game theory -- lecture 4 - accueil | loiseau/gametheory/slides/lecture4.pdfexamples • using game...
TRANSCRIPT
GameTheory--
Lecture4
PatrickLoiseauEURECOMFall2016
1
Lecture2-3recap• ProvedexistenceofpurestrategyNashequilibriumin
gameswithcompactconvexactionsetsandcontinuousconcaveutilities
• DefinedmixedstrategyNashequilibrium• ProvedexistenceofmixedstrategyNashequilibriumin
finitegames• Discussedcomputationandinterpretationofmixed
strategiesNashequilibrium
àNashequilibriumisnottheonlysolutionconceptàToday:Anothersolutionconcept:evolutionarystable
strategies
2
Outline
• Evolutionarystablestrategies
3
Evolutionarygametheory• Gametheoryßà evolutionarybiology• Idea:– Relatestrategiestophenotypes ofgenes– Relatepayoffstogeneticfitness– Strategiesthatdowell“grow”,thosethatobtainlowerpayoffs“dieout”
• Importantnote:– Strategiesarehardwired,theyarenotchosenbyplayers
• Assumptions:– Withinspeciescompetition:nomixtureofpopulation
4
Examples• Usinggametheorytounderstandpopulationdynamics
– Evolutionofspecies– Groupsoflionsdecidingwhethertoattackingroupanantelope– Antsdecidingtorespondtoanattackofaspider
– TCPvariants,P2Papplications
• Usingevolutiontointerpreteconomicactions– Firmsinacompetitivemarket– Firmsarebounded,theycan’tcomputethebestresponse,but
haverulesofthumbsandadopthardwired(consistent)strategies
– Survivalofthefittest==riseoffirmswithlowcostsandhighprofits
5
Asimplemodel• Assumesimplegame:two-playersymmetric• Assumerandomtournaments– Largepopulationofindividualswithhardwiredstrategies,picktwoindividualsatrandomandmakethemplaythesymmetricgame
– Theplayeradoptingthestrategyyieldinghigherpayoffwillsurvive(andeventuallygainnewelements)whereastheplayerwho“lost”thegamewill“dieout”
• Startwithentirepopulationplayingstrategys• Thenintroduceamutation:asmall groupofindividualsstartplayingstrategys’
• Question:willthemutantssurviveandgrowordieout? 6
Asimpleexample(1)
• Haveyoualreadyseenthisgame?• Examples:– Lionshuntinginacooperativegroup– Antsdefendingthenestinacooperativegroup
• Question:iscooperationevolutionarystable?
2,2 0,33,0 1,1
C
D
Cooperate Defect
ε 1- ε
Player1
Player2
7
Asimpleexample(2)Playerstrategyhardwiredè C
“SpatialGame”
Allplayersarecooperativeandgetapayoffof2
Whathappenswithamutation?
8
Asimpleexample(3)Playerstrategyhardwiredè C
Focusyourattentiononthisrandom“tournament”:
• Cooperatingplayerwillobtainapayoffof0• Defectingplayerwillobtainapayoffof3
Survivalofthefittest:DwinsoverC
Playerstrategyhardwiredè D
9
Asimpleexample(4)Playerstrategyhardwiredè C
Playerstrategyhardwiredè D
10
Asimpleexample(5)Playerstrategyhardwiredè C
Playerstrategyhardwiredè D
11
Asimpleexample(6)Playerstrategyhardwiredè C
Playerstrategyhardwiredè D
Asmallinitialmutationisrapidlyexpandinginsteadofdyingout
Eventually,Cwilldieout
à Conclusion:CisnotES
Remark:wehaveassumedasexualreproductionandnogeneredistribution 12
ESSDefinition1[MaynardSmith1972]
13
Definition1: EvolutionarystablestrategyInasymmetric2-playergame,thepurestrategyŝ isES(inpurestrategies)ifthereexistsε0 >0suchthat:
forallpossibledeviationss’andforallmutationsizesε <ε0.
[ ] [ ] [ ] [ ]),()ˆ,()1(),ˆ()ˆ,ˆ()1( ssussussussu ¢¢+¢->¢+- eeeePayofftoESŝ Payofftomutants’
ESstrategiesinthesimpleexample
• IscooperationES?Cvs.[(1-ε)C+εD]à (1-ε)2+ε0=2(1-ε)Dvs.[(1-ε)C+εD]à (1-ε)3+ε1=3(1-ε)+ ε3(1-ε)+ ε >2(1-ε)
èCisnotES becausetheaveragepayofftoCislowerthantheaveragepayofftoD
èAstrictlydominatedisneverEvolutionarilyStable– Thestrictlydominantstrategywillbeasuccessfulmutation
2,2 0,33,0 1,1
C
D
Cooperate Defect
ε 1- ε
Player1
Player2
1- ε ε ForCbeingamajorityForDbeingamajority
14
ESstrategiesinthesimpleexample
• IsdefectionES?Dvs.[εC+(1-ε)D]à (1-ε)1+ε3=(1-ε)+3εCvs.[εC+(1-ε)D]à (1-ε)0+ε2=2ε(1-ε)+3>2 ε
è DisES:anymutationfromDgetswipedout!
2,2 0,33,0 1,1
C
D
Cooperate Defect
ε 1- ε
Player1
Player2
1- ε ε ForCbeingamajorityForDbeingamajority
15
Anotherexample(1)
• 2-playerssymmetricgamewith3strategies• Is“c”ES? cvs.[(1-ε)c+εb]à (1-ε)0+ε 1=ε
bvs.[(1-ε)c+εb]à (1-ε)1+ε 0=1- ε >ε
è “c”isnotevolutionarystable,as“b”caninvadeit
• Note:“b”,theinvader,isitselfnotES!– ItisnotnecessarilytruethataninvadingstrategymustitselfbeES– Butitstillavoidsdyingoutcompletely(growsto50%here)
2,2 0,0 0,00,0 0,0 1,10,0 1,1 0,0
a
b
c
a b c
16
Anotherexample(3)
• Is(c,c)aNE?
2,2 0,0 0,00,0 0,0 1,10,0 1,1 0,0
a
b
c
a b c
17
Observation
• Ifs isnotNash(thatis(s,s) isnotaNE),thens isnotevolutionarystable (ES)
Equivalently:• Ifs isES,then(s,s) isaNE
• Question:istheoppositetrue?Thatis:– If(s,s) isaNE,thens isES
18
Yetanotherexample(1)
• NEofthisgame:(a,a) and(b,b)• Isb ES? bà 0
aà (1-ε)0+ε 1=ε >0
è (b,b) isaNE,butitisnotES!• ThisrelatestotheideaofaweakNE
è If(s,s) isastrictNE thens isES
1,1 0,00,0 0,0
a
b
a b
ε 1- ε
Player1
Player2
19
StrictNashequilibrium
• WeakNE:theinequalityisanequalityforatleastonealternativestrategy
• StrictNEissufficientbutnotnecessaryforES20
Definition: StrictNashequilibriumAstrategyprofile(s1*,s2*,…,sN*)isastrictNashEquilibriumif,foreachplayeri,
ui(si*,s-i*)>ui(si,s-i*)forallsi ≠si*
ESSDefinition2
21
Definition2: EvolutionarystablestrategyInasymmetric2-playergame,thepurestrategyŝ isES(inpurestrategies)if:
A)
AND
B)
sssussuss
¢"¢³ )ˆ,()ˆ,ˆ(mEquilibriuNash symmetric a is )ˆ,ˆ(
),(),ˆ( then )ˆ,()ˆ,ˆ( if
ssussussussu¢¢>¢¢=
Linkbetweendefinitions1and2
• Proofsketch:
22
TheoremDefinition1Definition2
€
⇔
Recap:checkingforESstrategies
• WehaveseenadefinitionthatconnectsEvolutionaryStabilitytoNashEquilibrium
• Bydef 2,tocheckthatŝ isES,weneedtodo:
– Firstcheckif(ŝ,ŝ)isasymmetric NashEquilibrium– Ifitisastrict NE,we’redone– Otherwise,weneedtocomparehowŝperformsagainstamutation,andhowamutationperformsagainstamutation
– Ifŝperformsbetter,thenwe’redone
23
Example:Is“a”evolutionarystable?
• Is(a,a)aNE?Isitstrict?• Is“a”evolutionarystable?
1,1 1,11,1 0,0
a
b
a b
ε 1- ε
Player1
Player2
24
Evolutionofsocialconvention• Evolutionisoftenappliedtosocialsciences• Let’shavealookathowdrivingtotheleftorrighthandsideoftheroadmightevolve
• WhataretheNE?aretheystrict?WhataretheESS?
• Conclusion:wecanhaveseveralESS– Theyneednotbeequallygood
2,2 0,00,0 1,1
L
R
L R
25
ThegameofChicken
• Thisisasymmetric coordinationgame• Biologyinterpretation:– “a”:individualsthatareaggressive– “b”:individualsthatarenon-aggressive
• WhatarethepurestrategyNE?– Theyarenotsymmetricà nocandidateforESS
0,0 2,11,2 0,0
a
b
a b
26
ThegameofChicken:mixedstrategyNE
• What’sthemixedstrategyNEofthisgame?– MixedstrategyNE=[(2/3,1/3),(2/3,1/3)]è Thisisasymmetric NashEquilibrium
èInterpretation:thereisanequilibriuminwhich2/3ofthegenesareaggressiveand1/3arenon-aggressive
• IsitastrictNashequilibrium?• IsitanESS?
0,0 2,11,2 0,0
a
b
a b
27
Remark
• Amixed-strategyNashequilibrium(withasupportofatleast2actionsforoneoftheplayers)canneverbeastrictNashequilibrium
• ThedefinitionofESSisthesame!
28
ESSDefinition2bis
29
Definition2: EvolutionarystablestrategyInasymmetric2-playergame,themixedstrategyŝisES(inmixedstrategies)if:
A)
AND
B)
sssussuss
¢"¢³ )ˆ,()ˆ,ˆ(mEquilibriuNash symmetric a is )ˆ,ˆ(
),(),ˆ( then )ˆ,()ˆ,ˆ( if
ssussussussu¢¢>¢¢=
ThegameofChicken:ESS
• MixedstrategyNE=[(2/3,1/3),(2/3,1/3)].• IsitanESS?weneedtocheckforallpossiblemixed
mutationss’:• Yes,itis(doitathome!)
• Inmanycasesthatariseinnature,theonlyequilibriumisamixedequilibrium– Itcouldmeanthatthegeneitselfisrandomizing,whichis
plausible– Itcouldbethatthereareactuallytwotypessurvivinginthe
population(cf.ourinterpretationofmixedstrategies)
0,0 2,11,2 0,0
a
b
a b
u(s, !s )> u( !s , !s ) ∀ !s ≠ s
30
Hawksanddoves
DoveHawk
31
TheHawksandDovegame(1)
• Moregeneralgameofaggressionvs.non-aggression– Theprizeisfood,anditsvalueisv>0– There’sacostforfighting,whichisc>0
• Note:we’restillinthecontextofwithinspicescompetition– Soit’snotabattleagainsttwodifferentanimals,hawksand
doves,wetalkaboutstrategies• “Actdovishvs.acthawkish”
• WhataretheESS?Howdotheychangewithc,v?
(v-c)/2,(v-c)/2 v,00,v v/2, v/2
H
D
H D
32
TheHawksandDovegame(2)
• CanwehaveaESpopulationofdoves?• Is(D,D) aNE?– No,hence“D”isnotESS– Indeed,amutationofhawksagainstdoveswouldbeprofitableinthatitwouldobtainapayoffofv
(v-c)/2,(v-c)/2 v,00,v v/2, v/2
H
D
H D
33
TheHawksandDovegame(3)
• CanwehaveaESpopulationofHawks?• Is(H,H) aNE?Itdepends:itisasymmetricNEif(v-c)/2≥0
• Case1:v>cè (H,H) isastrict NEè “H” isESS• Case2:v=cè (v-c)/2=0è u(H,H) =u(D,H) -- (H,H)isaweakNE
– Isu(H,D)=v largerthanu(D,D)=v/2?Yesè “H” isESS
èHisESS ifv≥c• Iftheprizeishighandthecostforfightingislow,thenyou’llsee
fightsarisinginnature
(v-c)/2,(v-c)/2 v,00,v v/2, v/2
H
D
H D
34
TheHawksandDovegame(4)
• Whatifc>v?– “H”isnotESSand“D”isnotESS(theyarenotNE)
• Step1:findamixedNE
• Step2:verifytheESScondition
(v-c)/2,(v-c)/2 v,00,v v/2, v/2
H
D
H D
s 1− s
35
TheHawksandDovegame:results
• Incasev<cwehaveanevolutionarilystablestateinwhichwehavev/chawks1. Asv↗wewillhavemorehawksinESS2. Asc↗wewillhavemoredovesinESS
• BymeasuringtheproportionofHandD,wecangetthevalueofv/c
• Payoff: E u(D, s)[ ] = E u(H, s)[ ] = 0 vc+ 1− v
c"
#$
%
&'v236
Onelastexample(1)
• Assume1<v<2– ~Rock,paper,scissors
• OnlyNE:ŝ =(1/3,1/3,1/3)– mixed,notstrict• IsitanESS?
– Supposes’=R– u(ŝ,R)=(1+v)/3<1– u(R,R)=1
• Conclusion:NotallgameshaveanESS!
1,1 v,0 0,v0,v 1,1 v,0v,0 0,v 1,1
R
P
S
R P S
37