QE2013_AC-3_ECE580-3
(i)
Solution 1 :
To remove absolute values, each variable is separated into a positive component and a negative component. For $ x_1 $ for example:
$ x_1 = x_1^+ - x_1^-, \ \ \ \ x_1^+ ,x_1^- \ge 0, \ at\ least\ one\ of\ x_1^+,\ x_1^-\ is\ 0 $
Then $ |x_1| = x_1^+ + x_1^- $
Therefore the given problem can be converted into a linear programming problem:
$ maximize -x_1^+ -x_1^- -x_2^+ -x_2^- -x_3^+ -x_3^- \\ subject\ to \\ \begin{bmatrix} 1 & -1 & 1 & -1 & -1 & 1 \\ 0 & 0 & -1& 1 & 0 & 0 \end{bmatrix} \begin{bmatrix} x_1^+ \\ x_1^- \\ x_2^+ \\ x_2^- \\ x_3^+ \\ x_3^- \end{bmatrix} = \begin{bmatrix} 2 \\ 1 \end{bmatrix} \\ x_1^+ ,x_1^- ,x_2^+ ,x_2^- ,x_3^+ ,x_3^- \ge 0 $
This problem shares the same solution of the original problem if for i = 1,2,3, at least one of $ x_i^+\ and\ x_i^-\ is\ 0. $
$ -x_2^+ +x_2^- = 1 \Rightarrow x_2^+ = 0,\ x_2^- = 1 $
Therefore the equality constraint can be simplified as
$ x_1^+ -x_1^- -x_3^+ +x_3^- = 3 $
There are 4 cases to consider:
Case 1: $ x_1^+ = x_3^+ = 0 $
This is equivalent to finding $ x_1^-, x_3^- \ge 0\ to\ maximize\ -x_1^- -x_3^-\ subject\ to\ -x_1^- +x_3^- = 3 $
Optimal solution is $ x_1^- = 0, x_3^- = 3 $, in this case $ -x_1^+ -x_1^- -x_2^+ -x_2^- -x_3^+ -x_3^- = -4 $
The other 3 cases can be solved similarly:
Case 2: $ x_1^+ = x_3^- = 0 $
No solution as $ -x_1^- -x_3^+ = 3 $ cannot be satisfied (left side is non-positive).
Case 3: $ x_1^- = x_3^+ = 0 $
Every feasible solution is optimal in this case (i.e., any $ x_1^+, x_3^- $ satisfying $ x_1^+ +x_3^- = 3, x_1^+, x_3^-\ge 0 $), $ -x_1^+ -x_1^- -x_2^+ -x_2^- -x_3^+ -x_3^- = -4 $
Case 4: $ x_1^- = x_3^- = 0 $
Optimal solution is $ x_1^+ = 3, x_3^+ = 0 $, in this case $ -x_1^+ -x_1^- -x_2^+ -x_2^- -x_3^+ -x_3^- = -4 $
Therefore, the optimal solution is $ x_1^- = x_3^+ = 0,\ x_2^+ = 0,\ x_2^- = 1, $ and any $ x_1^+, x_3^- $ satisfying $ x_1^+ +x_3^- = 3, x_1^+, x_3^-\ge 0 $. Or in terms of the original variables: $ x_2 = -1 $, Any $ x_1, x_3\ satisfying\ x_1 \ge 0, x_3 \le 0, |x_1| + |x_3| = 3 $.
In this case the objective function is $ -x_1^+ -x_1^- -x_2^+ -x_2^- -x_3^+ -x_3^- = -4 $.
Solution 2 :
$ x_i = x_i^+ - x_i^-, \ \ \ \ x_i^+ ,x_i^- \ge 0 $
$ |x_i| = x_i^+ + x_i^- $
Then the problem is converted into a linear programming problem:
$ min x_1^+ x_1^- +x_2^+ +x_2^- +x_3^+ +x_3^- \\ subject\ to \\ \begin{bmatrix} 1 & -1 & 1 & -1 & -1 & 1 \\ 0 & 0 & -1& 1 & 0 & 0 \end{bmatrix} \begin{bmatrix} x_1^+ \\ x_1^- \\ x_2^+ \\ x_2^- \\ x_3^+ \\ x_3^- \end{bmatrix} = \begin{bmatrix} 2 \\ 1 \end{bmatrix} \\ x_1^+ ,x_1^- ,x_2^+ ,x_2^- ,x_3^+ ,x_3^- \ge 0 $
$ \begin{array}{|c|c|c|c|c|c|c|} x_1^+ & x_1^- & x_2^+ & x_2^- & x_3^+ & x_3^- & b\\ \hline 1 &-1 & 1 &-1 &-1 & 1 & 2\\ 0 & 0 &-1 & 1 & 0 & 0 & 1\\ 1 & 1 & 1 & 1 & 1 & 1 & 0\\ \end{array} $
bring an artificial variable $ x_4 $
phase 1:
$ \begin{array}{|c|c|c|c|c|c|c|c|} x_1^+ & x_1^- & x_2^+ & x_2^- & x_3^+ & x_3^- & x_4 & b\\ \hline 1 &-1 & 1 &-1 &-1 & 1 & 0 & 2\\ 0 & 0 &-1 & 1 & 0 & 0 & 1 & 1\\ 0 & 0 & 0 & 0 & 0 & 0 & 1 & 0\\ \end{array} $
$ \begin{array}{|c|c|c|c|c|c|c|c|} x_1^+ & x_1^- & x_2^+ & x_2^- & x_3^+ & x_3^- & x_4 & b\\ \hline 1 &-1 & 1 &-1 &-1 & 1 & 0 & 2\\ 0 & 0 &-1 & 1 & 0 & 0 & 1 & 1\\ 0 & 0 & 1 &-1 & 0 & 0 & 0 &-1\\ \end{array} $
$ \begin{array}{|c|c|c|c|c|c|c|c|} x_1^+ & x_1^- & x_2^+ & x_2^- & x_3^+ & x_3^- & x_4 & b\\ \hline 1 &-1 & 0 & 0 &-1 & 1 & 1 & 3\\ 0 & 0 &-1 & 1 & 0 & 0 & 1 & 1\\ 0 & 0 & 0 & 0 & 0 & 0 & 1 & 0\\ \end{array} $
phase 2:
$ \begin{array}{|c|c|c|c|c|c|c|} x_1^+ & x_1^- & x_2^+ & x_2^- & x_3^+ & x_3^- & b\\ \hline 1 &-1 & 0 & 0 &-1 & 1 & 3\\ 0 & 0 &-1 & 1 & 0 & 0 & 1\\ 1 & 1 & 1 & 1 & 1 & 1 & 0\\ \end{array} $
$ \begin{array}{|c|c|c|c|c|c|c|} x_1^+ & x_1^- & x_2^+ & x_2^- & x_3^+ & x_3^- & b\\ \hline 1 &-1 & 0 & 0 &-1 & 1 & 3\\ 0 & 0 &-1 & 1 & 0 & 0 & 1\\ 0 & 2 & 2 & 0 & 2 & 0 &-4\\ \end{array} $
optimal solution: $ x^* = [3, -1, 0] $, cost = 4
(ii)
Solution:
The objective is the same as minimizing $ x_1^+ +x_1^- +x_2^+ +x_2^- +x_3^+ +x_3^- $
Therefore using the Asymmetric Form of Duality, the dual problem is: $ maximize\ \lambda^T \begin{bmatrix} 2 \\ 1 \end{bmatrix} \\ subject\ to \\ \lambda^T \begin{bmatrix} 1 & -1 & 1 & -1 & -1 & 1 \\ 0 & 0 & -1& 1 & 0 & 0 \end{bmatrix} \le \begin{bmatrix} 1 \\ 1 \\ 1 \\ 1 \\ 1 \\ 1 \end{bmatrix} $
After simplifying, this becomes:
$ maximize\ 2 \lambda_1 + \lambda_2 \\ subject\ to \\ |\lambda_1| \le 1 \\ |\lambda_1 - \lambda_2| \le 1 $
Therefore the optimal solution is $ \lambda_1 = 1, \lambda_2 = 2 $. In this case $ 2 \lambda_1 + \lambda_2 = 4 $.
