MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

For $k, n \in \mathbb{N}$, let $\mathcal{C}_n \mathbb{R}^k$ denote the configuration space of $n$ distinct points in $\mathbb{R}^k$.

  • (1) Is there a description of the tangent space $T_C \mathcal{C}_n \mathbb{R}^k$ in terms of the configuration $C$?

Equipping $\mathbb{R}^k$ with the usual metric, a regression line $l_C$ of a collection $C \in \mathcal{C}_n \mathbb{R}^k$ is a line minimizing the quantity $E_C = \sum\limits_{p \in C} d(p, l_C)^2.$ We can see this as variational problem $E_C: M^k \rightarrow \mathbb{R}$ where $M^k$ is the parameter space of all lines in $\mathbb{R}^k$.

  • (2) Is there an explicit parametrization of $M^k$?

Without this knowledge, I'm not sure how to proceed to check whether $E_C$ is a Morse function.

[Note for $k=2$: given $C \in \mathcal{C}_n \mathbb{R}^k$, since $n<\infty$ we can always find an angle $\theta$ such that a rotation of our axes by $\theta$ yields coordinates $(x,y)$ on $\mathbb{R}^2$ for which the $x$-values of the $p \in C$ are all distinct, bringing us back to function-fitting and the usual least-squares regression which minimizes only the distances in the $y$ direction.]

share|cite|improve this question
a quick note (this is presumably known to the OP): this is NOT the usual least squares regression, which minimizes the sum of squares of the $y$-displacements of the points with respect to the fit line. – Igor Rivin Jun 6 '11 at 0:58
thanks for the clarification - I've made the appropriate edits above – Alexander Moll Jun 6 '11 at 3:18
up vote 5 down vote accepted

By calculus, the line $l_C$ is 'the' major axis of the ellipse of inertia of the finite point set $C$. (The reason for the quotes around 'the' in the previous sentence is that, if the ellipse of inertia is a circle, then $l_C$ is not well-defined; any line through the center of mass will minimize $E_C$.) I'm not sure what more 'intuitive' description you want, especially given that the map is not well-defined everywhere. (It is easy to see that you couldn't get a unique line in every case. For example, take the 3 vertices of an equilateral triangle.)

What you are asking for in the second part is just finding a direction so that orthogonal projection onto a line in that direction won't send any two of the points to the same point. To do this, it suffices to note that the vectors joining pairs are a finite set of directions, so you just choose a direction that is not orthogonal to any of those directions and you are done.

Finally, I don't see that going to the 1-point compactification of the plane is going to be useful in this problem, so asking about describing the lines as circles on the 2-sphere doesn't seem to be germane.

Added later: After this response, Alexander changed the question from one about lines in the plane ($k=2$) to lines in $\mathbb{R}^k$. The question about whether $E_C$ is a Morse function on $M^k$ in general has the answer 'no' when the ellipsoid of inertia of $C$ does not have $k$ distinct eigen-axes and 'yes' when it does. (The critical points are on the $(k{-}1)$-sphere of lines through the center of mass of $C$.)

Since $\mathcal{C}_n\mathbb{R}^k$ is an isometric $S_n$-quotient of an open subset of the product of $\mathbb{R}^k$ with itself $n$ times, the usual description of the tangent space applies, so there's nothing more to say about that, I guess.

As for $M_k$, the space of lines in $\mathbb{R^k}$ this can be described by starting with the double cover $\tilde M_k$ consisting of the oriented lines in $\mathbb{R}^k$. This latter space is essentially the tangent bundle of $S^{k-1}$, which consists of pairs $(u,v)$ with $u\in S^{k-1}$ and $v\in\mathbb{R}^k$ satisfying $u\cdot v = 0$. The idea is that $u$ is the direction of the line and $v$ is its point of closest approach to the origin. Thus, the line is parametrized in the form $v + t u$. Now, to get $M_k$, you divide by the relation $(u,v)\simeq (-u,v)$.

share|cite|improve this answer
Thank you for the response - funnily enough, I've been working through Arnold's book on mechanics, and the ellipsoid of inertia is only a handful of pages in my future! I've edited the original statement accordingly - my main interest is in a description of the space $M^k$ and, if a parametrization isn't easy to write down, whether the collection of all $E_C$ might help describe that space. – Alexander Moll Jun 6 '11 at 5:48
again, great comments - thank you for your help! – Alexander Moll Jun 6 '11 at 15:09

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.