0

So it is well known that if $y$ is a boundary point for a $k$-manifold $M$ in $\Bbb R^n$ then a vector $\vec v$ of $\Bbb R^n$ is tangent to $M$ at $y$ if and only if there exist a parametrized curve $\gamma$ defined in a open neighborhood of $0$ that carries the right or left part of this neighborhood into $M$ such that $$ y=\gamma (0)\,\,\,\text{and}\,\,\,\vec v=\gamma'(0) $$

So I ask if this result is still true for corner points of a $k$-Manifold of class $C^r$ for any $r\ge 1$ with corners that is defined to be a $k$-manifold having some local patch defined in an open set of $\Bbb R^{k-h}\times[0,+\infty)^{h}$ for any $h\le k$.

To follow some reference about well know definitions and elementary results about manifolds: I put here them to show the formalism I use about manifolds; moereover I point out that theese reference (from the text Analysis on Manifolds by James Munkres) are for manifolds with boudary but reading them it is not hard to see that the definition and results can be easily (is this false?) extended to a manifolds with corners. So if you like click here to see it, otherwise (provided you like it!) see my proof attempt.

MY PROOF ATTEMPT

So let be $\gamma$ a parametrized curve defined in a open neighborhood of $0$ that carries the right (or left) part of this neighborhood into $M$ so that $\gamma(0)$ is a corner point for $M$ and we shall prove that that its derivative $\gamma'$ at $\gamma (0)$ is a tangent vector to $M$. So let be $\alpha:U\rightarrow V$ a coordinate patch about $\gamma(0)$ and thus putting $$ y=\alpha(x):=\gamma(0)\,\,\,\text{and}\,\,\,\vec v:=\frac{d}{dt}\gamma(t)|_{t=0}\,\,\,\text{and thus}\,\,\,\vec u:=D\alpha^{-1}(y)\cdot\vec v $$ we observe that $$ \alpha_*(x;\vec u)=(y;D\alpha(x)\cdot\vec u)=(y;D\alpha(x)\cdot D\alpha^{-1}(y)\cdot\vec v)=(y;D(\alpha\circ\alpha^-1)(y)\cdot\vec v)=(y;\vec v) $$ and so we conclude that $\vec v$ is a tangent vector to $M$ at $y$.

So let be $y$ a corner point for a $k$-manifold $M$ in $\Bbb R^n$ and we shall prove there exist a parametrized curve having the properties described above. So is $y$ is a corner point for $M$ coordinate patch $\alpha:U\rightarrow V$ defined in a set $U$ that is open in $\Bbb R^{k-h}\times[0,+\infty)^{h}$ for any $h\le k$ but not in $\Bbb R^k$ so that $$ y=\alpha(x) $$ for some $x\in\text{bd}\Big(\Bbb R^{k-h}\times[0,+\infty)^{h}\Big)$. Now if $(y,\vec v)$ is a tangent vector to $M$ at $y$ then there exist $\vec u\in\Bbb R^k$ such that $\vec v=D\alpha(x)\cdot\vec u$ and moreover since $U$ is open in $\Bbb R^{k-h}\times[0,+\infty)^{h}$ there exist an open cube $C(x,r)$ centred at $x$ and of radius $r>0$ such that. $$ C(x;r)\cap \Bbb R^{k-h}\times[0,+\infty)^{h}\subseteq U $$ and so first we suppose that $h=1$. Well in this case observing that $x+t\vec u\in C(x,r)$ iff $x_i+tu_i\in(x_i-r,x_i+r)$ and so iff $tu_i\in(-r,r)$ let be $$ \epsilon:=\min\Big\{\frac{r}{|u_i|}:i=1,...,k\Big\} $$ so that $$ x+t\vec u\in C(x;r)\cap\Bbb R^{k-h}\times[0,+\infty)^{h} $$ for any $t\in[0,\epsilon)$ or $t\in (-\epsilon,0]$ if $u_k\ge 0$ or $u_k\le 0$ respectively. So observing that $$ D\alpha(x+t\vec u)|_{t=0}=D\alpha(x+0\cdot\vec u)\cdot\vec u=\vec v $$ we conclude the result holds for $h=1$. So unfortunately if $h>1$ then $$ x+t\vec u\notin C(x;r)\cap\Bbb R^{k-h}\times[0,+\infty)^{h} $$ because it could be that $x_{i_1}=x_{i_2}=0$ for $i_1,i_2=1,...,k$ and $u_{i_1}<0$ and $u_{i_2}>0$ so I do not be able to implement the above arguments. So could someone help me, please?

  • 1
    Hint: The claim you are trying to prove is simply false, already for $k=2$ and $M$ equal the positive quadrant in ${\mathbb R}^2$. (Assuming that tangent space to a manifold with corners is defined so that it is a vector space.) I suggest you draw picture and try to find an example. – Moishe Kohan May 16 '21 at 23:52
  • @MoisheKohan Okay, so the statement is false. Anyway is true that the tangent space at a corner point $p$ defined as $$D\alpha(x)\cdot\vec v$$ where $x=\alpha^{-1}(p)$ and $\vec v\in\Bbb R^k$ is independet from the choice of the local patch $\alpha$, right? – Antonio Maria Di Mauro May 19 '21 at 10:05
  • I do not understand your definition but a better one would be as the set of equivalence classes of smooth curves $\alpha: [0,1)\to M$ such that $\alpha(0)=p$, where the equivalence relation is the usual one. This space is not a vector space but has an extra structure making it isomorphic to $R^{k-h}\times [0,\infty)^h$. – Moishe Kohan May 19 '21 at 15:23
  • @MoisheKohan Which is the usual equivalence relation? Sorry, but I do not know it because (it seems) I define the tangent space in a little different way: if you click in the link that I posted in the question above you can see it at the end of the linked page. Anyway (provided I do not tire you) could I ask your assistance here, please? – Antonio Maria Di Mauro May 19 '21 at 15:28
  • The usual definition is that two smooth curves $\alpha, \beta$ such that $\alpha(0)=\beta(0)=p$ are equivalent if for each chart $\phi$ at $p$, $(\phi\circ \alpha)'(0)=(\phi\circ \beta)'(0)$. – Moishe Kohan May 20 '21 at 02:45

0 Answers0