Exercise 2.20*

Consider the Fourier-Motzkin elimination algorithm.

(a)
Suppose that the number m of constraints defining a polyhedron P is even. Show, by means of an example, that the elimination algorithm may produce a description of the polyhedron Πn1(P) involving as many as m24 linear constraints, but no more than that.
(b)
Show that the elimination algorithm produces a description of the one-dimensional polyhedron Π1(P) involving no more than m2n1 22n2 constraints.
(c)
Let n = 2p + p + 2, where p is a nonnegative integer. Consider a polyhedron in n defined by the 8 ( n 3 ) constraints ±xi ± xj ± xk 1,1 i < j < k n

where all possible combinations are present. Show that after p eliminations, we have at least

22p+2

constraints. (Note that this number increases exponentially with n. )

Answers

(a)
Let a be the number of inequalities of type (2.4), b be the number of inequalities of type (2.5) and c be the number of inequalities of type (2.6). Since we have to make sure that each of the inequalities of type (2.5) is compared to each of the inequalities of type (2.4), the resulting number of constraints in Q is a b + c (where we have a,b,c 𝔑 : a + b + c = m). We argue that this amount reaches its maximum for a = b = m 2 and c = 0.

Proof. We induct on the original number of constraints m.

  • For m = 2 we consider all possible cases of a + b + c = 2 and obtain

    22 4 [1 1 + 0 = 1 0 1 + 1 = 1 1 0 + 1 = 1 .

    where in we have not considered the case of two inequalities of type (2.8) as this would result in a trivial projection.

  • Now suppose inductively that we have proven this for some m > 2. We then have, for all a,b,c 𝔑 : a + b + c = m + 1, c + 1m + 1:

    a + b + c = (a + b + c 1) + 1 m2 4 + 1 (m + 1)2 4 .

Now we give a very trivial example of the sharpness of the bound in case that m is even. Consider the two-dimensional polyhedron with m = 2 constraints:

{ [x y ] | [ 1 1 1 1 ] [x y ] [ 0 1 ] }

PIC
Figure 1: Polyhedron defined by y x (red area) and 1 x y (blue area) and its projection.

We then obtain inequalities

{ 1 x y y x 1x x,

resulting in an unbounded one-dimensional polyhedron (, 1 2 ) with 224 = 1 constraint.

Proof. We induct on the dimension n of the polyhedron.

  • Let n = 2. Then we only apply the projection once and the assertion follows from part (a) of this exercise.

  • Now suppose that we have a polyhedron in n + 1 dimensions. Applying the Fourier-Motzkin elimination algorithm, we obtain a polyhedron Πn(P) in n dimensions with at most m24 constraints. We can apply the induction hypothesis and conclude that the Fourier-Motzkin projection Π1 (Πn(P)) = Π1(P) involves at most

    (m2 4 )2n1 22n2 = m2n 22n+12

    constraints.

Applying the previous part to the number of constraints at dimension n, we obtain [8 n 3 ] 2p 22p+12 = 2 [32p2p+1+2 ] n 3 2p = 22p+2 n 3 2p 22p+2 .

User profile picture
2022-01-30 07:39
Comments