The convex combinations of two linearly independent vectors in $\mathbb{R}^3$ span a line. The convex combinations of three linearly independent vectors in $\mathbb{R}^3$ span a solid triangle. The convex combinations of four vectors in general position in $\mathbb{R}^3$ span something a lopsided tetrahedron, I think. How can we describe the convex combinations of five vectors in $\mathbb{R}^3$? Can we do better than saying it is "the convex hull of the union of the lopsided tetrahedra which are the combinations of any four of them taken at a time"? I.e.

$$C(x_1, \dotsc, x_5) = \operatorname{Hull}\left( \cup_{1\leq i \leq 5} T_i \right),\, \text{for } T_i = C(x_1, \dotsc, \hat{x}_i, \dotsc, x_5)?$$

(Here $C$ is a function mapping vectors to the set of all their convex combinations.)

More generally, what about $n+2$ vectors in $\mathbb{R}^n$?

  • 3
    $\begingroup$ The convex hull of five points in $\mathbb R^3$ is, in nondegenerate cases, either two tetrahedra glued together (i.e. a triangular bipyramid), or just a tetrahedron (if one of the five points can be expressed as a convex combination of the other four). $\endgroup$ – Rahul May 14 '14 at 4:40

Thanks to Carathéodory's theorem, you don't need to take convex hull after union: $$C(x_1, \dotsc, x_5) = \bigcup_{1\leq i \leq 5} T_i\quad \text{for } T_i = C(x_1, \dotsc, \hat{x}_i, \dotsc, x_5)$$ (Similarly in higher dimensions: every point of convex hull is a convex combination of some $(n+1)$ vertices).

However, this representation is somewhat redundant: the tetrahedra overlap a lot. As Rahul said in a comment, the set is actually the union of 1 or 2 tetrahedra (apart from degenerate cases). Here's a geometrical explanation: imagine you are at one of these five points. The other four form a tetrahedron $T$.

If you are inside of $T$, then $C = T$. If you are outside, then when looking at $T$, you see either $1$, $2$, or $3$ of its faces.

  • If you see $1$ face of $T$, then $C$ is the union of $T$ and the tetrahedron built on this face, using your point as a vertex.
  • If you see $3$ faces of $T$, then $C$ is the tetrahedron built on the face you don't see, using your point as a vertex.
  • If you see $2$ faces of $T$, they are separated by an edge. Call this edge $PQ$. The other two vertices of $T$ form an edge $RS$ which you don't see (except for the vertices $R,S$ themselves). Consider the triangle $YRS$ (where $Y$ is you), which divides $T$. Build two tetrahedra with the base $YRS$: one with vertex $P$, the other with vertex $Q$. Their union is $C$.

One could translate the above argument into inequalities and cases with subcases, but who would want to read that?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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