Due: Thursday, March 5, 2009.
Penalty for late homeworks: 10% for each day or part of a day.
Given the complete graph
on
vertices,
a clustering
of the vertices is obtained by choosing an integer
and a partition of the vertices into
sets
satisfying:
for
and
.
Note that
is not fixed.
The incidence vector of this clustering is defined by
Let
be the set of incidence vectors of clusterings
for
.
Let edge
have weight
.
The clustering problem for this set of edge weights is then
The edge weights
can be positive or negative.
This problem is the subject of questions 2-4.
- Let
be points in
.
- Prove that if the points are linearly independent then they are affinely independent.
- Assume all
points satisfy the equality
, where
and
.
Prove that if the points are affinely independent then they are linearly independent.
- Show that the inequality
 |
(1) |
defines
a facet of the convex hull of
.
(You may assume
that the dimension of
is
, that is,
is full dimensional.)
- Choose a seed and solve the clustering problem in
http://www.rpi.edu/~mitchj/matp6620/hw4/clr1.mod
using a cutting plane algorithm, adding violated triangle inequalities of the
form (1) and
- Another random clustering problem is in
http://www.rpi.edu/~mitchj/matp6620/hw4/clr2.mod
with all the triangle inequalities (1-3) included.
For this problem, the triangle inequalities are not sufficient.
Find a valid inequality for the clustering problem that is violated by the optimal
solution to the LP relaxation with all the triangle inequalities included.
John Mitchell
2009-02-22