Darwinism in quantum systems? A. Iqbal and A.H. Toor 2 0 Department of Electronics, Quaid-i-Azam University, 0 Islamabad 45320, Pakistan. 2 email: [email protected] n a February 1, 2008 J 0 1 Abstract 5 We investigate the role of quantum mechanical effects in the central v 5 stabilityconceptofevolutionarygametheoryi.e. anEvolutionarilyStable 8 Strategy(ESS).Usingtwoandthree-playersymmetricquantumgameswe 0 show how the presence of quantum phenomenon of entanglement can be 3 crucial to decide the course of evolutionary dynamics in a population of 0 interacting individuals. 1 0 / 1 Introduction h p - Many interesting results in the recently developed quantum game theory [1, 2, t n 3, 4] are about the most fundamental idea in noncooperative game theory [5] a i.e. the Nash equilibrium (NE). A strategy profile is a NE [6] if no player can u gain by unilaterally deviating from it. The implicit assumption behind NE is q that players make their choices simultaneously and independently. This idea : v also assumes that each player participating in a game behaves rational and i X searches to maximize his own payoff. In situations where evolution of complex behavior occurs further refinements of NE concept are required [7], especially, r a whenmultipleNEappearintheanalysisofagame. Arefinementisthenneeded toprefersome NEoverthe others. Due attentionhasalreadybeengiventoNE intherecentworks[2,4]onquantumgamesandthisdevelopmenthasmotivated us to study certain refinement notions of NE in quantum games. Refinements of NE in classical game theory are popular as well as numerous [8]. Speaking historically, the set of refinements became so large that eventually almost any NE could be justified in terms of someone or other’s refinement [7]. An interesting, and fruitful as well, refinement of NE was introduced by Maynard Smith in 1970’s that became the central notion of evolutionary game theory. InhisbookEvolutionandtheTheoryofGames [9]hedivertedattention awayfromelaboratedefinitionsofrationalityandpresentedanevolutionaryap- proach in classical game theory. The evolutionary approach can be seen as a 1 large population model of adjustment to a NE i.e. an adjustment of popula- tion segments by evolution as opposed to learning. Contrary to classical game theory, in evolutionary game theory the individuals of a population, subject to evolution, are not assumed to act consciously and rationally. Many successful applications of evolutionary game theory appeared in mathematical biology to predict the behavior of bacteria and insects that can hardlybe said to think at all. Themostimportantfeatureofevolutionarygametheoryisthattheassump- tionofrationalplayers,originatingfromclassicalgametheory,does notremain crucial. This important aspect appears when players’ payoffs are equated to their reproductive success. The concept of evolutionary stability stimulated the development of evolu- tionarygametheorythatestablishesalinkbetweengametheoryandthetheory of evolution. Presently the ESS theory is the central model of evolutionary dy- namics of a populations of interacting individuals. It asks, and finds answer to it, a basic question: which states, during the course of selection process, of a given population are stable against perturbations induced by mutations [10]. The ESS theory is based on Darwin’s idea of natural selection; which is shown to be describable as an algorithm called replicator dynamic [7]. Iterations of selections from randomly mutating replicators is an important feature of the dynamic. Speaking the language of game theory, the replicator dynamic says that in a population the proportion of players which play better strategies in- creasewith time. When replicatordynamic is underlying processof a game the ESSs are shown to be stable against perturbations [10]. In other words ESSs are, then, rest points of the replicator dynamic. Recent developments in quantum games provide an incentive to look at the mathematicaltheoryofevolution,with the centralidea ofanESS,in a broader picture given by Hilbert structure of strategy space in the new theory. This incentive is drivenby a question: how game-theoreticalmodels, of evolutionary dynamicsina population,shapethemselves inthe new settingsprovidedto the game theory recently by quantum mechanics? Our motivation is that quan- tum mechanical effects, especially entanglement, may have decisive role in the evolutionarydynamicsthathavealreadybeensuccessfully,andalsoquiterigor- ously,modelledusingthe classicalgametheory. Tostudy evolutioninquantum settings we have chosen the ESS idea mostly for simplicity and beauty of the concept. We ask questions like, how ESSs are affected when a classical game playedinapopulationchangesitselftooneofitsquantumforms? Howpureand mixedESSsaredistinguishedfromoneanotherwhensuchachangeinformofa game takes place? And most importantly, how evolutionary dynamics becomes linked to quantum entanglement present in games that are played in quantum settings? In earlier papers [11, 12, 13] we showed that the presence of entanglement, in asymmetric as well as symmetric bimatrix games, can disturb the evolution- ary stability expressed by the idea of ESS. So that, evolutionary stability of a symmetric NE can be made to appear or disappear by controlling entangle- ment in symmetric and asymmetric bimatrix games. We found a consideration of symmetric games more appropriate because the notion of an ESS was origi- 2 nally investigated[9]for pair-wisesymmetric contests. Inpresentletter we find examplesoftwoandthree-playergameswhereentanglementchangesevolution- arystabilityofasymmetric NE.We assumeinitialquantumstatesinthe same, originally suggested, simpler form presented in the scheme [3] that tells how to play a quantum game. With initial states in this form, we show that two and three-player games are distinguished in an interesting aspect i.e. entangle- ment can change evolutionary stability of pure strategies in two-player games. However,foramixedstrategyitcanbedonewhenthenumberofplayersarein- creasedfromtwotothree. Beforecomingtoquantumsettingswefirstdescribe, mathematically, the concept of an ESS in classical evolutionary game theory. 2 Evolutionary stability MaynardSmith introducedthe idea ofan EvolutionarilyStable Strategy (ESS) in a seminal paper ‘The logic of animal conflict’ [14]. In rough terms, an ESS is a strategy which, if playedby almost all members of a population, cannot be displaced by a small invading group playing any alternative strategy. Suppose pairs of individuals are repeatedly drawnat randomfrom a largepopulation to play a symmetric two-persongame. Let the game between the individuals be a symmetricbimatrixgamerepresentedbytheexpressionG=(M,MT)whereM representsthepayoffmatrixandT itstranspose. InusualnotationP(x,y)gives the payoff to a x-player against a y-player for a symmetric pair-wise contest. A strategy x is said to be an ESS if for each mutant strategy y there exists a positiveinvasion barrier suchthatifthepopulationshareofindividualsplaying the mutant strategy y falls below this barrier, then x earns a higher expected payoff than y. Mathematically speaking, x is an ESS when for each strategy y =x the inequality P[x,(1 ǫ)x+ǫy]>P[y,(1 ǫ)x+ǫy] should hold for all 6 − − sufficientlysmallǫ>0;where,forexample,theexpressiononthe left-handside ispayofftostrategyxwhenplayedagainstthemixedstrategy(1 ǫ)x+ǫy. This − condition for an ESS can be shown [9] easily to be equivalent to the following two requirements 1. P(x,x) > P(y,x) 2. If P(x,x) = P(y,x) then P(x,y)>P(y,y) (1) It becomes apparent that an ESS is a symmetric NE but also possesses a stability propertyagainstmutations. Condition1 in the abovedefinition shows that(x,x)isaNEforthebimatrixgameG=(M,MT)ifxisanESS.However, the reverse is not true. If (x,x) is a NE, then x is an ESS only if x satisfies condition2inthedefinition(1). TheESSconditiongivesarefinementontheset of symmetric Nash equilibria [7]. Its essential feature is that, apart from being a symmetric NE, it is robustagainsta small number of mutants appearing in a population playing an ESS [10]. Now we considerquantum settings, involvingtwo players,to investigate the concept of evolutionary stability. 3 3 Two-player case We use the quantization scheme suggested by Marinatto and Weber [3] for the two-player quantum game of Battle of Sexes. In this scheme the players’ tac- tics consist of deciding the classical probabilities of applying two unitary and Hermitian operators (the identity I and the inversion operator C) on an ini- tial quantum strategy in 2 2 dimensional Hilbert space; that can be obtained ⊗ fromasystemoftwo qubits. The tactics phase is similarto probabilisticchoice between pure strategies in the classicalgame theory. Interestingly, the classical formofthegameisreproducedbymakingtheinitialquantumstateunentangled [3]. There are some different opinions [15, 16] concerning the use of the term strategy in Marinattoand Weber’s scheme. EarlierEisert,Wilkins, and Lewen- stein proposed a scheme [2] where choosing a move corresponds to a strategy. Expanding on this work, Marinatto and Weber presented a different approach and preferred to call tactics the process of choosing a move when an initial strategy, in the form of a quantum state, is forwarded to the players. In this paperwe will referto MarinattoandWeber’s tactics as strategiesormovesand their initial strategy as initial quantum state. We then find how entanglement affects evolutionary stability in the circumstances that a quantum version of a game can be reduced to its classical form by removing entanglement. Because the classicalgame correspondsto an unentangled initial quantum state, a com- parison between ESSs in classical and quantized versions of the game can be made by maneuvering the initial quantum state, in some particular form. The scheme for two-player quantum game is shown in fig. 1. Consider a two-player symmetric game given by the matrix Bob’sstrategy S S 1 2 Alice′s strategy S (α,α) (β,γ) (2) 1 S (γ,β) (δ,δ) 2 and played via the initial state ψ = a S S +b S S where a2+ b2 = | ini | 1 1i | 2 2i | | | | 1. A unitary and Hermitian operator C used in the scheme is defined as [3] C S = S , C S = S andC† =C =C−1. Letone of the playerschooses 1 2 2 1 | i | i | i | i hisstrategybyimplementing the identity operatorI withprobabilitypandthe operatorC with probability (1 p), on the initial state ρ that correspondsto − in ψ . Similarly, suppose the second player applies the operators I and C with | ini probabilities q and (1 q) respectively. The final density matrix is written as − [3] ρ = Pr(U )Pr(U )[U U ρ U† U†] (3) fin A B A⊗ B in A⊗ B U=I,C X where the unitary and Hermitian operator U is either I or C. Pr(U ), Pr(U ) A B are the probabilities with which players A and B apply the operator U on 4 Figure 1: The scheme to play a two-player quantum game. the initial state, respectively. It is seen that ρ corresponds to a convex fin combination of all possible quantum operations. Payoffoperators for Alice and Bob are [3] (P ) =α,α S S +β,γ S S +γ,β S S +δ,δ S S A,B oper 1 1 1 2 2 1 2 2 | i | i | i | i The payoffs are then obtained as mean values of these operators i.e. P = A,B Tr (P ) ρ . Because the quantum game is symmetric using the initial A,B oper fin state ψ and the payoff matrix (2), there is no need for subscripts. We, (cid:2) | ini (cid:3) ⋆ therefore, write the payoff to a p player against a q player as P(p,q). When p is a NE we find the following payoff difference [12] P(p⋆,p⋆) P(p,p⋆)=(p⋆ p)[ a2(β δ)+ − − | | − b2(γ α) p⋆ (β δ)+(γ α) ] (4) | | − − { − − } Now the ESS conditions for the pure strategy p=0 are given as 1. b2 (β δ) (γ α) >(β δ) | | { − − − } − 2. If b2 (β δ) (γ α) =(β δ) | | { − − − } − then q2 (β δ)+(γ α) >0 (5) { − − } 5 where1isthe NE condition. Similarlythe ESSconditionsforthe purestrategy p=1 are 1. b2 (γ α) (β δ) >(γ α) | | { − − − } − 2. If b2 (γ α) (β δ) =(γ α) | | { − − − } − then (1 q)2 (β δ)+(γ α) >0 (6) − { − − } Becausetheseconditionsforboththepurestrategiesp=1andp=0dependon b2, therefore, there can be examples of two-player symmetric games for which | | the evolutionary stability of pure strategies can be changed while playing the gameusinginitialstateintheform ψ =a S S +b S S . However,forthe mixed NE, given as p⋆ = |a|2(β−δ)+|b||2(iγn−iα), t|he1co1riresp|on2di2nig payoff difference (β−δ)+(γ−α) 4 becomes identically zero. From the second condition of an ESS we find, for ⋆ the mixed NE p, the difference ⋆ 1 P(p,q) P(q,q)= − (β δ)+(γ α) × − − [(β δ) q (β δ)+(γ α) b2 (β δ) (γ α) ]2 (7) − − { − − }−| | { − − − } ⋆ Therefore, the mixed strategy p is an ESS when (β δ)+(γ α) >0. This condition, making the mixed NE p⋆ an ESS, is ind{epen−dent of b−2 [1}7]. So that, | | inthissymmetrictwo-playerquantumgame,evolutionarystabilityofthemixed ⋆ NE p cannot be changedwhen the game is playedusing initial quantum states of the form ψ =a S S +b S S . | ini | 1 1i | 2 2i Therefore, evolutionary stability of only the pure strategies can be affected, with the chosen form of the initial states, for the two-player symmetric games. Examples of the games with this property are easy to find. The class of games for which γ = α and (β δ) < 0 the strategies p = 0 and p = 1 remain NE − for all b2 [0,1]; but the strategy p = 1 is not an ESS when b2 = 0 and the | | ∈ | | strategyp=0 isnotanESSwhen b2 =1. Inanearlierletter [12]wefoundan | | exampleofaclassofgamesforwhichapurestrategy,thatisanESSclassically, does not remain ESS for a particular value of b2,even though it remains a NE | | for all possible range of b2. | | Tofindexamplesofsymmetricquantumgames,whereevolutionarystability of the mixed strategies may also be affected by controlling the entanglement, we now increase the number of players from two to three. 4 Three-player case Inextendingthetwo-playerschemetoathree-playercase,weassumethatthree players A,B, and C play their strategies by applying the identity operator I withtheprobabilitiesp,qandrrespectivelyontheinitialstate ψ . Therefore, | ini 6 they apply the operator C with the probabilities (1 p),(1 q) and (1 r) − − − respectively. The final state then corresponds to the density matrix ρ = Pr(U )Pr(U )Pr(U ) U U U ρ U† U† U† (8) fin A B C A⊗ B ⊗ C in A⊗ B⊗ C UX=I,C h i where the 8 basis vectors are S S S , for i,j,k = 1,2. Again we use initial i j k | i quantumstateintheform ψ =a S S S +b S S S ,where a2+ b2 =1. | ini | 1 1 1i | 2 2 2i | | | | Itisaquantumstatein2 2 2dimensionalHilbertspacethatcanbeprepared ⊗ ⊗ fromasystemofthreetwo-statequantumsystemsorqubits. Similartothetwo- player case, we define the payoff operators for the players A, B, and C as (P ) = A,B,C oper α ,β ,η S S S S S S +α ,β ,η S S S S S S + 1 1 1| 1 1 1ih 1 1 1| 2 2 2| 2 1 1ih 2 1 1| α ,β ,η S S S S S S +α ,β ,η S S S S S S + 3 3 3| 1 2 1ih 1 2 1| 4 4 4| 1 1 2ih 1 1 2| α ,β ,η S S S S S S +α ,β ,η S S S S S S + 5 5 5| 1 2 2ih 1 2 2| 6 6 6| 2 1 2ih 2 1 2| α ,β ,η S S S S S S +α ,β ,η S S S S S S (9) 7 7 7| 2 2 1ih 2 2 1| 8 8 8| 2 2 2ih 2 2 2| where α ,β ,η for 1 l 8 are 24 constants of the matrix of this three-player l l l ≤ ≤ game. Payoffs to the players A,B, and C are then obtained as mean values of these operators P (p,q,r)=Trace (P ) ρ (10) A,B,C A,B,C oper fin (cid:2) (cid:3) Here,similartotwoplayercase,theclassicalpayoffscanbeobtainedbymaking the initial quantum state unentangled and fixing b2 = 0. To get a symmetric | | game we define P (x,y,z) as the payoff to player A when players A, B, and C A play the strategies x,y and z respectively. Following relations make payoffs to the players a quantity that is identity independent but depends only on their strategies P (x,y,z)=P (x,z,y)=P (y,x,z) A A B =P (z,x,y)=P (y,z,x)=P (z,y,x) (11) B C C For these relations to hold we need following replacements for β and η i i β α β α β α β α 1 → 1 2 → 3 3 → 2 4 → 3 β α β α β α β α 5 → 6 6 → 5 7 → 6 8 → 8 η α η α η α η α 1 → 1 2 → 3 3 → 3 4 → 2 η α η α η α η α (12) 5 → 6 6 → 6 7 → 5 8 → 8 7 Also,itisnownecessarythatweshouldhaveα =α andα =α .Asymmetric 6 7 3 4 game between three players, therefore, can be defined by only six constants. We take these to be α ,α ,α ,α ,α and, α . The payoff to a player now 1 2 3 5 6 8 becomesonlyastrategydependentquantityandbecomesidentityindependent. Nosubscriptsaretherefore,needed. Payofftoapplayer,whenothertwoplayers ⋆ play q and r, cannow be written as P(p,q,r). A symmetric NE p canbe found ⋆ ⋆ ⋆ ⋆ ⋆ from the Nash condition P(p,p,p) P(p,p,p) 0 i.e. − ≥ P(p⋆,p⋆,p⋆) P(p,p⋆,p⋆)=(p⋆ p)[p⋆2(1 2 b2)(σ+ω 2η)+ − − − | | − 2p⋆ b2(σ+ω 2η) ω+η + ω b2(σ+ω) ] 0 (13) | | − − −| | ≥ n o n o where (α α )=σ, (α α )=η, and (α α )=ω. Three possible NE are 1 2 3 6 5 8 − − − found as ⋆ (ω−η)−|b|2(σ+ω−2η) ±√{(σ+ω)2−(2η)2}|b|2(1−|b|2)+(η2−σω) p1 = { }(1−2|b|2)(σ+ω−2η) ⋆ (14) p =0 (cid:26) 2 (cid:27) ⋆ p =1 3 ⋆ ⋆ ⋆ ⋆ ⋆ ⋆ Clearly the mixed NE p makes the difference P(p,p,p) P(p,p,p) identically 1 zeroand twovalues for p⋆ canbe found for a given b2. p⋆−, p⋆ arepure strategy ⋆ 1 | | 2 3 NE.We notice thatp comes outasaNE without imposingfurther restrictions 1 onthematrixofthesymmetricthree-playergame. However,thepurestrategies ⋆ ⋆ p and p can be NE when further restriction are imposed on the matrix of the 2 3 game. For example, p⋆ canbe a NE providedσ (ω+σ) b2 for all b2 [0,1]. Similarly p⋆ can be NE3 when ω (ω+σ) b2. ≥ | | | | ∈ 2 ≤ | | Now the question we ask: how the evolutionary stability of these three NE can be affected while playing the game via the initial quantum states given in the form ψ = a S S S +b S S S ?. For the two-player asymmetric | ini | 1 1 1i | 2 2 2i game of Battle of Sexes we showed that out of the three NE only two can be evolutionarily stable [11]. In classical evolutionary game theory the concept of an ESS is well known to be extended to multi-player case. When mutants are allowed to play only one strategy the definition of an ESS for three-player case is written as [18] 1. P(p,p,p) > P(q,p,p) 2. If P(p,p,p) = P(q,p,p) then P(p,q,p)>P(q,q,p) (15) Here p is a NE if it satisfies the condition 1 against all q =p. For our case the ⋆ ⋆ 6 ESS conditions for the pure strategies p and p can be written as follows. For 2 3 ⋆ example, p =0 is an ESS when 2 8 1. σ b2 > ω a2 | | | | 2. If σ b2 = ω a2 then ηq2(a2 b2)>0 (16) | | | | − | | −| | ⋆ ⋆ where1is NEconditionforthe strategyp =0. Similarlyp =1is anESS[19] 2 3 when 1. σ a2 > ω b2 | | | | 2. If σ a2 = ω b2 then η(1 q)2(a2 b2)>0 (17) | | | | − | | −| | and both the pure strategies p⋆ and p⋆ are ESSs when a2 = b2. Examples of 2 3 | | | | three-playersymmetricgamesareeasytofindforwhichapurestrategyisaNE forthewholerange b2 [0,1],butdoesnotremainanESSforsomeparticular | | ∈ value of b2. An example ofa class ofsuchgames is for whichσ =0,ω <0 and η 0. I|n|this class the strategy p⋆ = 0 is a NE for all b2 [0,1] but not an ≤ 2 | | ∈ ESS when b2 =1. | | ⋆ However, the mixed strategy NE p forms the most interesting case. It 1 ⋆ ⋆ ⋆ ⋆ ⋆ makes the payoff difference P(p ,p ,p ) P(p,p ,p ) identically zero for every 1 1 1 1 1 − ⋆ ⋆ ⋆ ⋆ strategy p. Now p is an ESS when P(p ,q,p ) P(q,q,p ) >0 but 1 1 1 1 − n o ⋆ ⋆ ⋆ P(p ,q,p ) P(q,q,p ) 1 1 1 − = (p⋆ q)2 (σ+ω)2 (2η)2 b2(1 b2)+(η2 σω) (18) 1 ± − { − }| | −| | − q ⋆ ⋆ Therefore,outofthetwopossibleroots(p ) and(p ) ,thatmakethedifference 1 1 1 2 ⋆ ⋆ ⋆ P(p ,q,p ) P(q,q,p ) greater than and less than zero respectively, of the 1 1 1 − quadratic equation p⋆2(1 2 b2)(σ+ω 2η)+ 1 − | | − 2p⋆ b2(σ+ω 2η) ω+η + ω b2(σ+ω) =0 (19) 1 | | − − −| | n o n o ⋆ only (p ) can exist as an ESS. When the square root term in the equation 1 1 (18) becomes zero we have only one mixed NE, that is not an ESS. Therefore, out of four possible NE in this three-player game only three can be ESSs. An interestingclassofthree-playergamesisonewithη2 =σω. Forthesegamesthe ⋆ (w−η)−|b|2(σ+ω−2η) ±|a||b||σ−ω| mixedNEarep1 = { (1−2|b|2)(σ+ω}−2η) and,whenplayedclassically, wecangetonlyone mixedNE thatis notanESS.Howeverforall b2,different | | from zero, we generally obtain two NE out of which one can be an ESS. Similar to the two-player case, the NE in a three-player symmetric game importantfromthepointofviewofevolutionarystabilityarethosethatsurvive 9 achangebetweentwoinitialstates;onebeingunentangledcorrespondingtothe classical game. Suppose p⋆ remains a NE for b2 =0 and some other non-zero 1 b2. Itispossiblewhen(σ ω)(2p⋆ 1)=0. On|e|possibilityisthestrategyp⋆ = 1 | | − 1− 2 remaining a NE for all b2 [0,1]. It reduces the defining quadratic equation ⋆ | | ∈ ⋆ ⋆ ⋆ for p to σ + ω + 2η = 0 and makes the difference P(p ,q,p ) P(q,q,p ) 1 1 1 1 independent of b2. Therefore the strategy p⋆ = 1, even when rem−aining a NE | | 2 for all b2 [0,1], can not be an ESS in only one version of the symmetric | | ∈ three-player game. For the second possibility σ = ω the defining equation for ⋆ p is reduced to 1 (1 2 b2) p⋆ (η−σ)− η2−σ2 p⋆ (η−σ)+ η2−σ2 =0 − | | ( 1− 2(η−pσ) )( 1− 2(η−pσ) ) (20) for which P(p⋆,q,p⋆) P(q,q,p⋆)= 2(p⋆ q)2 b2 1 η2 σ2 (21) 1 1 1 1 − ± − | | − 2 − (cid:12) (cid:12) (cid:12) (cid:12)p HerethedifferenceP(p⋆1,q,p⋆1) P(q,q,p⋆1)stilldep(cid:12)(cid:12)endson(cid:12)(cid:12)b2andbecomeszero − | | for b2 = 1. Thereforeforthe classofgamesforwhichσ =ω andη >σ, oneof | | 2 themixedstrategies(p⋆) ,(p⋆) remainsaNEforall b2 [0,1]butnotanESS 1 1 1 2 | | ∈ when b2 = 1. In this class of three-player quantum games the evolutionary | | 2 stabilityofamixedstrategycan,therefore,bechangedwhilethegameisplayed using initial quantum states in the form ψ =a S S S +b S S S . | ini | 1 1 1i | 2 2 2i 5 Discussion The fact that classical games are played in natural macroscopic world is well known for a long time. Evolutionary game theory is a subject, growing out of such studies and, dealing mostly with games played in the animal world. Recent work in biology [20] suggests nature playing classical games at micro- level. Bacterial infections by viruses have been presented as classical game-like situations where nature prefers the dominant strategies. The concept of evo- lutionary stability, without assuming rational and conscious individuals, gives game-theoretical models of stable states for a population of interacting indi- viduals. Darwinian idea of natural selection provides physical ground to these models of rationality studied in evolutionary game theory. Why there is need to study evolutionary stability in quantum games? We find it interesting that someentirely quantumaspectlike entanglementcanhavea deciding roleabout which stable states of the population should survive and others should not. It means that the presence of quantum interactions, in a population undergo- ing evolution, can alter its stable states resulting from evolutionary dynamics. 10