A curve is a trip, not just a picture
Up to now a curve has mostly been a shape: the set of points satisfying an equation, like a circle x^2 + y^2 = r^2 or a parabola y = x^2. That view is fine for asking which points belong, but it goes silent the moment you ask how was it drawn or how far is it around. The fix is to stop staring at the finished picture and instead watch a point move. We let a single moving point depend on a clock value t, and write its position as a vector that changes with t. That moving record is a parametrized curve, and t is its parameter — most vividly, time.
Concretely, a plane curve is a pair of ordinary functions packaged together: gamma(t) = (x(t), y(t)). As t runs over an interval, the point gamma(t) sweeps out the shape. The unit circle becomes gamma(t) = (cos t, sin t) for t from 0 to 2 pi — a point marching counterclockwise, returning home after one lap. A spatial curve just adds a third slot, gamma(t) = (x(t), y(t), z(t)); the helix gamma(t) = (cos t, sin t, t) is the unit circle that forgot to stop climbing, a corkscrew rising as it turns. The same machinery handles plane and space with one extra coordinate, exactly as it did when vectors first carried you off the page.
Velocity: the tangent vector falls out of calculus
Here is where the calculus you brought up this ladder earns its keep. If gamma(t) is a position changing with time, its derivative gamma'(t) — differentiate each component separately — is the velocity vector: gamma'(t) = (x'(t), y'(t)). Geometrically this arrow is tangent to the curve, pointing in the instantaneous direction of travel, and its length tells you the speed. For the circle gamma(t) = (cos t, sin t), the velocity is gamma'(t) = (-sin t, cos t): always perpendicular to the position arrow, always of length 1, the steady whirl of something going round at unit speed. The derivative we knew as 'slope of a graph' has quietly become a direction in space.
But there is an honest worry. What if the moving point stops dead — gamma'(t) = 0 at some instant? Then there is no velocity arrow, no direction of travel, and the curve can do something ugly: pin a sharp corner or a sudden cusp where you would never have guessed from the smooth-looking formula. The curve gamma(t) = (t^2, t^3) is the classic trap; at t = 0 its velocity vanishes and the picture has a genuine cusp, a sharp point. To rule this out we demand a regular curve: one whose velocity gamma'(t) is never the zero vector. Regularity is the promise that the trip never halts, so the curve always has a well-defined direction at every point.
Arc length: adding up the tiny steps
Now the real question: how long is the curve from t = a to t = b? Imagine freezing the trip at a great many instants and connecting the dots with tiny straight segments. Each little segment is a step the moving point took in a sliver of time dt, and its length is (speed) times dt — that is, |gamma'(t)| dt, the magnitude of the velocity times the time elapsed, exactly 'distance equals speed times time' for one heartbeat. Add up all those slivers and let them shrink to nothing, and the sum becomes an integral. That limit of summing tiny steps is the whole idea of an integral, met again as a length.
arc length from a to b L = integral from a to b of |gamma'(t)| dt in the plane, gamma'(t) = (x'(t), y'(t)), so |gamma'(t)| = square root of ( x'(t)^2 + y'(t)^2 ) L = integral from a to b of square root of ( x'(t)^2 + y'(t)^2 ) dt
Let us check it on something we already trust. Take the unit circle gamma(t) = (cos t, sin t) once around, t from 0 to 2 pi. Its speed is |gamma'(t)| = |(-sin t, cos t)| = square root of (sin^2 t + cos^2 t) = 1 — constant unit speed, by the Pythagorean identity. So L = integral from 0 to 2 pi of 1 dt = 2 pi, the circumference of a unit circle. The machinery reproduces the answer you have known since the radian was introduced — reassuring, and exactly what an honest definition should do: agree with the old cases before it conquers new ones.
Why speed shouldn't change the length
A road has one length no matter whether you stroll or sprint along it. So a good notion of arc length had better not depend on which parametrization we chose to trace the curve — only on the path itself. This is exactly the worry raised earlier: the same circle can be run at double speed by (cos 2t, sin 2t). Does the integral notice? Run that double-speed circle for t from 0 to pi (half the time, since you go twice as fast): the speed is now 2, and L = integral from 0 to pi of 2 dt = 2 pi again. The faster speed and the shorter time interval cancel perfectly. The length survived the change of clock.
This is not luck; it is the change-of-variables rule from your integration toolkit doing its job. Choosing a new clock is a reparametrization — a smooth, strictly increasing (or decreasing) relabeling of the parameter, say t = h(s). When you substitute it into the arc-length integral, the extra factor h'(s) that calculus produces is exactly the factor that adjusts the speed, and the two changes cancel. So arc length is an invariant of the curve, a property of the path and not of the timetable. That is precisely the kind of quantity worth building a geometry on: one the observer's choices cannot fake.
The honest parametrization: travel at unit speed
Since arc length belongs to the curve itself, why not use it as the clock? Instead of some arbitrary t, parametrize by the distance already travelled along the curve, usually written s. This is the arc-length parametrization, and it has a beautiful defining property: a point parametrized by arc length always moves at unit speed, |gamma'(s)| = 1. The reason is almost a tautology — when your clock is the odometer, one unit of clock buys exactly one unit of distance, so speed is forced to be 1. It is the curve described by an ant crawling along it at a steady one-centimetre-per-second, never hurrying, never dawdling.
Why bother? Because unit speed strips away everything about how fast we happened to trace the curve and leaves only what the curve is. At unit speed the velocity gamma'(s) is automatically a unit tangent vector — direction with no contaminating speed information — and that clean tangent is the seed from which the next guides grow curvature, the Frenet frame, and torsion. Arc-length parametrization is the lens that makes a curve's intrinsic shape visible, with the accidents of timing wiped off the glass.
What you can now do, and where it leads
- Write a curve as a moving point gamma(t) = (x(t), y(t)) (add a z(t) in space), seeing the trip, not just the set of points it lands on.
- Differentiate to get the velocity gamma'(t), the tangent direction; check the curve is a regular curve by confirming gamma'(t) is never the zero vector.
- Measure length by integrating speed: L = integral of |gamma'(t)| dt, the limit of adding up tiny straight steps.
- Trust that length is a reparametrization-invariant of the curve; in principle switch to the arc-length parametrization to travel at unit speed and read off the unit tangent.
This same circle of ideas extends past the Cartesian world: a curve given in polar coordinates as r = f(theta) is secretly the parametrized curve (f(theta) cos theta, f(theta) sin theta), so its arc length is the very same integral of speed once you compute the velocity. The recipe — position, differentiate, take magnitude, integrate — is robust enough to handle spirals, polar curves, and helices alike, because it never cared what coordinate dress the curve was wearing.
And it sets up everything ahead. With a unit-speed curve in hand, the next guide asks *how sharply does the tangent direction turn?* — and the rate of that turning is curvature, kappa. After that, the tangent grows companions (the normal and binormal) into the Frenet frame, the twisting out of the plane becomes torsion, tau, and the fundamental theorem of curves will reveal that kappa and tau between them pin down a space curve completely, up to where you place it. Every step of that story rests on the modest move you have just made: treating a curve as motion, and measuring it from the inside.