# How do I calculate the length of a curve given in parameterized form?

I am given a function describing a curve:

$f(t) = (f_1(t), f_2(t))',\quad t \in \mathbb{R},\quad f_1, f_2: \ \mathbb{R} \rightarrow \mathbb{R}\ .$

How would I calculate the length of that curve corresponding to a given $t$-interval $[a, b]$?

-
Looks like it'd take a miracle to get $x$ from $r$ with known elementary functions. If there's some overarching problem you're working on that lurks behind all this write-up, I urge you to present it clearly if you want meaningful responses. – anon Jul 25 '11 at 14:58
The lower part is pretty much my problem. I have a curve defined by two functions $f_1$ and $f_2$ and I need to work with that curve, e.g. determine its length in a given interval. – Maria Gordova Jul 25 '11 at 15:05
It's utterly hopeless to get a general closed form solution for $x$ in terms of the variable $r$. But if $r$ is kind of large, like $\gt 6$, the exponential makes virtually no difference. Without much trouble you can find excellent approximation for $x$ in terms of $r$. It is probably best to break up range into say $3$ parts, find different excellent approximating formulas over each part. – André Nicolas Jul 25 '11 at 15:16
@Maria Gordova: Perhaps you might want to take a look at arclength? If you use the formula for parametric curves then you don't need to bother with finding a new $g$ function. Also, in English, we might say we "invert" a function if we go from $r=f(x)$ to $x=f^{-1}(r)$, but for mathematical expressions in general, saying you're "manipulating" or "rearranging" it is okay. – anon Jul 25 '11 at 15:24
It would help if you could tell us what you are trying to do and where the problem came from. If you are doing this for a computer program or something similar, a numerical approximation might do well, or maybe you could change your functions slightly for something similar which will be possible to invert. Finding a symbolic answer is in general impossible or at least difficult, so if you don't need that you can get away much cheaper. – Samuel Jul 25 '11 at 15:41

The required length is given by the formula $$L(\gamma|a,b)\ =\ \int_a^b\sqrt{\dot f_1^2(t)+\dot f_2^2(t)}\ dt\ .$$ There are two possibilities:
${\it Either}$ you find an elementary primitive $\Sigma(t)$ of the function $\sigma(t):=\sqrt{\dot f_1^2(t)+\dot f_2^2(t)}$. In this case the length is simply $=\Sigma(b)-\Sigma(a)$. Examples for such $\gamma$ are, e.g., parabolas, circles, logarithmic spirals, but not ellipses.
${\it Or}$ you have to resort to a numerical procedure. The simplest is to put $t_k:=a+k {b-a \over N}$ $\ (0\leq k\leq N)$ for some large $N$ and compute the following approximation: $$L(\gamma|a,b)\ \dot=\ {b-a \over N}\Bigl({1\over 2}\bigl(\sigma(t_0)+\sigma(t_N)\bigr)+\sum_{k=1}^{N-1} \sigma(t_k)\Bigr)\ .$$
I can't follow the part where I need to find an elementary primitive. Can you explain, why it's not simply the formula mentioned at Wikipedia (ie. $L(a,b)\ =\ \int_a^b\sqrt{f_1'(t)^2+f_2'(t)^2}\ dt$)? – Maria Gordova Jul 25 '11 at 16:54
@Maria: this is no different for arc length than other integrals. There are many functions, like $\exp(-x^2)$ that we can't integrate in closed form. Not knowing your $f_1$ and $f_2$ we may have the same problem. But you can always do it numerically. – Ross Millikan Jul 25 '11 at 17:12