Cubic 2

in Eq. (5.55).
This inequality is satisfied automatically since A and B are se-

lected such that/^ < 0 and f'B > 0. Furthermore, the sufficiency condition
(when A = 0) requires that Q > 0, which is already satisfied. Now we com-
pute X* using Eq. (5.57) and proceed to the next stage.
Stage 4. The value of X* found in stage 3 is the true minimum of h(k) and
may not be close to the minimum of/(X). Hence the following convergence
criteria can be used before choosing X* « X*:
A(X*) - / ( X * )
/(X*) *e' (5 60)
-
^- = S7V/ < C2 (5.61)
where e{ and e2 are small numbers whose values depend on the accuracy de-
sired. The criterion of Eq. (5.61) can be stated in nondimensional form as
S7V/
s E <5 62)
iSfWF * ' -
If the criteria stated in Eqs. (5.60) and (5.62) are not satisfied, a new cubic
equation
h'(\) = af + b'\ + c'X2 + d'\3
can be used to approximate/(X). The constants a', b'', cr, and d' can be
evaluated by using the function and derivative values at the best two points
out of the three points currently available: A, B, and X*. Now the general
formula given by Eq. (5.54) is to be used for finding the optimal step size
X*. If/'(X*) < 0, the new points A and B are taken as X* and B, respectively;
otherwise [if/'(X*) > 0], the new points A and B are taken as A and X*, and
Eq. (5.54) is applied to find the new value of X*. Equations (5.60) and (5.62)
are again used to test for the convergence of X*. If convergence is achieved,
X* is taken as X* and the procedure is stopped. Otherwise, the entire procedure
is repeated until the desired convergence is achieved.
The flowchart for implementing the cubic interpolation method is given in
Fig. 5.17.
Example 5.11 Find the minimum of/ = X5 — 5X3 — 2OX + 5 by the cubic
interpolation method.
Enter with X, S and t0
Set fA = AO), a = f (0) and A = 0

Z indicates counter for
number of refits
SetfB= f[t0) and fB=Hh)
Yes Set I = 0
IS/B>0?
B=t0
No
Calculate
Set Set/W B
Z = (3(fA -Z 5 V(B-A)) + f'A + f'B
^o = 2£Q f'A-f'B
A = t0 Q = a2-rArB)j
X* = A + (f'A + Z ± Q)(B-AWA + f'B+ 2Z)
Set Z = Z + 1
Is|^)-^)U
No
/(X*)
Yes
Taker = X* Yes is No
and Stop i£f,i~r^
Yes
ls/>/ m a x ?
No
Yes Set
Is S ^ V ^ l x .
fA = fix*)
No
Set _ fk=m*)
B = X*
fB=f^}
Figure 5.17 Flowchart for cubic interpolation method.

SOLUTION Since this problem has not arisen during a multivariable opti-
mization process, we can skip stage 1. We take A = O and find that
^-
dX
(X = A = 0) = 5X4 - 15X2 - 20 = -20 < 0
x=o
To find B at which df/d\ is nonnegative, we start with t0 = 0.4 and evaluate

the derivative at t0, 2t0, 4t0, . . .. This gives
/'(*o = 0.4) = 5(0.4)4 - 15(0.4)2 - 20.0 = -22.272

f(2t0 = 0.8) = 5(0.8)4 - 15(0.8)2 - 20.0 = -27.552
/'(4f0 = 1.6) = 5(1.6)4 - 15(1.6)2 - 20.0 = -25.632
/'(8*o = 3.2) = (3.2)4 - 15(3.2)2 - 20.0 = 350.688
Thus we find thatf
A = 0.0, fA = 5.0, f'A = -20.0

B = 3.2, fB = 113.0, f'B = 350.688
A < X* < B
Iteration 1
To find the value of X* and to test the convergence criteria, we first compute
Z and Q as:
3(5 1130)
Z = '° ~ - 20.0 + 350.688 = 229.588
Q = [229.5882 + (20.0) (350.688)] " 2 = 244.0
Hence
-, - ( - 2 0 - 0 + 229.588 + 244.0 \
X = 3 2
* " ( - 2 0 . 0 + 350.688 + "459.176J = L84 OT
~0A396
By discarding the negative value, we have
X* = 1.84
1
As/' has been found to be negative at X = 1.6 also, we can take A = 1.6 for faster convergence.
Convergence criterion: If X* is close to the true minimum, X*, then/'(X*)
= df(k*)/d\ should be approximately zero. Since/' = 5X4 - 15X2 - 20,
/'(X*) = 5(1.84)4 - 15(1.84)2 - 20 = -13.0
Since this is not small, we go to the next iteration or refitting. As/'(X*) < 0,
we take A = X* and
fA =/(X*) = (1.84)5 - 5(1.84)3 - 20(1.84) + 5 = -41.70
Thus
A = 1.84, fA = -41.70, f'A = -13.0

B = 3.2, fB = 113.0, f'B = 350.688
A < X* < B
Iteration 2
Z = 3 4
^ On 7 J - I 1 QA ^ " 13
° + 350 688 =
- -3.312
Q = [(-3.312) 2 + (13.0)(350.688)]l/2 = 67.5
Hence
Convergence criterion:
/'(X*) = 5.0(2.05)4 - 15.0(2.05)2 - 20.0 = 5.35
Since this value is large, we go the next iteration with B = X* = 2.05 [as
/'(X*) > 0] and
fB = (2.05)5 - 5.0(2.05)3 - 20.0(2.05) + 5.0 = -42.90
Thus
A = 1.84, fA = -41.70, f'A = -13.00

B = 2.05, fB = -42.90, f'B = 5.35
A < X* < B
Iteration 3
3.0(-41.70 H- 42.90)
Z 13 + 5 35 9 49
- (2.05-!.84) " °° ' " '
Q = [(9.49)2 + (13.0)(5.35)]1/2 = 12.61
Therefore,
Convergence criterion:
/'(X*) = 5.0(2.0086)4 - 15.0(2.0086)2 - 20.0 = 0.855
Assuming that this value is close to zero, we can stop the iterative process and
take
X* ~ X* = 2.0086
5.12 DIRECT ROOT METHODS
The necessary condition for/(X) to have a minimum of X* is that/'(X*) =

0. The direct root methods seek to find the root (or solution) of the equation,
/'(X) = 0. Three root-finding methods—the Newton, the quasi-Newton, and
the secant methods—are discussed in this section.
5.12.1 Newton Method

Consider the quadratic approximation of the function/(X) at X = X, using the
Taylor's series expansion:
/(X) =f(\) +/'(X 1 )(X - X,) + l2f"(\)(\ - X,)2 (5.63)
By setting the derivative of Eq. (5.63) equal to zero for the minimum of/(X),
we obtain
/'(X) = /'(X/) + /"(X;)(X - \) = 0 (5.64)
If X1- denotes an approximation to the minimum of/(X), Eq. (5.64) can be

rearranged to obtain an improved approximation as

Cubic 2

Uploaded by

Copyright:

Available Formats

Cubic 2

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cubic 2

Uploaded by

Copyright:

Available Formats

in Eq. (5.55).

This inequality is satisfied automatically since A and B are se-

^- = S7V/ < C2 (5.61)

h'(\) = af + b'\ + c'X2 + d'\3

Set fA = AO), a = f (0) and A = 0

Figure 5.17 Flowchart for cubic interpolation method.

To find B at which df/d\ is nonnegative, we start with t0 = 0.4 and evaluate

/'(*o = 0.4) = 5(0.4)4 - 15(0.4)2 - 20.0 = -22.272

Thus we find thatf

A = 0.0, fA = 5.0, f'A = -20.0

Q = [229.5882 + (20.0) (350.688)] " 2 = 244.0

By discarding the negative value, we have

/'(X*) = 5(1.84)4 - 15(1.84)2 - 20 = -13.0

fA =/(X*) = (1.84)5 - 5(1.84)3 - 20(1.84) + 5 = -41.70

A = 1.84, fA = -41.70, f'A = -13.0

Q = [(-3.312) 2 + (13.0)(350.688)]l/2 = 67.5

/'(X*) = 5.0(2.05)4 - 15.0(2.05)2 - 20.0 = 5.35

fB = (2.05)5 - 5.0(2.05)3 - 20.0(2.05) + 5.0 = -42.90

A = 1.84, fA = -41.70, f'A = -13.00

/'(X*) = 5.0(2.0086)4 - 15.0(2.0086)2 - 20.0 = 0.855

5.12 DIRECT ROOT METHODS

The necessary condition for/(X) to have a minimum of X* is that/'(X*) =

5.12.1 Newton Method

/(X) =f(\) +/'(X 1 )(X - X,) + l2f"(\)(\ - X,)2 (5.63)

/'(X) = /'(X/) + /"(X;)(X - \) = 0 (5.64)

If X1- denotes an approximation to the minimum of/(X), Eq. (5.64) can be

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.