Loading [MathJax]/jax/element/mml/optable/BasicLatin.js

2024年1月23日 星期二

Solution to Partial Differential Equations: Methods and Applications (Robert McOwen) Section 2.3

Solution to Partial Differential Equations: Methods and Applications (Robert McOwen) Section 2.3

    1. For complex-valued functions u and v on Ω, let u,v=Ωuˉvdx, where ˉv denotes the complex conjugate of v. If the coefficients aα(x) in (32a) are complex-valued functions, define the adjoint L so that Lu,v=u,Lv for all uCm(Ω), vCm0(Ω).
    2. For (complex) vector-valued functions u(x)=(u1(x),,uN(x)) on Ω, let u,v=Ω(u1ˉv1++uNˉvN)dx. If the coefficients aα(x) in (32a) are N×N matrix-valued functions, then (32a) defines a system of mth-order operators. Define the adjoint L so that Lu,v=u,Lv for all uCm(Ω,RN), vCm0(Ω,RN).
  1. Solution
    1. From (32a), we find that

      Lu,v=Ω(Lu)(ˉv)dx=Ω(|α|maα(x)Dαu)ˉvdx=|α|mΩ(Dαu)(aα(x)ˉv)dx=|α|m(1)|α|Ω(u)(Dα(aα(x)ˉv))dx=Ω(u)[|α|m(1)|α|Dα(aα(x)ˉv)]dx:=u,Lv,

      where the adjoint operator L is given by

      Lv:=¯|α|m(1)|α|Dα(aα(x)ˉv)=|α|m(1)|α|Dα(¯aα(x)v)for vCm0(Ω).

    2. Suppose aα(x)=[aijα(x)]1i,jN is a N×N matrix-valued function. Then we have

      Lu,v=Ω[(Nj=1|α|ma1jα(x)Dαuj)ˉv1++(Nj=1|α|maNjα(x)Dαuj)ˉvN]dx=Ω(Nj=1|α|ma1jα(x)Dαuj)ˉv1dx++Ω(Nj=1|α|maNjα(x)Dαuj)ˉvNdx=Nj=1|α|m[Ω(Dαuj)(a1jα(x)ˉv1)dx++Ω(Dαuj)(aNjα(x)ˉvN)dx]=Nj=1|α|m(1)|α|[Ω(uj)Dα(a1jα(x)ˉv1)dx++Ω(uj)Dα(aNjα(x)ˉvN)dx]=Nj=1Ω(uj)|α|m(1)|α|Dα(a1jα(x)ˉv1++aNjα(x)ˉvN)dx=u,Lv,

      where the adjoint operator L is given by

      Lv=¯|α|m(1)|α|Dα(aTα(x)ˉv)=|α|m(1)|α|Dα(¯aTα(x)v)for vCm0(Ω,RN).

      Here AT denotes the tranpose of the matrix A.

  2. Let Lu=uμη as in Example 1. Derive the transmission condition (44) along the characteristic μ=0.
  3. SolutionSuppose u is a weak solution of Lu=0, i.e., Ωuμηvdμdη=Ω(Lu)vdμdη=0 for vC20(Ω). Recall that the operator L is self-adjoint, i.e., L=L. From (40), we have

    0=Ωu(Lv)dμdη=Ω+u+(Lv)dμdη+Ωu(Lv)dμdη=Ω+u+vμηdμdη+Ωuvμηdμdη=0u+vμηdηdμ+0uvμηdηdμ=0u+ηvμdηdμ0uηvμdηdμ=0u+ηvμdμdη0uηvμdμdη=[u+ηv|μ=μ=00u+μηvdμ]dη[uηv|μ=0μ=0uμηvdμ]dη=u+η(0,η)v(0,η)dηuη(0,η)v(0,η)dη=[u+η(0,η)uη(0,η)]v(0,η)dη.

    Thus, the transmission condition is

    [u+η(0,η)uη(0,η)]v(0,η)dη=0for vC20(Ω).


  4. Consider the first-order equation ut+cux=0.
    1. If fC(R), show that u(x,t)=f(xct) is a weak solution.
    2. Can you find any discontinuous weak solutions?
    3. Is there a transmission condition for a weak solution with jump discontinuity along the characteristic x=ct?
  5. Solution
    1. To show u(x,t)=f(xct) is a weak solution, it suffices to verify that R2f(xct)(Lv)(x,t)dxdt=0 for vC10(R2), where L=tcx is the adjoint operator L=t+cx. Let s=xct, t=t and ˜v(s,t)=v(s+ct,t) for s,tR2. Clearly, ˜vC10(R2) and satisfies ˜vt(s,t)=vt(s+ct,t)+cvx(s+ct,t). Thus, the improper double integral becomes

      R2f(xct)(Lv)(x,t)dxdt=f(xct)(vt(x,t)+cvx(x,t))dxdt=f(s)˜vt(s,t)dsdt=f(s)˜vt(s,t)dtds=f(s)0ds=0for vC10(R2).

      Here the Jacobian is 1. Therefore, u(x,t)=f(xct) is a weak solution for ut+cux=0.
    2. When f is discontinuous but Lebesgue integrable, u(x,t)=f(xct) is still a weak solution because the computation in part (a) also holds true.
    3. Suppose u is a weak solution with jump discontinuity along the characteristic x=ct. Let ˜u(s,t)=u(s+ct,t). Then ˜u is a weak solution of ˜L˜u:=˜ut(s,t)=0 with jump discontinuity along the line s=0, where ˜L=t. Clearly, the adjoint operator ˜L of ˜L is t. Let R2±={(s,t)R2:±s>0,tR} and ˜u±=˜u|R2±. Then for any test function vC10(R2), we have

      0=R2˜u(s,t)˜Lv(s,t)dsdt=R2˜u(s,t)(tv(s,t))dsdt=0˜u+(s,t)vt(s,t)dtds0˜u(s,t)vt(s,t)dtds=0[˜u+(s,t)v(s,t)|t=t=+˜u+t(s,t)v(s,t)dt]ds+0[˜u(s,t)v(s,t)|t=t=+˜ut(s,t)v(s,t)dt]ds=0˜u+t(s,t)v(s,t)dtds+0˜ut(s,t)v(s,t)dtds=0.

      Thus, there is no transmission condition for a weak solution with jump discontinuity along the characteristic x=ct.

  6. If fL1loc(Ω), define Ff,vΩf(x)v(x)dx. Show that Ff is a distribution in Ω.
  7. SolutionLet a,bR and v,wD(Ω). It is easy to verify that

    Ff(av+bw):=Ff,av+bw=Ωf(x)[av(x)+bw(x)]dx=aΩf(x)v(x)dx+bΩf(x)w(x)dx=aFf,v+bFf,w=aFf(v)+bFf(w),

    which shows Ff is a linear mapping on D(Ω). Moreover, let {vj}j=1 be a sequence of functions in D(Ω) with supp(vj)K for some compact set KΩ (independent of j) and Dαvj0 uniformly in K. Then for any given ε>0, there exists a NN such that |vj(x)|<ε(K|f(x)|dx)1 for j>N and xK. For j>N, we have

    |Ff(vj)0|=|Ff,vj|=|Ωf(x)vj(x)dx|=|Kf(x)vj(x)dx|K|f(x)||vj(x)|dxK|f(x)|ε(K|f(x)|dx)1dx=ε.

    This implies lim. Therefore, F_f is continuous on \mathcal D(\Omega) and F_f is a distribution in \Omega.

    1. If \xi_1,\dots,\xi_k are points in \Omega and a_1,\dots,a_k are real numbers, show that F\equiv a_1\delta_{\xi_1}+\cdots+a_k\delta_{\xi_k} is a distribution in \Omega.
    2. For \Omega=\mathbb R^n, find an infinite sequence \{\xi_k\}_{k=1}^\infty for which f\equiv a_1\delta_{\xi_1}+\cdots is a distribution. (This gives an example of a distribution that cannot be realized as a distributional derivative of an integrable function.)
  8. Solution
    1. Let a,b\in\mathbb R and v,w\in\mathcal D(\Omega). It is easy to verify that

      \begin{aligned}F(av+bw)&=\langle F,av+bw\rangle=\left\langle\sum_{i=1}^ka_i\delta_{\xi_i},av+bw\right\rangle=\int_\Omega\!\sum_{i=1}^ka_i\delta_{\xi_i}(x)[av(x)+bw(x)]\,\mathrm dx\\&=\sum_{i=1}^ka_i\int_\Omega\!\delta_{\xi_i}(x)[av(x)+bw(x)]\,\mathrm dx=\sum_{i=1}^ka_i[av(\xi_i)+bw(\xi_i)]\\&=a\sum_{i=1}^ka_iv(\xi_i)+b\sum_{i=1}^ka_iw(\xi_i)=a\sum_{i=1}^ka_i\int_\Omega\!\delta_{\xi_i}(x)v(x)\,\mathrm dx+b\sum_{i=1}^ka_i\int_\Omega\!\delta_{\xi_i}(x)w(x)\,\mathrm dx\\&=a\int_\Omega\!\sum_{i=1}^ka_i\delta_{\xi_i}(x)[v(x)]\,\mathrm dx+b\int_\Omega\!\sum_{i=1}^ka_i\delta_{\xi_i}(x)[w(x)]\,\mathrm dx=a\left\langle\sum_{i=1}^ka_i\delta_{\xi_i},v\right\rangle+ba\left\langle\sum_{i=1}^ka_i\delta_{\xi_i},w\right\rangle\\&=a\langle F,v\rangle+b\langle F,w\rangle=aF(v)+bF(w),\end{aligned}

      which shows F is a linear mapping on \mathcal D(\Omega). Moreover, let \{v_j\}_{j=1}^\infty be a sequence of functions in \mathcal D(\Omega) with \text{supp}(v_j)\subseteq K for some compact set K\subsetneq\Omega (independent of j) and D^\alpha v_j\to0 uniformly in K. Since K\cap\{\xi_1,\dots,\xi_k\} is finite, we may write it as \{\xi_{j_1},\dots,\xi_{j_r}\} for r\in\mathbb N. Set A=\max\limits_{1\leq i\leq r}|a_{j_i}|. Then for any \varepsilon>0, there exists N\in\mathbb N such that |v_j(x)|\leq\varepsilon/(rA) for j>N and x\in K. For j>N, we have

      \begin{aligned}|F(v_j)-0|&=|\langle F,v_j\rangle|=\left|\int_\Omega\!F(x)v_j(x)\,\mathrm dx\right|=\left|\int_\Omega\!\sum_{i=1}^ka_i\delta_{\xi_i}(x)v_j(x)\,\mathrm dx\right|\\&\leq\sum_{i=1}^r|a_{j_i}|\left|\int_K\!\delta_{\xi_{j_i}}(x)v_j(x)\,\mathrm dx\right|=\sum_{i=1}^r|a_{j_i}||v_j(\xi_{j_i})|\leq\sum_{i=1}^rA\cdot\frac{\varepsilon}{rA}=\varepsilon.\end{aligned}

      This implies \displaystyle\lim_{j\to\infty}F(v_j)=0. Therefore, F is continuous on \mathcal D(\Omega) and F is a distribution in \Omega.
    2. Choose \xi_j=je_1:=(j,0,\dots,0) for j\in\mathbb N, and then we prove F=\sum_{i=1}^{\infty}a_i\delta_{\xi_i} is a distribution in \mathbb R^n. Let a,b\in\mathbb R and v,w\in\mathcal D(\mathbb R^n). Then there exists a compact subset K\subsetneq\mathbb R^n such that \text{supp}(v)\subseteq K and \text{supp}(w)\subseteq K. Since K is bounded, K\cap\{\xi_j\}_{j=1}^\infty is finite and K\cap\{\xi_j\}_{j=1}^\infty=\{\xi_{j_1},\dots,\xi_{j_r}\} for r\in\mathbb N. Then one may check that

      \begin{aligned}F(av+bw)&=\langle F,av+bw\rangle=\left\langle\sum_{j=1}^\infty a_j\delta_{\xi_j},av+bw\right\rangle=\int_{\mathbb R^n}\!\sum_{j=1}^\infty\!a_j\delta_{\xi_j}(x)[av(x)+bw(x)]\,\mathrm dx\\&=\int_K\!\sum_{i=1}^ra_{j_i}\delta_{\xi_{j_i}}(x)[av(x)+bw(x)]\,\mathrm dx=\sum_{i=1}^ra_{j_i}\int_K\!\delta_{\xi_{j_i}}(x)[av(x)+bw(x)]\,\mathrm dx\\&=\sum_{i=1}^ra_{j_i}[av(\xi_{j_i})+bw(\xi_{j_i})]=a\sum_{i=1}^ra_{j_i}v(\xi_{j_i})+b\sum_{i=1}^ra_{j_i}w(\xi_{j_i})\\&=a\sum_{i=1}^ra_{j_i}\int_K\!\delta_{\xi_{j_i}}(x)v(x)\,\mathrm dx+b\sum_{i=1}^ra_{j_i}\int_K\!\delta_{\xi_{j_i}}(x)w(x)\,\mathrm dx\\&=a\int_K\!\sum_{i=1}^ra_{j_i}\delta_{\xi_{j_i}}(x)v(x)\,\mathrm dx+b\int_K\!\sum_{i=1}^ra_{j_i}\delta_{\xi_{j_i}}(x)w(x)\,\mathrm dx\\&=a\int_{\mathbb R^n}\!\sum_{j=1}^\infty a_j\delta_{\xi_j}(x)v(x)\,\mathrm dx+b\int_{\mathbb R^n}\!\sum_{j=1}^\infty a_j\delta_{\xi_j}(x)w(x)\,\mathrm dx\\&=a\left\langle\sum_{j=1}^\infty a_j\delta_{\xi_j},v\right\rangle+b\left\langle\sum_{j=1}^\infty a_j\delta_{\xi_j},w\right\rangle\\&=a\langle F,v\rangle+b\langle F,w\rangle=aF(v)+bF(w),\end{aligned}

      which shows F is a linear mapping on \mathcal D(\mathbb R^n). Moreover, let \{v_k\}_{k=1}^\infty be a sequence of functions in \mathcal D(\mathbb R^n) with \text{supp}(v_k)\subseteq K for some compact set K\subsetneq\mathbb R^n (independent of j) and D^\alpha v_k\to0 uniformly in K. Since K is bounded, K\cap\{\xi_j\}_{j=1}^\infty is finite and K\cap\{\xi_j\}_{j=1}^\infty=\{\xi_{j_i}\}_{i=1}^r for r\in\mathbb N. Set A=\max\limits_{1\leq i\leq r}|a_{j_i}|. Then for any \varepsilon>0, there exists N\in\mathbb N such that |v_k(x)|<\varepsilon/(rA) for k>N and x\in K. For j>N, we have

      \begin{aligned}|F(v_k)-0|&=|\langle F,v_k\rangle|=\left|\int_{\mathbb R^n}F(x)v_k(x)\,\mathrm dx\right|=\left|\int_{\mathbb R^n}\!\sum_{j=1}^\infty a_j\delta_{\xi_j}(x)v_k(x)\,\mathrm dx\right|\\&=\left|\int_K\!\sum_{i=1}^ra_{j_i}\delta_{\xi_{j_i}}(x)v_k(x)\,\mathrm dx\right|=\left|\sum_{i=1}^ra_{j_i}v_k(\xi_{j_i})\right|\leq\sum_{i=1}^r|a_{j_i}||v_k(\xi_{j_i})|\leq\sum_{i=1}^rA\cdot\frac{\varepsilon}{rA}=\varepsilon.\end{aligned}

      This implies \displaystyle\lim_{k\to\infty}F(v_k)=0. Therefore, F is continuous on \mathcal D(\mathbb R^n) and F is a distribution in \mathbb R^n.

    1. Prove the Lemma on p. 67 when f\in C(\mathbb R^n) has a compact support.
    2. Find an example to show that the result fails to hold if g\in L_{\text{loc}}^1(\mathbb R^n) is replaced by g\in\mathcal D'(\mathbb R^n).
  9. Solution
    1. Suppose that K=\text{supp}(f) is a compact subset of \mathbb R^n. Let x\in\mathbb R^n and y\in\bar B_1(x) arbitrarily and define h(z)=f(x-z)-f(y-z) for z\in\mathbb R^n. Clearly, h is continuous function compact support, and

      \text{supp}(h)\subseteq(x-K)\cup(\bar B_1(x)-K):=\tilde K,

      where

      x-K=\{x-z\in\mathbb R^n\,:\,z\in K\} and \bar B_1(x)-K=\{y-z\in\mathbb R^n\,:y\in\bar B_1(x),\,z\in K\}.

      On the other hand, f is uniformly continuous in \mathbb R^n. Then for any \varepsilon>0, there exists \delta=\delta(\varepsilon)>0 such that \displaystyle|f(x-z)-f(y-z)|<\varepsilon\left(\int_{\tilde K}\!|g(z)|\,\mathrm dz\right)^{-1} whenever |x-y|<\delta and z\in\mathbb R^n. Thus, we find that

      \begin{aligned}|(f\ast g)(x)-(f\ast g)(y)|&=\left|\int_{\mathbb R^n}\![f(x-z)-f(y-z)]g(z)\,\mathrm dz\right|\\&=\left|\int_{\tilde K}[f(x-z)-f(y-z)]g(z)\,\mathrm dz\right|\\&\leq\int_{\tilde K}|f(x-z)-f(y-z)||g(z)|\,\mathrm dz<\varepsilon.\end{aligned}

      This shows \displaystyle\lim_{y\to x}(f\ast g)(y)=(f\ast g)(x), which means f\ast g\in C(\mathbb R^n).
    2. For n=1, let g=\delta'\in\mathcal D'(\mathbb R) and f(x)=|x| for x\in\mathbb R. Then one may check that

      \begin{aligned}(f\ast g)(x)&=\int_{\mathbb R}\!f(x-y)g(y)\,\mathrm dy=\int_{\mathbb R}\!|x-y|\delta'(y)\,\mathrm dy\\&=-\int_{\mathbb R}\!\frac{\mathrm d}{\mathrm dy}|x-y|\delta(y)\,\mathrm dy=\int_{\mathbb R}\text{sgn}(x-y)\delta(y)\,\mathrm dy=\text{sgn}(x),\end{aligned}

      which is not continuous on \mathbb R.

  10. Let \Gamma be a hypersurface (such as a sphere) in \mathbb R^n, let a(z) be a continuous function of z\in\Gamma, and let \mathrm dz denote the surface measure on \Gamma.
    1. Show that \displaystyle\langle F,v\rangle=\int_\Gamma\!a(z)v(z)\,\mathrm dz is a distribution in \mathbb R^n. (It is natural to denote this distribution by F=a\delta_\Gamma.)
    2. Suppose we can choose a unit normal \nu along \Gamma. Formulate a definition for the "conormal distribution" a\partial_\nu\delta_\Gamma in \mathbb R^n.
  11. Solution
    1. Note that \Gamma is a hypersurface in \mathbb R^n so \Gamma is a closed set in \mathbb R^n. For c_1,c_2\in\mathbb R and v_1,v_2\in\mathcal D(\mathbb R^n), we have

      \begin{aligned}\langle F,c_1v_1+c_2v_2\rangle&=\int_{\Gamma}\!a(z)[c_1v_1(z)+c_2v_2(z)]\,\mathrm dz\\&=c_1\int_{\Gamma}\!a(z)v_1(z)\,\mathrm dz+c_2\int_{\Gamma}\!a(z)v_2(z)\,\mathrm dz\\&=c_1\langle F,v_1\rangle+c_2\langle F,v_2\rangle,\end{aligned}

      which shows F is a linear mapping on \mathcal D(\mathbb R^n). Here we have used the fact that v_1,v_2\in C^\infty(\Gamma)\subsetneq\mathcal D(\mathbb R^n). Moreoever, let \{v_j\}_{j=1}^\infty be a sequence of functions in \mathcal D(\mathbb R^n) with \text{supp}(v_j)\subseteq K for some compact set K\subsetneq\mathbb R^n (independent of j) and D^\alpha v_j\to0 uniformly in K. Since a is continuous on the compact set \Gamma\cap K, the maximum of |a| over \Gamma\cap K is finite and can be denoted by A:=\max\limits_{z\in\Gamma\cap K}a(z). For any \varepsilon>0, there exists N\in\mathbb N such that

      \displaystyle|v_j(x)|\leq\varepsilon\left[A\int_{\Gamma\cap K}\!\,\mathrm dz\right]^{-1} for j> N and x\in K.

      For j>N, we have

      \begin{aligned}|\langle F,v_j\rangle-0|&=\left|\int_{\Gamma}\!a(z)v_j(z)\,\mathrm dz\right|=\left|\int_{\Gamma\cap K}\!a(z)v_j(z)\,\mathrm dz\right|\\&\leq\int_{\Gamma\cap K}\!|a(z)||v_j(z)|\,\mathrm dz\leq\int_{\Gamma\cap K}\!A\cdot\varepsilon\left[A\int_{\Gamma\cap K}\!\,\mathrm dz\right]^{-1}\,\mathrm dz=\varepsilon.\end{aligned}

      This implies \displaystyle\lim_{j\to\infty}F(v_j)=0. Therefore, F is continuous on \mathcal D(\mathbb R^n) and F is a distribution in \mathbb R^n.
    2. Suppose a\in C^1(\Gamma). Then we define the conformal distribution a\partial_\nu\delta_\Gamma by

      \displaystyle\langle a\partial_\nu\delta_\Gamma,v\rangle=-\int_\Gamma\nabla(a(z)v(z))\cdot\nu(z)\,\mathrm dz\quad\text{for}~v\in\mathcal D(\mathbb R^n).

      For c_1,c_2\in\mathbb R and v_1,v_2\in\mathcal D(\mathbb R^n), we have

      \begin{aligned}\langle a\partial_\nu\delta_\Gamma,c_1v_1+c_2v_2\rangle&=-\int_\Gamma\!\nabla(a(z)[c_1v_1(z)+c_2v_2(z)])\cdot\nu(z)\,\mathrm dz\\&=-c_1\int_\Gamma\!\nabla(a(z)v_1(z))\cdot\nu(z)\,\mathrm dz-c_2\int_\Gamma\!\nabla(a(z)v_2(z))\cdot\nu(z)\,\mathrm dz\\&=c_1\langle a\partial_\nu\delta_\Gamma,v_1\rangle+c_2\langle a\partial_\nu\delta_\Gamma,v_2\rangle,\end{aligned}

      which shows \langle a\partial_\nu\delta_\Gamma,\cdot\rangle is a linear mapping on \mathcal D(\mathbb R^n). Here we have used the fact that v_1,v_2\in C^\infty(\Gamma)\subsetneq\mathcal D(\mathbb R^n). Moreover, let \{v_j\}_{j=1}^\infty be a sequence of functions in \mathcal D(\mathbb R^n) with \text{supp}(v_j)\subseteq K for some compact set K\subsetneq\mathbb R^n (independent of j) and D^\alpha v_j\to0 uniformly in K. Since a is continuously differentiable on the compact set \Gamma\cap K, \|a\|_{C^1(\Gamma\cap K)}:=\max\limits_{z\in\Gamma\cap K}|a(z)|+\max\limits_{z\in\Gamma\cap K}|\nabla a(z)| is finite. For any \varepsilon>0, there exists N\in\mathbb N such that

      \displaystyle v_j(x)\leq\frac\varepsilon2\left[\|a\|_{C^1(\Gamma\cap K)}\int_{\Gamma\cap K}\mathrm dz\right]^{-1},\quad|\nabla v_j(x)|\leq\frac\varepsilon2\left[\|a\|_{C^1(\Gamma\cap K)}\int_{\Gamma\cap K}\mathrm dz\right]^{-1}\quad\text{for}~j>N~\text{and}~x\in K.

      For j>N, we have

      \begin{aligned}|\langle a\partial_\nu\delta_\Gamma,v_j\rangle|&=\left|\int_\Gamma\!\nabla(a(z)v_j(z))\cdot\nu(z)\,\mathrm dz\right|=\left|\int_\Gamma\![v_j(z)\nabla a(z)+a(z)\nabla v_j(z)]\cdot\nu(z)\,\mathrm dz\right|\\&=\left|\int_{\Gamma\cap K}\![v_j(z)\nabla a(z)+a(z)\nabla v_j(z)]\cdot\nu(z)\,\mathrm dz\right|\leq\int_{\Gamma\cap K}(|v_j(z)||\nabla a(z)|+|a(z)||\nabla v_j(z)|)\,\mathrm dz\\&\leq\int_{\Gamma\cap K}\|a\|_{C^1(\Gamma\cap K)}(|v_j(z)|+|\nabla v_j(z)|)\,\mathrm dz\\&\leq\|a\|_{C^1(\Gamma\cap K)}\int_{\Gamma\cap K}\varepsilon\left[\|a\|_{C^1(\Gamma\cap K)}\int_{\Gamma\cap K}\mathrm dz\right]^{-1}\,\mathrm dz=\varepsilon.\end{aligned}

      This implies \displaystyle\lim_{j\to\infty}\langle a\partial_\nu\delta_\Gamma,v_j\rangle=0. Therefore, \langle a\partial_\nu\delta_\Gamma,\cdot\rangle is continuous on \mathcal D(\mathbb R^n) and F is a distribution in \mathbb R^n.

  12. Let

    f_n(x)=\begin{cases}\displaystyle\frac n2&\displaystyle\text{for}~-\frac1n<x<\frac1n;\\0&\displaystyle\text{for}~|x|\geq\frac1n.\end{cases}

    Show that f_n(x)\to\delta(x) as distributions on \mathbb R.
  13. SolutionFor v\in C_0^\infty(\mathbb R), we find that

    \displaystyle\int_{\mathbb R}\!f_n(x)v(x)\,\mathrm dx=\frac n2\int_{-1/n}^{1/n}\!v(x)\,\mathrm dx.

    Since v is continuous on \mathbb R, for any \varepsilon>0, there exists \delta>0 such that |v(x)-v(0)|<\varepsilon whenever |x|<\delta. Then for n>1/\delta, we have

    \displaystyle\left|\int_{\mathbb R}\!f_n(x)v(x)\,\mathrm dx-v(0)\right|=\frac n2\left|\int_{-1/n}^{1/n}\![v(x)-v(0)]\,\mathrm dx\right|\leq\frac n2\int_{-1/n}^{1/n}|v(x)-v(0)|\,\mathrm dx<\frac n2\int_{-1/n}^{1/n}\!\varepsilon\,\mathrm dx=\varepsilon.

    This shows \displaystyle\lim_{n\to\infty}\int_{\mathbb R}f_n(x)v(x)\,\mathrm dx=v(0)=\int_{\mathbb R}\!\delta(x)v(x)\,\mathrm dx, i.e., f_n(x)\to\delta(x) as distributions on \mathbb R.

  14. If f_n(x) and f(x) are integrable functions such that for any compact set K\subset\Omega we have \displaystyle\int_K\!|f_n(x)-f(x)|\,\mathrm dx\to0 as n\to\infty, then f_n\to f as distributions.
  15. SolutionLet v\in C_0^\infty(\mathbb R) and K=\text{supp}(v). Then we observe that

    \begin{aligned}|\langle f_n,v\rangle-\langle f,v\rangle|&=|\langle f_n-f,v\rangle|=\left|\int_\Omega\!(f_n(x)-f(x))v(x)\,\mathrm dx\right|\\&=\left|\int_K\!(f_n(x)-f(x))v(x)\,\mathrm dx\right|\leq\int_K\!|f_n(x)-f(x)||v(x)|\,\mathrm dx\\&\leq\|v\|_{C^0(K)}\int_K\!|f_n(x)-f(x)|\,\mathrm dx\to0\quad\text{as}~n\to\infty.\end{aligned}

    This means \displaystyle\lim_{n\to\infty}\langle f_n,v\rangle=\langle f,v\rangle for v\in\mathcal D(\Omega), i.e., f_n\to f as distributions on \Omega.

  16. Let a\in\mathbb R, a\neq0.
    1. Find a fundamental solution for L=\mathrm d/\mathrm dx-a on \mathbb R (i.e., solve \mathrm dF/\mathrm dx-aF=\delta).
    2. Show that a fundamental solution for L=\mathrm d^2/\mathrm dx^2-a^2=(\mathrm d/\mathrm dx+a)(\mathrm d/\mathrm dx-a) on \mathbb R is given by

      F(x)=\begin{cases}a^{-1}\sinh ax&\text{if}~x>0;\\0&\text{if}~x<0.\end{cases}

  17. Solution
    1. Multiplying equation by e^{-ax}, it gives

      \displaystyle\frac{\mathrm d}{\mathrm dx}(e^{-ax}F)=e^{-ax}\frac{\mathrm dF}{\mathrm dx}-ae^{-ax}F=e^{-ax}\delta(x)=\delta(x),

      which implies

      e^{-ax}F(x)=H(x)=\begin{cases}1&\text{if}~x>0;\\0&\text{if}~x<0.\end{cases}

      Hence F(x)=e^{ax}H(x)=\begin{cases}e^{ax}&\text{if}~x>0;\\0&\text{if}~x<0.\end{cases}

      Note that the adjoint operator L' of L is -\mathrm d/\mathrm dx-a. It is easy to verify that

      \begin{aligned}\langle LF,v\rangle&=\langle F,L'v\rangle=\int_{\mathbb R}F(x)\left(-\frac{\mathrm dv}{\mathrm dx}-av\right)\,\mathrm dx\\&=-\int_0^\infty\!e^{ax}\frac{\mathrm dv}{\mathrm dx}\,\mathrm dx-a\int_0^\infty\!e^{ax}v(x)\,\mathrm dx\\&=-e^{ax}v(x)\Big|_0^\infty+a\int_0^\infty\!e^{ax}v(x)\,\mathrm dx-a\int_0^\infty\!e^{ax}v(x)\,\mathrm dx\\&=v(0)=\langle\delta,v\rangle,\end{aligned}

      which shows LF=\delta.
    2. Note that L is self-adjoint, i.e., L=L'. Then for v\in\mathcal D(\mathbb R), it is easy to verify that

      \begin{aligned}\langle LF,v\rangle&=\langle F,L'v\rangle=\int_{\mathbb R}F(x)\left(\frac{\mathrm d^2v}{\mathrm dx^2}-a^2v(x)\right)\,\mathrm dx\\&=\int_0^\infty\!a^{-1}\sinh(ax)\left(\frac{\mathrm d^2v}{\mathrm dx^2}-a^2v(x)\right)\,\mathrm dx\\&=\left.a^{-1}\sinh(ax)\frac{\mathrm dv}{\mathrm dx}\right|_0^\infty-\int_0^\infty\!\cosh(ax)\frac{\mathrm dv}{\mathrm dx}\,\mathrm dx-\int_0^\infty\!a\sinh(ax)v(x)\,\mathrm dx\\&=-\cosh(ax)v(x)\Big|_0^\infty+\int_0^\infty\!a\sinh(ax)v(x)\,\mathrm dx-\int_0^\infty\!a\sinh(ax)v(x)\,\mathrm dx\\&=v(0)=\langle\delta,v\rangle,\end{aligned}

      which implies LF=\delta.

  18. Regularity of u defined by (60):
    1. If f(x)\in L^1(\mathbb R) has a compact support, show that (60) defines a continuous weak solution of u''=f(x).
    2. If, in addition to (a), f(x) is a bounded function on \mathbb R, show u\in C^1(\mathbb R), and \displaystyle u'(x)=\frac12\left(\int_{-\infty}^x\!f(y)\,\mathrm dy-\int_x^\infty\!f(y)\,\mathrm dy\right).
    3. If, in addition to (a), f(x) is continuous, show that u is in fact a classical solution: u\in C^2(\mathbb R) with u''=f.
  19. Solution
    1. Since f\in L^1(\mathbb R) has a compact support, f\in L_{\text{loc}}^1(\mathbb R). On the other hand, \displaystyle g(x)=\frac12|x|\in C(\mathbb R). Thus, by Lemma at Page 67, \displaystyle u(x)=(g\ast f)(x)=\frac12\int_{\mathbb R}\!|x-y|f(y)\,\mathrm dy is continuous on \mathbb R.
    2. From (60), we have

      \displaystyle u(x)=\frac12\int_{-\infty}^x\!(x-y)f(y)\,\mathrm dy+\frac12\int_x^\infty\!(y-x)f(y)\,\mathrm dy.

      A direct computation gives

      \begin{aligned}u(x+h)-u(x)&=\frac12\int_{-\infty}^{x+h}\!(x+h-y)f(y)\,\mathrm dy-\frac12\int_{-\infty}^x\!(x-y)f(y)\,\mathrm dy\\&\quad+\frac12\int_{x+h}^\infty\!(y-x-h)f(y)\,\mathrm dy-\frac12\int_x^\infty\!(y-x)f(y)\,\mathrm dy\\&=\frac h2\int_{-\infty}^{x+h}\!f(y)\,\mathrm dy+\frac12\int_x^{x+h}\!(x-y)f(y)\,\mathrm dy-\frac h2\int_{x+h}^\infty\!f(y)\,\mathrm dy-\frac12\int_x^{x+h}\!(y-x)f(y)\,\mathrm dy.\end{aligned}

      Since f is bounded, there exists M>0 such that |f(y)|\leq M for y\in\mathbb R. Then it is easy to observe that

      \displaystyle\left|\int_x^{x+h}\!\pm(x-y)f(y)\,\mathrm dy\right|\leq\left|\int_x^{x+h}\!(y-x)M\,\mathrm dy\right|=\frac{Mh^2}2,

      which implies \displaystyle\lim_{h\to0}\frac1h\int_x^{x+h}\!\pm(x-y)f(y)\,\mathrm dy=0 by the squeeze theorem. Therefore, by the definition, we find

      \begin{aligned}u'(x)&=\lim_{h\to0}\frac{u(x+h)-u(x)}h\\&=\frac12\int_{-\infty}^{x+h}\!f(y)\,\mathrm dy-\frac12\int_{x+h}^\infty\!f(y)\,\mathrm dy+\frac1{2h}\int_x^{x+h}(x-y)f(y)\,\mathrm dy-\frac1{2h}\int_x^{x+h}\!(y-x)f(y)\,\mathrm dy\\&=\frac12\int_{-\infty}^x\!f(y)\,\mathrm dy-\frac12\int_x^\infty\!f(y)\,\mathrm dy.\end{aligned}

      Here we have used the fact that functions \displaystyle\int_x^\infty\!f(y)\,\mathrm dy and \displaystyle\int_{-\infty}^x\!f(y)\,\mathrm dy are continuous in x because f\in L^1(\mathbb R) is bounded with compact support. This also shows u'\in C(\mathbb R), and hence u\in C^1(\mathbb R).
    3. When continuous function f\in L^1(\mathbb R) has a compact support, f is bounded on \mathbb R. Thus, by (b), u\in C^1(\mathbb R). Moreover, by the fundamental theorem of calculus, functions \displaystyle\int_{-\infty}^x\!f(y)\,\mathrm dy and \displaystyle\int_x^\infty\!f(y)\,\mathrm dy are differentiable in x, and then u' is also differentiable. By the fundamental theorem of calculus, we arrive at \displaystyle u''(x)=\frac12f(x)-\frac12\cdot(-f(x))=f(x) for x\in\mathbb R. Since f\in C(\mathbb R), u''\in C(\mathbb R) and then u\in C^2(\mathbb R).

    1. Use the fundamental solution (59) to solve the initial u''=f(x) for x>0 with u(0)=u_0 and u'(0)=u_0', where f\in C^\infty([0,\infty)) and f=O(|x|^{-2-\epsilon}) as |x|\to\infty.
    2. Do the same for the boundary value problem u''=f(x) for 0<x<\ell with u(0)=0=u(\ell).
  20. Solution
    1. Extend \tilde f by \tilde f(x)=f(x) if x>0; 0 if x\leq0. Then we consider the equation \tilde u''(x)=\tilde f(x) for x\in\mathbb R. Plugging (59) into (56), we obtain the particular solution

      \begin{aligned}u_p(x)&=(F\ast f)(x)=\int_{\mathbb R}F(x-y)\tilde f(y)\,\mathrm dy\\&=\frac12\int_{\mathbb R}|x-y|\tilde f(y)\,\mathrm dy=\frac12\int_0^\infty\!|x-y|f(y)\,\mathrm dy\quad\text{for}~x\in\mathbb R.\end{aligned}

      To fulfill the initial condition, we need to find a function u_h satisfying

      \begin{aligned} &u_h''(x)=0,\\&u_h(0)=u_0-u_p(0)=u_0-\frac12\int_0^\infty\!yf(y)\,\mathrm dy,\\&u_h'(0)=u_0'-u_p'(0)=u_0'+\frac12\int_0^\infty\!f(y)\,\mathrm dy.\end{aligned}

      Here we have used the result of Exercise 11(b). It is easy to obtain

      \displaystyle u_h(x)=u_h(0)+u_h'(0)x=u_0-\frac12\int_0^\infty\!yf(y)\,\mathrm dy+\left(u_0'+\frac12\int_0^\infty\!f(y)\,\mathrm dy\right)x\quad\text{for}~x\in\mathbb R.

      Thus, we find

      \begin{aligned}\tilde u(x)&=u_h(x)+u_p(x)\\&=u_0-\frac12\int_0^\infty\!yf(y)\,\mathrm dy+\left(u_0'+\frac12\int_0^\infty\!f(y)\,\mathrm dy\right)x+\frac12\int_0^\infty\!|x-y|f(y)\,\mathrm dy\quad\text{for}~x\in\mathbb R.\end{aligned}

      In particular, the solution is u(x)=\tilde u\Big|_{x\geq0}(x) for x\geq0.

      It is remark that the solution u can be represented as

      \begin{aligned} u(x)&=u(0)+\int_0^x\!u'(t)\,\mathrm dt=u_0+\int_0^x\!\left[u'(0)+\int_0^t\!u''(s)\,\mathrm ds\right]\,\mathrm dt\\&=u_0+u_0'x+\int_0^x\!\int_0^t\!f(s)\,\mathrm ds\,\mathrm dt=u_0+u_0'x+\int_0^x\!\int_s^x\!f(s)\,\mathrm dt\,\mathrm ds\\&=u_0+u_0'x+\int_0^x\!(x-s)f(s)\,\mathrm ds\quad\text{for}~x\geq0.\end{aligned}.

    2. Extend \tilde f by \tilde f(x)=f(x) if 0<x<\ell; 0 if x\leq0 or x\geq\ell. Then we consider the equation \tilde u''(x)=\tilde f(x) for x\in\mathbb R. Plugging (59) into (56), we obtain the particular solution

      \begin{aligned}u_p(x)&=(F\ast f)(x)=\int_{\mathbb R}\!F(x-y)\tilde f(y)\\&=\frac12\int_{\mathbb R}\!|x-y|\tilde f(y)\,\mathrm dy=\frac12\int_0^\ell\!|x-y|f(y)\,\mathrm dy\quad\text{for}~x\in\mathbb R.\end{aligned}

      To fulfill the boundary condition, we need to find a function u_h satisfying

      \begin{aligned} &u_h''(x)=0,\\&u_h(0)=-u_p(0)=-\frac12\int_0^\ell\!yf(y)\,\mathrm dy,\\&u_h(\ell)=-u_p(\ell)=-\frac12\int_0^\ell\!(\ell-y)f(y)\,\mathrm dy.\end{aligned}

      It is easy to obtain

      \begin{aligned}u_h(x)&=-\left(\frac1{2\ell}\int_0^\ell\!(\ell-y)f(y)\,\mathrm dy\right)x-\left(\frac1{2\ell}\int_0^\ell\!yf(y)\,\mathrm dy\right)(\ell-x)\\&=\frac{x}{\ell}\int_0^\ell\!(2y-\ell)f(y)\,\mathrm dy-\frac12\int_0^\ell\!yf(y)\,\mathrm dy\quad\text{for}~x\in\mathbb R.\end{aligned}

      Thus, we find

      \begin{aligned}\tilde u(x)&=u_h(x)+u_p(x)\\&=\frac{x}{\ell}\int_0^\ell\!(2y-\ell)f(y)\,\mathrm dy-\frac12\int_0^\ell\!yf(y)\,\mathrm dy+\frac12\int_0^\ell\!|x-y|f(y)\,\mathrm dy\quad\text{for}~x\in\mathbb R.\end{aligned}

      In particular, the solution is u(x)=\tilde u\Big|_{0\leq x\leq\ell}(x) for 0\leq x\leq\ell.

      Note that u(\ell)=0=u(0), which implies \displaystyle\int_0^\ell\!u'(t)\,\mathrm dt=0, and hence

      \displaystyle0=\int_0^\ell\!\left[u'(0)+\int_0^t\!u''(s)\,\mathrm ds\right]\,\mathrm dt=\ell u'(0)+\int_0^\ell\int_0^t\!f(s)\,\mathrm ds\,\mathrm dt.

      Thus, \displaystyle u'(0)=-\frac1\ell\int_0^\ell\!\int_0^t\!f(s)\,\mathrm ds\,\mathrm dt. Therefore, the solution can be represented as

      \begin{aligned}u(x)&=u(0)+\int_0^x\!u'(t)\,\mathrm dt=\int_0^x\!\left[u'(0)+\int_0^t\!u''(s)\,\mathrm ds\right]\,\mathrm dt\\&=-\left(\frac1\ell\int_0^\ell\!\int_0^t\!f(s)\,\mathrm ds\,\mathrm dt\right)x+\int_0^x\!\int_0^t\!f(s)\,\mathrm ds\,\mathrm dt\\&=-\frac x\ell\int_0^\ell\!(\ell-s)f(s)\,\mathrm ds+\int_0^x\!(x-s)f(s)\,\mathrm ds\quad\text{for}~0\leq x\leq\ell.\end{aligned}


    1. Using \delta(\mu,\eta)=\delta(\mu)\delta(\eta), show that each of the following functions is a fundamental solution of L=\partial^2/\partial\mu\partial\eta:

      \begin{aligned} &F_1(\mu,\eta)=H(\mu)H(\eta),\quad F_2(\mu,\eta)=-H(\mu)H(-\eta),\\&F_3(\mu,\eta)=-H(-\mu)H(\eta),\quad F_4(\mu,\eta)=H(-\mu)H(-\eta).\end{aligned}

    2. Use part (a) with the change of variables \mu=x+ct, \eta=x-ct to obtain four distinct fundamental solutions for the one-dimensional wave operator, each having support in one of the wedges determined by the lines x=\pm ct.
  21. Solution
    1. We firstly note that L is self-adjoint, i.e., L=L'. For v\in\mathcal D(\mathbb R^2), it is easy to verify that

      \begin{aligned}\langle LF_1,v\rangle&=\langle F_1,L'v\rangle=\int_{\mathbb R^2}H(\mu)H(\eta)v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta=\int_0^\infty\!\int_0^\infty\!v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta\\&=\int_0^\infty\!v_\eta(\mu,\eta)\Big|_{\mu=0}^{\mu=\infty}\,\mathrm d\eta=-\int_0^\infty\!v_\eta(0,\eta)\,\mathrm d\eta=-v(0,\eta)\Big|_{\eta=0}^\infty=v(0,0)=\langle\delta,v\rangle,\\\langle LF_2,v\rangle&=\langle F_2,L'v\rangle=-\int_{\mathbb R^2}\!H(\mu)H(-\eta)v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta=-\int_{-\infty}^0\!\int_0^\infty\!v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta\\&=-\int_{-\infty}^0\!v_{\eta}(\mu,\eta)\Big|_{\mu=0}^{\mu=\infty}\,\mathrm d\eta=\int_{-\infty}^0\!v_\eta(0,\eta)\,\mathrm d\eta=v(0,\eta)\Big|_{\eta=-\infty}^{\eta=0}=v(0,0)=\langle\delta,v\rangle,\\\langle LF_3,v\rangle&=\langle F_3,L'v\rangle=-\int_{\mathbb R^2}H(-\mu)H(\eta)v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta=-\int_0^\infty\!\int_{-\infty}^0\!v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta\\&=-\int_0^\infty\!v_{\eta}(\mu,\eta)\Big|_{\mu=-\infty}^{\mu=0}\,\mathrm d\eta=-\int_0^\infty\!v_\eta(0,\eta)\,\mathrm d\eta=-v(0,\eta)\Big|_{\eta=0}^{\eta=\infty}=v(0,0)=\langle\delta,v\rangle\\\langle LF_4,v\rangle&=\langle F_4,L'v\rangle=\int_{\mathbb R^2}H(-\mu)H(-\eta)v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta==\int_{-\infty}^0\!\int_{-\infty}^0\!v_{\mu\eta}(\mu,\eta)\,\mathrm d\mu\,\mathrm d\eta\\&=\int_{-\infty}^0\!v_\eta(\mu,\eta)\Big|_{\mu=-\infty}^{\mu=0}\,\mathrm d\eta=\int_{-\infty}^0\!v_\eta(0,\eta)\,\mathrm d\eta=v(0,\eta)\Big|_{\eta=-\infty}^{\eta=0}=v(0,0)=\langle\delta,v\rangle.\end{aligned}

      This shows F_i is a fundamental solution of LF_i=\delta for i=1,2,3,4.
    2. By (a) with the change of variables \mu=x+ct and \eta=x-ct, the fundamental solutions for -4c^2\tilde u_{\mu\eta}(\mu,\eta)=u_{tt}(x,t)-c^2u_{xx}(x,t)=\delta are

      \begin{aligned} &U_1(x,t)=-(2c)^{-1}H(x+ct)H(x-ct),\quad U_2(x,t)=(2c)^{-1}H(x+ct)H(-x+ct),\\&U_3(x,t)=(2c)^{-1}H(-x-ct)H(x-ct),\quad U_4(x,t)=-(2c)^{-1}H(-x-ct)H(-x+ct).\end{aligned}


  22. If K(x,t) is a distribution in \mathbb R^n depending on the parameter t, then we say K depends continuously on t if, for every v\in C_0^\infty(\mathbb R^n), \displaystyle f(t)\equiv\langle K(\cdot,t),v\rangle=\int\!K(x,t)v(x)\,\mathrm dx defines a continuous function of t. Use this idea to define K is continuously differentiable in t.
  23. SolutionK is continuously differentiable in t if for every v\in C_0^\infty(\mathbb R^n), \displaystyle f(t)=\int_{\mathbb R^n}\!K(x,t)v(x)\,\mathrm dx defines a continuously differentiable function of t.

  24. According to Section 1.2, a weak solution of u_y+(G(u))_x=0 in a rectangle \Omega=(a,b)\times (c,d) satisfies

    \displaystyle\frac{\mathrm d}{\mathrm dy}\int_{x_1}^{x_2}\!u(x,y)\,\mathrm dx+G(u(x,y))\Big|_{x=x_1}^{x=x_2}=0

    for every [x_1,x_2]\subset(a,b) and c<y<d. Show that this implies

    \displaystyle\int_\Omega\!uv_y+G(u)v_x\,\mathrm dx\,\mathrm dy=0\quad\text{for every}~v\in C_0^1(\Omega)

    (i.e., the notion of weak solution of Section 1.2 is consistent with that of Section 2.3).
  25. SolutionLet x_i=a+(b-a)i/m for i=0,1,\dots,m and y_j=c+(d-c)j/n for j=1,\dots,n. Set \Omega_{ij}=[x_{i-1},x_i]\times[y_{j-1},y_j] for 1\leq i\leq n and 1\leq j\leq m. Then \{\Omega_{ij}\}_{1\leq i\leq n,1\leq j\leq m} forms a partition of \Omega. Put \Delta x=(b-a)/m and \Delta y=(d-c)/n. Then using the summation by parts, we have

    \begin{aligned} &\quad\sum_{i=1}^m\sum_{j=1}^n\left[u(x_i,y_j)\frac{v(x_i,y_j)-v(x_i,y_{j-1})}{\Delta y}+G(u(x_i,y_j))\frac{v(x_i,y_j)-v(x_{i-1},y_j)}{\Delta x}\right]\Delta y\Delta x\\&=\sum_{i=1}^m\left[\frac{u(x_i,y_n)v(x_i,y_n)-u(x_i,y_1)v(x_i,y_0)}{\Delta y}-\sum_{j=2}^nv(x_i,y_{j-1})\frac{u(x_i,y_j)-u(x_i,y_{j-1})}{\Delta y}\right]\Delta y\Delta x\\&\quad+\sum_{j=1}^n\left[\frac{G(u(x_m,y_j))v(x_m,y_j)-G(u(x_1,y_j))v(x_0,y_j)}{\Delta x}-\sum_{i=2}^mv(x_{i-1},y_j)\frac{G(u(x_i,y_j))-G(u(x_{i-1},y_j))}{\Delta x}\right]\Delta y\Delta x\\&=-\sum_{i=1}^m\sum_{j=2}^n\left[v(x_i,y_{j-1})\frac{u(x_i,y_j)-u(x_i,y_{j-1})}{\Delta y}\right]\Delta y\Delta x-\sum_{i=2}^m\sum_{j=1}^n\left[v(x_{i-1},y_j)\frac{G(u(x_i,y_j))-G(u(x_{i-1},y_j))}{\Delta x}\right]\Delta y\Delta x\\&=-\sum_{i=1}^{m-1}\sum_{j=1}^{n-1}\left\{v(x_i,y_j)\left[\frac{u(x_i,y_{j+1})-u(x_i,y_j)}{\Delta y}+\frac{G(u(x_{i+1},y_j))-G(u(x_i,y_j))}{\Delta x}\right]\right\}\Delta y\Delta x\end{aligned}

    Here we have used the fact that v\in C_0^\infty(\Omega) so v(x_0,y_j)=v(x_m,v_j)=v(x_i,y_0)=v(x_i,y_n)=0 for 1\leq i\leq m and 1\leq j\leq n. Note that

    \displaystyle G(u(x_{i+1},y_j))-G(u(x_i,y_j))=-\int_{x_i}^{x_{i+1}}u_y(x,y_j)\,\mathrm dx\quad\text{for}~1\leq i\leq m-1,~1\leq j\leq n-1.

    Then the double sum becomes

    \begin{aligned} &\quad-\sum_{i=1}^{m-1}\sum_{j=1}^{n-1}\left\{v(x_i,y_j)\left[\frac{u(x_i,y_{j+1})-u(x_i,y_j)}{\Delta y}-\frac1{\Delta x}\int_{x_i}^{x_{i+1}}\!u_y(x,y_j)\,\mathrm dx\right]\right\}\Delta y\Delta x\\&\to\int_a^b\!\int_c^d\!v(x,y)\left[u_y(x,y)-u_y(x,y)\right]\,\mathrm dx\,\mathrm dy=0\quad\text{as}~n,m\to\infty.\end{aligned}

    This shows \displaystyle\int_\Omega\!uv_y+G(u)v_x\,\mathrm dx\,\mathrm dy=0 for v\in C_0^1(\Omega).

    Warning: This problem is quite difficult so I cannot promise this solution is correct. Please provide some references or theorem that I may use to make this solution better.


  26. The mth-order operator (32a) is elliptic at x if its principal symbol (32c) has no nonzero characteristics [i.e., \sigma_L(x,\xi)\neq0 for all \xi\in\mathbb R^n\backslash\{0\}]. In this case, show that m must be an even integer.
  27. SolutionSuppose by contradiction that m is an odd integer. Then \displaystyle\sigma_L(x,\xi)=\sum_{|\alpha|=m}a_\alpha(x)\xi^\alpha is a polynomial of odd degree. Since a_\alpha(x) is not all zero, we may assume a_{e_1}(x)=a_{(m,0,\dots,0)}(x)\neq0 by rearranging the indices. Thus, we have

    \displaystyle\sigma_L(x,\xi)=a_{me_1}(x)\xi_1^m+\sum_{|\alpha|=m,\,\alpha\neq me_1}a_\alpha(x)\xi^\alpha.

    Since m is odd and a_\alpha(x) are real numbers, by the fundamental theorem of algebra, there exists \xi_1=\xi_1(\xi_2,\dots,\xi_m,x) for any (\xi_2,\dots,\xi_m)\neq0. This means the principal symbol has nonzero characteristics, which leads a contradiction. Hence m must be an even integer.

沒有留言:

張貼留言