Weak Duality Theorem (according to my course notes)

Jamin2112
Messages
973
Reaction score
12

Homework Statement



Weak Duality Theorem. If x ε ℝn is feasible for P and x ε ℝm is feasible for D, then

cTx ≤ yTAx ≤ bTy.

Thus, if P is unbounded, then D is necessarily infeasible, and if D is unbounded, then P is necessarily infeasible. Moreover, if cTx = bTy with x feasible for P and y feasible for D, then x must solve for P and y must solve for D.

Homework Equations



c ε ℝn, b ε ℝm A ε ℝm x n

P: maximize cTx subject to Ax ≤ b, 0 ≤ x, where the inequalities Ax ≤ b are to be interpreted componentwise.

D: minimize bTy subject to ATy ≥ c, y ≥ 0.

The Attempt at a Solution



So the "proof" that follows in my course notes starts with

Let x ε ℝn be feasible for P and y ε ℝm be feasible for D.

Then it basically gives a simple verification of the inequality and says that the other results obviously follow.

I'm confused by the logic here. The theorem starts by choosing x and y feasible for P and D. Then it states that if P is unbounded, the inequality shows that there could be no feasible y ...... which is strange because the inequality only works for a feasible x and y.Could someone clear this up for me?
 
Last edited:
Physics news on Phys.org
The hypothesis for the theorem asserts that there exist x which is feasible for P and there exists y which is feasible for D. By that hypothesis, it is impossible for either P or D to be infeasible.
 
HallsofIvy said:
The hypothesis for the theorem asserts that there exist x which is feasible for P and there exists y which is feasible for D. By that hypothesis, it is impossible for either P or D to be infeasible.

Can that be gathered from looking at the inequality? My point is that the inequality is only relevant for feasible x and y; hence I don't see how it illustrates a point about the infeasibility of y.

From my course notes:

screen-capture-7-4.png
 
Jamin2112 said:
Can that be gathered from looking at the inequality? My point is that the inequality is only relevant for feasible x and y; hence I don't see how it illustrates a point about the infeasibility of y.

From my course notes:

screen-capture-7-4.png

Look at what the Theorem says: it says IF x is feasible for P and y is feasible for D, then cx ≤ by. OK, so think about it: if P is unbounded, there cannot be any y that is feasible for D (because a feasible y produces an upper bound on cx for all feasible x). Similarly for the other part of the statement.

BTW: it is possible for both P and D to be infeasible, as some examples show.

RGV
 
Ray Vickson said:
Look at what the Theorem says: it says IF x is feasible for P and y is feasible for D, then cx ≤ by. OK, so think about it: if P is unbounded, there cannot be any y that is feasible for D (because a feasible y produces an upper bound on cx for all feasible x). Similarly for the other part of the statement.

BTW: it is possible for both P and D to be infeasible, as some examples show.

RGV

Here's the logic of the first part of the theorem.

Premises:
(1) Ax ≤ b
(2) ATy ≤ c
Logical Consequence:
cTx ≤ bTy.

Here's the logic of the second part of the theorem:

Premises:
(1) For all M ε ℝ there exists an x ε ℝn such that cTx > M.
Logical Consequence:
There exists no y ε ℝm such that ATy ≤ b.


Those seem like unrelated parts of the theorem, each requiring it's own proof. Maybe I'll need to look at it more closely.
 
The second part is not part of the main theorem, but is a sole consequence of the theorem. It does not really need a formal proof, since it is just a logical consequence of the theorem itself. The theorem implies that IF P and are both feasible, we have an upper bound on cx and a lower bound on by. Thus, if it happens that cx is unbounded it MUST be the case that D is i feasible (for when we sat that cx is unbounded we are implicitly assuming that P is feasible).

RGV
 
There are two things I don't understand about this problem. First, when finding the nth root of a number, there should in theory be n solutions. However, the formula produces n+1 roots. Here is how. The first root is simply ##\left(r\right)^{\left(\frac{1}{n}\right)}##. Then you multiply this first root by n additional expressions given by the formula, as you go through k=0,1,...n-1. So you end up with n+1 roots, which cannot be correct. Let me illustrate what I mean. For this...
Back
Top