17ipm

# 17ipm - EE236C(Spring 2008-09 17 Primal-dual interior-point...

This preview shows pages 1–5. Sign up to view the full content.

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: EE236C (Spring 2008-09) 17. Primal-dual interior-point methods • cone programming • logarithmic barrier function and central path • symmetrization • Nesterov-Todd scaling • path-following algorithm • quadratic cone program 17–1 Cone program primal problem minimize c T x subject to Gx + s = h Ax = b s followsequal s followsequal is inequality with respect to proper convex cone C dual problem maximize − h T z − b T y subject to A T y + G T z + c = 0 z followsequal * z followsequal * is generalized inequality with respect to dual cone C * Primal-dual interior-point methods 17–2 Primal-dual path-following methods closely related to barrier methods (EE236B) • methods follow central path to find approximate solution • steps are computed by linearizing central path equations incorporate several modifications to improve efficiency and robustness • symmetric treatment of primal and dual iterates • barrier parameter is updated after each Newton step; no distinction between outer and inner iterations • more aggressive step sizes • infeasible starting points • higher-order approximation of central path • (some algorithms) detect infeasibility Primal-dual interior-point methods 17–3 Three standard cones we will limit the discussion to three types of cones nonnegative orthant { u ∈ R p | u i ≥ , i = 1 , . . . , p } second-order cone { ( u , u 1 ) ∈ R × R p- 1 | bardbl u 1 bardbl 2 ≤ u } semidefinite cone { u ∈ R p ( p +1) / 2 | mat ( u ) followsequal } ( mat ( u ) is symmetric matrix of order p constructed from u ; see below) these cones are self-dual, so we can drop the subscript in z followsequal * Primal-dual interior-point methods 17–4 Notation for symmetric matrices symmetric matrix as a vector if U = U 11 U 21 ··· U p 1 U 21 U 22 ··· U p 2 . . . . . . . . . . . . U p 1 U p 2 ··· U pp ∈ S p , then vec ( U ) = √ 2 parenleftbigg U 11 √ 2 , U 21 , . . . , U p 1 , U 22 √ 2 , U 32 , . . . , U p 2 , . . . , U pp √ 2 parenrightbigg • vec ( U ) ∈ R p ( p +1) / 2 are lower triangular entries in column-major order • off-diagonal entries are scaled by √ 2 so that tr ( UV ) = vec ( U ) T vec ( V ) Primal-dual interior-point methods 17–5 vector as symmetric matrix if u = ( u 1 , u 2 , . . . , u p ( p +1) / 2 ) ∈ R p ( p +1) / 2 , then mat ( u ) = 1 √ 2 √ 2 u 1 u 2 ··· u p u 2 √ 2 u p +1 ··· u 2 p- 1 . . . . . . u p u 2 p- 1 ··· √ 2 u p ( p +1) / 2 • mat ( u ) ∈ S p is the symmetric matrix of order p constructed from u • off-diagonal entries are scaled by √ 2 so that u T v = tr ( mat ( u ) mat ( v )) Primal-dual interior-point methods 17–6 semidefinite constraints in vector notation • suppose C is the cone of vectorized positive semidefinite matrices C = { u ∈ R p ( p +1) / 2 | mat ( u ) followsequal } = { vec ( U ) | U ∈ S p + } • for G i ∈ S p , h ∈ S p , define G = bracketleftbig vec (...
View Full Document

{[ snackBarMessage ]}

### Page1 / 15

17ipm - EE236C(Spring 2008-09 17 Primal-dual interior-point...

This preview shows document pages 1 - 5. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online