Arc Length and Reparametrization
A parametrized
curve comes with a built-in speedometer: the speed
\lVert \mathbf{r}'(t) \rVert. If you know your speed at every instant,
then the total distance you travel is just the accumulated speed — speed integrated over time.
That single idea is the whole of arc length.
L = \int_a^b \lVert \mathbf{r}'(t) \rVert \, dt.
This is the same formula you met for the length of a graph — see
arc length by
integration — now written in the cleaner vector language: chop the curve into tiny
pieces, each of length \lVert \mathbf{r}'(t)\rVert\,dt (speed × little
time), and add them all up.
The arc-length function
Instead of only the total length, track how much length has been laid down by time
t. Fix a starting time a and define the
arc-length function
s(t) = \int_a^t \lVert \mathbf{r}'(u) \rVert \, du.
It measures the distance travelled along the curve from the start up to parameter
t. By the Fundamental Theorem of Calculus, differentiating an integral
with respect to its upper limit gives back the integrand:
\frac{ds}{dt} = \lVert \mathbf{r}'(t) \rVert = v(t).
So ds/dt is the speed — a clean restatement of "distance
accumulates at a rate equal to speed." For a regular curve the speed is always positive, so
s(t) is strictly increasing: it never doubles back, and
therefore it can be inverted. That invertibility is the door to the most natural way of all to
describe a curve.
Unit-speed (arc-length) parametrization
Because s(t) is strictly increasing, we can turn the story around and
use the distance travelled, s, as the parameter instead of time
t. Invert to get t(s) and feed it back in:
\boldsymbol{\gamma}(s) = \mathbf{r}\big(t(s)\big).
This reparametrized curve has a magical property. By the chain rule and
ds/dt = \lVert \mathbf{r}'\rVert,
\boldsymbol{\gamma}'(s) = \mathbf{r}'(t)\,\frac{dt}{ds} = \frac{\mathbf{r}'(t)}{\lVert \mathbf{r}'(t)\rVert} = \mathbf{T}, \qquad \lVert \boldsymbol{\gamma}'(s)\rVert = 1.
A curve parametrized by arc length moves at unit speed: it covers one unit of
length for every unit of parameter. Its velocity is the unit tangent
\mathbf{T}. Nothing is wasted on bookkeeping how fast the point moves —
all the remaining information is pure shape.
Every regular curve can be reparametrized by arc length. In the arc-length parameter
s:
-
the speed is constant and equal to one:
\lVert \boldsymbol{\gamma}'(s)\rVert = 1;
-
the velocity coincides with the unit tangent,
\boldsymbol{\gamma}'(s) = \mathbf{T}(s);
-
differentiation with respect to s measures change per unit of
length travelled, which is exactly what curvature will need.
Time is arbitrary — you can traverse the same track slowly, quickly, or in fits and starts, and
the geometry of the track doesn't care. Arc length is the one parametrization that every
observer agrees on, because it's measured by the curve itself: lay a tape measure along it.
When we come to curvature, we'll want "how fast the direction turns" — but
fast with respect to what? If we used time, a curve driven quickly would seem to bend
more sharply than the same curve driven slowly, which is absurd. Measuring the turning per unit
of length strips the speed out and leaves a number that belongs to the shape alone.
That's why differential geometers reach for the arc-length parameter again and again: it turns
messy formulas into clean ones by removing the one thing that was never geometric — the clock.
Worked example 1 — the circle
Take the circle \mathbf{r}(t) = (R\cos t,\ R\sin t). Its speed is the
constant R, so the arc-length function is
s(t) = \int_0^t R\,du = R\,t \qquad\Longrightarrow\qquad t(s) = \frac{s}{R}.
Substitute back to get the unit-speed circle:
\boldsymbol{\gamma}(s) = \left(R\cos\frac{s}{R},\ R\sin\frac{s}{R}\right), \qquad \lVert \boldsymbol{\gamma}'(s)\rVert = 1.
The full circumference falls out for free: when the point returns to the start,
t = 2\pi, so s = 2\pi R — the length of the
circle.
Worked example 2 — the helix, and a line
The helix \mathbf{r}(t) = (\cos t,\ \sin t,\ c\,t) has constant speed
\sqrt{1 + c^2}, so
s(t) = \sqrt{1 + c^2}\;t \qquad\Longrightarrow\qquad t(s) = \frac{s}{\sqrt{1 + c^2}},
and the arc-length helix is
\boldsymbol{\gamma}(s) = \big(\cos\frac{s}{k},\ \sin\frac{s}{k},\ \frac{c s}{k}\big)
with k = \sqrt{1 + c^2}. Whenever the speed is a constant, the
reparametrization is nothing more than a rescaling of the parameter — the easy case.
A straight line makes it even plainer. Let
\mathbf{r}(t) = \mathbf{p} + t\,\mathbf{v} with
\lVert \mathbf{v}\rVert = L. Then
s(t) = Lt, and the unit-speed line is
\boldsymbol{\gamma}(s) = \mathbf{p} + s\,\frac{\mathbf{v}}{\lVert \mathbf{v}\rVert} = \mathbf{p} + s\,\hat{\mathbf{v}},
which just walks along the line one unit of length at a time — exactly what "unit speed" should
mean.
Two traps live in this section. First, arc length does not depend on the
parametrization. Whether you crawl or sprint around a circle, its length is still
2\pi R; the integral \int \lVert \mathbf{r}'\rVert\,dt
automatically corrects for speed, because a faster parametrization has a larger
\lVert \mathbf{r}'\rVert but a proportionally shorter time interval.
Length is a property of the image and orientation, not of the clock.
Second — and this is where many students come unstuck — you usually cannot solve
s(t) for t in closed form. The
unit-speed reparametrization always exists in principle (that's the theorem), but for
most curves the integral s(t) = \int_a^t \lVert \mathbf{r}'\rVert\,du
has no elementary antiderivative, and even when it does, inverting it can be impossible with
formulas. The ellipse is the classic culprit: its arc-length integral is an elliptic
integral, which is precisely the historical example that has no closed form. So arc-length
parametrization is a superb theoretical tool — we prove things with it — but in practice
you rarely write \boldsymbol{\gamma}(s) down explicitly. The circle,
helix and line are lucky exceptions because their speed is constant.
Speed builds up into length
For the ellipse \mathbf{r}(t) = (2\cos t,\ \sin t) the speed
\lVert \mathbf{r}'(t)\rVert = \sqrt{4\sin^2 t + \cos^2 t} is not
constant — it pulses up at the flat ends of the ellipse and drops at the pointy ones. The lower
curve is that speed; the upper curve is the running arc length
s(t) = \int_0^t \lVert \mathbf{r}'\rVert\,du. Wherever the speed is high
the length climbs steeply; the slope of the length curve is exactly the speed below it, because
ds/dt = \lVert \mathbf{r}'\rVert. The total height reached at
t = 2\pi is the perimeter of the ellipse — a number with no
elementary closed form.