TURING INSTABILITY IN A PUBLIC GOODS GAME
Derick O. Poku
A Thesis Submitted to the University of North Carolina Wilmington in Partial Fulfillment of the Requirements for the Degree of Master of Science
Department of Mathematics and Statistics
University of North Carolina Wilmington
2010
Approved by
Advisory Committee
Nolan Mcmurray Wei Feng
Xiaojie Hou
Chair
Accepted by
Dean, Graduate School TABLE OF CONTENTS
ABSTRACT ...... iii DEDICATION ...... iv ACKNOWLEDGMENTS ...... v LIST OF FIGURES ...... vi 1 INTRODUCTION ...... 1 2 THE STABILITY OF THE EQUILIBRIA OF THE O.D.E SYSTEM 6 2.1 Classification of the Equilibrium Solutions ...... 6 2.2 Equilibria and their Stability ...... 10 3 THE TURING INSTABILITY ...... 15 3.1 Turing Mechanisms ...... 15 3.2 Linearization and Turing Instability of the Reaction-Diffusion model ...... 16 4 NUMERICAL EXAMPLE ...... 25 5 CONCLUSION ...... 30 6 APPENDIX ...... 31 6.1 Derivation of Bump Function B(x) ...... 31 REFERENCES ...... 33
ii ABSTRACT
In economic theory a good that is non-rivaled and non-excludable is referred to as a public good. For the sustainability of these public goods, cooperation among beneficiaries is paramount. Unfortunately, there are bound to be non-cooperators or free-riders and an associated cost for the altruistic behaviour of cooperators. We study a model that represents the interaction of cooperators and free-riders in the consumption of a public good. We determine the conditions under which spatially homogeneous equilibrium solutions to the model without the diffusion terms of cooperators and free-riders are stable to small perturbations. We extend our results to determine the threshold of the diffusion rate of the cooperators above which solutions to the full reaction-diffusion model is unstable (Turing Instability) to non-homogeneous perturbation.The analysis shows that under certain conditions with spatial diffusion, the stable coexistence of the cooperators and the free-riders is impossible in the long run.
iii DEDICATION
Dedicated to my Mum.
iv ACKNOWLEDGMENTS
I would like to thank my advisor Dr. Xiaojie Hou who guided me through this whole process. To him I owe my gratitude.
I would also like to thank Dr. Wei Feng for her help in streamlining the conditions (8) and (9) on page 11 and also thank her the many joyous discussions on population dynamics. Thanks also should be given to Dr. Nolan McMurray for agreeing to be on my thesis committee.
My gratitude also goes to the members and staff of the Department of Mathematics and Statistics, University of North Carolina, Wilmington, for making my period of studies a pleasant one.
I would also like to thank William Reid Peters for translating the sketches I had for some of the figures in this paper into postscript document format.
v LIST OF FIGURES
1 Unstable equilibrium point, trajectories going away from the origin. . 7 2 Stable equilibrium point, trajectories going towards the origin. . . . . 8 3 Saddle equilibrium point,trajectories going towards or away from the origin...... 8 4 a) Real part of both eigenvalues are negative. b) Real part of eigen- values have opposite signs...... 9
5 Intersection of null clines L1 and L2 formed by solving system ( 5 ). . 11 6 Vector fields of (5) under assumptions (6) , (7) , (8) and (9). . . . . 14
2 7 Plot of h(θ ) defined by (27). When the diffusion coefficient d2 of
2 the free-riders increases beyond the critical value dc, h(θ ) becomes negative...... 23 8 The coexistent state is stable to non-homogenous pertubation when the diffusion rate of the free-riders is below the diffusion threshold
dc: (a) initial state of the coexistent solution and (b) the asymptotic behaviour of the coexistent state...... 28 9 The coexistent state is unstable to non-homogenous pertubation when
the diffusion rate of the free-riders is above the diffusion threshold dc: (a) initial state of the coexistent solution and (b) the asymptotic behaviour of the coexistent state...... 29 10 The graph of c(t)...... 31 11 The graph of f(t)...... 32 12 The graph of g(x)...... 32 13 The graph of h(x)...... 33 14 The graph of m(x) ...... 33 15 The graph of B(x) ...... 33
vi 1 INTRODUCTION
In biology Turing instability is the presage of pattern formulation. Alan Turing (1952) postulated that chemicals called morphogens through reaction and diffusion could create spatial patterns in chemical concentrations through an instability pro- cess. Turing’s theory states that diffusion which is generally seen as a stabilizing process could destabilize an otherwise stable reaction-diffusion system and trigger a pattern formation. Spontaneous pattern formation has been observed in chemical and physical systems ranging from hydrodynamical phenomena like ripples in the sand and streaks of colours in animal furs [17]. Research on Turing instability in the context of competition for a public good is very new. This paper is devoted to the study of pattern formation in a social dilemma modelled as a game which represents the competition for resources.
The front page of a newspaper today may contain a report of violent attack by a group of people on a foreign soil, a domestic political scandal or a geopolitical issue of trade tariffs between an entrenched Superpower and a fledgling Economy. The inside pages may contain agitations by various social groups to sway the course of government decisions on specific policies. The business section may be full of deals relating to merger and acquisitions companies. The aforementioned scenarios have a common denominator which is a conflict of interest between groups of people such as governments, businesses and social networks . The theoretical models that represents these conflicts are termed games [6].
Mathematical theory of games termed Game theory was invented by John von Neumann and Oskar Morgenstern (1944). Game theory is the study of the ways in which strategic interactions among rational players produce outcomes with respect to the preferences of those players, none of which might have been intended by any
1 of them [16]. The interaction could be viewed from the angle of non-cooperation and cooperation amongst players. The development of “Prisoner’s Dilemma” [14] and John Nash’s papers on the definition and existence of equilibrium laid the foun- dations for modern non-cooperative game theory. Simultaneously, cooperative game theory reached important results in papers by Nash and Shapley on bargaining games [13].
An important question that arises in a game of conflict is the effect of cooperation among participants. Suppose a game represents the production/conservation and consumption of a public good. Countries have the opportunity to invest a certain amount of their Gross Domestic Product (GDP) to sustain the good. A return on investment would be an additional increase in GDP for the next fiscal year by the average of twice the sum of all contributions by investors. John Stuart Mill a 19th century political economist stipulated that the rational behaviour would be that individual countries in their selfish act to maximize utility would not make a contribution [3]. Evidently a non-contributor (“free rider”) tends to gain if at least one country makes a contribution. But the most prudent strategy is for each to make a contribution. The net return on investment is maximized for all.
Whereas some participants in a game similar to the scenario described earlier may realise the benefits of cooperation and act irrationally, others may act otherwise. We note that the cooperators (investors) may incur a cost for their altruistic behaviour. As an example of such a cost, suppose only 1 out of N (where N > 2) countries make a contribution of P of her GDP then the return on investment will be
2P N
2 and the cost of altruism incurred is
P (N − 2) N
Another twist to this game is the ability of players to migrate to other areas of a spatial domain to exploit a public good. The dynamics of this game raises some questions that needs to be addressed:
• Could the cooperators and “free riders” coexist in this game?
• Under what conditions will such coexistence occur?
• What is the effect of the mobility of the participants on the coexistence state?
• What is the effect of mobility on the spatial domain in which the game is played?
• What role does the cost of altruistic behaviour play in this game?
To answer these questions we examine the following public good game modelled as a reaction diffusion system:
∂u ∂2u u + av = d + u(1 − − α), 1 2 ∂t ∂x k(u) 2 ∂v ∂ v bu + v + = d2 + v(1 − ), (x, t) ∈ R × R , (1) ∂t ∂x2 k(u) u(x, 0) = u0(x), v(x, 0) = v0(x) where u = u(x, t), v = u(x, t) are two participating groups in the game and u0(x),v0(x) are nonnegative bounded smooth functions in R and a, b, α, d1, d2 are positive constants. The system is a continuous spatial-temporal version of a public
3 goods game [17] derived from the Lotka-Volterra competition model and describes the interaction between the group of players u(x, t) and v(x, t). The cooperators u employ an altruistic strategy in order to ensure sustenance of the public good; whiles the strategy of v (“free riders”) is to exploit the public good without investing. The function k(u) = 1 + ku (where k > 0) represents the investment contributed by the cooperators u and shared with the “free riders” v. The cost of altruism for the coop- erators is measured by α where 0 < α < 1 [2]. The effects of v on u is represented by the positive constant a and vice versa the constant b. The constants d1 > 0, d2 > 0 are respectively the diffusion rates (or mobility) of cooperators u and “free riders” v.
We assume zero flux boundary conditions for our system and restrict our domain to a one-dimensional finite interval where x ∈ R and 0 < x < L. The zero flux condition guarantees that there are no external input to our model. The model becomes self-organizing and all inferences made could be attributed to the domain under consideration [10]. An example of zero flux condition is when no new player can join nor leave the domain in which the game is played.
In Chapter 2, we determine the equilibrium states of the reaction diffusion model and conditions for asymptotic stability of these states without the diffusion coefficients. It will be shown that the homogeneous equilibrium state of coexistence is stable to small perturbations when some restrictions are imposed on the relationships between the positive constants a, b, k and α of our model.
In Chapter 3, we derive the necessary and sufficient conditions under which diffusion driven instability or Turing Instability may occur.
We present an example of our system that undergoes a Turing Instability in Chapter
4 4. The numerical example will show how our model goes from a homogeneous steady state to a Turing bifurcation.
Chapter 5 is a summary of this work and future direction and Chapter 6 is devoted to supporting computations in this paper.
5 2 THE STABILITY OF THE EQUILIBRIA OF THE O.D.E SYSTEM
In this chapter we study the stability of the coexistence equilibrium. First, the conditions for the coexistence equilibrium state is derived. We assume that the density of u and v are only time dependent and the basic global in time solutions to system (1) exist. Secondly, we linearise the vector field of (1) and further derive the stability of the co-existence equilibrium state. The existence and stability of the coexistent state shows that cooperators and “free riders” will sustain their winning strategy with the evolution of time.
2.1 Classification of the Equilibrium Solutions
We recall from Ordinary Differential Equations the classification of equilibrium solu- tions as stable, asymptotically stable or unstable by considering the following system
∂x = a x + a y + P (x, y), ∂t 11 12 (2) ∂y = a21x + a22y + Q(x, y) ∂t where |P (x, y)|, |Q(x, y)| ≤ β(|x|2 + |y|2) and β is some positive constant. The term (|x|2 + |y|2) is assumed to be very small and the origin, (0, 0), is an equilibrium solution1 of system (2).
To perform the classifications we state some definitions by considering the differen- tial equation
1 ∂u ∂v Let the system of ordinary differential equations, ∂t = f(u, v), ∂t = g(u, v) be defined in some n region U in R .Then a point (u0, v0) is called an equilibrium point of the system if f(u0, v0) = g(u0, v0) = 0 .
6 ∂x = f(x, t), x ∈ n. ∂t R
Definition 1 An equilibrium point x0 is stable iff for all > 0 there exist δ > 0 such that for a homogeneous pertubation p0 where |x − p0| < δ then |x0 − p0| < .
Definition 2 An equilibrium point x0 is asymptotically stable iff it is both stable and for a homogeneous pertubation p0 , |x0 − p0| → 0 as t → ∞.
We use the coefficients of system (2) to define matrix A
a a 11 12 A = (3) a21 a22
with the following eigenvalues λ1, λ2. Depending on the signs of the λ1 and λ2, we have the following classifications:
Case 1: 0 < λ1 < λ2
If 0 < λ1 < λ2 , then the equilibrium point (0, 0) is unstable. The local topological structure in the neighbourhood of (0, 0) is illustrated in figure (Fig. 1).
<1, λ > 1 eigenvectors y <1, > } λ2 <1, λ2>
<1, λ1>
0 x
Figure 1: Unstable equilibrium point, trajectories going away from the origin.
7 Case 2: λ1 < λ2 < 0
If λ1 < λ2 < 0 , then the equilibrium point (0, 0) is asymptotically stable. See figure (Fig. 2).
<1, λ > 1 eigenvectors y <1, > } λ2 <1, λ2>
<1, λ1>
0 x
Figure 2: Stable equilibrium point, trajectories going towards the origin.
Case 3: λ1 < 0 < λ2
If λ1 < 0 < λ2 , then the equilibrium (0, 0) is a saddle point. A saddle point is always unstable. See figure (Fig. 3).
y y
0 0 x x
a) b)
Figure 3: Saddle equilibrium point,trajectories going towards or away from the origin.
Case 4: λ1 and λ2 are two conjugate eigenvalues
If the real part of both eigenvalues denoted Re(λ1), Re(λ2) are negative then the equilibrium (0, 0) is an asymptotically stable point. Unstable otherwise. See figure
8 (Fig. 4) for the spiral trajectories.
y y
0 x 0 x
a) b)
Figure 4: a) Real part of both eigenvalues are negative. b) Real part of eigenvalues have opposite signs.
Define matrix J as ∂f ∂f (u0, v0) (u0, v0) ∂u ∂v J = ∂g ∂g (u , v ) (u , v ) ∂u 0 0 ∂v 0 0 where the ijth entry is the partial derivative of function f(u, v) or g(u, v) evaluated at an equilibrium point (u0, v0). The real parts Re(λ1), Re(λ2) of the eigenvalues of matrix J is negative (ie. (u0, v0) is stable) if J satisfies the following theorem.
Theorem 1 An equilibrium point (u0, v0) is stable if both conditions a) and b) si- multaneously hold
a) trace(J) = ∂f + ∂g < 0 ∂u u=u0,v=v0 ∂u u=u0,v=v0
b) det(J) = ∂f ∂g − ∂f ∂g > 0 ∂u ∂v u=u0,v=v0 ∂v ∂u u=u0,v=v0
Proof: see ([7]).
9 2.2 Equilibria and their Stability
Definition 3 Let the system of ordinary differential equations,
∂u = f(u, v), ∂t ∂v = g(u, v) ∂t
n be defined in some region U in R .Then a point (u0, v0) is called an equilibrium point of the system if f(u0, v0) = g(u0, v0) = 0 [11].
Consider system (1) with the diffusion coefficients set to zero,
∂u u + av = u 1 − − α , ∂t k(u) ∂v bu + v = v 1 − , (4) ∂t k(u) u(x, 0) = u0(x), v(x, 0) = v0(x)
The equilibrium states are determined by
. u+av f(u, v) = u 1 − k(u) − α = 0, (5) . bu+v g(u, v) = v 1 − k(u) = 0
We solve (5) to get the following equilibria: 1 − α (0, 0) , (0, 1) , , 0 and 1 − k + αk α + a − 1 b − αb − 1 , ab + k − ak − αk − 1 ab + k − ak − αk − 1 which are illustrated in figure (5).
The point (0, 0) which corresponds to the origin of (5) is when no participants are in the game, (0, 1) is the state of the game that is purely made up of “free riders” v
10 1−α and have reached their saturation and ( 1−k+αk , 0) is when the participants are solely α+a−1 b−αb−1 cooperators u. The last scenario, ( ab+k−ak−αk−1 , ab+k−ak−αk−1 ), is the coexistence between u and v.
α + a – 1 u0 = ab + k – ak – αk – 1 v b – αb – 1 v0 = ab + k – ak – αk – 1 1 – α a
1
(u0 , v0 )
0 1 – α 1 u 1 – k + αk b – k L1 L2
Figure 5: Intersection of null clines L1 and L2 formed by solving system ( 5 ).
Throughout the rest of this paper, we make the following assumptions:
b > k (6)
1 − α 1 1 < , or b < (7) 1 − k + αk b − k 1 − α 1 − α > 1, or a < 1 − α (8) a
ab < 1 (9)
Next, we analyse the stability of the equilibria corresponding to the notions of section (2.1). Consider matrix J where the ijth entry is the partial derivative of function f(u, v) or g(u, v) of system (5),
2u+av+ku2 au 1 − 2 − α − J = (1+ku) 1+ku . v(b−kv) 1+ku−bu−2v − (1+ku)2 1+ku
11 To determine the stabilities of the four equilibrium points found in section (2.1), we apply theorem (2) to derive the following propositions.
Proposition 1 Under the conditions (6) , (7) , (8) and (9), the equilibrium point (0, 0) is unstable.
Proof 1 − α 0 J(0, 0) = 0 1 trace(J(0, 0)) = 2 − α > 0 since by the earlier assumption 0 < α < 1 . Theorem (2) implies that (u0, v0) = (0, 0) is an unstable equilibrium point.
Proposition 2 Assuming the conditions of proposition (1), the equilibrium point (0, 1) is unstable
Proof 1 − a − α 0 J(0, 1) = . k − b −1
The determinant det(J(0, 1)) = a + α − 1 < 0, since by equation (8), a < 1 − α .
Evidently, theorem (2) makes (0, 1) an unstable equilibrium point.
Proposition 3 Assuming the conditions of proposition (1), the equilibrium point