27.06.2013 Views

Evolution and Optimum Seeking

Evolution and Optimum Seeking

Evolution and Optimum Seeking

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

A Multimembered <strong>Evolution</strong> Strategy 139<br />

With the new quantities ' <strong>and</strong> , Equation (5.24) for v becomes<br />

p !<br />

2 n<br />

v = erf ; p<br />

2<br />

"<br />

1 ; exp<br />

n<br />

; 2 n2<br />

!#<br />

2<br />

Since the argument of the error function increases as n, thenumber of variables, the<br />

approximation<br />

erf(y) ' 1 ; 1<br />

p exp (;y<br />

y 2 )<br />

can be used to give<br />

<strong>and</strong> with<br />

nally<br />

v =1; n p 2<br />

lim<br />

n!1<br />

1+ 1<br />

n<br />

v 1;n =exp<br />

The desired relation = (' )isthus<br />

=1+<br />

p ~'<br />

p 2<br />

exp<br />

2<br />

4<br />

~'<br />

p2<br />

in which, from Equation (5.27),<br />

~' =<br />

! 2 3<br />

5<br />

"<br />

erf<br />

n<br />

p 2<br />

~'<br />

p2<br />

'<br />

for n 1<br />

= e<br />

!<br />

!<br />

1 ; h 1 ; exp ;p 2<br />

+ 2 exp<br />

i<br />

p 2<br />

!<br />

#<br />

; 1<br />

(5.29)<br />

Pairs of values obtained iteratively are shown in Figure 5.13 together with simulation<br />

results for the cases of \survival" <strong>and</strong> \extinction" of the parent (n = 100 b = 100,<br />

average over 10 000 successful generations).<br />

As in the case of the sphere model, the deviations can be attributed to the simplifying<br />

assumptions made in deriving the approximate theory. For = 1' is always zero if<br />

the parent is not included in the selection. The transition to the inclined plane model is<br />

correctly reproduced in this respect. Negative rates of progress cannot occur.<br />

The position of the maxima ' max = ' ( = opt) at constant are obtained in the<br />

same way as for the sphere model. The condition to be added to Equation (5.29) is<br />

c exp (; + )[1; exp (; + )] ;1 ; 1<br />

h erf(' + )+2exp( + ) ; 1 ih 1+2' +2 i + 2<br />

p ' + exp (;' +2 )<br />

in which the following new quantities are introduced again for compactness:<br />

+<br />

!<br />

!<br />

+2exp( + ) ! =0<br />

(5.30)

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

Saved successfully!

Ooh no, something went wrong!