it ""'-""ilrt
I'otlotos itnnt'eilícr,td1¡tÌtal X*:
V*.aiuii,l
Ñnraton-I{cotttot'ouiclt' ctssunltttons,Gy¡lg c,rxl T'apiu l4l
as weli ás otlrcts
t51-
t10l proaiúed, th,e J'o\totiitr,g Íto:rmcl Jo:r Neuloru's ntel'ho¡J,t -
t.,uì- -
(1-
(]2lngz'-r Jor all
n, >- o,1-ozo
wlticlt, cant¿ot be ,ím4troued,. Thc¡,t
is
tl¿e o'¡'cler o,f con"uergen'crt oJ Neutott')s '¡n'el'- äotlis
l,wo tulrcre (¿ri (4.13)) tlte Ítcttleu-trVernef nt'eth.ocl lms ot'iler th¡'ee.t"i
Note utioi tiu.ct rue'lrc,ue shorun (by (a.13)) Lhe enisl'anca oJ infinitclu^,nrrrry'irrllrocl,s
for''øelil,2)
tuh,eret¡'c
up'per bou'tt'cls tt're Lesst¡un tlt'at
of Halley's 'ntetiJt'ocl(u : L
tlrcn).}IEIìI]RJìNCI'S
1, r\rg¡rfos, I. I\., ena<bQlfu crpuiioils antt tL¡:;pIítations lo Chrndtascithtu"s tut<i tdûlcdcqualions,
Íjtrll. r\rrstr:aì, lf aLlt. Soc., 3g (.t 9S5), 2-t5-292'
2, -, Ou tt cllss of ¡tottline¡rr it'tlcgral equirtiotts ctrisittcl irt ¡tt:nlrott !rttttsporl. Accluationcs l,IaLhcrn¿rtictrc, Jli (1988), 99-1 11'
3. Chcn, l)., Itlnloioniclt'-Osltóín:;1,i coilûaìgcn('c lhcorcnts and o¡'tlitnal u¡or ltotn<Is for,Iurntl's
itc't.atitie ntelhatls, lrtcru. .J. Coilrputcr J\talh., 3l (3 -'r 'l) (1!90)' 221 --2:li:.
4. GIagg, \\¡. ll. atrrl l'apia, lì. t\.,0'ptintcl error boun<ls fun'lhe Netulott-l{ct¡tlorc¡oiclt l.luorcn, SJÀ;U.f. Nunrcr'.,\nal., lt I (1974), 10-líl'
5. I{atrttuovich, L' V' antl Ákilor', G' })" lìrulc/inrt¡tl
"lrlnlu'sis i¡t \tort¡t¿tl Spalcs' Pcrgarrotr Prcss, Ncrv YoìÌi, 196!ì.
(ì. OsL'orvsiii, A. ìI., Sglrrliott o[ ]Ìqttcrlions nrrd ,S1,'s/rnts of lÌc1tt<tliotts, Âcarlcnric Ptcss, New
Yolk, 19(i(ì.
7. I)otr.a, ¡i.4. ¿¡rl l,ialt, \.., shru,p trror bctuntls for N¿¡D¿on,.r ¡nelht¡tl, Nurrct'. \lath., lÌ4,
(.1 9{i()), 63-72.
g. Wcr.ncr.,'\V,, Snrne intprcnentcnls ol'clttssìtul ìleì'(liúc ntdltotls fctr ilic søIulions of nottlitttar eclttrrliotts. l,cctÙ-r'e llOtes in I,Iat.li., Nrrrnci i¿:rù solttlíott of rtortlitteat cquali<ttrs. Ìtroccttlittgs, Illetttcn, 878, (19tì0), 'l'2'1 -'1'lO'
g. y¿¡lliarnoto, 'l'., À cnrttiir.eetu:c lhect¡e¡tt for Netuk;rt-lilr'c ntclJtotls in Bantu'lt .spcccs, Nunrct' l\'Iath., 5l (19t17), 545-:iir7.
10. Zabr.ejl<o, lr. lr. rn<l Ngìlell, ,t. 1;., ',l'Ìtt nrtjorturl ¡tttlltod in lhe lhtotu of -'\rurlon-linttlotopítlt trppt'ot:intulÌ<tns r¿ntl lltc Ptnlr r,r;ril eslitt¡¡tles. Nrtrutlr. lìtrtlct. ,\rirl . atrd Optinriz , f) (lf)ij7i,
l!71 -(:ill4'
tRÌìvulì Ð'AN/\t,ysE lrlllrfÉIltQuE
n'r
ÐIì T-IrÉ()Itììì IJH r,'ÀppÌt0xlur\T.lolrjX'omc 9Í1, Iiìo
I,
IÐÐ4, ¡rp. t5-p3ON A CLASS OF FITNESS FUNCTIONS FOR GENETIC AL GORITHMS USING
PROPORTIONAL SELECTION
RIIÁìì,ION-IjRNö BAI,ÁZS
(Cìu j-Napoca)
1. tN'rnonllf.'iloN
74 l. l(. Àrsvros, ìf . À. 'fabat¿rbai, l). (ìltctt
is
inaertittle ct"ìLd .frctnt, the estriltùGteP(X*) - P(r'r¡ -
P',(Y'r+ (X'r -
Y'F))d(x',¡- Y*),
lìeceilcd B X 19U2
^ -[n [2]
Grefenstetteand Bakel
cliscussedthe impact of the
fitrressftlirctions
onthe
behavioul of genel,ic algorithms.lhe¡i
shorved situations rn'hele Hollancl's schematheoremlsae
¡ã1¡ cloesnot
have a cleal iritelpr,e-tation
ancl su_ggested,for
alather
large class of genetic algor,itìrrns aiirn-
plebut
useful chalacterization ofthe implicit
pãrallelìsrnl- rn
the present pâpel we_ s1,udy a clais of fitness functionsfor
genetic algorithms usingproportionâl
seie¿tion.'Ihe first
sectjon containsîhe
re- hrns using a monotonic fjtness func-2. GNI,N[I'¡'IC AI,GOIII';'IINS USINç A NTONÛ'¡'(}NIc FI'|¡{IìSS IìUNC'I'TON
^ß*
r) A tf ON OTONIC S ll ¡,tr(ìTION
^LG(XrI'tìÉIlU
,
Selgcti_on is probablythe
rnostirnpoltant
stepin
a gen beca,nseit
clel,err'ines r.vhichinrlividrials rvill cdltribuie
amonnt)to
l,he creatjonof a
lrer.vpopulation. ¡ls jn [2]
rveselection
to
be par,titionedinto twô
steps :eàch
cÌ'ea-
rentlale on the selectjon algolithrn,
algolithrn
issiven.
it'lius ive sìra,lirdividual as .i,yell &s the effect of such er,planes.
ilhis
latter,rvjll lie
cìr¿lr,ac_morrent f by the targel, sam,pliu,g rale t,sr'(H,r)
: I ,.y Yy+,
,n(H, l)14
i ) t p il ! ù1 r:!i I ct [ ],[ ut hetnt.tI itrLI Scicltcrs, {:Qtlt¿l ott U ttioersilg
Lontlott, Oli 73505, Ii.S.A.
))r.ptttlmettÍ ol Ìlalh,5ci¿rlc¿s
l-lrt i¡r¿'¡.silil o I Ã t )i<i ti -<tt s
I¡ttUtllct¡iIl<, A Il 7 2'10 1, rJ'S.A
16 i\I. E. llalrizs
\'r'heÌe
n(Htt)
denotcsthe nurnlrcÌ of
repleseul,ati\¡esof
h¡'pet'planeIl in
the
populationat
rnonìerìtI
(clenntecl b,1. P(f)).rr'ol a
problem forrnulat'edin terms of an
objectir.efunctiotl
,f, thetarget
sarnpling rate is girrenb'l'
cornposing trvo functions:
a filn,ess ,frunc-tion
u,and a
selention alç¡ot'ith,nt, s,that
istu'(n,t) :
s('u(r, f), ú).For the ¡est of
this
paper rve sllall consiclelthat
the fitness does nt¡t d.êltenclolt
¿,i.e.
lsr(e;,Í) :
s(t¿(æ), l).In
sil,ual,ions s'hen theolljectivc
function is to be minirnizecl or whenit
can take on negativc l'a,lues the fitness frurction js obtainecl b1'a trans-lornration of tìle objective function
suchthat
u,(*):
/r(./(er)).llowcver
these atenot the
onl¡r ¡¿¿ootlsfor
using' f itness functjons. Às we shall see ,thev greatly
influencethe behavionr of
genetic algolithnrs.llre
t¡est hltou'n sclcctionalgolithm
is proportional selection defincclby
úsr(er,
f) -
tt'(n)il'(t)u'ltete,i7(l) denol,es
the
average fitncss of the indivjduals ir-L tltc.¡ropulation al, rnornentf.
Thefilst chalacterization of
gcnetic algorithnts,given by Hollancl [3] is
baseclon propoltional
selection.In [2]
Grefenstette and.Ilaket
shol'ecltbat
evcnfol verv
simplc fitness functions (linear ones) theinterpletation
of Hollancl 's Sclterna 'Ihe- orern isnot
clcar. Accoldingto
themthis
problemjs
clueto the fact
that,the
l,heotern refersto
the fitnessfunction,
lvhichis
a desìgn pararnetel ofthe
geneticalgolithm, instead of gir.ing a
chalacterizabionin terrns
ofthe obiective function. they
suggestthat
charactelizationsof
genel,icalg-olithms shoulcl
st,ate((hor.the
space dcfinedb]'the
objcctive funci,ionis
sealched Jrvthc
genctio algolithrn').In
the rest ofthis
sectiorr ure gir.e a shcllt plesental,ion ofthc
l'csultsin [2]. flo
makc cliscussion sirnplein
the follorvings s'e shall c,onsider' (rvit-hout
loss of genelalit¡r)that /
isto
bc maximized.Dot¡rtitttloN 1.1.
A
f,itness Ju,nction,u is
r¡'¿onotoníc iJ' thc follotoi,n'g condition, ltolds:T¿(ø)
(
u(u)i.fÎ /(r) <
/(y).Note 1.1.
As
shownin [2] this
classof
rnonotonic Ïii,ness functions inclucle manyfreqlentl-v
used fitness ïunctions, tlrusit
constitutes a rta-tural subject of
sturl¡..DnrrnnroN
1.2.A
selection algoritltm,is
n't o n' rs t o ni c iJ it
ussignsu
target sunpLin,g rate to each'ittdiaidual,
at,alty
momeltt s'uch tllo,ttsr(n,t) (
/sr(r¡,t) ifJ
u,(n)<
u(y).Note l.2..Plopor'1,icural
selcctìon,
selection b-v rattliìnp¡,as rvoll
asnrân.\'
other
l<norvn sclecbionalgorithms ale
monotonic.l.he.results
pr.eserrteclin the
follor.l,ingsrvill
concern genetic algo_ritltm,s usllg a
monotonic selectionalgoritrin and a
rnonolonic fitn"ess f-u19i]9n' 1.ogive the central
char-actefizationof tl'ris sectio¡
one rnoïedefinition is
neecled.Dn¡'rNrrron
l.B.
Ghen, the p.tpsttlntion,p(t)
Loe say r,ltnt the lryperltlaneH,
i.s d,o,t i,,
tt,t ccl
by the hy1te.íptaneHr,
ìtenotingit by A, {
o,,H,6
max {/(r)lr
ett, n p(r)} ç rnin{/(r) lr
e EI,n
p(ú)}.Non
$'e can gir.ethe following lesult (t2l)
:Trrnonnu 1.1
. rn
cr,ry1 ç¡enetic argotitrrn't, usinga
motroto,nic selection a,Igorithm andu
mt¡norc¡nic fitricss Juctctiott,,for
any "ltyperltl,a,nes, EIr,Í1,
,in,P(t)
E[, 4 ø,,H2+
Lsr(IIr, ü)ç
úsr(Er, t)t'he ploof of Theoren
1.1is
basecl onthe rlefinitions
andis
straightfor- wal'd.s a
x.eaker characterizationof
thenn¡'rx*rox
r.4.4,rtrlyterprcnt,e Fr,is
con,¡tr eterq
d ont,,ino,t,ed,b1y cr,ttotlret ltEperplane
Hr,
ãcnàtelt byH, < ;Hrii.f
ittnax{./(ø)ln, e
Hr,\ (
rnin{./(rr,-) ln etrIrl.
Tire follou'itrg
cor,olla,r.y holds :conolr,¡.nv r. r.
rn, cLny gener,i,c ctrgoritrnn u,s,itttlu
tnonotonic serection algorithnt cut,ll o, ntonototticfilnels
functioit,, Jor cnty h.yþe,rltlarnesH,
rnd, n{,IIr < Hz
=+ V(f)tsr(I{r, ú)( fsr(ãr,
t).lhis lesult
statesthat
unc{er,the
gir.en conditions -H, glorvsnt
rãu*t as_fasb as- .Hl doesin
arr.\¡ g^euerartion,in
any genebic aigoriihmof the
con- siderecl class.fn the
encl of llre,ir paper, Grefenstette anclBal
searchfol'
corrclitions n,hich allow chalactelizations offhe s
lectionalgolithms.
X'henext
sect,ion prcscnts soureof oul
dilec_tion.
3. GüL\Iì'I'IÍ .1\Lf,iOIüTIIiU,S USING t'Ii{)I,OtÈ'i,[ONl\L S]ìf,ntìl,ION
. trn the ìrlevious
sect,iona clmractcrization of
genet,ic algorithms usinga
Dronotoriic selectionalgolitlun and a
monobonic fibnessïunction
was given.
This
coversa
bloacì classof
genctic algorithms"sc,i i"
p.àc_2
-
c. 1080On Genctic Algor.itlrrrrs t7
tice. Ilol,ever
these l'esulls canlleither gile
bouncls onthe glorvth
of the¡epreserrt,a|irres
of
h¡'pelplanes,llol' captlr'e sensitiyit¡'
aspectstlÏ
genetrcalgolithns.
we
claiur tha1, one possiblc l'eàsonfor
these cleficiellcjesis thal' in the
clttss co.sicLeÌeclsoÌle;clection
aly,^olitlt'lSnla¡' "\'oì'ì(
.-tgàillst" Solne c1¡alitiesof celtain
fitncssfullctions.
'l'oillustlate this
lct, us collñiclel tnefitness function
æo
:
log(J(ø))- 1,
e<
J(t':){
a¿fol all
possìble o(rvhich
js sirlilar
19¿¿:
Ò- tog(/(r))
co¡sitlelecli1
l2.l) rvhele/ is to
bernaxirnizecl.
this
litnessfu¡ci,io1
has soure tlice pr-opertiesI'hich
rnakeit
aplleáù-Iing,
such asa. il,
r.ecluccsthe
clanger o1 pl',eÌra,trU'e Con\¡el'getlce ll.-v clatnping out cliffererrces l:ets'celrlalge
I'alucrs of .b.
mahes^
ts^t"*1, ãifferencebe
J'(r)-'Iìor a
g^enetic^lgnt:itht
anal
the
fitness fùnc1,ion ?l,o 1\¡eca
mo-notonic
sclecbionntg'oilttrins I ut
Let for
jnstancetbc
tar:ge1, satnplinelale
be clefinetlìly
üsro(r:'
q: !!-,
MÁT)
rvlre}o
][r(t) : ruar{./(r)ln
e P(t,)}.s[bstjtul,jug
?¿oftÌ'
¿¿\\t¡
o.btain¿rrn.tl
.il *.¡lst'o(;1,
,, :
*
rr,,: jl(Ð '
îhe
genetic algor,ithur using tsi'o aurl ir,o uses âlinear
selectjon rvhich laoksthe
a,bove-nen1'jonctl pl'ollerties.For
1,he l,càson lireniionecl ¿ncl illusl,ratecl above rveshall
clisoussgenetrc'-alg'ot'i1,hms
s,hich
usesplopoli,ional
selection. Obviouslythis
se-iu.6or-, aìgit1i1,ì¡n
is
rnonoton;c ãndin
atlclitioarrit
hasthe
follorvillg pro-On lìclrctic Âlgoli [ìrnrs 19
Note 3.1. Oþrriousl¡r
all
1,he lesultsin the
plevions section are v¿liclfor l'SGA-s using
;u rnotlotonic fit'nessfutlction'
r\s shou,lt;ñ
¡21, r'esultssirnilal to
those prese,nteclin thc
previoussoctjot
can beo'lrtaiñcci
ronotonic fitñess functions anclstrictly
rnonotonic selectr'onal he
follorvingsby
rnonotonic fitness iuncl,ion u,c shallrnean
t¿'¿c ones. OllviorLsl¡t pl'opoltion¿ì'l Se-lection js
sl,r'jtl,lvrloltotonic.
Tret
¡s
norv tr'.yto give
a chalactel'jzationfor llSGÄ-s
usitlg ar lnollo-tonicfitness function.lüi.h
propeltjcs sirnilal to those of tt'o a'bove.fn
ol'dLerto
c1ot}is
rve neecl sorne rnol'e ilefinitiolts.J)nnrrqrrroN
3.\.
lVe søu thctt ttis
rc cottlec) Jilness .furtctio'tt,iJ for
øn'y#t¡ fiz
ancl n,[./(ø,),
l(nr), :f@');
I'L)>
ona-here
ln, y, z; h,l
denoles lhe secontl r,trtler clíticlect difference' aJ t'he tun'ctionIt in, nr,
fizt ïs.Tlrc fit'ness Junctiott, t¿ i,s su,ícl, Io be cottcctue
if
Jr'traN!
tr1, n, and n"l/(rr),
J(.t:r),J(rr); øl <
0.the followins
trvo theoÏerusgive
aplopelty
of PSG-{-s usìng a mo-notonic
ancl convex, r,espectivcly .-(]oncavefitness
furtction,
rvhich mahes them jnteresting frotn thepoint
of vierv of s1.¡clyof
sensitiveness.Tlrnonn¡vr
3.I'
ß'ot' r¿tr'SGÄ
trsirtrttt
ntrtttolottic att'd conuet;^Jilness .frr,tr,ct'ionu,
Jor tt'rx'yqi r,
ccn'clr"
ítt'P(ti
suclt' tlmt/(¿i) < /(r:r) <./(rt)
JTør)
- l(*r)
> l@n)-/(rr) - tst'(n,t) - tsr(r,f) )
fsr'( üz,t)-
Úsr('rt,f)'I'he col'r'esponding
result
Tol: conc¿ut'e fitnessf¡nctions is
given loyTrrnottnu 3.2. For
rl,PSGÄ
l;s'i"tl(l (c nt'ott'ottttt'ic tt'n'd concaue Jitness fu,ttction u,, Jor tutuy ç)1)t;,
ctntlr,
'i,ttP(ti
stt'clt, lltatT(*t) <
J@")< f(nt)
l(r") -J(ør)
<î\rr) - T(sr) - lsr(r'
t')- tsr(n^f) (
fsr'(ør,t) - tsr(trnt\
Since
the
lrroofsof the trvo
theolems at'c absolutely analogous weshall only give
1,heproof for tlrc lattel
one.ProoJ. (lìheorem
3.2.)
Sinceu js
conc¿ù\re'ï'e
ha,\'ellØ;'), !(n"), Jþ;'); øl <
o,r,vhich
by the
clefinitiónof the
clividecl cliffelenee is u,(n")- tt(ur)
u'(ur)-
u(nr).f(uo)
- l(*,) l(a,\ - lþ') <0.
l@) -
l@,)18 4
perty
lI. l.l. ì:lal hz-s
For
any lr1'perpklneII , rú any
rnoment f/sr'(e;.
fl
tsr(
lI.l,\ - ç fr¡1
n(H, t)-
ru(r)_s. - Èn
uQ)n(H,t)u'(H, t,)
tr(t)
rvhere
u(Hrt) is the
¿ùvela,gefitness of the
lepresetrtat,ivesof
-È1in P(fl'
lYe
shali cád suchgeneticäIgoûthms
Proporl,ional Selection GoneticAl-
gorithrns
(PSGA).2D
llf . E. Ilalázs 6 I
Using the
rr¡-pothesisthat l(*rl < I(nr) </(nr.)
we oLrtain u(n")-
nt('T,\l@"\ - l@,)
wìrich, sinc¿ ø
is
monotonic,is
eç¡uir.¿i,lautto
a nrcrrrbcr ttll
p(t)
snclr tll¿1, f/.1.n¡in\_
2'
j-,
Lrrt_tb¡'
.¡,'l'i'o ,',n,rlìrLí';i
þ(ú)n ])(l)Ì
for,i : l, 2. B. '- -
\v,'cle
of the
co¡rclusiolr,of the
ilreoreru¡,vlriclr
is the
sa,rnea,s l(u?^") *
'f(ø1*^")(
2/(ei't"), ./(æi-")
-
,/(e,à",") < ./(ø1,,,,,)_
.f(,,1.,n_*).
tsJ'
the
coneavitvof
tJre fjtnessfunctjon
¿¿ilrjs
inrpplies t.t(nl'"^)- u,(r|",) {
¿(r1,,r,,)_
¿( ø.1,n*),u'hich is
equivalentto
O¡l Genctic
^lgolitht¡ s
tt(r)
, tt([I r, y)
27
nr(nrl
- tr,(rr) ç þt(nrl - u,{*r\l fl.ì_Jt.¿
.Í(r,t -
J@,1Tlre conclition
/(ør) -
T@r)<,/(rr) _ l(*r)
meansilrat l(nrl -
.l(n,ll(n,l -
l@,1by
rvhich the previous eclualit¡. l:ecornesu(nrl -,u(ør) ç u(rrl - u,{rrl.
Divicling
bhis inecluality b), ,r?(f) rveolttain
tst'(rr,
t) _
tsr.(rnf) (
fsr(e;¿,tl _
lsr(xr,tl,
s.an
intuil;ivc
cxptr..r,rratiorr rvhy sca_in
sclectionpr.actice. 1
"vex
ancl concal¡e selecl,ionalgolithr is that in oul future l,olh ive int
seusili,r'if¡'of
¡nclrLlcls of a cìass of g{ilsl.
sto¡rin tìlis
tlil,cc,l iort n orrlrl l¡e, l-,r¡'llrc lif
ncss f tulcLjt¡u [or, 1 ltc sarrrcotonic and concave fitncss function,
the
con.r,erx case bcirrg siruilar.'if ,i,::;'ií:"i;"Y(Tf ,i:;;"':r:;J'p,'z
max
{/(r)ln
eil, 0 p(r)} f max{/(ø lln
eH, n
I_,(r)} <(
2rnin{/(r)ln
efI, n
J?(ú)}+
+ tsr(ÍIr, t) _
tsr(Hr, ú)(
tsr,(tr ,, t) _ tsr(Iil.t)
u(IIr,t) _ u,(IIr,t) < u(H,t) _,(Ht,t),
wlrich dtivi¡teet
by û,(t)()
0) givestsr(fIy t) _
tst(Ilz,f) <
ts{}7 ,,t) _ tsr(II[
t).riTo"
o';
'i;(íili.ïi" ,fii,ï,,íi,,!,1
, !;,,"îi,:, tt4H1,t)
\-
./_)i: I
;Ur't)'
t.r(øi'n')Sincc
r¿is
monot,onic -w,e havcthat
is,Ð,^iffi-,p,,, ffi* "\,
lsr(Hr,
t) ) I
-l- ,r,f" irrrrr, ..f ttt)
,tt(II,
t)u'(æ)
-
(.l( e,l"'')
- .l(øi"'))
lsr(IIr,, t),f vDìin -_tnAr,
tt¡here
tn',,'t
t JI
l(a) - J(,ti,")
J ø e ferfrin,;¡lÌ'ax) : Illtllrt(ri)")
I Ol¡ Cìeuctic
^li3-orillulì.s
4. {:O\(:f,TtSIONsi ÀNI) IrU't'UIIIì \tOltIi
c.) l{ . E. Ilalázs
'I¡rnonnlr 3.5. Ior
¿rPSGA
lrsitr{Ja
?nt)motonic (x,tr{Í, concalie filness funct'ion xL ctlxd,for
lryTterpianesIL
&n(t "Hz i.nP(t)
sualt tlre,t I-11(r,.rIlrllte
Jollowing in'ec1uctl'ity
is
trnLe :f
^"('!t"'
'lto*l tsr'(I1.,t)> l1 ¡
1!t" .-- - '
(./(oi"")
-,/(n'Ì'^*))ltsr(I1.,
t),I
rr(II,, f)
J,
(,jt"t, ,!'**l-
rrri rrJ l(ri"") -
,r¿(ø)lø
e (a;f,r,,,fi,,-] Ì
.u)tte?'?
tltu
{ ¡*i*,¡ 1¡") I'
Note 3.7. Since the l':ìhtcs of
/alc
consitlclcdto
be l¡ountlcct artd t¿ is rnonotonicboth of
i,hc consicleleclminitnurns
exisb.\Ye give 1,he proof
fol
Tlteorem 3.5,for thc
cotrespontllrLg onervith
convexfitness function the proof is
arialogous.Prctof. ('l'heolem
lì.5.) Sincc
r¿is a
cìonc&\refitness function,
usingthe sane
r"Lcltationas
aì:or.e 'vve have [.f(ci"""),./(rt!""'),Í(n); u] (
0for
^nl n:l(ø)
e (/(a.,ï''"), "f(¿]"n')1,that
is. fl'hc
main poìnt
of tJris paper' ìsthat thcle is a rather
large class of genetìc algorithr-nsfor-n'hich
somesensitivilr- plopelties ca¡ lié
estaSlis-hed
cle.pending onthe natulc of lhe
Lrnclelil'ìirg'^fitnessfrnctio¡. As in orll
rnairì r'efelence[2]
lirelcsulls
are folrnulaicci'in ter,ms ofthe
objegtiye function.llhe intclpletatìon of thc
mentionedresulls shol. that
fclr I,SGA-s usitrg nlonotonÌc fitnessfunctions tle lattels
shoulcl tre chosento lte
con-opclat
conr¡exin tlie
aclr.anced ones. Thisuseful
tnessfu¡ctiol
clependitrg onnehow
ogress oT sc¿lrch. Conceriring¿rt
least l-hich
mar' :¡r.ise :a. tfotY to
clesigna
(parametrìzetl) fitnessfunction to
e,xploit the preserrtecl resulis ?oultl
the_palametcr l'liich dirccts scalirc' be
ancl þorr'shou
cluring 1,hé operaLion ofthe
genetic algor.ithrn?lialler,\\'e sh¿rll
tlv to
give sonre possìblè anss,elsto
t,fuesetï,o
IÌIjIì]JIìENCìJS
1. Iìalázs, \'I. li., Ott an l.):t¡terinrcrtl for tjsirtq (]enctic AlooLitlLnts for Soluin(t ]iqLLnliorts, subrnit- tccl to Slrrdia lirir'. llaìrcs-tìol¡.li (19()l ).
2. (ìr'cIcrrstcLtc, .I. J. anrl lìul<ct., J,li., Itotu Genctic A|got i\tnts ll:orl¡ ; A Ot ilical l.ool: ul IntpIi_
cil Pcu'rtLLcIisnr, l)r'occcrìings of I(jÇ;\,g9, lirl. Scìralfcr. (J ggg).
3. F{olland, J.Il., AdopLctliott in nalural ancl utlil'icial srTslcnrs, rinn .ilbor,: I'nir.cr.sit.r. I\Iic¡iga¡
l:,r'css (197õ).
23
u(n) -
tl(r12""')l@) -
"f(ø'j'"')_
,u(rä"'")-
ø(ø'f'o*),/(r|"") - .f(r'i"") (0
for any # : f(rc) e (/(øÏ""),"f(r'i'-.)Í(n) -"f(øi''.)
î'his is tlte
sarnclvith
1r}rrt, r|.exl
tn'(H",1) > tsr(Hþl) f
,ITI U (./( øä""')-
./( rìi'""))l'ìcccivotl 27 \ III 199:i llrtlttl-JlrLl qri L-rtItcr silg Dtpru Ltttcttl r;l llulltentaLics rrt:ti
IttIotntulics 51r. .l/. Iioqrrlttirctttttr 7,
3400 Clnj-Ìt-a¡trsta, lìo¡ttìn ia
'l','hich
is the
sameu'ith
/-lrllll,.lll&fr
.tsr(I{r,t) >
L+t,li,'
_'-"' _'(/(r,1,,^)_./(øi.-)).
fsr(
flr' 1) tr(ll
t, l).l'his
latter
incqtLal.ib¡. is equivalent to the oüc ì\'c ltattto
proYe, i.e._ (.'Ðt"t,"tjto*l
tsr.(t1,,,,,
It * ""*
(,ftn.i"") --.f(,rÏ'o-)) Itsr'(//,,
l),Thus
fol
I,SGA-s u'hich uses ¿ù rnonotonic antl cither' ¿ìi oonvcxor
acronca\¡e litness