I'm having trouble deriving arc length of functions. I'm using a vector-oriented approach, if that's alright. I'm doing this because I don't understand the $1+(f')^2$ part of the standard formula, and I don't like using formulae I don't understand.

So far I have this: Let $d(\vec{u},\vec{v})$ be the Euclidean distance function between two vectors $\vec{u}$ and $\vec{v}$, defined by $$d(\vec{u},\vec{v})=\left \| \vec{v}-\vec{u} \right \|$$ Let $f(x)$ be any "reasonable" function. (By "reasonable", I mean continuously differentiable on the closed interval $[a,b]$, with no singularities, crazy fractal stuff, etc. I think the proof works as long as it's continuous, actually. I could be wrong.) To obtain the arc length of $f(x)$ on the interval $[a,b]$, we have to rectify the curve generated by $f(x)$. We add up the distances between infinitesimal parts of $f$: $$\int_{a}^{b}d(f(x),f(x+h))dx$$

We are working in 2D, so here's the explicit formula for $d(\vec{u},\vec{v})$. Let $\vec{u}=\left \langle x_1,y_1 \right \rangle$ and $\vec{v}=\left \langle x_2,y_2 \right \rangle$ $$\vec{v}-\vec{u}=\left \langle x_2-x_1,y_2-y_1 \right \rangle$$ $$d(\vec{u},\vec{v})=\sqrt{(x_2-x_1)^2+(y_2-y_1)^2}$$ Now, let's substitute $\vec{u}=\left \langle a,f(a) \right \rangle$ and $\vec{v}=\left \langle a+h,f(a+h) \right \rangle$ where $a$ is some number on $f$ and $h$ is a small number (infinitesimal, you know?). $$d(\vec{u},\vec{v})=\lim_{h \to 0}\sqrt{((a+h)-(a))^2+(f(a+h)-f(a))^2}$$

That's where I hit a hitch. I'm not sure how to get rid of that h and replace it with a derivative or something more practical. Oh, and $((a+h)-a)^2=h^2$, right? $h$ is infinitesimal. Is squaring it an issue? Or does it become so small so as to be considered nothing?


If you multiply and divide by $h$ outside your square root, it becomes

\begin{align*} h \sqrt{\frac{\big((a + h) - a\big)^2}{h^2} + \frac{\big(f(a + h) - f(a)\big)^2}{h^2}} &= h \sqrt{1 + \left(\frac{f(a + h) - f(a)}{h}\right)^2} \end{align*}

Now if you partition your interval into $n$ little bits of length $h$ (replacing $h$ by the more suggest $\delta x$), and add up all these approximate lengths, you get

$$\sum_{i = 1}^n \sqrt{1 + \left(\frac{f(x_i + \Delta x) - f(x_i)}{h}\right)^2} \Delta x$$

Now you should recognize this as a Riemann sum, and the term inside the square root will converge to $\sqrt{1 + f'(x)^2}$ if you have a "nice" function.

  • $\begingroup$ "Multiply and divide by $h$ outside [the] square root"? I'm confused. $\endgroup$ – Aieou Dec 31 '15 at 8:47
  • $\begingroup$ @Aieou Write $h/h$, and move the denominator inside the square root and leave the numerator as is. $\endgroup$ – user296602 Dec 31 '15 at 8:47
  • $\begingroup$ Okay, let me see if I got this right. $\frac{h}{h}=1$, so obviously it's "free": multiplying by it does not actually change the equation. Then, as the next step: $\frac{h}{h}=\frac{h}{1}\frac{1}{h}$ Then we multiply the second half into the square root. This is the part that confuses me. Let's define $p$ as all of the stuff currently under the square root, for brevity's sake. You're telling me that $$\frac{1}{h}\sqrt{p}=\sqrt{\frac{p}{h^2}}$$ [comment continued] $\endgroup$ – Aieou Dec 31 '15 at 9:02
  • $\begingroup$ After some testing, this does appear to be true, and everything works out from there. $\frac{h^2}{h^2}$ is definitely 1, and the term with $f$ in it is definitely the definition of the derivative. I guess what confuses me is how despite that multiplying by $\frac{h}{h}$ is definitely valid and doesn't really change anything, it somehow makes a 1 appear out of "nowhere" and turns that difference in a function into a derivative. Very strange. $\endgroup$ – Aieou Dec 31 '15 at 9:02
  • $\begingroup$ @Aieou It's just a different phrasing of "factor an $h^2$ from each term inside the square root." I find it more natural to introduce it as I did in the answer, but they are equivalent ideas. $\endgroup$ – user296602 Dec 31 '15 at 9:03

An answer to a related question can be found here

From that you will see how to arrive at the well known result that the length of a parametric curve $c(t)= (c_1(t), \dots, c_n(t))$ is given by $$\ell(c) =\int_a^b ||c^\prime(t)|| \, dt$$

The special case you are considering in the heading is the case where the curve is given by a graph, i.e. $c(t) = (t, f(t))$. It is easy to see that in this case the previous formula reduces to


  • $\begingroup$ Wouldn't it be $c(t)=(t,f(t))$? $\endgroup$ – Aieou Dec 31 '15 at 9:18
  • $\begingroup$ @Aieou yes, of course. Thank you! $\endgroup$ – Thomas Dec 31 '15 at 9:26

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.