1 chapter 5 one dimensional search. 2 chapter 5 unidimensional search (1) if have a search...

28
1 Chapter 5 Chapter 5 One Dimensional Search

Upload: zoe-clark

Post on 18-Jan-2016

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

1

Ch

apte

r 5

Chapter 5

One Dimensional Search

Page 2: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

2

Ch

apte

r 5

Unidimensional Search

(1) If have a search direction, want to minimize in that direction by numerical methods

(2) Search Methods in General2.1. Non Sequential – Simultaneous evaluation of f at n points – no good (unless on parallel computer).2.2. Sequential – One evaluation follows the other.

Page 3: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

3

Ch

apte

r 5

(3) Types of search that are better or best is often problem dependent. Some of the types are:

a. Newton, Quasi-Newton, and Secant methods.b. Region Elimination Methods (Fibonacci, Golden

Section, etc.).c. Polynomial Approximation (Quadratic Interpolation,

etc.).d. Random Search

(4) Most methods assume (a) a unimodal function, (b) that the min is bracketed at the start and (c) also you start in a direction that reduces f.

Page 4: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

4

Ch

apte

r 5

To Bracket the Minimum

)()( untilx doubling Continue

)()( Compute .2

2let ),()( If

let ),()( If

)( and )( Compute 1.

)1()0()()0(

)0()1(

)0()0(

)0()0(

0)0(

kk

NEW

OLDNEW

OLDNEW

xxfxxf

xxfxf

xxxfxxf

xxxfxxf

xxfxf

Page 5: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

5

Ch

apte

r 5

points)closest

the(using f(x) minimum thegivingpoint on the

bracket a keep you to enables point that theDiscard

.,,, points spacedequally 4 have nowYou

)( Compute 3.

)1()2()2

12()3(

)2()1(

xxxx

xxf kk

Page 6: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

6

Ch

apte

r 5

1. Newton’s Method

Newton’s method for an equation is

)(

)(

)(

)()(

0))(()()(

0

00

0

00

000

xf

xfxxor

xf

xfxx

xxxfxfxf

Application to Minimization

The necessary condition for f(x) to have a local minimumis f′(x) = 0. Apply Newton’s method.

)(

)()(

)()()1(

k

kkk

xf

xfxx

Page 7: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

7

Ch

apte

r 5

ExamplesMinimize

2

1)0(

2

1)0(

2

)0(21)0()1(

2

21

2210

222

2

2)(

2)(

)(

a

ax

a

ax

a

xaaxx

axf

xaaxf

xaxaaxf

Minimize

Continue

x

x

xxxx

xxf

xxxf

xxxf

100.0212

231,1at xStart

212

23

212)(

24)(

1)(

)1((0)

2

4)0()1(

2

3

24

Page 8: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

8

Ch

apte

r 5

Advantages of Newton’s Method

(1) Locally quadratically convergent (as long as f′(x) is positive – for a minimum).

(2) For a quadratic function, get min in one step.

Disadvantages

(1) Need to calculate both f′(x) and f″(x)(2) If f″(x)→0, method converges slowly(3) If function has multiple extrema, may not converge

to global optimum.

Page 9: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

9

Ch

apte

r 5

2. Finite-Difference Newton Method

Replace derivatives with finite differences

2

)()1(

)()(2)(2

)()(

hhxfxfhxf

hhxfhxf

xx kk

DisadvantageNow need additional function evals (3 here vs. 2 for Newton)

Page 10: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

10

Ch

apte

r 5

3. Secant(Quasi-Newton) Method

Analogous equation to (A) is

)(0)()( )()( Bxxmxf kk

The secant approximates f″(x) as a straight line

)()(

)()(

)()()1(

)()(

)()(

)()(

)(

)()(

pq

pq

kkk

pq

pq

xx

xfxf

xfxx

xx

xfxfm

Page 11: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

11

Ch

apte

r 5

Start the Secant method by using 2 points spanning x at which first derivatives are of opposite sign.

For next stage, retain either x(q) or x(p) so that the pair of derivatives still have opposite sign.

Page 12: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

12

Ch

apte

r 5

Order of Convergence

Can be expressed in various ways. Want to consider how

10*)(

*)1(

*)(

ccxx

xx

Linear

kasxx

k

k

k

usually slow in practice

Page 13: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

13

Ch

apte

r 5

10*)(

*)1(

pccxx

xx

POrder

pk

k

Fastest in practiceIf p = 2, quadratic convergencep = 1.32 ?

)0(0lim*)(

*)1(

kascandcr

xx

xx

rSuperlinea

kkk

k

k

Usually fast in practice

Some methods can show theoretically whatthe order is.

Page 14: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

14

Ch

apte

r 5

Page 15: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

15

Ch

apte

r 5

Page 16: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

16

Ch

apte

r 5

Page 17: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

17

Ch

apte

r 5

Page 18: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

18

Ch

apte

r 5

Page 19: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

19

Ch

apte

r 5

Quadratic Interpolation

Approximate f(x) by a quadratic function.Use 3 points

2333

2222

2111

321

2

)(

)(

)(

)(),(,2

*20)(:

)(

cxbxaxf

cxbxaxf

cxbxaxf

xfxf)f(xc

bxsocxbxfMinimize

cxbxaxf

c b,a, for equations ussimultaneo 3 Solve

points3theatf(x)Evaluate

Page 20: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

20

Ch

apte

r 5

21 1 1 1

22 2 2 2

23 3 3 3

2 21 1 1 1

2 22 2 2 2

2 23 3 3 3

1 ( ) 1 ( )1 ( ) 1 ( )1 ( ) 1 ( )

1 1

1 1

1 1

f x x x f xf x x x f xf x x x f x

b cx x x x

x x x x

x x x x

(or use Gaussian elimination)

Page 21: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

21

Ch

apte

r 5

2 2 2 22 3 3 2 1 3 3 1

2 21 2 2 1

2 2 2 2 2 21 3 2 2 3 1 3 2 1

2 2 2 2 21 2 3 2 3 1 3 2

1 ( ) ( ) 1 ( ) ( )

1 ( ) ( ) : Numerator

( ) ( ) ( ) ( )

( )( ) ( )( ) ( )( ) :

Denominator

b f x x f x x f x x f x x

f x x f x x

c f x x x f x x x f x x x

f x x x f x x x f x x x

c

bx

2*

Page 22: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

22

Ch

apte

r 5

Page 23: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

23

Ch

apte

r 5

Page 24: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

24

Ch

apte

r 5

Page 25: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

25

Ch

apte

r 5

Page 26: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

26

Ch

apte

r 5

Page 27: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

27

Ch

apte

r 5

Page 28: 1 Chapter 5 One Dimensional Search. 2 Chapter 5 Unidimensional Search (1) If have a search direction, want to minimize in that direction by numerical

28

Ch

apte

r 5