I have a smooth $k$-surface $S\subset \mathbb{R}^n$ and two charts $\varphi_1:I_t^n\to U_1\subset S$, $\varphi_2:I_\tau^n\to U_2\subset S$ with $U_1\cap U_2\neq \emptyset$ ($I^n$ is the unit open cube in $\mathbb{R}^n$). Picked a point $\underbrace{x_0}_{\in U_1\cap U_2}=\varphi_1(\underbrace{t_0}_{\in I_t^n})=\varphi_2(\underbrace{\tau_0}_{\in I_\tau^n})$ in such intersection, we know that if the two charts have positive transitions at $x_0$, then the induced two frames in $x_0$ by the two charts have the same orientation (and viceversa). Matematically this statement is written as:

$\text{sign}\det \left([\varphi_2^{-1}\circ \varphi_1]'(t_0)\right)=\text{sign}\det \left([\varphi_1^{-1}\circ \varphi_2]'(\tau_0)\right)>0\iff$ the two frames $F_{\varphi_1}=\{\mathbf{e}_1,...,\mathbf{e}_n\}$ and $F_{\varphi_2}=\{\mathbf{b}_1,...,\mathbf{b}_n\}$ induced by the two charts in $x_0$ are such that $\text{sign}\det \left(M_{F_{\varphi_1}\to F_{\varphi_2}}\right)=\text{sign}\det \left(M_{F_{\varphi_2}\to F_{\varphi_1}}\right)>0$.

I remind that:

  1. $\left\{\begin{matrix} \mathbf{b}_1=a_{11}\mathbf{e}_1+...+a_{1n}\mathbf{e}_n \\ ...\\ \mathbf{b}_n=a_{n1}\mathbf{e}_1+...+a_{nn}\mathbf{e}_n \end{matrix}\right.\iff M_{F_{\varphi_1}\to F_{\varphi_2}}=\left(\begin{matrix}a_{11} & ...&a_{1n}\\ ...&...&...\\ a_{n1} & ...&a_{nn}\end{matrix}\right)$;
  2. $M_{F_{\varphi_2}\to F_{\varphi_1}}=M_{F_{\varphi_1}\to F_{\varphi_2}}^{-1}$;
  3. $\mathbf{e}_i=\varphi_1'(t_0)\cdot \left(\begin{matrix}0_1\\...\\1_i\\...\\0_n\end{matrix}\right)$, $i=1,...,n$;
  4. $\mathbf{b}_i=\varphi_2'(\tau_0)\cdot \left(\begin{matrix}0_1\\...\\1_i\\...\\0_n\end{matrix}\right)$, $i=1,...,n$.

Now, my question...

I would prove also the following proposition for the tangent space at $S$ in $x_0$ (namely $TS_{x_0}$):

The two charts have positive transitions at $x_0\iff$ the induced two frames for $TS_{x_0}$ by the two charts have the same orientation.

My book (Zorich, Mathematical Analysis II, 1st ed., Page 173) says that it's true but I can't find a plausible reason to prove it.


(I believe you want $I^k$, not $I^n$ and all the indexes ranging from $1,\ldots, k$.)

Claim: The matrix $M_{F_{\varphi_1}\to F_{\varphi_2}}$ and the matrix $[\varphi_1^{-1}\circ \varphi_2]'(t_0)$ are transposes of each other.

Proof: Consider the image of the vector $\left(\begin{matrix}0_1\\...\\1_i\\...\\0_k\end{matrix}\right)$ under $[\varphi_1^{-1}\circ \varphi_2]'$. The idea of the proof is that by chain rule, we can first map it by $\varphi_2'(\tau_0)$ (and it goes to $\mathbf{b}_i$), and then take the resulting vector


and map it by $(\varphi_1^{-1})'(x_0)$ considered as a map from $TS_{x_0}$ to $TI^k_{\tau_0}$, which is the inverse of $(\varphi_1)'(t_0)$ (considered as map to $TS_{x_0}$), and so sends it to $\left(\begin{matrix}a_{i1}\\...\\a_{ik}\end{matrix}\right)$ (since it's linear and sends each $\mathbf{e}_j$ to $\left(\begin{matrix}0_1\\...\\1_j\\...\\0_k\end{matrix}\right)$).

Thus the matrix of $[\varphi_1^{-1}\circ \varphi_2]'$ has $i$th column equal to $\left(\begin{matrix}a_{i1}\\...\\a_{ik}\end{matrix}\right)$, proving the claim.

This is a rigorous argument once one knows the appropriate chain rule. To avoid this more advanced version of chain rule, one can argue as follows:

Extend each $\phi_1$ and $\phi_2$ to maps $\Phi_1$ and $\Phi_2$ from $I^n \to \mathbb{R}^n$ (using Proposition on page 162). Moreover, extend $(\mathbf{e_1},\ldots, \mathbf{e_k})$ to a basis or $\mathbb{R}^n$. Then we write $\Phi_1'(t_0)$ using the standard basis on $TI^k_{t_0}$ and this newly constructed basis on $\mathbb{R}^n$. We have that $\Phi_1'(t_0)$ is block upper-triangular, with a $k$ by $k$ upper left block being identity, corresponding to the fact that

$$[\Phi_1'(t_0)](\left(\begin{matrix}0_1\\...\\1_j\\...\\0_n\end{matrix}\right))=[\phi_1'(t_0)](\left(\begin{matrix}0_1\\...\\1_j\\...\\0_k\end{matrix}\right))=\mathbf{e}_j$$ for $j=1, \ldots, k$.

Similarly, $\Phi_2'(\tau_0)$ is block upper-triangular, with a $k$ by $k$ upper left block being $\left(M_{F_{\varphi_1}\to F_{\varphi_2}}\right)^T$, corresponding to

$$[\Phi_2'(\tau_0)](\left(\begin{matrix}0_1\\...\\1_j\\...\\0_n\end{matrix}\right))=[\phi_2'(\tau_0)](\left(\begin{matrix}0_1\\...\\1_j\\...\\0_k\end{matrix}\right))=\mathbf{b}_j=\sum a_{ji}\mathbf{e}_i$$ for $j=1, \ldots, k$.

Then, by chain rule applied to $[\Phi_1^{-1} \cdot \Phi_2]$ we have that the $n$ by $n$ matrix $[\Phi_1^{-1} \cdot \Phi_2 (\tau)]'$ is composition of two $n$ by $n$ matrices $ [\Phi_2'] (\tau_0)$ and $[\Phi_1^{-1}]'(x_0)$. Both of these are block-upper-triangular, with the $k$ by $k$ left upper block of $ [\Phi_2'] (\tau_0)$ equal to $\left(M_{F_{\varphi_1}\to F_{\varphi_2}}\right)^T$ and the $k$ by $k$ left upper block of $[\Phi_1^{-1}]'(x_0)$ equal to inverse of identity, i.e. identity. Then we conclude that the $k$ by $k$ left upper block of $[\Phi_1^{-1} \cdot \Phi_2]$, being product of these two $k$ by $k$ blocks, is also just $\left(M_{F_{\varphi_1}\to F_{\varphi_2}}\right)^T$.

But this block is of course just $[\varphi_1^{-1}\circ \varphi_2]'(\tau_0)$, since the map $[\Phi_1^{-1} \cdot \Phi_2]$ restricted to $I^k$ is $[\varphi_1^{-1}\circ \varphi_2]$.

This reproves the claim (using onluy chain rule for maps between open subsets of $\mathbb{R}^n$).

Now from the claim the result follows, since transpose matrices have the same determinant.

  • $\begingroup$ No, the matrix $[\varphi_1^{-1}\circ\varphi_2]'(t_0)$ has dimension $n\times n$ and the matrix $ M_{F_{\varphi_1}\to F_{\varphi_2}} $ has dimension $k\times k$. In fact, the frame generated by a chart at a point is composed by $n$ linear independent vectors, but only $k$ of these $n$ are in (generates) $TS_{x_0}$. $\endgroup$
    – Nameless
    Sep 2 '20 at 8:41
  • $\begingroup$ ...at least, this is what I have understood from what I have studied, of course I can be wrong... $\endgroup$
    – Nameless
    Sep 2 '20 at 8:49
  • $\begingroup$ Usual definition of a chart says it is a homeomorphism to (or, sometimes, from) a piece of a surface/manifold. As a homeomorphism, it better map between things that have the same dimension, and you are working with a k-surface. I believe Zorich is using the same definition. $\endgroup$
    – Max
    Sep 2 '20 at 12:41
  • $\begingroup$ There could be differences in definitions, but any reasonable definition would hard to square away with what you have written in the question (i.e. $\varphi_1:I_t^n\to U_1\subset S$ etc.) without modifying something. $\endgroup$
    – Max
    Sep 2 '20 at 12:43
  • $\begingroup$ Zorich uses the following def. for smooth k-surfaces: A set $S\subset\mathbb{R}^n$ is a smooth k-surface if for each point $x\in S$ there exists a diffeomorphism $\varphi:I^n\to U(x)$ such that every point of $U(x)$ is described by $(t_1,...,t_k,0,...,0)$ under $\varphi$. In other words, the diffeomorphism is between points of $\mathbb{R}^n$ and the points of the surface have the property that they are the image of points in $I^n$ with the last n-k coordinates equals to 0. $\endgroup$
    – Nameless
    Sep 2 '20 at 13:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.