participation costs dismiss the advantage of heterogeneous networks in evolution of cooperation
DESCRIPTION
Presentation slides for the following two papers (mainly (1)): (1) Masuda. Proceedings of the Royal Society B: Biological Sciences, 274, 1815-1821 (2007). (2) Masuda and Aihara. Physics Letters A, 313, 55-61 (2003).TRANSCRIPT
Naoki Masuda (University of Tokyo)
Ref: Masuda. Proc. R. Soc. B, 274, 1815-‐1821 (2007).
Also see Masuda & Aihara, Phys. LeK. A, 313, 55-‐61 (2003).
ParNcipaNon costs dismiss the advantage of heterogeneous
networks in evoluNon of cooperaNon
Prisoner’s DilemmaCooperate Defect
Cooperate (3, 3) (0, 5)
Defect (5, 0) (1, 1)
SelfOpponent
unique Nash equilibrium
CC
C C
CD
DD D
D
DC
D C
DD
DD D
D
DD
D D
DD
DD D
D
Mechanisms for cooperaNon• Kin selecNon (Hamilton, 1964)
• Direct reciprocity (Trivers, 1971; Axelrod & Hamilton 1981)
• Iterated Prisoner’s dilemma
• Group selecNon (Wilson, 1975; Traulsen & Nowak, 2006)
• SpaNal reciprocity (Axelrod, 1984; Nowak & May, 1992)
• Indirect reciprocity (Nowak & Sigmund, 1998)
• Image scoring
• Network reciprocity (Lieberman, Hauert & Nowak, 2005; Santos et al., 2005, 2006; Ohtsuki et al., 2006)
• Others (punishment?, voluntary parNcipaNon etc.)
Iterated Prisoner’s Dilemma
• Players randomly interact with others
• Discount factor w (0 ≤ w ≤ 1) to specify the prob. that the next game is played in a round
A’s accumulated payoff = 3 + 5w + 1w2 + 3w3 + 1w4 + 0w5 + …
A plays C D D C D C
B plays C C D C D D
A gets 3 5 1 3 1 0
acNon C D
C (3, 3) (0, 5)
D (5, 0) (1, 1)
• SelecNon based on accumulated payoff aher each round
• Replicator dynamics
• Best-‐response dynamics
• Nice, retalitatory, and forgiving strategies (e.g. Tit-‐for-‐Tat) are generally strong (but not the strongest).
SpaNal Prisoner’s Dilemma(Axelrod, 1984; Nowak & May, 1992)• e.g. square laice
• Either cooperator or defector on each vertex
• Each player plays against all (4 or 8) neighbors.
C:24 C:18 D:24 D:12
C:24 C:21 C:12 D:16
C:24 C:21 C:12 D:16
C:21 C:15 D:20 D:16
• Successful strategies propagate aher one generaNon.
• Result: Cs form “clusters” to resist invasion by Ds.
• Note: 2-‐neighbor CA. More complex than 1-‐neighbor dynamics such as spin (opinion) dynamics and disease dynamics
PD on the WaKs-‐Strogatz small-‐world network(Masuda and Aihara, Phys. LeK. A, 2003)
Note: The degrees of all the nodes are the same
regardless of the rewiring.
acNon C D
C (1, 1) (0, T)
D (T, 0) (0, 0)p = 0 p: small p ≈ 1
0
0.2
0.4
0.6
0.8
1
1 1.5 2 2.5
%C
T
p=0p=0.01p=0.9
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400
%C
generation
T=1.1
0
0.2
0.4
0.6
0.8
1
0 100 200 300 400
%C
generation
T=1.70
0.2
0.4
0.6
0.8
1
0 100 200 300 400
%C
generation
T=3.0
• 1-‐dim ring
• QualitaNvely the same results for 2-‐dim networks
small p
small p
small p
large p large p
large p
Clustering helps cooperaNonmore C more D
Small distance (L) accelerates whatever propagaNon
slow fast
Social dilemma games on scale-‐free networks?
Scale-‐free networks promote cooperaNon
acNon Cooperate Defect
Cooperate (1, 1) (S, T)
Defect (T, S) (0, 0)
Originally by Santos, Pacheco & Lenaerts, PNAS 2006T
S
(a)
0 1 2-1
0
1
T
S
(b)
0 1 2-1
0
1
Regular RG(or the complete graph) scale-‐free
stag hunt
snowdrihno dilemma
PD
Two assumpNons underlie the enhanced coopraNon in SF nets
AssumpNon 1: addiNve payoff scheme
• AddiNve: add the payoffs gained via all the neighbors
• Nowak, Bonhoeffer & May 1994; Abramson & Kuperman 2001; Ebel & Bornholdt, 2002; Ihi, Killingback & Doebeli 2004; Durán & Mulet, 2005; Santos et al., 2005; 2006; Ohtsuki et al., 2006
• Average: divide the summed payoffs by the number of neighbors
• Kim et al., 2002; Holme et al., 2003; Vukov & Szabó, 2005; Taylor & Nowak, 2006
acNon Cooperate Defect
Cooperate (3, 3) (0, 5)
Defect (5, 0) (1, 1)
k = 1001 2v v
10 90
1 k = 22
= C= D
addiNve
payoff scheme
average
3×10+0×90 = 30
(3×10+0×90)/100 = 0.3
5×2 = 10
(5×2)/2 = 5
• Average payoff diminishes cooperaNon in heterogeneous networks (Santos & Pacheco, J. Evol. Biol., 2006).
• Originate from the translaNon invariance of replicator dynamics
AssumpNon 2: PosiNvely biased payoffs
→ Nme rescaling
→ parNcipaNon cost
acNon C D
C (1, 1) (0, T)
D (T, 0) (0, 0)
acNon C D
C (1, 1) (S, T)
D (T, S) (0, 0)
✓a bc d
◆!
✓a� h b� hc� h d� h
◆
✓a bc d
◆!
✓ka kbkc kd
◆
✓a bc d
◆!
✓a+ k b+ kc d
◆
C D
C a b
D c d
!!(⇡C = axC + bxD
⇡D = cxC + dxD
h⇡i = ⇡CxC + ⇡DxD(xC = xC (⇡C � h⇡i)xD = xD (⇡D � h⇡i)
Reason for enhanced cooperaNon
• Hubs earn more than leaves.
• C on hubs (with at least some C neighbors) are stable.
• C spreads from hubs to leaves.
acNon C D
C (1, 1) (0, T)
D (T, 0) (0, 0)
k = 1001 2v v
10 90
1 k = 22
= C= D
Payoff matrix is not invariant on heterogeneous networks
acNon C D
C (1, 1) (0, T)
D (T, 0) (0, 0)
→ Nme rescaling
→ parNcipaNon cost
✓a bc d
◆!
✓a� h b� hc� h d� h
◆
✓a bc d
◆!
✓ka kbkc kd
◆
✓a bc d
◆!
✓a+ k b+ kc d
◆
C D
C a b
D c d
!!(⇡C = axC + bxD
⇡D = cxC + dxD
h⇡i = ⇡CxC + ⇡DxD(xC = xC (⇡C � h⇡i)xD = xD (⇡D � h⇡i)
✔
NG
NG
Our assumpNons• AddiNve payoff scheme
• Introduce the parNcipaNon cost
• Do numerics
• N = 5000 players
• Each player plays against all the neighbors.
• Replicator-‐type update rule: player i copies player j’s strategy with prob
(πj-‐πi)/[max(ki,kj) * (max possible payoff – min possible payoff)]
T
h
(a)
0.7 1 1.3 1.6
0
1
2
3
T
h
(a)
0.7 1 1.3 1.6
0
1
2
3cf
0
0.25
0.5
0.75
1
Simplified prisoner’s dilemma
acNon C D
C (1-h, 1-h) (-h, T-h)D (T-h, -h) (-h, -h)
regular random net
scale-‐free net 3 (roughly separated) regimes
Strong influence of iniNal cnds due to long transients
Somewhat reduced cooperaNon
Enhanced cooperaNon (prev results)
1
2
3
From leaves to hubs. PD payoff structure is most relevant.
T
h
(a)
0.7 1 1.3 1.6
0
1
2
3 1
2
3
Strategy spreads from stubborn leaves to hubs.
From hub cooperators to leaves.
-50
0
50
100
0 50 100 150 200
gene
ratio
n pa
yoff
# neighbors
(a)
0
50
10 100
# fli
ps
# neighbors
(b)
T = 1.5
h = 0h = 0.23h = 0.2
h = 0.24h = 0.25
h = 0.3h = 0.5
h = 0h = 0.2
h = 0.23
h = 0.24h = 0.25
h = 0.3
h = 0.5
General matrix game• Homogeneous (in degree) → 2 parameters (S
and T)
• e.g. well-‐mixed, square laice, regular random graph
• Heterogeneous → 3 parameters (S, T, and h)
• e.g. ER random graph, scale-‐free
• PD, snowdrih game, hawk-‐dove game included
acNon C D
C (1-h, 1-h) (S-h, T-h)D (T-h, S-h) (-h, -h)
T
S
(a)
0 1 2-1
0
1
T
S
(c)
0 1 2-1
0
1
T
S
(e)
0 1 2-1
0
1
T
S
(d)
0 1 2-1
0
1
T
S
(b)
0 1 2-1
0
1
1 2
3
Regular RG (h = 0) SF (h = 0) SF (h = 0.5)
SF (h = 1) SF (h = 2)
stag hunt
snowdrihno dilemma
PD
cf
0
0.25
0.5
0.75
1
1
2
consistent with Santos et al., PNAS (2006)
Thoughts about the payoff bias
• Naturally understood as the parNcipaNon cost
• Payoffs may be negaNve in many pracNcal situaNons.
• Environmental problems?
• InternaNonal relaNons?
• When one is ‘forced’ to play games
Conclusions• Games with parNcipaNon costs on networks
• More C for small parNcipaNon cost h (previous work).
• Networks determine dynamics for small and large h.
• Payoff matrix is most relevant for intermediate h.
• Think twice about the use of simplified PD payoff matrices.