Problem set 12

Integration on manifold (2 dim case)

Problem 1

Let MR3M\subset \mathbb{R}^3 be a surface, let x(u,v)=(x1(u,v)x2(u,v)x3(u,v))\mathbf{x}(u,v)=\begin{pmatrix}x_1(u,v)\\x_2(u,v)\\x_3(u,v)\end{pmatrix} be parametrization on UU and x(s,t)=(x1(s,t)x2(s,t)x3(s,t))\mathbf{x}(s,t)=\begin{pmatrix}x_1(s,t)\\x_2(s,t)\\x_3(s,t)\end{pmatrix} be parametrization on VV.

Let xu=((x1)u(x2)u(x3)u)\mathbf{x}_u = \begin{pmatrix}(x_1)_u\\ (x_2)_u\\ (x_3)_u\end{pmatrix}, where (xi)u(x_i)_u means derivative xiu(u,v)\frac{\partial x_i}{\partial u}(u,v), and xv=((x1)v(x2)v(x3)v)\mathbf{x}_v = \begin{pmatrix} (x_1)_v \\ (x_2)_v \\ (x_3)_v \end{pmatrix}.

(1) Let (u(s,t)v(s,t))\begin{pmatrix} u(s,t)\\v(s,t)\end{pmatrix} be change of coordinate, and J:=(usutvsvt)J:=\begin{pmatrix} \frac{\partial u}{\partial s}& \frac{\partial u}{\partial t}\\ \frac{\partial v}{\partial s} & \frac{\partial v}{\partial t}\end{pmatrix} be its Jacobi matrix. Verify that (xuxv)=J(xsxt)\begin{pmatrix}\mathbf{x}_u \\ \mathbf{x}_v\end{pmatrix} = J \begin{pmatrix} \mathbf{x}_s \\ \mathbf{x_t}\end{pmatrix}.

Let E=xu,xvE = \langle \mathbf{x}_u,\mathbf{x}_v\rangle be square length of xu\mathbf{x}_u, G=xv,xvG = \langle \mathbf{x}_v,\mathbf{x}_v\rangle be square length of xv\mathbf{x}_v, and F=xu,xvF = \langle \mathbf{x}_u,\mathbf{x}_v\rangle. Let g=(EFFG)g = \begin{pmatrix}E&F\\F&G\end{pmatrix}.

(2) Let E~=xs,xs\tilde E = \langle \mathbf{x}_s,\mathbf{x}_s\rangle, G~=xt,xt\tilde{G} = \langle \mathbf{x}_t,\mathbf{x}_t\rangle and F~=xs,xt\tilde{F} = \langle \mathbf{x}_s,\mathbf{x}_t \rangle be the corresponding data on VV, and g~:=(E~F~F~G~)\tilde{g}:=\begin{pmatrix}\tilde{E}&\tilde{F}\\ \tilde{F} &\tilde{G} \end{pmatrix}. Show that g~=JTgJ\tilde g = J^{T}g J.

(3) Consider the two form det(g)dudv\sqrt{\det(g)}du\wedge dv. Assume detJ>0\det J> 0. Show that detg  dudv=detg~  dsdt\sqrt{\det g}\;du\wedge dv = \sqrt{\det \tilde{g}}\;ds\wedge dt.

If one can always choose parametrizations such that detJ>0\det J>0 for every change of coordinate, we call such MM oriented. We always assume a surface is oriented. The above exercise indicates that one can define a positive measure dVdV on MM, which is given on any parametrization UU by detg  dudv\sqrt{\det g}\;du\wedge dv. This is well defined on MM because on any coordinate it is of the same form by (3).

Let MM be a compact surface, let (Ui)i=1N(U_i)_{i=1}^N be its coordinate charts. On each UiU_i we parametrize MM by x(ui,vi)\mathbf{x}(u_i,v_i). Let dVdV be the area measure on MM. Now we define the integral Mf  dV\int_{M}f\;dV for fC(M)f\in \mathscr{C}^{\infty}(M).

Choose partition of unity (ρi)i=1N(\rho_i)_{i=1}^N subordinate to (Ui)i=1N(U_i)_{i=1}^N. We know that ρif\rho_i f is supported on UiU_i. Then we define

Mf  dV:=i=1NUiρi(ui,vi)f(ui,vi)detgi  duidvi\int_{M} f \; dV:= \sum_{i=1}^N \int_{U_i}\rho_i(u_i,v_i)f(u_i,v_i)\sqrt{\det g_i}\;du_idv_i

. One can check that this does not depend on choice of partition of unity.

Smoothing operators

The statements are given on surface, the same argument will work on compact smooth manifolds with integration defined. If you are really uncomfortable with integration on surfaces, you can replace all “MM” by an open set UU in Rn\mathbb{R}^n.

Let MM be a compact smooth surface. Let KC(M×M)K\in \mathscr{C}^{\infty}(M\times M). Define TK:C(M)C(M)T_K:\mathscr{C}^{\infty}(M)\to \mathscr{C}^{\infty}(M) given by TKf(x)=MK(x,y)f(y)dV(y)T_K f(x) = \int_{M}K(x,y)f(y)dV(y).

Problem 2

(1) Let TK1,TK2T_{K_1},T_{K_2} be smoothing operators. Show that TK1TK2T_{K_1}\circ T_{K_2} is a smoothing operator. In other words, TK1TK2=TKT_{K_1}\circ T_{K_2} = T_{K'} for some KC(M×M)K'\in \mathscr{C}^{\infty}(M\times M).

Denote KK' in (1) by K1K2K_1\circ K_2.

(2) Let Vol(M)=MdV\mathrm{Vol(M)} =\int_{M}dV be the volume (area) of MM. Show that if max(x,y)M×MK(x,y)<ϵV\max_{(x,y)\in M\times M}{|K(x,y)|}<\frac{\epsilon}{V} for some 0<ϵ<10<\epsilon < 1 then ITKI-T_K is invertible.

Differential operator on manifold

Problem 3

Let U,VU,V be open sets in R2\mathbb{R}^2. φ:VU\varphi:V\to U be a diffeomorphism (bijective, smooth, det\det(Jacobi matrix)0\neq 0). Let f:URf:U\to \mathbb{R} be a smooth function. Then fφf\circ \varphi is a smooth function on VV, we denote this by φf\varphi^* f. Since φ is a smooth diffeo, we can do samething for φ1\varphi^{-1}, so that φ\varphi^* and (φ1)(\varphi^{-1})^* give bijection C(U)\mathscr{C^{\infty}}(U) and C(V)\mathscr{C}^{\infty}(V).

(1) Check that (φ1)Djαφ=((φ1)Djφ))α(\varphi^{-1})^*D_j^{\alpha}\varphi = ((\varphi^{-1})^*D_j\varphi))^{\alpha}.

(2) Let PP be a differential operator with degree mm on VV, we know that (φ1)Pφ(\varphi^{-1})^*P\varphi^* is a differential operator on UU. The idea is, for fC(U)f\in \mathscr{C}^{\infty}(U), we first “pull back” ff to a smooth function on VV, apply PP, then “pull back” back to UU. Show that ((φ)1)Pφ((\varphi)^{-1})^{*}P \varphi^* is also degree mm.

Hint: Write Pf=αmaα(s,t)Dαf(s,t)Pf = \sum_{|\alpha| \leq m}a_{\alpha}(s,t)D^{\alpha}f(s,t). Write φ:VU\varphi:V\to U by (uv)=φ((st))=(u(s,t)v(s,t))\begin{pmatrix}u\\v\end{pmatrix} = \varphi(\begin{pmatrix}s\\t\end{pmatrix}) = \begin{pmatrix} u(s,t)\\v(s,t)\end{pmatrix}. Then (1) can be verified directly, and (2) reduce to check the case P=DkP = D_k. By the chain rule, we have Dsφf(s,t)=Ds(f(u(s,t),v(s,t)))=usDuf(u(s,t),v(s,t))+vsDvf(u(s,t),v(s,t))=usφDuf+vsφDvfD_s\varphi^*f(s,t) = D_s(f(u(s,t),v(s,t))) = \frac{\partial u}{\partial s}D_u f(u(s,t),v(s,t))+\frac{\partial v}{\partial s}D_vf(u(s,t),v(s,t)) = \frac{\partial u}{\partial s}\varphi^*D_u f + \frac{\partial v}{\partial s}\varphi^* D_v f.

Problem 4

Let f,gCc(U)f,g\in \mathscr{C}_c^{\infty}(U), complex valued. Let Ufgˉ  dm\int_{U}f\bar{g}\;dm be the inner product. Let P=αmaα(x)DαP = \sum_{|\alpha|\leq m}a_\alpha(x)D^{\alpha}, define its adjoint PP^* by requiring (Pf,g)=(f,Pg)(Pf,g) = (f,P^*g). Calculate PP^* using integration by parts. The PP^* is called the formal adjoint of PP. Note that here it only applies on test functions.