5

I'm reading Geiges' notes. (https://arxiv.org/pdf/math/0307242.pdf) In the proof of Theorem 2.44 on page 17, the existence of the contact version Darboux coordinate is reduced to solving $H_t$ for each $t$, the PDE near the origin of $\mathbb{R}^{2n+1}$ $$\dot{\alpha}_t (R_{\alpha_t})+dH_t(R_{\alpha_t} )= 0$$ where $\alpha_t$ is a $1$-parameter family of contact forms and $R_{\alpha_t}$ is the corresponding reeb vector field. And he said that this equation always has a solution by integration if the neighborhood is small enough so that $R_{\alpha_t}$ has no closed orbit.

My question is why this is obvious? What I know is that this equation is a quasilinear first order PDE and can possibly be solved by the method of characteristics. But I can't find a reference that contains a clear statement when this kind of equation can be solved.

Thank you.

Chris Kuo
  • 1,849
  • 11
  • 14

2 Answers2

3

First, let us prove the following general result:

Proposition. Let $M$ be a manifold and $p\in M$, let $X$ be a vector field on $M$ and let $g\colon M\rightarrow\mathbb{R}$ smooth. If $X(p)\neq 0$, then there exists $U$ an open neighborhood of $p$ in $M$ and $f\colon U\rightarrow\mathbb{R}$ smooth such that: $$\mathrm{d}f(X)=g_{\vert U}.$$ Furthermore, if $g(p)=0$, one may assume that $f(p)=0$ and $\mathrm{d}f_p=0$.

Proof. Using the straightening theorem, there exists $(U,\phi)$ a chart of $M$ around $p$ such that $\phi_*X=\frac{\partial}{\partial x_1}$. Furthermore, one may assume that for $(x_1,\ldots,x_n)\in\phi(U)$, one has the following property: $$s\in[\min(0,x_1),\max(0,x_1)]\Rightarrow(s,x_2,\ldots,x_n)\in\phi(U)\tag{$\star$}.$$ Therefore, one can define a smooth map $F\colon\phi(U)\rightarrow\mathbb{R}$ by the following formula: $$F(x_1,\ldots,x_n):=\int_{0}^{x_1}g(\phi^{-1}(s,x_2,\ldots,x_n))\,\mathrm{d}s.$$ Notice that one has $F(0)=0$ and $\frac{\partial F}{\partial x_1}=g\circ\phi^{-1}$.

In this section, assume that $g(p)=0$, then there exists constants $a_2,\ldots,a_n$ such that $\mathrm{d}F_0=\sum\limits_{i=2}^na_i\mathrm{d}x_i$ and let us define $\overline{F}\colon\phi(U)\rightarrow\mathbb{R}$ in the following fashion: $$\widetilde{F}(x_1,\ldots,x_n):=F(x_1,\ldots,x_n)-\sum_{i=2}^na_ix_i.$$ Notice that $\widetilde{F}(0)=0$, $d\widetilde{F}_0=0$ and $\frac{\partial\widetilde{F}}{\partial x_1}=g\circ\phi^{-1}$, so that one can assume that $F(0)=0$ and $\mathrm{d}F_0=0$.

Finally, with $f=F\circ\phi$, using the chain rule, for all $x\in U$, one has: $$\mathrm{d}f_x(X(x))=(\mathrm{d}F_{\phi(x)}\circ T_x\phi)(X(x))=\mathrm{d}F_{\phi(x)}\left(\frac{\partial}{\partial x_1}_{\big\vert\phi(x)}\right)=\frac{\partial F}{\partial x_1}_{\big\vert\phi(x)}=g(x).$$ In addition, if $g(p)=0$, then one has $\mathrm{d}f_p=\mathrm{d}F_0\circ T_p\phi=0$. Whence the result. $\Box$

Remark. When I say that $(U,\phi)$ a chart aroud $p$, I mean $\phi(p)=0$.

Remark. Let $\|\cdot\|$ be the product norm on $\mathbb{R}^n$, since $0\in\phi(U)$ is open, there exists $\varepsilon>0$, s.t. $B(0,\varepsilon)\subset U$. Let $x\in B(0,\varepsilon)$ and $s\in[\min(0,x_1),\max(0,x_1)]$, then notice that $|s|\leqslant|x_1|$, so that one has: $$(s,x_2,\ldots,x_n)\in B(0,\varepsilon).$$ Therefore, shrinking $U$ to $\phi^{-1}(B(0,\varepsilon))$ establishes the technical assumption $(\star)$.


For all $t\in[0,1]$, since $R_{\alpha_t}(0)\neq 0$ (for example: $R_{\alpha_t}(0)\not\in\xi_0$) applying the result to the map $-\dot{\alpha_t}(R_{\alpha_t})$, there exists $U_t$ an open neighborhood of $0$ in $\mathbb{R}^{2n+1}$ and a map $H_t\colon U_t\rightarrow\mathbb{R}$ such that: $$\dot{\alpha_t}(R_{\alpha_t})+\mathrm{d}H_t(R_{\alpha_t})=0.$$ To conclude, one has to see that there exists a single neighborhood on which all $H_t$, $t\in[0,1]$ are defined. For all $t\in[0,1]$, let us define the following time of existence: $$\varepsilon_t:=\sup\{\varepsilon>0\textrm{ s.t. }B(0,\varepsilon)\subset U_t\},$$ then $t\mapsto\varepsilon_t$ is a lower semicontinuous function and by compactness of $[0,1]$, one has $\varepsilon:=\inf\limits_{t\in[0,1]}\varepsilon_t>0$. Finally, for all $t\in[0,1]$, $H_t$ is defined on $B(0,\varepsilon)$.

fresh
  • 343
  • 1
  • 10
C. Falcon
  • 19,553
0

My answer is probably late... But I would like to know it there is any error in what I did.

I use the following theorem: Let $M$ be a smooth manifold. Suppose we are given an embedded hypersurface $S\subseteq M$, a smooth vector field $V$ on $M$, that is nowhere tangent to $S$, smooth functions $b,g:M\to\mathbb{R}$ and $\varphi:S\to\mathbb{R}$. Then for some neighbourhood $U$ of $S$ in $M$, there exists a unique smooth solution $u:U\to\mathbb{R}$ to the Cauchy problem: $$Vu+bu=g$$ $$u\vert_S=\varphi$$ The proof of this statement can be found in Introduction to Manifolds by the professor Lee.

I've also read the same notes so I've tried to avoid mentioning those steps that are not part of your doubt ...

Let $p_0\in M$ and $\varphi=(x^1,\cdots,x^n,y^1,\cdots,y^n,z)$ the smooth coordinates mentioned in the beginning of the proof.

  1. The manifold is $J\times U_1$, here $J$ is a bounded and open interval containing $[0,1]$, $U_1$ a neighbourhood of $p_0$, recall that $U_1$ is chosen in such a way $\alpha_s$ is a contact form on $U_1$, thus the smoothness of $\alpha_t$ and the condition at $p_0$ given in the notes, implies that $R(t,p)=R_{\alpha_t}\vert_{p}$ is a smooth time dependent vector field, we can add the condition $Rz\neq 0$ to the problem of finding $J$ and $U_1$.

  2. $g(t,p)= \alpha_0\vert_{p}(R(t,p))-\alpha_1\vert_{p}(R(t,p))$ , $b,\varphi = 0$ and $V=\hat{R}$, where $\hat{R}$ is the smooth vector field defined by just changing $\vert_{p}$ to $\vert_{(t,p)}$ in the expression of $R(t,p)$, i.e. if $R(t,p)=a(t,p)\partial_z\vert_{p}$ then $\hat{R}(t,p)=a(t,p)\partial_z\vert_{(t,p)}$. Moreover, $S=z^{-1}(0)$ is a hypersurface in $J\times U_1$ s.t. $\hat{R}$ is not tangent in any point.

By the theorem, we know there exists $u$ defined on some neighbourhood of $S$ s.t. $\hat{R}u=g$ and $u\vert_S=0$. So, by the openness of the neighbourhood of $S$ and compactness of any bounded interval, we can assume that $u$ is defined on $I\times U_2$, here $[0,1]\subseteq I\subseteq J$ and $U_2\subseteq U_1$, thus we define $H_t:U_2\to\mathbb{R}$ s.t. $H_t(p)=u(t,p)$, and for obvious reasons $\{H_t\}_{t\in I}$ is a smooth family of smooth functions.

We also got $H_t(p_0)=0$($(t,p_0)\in S$) and $dH_t\vert_{p_0}=0$($du_{(t,p_0)}=0$ since $g(t,p_0)=0$).

The reason I use that theorem it's because we need to prove the smoothness of the family $\{H_t\}$, since $\{X_t=H_tR_{\alpha_t}+Y_t\}$ needs to be a smooth family in order to define a smooth flow.

no name
  • 11
  • 3