While studying for my exam, I've run into some problems understanding what a smooth curve in $\mathbb{R}^2$ is.

I first thought that, intuitively, I could think of a piece of string on a piece of paper, where the string is the curve and my paper is the $\mathbb{R}^2$ space.

But in my book, it says the following:

A smooth curve in $\mathbb{R}^2$ is every subset $\Gamma$ of $\mathbb{R}^2$ which can be written as $\Gamma = \phi[a,b]$, where $\phi:[a,b] \rightarrow \mathbb{R}^2$ $(a<b)$ has the following properties:

  1. $\phi$ is a bijection of $[a,b]$ on $\Gamma$
  2. $\phi \in C^1[a,b]$
  3. $\phi'(t) \neq 0$ $\forall a<t<b$

Now my problem lies in these properties:

I googled examples of smooth curves and was returned images of ellipsoids and circles (among other shapes).
As far as I know, to represent a circle or ellipsoid, every value in $[a,b]$ would need have 2 values in $\mathbb{R}^2$ and because of this it's not a bijection.
Another problem arises in the fact that the derivative cannot be 0, as this is the case in 2 points in aforementioned shapes.

  • $\begingroup$ You're quite mistaken in saying the derivative must be $0$ in the case of a circle or an ellipse. If $[a,b]=[0,2\pi]$ and $\varphi(x)=(\cos x, \sin x)$, then the curve is a circle, and the derivative is $\varphi'(x)=(-\sin x, \cos x)\ne(0,0)$. But I agree that the part about being one-to-one is mistaken. ${}\qquad{}$ $\endgroup$ May 19, 2014 at 15:47
  • $\begingroup$ The curve $(x,y)=(t^2-1,t^3-t)$ is by reasonable definitions smooth, but is not one-to-one since its values at $t=1$ and $t=-1$ are equal. $\endgroup$ May 19, 2014 at 15:57

3 Answers 3


You've got a couple of mis-interpretations here; let me clear them up.

  1. You are thinking of graphs $\{(x,f(x))\mid x\in[a,b]\}$, where $f$ is a function $[a,b]\to\mathbb{R}$, instead of more general subsets of the plain. Here, $\phi$ is a parameterization of your curve: it is a function $\phi(t)=(x(t),y(t))$. The condition that this be a bijection has nothing at all to do with whether or not each $x$-coordinate has at most one corresponding $y$-coordinate; rather, it is simply whether each point $(x,y)$ has at most one $t$ such that $\phi(t)=(x,y)$.

  2. The derivative of $\phi$ is not the slope of the tangent line; rather, it is a direction vector for the tangent line. If we write $\phi(t)=(x(t),y(t))$ as above, then $\phi'(t)=(x'(t),y'(t))$. The condition that $\phi'$ be non-zero is actually the condition that the derivative is never the zero vector. At no point on the unit circle, with its usual parameterization $\phi(t)=(\cos t,\sin t)$ for $0\leq t\leq 2\pi$, is the derivative the $0$ vector.

I will note, however, that the definition you're given DOES preclude considering circles as smooth curves; not for the reasons you suggest, however. The problem with circles is that there is no way to come up with a parameterization from a closed interval to a circle without at least two $t$-values giving the same point; in the classic parameterization on $[0,2\pi]$, these are $t=0$ and $t=2\pi$.

You can think of the definition you've been given as being for a non-closed smooth curve. You can give a (barely!) more general definition which wouldn't exclude circles/other closed curves, as follows: simply relax the bijection condition to say that we need the restriction of $\phi$ to $[a,b)$ to be a bijection, rather than the entire map to be a bijection. This allows us to start and end at the same point.

  • $\begingroup$ By this definition a figure-eight that crosses itself at a right angle would not be a smooth curve. I wonder why that's considered reasonable as a definition. $\endgroup$ May 19, 2014 at 15:49
  • $\begingroup$ @MichaelHardy Yeah, hard to say. In multivariable calc courses, they frequently deal with non-crossing curves as a special case; I guess they're just restricting their attention to those, then considering curves that cross a finite number of times in a piecewise fashion? $\endgroup$ May 19, 2014 at 15:50
  • $\begingroup$ @Hardy It is one of the possible reasonable definitions; it is usually stated as "an injective immersion of an interval". And by the way the figure eight can be parametrized in such a way that it is indeed an injective immersion of an open interval. So it is less restrictive than an embedded 1-submanifold and more specific than a generic parametrized curve (e.g. it has one tangent line at each point)... $\endgroup$
    – N. Ciccoli
    May 21, 2014 at 10:14

After doing some more research, I've come to the conclusion that the images google returned were images of PIECEWISE smooth curves, which is why I got confused. I'm going to leave this question though since it could help out someone who made the same mistake as me


Just to add a word about condition 3 (this is usually stated by saying that your curve is regular). This is not absolutely necessary and one could, in principle, develop a theory of curves having also non-regular points. What is interesting is to consider which properties depend on regularity. The most important one is the fact that a differentiable path admits arclength parametrization if and only if it is regular at every point. The second one is that at non regular points one does not have a Frénet frame.

The usual example is the cuspidal curve $F(t)=(t^3,t^2)$ which is not regular at $t=0$ (remark that one may still define a tangent line at that point by using a concept of weak tangent, i.e. the common limit, if it exists of left and right tangent lines).


You must log in to answer this question.

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