Scribd
Upload
265 views15 pages

Hooke Jeeves Method

The Hooke-Jeeves algorithm is a method for minimizing functions in multiple dimensions. It works by iteratively taking steps in the direction of improvement found from the previous step, rather than restarting from scratch at each new point. The key steps are: (1) test points around the initial point and save the best, (2) move in the direction of improvement until no further improvement is seen, (3) repeat steps 1 and 2 until no direction of improvement is found, then (4) reduce the step size and repeat the process.

Uploaded by

coro kakus
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
265 views15 pages

Hooke Jeeves Method

The Hooke-Jeeves algorithm is a method for minimizing functions in multiple dimensions. It works by iteratively taking steps in the direction of improvement found from the previous step, rather than restarting from scratch at each new point. The key steps are: (1) test points around the initial point and save the best, (2) move in the direction of improvement until no further improvement is seen, (3) repeat steps 1 and 2 until no direction of improvement is found, then (4) reduce the step size and repeat the process.

Uploaded by

coro kakus
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 15

Minimizing in 3D

Method 3: Hooke-Jeeves
The Premise
The Hooke-Jeeves algorithm, also known as
“pattern search”, keeps track of the direction of
travel as the process moves from point to point,
instead of starting over from scratch at each
new point.
The First Step
1. Test your initial point in one dimension only,
for example by testing (x1 + .1, x2) and (x1 - .1,
x2).

? ?

Save the best value of x1.


The First Step
1. Test your initial point in one dimension only,
for example by testing (x1 + .1, x2) and (x1 - .1,
x2).
?

?
Save the best value of x1.
2. Repeat for x2, using the best value of x1.
The First Step
1. Test your initial point in one dimension only,
for example by testing (x1 + .1, x2) and (x1 - .1,
x2). ?
? ?
?
Save the best value of x1.
2. Repeat for x2, using the best value of x1.
Now you have a new point made up of the
new values of x1 and x2.
Practice Problem 1
Write a program that, given a function f(x1, x2) and
an initial point, tests the function in all four
directions with a step of 0.1. Your program should
print:
1) The original point
2) The improved point
3) The vector between them, found by subtracting
(improved – original).

Test your code using f(x1, x2) = (x1 – 3)2 + (x2 + 1)2.
The Second Step
With the brute force method, we would just
repeat the process from the new point.
Instead, with Hooke-Jeeves, we use the vector
from the old point to the new point, and we
search in that direction until we stop seeing
improvement.
Practice Problem 2
Write a program that, given a function, a starting
point, and a vector, moves the point in the direction
of the vector until the function value stops
decreasing.

To test your code,


Use f(x1, x2) = (x1 – 3)2 + (x2 + 1)2
1. Choose a starting point and run it through your
code from problem 1 to find a vector.
2. Run the starting point and vector through your
code for problem 2.
3. Repeat until the vector in step 1 returns (0, 0).
The Third Step
Once you stop seeing improvement in the
direction of the original vector, you repeat the
first and second steps: cast about to find a
better point, then travel in that direction.

When the first step returns the original point as


the best point (in other words, the vector is 0),
then you move on to the next step.
Practice Problem 3
Using the function
f(x1, x2) = (x1 + x2)2 + (sin(x1 + 2))2 + (x2)2 + 10,
3a. Move between your first and second programs
(from problems 1 and 2), with an interval of 0.1,
until the vector returns 0. Transfer results between
programs by hand at first.

3b. Then combine your two sets of code into one


program that, given a function and starting point,
will run the first two steps of the Hooke-Jeeves
procedure and return the next starting point.
Repeat until the vector returns 0.
The Fourth Step
The fourth step is to decrease the incremental
interval. If you started by adding and subtracting
0.1, now you might add and subtract 0.01.

Then you repeat everything, including reducing


the interval, until you have reached the desired
level of accuracy and there is no more
improvement.
The Steps of Hooke-Jeeves
1. Cast about your original point to find the best
point in its neighborhood.
2. Travel along that vector until you stop seeing
improvement.
3. Repeat steps 1 and 2 until step 1 returns a
vector of (0, 0).
4. Reduce the interval.
5. Repeat steps 1-4 until the interval is small
enough to achieve the required tolerance.
The Steps of Hooke-Jeeves
Work in groups or alone to outline the basic
loop structures* of a program that would run
the entire Hooke-Jeeves method from start to
finish (plan the program; do not write the
program!):

*meaning: what kind of loop, where to start, when to stop, what


gets done inside the loop
Practice Problem 4
Insert your code from Problem 3 into a loop that
will run as long as the vector is nonzero.

Then, using f(x1, x2) = (x1 + x2)2 + (sin(x1 + 2))2 +


(x2)2 + 10, run your new code with an interval of
1 to get a new point; from that starting point
repeat with a new interval of .1; continue to
repeat, reducing intervals, until you reach an
interval of 0.0001.
Practice Problem 5
Insert your code from Problem 4 into a loop that
will run Hooke-Jeeves with successive interval
widths from 1 to 0.000001. This program, when
completed, will execute the entire Hooke-Jeeves
algorithm with one input line.

Test it on
f(x1, x2) = (x1 + x2)2 + (sin(x1 + 2))2 + (x2)2 + 10.
Document and save this program!

You might also like