Back
With help from https://en.wikipedia.org/wiki/Frenet–Serret_formulas
Let a curve be defined as $\vec{r}(t) : \mathbb{R} \rightarrow \mathbb{R}^n$
$\vec{r} = \{ r_i(t), 1 \le i \le n \}$
Velocity: $\vec{v} = \frac{\partial \vec{r}}{\partial t}$
Speed: $\sigma = |\vec{v}| = \sqrt{ \underset{j}{\Sigma} \left( \frac{\partial r_j}{\partial t} \right)^2 }$
$\sigma \ge 0$
$\frac{\partial \sigma}{\partial t} = \frac{1}{2 \sigma}$
Arclength: $s(t) := \int_{t_0}^t \sigma(t') dt'$
$\frac{\partial s}{\partial t} = \sigma$
$t = s^{-1}( s(t) )$
Notice the ambuguity that $(s)$ is a parameter (and can be replaced by any other letter) while $s^{-1}(\cdot)$ is the inverse of the arclength function.
Arclength inverse:
$t(s) = s^{-1}( s )$
Reparameterization by arclength: $r_i(s) := r_i(t(s))$
Notice that $r_i(t)$ and $r_i(s)$ are represented identically with the only difference being the parameter variable, however using a $t$ denotes the original function while using a $s$ denotes the arclength parameterization of the function.
Velocity wrt arclength:
$\frac{\partial \vec{r}}{\partial s} = \frac{\partial \vec{r}}{\partial t} \frac{\partial t}{\partial s} = \frac{1}{\sigma} \vec{v}$
un-normalized tangent:
$\vec{\bar{e}}_1 = \frac{\partial \vec{r}}{\partial s}$
un-normalized tangent length:
$|\vec{\bar{e}}_1| = |\frac{\partial \vec{r}}{\partial s}| = \frac{1}{\sigma} |\vec{v}| = \frac{\sigma}{\sigma} = 1$
tangent:
$\vec{e}_1 = unit(\vec{\bar{e}}_1) = \vec{\bar{e}}_1 = \frac{\partial \vec{r}}{\partial s} = \frac{1}{\sigma} \vec{v}$
$\vec{T} = \vec{e}_1$
tangent orthogonality to derivative:
$\vec{T} \cdot \vec{T} = 1$
$\frac{\partial}{\partial s} (\vec{T} \cdot \vec{T}) = 0$
$\frac{\partial \vec{T}}{\partial s} \cdot \vec{T} = 0$
$\frac{\partial^2 \vec{r}}{\partial s^2} \cdot \frac{\partial \vec{r}}{\partial s} = 0$
un-normalized normal:
$\vec{\bar{e}}_2 = \frac{\partial^2 \vec{r}}{\partial s^2} - (\frac{\partial^2 \vec{r}}{\partial s^2} \cdot \vec{e}_1) \vec{e}_1$
$= \frac{\partial^2 \vec{r}}{\partial s^2} - (\frac{\partial^2 \vec{r}}{\partial s^2} \cdot \frac{\partial \vec{r}}{\partial s}) \frac{\partial \vec{r}}{\partial s}$
$= \frac{\partial^2 \vec{r}}{\partial s^2}$
normal:
$\vec{N} = \vec{e}_2$
$= unit( \vec{\bar{e}}_2 )$
$= unit( \frac{\partial^2 \vec{r}}{\partial s^2} )$
$= \frac{
\frac{\partial^2 \vec{r}}{\partial s^2}
}{
\left| \frac{\partial^2 \vec{r}}{\partial s^2} \right|
}$
curvature:
$\chi_1 = \frac{1}{\left|
\frac{\partial^2 \vec{r}}{\partial s^2}
\right|}$
$\kappa = \chi_1$
So $\frac{\partial}{\partial s} \vec{e}_1 = \chi_1 \vec{e}_2$
i.e. $\frac{\partial}{\partial s} \vec{T} = \kappa \vec{N}$
Normal derivative:
$\frac{\partial}{\partial s} \vec{e}_2
= \frac{\partial}{\partial s} \left(
\frac{
\vec{\bar{e}}_2
}{
\left|
\vec{\bar{e}}_2
\right|
}
\right)
$
$ = \frac{
\frac{\partial}{\partial s} \vec{\bar{e}}_2
\left| \vec{\bar{e}}_2 \right|
-
\vec{\bar{e}}_2
\frac{\partial}{\partial s} \left| \vec{\bar{e}}_2 \right|
}{
\left| \vec{\bar{e}}_2 \right|^2
}
$
$ =
\chi_1^2 (
\frac{\partial^3 \vec{r}}{\partial s^3}
\left| \vec{\bar{e}}_2 \right|
-
\frac{\partial^2 \vec{r}}{\partial s^2}
\frac{\partial}{\partial s} \left| \vec{\bar{e}}_2 \right|
)$
Frenet frame:
$\frac{\partial}{\partial s} \left[ \begin{matrix}
\vec{T} \\ \vec{N} \\ \vec{B}
\end{matrix} \right] = \left[ \begin{matrix}
0 & \chi_1 & 0 \\
-\chi_1 & 0 & \chi_2 \\
0 & -\chi_2 & 0 \\
\end{matrix} \right] \left[ \begin{matrix}
\vec{T} \\ \vec{N} \\ \vec{B}
\end{matrix} \right]$
Arclength generalized by coordinate system / line element:
Start in our Cartesian basis:
$\frac{\partial s}{\partial t} = \sqrt{ \underset{i}{\Sigma} \left( \frac{\partial r^i}{\partial t} \right)^2 }$
Apply a change-of-basis:
Let $r^i = {e^i}_j x^j$
$\frac{\partial s}{\partial t} = \sqrt{ \underset{i}{\Sigma} \left( \frac{\partial}{\partial t}( \underset{j}{\Sigma} {e^i}_j x^j ) \right)^2 }$
Assuming the parameterization is independent of the change-of-basis:
$\frac{\partial s}{\partial t} = \sqrt{ \underset{i}{\Sigma} \left( \underset{j}{\Sigma} {e^i}_j \frac{\partial x^j}{\partial t} \right)^2 }$
$\frac{\partial s}{\partial t} = \sqrt{
\underset{i}{\Sigma} \left(
\frac{\partial x^i}{\partial t}
\underset{k}{\Sigma} \left(
{e^k}_i
\underset{j}{\Sigma} \left(
{e^k}_j
\frac{\partial x^j}{\partial t}
\right)
\right)
\right)
}$
$\frac{\partial s}{\partial t} = \sqrt{
\underset{i,j,k}{\Sigma} \left(
{e^k}_i
{e^k}_j
\frac{\partial x^i}{\partial t}
\frac{\partial x^j}{\partial t}
\right)
}$
Let $g_{ij} = \underset{k}{\Sigma} {e^k}_i {e^k}_j$:
$\frac{\partial s}{\partial t} = \sqrt{
\underset{i,j}{\Sigma} \left(
g_{ij}
\frac{\partial x^i}{\partial t}
\frac{\partial x^j}{\partial t}
\right)
}$
As matrices:
$\frac{\partial s}{\partial t} = \sqrt{ | \vec{r} |^2 }$
$\frac{\partial s}{\partial t} = \sqrt{ \textbf{r}^T \textbf{r} }$
$\frac{\partial s}{\partial t} = \sqrt{ (\textbf{E} \textbf{x})^T \textbf{E} \textbf{x} }$
$\frac{\partial s}{\partial t} = \sqrt{ \textbf{x}^T \textbf{E}^T \textbf{E} \textbf{x} }$
Let $\textbf{G} = (g_{ij})$ be defined as $\textbf{G} = \textbf{E}^T \textbf{E}$, so $g_{ij} = \vec{e}_i \cdot \vec{e}_j$, for $\vec{e}_i$ the i'th column vector of $\textbf{E}$.
Now we are looking at a norm weighted by the metric:
$\frac{\partial s}{\partial t} = \sqrt{ \textbf{x}^T \textbf{G} \textbf{x} }$
A concise way of writing this is:
$ds = \sqrt{
\underset{i,j}{\Sigma} \left(
g_{ij}
\frac{\partial x^i}{\partial t}
\frac{\partial x^j}{\partial t}
\right)
} dt$
$ds^2 = g_{ij} \frac{\partial x^i}{\partial t} \frac{\partial x^j}{\partial t} dt^2$
$ds^2 = g_{ij} dx^i dx^j$
Back