3

Let $X = \{(x,y,z) \in R^3: z^4 = x^2+y^2, z \geq 0\}$. First I'm trying to prove that this is not a smooth submanifold of $R^3$. Clearly the problematic point is $(0,0,0)$ however I haven't been able to prove formally that this point fails. For reference I'm trying to do an elementary proof of this, so this should basically be a result of the definition of a submanifold. I tried to asume the existence of a chart $\varphi$ of a neighborhood of $(0,0,0)$ to $R^2$ and with this try to consider the projection $\pi$ from $X$ to $R^2$ and consider the map $\pi \circ \varphi^{-1}$ and use inverse function theorem, but the derivative of this function is not invertible, so I didn't know how to proceed.

Second for the differential structure I took the atlas generated by the chart $\phi(x,y,z) = (x^{1/3},y^{1/3})$. Now with this atlas the inclusion map is differentiable but not smooth, so something like this should work, but I don't know exactly what chart to take. Any help would be greatly appreciated.

H4z3
  • 1,153

2 Answers2

1

A Special Chart

Locally, a chart of $X$ could be $$\begin{align*} \phi \colon X &\to \mathbb{R}^{2} \\ (x,y,z) &\mapsto (u,v), \end{align*}$$ such that $$ x=ue^{-\frac{1}{u^2+v^2}},y=ve^{-\frac{1}{u^2+v^2}}.$$ We define $e^{-\frac{1}{0^2+0^2}}$ to be zero.

To show that this is indeed a homeomorphism, it suffices to prove the following map $$\begin{align*} F \colon \mathbb{R}^{2} &\to \mathbb{R}^{2} \\ (u,v) &\mapsto \exp \left(-\frac{1}{u^{2}+v^{2}} \right)\cdot (u,v) \end{align*}$$ is a homeomorphism. We only need to prove $F^{-1}$ is continuous at the origin.

The map $F^{-1}$ is continuous at $0$ iff $\forall \epsilon>0, \exists \delta >0$, such that $F^{-1}(B_\delta (0))\subset B_\epsilon(0)$. It is equivalent to proving $\forall \epsilon>0, \exists \delta >0$, such that $e^{-\frac{2}{r^2}}r^2<\delta^2\implies r^2<\epsilon^2$. Choosing $\delta^2=\frac{1}{2} \epsilon^2 e^{-\frac{2}{\epsilon^2}}$ will do the job.

A chart of $\mathbb{R}^{3}$ is $$ \begin{align*} \operatorname{id} \colon \mathbb{R}^{3} &\to \mathbb{R}^{3} \\ (x,y,z) &\mapsto (x,y,z). \end{align*} $$

Now, the representation of the inclusion map is $$\begin{align*} \operatorname{id}\circ i \circ \phi^{-1} \colon \phi(X) &\to \mathbb{R}^{3} \\ (u,v) &\mapsto (u\exp\left(-\frac{1}{u^2+v^2}\right),v\exp\left(-\frac{1}{u^2+v^2}\right),\exp\left(-\frac{1}{2(u^2+v^2)}\right)(u^2+v^2)^{1/4}). \end{align*}$$

Smoothness

Now, we will prove that $z=\exp\left(-\frac{1}{2(u^2+v^2)}\right)(u^2+v^2)^{1/4}$ is smooth at $(0,0)$. In fact, the higher-order derivatives of $z$ are all $0$. The idea is the same as taking the n-th derivative of $f(x)=e^{-\frac{1}{x^2}}$ at $0$.

Let's first calculate the first-order derivative of $z$ at the origin.

$$\begin{aligned}\frac{\partial z}{\partial u}\Bigg|_{(0,0)}&=\lim_{u\to 0}\frac{\exp\left(-\frac{1}{2(u^2+0^2)}\right)(u^2+0^2)^{1/4}-0\cdot (0^2+0^2)^{1/4}}{u}\\ &=\pm \lim_{|u|\to 0^+}\frac{\exp\left(-\frac{1}{2u^2}\right)}{\sqrt{|u|}}\\ \left(t=\frac{1}{\sqrt{|u|}}\right):\;&= \pm \lim_{t\to +\infty} \frac{t}{\exp \left({\frac{t^4}{2}}\right)}\\ \left( \text{L'Hôpital's rule} \right):\;&=\pm \lim_{t \to \infty} \frac{1}{2 e^{\frac{t^4}{2}} t^3}\\ &= 0 \end{aligned}$$

Similarly, we have $$\frac{\partial z}{\partial v}\Bigg|_{(0,0)}=0.$$

We can now use mathematical induction to prove $$z^{(m,n)}|_{(0,0)}\equiv \frac{\partial^{m+n} z}{\partial^m u \;\partial^n v}\Bigg|_{(0,0)}=0.$$

When $(u,v)\neq (0,0)$, the $(m,n)$-th derivative has the form $$z^{(m,n)}\equiv \frac{\partial^{m+n} z}{\partial^m u \;\partial^n v}=\exp\left(-\frac{1}{2(u^2+v^2)}\right)\frac{P(u,v)}{(u^2+v^2)^{\frac{s}{4}}},$$

where $s$ is an integer and $P(u,v)$ is some polynomial of $u,v$.

For example, $$z^{(2,1)}\equiv \frac{\partial^{2+1} z}{\partial^2 u \;\partial^1 v}=\frac{\exp\left(-\frac{1}{2 \left(u^2+v^2\right)}\right)}{(u^2+v^2)^{23/4}} \cdot \frac{v}{8} \left(15 u^8+3 u^6 \left(13 v^2+34\right)+u^4 \left(27 v^4+180 v^2-76\right)+u^2 \left(-3 v^6+54 v^4-68 v^2+8\right)-6 v^8-24 v^6+8 v^4\right)$$

Given $z^{(m,n)}|_{(0,0)}=0$, we have $$\begin{aligned}z^{(m+1,n)}|_{(0,0)}&=\lim_{u\to 0}\frac{z^{(m,n)}(u,0)-0}{u}\\ &=\lim_{u\to 0} \exp\left(-\frac{1}{2(u^2+0^2)}\right)\frac{P(u,0)}{u\cdot (u^2+0^2)^{\frac{s}{4}}}\\ \left(t=\frac{1}{\sqrt{|u|}}\right):\;&= \pm P(0,0)\cdot \lim_{t \to \infty}\frac{t^{s+2}}{\exp \left({\frac{t^4}{2}}\right)}\\\left(\text{L'Hôpital's rule} \right):\;&=\pm P(0,0)\cdot \lim_{t \to \infty}\frac{(s+2)!}{\exp \left({\frac{t^4}{2}}\right)p(t)}\\&=0,\end{aligned}$$ where $p(t)$ is just a polynomial. Similarly, you can prove $z^{(m,n+1)}|_{(0,0)}=0$.

Therefore, by the principle of induction, $z^{(m,n)}|_{(0,0)}=0$ for all integers $m,n$, and the $z(u,v)$ is smooth at the origin.

Intuitions

Next, we will interpret our results, and gain some intuition.

First, let's see what $z^4=x^2+y^2$ looks like: enter image description here

Clearly, the origin is a critical point.

Let's see what $z^4=\exp\left(-\frac{2}{u^2+v^2}\right)(u^2+v^2)$ looks like: enter image description here

The origin is now smooth.

You may wonder why these two charts are so different; this is because our homeomorphism $F$ is not a diffeomorphism.

To make this point more precise, let's calculate the Jacobian of $F$ at the origin: $$ JF|_{(0,0)}= \begin{pmatrix} 0 & 0 \\ 0 & 0 \\ \end{pmatrix} ,$$ so $JF^{-1}|_{(0,0)}$ doesn't exist. That is to say, $F^{-1}$ is only continuous, but not differentiable.

However, any point other than $(0,0)$ is differentiable: $$ JF|_{(u,v)\neq (0,0)}= \frac{e^{-\frac{1}{u^2+v^2}}}{(u^2 + v^2)^2}\begin{pmatrix} u^4+2 u^2 \left(v^2+1\right)+v^4 & 2 u v \\ 2 u v & u^4+2 v^2( u^2+1)+v^4 \end{pmatrix}, $$ and $$\det JF|_{(u,v)\neq (0,0)}= \frac{e^{-\frac{2}{u^2+v^2}} \left(u^2+v^2+2\right)}{u^2 +v^2}>0.$$

By the Inverse function theorem, $F^{-1}|_{(u,v)\neq (0,0)}$ is smooth.

TRZ
  • 90
  • I do like the answer but I'm still looking for a way to define an atlas for which the inclusion map is smooth – H4z3 Jun 08 '24 at 07:26
  • I know that you can define such atlas it's just that I don't know how, the atlas will probably have only one chart, and like I said in my answer you can define an atlas in which the inclusion is differentiable, I'm just looking for one in which is smooth – H4z3 Jun 08 '24 at 07:38
  • I have edited my answer, I think it will work now. The problem with $\phi(x,y,z) = (x^{1/3},y^{1/3})$ is that when $x\to 0,y\to 0$, our $z$ is of order $r^{3/2}$, but you need it to be $z\sim r^{\infty}$ to be smooth at the origin. Then $e^{-1/r^2}$ is a good factor to achieve that goal. – TRZ Jun 08 '24 at 10:33
  • Ok thanks I think that something like this is what I was looking for. Just one question, when you get $id \circ i \circ \phi^{-1}(u,v) = (ue^{-1/{u^2+v^2}},ve^{-1/{u^2+v^2}},e^{-1/2{u^2+v^2}}(u^2+v^2)^{1/4}$, why is $e^{-1/2{u^2+v^2}}(u^2+v^2)^{1/4}$ smooth? Also I noticed that you deleted your argument for which $X$ was not a submanifold, what was this argument again? – H4z3 Jun 08 '24 at 17:43
  • I don't think your answer works because it doesn't look like the function I just pointed out can be smooth – H4z3 Jun 08 '24 at 18:17
  • I have added the proof of the smoothness of $z$ at $(0,0)$ and some intuition. My argument that $X$ was not a submanifold only holds in the standard $(x,y)$ chart situation. However, it is a submanifold in this $(u,v)$ chart we have constructed. These two charts are not compatible. Hope everything is clear now. – TRZ Jun 09 '24 at 05:32
  • I really appreciate the answer thanks! – H4z3 Jun 09 '24 at 14:37
0

I assume you want to show that $X$ has no topology and smooth structure for which the inclusion $X\rightarrow \mathbf{R}^3$ is a smooth immersion, i.e. it is not an immersed submanifold (there is also the notion of an embedded submanifold).

It is really best not to think about atlases and smooth structures for such a problem. Graph a cross-section $C$ of $X$, say $z^4 = x^2$, $z\ge 0$, in the $xz$-plane, and you may gain more intuition. There is a cusp at $(0,0)$. If the inclusion were an immersion, for each tangent vector $v$ at $(0,0)$, we would be able to find a smooth curve $\gamma$, $\gamma(0)=(0,0)$, with image in $C$ so that $\gamma'(0)=v$. It is fairly easy to see this is not the case. For the original problem, you could find a contradiction by proving the tangent space to $X$ at $(0,0,0)$ is one-dimensional.

  • I do like the answer but I'm still looking for a way to define an atlas for which the inclusion map is smooth – H4z3 Jun 08 '24 at 07:26
  • You probably want a homeomorphism $X\rightarrow \mathbf{R}^2$ with smooth inverse $f:\mathbf{R}^2 \rightarrow \mathbf{R}^3$. Multiplying the projection $\pi$ by some power of $r=\sqrt{x^2+y^2}$ should do the trick. I will let you figure out the details. For the cross-section $C$ I think $p(x,z)=|x|^{1/4}$ works. – C. icosahedra Jun 08 '24 at 07:56
  • Hey but what should be the chart? Like I said I was struggling with finding the particular chart – H4z3 Jun 08 '24 at 08:13
  • Any homeomorphism $\varphi:X\rightarrow Y$ between $X$ a topological space and $Y$ a smooth manifold provides $X$ with a unique smooth structure so that $\varphi$ is a diffeomorphism. In this case $Y=\mathbf{R}^2$ and the homeomorphism is itself a smooth chart. – C. icosahedra Jun 08 '24 at 08:20
  • Yes but I'm looking a homeomorphism for which the inclusion map is smooth. Like I said I found one for which is differentiable but not smooth – H4z3 Jun 08 '24 at 08:28
  • The point is with the right choice of homeomorphism, the resulting smooth structure will make the inclusion smooth – C. icosahedra Jun 08 '24 at 08:31
  • Yeah and I was asking for the particular homeomorphism that would work – H4z3 Jun 08 '24 at 23:30