Abstract
Recently we interpreted the notion of ESS for matrix games under time constraints and investigated the corresponding state in the polymorphic situation. Now we give two further static (monomorphic) characterizations which are the appropriate analogues of those known for classical evolutionary matrix games. Namely, it is verified that an ESS can be described as a neighbourhood invader strategy independently of the dimension of the strategy space in our non-linear situation too, that is, a strategy is an ESS if and only if it is able to invade and completely replace any monomorphic population which totally consists of individuals following a strategy close to the ESS. With the neighbourhood invader property at hand, we establish a dynamic characterization under the replicator dynamics in two dimensions which corresponds to the strong stability concept for classical evolutionary matrix games. Besides, in some special cases, we also prove the stability of the corresponding rest point in higher dimensions.
Similar content being viewed by others
Notes
In Křivan and Cressman (2017) and Garay et al. (2017) there are examples for the Hawk–Dove game and the Prisoner’s dilemma with distinct behaviors compared to the classical case. For example, if the cost of fighting is smaller than the value of the resource then the Hawk is the only equilibrium in the classical case. Contrary to this, if the Hawk–Hawk interaction is long enough compared to the other types of interactions then a mixed equilibrium also appears in addition to the pure Hawk equilibrium. Also, for instance, if the matrix of the classical Prisoner’s dilemma is taken as the time constraint matrix then the cooperator strategy proves to be an ESS for an appropriate payoff matrix. In Garay et al. (2018), there is an example for a mixed ESS such that the corresponding interior state is a locally asymptotically stable rest point of the replicator dynamics but, contrary to the classical case, it is not globally stable.
Although \(\rho _{\mathbf p}=\varrho _{\mathbf p}(1-\varepsilon ,\varepsilon , \mathbf p,\mathbf q)\) and \(\omega _{\mathbf p}=W_{\mathbf p}(1-\varepsilon ,\varepsilon , \mathbf p,\mathbf q)\), respectively, the aim of the use of symbols \(\rho \) and \(\omega \), respectively, is to emphasize the monomorphic approach. The notations \(\varrho \) and W, respectively, are reserved for the polymorphic approach (see Sect. 2.3). This will be useful if we investigate a notion with respect to both approaches.
For classical matrix games, a set of states corresponding to an interior ESS is always globally stable (Hofbauer and Sigmund 1998, the remark after Theorem 6.4.1, Exercises 6.4.3, 7.2.7 and 7.2.3).
In other places (e.g. Hofbauer and Sigmund 1998) the term evolutionarily stable state is used instead of PSS but we prefer the latter to avoid the confusion between the terms evolutionarily stable strategy and evolutionarily stable state.
The adjective “standard” refers the fact that the replicator dynamics is considered for the pure strategies \(\mathbf e_1,\ldots ,\mathbf e_N\).
This is always true in the two cases (i) \(n=1\) and \(\mathbf p^*\not =\mathbf p_1\) and (ii) \(\mathbf p^*,\mathbf p_1,\ldots ,\mathbf p_n\) are distinct pure strategies.
Let \(F_i=\{\mathbf q \in S_N\,:\,q_i=0\}\). Then \(F_1,\ldots ,F_N\) are the faces of \(S_N\) and \(F=\cup _{p_i>0} F_i\).
However, \(\rho _{\hat{\mathbf q}}(\mathbf p,\hat{\mathbf q}, \varepsilon (\eta ))\not =\rho _{\mathbf q}(\mathbf p,\mathbf q,\eta )\), but \(\rho _{\mathbf q}(\mathbf p,\mathbf q,\eta ) =\frac{1}{\eta }[\varepsilon \rho _{\hat{\mathbf q}}(\mathbf p,\hat{\mathbf q},\varepsilon )- (\varepsilon -\eta )\rho _{\mathbf p}(\mathbf p,\hat{\mathbf q},\varepsilon (\eta ))]\).
References
Akin E (1982) Exponential families and game dynamics. Can J Math 34:374–405
Apaloo J (1997) Revisiting strategic models of evolution: the concept of neighborhood invader strategies. Theor Popul Biol 52:71–77
Apaloo J (2006) Revisiting matrix games of evolution: the concept of neighborhood invader strategies. Theor Popul Biol 69:235–242
Apaloo J, Brown JS, Vincent TL (2009) Evolutionary game theory: ESS, convergence stability, and NIS. Evolut Ecol Res 11:489–515
Balkenborg D, Schlag KH (2001) Evolutionarily stable sets. Intern J Game Theory 29:571–595
Bomze IM, Pötscher BM (1989) Game theoretical foundations of evolutionary stability. Volume 161 of lecture notes in economics and mathematical systems free preview. Springer, Berlin
Bomze IM, Pötscher BM (1993) Letter to the editor. Theor Popul Biol 161:405
Bomze IM, Weibull J (1995) Does neutral stability imply Lyapunov stability? Games Econ Behav 11:173–192
Broom M, Ruxton GD (1998) Evolutionarily stable stealing: game theory applied to kleptoparasitism. Behav Ecol 9:397–403
Broom M, Rychtář J (2013) Game-theoretical models in biology. Mathematical and computational biology. Chapman & Hall/CRC, New York
Broom M, Luther RM, Ruxton GD, Rychtář J (2008) A game-theoretic model of kleptoparasitic behavior in polymorphic populations. J Theor Biol 255:81–91
Broom M, Luther RM, Rychtář J (2009) Hawk–Dove game in kleptoparasitic populations. J Comb Inf Syst Sci 4:449–462
Broom M, Crowe ML, Fitzgerald MR, Rychtář J (2010) The stochastic modelling of kleptoparasitism using a Markov process. J Theor Biol 254:266–272
Broom M, Cressman R, Křivan V (2019) Revisiting the “fallacy of averages” in ecology: expected gain per unit time equals expected gain divided by expected time. J Theor Biol. https://doi.org/10.1016/j.jtbi.2019.109993
Charnov EL (1976) Optimal foraging: attack strategy of a mantid. Am Nat 110:141–151
Cressman R (1990) Strong stability and density-dependent evolutionarily stable strategies. J Theor Biol 145:147–165
Cressman R (1992) The stability concept of evolutionary game theory: a dynamic approach. Volume 94 lecture notes in biomathematics. Springer, Berlin
Cressman R (2003) Evolutionary dynamics and extensive form games. MIT Press, Cambridge
Garay J, Móri TF (2010) When is the opportunism remunerative? Commun Ecol 11:160–170
Garay J, Varga Z, Cabello T, Gámez M (2012) Optimal nutrient foraging strategy of an omnivore: Liebig’s law determining numerical response. J Theor Biol 310:31–42
Garay J, Cressman R, Xu F, Varga Z, Cabello T (2015) Optimal forager against ideal free distributed prey. Am Nat 186:111–122
Garay J, Csiszár V, Móri TF (2017) Evolutionary stability for matrix games under time constraints. J Theor Biol 415:1–12
Garay J, Cressman R, Móri TF, Varga T (2018) The ESS and replicator equation in matrix games under time constraints. J Math Biol. https://doi.org/10.1007/s00285-018-1207-0
Ginzburg LR (1983) Theory of natural selection and population growth. Benjamin/Cummings, San Francisco
Hines WGS (1980) Strategy stability in complex populations. J Appl Prob 17:600–610
Hirsch MW, Smale S, Devaney RL (2004) Differential equations, dynamical systems, and introduction to chaos, 2nd edn. Elsevier, Amsterdam
Hofbauer J, Sigmund K (1988) The theory of evolution and dynamical systems. Cambridge University Press, Cambridge
Hofbauer J, Sigmund K (1998) Evolutionary games and population dynamics. Cambridge University Press, Cambridge
Hofbauer J, Schuster P, Sigmund K (1979) A note on evolutionarily stable strategies and game dynamics. J Theor Biol 81:609–612
Holling CS (1959) The components of predation as revealed by a study of small-mammal predation of the European pine sawfly. Can Entomol 9:293–320
Křivan V, Cressman R (2017) Interaction times change evolutionary outcomes: two-player matrix games. J Theor Biol 416:199–207
Maynard Smith J (1974) The theory of games and the evolution of animal conflicts. J Theor Biol 47:209–221
Maynard Smith J (1982) Evolution and the theory of games. Oxford University Press, Oxford
Maynard Smith J, Price GR (1973) The logic of animal conflict. Nature 246:15–18
Roughgarden J (1983) The theory of coevolution. In: Futuyama DJ, Slatkin M (eds) Coevolution. Sinauer, Sunderland, pp 33–64
Sirot E (2000) An evolutionarily stable strategy for aggressiveness in feeding groups. Behav Ecol 11:351–356
Taylor PD, Jonker LB (1978) Evolutionarily stable strategies and game dynamics. Math Biosci 40(1):145–156
Thomas B (1985) On evolutionarily stable sets. J Math Biol 22(1):105–115
Vickers G, Cannings C (1987) On the definition of an evolutionarily stable strategy. J Theor Biol 129:349–353
Zeeman EC (1980) Population dynamics from game theory. In: Nitecki Z, Robinson C (eds) Global theory of dynamical systems. Volume of 819 lecture notes in mathematics. Springer, New York
Zeeman EC (1981) Dynamics of evolution of animal conflicts. J Theor Biol 89:249–270
Acknowledgements
This research was supported by the EU-funded Hungarian Grant EFOP-3.6.1-16-2016-00008 (to T. Varga). The project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant Agreement No. 690817 (to J. Garay and T. Varga). T. F. Móri was supported by the Hungarian National Research, Development and Innovation Office NKFIH—Grant No. K125569. J. Garay was supported by the Hungarian National Research, Development and Innovation Office NKFIH (GINOP 2.3.2-15-2016-00057).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
In this part we cite or claim some technical statements for the convenience of the reader and prove the new assertions appeared in the previous sections.
1.1 Auxiliary statements
Proof of Lemma 2.3
(i)\(\Leftrightarrow \)(ii) Consider a NE \(\mathbf p.\) By (2.5), we have
Multiplying by \(p_{i}\) we get
Since \(\mathbf p=\sum _{i\in \mathrm {supp}(\mathbf p)}p_i\mathbf e_i\) it follows that if we take the sum of the previous inequalities for \(i\in \mathrm {supp}(\mathbf p)\) we obtain that
that is, the sum of the left-hand sides of the inequalities is equal to the sum of the right-hand sides of the inequalities. This is possibile only if there is an equality in (2.5) for every \(i\in \mathrm {supp}(\mathbf p)\).
Now assume that (2.6) holds and \(\mathbf q\) is an arbitrary strategy. It is clear that (2.6) is equivalent to the inequalities
Multiply by \(q_i\) the ith inequality in (A.13). Taking the sum from \(i=1\) to \(i=N\) we get that
which is equivalent to
This means that (2.5) holds for every \(\mathbf q\in S_N\). Therefore \(\mathbf p\) is a NE.
(i)\(\Leftrightarrow \)(iii) We have just seen that if \(\mathbf p\) is a NE then \(\omega _{\mathbf p}(\mathbf p,\mathbf e_i,0)=\omega _{\mathbf e_i}(\mathbf p, \mathbf e_i,0)\) for every \(i\in \mathrm {supp}(\mathbf p)\) which implies that \(\omega _{\mathbf p}(\mathbf p,\mathbf q,0)=\omega _{\mathbf q}(\mathbf p,\mathbf q,0)\) whenever \(\mathrm {supp}(\mathbf q)\subset \mathrm {supp}(\mathbf p)\).
If \(\omega _{\mathbf p}(\mathbf p,\mathbf q,0)\ge \omega _{\mathbf q}(\mathbf p,\mathbf q,0)\) for every \(\mathbf q\in S_N\) with equality if \(\mathrm {supp}(\mathbf q)\subset \mathrm {supp}(\mathbf p)\) then (2.5) holds, so \(\mathbf p\) is a NE. \(\square \)
Lemma A.1
(Garay et al. 2017, Lemma 2) The following system of nonlinear equations in n variables,
where the coefficients \(c_{ij}\) are positive numbers, has a unique solution in the unite hypercube \([0,1]^{n}\).
Remark A.2
In Garay et al. (2017), Lemma 2 \(c_{ij}\) is assumed to be positive for every i, j but considering the proof a bit further one can see the validity of the lemma with non-negative \(c_{ij}\)-s too.
As claimed by the next lemma, the solution of the previous equation system varies continuously with the coefficients.
Lemma A.3
[Garay et al. 2018, Lemma 6.2 (i)] The solution \(\mathbf {u}=(u_1,u_2,\ldots ,u_n)\in [0,1]^{n}\) of (A.14) is a continuous function in
Lemma A.4
(Garay et al. 2018, Corollary 6.3) Consider a population of phenotypes \(\mathbf p^*,\mathbf p_1,\ldots ,\mathbf p_n\in S_N\) with frequencies \(1-x,x_1,\ldots ,x_n\) where \(x=x_1+x_2+\cdots +x_n\).
- (i)
The active parts \(\varrho ^*\) and \(\varrho _i\) (\(i=1,\ldots ,n\)) of the different phenotypes [see the definition at (2.8)] continuously depend on \(\mathbf {x}=(1-x,x_1,\ldots ,x_n)\).
- (ii)
If \(\mathbf {y}\) is another frequency distribution such that \(\bar{\mathbf h}(\mathbf {y}) =\bar{\mathbf h}(\mathbf {x})=:\bar{\mathbf h}\), then both \(\bar{\varrho }(\mathbf {x})=\bar{\varrho }(\mathbf {y})\), \(\varrho ^*(\mathbf {x})=\varrho ^*(\mathbf {y})\) and \(\varrho _i(\mathbf {x})=\varrho _i(\mathbf {y})\) (\(1\le i\le n\)).
- (iii)
If \(\bar{\mathbf h}\) can be uniquely represented as a convex combination of \(\mathbf p^*,\mathbf p_1,\ldots ,\mathbf p_n\)Footnote 7 then \(x_i\) must be equal to \(y_i\) for every i.
Obviously, if there exists only two phenotypes with positive frequency then the average strategy of the active subpopulation is a convex combination of the two phenotypes. Conversely, if a strategy is a convex combination of the two phenotypes, does there exist a composition which mix the strategy? The next lemma gives the precise answer.
Lemma A.5
(Garay et al. 2018, Lemma 6.4) Let \(\mathbf p,\mathbf q\in S_2\). Denote by \(\varrho _{\mathbf p}(\varepsilon ),\varrho _{\mathbf q}(\varepsilon )\) the unique solution in \([0,1]\times [0,1]\) of the system
Furthermore, \(\bar{\varrho }(\varepsilon ):=(1-\varepsilon )\varrho _{\mathbf p}(\varepsilon )+ \varepsilon \varrho _{\mathbf q}(\varepsilon )\) and
Then \(\mathbf r(0)=\mathbf p\), \(\mathbf r(1)=\mathbf q\) and \(\mathbf r(\varepsilon )\) uniquely runs through the line segment between \(\mathbf p\) and \(\mathbf q\) as \(\varepsilon \) runs from 0 to 1 in such a way that \(0\le \varepsilon _1<\varepsilon _2\le 1\) implies that \(||\mathbf r(\varepsilon _1)-\mathbf p||< ||\mathbf r(\varepsilon _2)-\mathbf p||\).
1.2 Proof of the main theorems
Now, we are ready to prove the two important static characterizations of the notion of UESS.
Proof of Theorem 3.1
Assume that \(\mathbf p\) is a UESS and let \(\varepsilon _0\) be the uniform threshold number in Definition 2.1. Denote by F the union of those facesFootnote 8 of \(S_N\) which do not contain \(\mathbf p\).
Necessity. We claim that
is a suitable choice for \(\delta \) (do not forget a positive function continuous on a compact set has a positive minimum). Indeed, let \(\mathbf q\) be a strategy for which \(0<||\mathbf p-\mathbf q||<\delta \) and take the type \(\mathbf q\) with frequency \(\eta (\in (0,1])\). Denote by \(\hat{\mathbf q}\) the point in F which can be represented in a form \(\mathbf p+\tau (\mathbf q-\mathbf p)\) with some \(\tau >0\) (that is \(\hat{\mathbf q}\) is the common point of the boundary of F and the half line from \(\mathbf p\) through \(\mathbf q\)).
By Lemma A.5, for any \(\eta \in (0,1]\) there is precisely one \(\varepsilon =\varepsilon (\eta )\in (0,1]\) such that
From this fact, a similar argument (the uniqueness provided by Lemma A.1) as in the proof of Lemma A.4(ii) shows that \(\bar{\rho }(\mathbf p,\hat{\mathbf q},\varepsilon (\eta ))= \bar{\rho }(\mathbf p,\mathbf q,\eta )\) and \(\rho _{\mathbf p}(\mathbf p,\hat{\mathbf q},\varepsilon (\eta )) =\rho _{\mathbf p}(\mathbf p,\mathbf q,\eta )\),Footnote 9 respectively. Hence, one can readily infer that
and, similarly,
Considering the choice of \(\delta \) and Lemma A.5 we infer from the fact that \(||\mathbf p-\mathbf r\big (\hat{\mathbf q},\varepsilon (\eta )\big )|| =||\mathbf p-\mathbf r(\mathbf q,\eta )||\le ||\mathbf p-\mathbf q||<\delta \) that \(0<\varepsilon (\eta )\le \varepsilon _0\). We thus obtain
for every \(\mathbf q\) with \(||\mathbf q-\mathbf p||<\delta \) and for every \(\eta \in (0,1]\).
Sufficiency. Let us turn to the other direction. Let \(\hat{\mathbf q}\in F\) and let \(\mathbf q\not =\hat{\mathbf q}\) be a point on the segment between \(\mathbf p\) and \(\hat{\mathbf q}.\) We have seen in the proof of the previous direction that for any \(\eta \in (0,1]\) there is a unique \(\varepsilon =\varepsilon (\eta )\) such that \(\mathbf r(\hat{\mathbf q},\varepsilon )=\mathbf r(\mathbf q,\eta )\). We show that \(\varepsilon <\eta \).
Writing out the definition of \(\mathbf r(\hat{\mathbf q},\varepsilon )\) and \(\mathbf r(\mathbf q,\eta )\) their equality means
By Lemma A.5 there is a \(\theta \in (0,1)\) such that \(\mathbf r(\hat{\mathbf q},\theta )=\mathbf q\). Replace \(\mathbf q\) with this representation in (A.15). We get
Lemma A.4(iii) yields the coefficients of \(\mathbf p\) in the two sides to be equal:
Since \(\bar{\rho }(\mathbf p,\hat{ \mathbf q},\varepsilon ) =\bar{\rho }(\mathbf p,\mathbf q,\eta )\) (see the proof of the previous direction) it can be simplified by \(\bar{\rho }(\mathbf p,\hat{\mathbf q},\varepsilon ) =\bar{\rho }(\mathbf p,\mathbf q,\eta )\) which results in
Since the second term of the right-hand side is positive it is inferred that
Recalling that \(\rho _{\mathbf p}(\mathbf p,\hat{\mathbf q},\varepsilon ) =\rho _{\mathbf p}(\mathbf p,\mathbf q,\eta )\) (see the proof of the previous direction) one can conclude that \(\varepsilon <\eta \). This immediately implies if for some \(\varepsilon _0(\hat{\mathbf q})>0\) we have that
whenever \(0<\varepsilon <\varepsilon _0(\hat{\mathbf q})\) then, also,
whenever \(0<\eta <\varepsilon _0(\hat{\mathbf q})\), so such an \(\varepsilon _0(\hat{\mathbf q})\) is uniform on the segment between \(\hat{\mathbf q}\) and \(\mathbf p\).
It is clear from the argument made hitherto that the half of the unique \(\varepsilon >0\) for which \(||\mathbf r(\hat{\mathbf q},\varepsilon )-\mathbf p||=\delta \) is an appropriate choice for such an \(\varepsilon _0(\hat{\mathbf q})\) (if \(||\hat{\mathbf q}-\mathbf p||<\delta \) let \(\varepsilon _0(\hat{\mathbf q})\) set to be, say, 1 / 2). With this choice of \(\varepsilon _0(\hat{\mathbf q})\) we have
on the one hand, and
on the other, for any \(0<\varepsilon \le \text {(sic!)}\, \varepsilon _0(\hat{\mathbf q})\). Since, by Lemma A.3, \(\rho _{\mathbf q},\rho _{\mathbf p},\bar{\rho }\) are continuous in \((\mathbf q,\varepsilon )\in S_N\times [0,1]\) it follows that \(\omega _{\mathbf q},\omega _{\mathbf p},\bar{\omega }\) and \(\mathbf r(\mathbf q,\varepsilon )\) are also continuous. Consequently, for every \(\hat{\mathbf q}\in F\), there exists a \(\lambda =\lambda (\hat{\mathbf q})\) such that
whenever \(\hat{\mathbf s}\in S_N\) and \(||\hat{\mathbf s}-\hat{\mathbf q}|| <\lambda (\hat{\mathbf q})\). Taking into account the definition of \(\delta \) and Lemma A.5, we find it is also valid that
whenever \(0<\varepsilon \le \text {(sic!)}\, \varepsilon _0(\hat{\mathbf q})\). By the compactness of F, there are (finitely many) \(\hat{\mathbf q}_1\), \(\ldots \), \(\hat{\mathbf q}_k\in F\) such that the union of the open ball with centers at \(\hat{\mathbf q}_1\), \(\ldots \), \(\hat{\mathbf q}_k\) and with radius \(\lambda (\hat{\mathbf q}_1)\), \(\ldots \), \(\lambda (\hat{\mathbf q}_k)\) in turn covers F. Then
is a suitable choice for \(\varepsilon _0\) in the definition of the UESS. \(\square \)
Proof of Theorem 3.2
The necessity is clear. The problematic direction is the sufficiency. Assume that (3.1) holds with \(\varepsilon =1\) for any \(\mathbf u\) with \(0<||\mathbf u-\mathbf p||<\delta \) and let \(\mathbf q\) be an arbitrary element of \(S_n\) with \(0<||\mathbf q-\mathbf p||<\delta \). We validate that, for any \(\varepsilon \in (0,1]\), (3.1) holds. For \(\varepsilon =1\), this is just the assumption. For \(0<\varepsilon <1\), let
By Lemma A.5, \(\mathbf r(\varepsilon )\) is on the segment matching \(\mathbf p\) and \(\mathbf q\) such that
On the other hand, by Proposition 2.6, the mean fitness of the population consisting of \(\mathbf p\) and \(\mathbf q\) individuals with proportions \((1-\varepsilon )\) and \(\varepsilon \), respectively, corresponds to the fitness of a population consisting of only \(\mathbf r(\varepsilon )\) individuals which can be viewed as a population consisting of \(\mathbf p\) and \(\mathbf r(\varepsilon )\) individuals with proportions 0 and 1, respectively. Considering Proposition 2.6, the last interpretation also shows that the fitness of a \(\mathbf p\) individual in a population consisting of only \(\mathbf r(\varepsilon )\) individuals is equal to that in population of the \(\mathbf p,\mathbf q\) individuals. Formally, this means that \( {\bar{\omega }}(\mathbf p,\mathbf q,\varepsilon )=\omega _{\mathbf r(\varepsilon )}(\mathbf p,\mathbf r(\varepsilon ),1). \) and \(\omega _{\mathbf p}(\mathbf p,\mathbf r(\varepsilon ),1)= \omega _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon ). \) By assumption, \(\omega _{\mathbf r(\varepsilon )}(\mathbf p,\mathbf r(\varepsilon ),1)< \omega _{\mathbf p}(\mathbf p,\mathbf r(\varepsilon ),1)\) because \(||\mathbf r(\varepsilon )-\mathbf p||< \delta \). We immediately conclude that \({\bar{\omega }}(\mathbf p,\mathbf q,\varepsilon )< \omega _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon )\) which can be possible if and only if \(\omega _{\mathbf q}(\mathbf p,\mathbf q,\varepsilon )< \omega _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon )\). \(\square \)
Proof of Theorem 3.3
Assume that \(\mathbf p^*\in S_2\) is a UESS. We prove that the set of states \(\mathbf x=(x_1,\ldots ,x_n)\) with \(\sum [x_i \varrho _i(\mathbf x)/{\bar{\varrho }}(\mathbf x)] \mathbf p_i=\mathbf p^*\) is locally asymptotically stable. Consider the phenotypes \(\mathbf p_1,\ldots , \mathbf p_k,\mathbf p_{k+1},\ldots ,\mathbf p_n\) in the ascending order of the second coordinates, that is, \(p_1^2<p_2^2<\cdots<p_k^2\le p^*_2<p_{k+1}^2<\cdots <p_n^2\) where \(p_i^1\mathbf e_1+p_i^2\mathbf e_2=\mathbf p_i\) (\(i=1,2,\ldots ,n\)) and \(p^*_1 \mathbf e_1+p^*_2\mathbf e_2=\mathbf p^*\).
Assume that \(\mathbf x\in S_n\) is a state such that
where
and \(\delta \) comes from Theorem 3.2. This implies that
where
It can be assumed that the second coordinate \({\bar{h}}_2(\mathbf x)\) of \(\bar{\mathbf h}(\mathbf x)\) is strictly greater than \(p^*_2\) (the case when \({\bar{h}}_2(\mathbf x)\) is strictly smaller can be treated in a similar way and when \({\bar{h}}_2(\mathbf x)=p^*_2\) then \(\mathbf x\) is an equilibrium point of the replicator dynamics).
Observation 1
We prove that \(W_i>{\bar{W}}\) for \(1\le i\le k\) and \(W_i<{\bar{W}}\) for \(k+1\le i\le n\).
Recall the assumption that \(p_2^*<{\bar{h}}_2\). Consequently, if \(1\le i\le k\) then there is an \(0\le \alpha _i= \alpha _i(\mathbf x)\le 1\) such that \(\mathbf p^*=\alpha _i\mathbf p_i+(1-\alpha _i)\bar{\mathbf h}\) (it is easy to check that \(\alpha _i=(p_1^*-{\bar{h}}_1)/(p_i^1-{\bar{h}}_1)\)). Therefore we have
(Note that \(\alpha _i\varrho ^*/\varrho _i+(1-\alpha _i)\varrho ^*/{\bar{\varrho }}=1.\)) From this we immediately infer that
Similarly, if \(k+1\le i\le n\), then \(\bar{ \mathbf h}\) lies on the line segment connecting \(\mathbf p^*\) to \(\mathbf p_i\). So there is a \(0<\beta _i= \beta _i(\mathbf x)\le 1\) such that \(\bar{\mathbf h}=\beta _i \mathbf p^*+(1-\beta _i)\mathbf p_i\) (it is easy to check that \(\beta _i=({\bar{h}}_1-p_i^1)/(p_1^*-p_i^1)\)). Therefore we have that
which immediately implies that
In summary, if \(0<||\mathbf p^*-\bar{\mathbf h}(\mathbf x)||<\delta '\) and \(p_2^*<\bar{h}_2(x)\) then
for i with \(p_i^2\le p_2^*\) and \(x_i>0\), we have that \(\dot{x}_i=x_i[W_i(\mathbf x)-{\bar{W}}(\mathbf x)]>0\), and
for i with \(p_i^2>p_2^*\) and \(x_i>0\), we have that \(\dot{x}_i=x_i[W_i(\mathbf x)-{\bar{W}}(\mathbf x)]<0\), respectively.
Observation 2
Let \(\mathbf x,\mathbf y\in S_n\) be states such that
- (i)
\(x_1\le y_1,\ldots ,x_k\le y_k\) but \( \sum _{i=1}^kx_i<\sum _{i=1}^ky_i\);
- (ii)
\(y_{k+1}\le x_{k+1},\ldots ,y_n\le x_n\) but \(\sum _{i=k+1}^ny_i<\sum _{i=k+1}^nx_i\);
- (iii)
\(p_2^*<\bar{h}_2(\mathbf x)\), \(p_2^*<\bar{h}_2(\mathbf y)\); and
- (iv)
\(0<||\mathbf p^*-\bar{\mathbf h}(\mathbf x)||<\delta '\), \(0<||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||<\delta '\), respectively.
We prove that \(||\mathbf p^*-\bar{\mathbf h}(\mathbf x)||>||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||\). Suppose the contrary, that is, there are states \(\mathbf x,\mathbf y\) satisfying the previous conditions (i)–(iv), but with \(||\mathbf p^*-\bar{\mathbf h}(\mathbf x)||\le ||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||\). Consider a state \(\mathbf z^0\in S_n\) with \(z_i^0=0\) if \(1\le i\le k\) and with \(z_i^0\ge x_i\) if \(k+1\le i\le n\). Then \(p_{k+1}^2\le {\bar{h}}_2(\mathbf z^0)\). Hence
Now take \(S_n\ni (z_1,\ldots ,z_n)=\mathbf z=\mathbf z^0\) and start to increase the first coordinate of \(\mathbf z\) in the following way:
\(z_1\) cannot be greater, then \(x_1\);
as \(z_1\) is increasing \(z_{k+1},\ldots ,z_n\) is decreasing but \(z_i\) cannot be less, than \(x_i\) if \(k+1\le i\le n\), say, first we decrease \(z_n\) until \(z_n=x_n\) then we decrease \(z_{n-1}\) until \(z_{n-1}=x_{n-1}\) and so on;
if for some \(0<z_1\le x_1\) we have that \(||\mathbf p^*-\bar{\mathbf h}(\mathbf z)||=||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||\), then we stop;
if for every \(0<z_1\le x_1\) we have that \(||\mathbf p^*-\bar{\mathbf h}(\mathbf z)||>||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||\), then we set \(z_1\) to be \(x_1\) and start to increase \(z_2\) repeating the process replacing the index 1 with index 2;
if we do not find a \(\mathbf z\) with \(||\mathbf p^*-\bar{\mathbf h}(\mathbf z)||=|| \mathbf p^*-\bar{\mathbf h}(\mathbf y)||\) by moving \(z_2\), then we set \(z_2\) to be \(x_2\) and start to increase \(z_3\) and so on.
As \(\bar{\mathbf h}\) is continuous in \(\mathbf z\) and
we must find a \(\mathbf z\in S_n\) such that (i) \(z_i\le x_i\le y_i\) if \(1\le i \le k\) but \(\sum _{i=1}^kz_i<\sum _{i=1}^ky_i\), (ii) \(y_i\le x_i\le z_i\) if \(k+1\le i\le n\) but \(\sum _{i=k+1}^ny_i<\sum _{i=k+1}^nz_i\), (iii) \(||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||=||\mathbf p^*-\bar{\mathbf h}(\mathbf z)||\), (iv) \(p_2^*<{\bar{h}}_2(\mathbf z)\). Since \(p_2^*<{\bar{h}}_2(\mathbf y)\) also holds \(\bar{\mathbf h}(\mathbf z)\) must be equal to \(\bar{\mathbf h}(\mathbf y)\). As \({\bar{\varrho }}\) is the solution in [0, 1] of the equation \({\bar{\varrho }}=1/(1+\bar{\mathbf h}T{\bar{\varrho }} \bar{\mathbf h})\) we also have
Hence
Consequently, we get that
Since \(\sum _{i=1}^nz_i\frac{\varrho _i(\mathbf z)}{{\bar{\varrho }}(\mathbf z)}= \sum _{i=1}^ny_i\frac{\varrho _i(\mathbf y)}{{\bar{\varrho }}(\mathbf y)}=1\), \({\bar{\varrho }}(\mathbf y)={\bar{\varrho }}(\mathbf z)\) and \(\varrho _i(\mathbf y)=\varrho _i(\mathbf z)\) (\(i=1,\ldots ,n\)) it follows that \(\sum _{i=1}^n(z_i-y_i) \frac{\varrho _i(\mathbf z)}{{\bar{\varrho }}(\mathbf z)}=0\) and (A.19) can be continued as
which is a contradiction. This validates that \(||\mathbf p^*-\bar{\mathbf h}(\mathbf x)||>||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||\) providing that \(x_1\le y_1,\ldots ,x_k\le y_k\), \(\sum _{i=1}^kx_i<\sum _{i=1}^ky_i\), \(y_{k+1}\le x_{k+1}, \ldots ,y_n\le x_n\), \(\sum _{i=k+1}^ny_i<\sum _{i=k+1}^nx_i\), \(p_2^*<{\bar{h}}_2(\mathbf x)\), \(p_2^*<{\bar{h}}_2(\mathbf y)\), \(0<||\mathbf p^*-\bar{\mathbf h}(\mathbf x)||<\delta '\) and \(0<||\mathbf p^*-\bar{\mathbf h}(\mathbf y)||<\delta '\).
Conclusion
In summary, if \(0<||\bar{\mathbf h}(\mathbf x)-\mathbf p^*||<\delta '\) and \(p_2^*<\bar{h}_2(\mathbf x)\) it follows that for a positive \(x_i\) the expression \(x_i[W_i(\mathbf x)-{\bar{W}}(\mathbf x)]\) is strictly positive or strictly negative according as \(1\le i\le k\) or \(k+1 \le i \le n\) (Observation 1). This means that if \(x_i>0\) then \(x_i\) strictly increases for \(1\le i \le k\) and \(x_i\) strictly decreases for \(k+1\le i\le n\), respectively, which, by Observation 2, implies that \({\bar{h}}_2(\mathbf x)\) has to strictly decrease until reaching \(p_2^*\). If \({\bar{h}}_2(\mathbf x)<p_2^*\) then a similar argument shows that \({\bar{h}}_2(\mathbf x)\) tends to \(p_2^*\) strictly increasingly. Consequently, \(\bar{\mathbf h}(\mathbf x)\rightarrow \mathbf p^*\) in a monotone way. \(\square \)
Proof of Theorem 3.4
We trace back the proof to Garay et al. (2018, Theorem 4.2) which says that \(\mathbf p^*\) is UESS providing that the state \(\mathbf y \in S_2\) with
is a locally asymptotically stable rest point of the replicator dynamics with respect to \(\mathbf e_1,\mathbf e_2\in S_2\). (Here \(\varrho _{\mathbf e_1}=\varrho _{\mathbf e_1}(\mathbf y,\mathbf e_1,\mathbf e_2)\) and \(\varrho _{\mathbf e_2}=\varrho _{\mathbf e_2}(\mathbf y,\mathbf e_1,\mathbf e_2)\) are the solution of the equation system
and \({\bar{\varrho }}=y_1\varrho _{\mathbf e_1}+y_2\varrho _{\mathbf e_2}.\))
It is not too difficult to see that this statement remains true if we replace \(\mathbf e_1,\mathbf e_2\) with strategies \(\mathbf q_1\not =\mathbf q_2\) with \(q_1^2<p^*_2<q_2^2\) where \(\mathbf q_i=(q_i^1,q_i^2)=q_i^1\mathbf e_1+q_i^2\mathbf e_2\) and \(\mathbf p^*=(p_1^*,p_2^*)=p_1^*\mathbf e_1+p_2^*\mathbf e_2\).
It can be assumed that \(p_1^2<p^*_2<p_2^2\) or change the order of \(\mathbf p_1,\ldots ,\mathbf p_n\). For the replicator dynamics with respect to \(\mathbf p_1,\ldots ,\mathbf p_n\), consider the initial value problem with \(x_i(0)=0\) for \(i=3,4,\ldots ,n\) and with \(x_1(0),x_2(0)\) such that \(||\bar{\mathbf h}\big (\mathbf x(0)\big )-\mathbf p^*||<\delta \) where \(\delta \) comes from the remark about the local asymptotic stability of set G after Theorem 3.3. Because of the uniqueness of solutions of differential equations with continuously differentiable right-hand side [see e.g. p. 144 in Hirsch et al. (2004)] it follows that \(x_1(t)=y_1(t)\), \(x_2(t)=y_2(t)\) and \(x_i(t)=0\) (\(t\ge 0\), \(i=2,3,\ldots ,n\)) where \(y_1(t),y_2(t)\) is the solution of the initial value problem \(y_1(0)=x_1(0),\)\(y_2(0)=x_2(0)\) for the replicator dynamics with respect to \(\mathbf p_1,\mathbf p_2\). To finish the proof one should only apply (Garay et al. 2018, Theorem 4.2) for the replicator dynamics with respect to \(\mathbf p_1,\mathbf p_2\). \(\square \)
1.3 Proof of the statements of Sect. 4
Proof of Theorem 4.1
By Remark 2.4, we can assume that \(\mathbf p=\mathbf e_1\). Let F denote the surface of \(S_N\) determined by the vertices \(\mathbf e_2,\ldots ,\mathbf e_N\) that is \(F=\{\mathbf q\in S_N\,:\,q_1=0\}\). By Definition 2.2, we have that
for every \(\mathbf q \not =\mathbf p\), in particular, for every \(\mathbf q\in F\). Since \(F\times \{0\}\) is a compact set from the continuity of \(\omega _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon )-\omega _{\mathbf q} (\mathbf p,\mathbf q,\varepsilon )\) in \((\mathbf q,\varepsilon )\) we infer the existence of an \(\varepsilon _0>0\) such that
for any \(\hat{\mathbf q}\in F\) and \(0\le \eta \le \varepsilon _0.\)
We prove that Definition 2.1 holds with this \(\varepsilon _0\). To see this let \(\mathbf q\) be an arbitrary strategy distinct from \(\mathbf p\) and \( 0\le \varepsilon \le \varepsilon _0\). Take the strategy \(\hat{\mathbf q}\in F\) such that \(\mathbf q\) lies on the segment between \(\mathbf p\) and \(\hat{\mathbf q}\). Then, using Lemma A.5 as in the proof of Theorem 3.1, we infer that there is an \(\eta =\eta (\varepsilon )\le \varepsilon \) such that
where \({\bar{\rho }}(\mathbf p,\mathbf q,\varepsilon )= (1-\varepsilon )\rho _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon )+ \varepsilon \rho _{\mathbf q}(\mathbf p,\mathbf q,\varepsilon )\). Note that, as in the necessity part of the proof of Theorem 3.1, this implies that \({\bar{\rho }}(\mathbf p,\mathbf q,\varepsilon )= {\bar{\rho }}(\mathbf p,\hat{\mathbf q},\eta )\), \(\rho _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon )= \rho _{\mathbf p}(\mathbf p,\hat{\mathbf q},\eta )\) and \(\omega _{\mathbf p}(\mathbf p,\mathbf q,\varepsilon )= \omega _{\mathbf p}(\mathbf p,\hat{\mathbf q},\eta )\), respectively.
Therefore we have
By (A.20), it can be continued as
so by comparing the rightmost side with the leftmost one we get that \(\omega _{\mathbf q}(\mathbf p,\mathbf q,\varepsilon )<\omega _{\mathbf p}(\mathbf p,\mathbf q, \varepsilon )\) for every \(\mathbf q\not =\mathbf p\) and \(0\le \varepsilon \le \varepsilon _0\) which proves that \(\mathbf p\) is a UESS. \(\square \)
Proof of Lemma 4.3
\(\mathbf p^*\) being a UESS ensures the existence of an \(\varepsilon _0>0\) such that
for every \(\mathbf q\not = \mathbf p^*\) and every \(0<\varepsilon \le \varepsilon _0\). Let \(0<x\le \varepsilon _0\) and \(\mathbf x:=(x^*,x_1,\ldots ,x_n)\in [0,1]^{n+1}\) such that \(x_1+\cdots +x_n=x\) and \(x^*=1-x\). Consider \(\breve{\varrho }(\mathbf x)\), \(\breve{\mathbf h}(\mathbf x)\) and \(\breve{W}(\mathbf x)\), respectively, as in Propositions 2.6. It is clear that \(\breve{\mathbf h}(\mathbf x)\) is a convex combination of \(\mathbf p_1,\ldots ,\mathbf p_n\), so \(\breve{\mathbf h}(\mathbf x)\not =\mathbf p^*\). By Proposition 2.6,
If we set \(\varepsilon =x\) in (A.21), (A.22) shows that
Hence \(W^*(\mathbf x)>(1-x)W^*(\mathbf x)+x\breve{W}(\mathbf x)={\bar{W}}(\mathbf x)\) whenever \(0<x<\varepsilon _0\), that is, \(\varepsilon _0\) is a suitable choice for \(\delta \). \(\square \)
Proof of Corollary 4.5
Let \(\varepsilon _0\) be the same as in the previous proof. Assume that \(0<||(1,0,\ldots ,0)-\mathbf x||<\varepsilon _0/n\). Then
By the previous lemma, therefore, it is true that
whenever \(0<||(1,0,\ldots ,0)-\mathbf x||<\varepsilon _0/n\). \(\square \)
Proof of Lemma 4.6
-
(a)
See in Garay et al. (2018).
-
(b)
Note that, by Proposition 2.6 and 2.8, \(\bar{\mathbf h}(\hat{\mathbf x})=\mathbf p^*\) and
$$\begin{aligned} {\bar{W}}(\hat{\mathbf x})\big [=\breve{W}(\hat{\mathbf x})\big ]=\omega (\mathbf p^*)= \frac{\mathbf p^* A \mathbf p^* \rho (\mathbf p^*)}{1+\mathbf p^* T \mathbf p^* \rho (\mathbf p^*)}. \end{aligned}$$Since \(\hat{\mathbf x}\) is an interior rest point the right hand side of the replicator dynamics can vanish if and only if \(W_i(\hat{\mathbf x})={\bar{W}}(\mathbf x)\) for every i which, by Lemma 2.3, immediately implies that \(\mathbf p^*\) is a NE.
-
(c)–(d)
Since \(\hat{\mathbf x}\) is also a rest point it follows that if \(\hat{x}_i>0\) then
$$\begin{aligned} W_i(\hat{\mathbf x})={\bar{W}}(\hat{\mathbf x})= \frac{\mathbf p^* A \mathbf p^* \rho (\mathbf p^*)}{1+\mathbf p^* T \mathbf p^* \rho (\mathbf p^*)} \end{aligned}$$must hold. Therefore, if, contrary to our claim, \(\mathbf p^*\) is not a NE then the equilibrium condition is hurt only for some i with \(\hat{x}_i=0\). For such an index i, say \(i_0\), there is an \(\varepsilon >0\) such that
$$\begin{aligned} \underbrace{\frac{\mathbf p_{i_0} A \mathbf p^* \rho (\mathbf p^*)}{1+\mathbf p_{i_0}T \mathbf p^* \rho (\mathbf p^*)}}_{W_{i_0}(\hat{\mathbf x})}- \underbrace{\frac{\mathbf p^* A \mathbf p^* \rho (\mathbf p^*)}{1+\mathbf p^* T \mathbf p^* \rho (\mathbf p^*)}}_{{\bar{W}}(\hat{\mathbf x})}= \varepsilon >0. \end{aligned}$$So, by continuity, \(W_{i_0}(\mathbf x)-{\bar{W}}(\mathbf x)>\varepsilon /2\) in a bounded neighbourhood \({\mathbb {H}}\) of \(\hat{\mathbf x}\). Henceforth, the way of the proofs of c) and d) branch off:
- (c)
Since \(\hat{\mathbf x}\) is stable there is another neighbourhood H of \(\hat{\mathbf x}\) such that for any solution starting from H remains in \({\mathbb {H}}\) forever. Take an arbitrary \(\mathbf x\in H\) with \(x_{i_0}\not =0\) and consider this \(\mathbf x\) as an initial value. Then
$$\begin{aligned} \log x_{i_0}(t)&=\log x_{i_0}(0)+ \int _0^t\frac{\dot{x}_{i_0}(s)}{x_{i_0}(s)}\,\mathrm {d}s\\&= \log x_{i_0}(0)+ \int _0^tW_{i_0}(\mathbf x(s))-{\bar{W}}(\mathbf x(s))\,\mathrm {d}s\\&>\log x_{i_0}(0)+\int _0^t\frac{\varepsilon }{2}\,\mathrm {d}s= \log x_{i_0}(0)+t\frac{\varepsilon }{2}\rightarrow \infty \quad \text {as}\quad t\rightarrow \infty \end{aligned}$$contradicting that \(\mathbf x(t)\) remains in \({\mathbb {H}}\).
- (d)
Since \(\hat{\mathbf x}\) is the only member of the \(\omega \)-limit of \(\mathbf x(t)\) and \(\mathbf x(t)\) is bounded it follows that \(\mathbf x(t)\rightarrow \hat{\mathbf x}\), in particular, \(x_{i_0}(t)\rightarrow {\hat{x}}_{i_0}=0\). Hence, there is a \(t_0\) such that \(\mathbf x(t)\in {\mathbb {H}}\) whenever \(t\ge t_0\) from which we infer that \( \log x_{i_0}(t)\rightarrow \infty \) as in the proof of c) which contradicts \(x_{i_0}(t)\) tending to 0.
- (c)
\(\square \)
Proof of Theorem 4.7
By Corollary 4.5 the vector \((1,0,\ldots ,0)\in S_{n+1}\) is a PSS of the replicator dynamics which is an asymptotically stable rest point of the replicator dynamics in accordance with the remark immediate after Definition 4.2. \(\square \)
In the next proof we again apply Theorem 3.2.
Proof of Theorem 4.9
We use the notation of Proposition 2.6. Since Proposition 2.8 says that \((1,0,\ldots ,0)\in S_{n+1}\) is a rest point (moreover every point \(\mathbf x\) for which \(\bar{\mathbf h}(\mathbf x)=\mathbf p^*\) is a rest point) we should only prove the stability of \((1,0,\ldots ,0)\). By Lemma A.4, \(\bar{\mathbf h}(\mathbf x)\) is continuous in \(\mathbf x=(x^*,x_1,\ldots ,x_n)\). On the other hand, \(\mathbf p^*\) is a UESS. Therefore, \(\omega _{\mathbf p^*}\big (\mathbf p^*,\bar{\mathbf h}(\mathbf x),1\big ) \ge \omega _{\bar{\mathbf h}(\mathbf x)}\big (\mathbf p^*,\bar{\mathbf h}(\mathbf x),1\big )\) in a neighbourhood of \((1,0,\ldots ,0)\) (see Theorem 3.2). (The strict inequality is not sure at all because \(\bar{\mathbf h}(\mathbf x) =\mathbf p^*\) can occur!) By Proposition 2.6, \(\omega _{\mathbf p^*}\big (\mathbf p^*,\bar{\mathbf h}(\mathbf x),1\big )=W^*(\mathbf x)\) and \(\omega _{\bar{\mathbf h}(\mathbf x)}\big (\mathbf p^*,\bar{\mathbf h}(\mathbf x),1\big )={\bar{W}}(\mathbf x)\). Therefore, \(W^*(\mathbf x)-{\bar{W}}(\mathbf x)\ge 0\) holds at least in a neighbourhood of \((1,0,\ldots ,0)\).
From here, follow the proof of Hofbauer and Sigmund (1998, Theorem 7.2.4). Take the function \(P(\mathbf x)=-\log x^*\) which is positive semidefinite (that is \(P(\mathbf x)\ge 0\)) in a neighbourhood of \((1,0,\ldots ,0)\) in \(S_{n+1}\). As regards the derivative of \(P(\mathbf x)\) along the replicator dynamics we have
that is, \(\dot{P}(\mathbf x)\) is negative semidefinite in a neighbourhood of \((1,0,\ldots ,0)\). By Lyapunov’s theorem on stability [see for example (Hirsch et al. 2004, p. 194) or (Hofbauer and Sigmund 1998, Theorem 2.6.1)], this ensures the local stability of \((1,0,\ldots ,0)\) in the simplex \(S_{n+1}\). \(\square \)
Rights and permissions
About this article
Cite this article
Varga, T., Móri, T.F. & Garay, J. The ESS for evolutionary matrix games under time constraints and its relationship with the asymptotically stable rest point of the replicator dynamics. J. Math. Biol. 80, 743–774 (2020). https://doi.org/10.1007/s00285-019-01440-6
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00285-019-01440-6
Keywords
- Evolutionary stability
- Matrix game
- Time constraint
- Monomorphic
- Polymorphic
- Population game
- Replicator dynamics