As to compute how much a circle rotates when 'rolled' along a curve in $\mathbb{R}^2$, the most intuitive way to me to find the number of rotations is:


  • $S$ is the arc-length of the curve
  • $C$ is the circumference of the circle
  • $W$ is the total curvature of the curve

However, this seems to agree with:


  • $T$ is the arc-length of path of the center of the circle

Can anyone intuitively explain why the latter works as well?

Also I'm wondering whether $T/C$ still works if the circle is replaced by some closed curve (whereby $C$ is the arc-length of the closed curve and $T$ is the arc-length of path of the mass-center of the closed curve). Edit: After writing down the integrals, I think a sensible generalization might (rather than the mass-center) have more to do with the center of the osculating circle of the close curve at its current intersection with the curve it's being rolled on.

$\ $
Edit: In other words:
Let $\ c:[a,b]\to\mathbb{R}^2\ $ be some smooth curve along which a circle with radius $\text{abs}(r)$ is being rolled.
Let $\ \text{center}:[a,b]\to\mathbb{R}^2\ $ be the center of the circle given by: $$\text{center}(t)=c(t)+r\frac{\{c_2'(t),-c_1'(t)\}}{||c'(t)||_2}$$ That is the sign of $r$ determines on which side of the curve the circle is being rolled.

Expressed with integrals, the formulas for the total rotation are

$S/C+W/(2\pi)={\large\int_a^b}\dfrac{||c'(t)||_2}{2r\pi}dt+ {\huge\int_{\large a}^{\large b}}\dfrac{\det{\left( \begin{array}{cc} c_1'(t) & c_2'(t) \\ c_1''(t) & c_2''(t) \\ \end{array} \right)}}{||c'(t)||_2^2\cdot (2\pi)}dt$

$T/C = {\large\int_a^b}\dfrac{||\text{center}'(t)||_2}{2\,\text{abs}(r)\pi}\cdot\text{sign}\left(\dfrac{1}{r}+\dfrac{\det{\left( \begin{array}{cc} c_1'(t) & c_2'(t) \\ c_1''(t) & c_2''(t) \\ \end{array} \right)}}{||c'(t)||_2^3}\right)dt$

both of which are influenced by the signs of $r$ and the curvature determinant.


1 Answer 1


Can anyone intuitively explain why the latter works as well?

It's easier to understand if the curve is parametrized by arc length. Also, I find it easier to understand the relation $T=S+rW$ relating absolute lengths than the relation $T/C=S/C+W/(2\pi)$ relating counts of rotations.

So let $s$ parametrize the curve by arc length. I'll also assume that the rotating circle is along the "outside" of the curve the entire time, and that $\kappa$ is nonzero. At the point of tangency, there is the osculating circle with radius $1/\kappa(s)$. The rotating circle has radius $r$. So the center of the rotating circle is (for an infinitesimal moment) tracing a circular path with radius $r+1/\kappa(s)$.

Over a short length $ds$ within the curve, the length of the circular arc that the center travels through can be calculated using proportional reasoning:


enter image description here

Now integrate over $s$ and you get the arc length through which the center passes. That is, $$T=\int_0^S(1+r\kappa(s))\,ds$$

But break it up into two integrals: $$\begin{align} T&=\int_0^Sds+r\int_0^S\kappa(s)\,ds\\ T&=S+rW \end{align}$$

Now divide by $C$ to get the form you have observed.

If the rotating circle is along the "inside", then the same reasoning changes the integral for $T$ to $$T=\int_0^S(1-r\kappa(s))\,ds$$ If the curve has zero curvature throughout, then $$T=\int_0^Sds=S$$ And lastly for more complicated curves, if they can be broken up piecewise into curves of these three types, you are set.


You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .