Let \(f\) be a curve on the interval \([a,b]\) with the property that \(f'\) is continuous on \([a,b]\) as well.
Subdivide \([a,b]\) into smaller intervals of length \(\Delta x\text{.}\) Then, the length of each subdivision, \(L_i\text{,}\) can be approximated by the straight-line distance between the points \(f(x_{i-1})\) and \(f(x_i)\text{:}\)
\begin{equation*}
L_i = \sqrt{(x_i-x_{i-1})^2 + (y_i-y_{i-1})^2}
\end{equation*}
Since \(x_i-x_{i-1} = \Delta x\text{,}\) we can rewrite \(L_i\) as
\begin{equation*}
L_i = \sqrt{(\Delta x)^2 + (\Delta y_i)^2}
\end{equation*}
where \(\Delta y_i\) represents \(y_i-y_{i-1}\text{.}\)
From the Mean Value Theorem (see
Section 3.2), we know that
\begin{equation*}
f'(x_i) = \dfrac{f(x_i)-f(x_{i-1})}{x_i-x_{i-1}}
\end{equation*}
which can be rearranged to say
\begin{equation*}
f(x_i) - f(x_{i-1}) = f'(x_i)\cdot (x_i-x_{i-1})
\end{equation*}
Since \(f(x_i) - f(x_{i-1}) = \Delta y_i\) because it precisely measures the change in the output values and \(x_i-x_{i-1} = \Delta x\text{,}\) we can rewrite the Mean Value Theorem expression as
\begin{equation*}
\Delta y_i = f'(x_i)\cdot \Delta x
\end{equation*}
Replacing this in the expression for \(L_i\text{,}\) we have
\begin{equation*}
L_i = \sqrt{(\Delta x)^2 + (f'(x_i)\Delta x)^2}
\end{equation*}
Now we may simplify:
\begin{align*}
L_i \amp = \sqrt{(\Delta x)^2 + (f'(x_i)\Delta x)^2}\\
\amp = \sqrt{(\Delta x)^2[1 + (f(x_i))^2]} \\
\amp = \sqrt{1 + (f(x_i))^2)}\cdot \Delta x
\end{align*}
The approximation for the total length of the curve is given by
\begin{align*}
L \amp \approx \sum_{i=1}^n L_i\\
\amp \approx \sum_{i=1}^n \sqrt{1 + (f(x_i))^2)}\cdot \Delta x
\end{align*}
Taking the limit will gives an expression for the precise length of the curve:
\begin{align*}
L \amp = \lim_{n\to\infty}\sum_{i=1}^n \sqrt{1 + (f(x_i))^2)}\cdot \Delta x\\
\amp = \int_a^b \sqrt{1 + (f'(x))^2}\, dx
\end{align*}