CIS 677: Homework 4, due about Nov. 24

CIS 677: Homework 4, due about Nov. 24


  1. Problem 11.9 of the text.
  2. Let S be a set of points in E^d, in general position, so that no d+1 points of S are in a common hyperplane. Suppose we want to find the vertices of the convex hull of S, the extreme points of S.

    Consider the following algorithm:

    E := {};
    for each p in S do
    	If p is in E, skip it.
    	Check if p is in conv(E);
    	If p is not in conv(E) then
    		use the proof that p is not in conv(E)
    			to find v in vert(conv(S))\E, and add v to E.
    od;
    
    1. Show that this algorithm can be implemented to run in O(nV) expected time, where V is the size of vert(conv(S)). Hint: use linear programming to find the "proof".
    2. For p in S, define depth(p, S) as 1 if p is an extreme point of S, so p is in vert(conv(S)), and otherwise depth(p, S) is 1+depth(p, S\vert(conv(S))). Show that the depth of every point in S can be computed in O(n2) expected time.
    3. Given a set H of n hyperplanes, a halfspace h is called irredundant if P(H) is not equal to P(H\{h}). Show that a halfspace can be proven irredundant in O(n) expected time. Show that the I irredundant halfspaces of H can be found in O(nI) expected time.
    4. Each irredundant halfspace of H determines a facet of P(H). Recall that a face of a polyhedron is either a facet, or a face of a facet. Show that all F faces of P(H) can be found in O(nF) expected time and O(n) space. The "constant factors" in O( ) may be exponential in d.
  3. Suppose the convex hull of a set S in general position is computed using a randomized incremental algorithm. The search problem is to find the facets of conv(Ri) seen by pi+1. Suppose that for each site p not in Ri, a facet of conv(Ri) seen by p is known. Show that this information solves the search problem.
    1. Show that this information can be updated as the algorithm proceeds, such that the total work done by the algorithm is within a constant factor of that of the "conflict graph" algorithm described in the text or the algorithm of Lecture 8.
    2. How much space does the algorithm need, if the expected number of facets of conv(Ri) is some number fi?

      Ken Clarkson
      November 1997