Problem set 8

Problem 1

Let TT be a distribution, show that if T=0T'=0, then TT is a constant, i.e. for every φCc\varphi \in \mathscr{C}_c^{\infty} (T,φ)=cφ(T,\varphi) = c\int \varphi for some constant cc.

Problem 2

In the lecture we showed that x[H(x)]=δ\frac{\partial }{\partial x}[H(x)] = \delta directly by definition. In this problem we take a closer look at this fact and get some intuition.

(1) Construct a “smooth version” of the Heviside function, i.e. find a function hh such that

h(x)={1,x>10,x<0smooth,x(ϵ,1+ϵ)h(x) =\begin{cases} 1,x>1\\ 0, x<0\\ smooth, x\in (-\epsilon,1+\epsilon)\end{cases}

(2) Similar to the construction of smooth good kernel, consider the family of functions hϵ(x)=h(xϵ)h_\epsilon(x) = h(\frac{x}{\epsilon}). Show that hϵHh_\epsilon \to H in D\mathcal D', i.e. for every φD\varphi \in \mathcal D, (Thϵ,φ)(H,φ)(T_{h_\epsilon},\varphi)\to (H,\varphi).

(3) What can you conclude if you take derivative on (2) ?

(4) Let χ(a,a)\chi_{(-a,a)} be the characteristic function of the interval (a,a)(-a,a), calculate ddx[χ(a,a)]\frac{d}{dx}[\chi_{(-a,a)}].

Problem 3

This problem is a high dimensional generalization of x[H(x)]=δ\frac{\partial }{\partial x}[H(x)] = \delta. In high dimension there are many directions to take derivative and things becomes more complicated but turns out to be more intuitive.

Let DD be a bounded domain in Rn\mathbb{R}^n. We say that DD has C1\mathscr{C}^1 boundary if For every x0Dx_0\in \partial D, there exists a C1\mathscr{C}^1 function ρ on an open neighbourhood UU of x0x_0 such that

  • ρ(x0)=0\rho(x_0) = 0.
  • dρ(x0)0d\rho(x_0)\neq 0.
  • DU={ρ<0}D\cap U = \{\rho < 0\}.

Here dd is the differential df=fx1dx1++fxndfndf = \frac{\partial f}{\partial x_1} dx_1 + \cdots + \frac{\partial f}{\partial x_n} df_n, which can be identified with the gradient of ff.

(1) Show that if DD is a bounded domain with C1\mathscr{C}^1 boundary, then there exists a C1\mathscr{C}^1 function ρ on Rn\mathbb{R}^n such that

  • D={ρ<0}D = \{\rho < 0\}.
  • ρ(x)=0\rho(x) = 0 and dρ(x)0d\rho(x) \neq 0 for every xDx\in \partial D.

Let χD\chi_D be the characteristic function of DD. We want to calculate j[χD]\partial_j[\chi_D], where we use j\partial_j to denote the partial derivative xj\frac{\partial}{\partial x_j} for simplicity.

(2) Let pDp\in \partial D with 1ρ(p)0\partial_1\rho(p)\neq 0. Then on a neighbourhood UU of pp, there exists ψ such that ρ(x)=0    x1=ψ(x)\rho(x) = 0\iff x_1 = \psi(x'), where xx' means (x2,,xn)(x_2,\dots,x_n). ALso, 1ρ(x0)>0\partial_1\rho(x_0) > 0 (resp. <0< 0)     \implies DD is given by x1<ψ(x)x_1<\psi(x') (resp. x1>ψ(x)x_1 > \psi(x')).

(3) Let pp and UU be from (2) such that DD is given by x1>ψ(x)x_1>\psi(x') on UU. Draw a picture of χD,h(x1ψ(x))\chi_D, h(x_1 - \psi(x')) and hϵ(x1ψ(x))h_\epsilon(x_1 - \psi(x')) where hh and hϵh_\epsilon are from Problem 2. Show that

χD=limϵ0hϵ(x1ψ(x))\chi_D = \lim_{\epsilon \to 0} h_\epsilon(x_1 - \psi(x'))

in UU pointwise and in the sense of distributions.

(4) Let φCc(U)\varphi \in \mathscr{C}_c^{\infty}(U). Show that (jχD,φ)=limϵ0νjhϵ(x1ψ(x))φ(x)dx=νj(x)φ(ψ(x),x)dx(\partial_j\chi_D,\varphi) = \lim_{\epsilon\to 0} \int \nu_j \cdot h_{\epsilon}'(x_1 - \psi(x'))\cdot \varphi(x)dx = \int \nu_j(x')\varphi(\psi(x'),x')dx', where ν=(1,xψ)=(1,2ψ,,nψ)\nu =(1,-\frac{\partial}{\partial x'}\psi) = (1,-\partial_2\psi,\dots,-\partial_n\psi). Where hϵ(t)h'_\epsilon(t) is the derivative of hϵh_\epsilon given by 1ϵh(tϵ)\frac{1}{\epsilon}h'(\frac{t}{\epsilon}).

(5) Recall that the Euclidean surface element dSdS on D\partial D is given by 1+ψ2dx\sqrt{1+|\psi'|^2}dx' on UU and n:=ν1+ψ2n:= \frac{\nu}{\sqrt{1+|\psi'|^2}} is the invards normal on UDU\cap \partial D. So (4) shows that jχD=njdS\partial_j \chi_D = n_j dS on UU (this means for every φCc(U)\varphi \in \mathscr{C}_c^{\infty}(U),...). Use partition of unity to conclude that

j[χD]=njdS\partial_j [\chi_D] = n_j dS

on Rn\mathbb{R}^n (or any neighbourhood of DD) <- This means for every φCc(Rn)\varphi \in \mathscr{C}_c^{\infty}(\mathbb{R}^n) or Cc(X)\mathscr{C}_c^{\infty}(X) for open XDX\supset D,...

(6) Prove the Gauss-Green formula

Ddivf  dx=Dfn  dS.\int_D \mathrm{div} f \; dx = -\int_{\partial D} f\cdot n\; dS.

(7) Let uu be a C1\mathscr{C}^1 function on a neighbourhood of DD. Show that j[uχD]=(ju)χD+unjdS\partial_j[u\chi_D] = (\partial_j u)\chi_D + u n_j dS.

(8) Prove the integration by parts formula in high dimensional case: Let ψC1(Rn)\psi\in \mathscr{C}^{1}(\mathbb{R}^n), then iTψ=Tiψ\partial_i T_{\psi} = T_{\partial_i\psi}.

Appendix: Surface measure

Case of graph of function

Let z=ψ(x,y)z = \psi(x,y) for C1\mathscr{C}^1 function ψ and dψ0d\psi\neq 0 everywhere, let VV be a domain on the x,yx,y-plane, then the graph of ψ determines a surface patch SS in R3\mathbb{R}^3. Let ff be a continuous function on R3\mathbb{R}^3, then the surface integral of ff on SS is given by Vf(x,y)1+ψx2+ψy2  dxdy\int_{V}f(x,y)\sqrt{1+\psi_x^2+\psi_y^2} \;dxdy. Similarly, in Rn\mathbb{R}^n and hypersurface determined by graph xn=ψ(x1,,xn)x_n = \psi(x_1,\dots,x_n) for (x1,,xn1)VRn1(x_1,\dots,x_{n-1})\in V\subset \mathbb{R}^{n-1}, we let (dS,f):=SfdS=Vf(x1,,xn1,ψ(x1,,xn1))1+1ψ2++n1ψ2dx1dxn1(dS,f) := \int_S f dS = \int_{V}f(x_1,\dots,x_{n-1},\psi(x_1,\dots,x_{n-1}))\sqrt{1+\partial_1\psi ^2+\cdots+\partial_{n-1}\psi^2}dx_1\cdots dx_{n-1}.

General case

Let UU be an open set on which 1ρ>0\partial_1 \rho > 0, let ψ be the function given in problem 3, so DD is given by x1<ψ(x)x_1<\psi(x') on UU. In other words, we have written D\partial D as graph of function on UU. Following previous discussion, let φ be a continuous function supported on UU, define (dS,φ)(dS,\varphi) to be VRn1φ(ψ(y),y)1+ψ(y)2dy\int_{V\subset \mathbb{R}^{n-1}} \varphi(\psi(y),y)\sqrt{1+|\psi'(y)|^2}dy, where VV is the corresponding domain for ψ.

Choose an open cover (Uj)jJ(U_j)_{j\in J} and use partition of unity (φ)j(\varphi)_j subordinate to UjU_j we can define (dS,f)=jJ(dS,φjf)(dS,f) = \sum_{j\in J}(dS,\varphi_j f).