Abstract
In this paper, we study evolutionary games and we examine the stability of the evolutionarily stable strategy (ESS) in the continuous-time replicator dynamics with distributed time delays. In many examples, the interactions between individuals take place instantaneously, but their impacts are not immediate and take a certain amount of time which is usually random. In this paper, we study the consequences of distributed delays on the stability of the replicator dynamics. The main results are that (i) under the exponential delay distribution, the ESS is asymptotically stable for any value of the rate parameter; (ii) the necessary and sufficient conditions for the asymptotic stability of the ESS under uniform and Erlang distributions; and (iii) for the discrete distributed delays, we derive a necessary and sufficient delay-independent stability conditions. We illustrate our results with numerical examples from the Hawk–Dove game.
Similar content being viewed by others
References
Alboszta J, Miekisz J (2004) Stability of evolutionarily stable strategies in discrete replicator dynamics with time delay. J Theor Biol 231(2):175179
Bélair J, Campbell SA (1994) Stability and bifurcations of equilibria in a multiple-delayed differential equation. SIAM J Appl Math 54:1402–1424
Bellman R, Cooke KL (1963) Differential difference equations. Academic Press, New York
Ben Khalifa N, El-Azouzi R, Hayel Y (2014) Delayed evolutionary game dynamics with non-uniform interactions in two communities. In: Proceedings of IEEE CDC, Los Angeles, California, USA, pp 3809–3814
Ben Khalifa N, El-Azouzi R, Hayel Y (2015) Random time delays in evolutionary game dynamics. In: Proceedings of IEEE CDC, Osaka, Japan, pp 3840–3845
Ben-Khalifa N, Hayel Y, El-Azouzi R, Mabrouki I (2016) Evolutionary games in interacting communities. Dyn Games Appl 1:1–26
Berezansky L, Braverman E (2006) On stability of some linear and nonlinear delay differential equations. J Math Anal Appl 314:391–411
Berezansky L, Braverman E (2007) Explicit exponential stability conditions for linear differential equations with several delays. J Math Anal Appl 332:246–264
Chen J, Latchman HA (1994) Asymptotic stability independent of delays: simple necessary and sufficient conditions. In: Proceedings of the American Control Conference, Baltimore, Maryland, USA, pp 1027–1031
Gopalsamy K (1992) Stability and oscillations in delay differential equations of population dynamics. Kluwer Academic Publisher, London
Hale JK, Huang WZ (1993) Global geometry of the stable regions for two delay differential equations. J Math Anal Appl 178:344–362
Hofbauer J, Sigmund K (1998) Evolutionary games and population dynamics. Cambridge University Press, Cambridge
Iijima R (2011) Heterogeneous information lags and evolutionary stability. Math Soc Sci 61:83–85
Iijima R (2012) On delayed discrete evolutionary dynamics. J Theor Biol 300:1–6
Kuang Y (1993) Delay differential equations: with applications in population dynamic. Academic Press, New York
Li X, Ruan S, Wei J (1999) Stability and bifurcation in delay-differential equations with two delays. J Math Anal Appl 236:254–280
Matsumoto A, Szidarovszky F (2012) An elementary study of a class of dynamic systems with two time delays. Cubo 14(3):103–113
Maynard Smith J (1982) Evolution and the theory of games. Cambridge University Press, Cambridge
Moreira JA, Pinheiro FL, Nunes A, Pacheco JM (2012) Evolutionary dynamics of collective action when individual fitness derives from group decisions taken in the past. J Theor Biol 298:8–15
Nowak MA (2006) Evolutionary dynamics: exploring the equations of life. Harvard University Press, Cambridge
Oaku H (2002) Evolution with delay. Jpn Econ Rev 53:114–133
Piotrowska MJ (2007) A remark on the ode with two discrete delays. J Math Anal Appl 329(1):664–676
Taylor PD, Jonker LB (1978) Evolutionarily stable strategies and game dynamics. Math Biosci 40:145–156
Tembine H, Altman E, El-Azouzi R (2007) Asymmetric delay in evolutionary games. In: Valuetools, Nantes, France, pp 1–8
Tembine H, Altman E, El-Azouzi R, Hayel Y (2011) Bio-inspired delayed evolutionary game dynamics with networking applications. Telecommun Syst 47:137–152
Vincent TL, Brown JS (2005) Evolutionary game theory, natural selection, and darwinian dynamics. Cambridge University Press, Cambridge
Weibull JW (1995) Evolutionary game theory. MIT Press, Cambridge
Wesson E, Rand R (2016) Hopf bifurcations in delayed rock–paper–scissors replicator dynamics. Dyn Games Appl 6(1):139–156
Wessonn E, Rand R, Rand D (2016) Hopf bifurcations in two-strategy delayed replicator dynamics. Int J Bifurc Chaos 26(1):1650006
Yi T, Zuwang W (1997) Effect of time delay and evolutionarily stable strategy. J Theor Biol 187(1):111–116
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix A: Proof of Theorem 1
We suppose \(\tau _\mathrm{max}<\frac{\pi ^2}{2D}\). If \(\lambda \) is a real solution of (5), then it is clear that \(\lambda \) cannot be positive. Let \(\lambda =u+iv\) be a complex root of (5) with \(u > 0\) and \(v>0\) (without loss of generality we assume that \(v>0\) since if \(u+iv\) a solution of (5) then \(u-iv\) is also a solution). We aim to find a contradiction, and hence we prove that no root of (5) with a positive real part can exist. We have:
Let \(\tau _\mathrm{max} v=2k\pi +\gamma \) with \(k \ge 0\) and \(0\le \gamma <2\pi \). We make this change of variable to explore the periodicity of sinus and cosinus functions. Hence, we get
Thus, we have \(v^2 \le \frac{2(k+1)D}{\tau _\mathrm{max}}\); and since \(\tau _\mathrm{max}<\frac{\pi ^2}{2D}\) we get \(v \tau _\mathrm{max}< (k+1)^{\frac{1}{2}}\pi \). But \(v \tau _\mathrm{max}=2k\pi +\gamma \) with \(0 \le \gamma < 2\pi \), which finally yields \(k=0\) and \(v\tau _\mathrm{max}<\pi \).
On the other hand, we have \(u=-\frac{D}{\tau _\mathrm{max}v}\int _{0}^{\tau _\mathrm{max}v}{e^{-\alpha \frac{u}{v}}\mathrm{cos}(\alpha )\hbox {d}\alpha }\). Let us study the sign of the right-hand side of this equation.
If \(\tau _\mathrm{max}v \le \frac{\pi }{2}\) then we obtain \(u<0\), this a contradiction.
If \(\tau _\mathrm{max}v>\frac{\pi }{2}\), then we have
We obtain \(u<0\) which is a contradiction with the initial assumption of \(u > 0\). This proves the sufficient condition.
Now let us prove the necessary condition. A switch from stability to instability or the inverse corresponds to the existence of a root with a zero real part. We have to prove that when \(\tau _\mathrm{max}=\frac{\pi ^2}{2D}\), the stability is lost and cannot be regained. Let \(\lambda \) a solution of (5), and let \(\lambda ^*=iw_0\) a pure imaginary solution of (5) and let \(\tau _\mathrm{max}^*\) the value of \(\tau _\mathrm{max}\) when \(\lambda =\lambda ^*\). We will show that as \(\tau _\mathrm{max}\) increases, the roots of the characteristic equation can cross the imaginary axis only from the left to the right, and therefore once the stability is lost, it cannot be regained. A stability switch occurs when \(\int _{0}^{v\tau _\mathrm{max}^*}{\mathrm{cos}(\alpha )\hbox {d}\alpha }=0\) and \(\int _{0}^{w_0\tau _\mathrm{max}^*}{\mathrm{sin}(\alpha )\hbox {d}\alpha } \not =0\) since zero is not a root of (5); which yields \(w_0\tau _\mathrm{max}^*=(2k+1) \pi \). When \(\tau _\mathrm{max}=\frac{\pi ^2}{2D}\) we check that the above two conditions are satisfied, and hence this value of \(\tau _\mathrm{max}\) corresponds to a pure imaginary root. Moreover, at \(\lambda =iv\) we have:
The derivative of \(\lambda (\tau _\mathrm{max})\) at \(\tau _\mathrm{max}^*\) is positive, which means that roots can cross the imaginary axis only from the left to the right. Since the solution of (3) is asymptotically stable when \(\tau _\mathrm{max}<\frac{\pi ^2}{2D}\), then the stability is lost at \(\tau _\mathrm{max}=\frac{\pi ^2}{2D}\) and cannot be regained (because there is no possibility that all roots have negative real parts when further increasing the value of the delay). This proves the necessary condition.
Appendix B: Proof of Theorem 2
We shall prove that all roots of (6) have negative real parts for any \(\beta >0\). Let \(\alpha \) be a real solution of (6), it is clear that \(\alpha \) cannot be positive. Let \(\lambda =u+iv\) be a complex solution of (6) with \(v>0\) (without loss of generality, we assume that \(v>0\), since if \(u+iv\) a solution of the characteristic equation, then \(u-iv\) is also a solution). We aim to prove that no root with a positive real part can exist. Separating real and imaginary parts in (6), we get:
Furthermore, we have:
with \(I=\int _{0}^{\infty }{e^{-(\beta +u)\frac{z}{v}}\mathrm{cos }(z)dz}\).
Taking into account the above equations, we can write (19) and (20) as follows:
The two above equations yield \(u=\frac{-\beta }{2}<0\). Therefore, the real parts of the roots are always negative, which results in the asymptotic stability.
Appendix C: Proof of Theorem 3
Let \(\lambda =u+iv\) be a solution of the characteristic equation with \(u>0\). We suppose without loss of generality that \(v>0\) (recall that if \(u+iv\) is a solution of the characteristic equation then \(u-iv\) is also a solution). We aim to prove that if \(u>0\), then \(\beta <\beta _c\). By substituting \(\lambda \) with \(u+iv\) and separating real and imaginary parts in the characteristic equation, we get :
with \(\mathrm{cos}(\theta )=\frac{\beta +u}{[(\beta +u)^2+v^2]^\frac{1}{2}}\) and \(\mathrm{sin}(\theta )=\frac{v}{[(\beta +u)^2+v^2]^\frac{1}{2}}\) (with \( 0 \le \theta < 2\pi \)). The above system can be written:
From the equations above, we get:
By virtue of (21), we get \(\mathrm{cos}(\theta )<\mathrm{cos}(\frac{\pi }{2k})\). In addition, we check that \(\frac{\beta }{[\beta ^2+v^2]^{\frac{1}{2}}}<\frac{\beta +u}{[(\beta +u)^2+v^2]^\frac{1}{2}}\) (recall that u is positive); and then we have \(\frac{\beta }{[\beta ^2+v^2]^{\frac{1}{2}}}<\mathrm {cos}(\theta )<\mathrm{cos}(\frac{\pi }{2k})\). From this inequality, we obtain:
On the other hand, we have \(v^2<D^2\frac{{\beta }^{2k}}{[\beta ^2+v^2]^k}\) and \(\frac{\beta ^{2k}}{[\beta ^2+v^2]^k}<\mathrm{cos}^{2k}(\frac{\pi }{2k})\), which yields:
Using (22) and (23), we get \(\beta <\beta _c=D \frac{\mathrm{cos}^{k+1}(\frac{\pi }{2k})}{\mathrm{sin}(\frac{\pi }{2k})}\). Therefore, we conclude that there is no root with a positive real part when \(\beta >\beta _c\), and the local asymptotic stability follows.
Now, let us prove the necessary condition. By implicit differentiation of the characteristic equation, we have:
Therefore, as \(\beta \) decreases, the roots of characteristic equation cross the imaginary axis only from the left to the right and once the stability is lost, it cannot be regained again. The stability is persistently lost at \(\beta ={\beta _c}\).
Appendix D: Proof of Theorem 4
The characteristic equation is given by:
Let \(\lambda =u+iv\) be a root of the characteristic equation with \(v>0\) (without loss of generality). We aim to prove that when \(u>0\), \(\tau >\tau _c\). By separating real and imaginary parts in the characteristic equation, we get:
where \(A=\gamma (p_1(a-b)-c+d)\) and \(B=\gamma (1-p_1)(a-b)\). We have the relation \(A>B\) and \(A+B<0\). First, we study the case when \(A<0\) and \(B<0\). From (24) and (25), we have:
The system above yields \(\tau >\tau _c\). Therefore, we conclude that when \(\tau <\tau _c\), the real parts of the roots of the characteristic equation are negative.
Rights and permissions
About this article
Cite this article
Ben-Khalifa, N., El-Azouzi, R. & Hayel, Y. Discrete and Continuous Distributed Delays in Replicator Dynamics. Dyn Games Appl 8, 713–732 (2018). https://doi.org/10.1007/s13235-017-0225-7
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13235-017-0225-7