ,称为“重标定”。它在许多不同类型的偏微分方程中被广泛使用。
2.1.9 Numerical schemes 2.1.9 数值方案
A central question in applications is to compute numerically solutions of PDEs, since most of them have no explicit solutions. It is of course not possible to evaluate a solution everywhere, namely for every time and position, since computers can only handle a finite number of informations. The starting point is thus to construct a “grid” and to “discretize” the space and time. 在应用中,一个核心问题是数值计算偏微分方程的解,因为大多数方程没有显式解。当然,无法在每个时间和位置评估解,因为计算机只能处理有限数量的信息。因此,起点是构建一个“网格”,并对空间和时间进行“离散化”。
Let h > 0h>0 and kk be (small) real numbers. We define the points x_(j)x_{j} by 让 h > 0h>0 和 kk 是(小)实数。我们通过以下方式定义点 x_(j)x_{j} :
t_(n)=nk,quad n inNt_{n}=n k, \quad n \in \mathbb{N}
The couples (t_(n),x_(j))_(n inN,j inZ)\left(t_{n}, x_{j}\right)_{n \in \mathbb{N}, j \in \mathbb{Z}} form a “grid” on which we want to evaluate the solution phi\phi to (2.3 2.4). We will denote by phi_(n)^(j)\phi_{n}^{j} a numerical approximation of phi(t_(n),x_(j))\phi\left(t_{n}, x_{j}\right). 这些夫妇 (t_(n),x_(j))_(n inN,j inZ)\left(t_{n}, x_{j}\right)_{n \in \mathbb{N}, j \in \mathbb{Z}} 形成了一个“网格”,我们希望在其上评估解决方案 phi\phi 对(2.3 2.4)。我们将用 phi_(n)^(j)\phi_{n}^{j} 表示 phi(t_(n),x_(j))\phi\left(t_{n}, x_{j}\right) 的数值近似。
The central question of “numerical analysis” is to find algorithms to evaluate phi_(n)^(j)\phi_{n}^{j} which are precise and fast. Of course phi_(n)^(j)\phi_{n}^{j} will in general not equal phi(t_(n),x_(j))\phi\left(t_{n}, x_{j}\right) but we want the error “数值分析”的核心问题是寻找精确且快速的算法来评估 phi_(n)^(j)\phi_{n}^{j} 。当然, phi_(n)^(j)\phi_{n}^{j} 通常不等于 phi(t_(n),x_(j))\phi\left(t_{n}, x_{j}\right) ,但我们希望误差
to be as small and possible, and we want to compute as fast as possible. 尽可能小,并且我们希望计算尽可能快。
Let us now detail a simple “numerical scheme”. We approximate del_(t)phi\partial_{t} \phi and del_(x)phi\partial_{x} \phi by 现在让我们详细介绍一个简单的“数值方案”。我们通过以下方式近似 del_(t)phi\partial_{t} \phi 和 del_(x)phi\partial_{x} \phi :
Starting with (phi_(0)^(j))_(j inZ)\left(\phi_{0}^{j}\right)_{j \in \mathbb{Z}}, we compute (phi_(1)^(j))_(j inZ)\left(\phi_{1}^{j}\right)_{j \in \mathbb{Z}} and iterate the process. 从 (phi_(0)^(j))_(j inZ)\left(\phi_{0}^{j}\right)_{j \in \mathbb{Z}} 开始,我们计算 (phi_(1)^(j))_(j inZ)\left(\phi_{1}^{j}\right)_{j \in \mathbb{Z}} 并迭代这个过程。
Consistence of the numerical scheme 数值方案的一致性
The error of consistence is the error made when we replace phi_(n)^(j)\phi_{n}^{j} by phi(t_(n),x_(j))\phi\left(t_{n}, x_{j}\right), namely, 一致性错误是指当我们用 phi(t_(n),x_(j))\phi\left(t_{n}, x_{j}\right) 替换 phi_(n)^(j)\phi_{n}^{j} 时所犯的错误,即,
Using Taylor’s formula, assuming that the derivative of phi_(0)\phi_{0} is uniformly bounded (which implies that the derivatives of phi\phi are uniformly bounded), we have 使用泰勒公式,假设 phi_(0)\phi_{0} 的导数是均匀有界的(这意味着 phi\phi 的导数是均匀有界的),我们有
Thus delta_(n,j)\delta_{n, j} is small, like k(h+k)k(h+k) provided kk and hh are small. The numerical scheme is thus said “consistent”. 因此 delta_(n,j)\delta_{n, j} 很小,就像 k(h+k)k(h+k) ,前提是 kk 和 hh 也很小。因此,这个数值方案被称为“一致的”。
Stability 稳定性
A numerical scheme is said “stable” if we can control the growth of norms. 一个数值方案被称为“稳定”,如果我们可以控制范数的增长。
Proposition 2.1.7. Assume that c > 0c>0 and that ck//h < 1c k / h<1, then we have 命题 2.1.7. 假设 c > 0c>0 和 ck//h < 1c k / h<1 ,则我们有
s u p_(l inZ)|phi_(n+1)^(l)| <= s u p_(l inZ)|phi_(n)^(l)|\sup _{l \in \mathbb{Z}}\left|\phi_{n+1}^{l}\right| \leq \sup _{l \in \mathbb{Z}}\left|\phi_{n}^{l}\right|
Proof. We have for any n >= 0n \geq 0 and j inZj \in \mathbb{Z}, 证明。对于任何 n >= 0n \geq 0 和 j inZj \in \mathbb{Z} ,
We note that 1-ck//h1-c k / h and ck//hc k / h positive and less than 1 provided c > 0c>0 and ck//h < 1c k / h<1. Thus 我们注意到 1-ck//h1-c k / h 和 ck//hc k / h 是正数且小于 1,前提是 c > 0c>0 和 ck//h < 1c k / h<1 。因此
If c < 0c<0, the maximum of phi_(n)^(j)\phi_{n}^{j} may increase at each step and develop “oscillations”. If c < 0c<0, we know that the solution travels “to the left”, thus the evolution of a value at x_(j)x_{j} depends on the value at x_(j+1)x_{j+1}. A numerical scheme which does not involve x_(j+1)x_{j+1} will fail. 如果 c < 0c<0 , phi_(n)^(j)\phi_{n}^{j} 的最大值可能在每一步增加并产生“振荡”。如果 c < 0c<0 ,我们知道解是“向左”移动的,因此在 x_(j)x_{j} 处的值的演变依赖于在 x_(j+1)x_{j+1} 处的值。一个不涉及 x_(j+1)x_{j+1} 的数值方案将会失败。
The condition ck//h < 1c k / h<1 is called a Courant-Friedrichs-Lewy condition (“CFL” condition). It occurs in many different numerical schemes for many different equations. It says that during a time step kk, the solution has travelled over a distance kck c which is smaller than the step of the grid hh. 条件 ck//h < 1c k / h<1 被称为 Courant-Friedrichs-Lewy 条件(“CFL” 条件)。它出现在许多不同方程的许多不同数值方案中。它表示在一个时间步 kk 内,解所经过的距离 kck c 小于网格步长 hh 。
The support of the numerical solution phi_(n)^(j)\phi_{n}^{j} increases by 1 on the right at each time step, namely with a velocity h//kh / k. The velocity of the evolution of the support of phi\phi is cc. We must have c < h//kc<h / k, which is the CFL condition. 数值解的支撑在每个时间步长右侧增加 1,即以速度 h//kh / k 。 phi\phi 的支撑演化速度为 cc 。我们必须满足 c < h//kc<h / k ,这就是 CFL 条件。
2.1.10 Convergence of the numerical scheme 2.1.10 数值方案的收敛性
Proposition 2.1.8. Let phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) with compact support. Let h_(m)h_{m} and k_(m)k_{m} be a sequence of positive real numbers, such that ck_(m)//h_(m) < 1c k_{m} / h_{m}<1. Let phi_(n,m)^(j)\phi_{n, m}^{j} be defined by 2.8 2.9) and let phi\phi be the solution of 2.3 2.4). Then, for any T > 0T>0, there exists a constant CC such that 命题 2.1.8. 设 phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) 具有紧支撑。设 h_(m)h_{m} 和 k_(m)k_{m} 为一列正实数,满足 ck_(m)//h_(m) < 1c k_{m} / h_{m}<1 。设 phi_(n,m)^(j)\phi_{n, m}^{j} 由 2.8 2.9) 定义,且 phi\phi 为 2.3 2.4) 的解。那么,对于任何 T > 0T>0 ,存在一个常数 CC 使得
s u p_(m rarr+oo)s u p_(nk_(m) <= T,j inZ)|phi_(n,m)^(j)-phi(t_(n),x_(j))| <= Ch\sup _{m \rightarrow+\infty} \sup _{n k_{m} \leq T, j \in \mathbb{Z}}\left|\phi_{n, m}^{j}-\phi\left(t_{n}, x_{j}\right)\right| \leq C h
In other words, the numerical solution converges to the true one with a speed O(h)O(h). 换句话说,数值解以速度 O(h)O(h) 收敛到真实解。
Proof. Let us fix some m > 0m>0 and forget the index mm in the notations. We note that 证明。我们固定一些 m > 0m>0 ,并在符号中忽略索引 mm 。我们注意到
s u p_(j inZ)|epsi_(n+1)^(j)| <= s u p_(j inZ)|epsi_(n)^(j)|+C_(0)k(h+k)\sup _{j \in \mathbb{Z}}\left|\varepsilon_{n+1}^{j}\right| \leq \sup _{j \in \mathbb{Z}}\left|\varepsilon_{n}^{j}\right|+C_{0} k(h+k)
This leads to 这导致了
s u p_(j inZ)|epsi_(n)^(j)| <= C_(0)nk(h+k) <= C_(0)T(h+k) <= Ch\sup _{j \in \mathbb{Z}}\left|\varepsilon_{n}^{j}\right| \leq C_{0} n k(h+k) \leq C_{0} T(h+k) \leq C h
where we have used nk <= Tn k \leq T and the CFL condition. 我们使用了 nk <= Tn k \leq T 和 CFL 条件。
Remarks. The previous proof uses only two ingredients: the consistence of the scheme (namely estimates on delta_(n)^(j)\delta_{n}^{j} and the stability of the scheme (estimates on the supremum). The consistence and the stability leads to the convergence of the scheme, which occurs for many different numerical schemes and many different equations. We sum up the proof by saying 备注。之前的证明只使用了两个要素:方案的一致性(即对 delta_(n)^(j)\delta_{n}^{j} 的估计)和方案的稳定性(对上确界的估计)。一致性和稳定性导致了方案的收敛,这在许多不同的数值方案和许多不同的方程中都发生。我们总结证明如下:
In applications, we usually are not in the whole space, but only in a bounded domain. Let a < ba<b be two real numbers and let us study the transport equation on [a,b][a, b]. We have to describe what happens at aa and bb. If c > 0c>0, the flow goes from the left to the right, it “enters” the domain at aa and “leaves” the domain at bb. We thus enforce a boundary condition at x=ax=a, and nothing at x=bx=b. This leads to 在应用中,我们通常不在整个空间中,而只是处于一个有界域中。设 a < ba<b 为两个实数,我们来研究 [a,b][a, b] 上的输运方程。我们必须描述在 aa 和 bb 发生的情况。如果 c > 0c>0 ,流动从左向右,它在 aa “进入”该域,在 bb “离开”该域。因此,我们在 x=ax=a 施加边界条件,而在 x=bx=b 则不施加任何条件。这导致了
{:[del_(t)phi+cdel_(x)phi=0","quad AA t >= 0","AA x in[a","b]],[phi(0","x)=phi_(0)(x)","quad AA x in[a","b]],[phi(t","a)=phi_(1)(t)","quad AA t >= 0]:}\begin{gathered}
\partial_{t} \phi+c \partial_{x} \phi=0, \quad \forall t \geq 0, \forall x \in[a, b] \\
\phi(0, x)=\phi_{0}(x), \quad \forall x \in[a, b] \\
\phi(t, a)=\phi_{1}(t), \quad \forall t \geq 0
\end{gathered}
where phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) and phi_(1)inC^(1)(R^(+))\phi_{1} \in C^{1}\left(\mathbb{R}^{+}\right)are two given functions. 其中 phi_(0)inC^(1)(R)\phi_{0} \in C^{1}(\mathbb{R}) 和 phi_(1)inC^(1)(R^(+))\phi_{1} \in C^{1}\left(\mathbb{R}^{+}\right) 是两个给定的函数。
Theorem 2.1.2. Let us assume that 定理 2.1.2. 让我们假设
thus psi(t,x)\psi(t, x) is independent on tt. 因此 psi(t,x)\psi(t, x) 独立于 tt 。
Let t >= 0t \geq 0 and x in[a,b]x \in[a, b]. Two cases arise. If x-ct > ax-c t>a, then we have 让 t >= 0t \geq 0 和 x in[a,b]x \in[a, b] 。出现两种情况。如果 x-ct > ax-c t>a ,那么我们有
On the contrary if x-ct < ax-c t<a, then we have, for any tau\tau such that 0 <= tau <= t0 \leq \tau \leq t, 相反,如果 x-ct < ax-c t<a ,那么对于任何 tau\tau ,只要 0 <= tau <= t0 \leq \tau \leq t ,我们有:
Thus, if the solution phi\phi exists, it is given by 2.16 . 因此,如果解 phi\phi 存在,则由 2.16 给出。
It remains to check that phi\phi is a solution. In order phi\phi to be C^(0)C^{0} at (t,x)=(t, x)=(0,a)(0, a), we must have 仍需检查 phi\phi 是否是一个解。为了 phi\phi 在 C^(0)C^{0}(t,x)=(t, x)=(0,a)(0, a) ,我们必须有
phi_(0)(a)=phi_(1)(0)\phi_{0}(a)=\phi_{1}(0)
In order phi\phi to be C^(1)C^{1} at (t,x)=(0,a)(t, x)=(0, a), we must have 为了 phi\phi 在 C^(1)C^{1} 时 (t,x)=(0,a)(t, x)=(0, a) ,我们必须拥有
del_(t)phi(0,a)=-cdel_(x)phi(0,a)\partial_{t} \phi(0, a)=-c \partial_{x} \phi(0, a)
The two relations (2.18) and (2.19) are called “compatibility conditions”. 这两个关系(2.18)和(2.19)被称为“兼容性条件”。
Under these two compatibility conditions, it can be checked that phi\phi satisfies (2.12) when x > a+ctx>a+c t, when x < a+ctx<a+c t and also when x=a+ctx=a+c t. 在这两个兼容性条件下,可以检查到 phi\phi 在 x > a+ctx>a+c t 、 x < a+ctx<a+c t 和 x=a+ctx=a+c t 时满足 (2.12)。
2.2 Transport equation in R^(n)\mathbb{R}^{n} 2.2 在 R^(n)\mathbb{R}^{n} 中的传输方程
2.2.1 Introduction 2.2.1 引言
Let n >= 1n \geq 1 be an integer. Let V(t,x)V(t, x) be a vector field on R^(n)\mathbb{R}^{n}, namely 设 n >= 1n \geq 1 为一个整数。设 V(t,x)V(t, x) 为在 R^(n)\mathbb{R}^{n} 上的一个向量场,即
In this section, we study the transport of quantities by the vector field V(t,x)V(t, x). 在本节中,我们研究由向量场 V(t,x)V(t, x) 运输的量。
Definition 2.2.1. Characteristics of the vector field V(t,x)V(t, x). 定义 2.2.1. 向量场 V(t,x)V(t, x) 的特征。
The solution X(t;s,x)X(t ; s, x) to the equations 方程的解 X(t;s,x)X(t ; s, x)
when it exists, is called the characteristics curve, or integral curves, or characteristics of the vector field V(t,x)V(t, x). Moreover, X(t;s,x)X(t ; s, x) is continuous in all its variables. 当它存在时,称为特征曲线,或积分曲线,或向量场 V(t,x)V(t, x) 的特征。此外, X(t;s,x)X(t ; s, x) 在其所有变量中是连续的。
Remark. The characteristics can be interpreted as follow. Imagine that a particle is at time ss at the position x inR^(n)x \in \mathbb{R}^{n} and that it evolves according to the velocity field V(t,x)V(t, x), namely that its speed at any time is equal to the value of VV at it current position. Then its position at time ss is X(t;s,x)X(t ; s, x). 备注。特征可以解释如下。想象一个粒子在时间 ss 处于位置 x inR^(n)x \in \mathbb{R}^{n} ,并且它根据速度场 V(t,x)V(t, x) 演变,也就是说它在任何时刻的速度等于其当前位置的 VV 值。那么它在时间 ss 的位置信息是 X(t;s,x)X(t ; s, x) 。 X(t;s,x)X(t ; s, x) is the position at time tt of the particle which at time ss is at xx. X(t;s,x)X(t ; s, x) 是在时间 tt 时粒子的位置,而在时间 ss 时粒子位于 xx 。
Let us now detail the properties of X(t;s,x)X(t ; s, x). 现在让我们详细说明 X(t;s,x)X(t ; s, x) 的属性。
Lemma 2.2.2. For any s,t,taus, t, \tau and xx, we have 引理 2.2.2。对于任何 s,t,taus, t, \tau 和 xx ,我们有
Proof. Note that (2.23) is a consequence of (2.22) by taking tau=t\tau=t. Now both terms of (2.22) are the position at time tau\tau of the particle which at time tt was at xx. 证明。注意到 (2.23) 是通过取 tau=t\tau=t 得到 (2.22) 的结果。现在 (2.22) 的两个项是粒子在时间 tau\tau 的位置,而该粒子在时间 tt 时位于 xx 。
We say that a scalar phi(t,x)\phi(t, x) is "transported by the velocity field V(t,x)V(t, x) ", if its values are constant along a characteristics, namely if, for any s,t,xs, t, x, 我们说标量 phi(t,x)\phi(t, x) "由速度场 V(t,x)V(t, x) 运输",如果它的值沿特征线是恒定的,即对于任何 s,t,xs, t, x ,
If phi(t,x)\phi(t, x) is C^(1)C^{1} in tt and xx, then, differentiating 2.24, we obtain 如果 phi(t,x)\phi(t, x) 在 C^(1)C^{1} 和 tt 中是 xx ,那么,对 2.24 进行求导,我们得到
This equation describes the transport of a quantity phi\phi by the vector field VV. 这个方程描述了量 phi\phi 在向量场 VV 中的传输。
2.2.2 Resolution of the transport equation 2.2.2 运输方程的求解
In this paragraph, we focus on the resolution of the transport equation 在这一段中,我们专注于运输方程的解
{:[del_(t)phi+V*grad_(x)phi=0],[phi(0","x)=phi_(0)(x)","quad AA x inR^(d)]:}\begin{gathered}
\partial_{t} \phi+V \cdot \nabla_{x} \phi=0 \\
\phi(0, x)=\phi_{0}(x), \quad \forall x \in \mathbb{R}^{d}
\end{gathered}
where phi_(0)(x)\phi_{0}(x) is a given C^(1)(R^(d))C^{1}\left(\mathbb{R}^{d}\right) function. 其中 phi_(0)(x)\phi_{0}(x) 是给定的 C^(1)(R^(d))C^{1}\left(\mathbb{R}^{d}\right) 函数。
Proposition 2.2.3. Existence and uniqueness of C^(1)C^{1} solutions. Let phi_(0)inC^(1)(R^(d))\phi_{0} \in C^{1}\left(\mathbb{R}^{d}\right). Let us assume that V inC^(0)([0,T],C^(1)(R^(d)))V \in C^{0}\left([0, T], C^{1}\left(\mathbb{R}^{d}\right)\right) and that V inL^(oo)([0,T]xxR^(n))V \in L^{\infty}\left([0, T] \times \mathbb{R}^{n}\right). Then the characteristics X(t;s,x)X(t ; s, x) are well defined, unique, and X inC^(1)([0,T]xx[0,T]xxR^(d))X \in C^{1}\left([0, T] \times[0, T] \times \mathbb{R}^{d}\right). Moreover, there exists a unique solution phi\phi to 2.27), 2.28) in C^(1)([0,T]xxR^(d))C^{1}\left([0, T] \times \mathbb{R}^{d}\right). This solution is given by 命题 2.2.3. C^(1)C^{1} 解的存在性和唯一性。设 phi_(0)inC^(1)(R^(d))\phi_{0} \in C^{1}\left(\mathbb{R}^{d}\right) 。我们假设 V inC^(0)([0,T],C^(1)(R^(d)))V \in C^{0}\left([0, T], C^{1}\left(\mathbb{R}^{d}\right)\right) 和 V inL^(oo)([0,T]xxR^(n))V \in L^{\infty}\left([0, T] \times \mathbb{R}^{n}\right) 。那么特征 X(t;s,x)X(t ; s, x) 是良好定义的、唯一的,并且 X inC^(1)([0,T]xx[0,T]xxR^(d))X \in C^{1}\left([0, T] \times[0, T] \times \mathbb{R}^{d}\right) 。此外,存在唯一解 phi\phi 对于 2.27)、2.28) 在 C^(1)([0,T]xxR^(d))C^{1}\left([0, T] \times \mathbb{R}^{d}\right) 中。该解由以下给出:
Proof. Let tt and xx be fixed, then 2.20 is an ordinary differential equation in tt, with an initial condition (2.21) prescribed at time ss. As VV is C^(1)C^{1} in xx and C^(0)C^{0} in tt, we can apply the Cauchy-Lifschitz theorem with parameters, which provides the existence, uniqueness and smoothness of the characteristics. 证明。设 tt 和 xx 为固定值,则 2.20 是 tt 中的一个常微分方程,初始条件在时间 ss 处规定为(2.21)。由于 VV 在 C^(1)C^{1} 中, C^(0)C^{0} 在 tt 中,我们可以应用带参数的 Cauchy-Lifschitz 定理,这提供了特征的存在性、唯一性和光滑性。
We note that phi\phi, defined by (2.29) is a solution, since, using (2.22), we have 我们注意到由(2.29)定义的 phi\phi 是一个解,因为使用(2.22),我们有
Moreover phi(0,x)=phi_(0)(x)\phi(0, x)=\phi_{0}(x). This solution is C^(1)C^{1} since XX is C^(1)C^{1}. 此外 phi(0,x)=phi_(0)(x)\phi(0, x)=\phi_{0}(x) 。这个解决方案是 C^(1)C^{1} ,因为 XX 是 C^(1)C^{1} 。
Let us turn to the uniqueness. If psi\psi is a solution then psi(t,X(t;0,x))\psi(t, X(t ; 0, x)) is constant in tt and is equal to psi(0,X(0;0,x))=phi_(0)(x)\psi(0, X(0 ; 0, x))=\phi_{0}(x), thus using (2.22), psi=phi\psi=\phi. 让我们转向独特性。如果 psi\psi 是一个解,那么 psi(t,X(t;0,x))\psi(t, X(t ; 0, x)) 在 tt 中是常数,并且等于 psi(0,X(0;0,x))=phi_(0)(x)\psi(0, X(0 ; 0, x))=\phi_{0}(x) ,因此使用(2.22), psi=phi\psi=\phi 。
2.3 Exercises 2.3 练习
Exercise 2.3.1. Check your knowledge of the Cauchy-Lifschitz theorem with parameters, and read again its proof. 练习 2.3.1。检查你对带参数的柯西-利夫希茨定理的知识,并再次阅读其证明。
Exercise 2.3.2. Let phi_(0)(x)inC^(1)(R)\phi_{0}(x) \in C^{1}(\mathbb{R}) be a given bounded function. 练习 2.3.2. 设 phi_(0)(x)inC^(1)(R)\phi_{0}(x) \in C^{1}(\mathbb{R}) 为一个给定的有界函数。
with initial data phi_(0)\phi_{0}. 带有初始数据 phi_(0)\phi_{0} 。
Hint: divide the equation by phi\phi, and study log phi-2t\log \phi-2 t. 提示:将方程除以 phi\phi ,并研究 log phi-2t\log \phi-2 t 。
Exrtise 2.3.3. Compute the solution to the following equations 练习 2.3.3。计算以下方程的解。
{:[del_(t)phi+del_(x)phi=0","," with ",phi_(0)(x)=sin x],[del_(t)phi+xdel_(x)phi=0","," with ",phi_(0)(x)=sin x","],[del_(t)phi+cos xdel_(x)phi=0","," with ",phi_(0)(x)=e^(x)],[del_(t)phi+psi(x)del_(x)phi=0","," with ",phi_(0)(x)=e^(x).]:}\begin{array}{crr}
\partial_{t} \phi+\partial_{x} \phi=0, & \text { with } & \phi_{0}(x)=\sin x \\
\partial_{t} \phi+x \partial_{x} \phi=0, & \text { with } & \phi_{0}(x)=\sin x, \\
\partial_{t} \phi+\cos x \partial_{x} \phi=0, & \text { with } & \phi_{0}(x)=e^{x} \\
\partial_{t} \phi+\psi(x) \partial_{x} \phi=0, & \text { with } & \phi_{0}(x)=e^{x} .
\end{array}
where psi(x)\psi(x) is a given function. 其中 psi(x)\psi(x) 是一个给定的函数。
Exercise 2.3.4. Compute the solution to the following equations 练习 2.3.4。计算以下方程的解。
{:[del_(t)phi+xdel_(x)phi+ydel_(y)phi=phi","quad" with "quadphi_(0)(x)=sin x cos y],[del_(t)phi-ydel_(x)phi+xdel_(y)phi=0]:}\begin{gathered}
\partial_{t} \phi+x \partial_{x} \phi+y \partial_{y} \phi=\phi, \quad \text { with } \quad \phi_{0}(x)=\sin x \cos y \\
\partial_{t} \phi-y \partial_{x} \phi+x \partial_{y} \phi=0
\end{gathered}
where psi(x)\psi(x) is a given function. 其中 psi(x)\psi(x) 是一个给定的函数。
Exercise 2.3.5. Let V(t,x)inC^(1)(R^(+)xxR^(n))V(t, x) \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}^{n}\right) be a bounded function, whose gradient is also bounded (uniformly in tt and xx ). Let phi_(0)(x)inC^(1)(R^(n))\phi_{0}(x) \in C^{1}\left(\mathbb{R}^{n}\right). Prove that there exists a unique solution phi inC^(1)(R^(+)xxR^(n))\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}^{n}\right) to 练习 2.3.5. 设 V(t,x)inC^(1)(R^(+)xxR^(n))V(t, x) \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}^{n}\right) 是一个有界函数,其梯度在 tt 和 xx 中也是有界的。设 phi_(0)(x)inC^(1)(R^(n))\phi_{0}(x) \in C^{1}\left(\mathbb{R}^{n}\right) 。证明存在唯一解 phi inC^(1)(R^(+)xxR^(n))\phi \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}^{n}\right) 。
with initial data phi(0,x)=phi_(0)(x)\phi(0, x)=\phi_{0}(x) (for every {:x inR^(n))\left.x \in \mathbb{R}^{n}\right). 使用初始数据 phi(0,x)=phi_(0)(x)\phi(0, x)=\phi_{0}(x) (对于每个 {:x inR^(n))\left.x \in \mathbb{R}^{n}\right) 。
Exercise 2.3.6. Using Python or another language, code the numerical scheme. Observe what happens if c < 0c<0 or if the CFL condition is not satisfied. By taking smaller and smaller kk and hh, check the speed of convergence. 练习 2.3.6。使用 Python 或其他语言,编码数值方案。观察如果 c < 0c<0 或 CFL 条件不满足时会发生什么。通过逐渐减小 kk 和 hh ,检查收敛速度。
Exercise 2.3.7. We consider the numerical scheme 练习 2.3.7。我们考虑数值方案
What is its error of consistence ? Is it stable? What is the corresponding CLF? Prove a result similar to Theorem 2.1.8. 它的一致性误差是多少?它稳定吗?相应的 CLF 是什么?证明一个类似于定理 2.1.8 的结果。
Exercise 2.3.8. Let u_(0)inC^(1)(R)u_{0} \in C^{1}(\mathbb{R}) and let f inC^(1)(R^(+)xxR)f \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right). Prove that there exists a unique classical solution to 练习 2.3.8。设 u_(0)inC^(1)(R)u_{0} \in C^{1}(\mathbb{R}) 和 f inC^(1)(R^(+)xxR)f \in C^{1}\left(\mathbb{R}^{+} \times \mathbb{R}\right) 。证明存在唯一的经典解。