Scribd
Upload
82 views

Homework Optimization

The document contains two homework problems from an operations research course. The first problem proves a theorem about basic feasible rays in standard form linear programs. The second problem analyzes how perturbing a dual feasible solution by subtracting a multiple of a basis vector affects constraint activities, finding it leaves constraints with indices in the basis active but makes others non-active.

Uploaded by

Eldr hakk (Eldrhakk)
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
82 views

Homework Optimization

The document contains two homework problems from an operations research course. The first problem proves a theorem about basic feasible rays in standard form linear programs. The second problem analyzes how perturbing a dual feasible solution by subtracting a multiple of a basis vector affects constraint activities, finding it leaves constraints with indices in the basis active but makes others non-active.

Uploaded by

Eldr hakk (Eldrhakk)
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

HOMEWORK 3

1. Exercice 3.2
Problem: Prove the following theorem:
Every basic feasible ray of standard-form (P) is an extreme ray of its feasible region.
Let us take zb a basic feasible ray of standard-form (P) with problem (P) being:

min c0 x
(P) Ax = b ;
x ≥ 0.
Let β and η be the basic partition associated with zb with a non-basic index ηj such that :

zbη = ej ∈ Rn−m

zbβ = −A−1
β Aηj ∈ R
m

z = 0 and zb ≥ 0
zb is a feasible ray, so Ab
We are going to prove the theorem by contradiction. Let us take z1 and z2 such that zb = z1 + z2 with
z1 6= µz2 , µ 6= 0.
We have Ab z = 0 so A(z1 + z2 ) = 0. Which means that Az1 + Az2 = 0.
However we have z1 6= µz2 and z1 6= 0, z2 6= 0 , so Az1 = 0 and Az2 = 0. This means that z1 , z2 ∈ Ker(A).
It must be noted that dim(Ker(A)) = n − m
 
I
We have seen in the lecture that is a basis of Ker(A).
−A−1β A
 η
I
Let us call xi each column of . Then we can write any element of Ker(A) as a linear combi-
−A−1 β Aη
nation of the xi . We can see that zb = xj
Pn−m Pn−m
Let z1 = i=1 αi xi and z2 = i=1 β i xi
Pn−m Pn−m
Thus zb = z1 + z2 = i=1 αi xi + i=1 βi xi
Pn−m
So zb = i=1 (αi + βi )xi
Pj−1 Pn−m
So zb = i=1 (αi + βi )xi + (αj + βj )xj + i=j+1 (αi + βi )xi

As we have seen zb = xj .
Pj−1 Pn−m
Thus xj = (αj + βj )xj + i=1 (αi + βi )xi + i=j+1 (αi + βi )xi

But the xi form a basis of Ker(A) so there are linearly independent. This means that there is no λi
Pn−m
with i = 1, ..., n − m such that i=1 λi xi = 0.

Thus αi + βi = 0 ∀i ∈ [1, .., n − m] \ {j}


Because zb = z1 + z2 ≥ 0 and z1 , z2 are rays, we have z1 ≥ 0 and z2 ≥ 0.
Then αi ≥ 0 ∀i ∈ [1, .., n − m] and βi ≥ 0 ∀i ∈ [1, .., n − m]

This means that αi = 0 ∀i ∈ [1, .., n − m] \ {j} and βi = 0 ∀i ∈ [1, .., n − m] \ {j}
And αj + βj = 1, so αj = 1 − βj

Date: October 2, 2021.


1
2 IOE 510 - HOMEWORK 3

Then, we can write z1 = αj xj and z2 = βj xj


So z1 = (1 − βj )xj
1−β
If we take µ = βj j , then z2 = µz1 .

This contradicts with our initial statement, therefore we cannot write zb = z1 + z2 with z1 6= µz2 , µ 6= 0.
This proves the theorem: Every basic feasible ray of (P) is an extreme ray of its feasible region.

2. Exercice 3.4
Let β and η be a basic partition for the standard-form problem:
min c0 x
(P) Ax = b;
x ≥ 0.
We associate to β a dual feasible solution ȳ 0 = Cβ0 A−1
β for the dual problem :

max y0 b
(D)
y0 A ≤ C 0 ;
We have y 0 Aβ = Cβ0 .
We take a basic index βl and we take w̄ = Hl with H = A−1
β

We consider ȳ˜ = ȳ − λw̄0


Then ȳ˜0 = ȳ 0 − λw̄ = Cβ0 A−1
β − λw̄.

So ȳ˜0 A = Cβ0 A−1


β [Aβ |Aη ] − λw̄[Aβ |Aη ]

We must then calculate ȳ˜0 A for three different cases : if the index i = l, i ∈ β \ {l} and i ∈ η.

First, if i = l:
(ȳ˜0 A)l = Cβ,l − λ
In that case we can see that λ has a direct impact on the activity. The bigger λ is, the further we are from
the equality. The constraint will be verified but it will not be active.

Second, if i ∈ β \ {l}.
Then, (ȳ˜0 A)i = Cβ,i − λA−1
β,l Aβ,i
But A−1β,l Aβ,i = 0 hence :
˜ 0
(ȳ A)i = Cβ,i
If we choose an index within β except l, the constraint is active.

Finally, if we have i ∈ η:
(ȳ˜0 A)i = Cβ,i Aβ,i Aη,i − λA−1
β,l Aη,i
So (ȳ˜0 A)i = Aη,i (Cβ,i Aβ,i − λA−1β,l )
The constraint is verified, but it will not be active as when λ increases we move even further from the
constraint objective.

We have seen that ȳ˜ is a feasible solution of (D) because the constraint is verified for all i. However,
the constraint is only active when we choose an index i ∈ β \ {l}.

You might also like