linear programming review
Example
- primal model
max 3 x 1 + 5 x 2 s . t . x 1 ≤ 4 ( π 1 ) 2 x 2 ≤ 12 ( π 2 ) 3 x 1 + 2 x 2 ≤ 18 ( π 3 ) x 1 , x 2 ≥ 0 \\\max\ \ 3x_1+5x_2 \\ s.t.\ \ x_1\le4(\pi_1) \\ 2x_2\le 12(\pi_2) \\ 3x_1+2x_2\le 18 (\pi_3) \\ x_1, x_2\ge0 max 3x1+5x2s.t. x1≤4(π1)2x2≤12(π2)3x1+2x2≤18(π3)x1,x2≥0 - dual model
m i n 4 π 1 + 12 π 2 + 18 π 3 s . t . π 1 + 3 π 3 ≥ 3 2 π 2 + 2 π 3 ≥ 5 π 1 , π 2 , π 3 ≥ 0 \\ min\ \ 4\pi_1+12\pi_2+18\pi_3 \\ s.t.\ \ \pi_1+3\pi_3\ge3 \\ 2\pi_2+2\pi_3\ge 5 \\ \pi_1,\pi_2,\pi_3\ge0 min 4π1+12π2+18π3s.t. π1+3π3≥32π2+2π3≥5π1,π2,π3≥0
对偶问题
对偶理论
- max z = c x s . t . A x ≤ b x ≥ 0 \\\max\ \ z=cx \\ s.t. \ \ Ax\le b \\ x\ge 0 max z=cxs.t. Ax≤bx≥0
- min w = b π s . t . A π ≥ c π ≥ 0 \\ \min w=b\pi \\s.t.\ \ A\pi\ge c \\ \pi \ge 0 minw=bπs.t. Aπ≥cπ≥0
- 弱对偶
x , π x,\pi x,π feasible ⇒ z = c x ≤ b π = w \Rightarrow z=cx\le b\pi=w ⇒z=cx≤bπ=w - 强对偶
x ∗ , π ∗ x^*,\pi^* x∗,π∗ optimal ⇒ z ≤ z ∗ = c x ∗ = b π ∗ = w ∗ ≤ w \Rightarrow z\le z^* =cx^* = b\pi^* =w^* \le w ⇒z≤z∗=cx∗=bπ∗=w∗≤w
对偶问题和原问题的解
- finite objective——finite objective
- unbounded——infeasible
max x s . t . x ≥ 1 x ≥ 0 − − − − − − − − − min y s . t . y ≥ 1 y ≤ 0 \\ \max\ \ x \\ s.t. \ \ x\ge 1 \\ x\ge 0 \\ --------- \\ \min \ \ y \\ s.t. \ \ y\ge1 \\ y\le0 max xs.t. x≥1x≥0−−−−−−−−−min ys.t. y≥1y≤0 - infeasible——infeasible
max 2 x 1 − x 2 s . t . x 1 − x 2 ≤ 1 − x 1 + x 2 ≤ − 2 x 1 . x 2 ≥ 0 − − − − − − − − − min y 1 − 2 y 2 s . t . y 1 − y 2 ≥ 2 − y 1 + y 2 ≥ − 1 y 1 , y 2 ≥ 0 \\ \max \ \ 2x_1-x_2 \\ s.t. \ \ x_1-x_2 \le 1 \\ -x_1+x_2 \le -2 \\ x_1.x_2 \ge 0 \\--------- \\ \min \ \ y_1-2y_2 \\ s.t. \ \ y_1-y_2\ge 2 \\ -y_1+y_2\ge -1 \\ y_1,y_2\ge 0 max 2x1−x2s.t. x1−x2≤1−x1+x2≤−2x1.x2≥0−−−−−−−−−min y1−2y2s.t. y1−y2≥2−y1+y2≥−1y1,y2≥0