Scribd
Upload
4 views2 pages

Convex Sets, Functions, Fenchel Duality, and Linear Programming

The document discusses the fundamentals of linear programming (LP), including the formulation of primal and dual problems, feasibility, and boundedness. It covers Farkas' Lemma, duality principles, and the simplex algorithm for solving LP problems. Additionally, it explains standard LP forms and basic solutions, emphasizing the convex nature of LP problems.

Uploaded by

Rana AL ALI
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
4 views2 pages

Convex Sets, Functions, Fenchel Duality, and Linear Programming

The document discusses the fundamentals of linear programming (LP), including the formulation of primal and dual problems, feasibility, and boundedness. It covers Farkas' Lemma, duality principles, and the simplex algorithm for solving LP problems. Additionally, it explains standard LP forms and basic solutions, emphasizing the convex nature of LP problems.

Uploaded by

Rana AL ALI
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Convex Sets, Functions, Fenchel Duality, and Linear Programming

Canonical Linear Programming (LP) Problem


Let K, N ∈ N∗ , L ∈ RK×N , b ∈ RK , and c ∈ RN .

Primal and Dual Problems


• Primal-LP:
min ⟨c, x⟩ subject to Lx ≥ b
x∈RN
+

• Dual-LP:
max ⟨b, y⟩ subject to L⊤ y ≤ c
y∈RK
+

The dual LP is the Fenchel dual of the primal LP.

Other LP Forms
• Equality constraint Lx = b can be rewritten as:
   
L b
x≥
−L −b

• If x ∈ RN (no non-negativity constraint), decompose x = x+ − x− with x+ , x− ∈ RN


+:

x̃ = (x+ , x− ) ∈ R2N
+ , c̃ = (c⊤ , −c⊤ )⊤ , L̃ = (L, −L)

min ⟨c̃, x̃⟩ subject to L̃x̃ ≥ b


x̃∈R2N
+

Feasibility and Boundedness


• Domain of the primal: A = {x ∈ RN
+ | Lx ≥ b}

• Domain of the dual: B = {y ∈ RK ⊤


+ | L y ≤ c}

• Primal is feasible if A ̸= ∅; bounded if inf x∈A ⟨c, x⟩ > −∞

• Dual is feasible if B ̸= ∅; bounded if supy∈B ⟨b, y⟩ < +∞

A and B are polyhedra, hence LP problems are convex optimization problems.

Farkas’ Lemma and Duality


Farkas Lemma: Either ∃x ∈ RN with Ax ≥ 0 and ⟨c, x⟩ < 0, or ∃y ∈ RK + such that A y = c.
⊤

Weak Duality: If x ∈ A and y ∈ B, then ⟨b, y⟩ ≤ ⟨c, x⟩.


Strong Duality: If both primal and dual are feasible, then µ = −µ∗ .
Complementary Slackness: (x̂, ŷ) is primal-dual optimal iff:

⟨c, x̂⟩ = ⟨b, ŷ⟩


[b − Lx̂]j ŷj = 0 ∀j
[L⊤ ŷ − c]i x̂i = 0 ∀i
Standard LP Form and Basic Solutions
Standard form:
min ⟨d, z⟩ subject to Az = b
z∈RM
+

Extreme Point: z ∈ C is an extreme point if z = (u + v)/2 with u, v ∈ C ⇒ u = v.


Basic Solution: z is basic if:

• z = 0; or

• I0 = {i | zi ̸= 0} and {ai }i∈I0 is linearly independent

Simplex Algorithm
−1 ⊤
Reduced cost: r = dJ − A⊤
J (AI ) dI
Steps:

1. Initialize with basic feasible solution w for index set I; let J = {1, ..., M } \ I

2. Compute r; if r ≥ 0, w is optimal

3. Choose j ∈ J with rj < 0

4. Compute ∆(j) = A−1


I aj

5. If ∆(j) ≤ 0, problem is unbounded


(j)
6. Else, compute step size zj = mini∈I˜j wi /∆i

7. Update:

wI ← wI − zj ∆(j)
wj ← z j
I ← (I \ {ij }) ∪ {j}, J ← {1, ..., M } \ I

Termination: If no degenerate vertex, simplex terminates in finite steps.


Initialization Trick: Minimize vj with Az + Dv = b, v ≥ 0, to get starting vertex.
P

You might also like