27.06.2013 Views

Evolution and Optimum Seeking

Evolution and Optimum Seeking

Evolution and Optimum Seeking

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

76 Hill climbing Strategies<br />

s (k) = 1 may lead to a point withaworse value of the objective function. The<br />

search diverges, e.g., when r 2 F (x (k) ) is not positive-de nite.<br />

It can happen that r 2 F (x (k) ) is singular or almost singular. The Hessian matrix<br />

cannot be inverted.<br />

Furthermore, it depends on the starting point x (0) whether a minimum, a maximum,<br />

or a saddle point is approached, or the whole iteration diverges. The strategy itself does<br />

not distinguish the stationary points with regard to type.<br />

If the method does converge, then the convergence is better than of linear order<br />

(Goldstein, 1965). Under certain, very strict conditions on the structure of the objective<br />

function <strong>and</strong> its derivatives even second order convergence can be achieved (e.g., Polak,<br />

1971) that is, the number of exact signi cant gures in the approximation to the minimum<br />

solution doubles from iteration to iteration. This phenomenon is exhibited in the solution<br />

of some test problems, particularly in the neighborhood of the desired extremum.<br />

All the variations of the basic procedure to be described are aimed at increasing the<br />

reliability of the Newton iteration, without sacri cing the high convergence rate. A distinction<br />

is made here between quasi-Newton strategies, which donotevaluate the Hessian<br />

matrix explicitly, <strong>and</strong>modi ed Newton methods, for which rst <strong>and</strong> second derivatives<br />

must be provided at each point. The only strategy presently known which makes use of<br />

higher than second order properties of the objective function is due to Biggs (1971, 1973).<br />

The simplest modi cation of the Newton-Raphson scheme consists of determining the<br />

step length s (k) by a line search in the Newton direction v (k) (Equation (3.28)) until the<br />

relative optimum is reached (e.g., Dixon, 1972a):<br />

F (x (k) + s (k) v (k) ) = min<br />

s fF (x (k) + sv (k) )g (3.30)<br />

To save computational operations, the second partial derivatives can be redetermined<br />

less frequently <strong>and</strong> used for several iterations. Care must always be taken, however, that<br />

v (k) always points \downhill," i.e., the angle between v (k) <strong>and</strong> ;rF (x (k) ) is less than<br />

90 0 . The Hessian matrix must also be positive-de nite. If the eigenvalues of the matrix<br />

are calculated when it is inverted, their signs show whether this condition is ful lled.<br />

If a negative eigenvalue appears, Pearson (1969) suggests proceeding in the direction of<br />

the associated eigenvector until a point isreached with positive-de nite r 2 F (x). Greenstadt<br />

(1967a) simply replaces negative eigenvalues by their absolute value <strong>and</strong> vanishing<br />

eigenvalues by unity. Other proposals have been made to keep the Hessian matrix positivede<br />

nite by addition of a correction matrix (Goldfeld, Qu<strong>and</strong>t, <strong>and</strong> Trotter, 1966, 1968<br />

Shanno, 1970a) or to include simple gradient steps in the iteration scheme (Dixon <strong>and</strong><br />

Biggs, 1972). Further modi cations, which operate on the matrix inversion procedure<br />

itself, have beensuggestedby Goldstein <strong>and</strong> Price (1967), Fiacco <strong>and</strong> McCormick (1968),<br />

<strong>and</strong> Matthews <strong>and</strong> Davies (1971). A good survey has been given by Murray (1972b).<br />

Very few algorithms exist that determine the rst <strong>and</strong> second partial derivatives numerically<br />

from trial step operations (Whitley, 1962 see also Wasscher, 1963c Wegge,<br />

1966). The inevitable approximation errors too easily cancel out the advantages of the<br />

Newton directions.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!