This preview shows pages 1–4. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: EE236A (Fall 200708) Lecture 13 Convergence analysis of the barrier method • complexity analysis of the barrier method – convergence analysis of Newton’s method – choice of update parameter μ – bound on the total number of Newton iterations • initialization 13–1 Complexity analysis we’ll analyze the method of page 12–21 with • update t + = μt • starting point x ∗ ( t (0) ) on the central path main result : #Newton iters is bounded by O ( √ m log( ǫ (0) /ǫ )) (where ǫ (0) = m/t (0) ) caveats: • methods with good worstcase complexity don’t necessarily work better in practice • we’re not interested in the numerical values for the bound—only in the exponent of m and n • doesn’t include initialization • insights obtained from analysis are more valuable than the bound itself Convergence analysis of the barrier method 13–2 Outline 1. convergence analysis of Newton’s method for ϕ ( x ) = tc T x − m summationdisplay i =1 log( b i − a T i x ) (will give us a bound on the number of Newton steps per outer iteration) 2. effect of μ on total number of Newton iterations to compute x ∗ ( μt ) from x ∗ ( t ) 3. combine 1 and 2 to obtain the total number of Newton steps, starting at x ∗ ( t (0) ) Convergence analysis of the barrier method 13–3 The Newton decrement Newton step at x : v = −∇ 2 ϕ ( x ) − 1 ∇ ϕ ( x ) = − ( A T diag ( d ) 2 A ) − 1 ( tc + A T d ) where d = (1 / ( b 1 − a T 1 x ) , . . . , 1 / ( b m − a T m x )) Newton decrement at x : λ ( x ) = radicalBig ∇ ϕ ( x ) T ∇ 2 ϕ ( x ) − 1 ∇ ϕ ( x ) = radicalBig v T ∇ 2 ϕ ( x ) v = parenleftBigg m summationdisplay i =1 parenleftbigg a T i v b i − a T i x parenrightbigg 2 parenrightBigg 1 / 2 = bardbl diag ( d ) Av bardbl Convergence analysis of the barrier method 13–4 theorem. if λ = λ ( x ) < 1 , then ϕ is bounded below and ϕ ( x ) ≤ ϕ ( x ∗ ( t )) − λ − log(1 − λ ) 0.2 0.4 0.6 0.8 1 0.5 1 1.5 2 2.5 3 3.5 4 • if λ ≤ . 81 , then ϕ ( x...
View
Full
Document
This note was uploaded on 09/26/2009 for the course CAAM 236 taught by Professor Dr.vandenber during the Spring '07 term at Monmouth IL.
 Spring '07
 Dr.Vandenber

Click to edit the document details