Week 3: Programming: convex program, linear program
1 Convex programming
1.1 Definition
min x f ( x ) s . t . x ∈ C \begin{aligned} \min_x& \quad f(x)\\ s.t.& \quad x\in C \end{aligned} xmins.t.f(x)x∈C is convex if f f f is a convex function and C C C is a convex set.
1.2 Local, Global optimality
- Local Optimality ⇏ \nRightarrow ⇏ Global Optimality
- If convex programming, Local Optimality ⇔ \Leftrightarrow ⇔ Global Optimality
- Unconstrained programming, Local Optimality ⇔ \Leftrightarrow ⇔ ∇ f = 0 \nabla f=0 ∇f=0 ⇔ \Leftrightarrow ⇔ Global Optimality
2 Linear programming
2.1 Definition of LP
min x c T x s . t . A x ≤ b C x = d \begin{aligned} \min_x& \quad c^Tx\\ s.t.& \quad Ax\leq b\\ & \quad Cx=d \end{aligned} xmins.t.cTxAx≤bCx=d
- Feasible set of LP is a Polyhedron
- LP can be :
- No optimal solution
- Infeasible
- Unbounded
- Finite optimal solution
- No optimal solution
2.2 Optimality of LP
Feasible set P = { x : A x ≤ b , C x = d } P=\{x: Ax\leq b, Cx=d\} P={x:Ax≤b,Cx=d}
Extreme point
- Definition 1: Not the convex combination of any other two other points in the polytope.
If ∃ y , z ∈ P \exists y,z \in P ∃y,z∈P and λ ∈ [ 0 , 1 ] \lambda \in [0,1] λ∈[0,1], such that x = λ y + ( 1 − λ ) z x=\lambda y+(1-\lambda) z x=λy+(1−λ)z, then x x x is not extreme. - Definition 2: It is the unique optimum for some cost vector c c c.
Basic feasible point
- Active constraints: Constraint a i a_i ai is active at x x x if a i T x = b i a_i^Tx=b_i aiTx=bi.
- Active set: A x = { a i : a i T x = b i } ∪ { d i : c i T x = d } \mathcal{A}_x=\{a_i: a_i^Tx=b_i\}\cup\{d_i: c_i^Tx=d\} Ax={ai:aiTx=bi}∪{di:ciTx=d}.
- Basic feasible solution (BFS): x x x is BFS is its active set A x \mathcal{A}_x Ax has n linearly independent vectors.
Extreme point and Basic feasible point
-
The following are equivalent:
(1) x x x is a BFS;
(2) x x x satisfies definition 1 of BFS;
(3) x x x satisfies definition 2 of BFS; -
Possible to have no extreme point:
P has no extreme point ⇔ \Leftrightarrow ⇔ P contains a full line.
Optima and extreme point (BFS)
- If an LP
- (a) has a finite optimum, and
- (b) its contraint polytope has at least one extreme point,
- then there is an extreme point which is optimal.
2.3 Converting LPs
Homework
TODO