Академический Документы
Профессиональный Документы
Культура Документы
1
Line Search Algorithms
Katta G. Murty, IOE 611 Lecture slides
78
above facts). Define
0 = 0
r = r1 + 2r1, for r = 1, 2, . . .
79
Methods Using 3-pt. Brackets
> 2 and f ()
= f (2) new bracket is either of the above.
81
2 + /2 if 2 1 < 3 2
=
3 1 > , define
2 /2 if 2 1 > 3 2
82
Methods Using 2-pt. Brackets
83
Line Search Method Based on Piecewise Linear Ap-
proximation
85
Newtons Method for line search
f 0(r )
r+1 = r 00
f (r )
assuming all f 00(r ) > 0. Not suitable if 2nd derivative 0 at a
point encountered.
Suitable if a near opt. sol. known, then function is locally
convex in the nbhd. of opt.
86
Inexact Line Search Procedures
is step
1. Sufficient Rate of Descent Condition: If
Cond. is that average rate
length choosen, new pt. is x + y.
be specified fraction of initial rate
of descent from x to x + y
of decrease in theta(x) at x in direction y. Select (0, 1) and
satisfy:
require
f (0) + f
f () 0(0)
(2.1) f 0(0)
f 0()
for some selected (, 1). Sats step must be long enough that
is specified fraction of its magnitude
rate of change in f at
at 0.
(2.2) f (0) + f
f () 0(0)
89
A simple procedure to satisfy these conds. uses 0 < 1 < 1 < 2
(1 = 0.2, 2 = 2 commonly used). plays role of in above.
satisfy f ()
Cond. 1 is equivalent to requiring step length
= f (0) +
`() 1 f 0(0).
> `(2 )
Cond. 2 is enforced by requiring f (2 ) = f (0) +
1 f 0(0).
2
=
Step 2: Take 1
where t is the smallest integer > 1 for which
t2
90