slater s condition proof
play

Slaters condition: proof p* = inf x f(x) s.t. Ax = b, g(x) 0 - PowerPoint PPT Presentation

Slaters condition: proof p* = inf x f(x) s.t. Ax = b, g(x) 0 e.g., inf x 2 s.t. e x+2 3 0 A = e.g., A = Picture of set A L(y,z) = Nonconvex example Interpretations L(x, y, z) = f(x) + y T (Axb) + z T g(x)


  1. Slater’s condition: proof • p* = inf x f(x) s.t. Ax = b, g(x) ≤ 0 e.g., inf x 2 s.t. e x+2 – 3 ≤ 0 • A = e.g., A =

  2. Picture of set A L(y,z) =

  3. Nonconvex example

  4. Interpretations L(x, y, z) = f(x) + y T (Ax–b) + z T g(x) • Prices or sensitivity analysis • Certificate of optimality

  5. Optimality conditions • L(x, y, z) = f(x) + y T (Ax–b) + z T g(x) • Suppose strong duality, (x, y, z) optimal

  6. Optimality conditions • L(x, y, z) = f(x) + y T (Ax–b) + z T g(x) • Suppose (x, y, z) satisfy KKT: g(x) ≤ 0 Ax = b z ≥ 0 z T g(x) = 0 0 ∈ ∂ f(x) + A T y + ∑ i z i ∂ g i (x)

  7. Using KKT • Can often use KKT to go from primal to dual optimum (or vice versa) • E.g., in SVM: α i > 0 <==> y i (x iT w + b) = 1 T w for any such i • Means b = y i – x i – typically, average a few in case of roundoff

  8. Set duality • Let C be a set with 0 ∈ conv(C) • C* = { y | x T y ≤ 1 for all x ∈ C } • Let K = { (x, s) | x ∈ sC } • K* = { (y, t) | x T y + st ≥ 0 for all X ∈ K }

  9. What is set duality good for? • Related to norm duality • Useful for helping visualize cones

  10. Duality of norms • Dual norm definition ||y|| * = max • Motivation: Holder’s inequality x T y ≤ ||x|| ||y|| *

  11. Dual norm examples

  12. Dual norm examples

  13. Dual norm examples

  14. Cuboctahedron

  15. ||y|| * is a norm • ||y|| * ≥ 0: • ||ky|| * = |k| ||y|| * : • ||y|| * = 0 iff y = 0: • ||y 1 +y 2 || * ≤ ||y 1 || * + ||y 2 || *

  16. Dual-norm balls • { y | ||y|| * ≤ 1 } = • Duality of norms:

  17. Visualizing cones • Suppose we have some weird cone in high dimensions (say, K = S + ) • Often easy to get a vector u in K ∩ K* – e.g., I ∈ S + , I ∈ S + * = S + • Plot K ∩ { u T x = 1 } and K* ∩ { u T x = 1 } instead of K, K* – saves a dimension

  18. Visualizing S + • Say, 2 x 2 symmetric matrices • Add constraint tr(X T I) = 1 • Result: a 2D set

  19. What about 3 x 3? • 6 parameters in raw form • Still 5 after tr(X)=1 • Try setting entire diagonal to 1/3 – plot off-diagonal elements

  20. Visualizing 3*3 symmetric semidefinite matrices

  21. Multi-criterion optimization • Ordinary feasible region • Indecisive optimizer: wants all of

  22. Buying the perfect car $K 0-60 MPG

  23. x* Pareto optimal = Pareto optimality

  24. Pareto examples

  25. Scalarization • To find Pareto optima of convex problem:

Recommend


More recommend