{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

Lecture 15 - Properties of the Simplex Method

# Lecture 15 - Properties of the Simplex Method - Recap the...

This preview shows pages 1–4. Sign up to view the full content.

Lecture 15 Properties of the Simplex Method UCSD Math 171A: Numerical Optimization Philip E. Gill http://ccom.ucsd.edu/~peg/math171 Friday, February 13th, 2009 Recap: the simplex method The simplex method moves from one vertex to another while reducing the objective function. Each iteration requires the solution of two systems of equations: A T k λ k = c and A k p k = e s where A k is the nonsingular working-set matrix. UCSD Center for Computational Mathematics Slide 2/46, Friday, February 13th, 2009 Example: minimize x R n c T x subject to Ax b where c = 2 1 ! , A = 1 1 0 1 1 0 - 1 0 - 1 1 , b = 1 0 0 - 2 - 2 UCSD Center for Computational Mathematics Slide 3/46, Friday, February 13th, 2009 x 1 x 2 x * #3 #1 #2 #5 #4 x 0 x 1 = x 0 + α 0 p 0

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

View Full Document
In this example, the simplex method moves to an adjacent vertex . We must show that A k +1 satisfies the properties of a working set. Result If A k is nonsingular, then A k +1 is nonsingular. Proof: At the end of iteration k , row s of A k is replaced by a T t . A k +1 = a T w 1 a T w 2 . . . a T t . . . a T w n row s UCSD Center for Computational Mathematics Slide 5/46, Friday, February 13th, 2009 The vectors a w 1 , a w 2 , . . . , a w s - 1 , a w s +1 , . . . , a w n are linearly independent. If A k +1 is singular, then row a t must be dependent on a w 1 , a w 2 , . . . , a w s - 1 , a w s +1 , . . . , a w n , i.e., a t = X i 6 = s a w i y i for some { y i } not all zero Multiplying by p T k gives a T t p k = X i 6 = s ( a T w i p k ) y i UCSD Center for Computational Mathematics Slide 6/46, Friday, February 13th, 2009 From the previous slide: a T t p k = X i 6 = s ( a T w i p k ) y i But p k is defined so that A k p k = e s , i.e., a T w i p k = 0 for all i 6 = s a T w s p k = 1 a T t p k = 0. UCSD Center for Computational Mathematics Slide 7/46, Friday, February 13th, 2009 If a T t p k = 0 then a T t x b t is not decreasing along p k a T t x b t cannot be a blocking constraint a contradiction. A k +1 is nonsingular. UCSD Center for Computational Mathematics Slide 8/46, Friday, February 13th, 2009
Two-dimensional example a 1 a 2 x k p k a T 2 p k = 0 a T 2 x b 2 is never blocking In general, there may be many working-set constraints with negative multipliers. Which index do we choose for s ? c T p k = d d α c T ( x k + α p k ) α =0 = initial rate of decrease of ( x ) = ( A T k λ k ) T p k = ( λ T k A k ) p k = λ T k ( A k p k ) = λ T k e s = ( λ k ) s < 0 UCSD Center for Computational Mathematics Slide 10/46, Friday, February 13th, 2009 From the previous slide c T p k = ( λ k ) s < 0 Choose the most negative ( λ k ) i This is called the “Dantzig rule” . c T x k +1 = c T x k + α k c T p k | {z } biggest change in per unit step Each choice of s gives a different direction p k .

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.

{[ snackBarMessage ]}