Proceedings of the 19th World Congress

The International Federation of Automatic Control

Cape Town, South Africa. August 24-29, 2014

On Networked Evolutionary Games

Part 1: Formulation ?
Hongsheng Qi ∗ Daizhan Cheng ∗ Hairong Dong ∗∗
∗
Key Laboratory of Systems and Control, Academy of Mathematics
and Systems Science, Chinese Academy of Sciences, Beijing 100190,
P.R.China (e-mail: [email protected], [email protected])
∗∗
Beijing Jiaotong University, Beijing 100044, P.R.China (e-mail:
[email protected])

Abstract: This paper presents a comprehensive modeling technique for networked evolutionary
games (NEG). Three kinds of network graphs are considered, which are (i) undirected graph
for symmetric games; (ii) directed graph for asymmetric games, and (iii) d-directed graph
for symmetric games with partial neighborhood information. Three kinds of fundamental
evolutionary games (FEGs) are discussed, which are (i) two strategies and symmetric (S-2);
(ii) two strategies and asymmetric (A-2); and (iii) three strategies and symmetric (S-3). Three
strategy updating rules (SUR) are introduced, which are (i) Unconditional Imitation (UI);
(ii) Fermi Rule (FR); (iii) Myopic Best Response Adjustment Rule (MBRA). Then we review
the fundamental evolutionary equation (FEE), and give the detailed formulation for different
models. Finally, the network profile dynamics (NPD) of NEGs are investigated via their FEE.

Keywords: Networked evolutionary game, fundamental evolutionary equation, network profile

dynamics, semi-tensor product of matrices

1. INTRODUCTION s (Laschov & Margaliot, 2012; Li & Sun, 2011a; Zhang

& Zhang, 2013); (iii) optimal control and game related
In the last four decades or so, the investigation of evolu- optimization (Laschov & Margaliot, 2012); (iv) network
tionary games (EG) has attracted a great attention from stability and stabilization (Li et al., 2013b); (v) tech-
scientists in cross disciplines, because evolutionary game nique for reducing complexity (Zhao et al., 2013); and
has wide background from biological systems (Taylor & (vi) various applications to control and signal processing
Jonker, 1978; Charnov, 1982), economical systems (Sug- etc. (Wang et al., 2012; Xu & Hong, 2013), just to quote
den, 1986), social systems (Ohtsuki et al., 2006), physical a few.
systems (Nowak & May, 1992), etc. In a very recent work, the STP has also been used to the
In recent researches, the topological relationship among modeling, analysis and control design of the NEGs (Cheng
players of an EG is mostly ignored. That is, assume each et al., Preprint2013). This paper is a development of Cheng
player gambles with all others. In many practical cases et al. (Preprint2013). It provided a comprehensive discus-
the situation is not like this. Therefore, in recent years sion for various NEGs. The NEGs discussed could have
the networked EG (NEG) becomes a hot topic. Roughly three different graphs (i) undirected graph, which is used
speaking, an NEG adds a graph with players as its nodes for the NEGs with symmetric fundamental network games
and sides describing the neighborhoods of each players. (FNG); (ii) directed graph, which is used for the NEGs
Then each player only gambles with its neighbors (Nowak with asymmetric FNGs; and (iii) d-directed graph, which
& May, 1992; Szabo & Toke, 1998; Santos et al., 2008). is used for symmetric games with partial neighborhood
Since there are no many proper tools to deal with NEG, information. Three kinds of FNGs are discussed, which
most of the researches are based on either simulations or are (i) each player has two strategies and the game is
statistics. symmetric (S-2); (ii) each player has two strategies and
the game is asymmetric (A-2); and (iii) each player has
Recently, the semi-tensor product (STP) has been pro- three strategies and the game is symmetric (S-3). Three
posed for investigating (Boolean and k-valued logical) strategy updating rules (SUR) are introduced, which are
networks and network-based games (Cheng et al., 2011, (i) Unconditional Imitation (UI); (ii) Fermi Rule (FR);
2012a). There are many other interesting developments (iii) Myopic Best Response Adjustment Rule (MBRA).
such as (i) topological structure of notworks (Fornasini & Though most of widely discussed kinds of NEGs will be
Valcher, 2013b; Hochma et al., 2013); (ii) controllability discussed in detail, the technique developed is applicable
and control design of various kinds of control network- for other cases.
? This work was supported in part by National Natural Science
Foundation (NNSF) of China under Grants 61074114, 61273013,
61104065, 61374168, and 61333001.

978-3-902823-62-5/2014 © IFAC 275

19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Then we review the fundamental evolutionary equation Proposition 3. Let X ∈ Rm be a column and M
(FEE) introduced in Cheng et al. (Preprint2013) and is a matrix. Then
construct the FEEs for various types of NEGs. X n M = (Im ⊗ M ) X. (2)
For statement ease, some notations and basic concepts are (iii) Swap matrix (Cheng et al., 2011, 2012a):
introduced first. Definition 4. A matrix W[m,n] ∈ Mm×n , defined
• Notations: by
(i) Mm×n : the set of m × n real matrices. W[m,n] = δmn [1, m + 1, · · · , (n − 1)m + 1;
(ii) Col(M ) (Row(M )) is the set of columns (rows) of 2, m + 2, · · · , (n − 1)m + 2;
M . Coli (M ) (Rowi (M )) is the i-th column (row) (3)
··· ;
of M .
(ii) Dk := {1, 2, · · · , k} , k ≥ 2. n, m + n, · · · , nm],
(iii) δni : thei-th column of the identity matrix In . is called the (m, n)-dimensional swap matrix.
(iv) ∆n := δni |i = 1, · · · , n . The basic function of the swap matrix is to
 k

P “swap” two vectors. That is,
(v) Υk = (r1 , · · · , rk ) ri ≥ 0, i = 1, · · · , k; ri = 1 is
i=1 Proposition 5. Let X ∈ Rm and Y ∈ Rn be two
called the set of k-th dimensional probabilistic columns. Then
vectors. W[m,n] n X n Y = Y n X. (4)
(vi) A matrix L ∈ Mm×n is called a logical matrix if
the set of columns of L, denoted by Col(L), are The rest of this paper is organized as follows: Section 2
k presents a mathematical framework for NEGs. Three basic
of the form of δm . That is,
Col(L) ⊂ ∆m . components of an NEG, namely, network graph, FNG, and
SUR, are discussed in detail. Section 3 is devoted to the
Denote by Lm×n the set of m×n logical matrices. FEE, which plays a key role in the investigation of NEGs.
(vii) If L ∈ Ln×r , by definition it can be expressed as FEEs of all players are building block for constructing
L = [δni1 , δni2 , · · · , δnir ]. For the sake of brevity, it strategy profile dynamics of the overall networks. Section
is briefly denoted as 4 is a brief conclusion.
L = δn [i1 , i2 , · · · , ir ].
2. MATHEMATICAL FRAMEWORK FOR NEG
(viii) A matrix L ∈ Mm×n is called a probabilistic
matrix if the columns of L are m-dimensional
probabilistic vectors. That is, This section is a comprehensive description of mathemati-
cal framework of NEGs. The main idea of which was firstly
Col(L) ⊂ Υm . proposed in Cheng et al. (Preprint2013).
The set of m×n probabilistic matrices is denoted Definition 6. A networked evolutionary game, denoted by
by Υm×n . ((N, E), G, Π), consists of three ingredients as:
(ix) If L ∈ Υm×n , if Col(L) = C1 ∪ C2 , where
C1 ⊂ ∆m and C2 ⊂ Υm \∆m , and |C2 |  |C1 |. (i) a network (graph) (N, E);
Then for notational compactness, we still use the (ii) a fundamental network game (FNG), G, such that if
shorthand (i, j) ∈ E, then i and j play the FNG with strategies
xi (t) and xj (t) respectively.
L = δm [i1 , i2 , · · · , in ], (iii) a local information based strategy updating rule
s (SUR).
where if Colk (L) = δm ∈ C1 , ik = s, else if
Colk (L) ∈ C2 , that is, Colk (L) = (r1 , · · · , rm ),
we express ik as In the following we describe these three ingredients one by
one.
ik = 1/(r1 ) + 2/(r2 ) + · · · + m/(rm ).
• Operators: 2.1 Network Graph
(i) Semi-tensor product of matrices (Cheng et al.,
2011, 2012a): We consider three kinds of network graphs.
Definition 1. Let M ∈ Mm×n and N ∈ Mp×q , (i) Undirected graph: N = {1, 2, · · · , n} (n ≤ ∞).
and t = lcm{n, p} be the least common multiple It represents n players. If (i, j) ∈ E, then i is
of n and p. The semi-tensor product of M and in the neighborhood of j, denoted by i ∈ U (j).
N , denoted by M n N , is defined as Simultaneously, j ∈ U (i).



M ⊗ It/n N ⊗ It/p ∈ Mmt/n×qt/p , (1) (ii) Directed graph: Note that the FNG is always played
where ⊗ is the Kronecker product. by two neighboring players. If the FNG is not sym-
metric, the directed edge is used to distinguish differ-
(ii) Khatri-Rao Product of matrices (Ljung & Söderström, ent roles of two players. Assume (i, j) ∈ E, i.e., there
1982) is an edge from i to j, then in the game i is player 1
Definition 2. Let M ∈ Mp×m , N ∈ Mq×m . and j is player 2. Note that such directed graph does
Then the Khatri-Rao Product is defined as not affect the definition of neighborhoods.
M ∗ N = [ Col1 (M ) n Col1 (N ) · · · (iii) D-directed graph: Assume the FNG is still symmetric,
Colm (M ) n Colm (N )] ∈ Mpq×m . but the graph is not symmetric. That is, if (i, j) ∈ E,
denoted by dot line arrow goes from i to j, it means

276
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

the information can go from i to j but not the other 1

direction. In this case there exist V (i) ⊂ U (i), such
that player i can use only part of its neighborhood 1
information, precisely, only the information from V (i)
can be used. 2
Definition 7. Consider an NEG with graph (N, E).
5 2
(i) If (i, j) ∈ E, then both i ∈ U (j) and j ∈ U (i) (no 5
matter whether the graph is directed or not). 3
(ii) If there exist α1 , · · · , αλ , such that i ∈ U (α1 ), α1 ∈
U (α2 ), · · · , αs ∈ U (j), where λ < s, then i is said to
be in the s-neighborhood of j, denoted by i ∈ Us (j). 4 3 4
Note that (i) if i ∈ Us (j), then j ∈ Us (i); (ii) if i ∈ Us (j), (a) (b)
then i ∈ Uh (j), h > s; (iii) it is assumed that i ∈ U (i).
Definition 8. A graph is said to be homogeneous if the 1
graph is undirected and each node has same degree, or the
graph is directed and each node has same in-degree and
same out-degree. If a graph is not homogeneous it is said
to be heterogeneous.
5 2
The following example shows different kinds of network
graphs.
Example 9. Assume there are 5 players. They form three
kinds of graphs as shown in Fig. 1.
(i) A cycle of undirected network graph shown in 4 3
Fig. 1 (a).
Consider the neighborhoods of 1: (c)
U (1) = {5, 1, 2}; U2 (1) = {4, 5, 1, 2, 3}.
(ii) A directed network graph shown in Fig. 1 (b).
Consider the neighborhoods of 1: Fig. 1. Three kinds of network graphs
U (1) = {1, 2}; U2 (1) = {5, 1, 2, 3}; In the following example we collect some commonly used
U3 (1) = {4, 5, 1, 2, 3}. FNGs. More details about the practical meanings can
(iii) D-directed network graph Fig. 1 (c). be found in Rasmusen (2007); Smith (1982); Benoit &
Consider the neighborhoods of 1: Krishna (1985).
U (1) = {1, 2, 3, 4, 5}; V (1) = {1, 3, 4}. Example 11. We consider three simplest kinds of FNGs as
follows.
2.2 Fundamental Network Game • S-2: The payoff bi-matrix of this kind of games is in
Table 1.
Definition 10. A fundamental network game (FNG) is a
game with two players, i.e., N = (i, j), and each player Table 1. S-2 Games
has the same set of strategies as P1 \P2 1 2
S = Si = Sj = (1, 2, · · · , k). 1 (R, R) (S, T )
2 (T, S) (P, P )
It is symmetric if the payoff functions satisfy:
ci (sp , sq ) = cj (sq , sp ), ∀sp , sq ∈ S.
It covers many well known games. For instance,
Otherwise it is asymmetric. (1) if 2R > T + S > 2P , it is the Game of Prisoner’s
Dilemma;
In asymmetric case, only the directed graph can be used
(ii) if R = b − c, S = b − c, T = b, P = 0, and
as the network graph.
2b > c > b > 0, it is the Snowdrift Game;
The overall payoff of player i is assumed to be the average (iii) if R = 12 (v − c), S = v, T = 0, P = v2 , and v < c,
of its payoffs with all neighbors. Precisely, it is the Hawk-Dove Game.
1 X • A-2: The payoff bi-matrix of this kind of games is in
ci (t) = cij (t), i ∈ N. (5) Table 2.
|U (i)| − 1
j∈U (i)\i
Table 2. A-2 Games
An FNG is determined by two key factors: (i) k: the P1 \P2 1 2
number of possible strategies; (ii) type: symmetric or 1 (A, B) (C, D)
asymmetric. So we classify the FNGs as: 2 (E, F ) (G, H)
• S-k: a symmetric game with k possible strategies;
• A-k: an asymmetric game with k possible strategies; It also covers many well known games. For instance,

277
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

(1) if A = H = a, B = G = b, C = D = E = F = 0, • Fermi Rule (FM) (Szabo & Toke, 1998; Traulsen

and a > b > 0, it is the Battle of The Sexes; et al., 2006). That is, randomly choose a neighbor-
(ii) if E > A > C = D > B > 0 > F , and hood j ∈ U (i). Comparing cj (t) with ci (t) to deter-
G = H = 0, it is the Game of Boxed Pigs; mine xi (t + 1) as
(iii) if A = b, B = −b, C = b, D = −b E = c, 
xj (t), with probability pt ,
F = −c, G = a, H = −a, and a > b > c > 0, it xi (t + 1) = (12)
is the Game of Battle of the Bismark See, xi (t), with probability 1 − pt ,
(iv) if A = D = F = G−a, B = C = E = H = a, and where pt is decided by the Fermi function
a 6= 0, it is the Game of Matching the Pennies. 1
• S-3: The payoff bi-matrix of this kind of games is in pt = .
1 + exp(−ζ(cj (t) − ci (t)))
Table 3.
The parameter ζ > 0 can be chosen arbitrarily. For
Table 3. S-3 Games simplicity, throughout this paper we set ζ = ∞. Then
P1 \P2 1 2 3

1, cj (t) > ci (t),
1 (A, A) (B, C) (D, E) pt =
2 (C, B) (F, F ) (G, H) 0, cj (t) ≤ ci (t).
3 (E, D) (H, G) (I, I) This method leads to a probabilistic k-valued logical
dynamics.
Some examples are • Myopic Best Response Adjustment Rule (MBRA)
(1) if A = F = I = 0, B = E = G = a, (Young, 1993): Assume
C = D = H = −a, and a 6= 0, it is the Game of ci (xi = x∗ , xj = xj (t), j ∈ U (i)\{i})
Rock-Scissor-Paper;
(ii) if E = a, A = b, F = c, I = 0, B = G = H = = max ci (xi = x, xj = xj (t), j ∈ U (i)\{i}) , (13)
x∈S
D = d, C = e, and a > b > c > 0 > d > c, it is then we choose
the Benoit-Krishna Game.
xi (t + 1) = x∗ . (14)
2.3 Strategy Updating Rule When the strategies with best payoff are not u-
nique, say, the set of best strategies is
Denote by xi (t) the strategy of player i at time t. Then S ∗ = {x∗1 , · · · , x∗r } ⊂ S, (15)
SUR is a rule which uses the local information to decide
its next strategy. Precisely, we may use the following 2 options:

(i) First MBRA (MBRA-1): Choose one correspond-
xi (t + 1) = fi {xj (t), cj (t) j ∈ U (i)} , t ≥ 0, i ∈ N. ing to a priority. For instance (as a default),
(6)
xi (t + 1) = min{x∗j ∈ S ∗ }. (16)
There are some commonly used SURs. This method leads to a deterministic k-valued
• Unconditional Imitation (UI) (Nowak & May, 1992): logical dynamics.
The strategy of player i at time t + 1, i.e., xi (t + 1), (ii) Second MBRA (MBRA-2): Choose one with e-
is selected as the best strategy from strategies of qual probability for best strategies. That is,
neighborhood players j ∈ U (i) at time t. Precisely, 1
xi (t + 1) = x∗j (t), with probability piµ = ,
if r
µ = 1, · · · , r.
j ∗ = argmaxj∈U (i) cj (x(t)), (7)
(17)
then
This method leads to a probabilistic k-valued
xi (t + 1) = xj ∗ (t). (8) logical dynamics.
When the players with best payoff are not unique,
say 3. FUNDAMENTAL EVOLUTIONARY EQUATION
argmaxj∈U (i) cj (x(t)) := {j1∗ , · · · , jr∗ }, (9)
we may use the following 2 options: Observing equation (6), since cj (t) depends on U (j) and
(i) First Unconditional Imitation (UI-1): Choose one U (j) ⊂ U2 (i), (6) can be rewritten as
corresponding to a priority. For instance (as a


xi (t + 1) = fi {xj (t) j ∈ U2 (i)} , t ≥ 0, i = 1, 2, · · · , n.
default), (18)
j ∗ = min{µ | µ ∈ argmaxj∈U (i) cj (x(t))}. (10) We call (18) the fundamental evolutionary equation (FEE).
This method leads to a deterministic k-valued One sees easily that the overall network evolutionary dy-
logical dynamics. namics, called the network profile dynamics (NPD), is
(ii) Second Unconditional Imitation (UI-2): Choose completely determined by the FEEs. It is also obvious
any one with equal probability. That is, that the FEEs are determined by network graph, FNG,
1 and SUR. In the following we use some examples to depict
xi (t + 1) = xjµ∗ (t), with probability piµ = , the procedure for constructing FEEs.
r
µ = 1, · · · , r. Example 12. Consider an NEG. Assume the network
(11) graph is Fig.1 (a); FNG is S-2 with R = S = −1, T = 2,
This method leads to a probabilistic k-valued P = 0 (the game of Snowdrift). SUR is UI-1. Then the
logical dynamics. FEE can be determined via Table 4.

278
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

Table 4. Payoffs → Dynamics (Example 12) Table 5. Payoffs → Dynamics (Example 13)
Profile 11111 11112 11121 11122 11211 11212 11221 11222 Profile 11111 11112 11121 11122 11211 11212 11221 11222
ci−1 -1 -1 -1 -1 -1 -1 -1 -1 c1 2 2 2 2 2 2 2 2
1 1
ci -1 -1 -1 -1 2 2 1 1 c2 5/3 1 5/3 1 1 3
1 3
ci+1 -1 -1 2 1 -1 -1 1 0 1 1 1 1
c3 3/2 1 2 2
0 0 2 2
fi 1 1 2 2 2 2 2 2 1 1
c4 1 2
0 1 2
0 1 2
.. 1 1 1 1
c5 3/2 0 2 2
3/2 0 2 2
. f1 1 1 1 1 1 1 1 1
Profile 22111 22112 22121 22122 22211 22212 22221 22222 f2 1 1 1 1 1 1 1 ( 31 , 32 )
ci−1 1 1 1 1 0 0 0 0 f3 1 1 1 1 1
(2,2) 1 1 1
(2,2) 1 1
(2,2) 2
ci -1 -1 -1 -1 1 1 0 0 f4 1 1 1 2 1 1 2 2
ci+1 -1 -1 2 1 -1 -1 1 0 f5 1 1 1 ( 12 , 12 ) 1 ( 21 , 12 ) 1 2
fi 2 2 2 2 2 2 2 2
..
.
Now identify 1 ∼ δ21 and 2 ∼ δ22 , then we can express f Profile 22111 22112 22121 22122 22211 22212 22221 22222
into its algebraic form as c1 1
2
1 1
2
1 1 1 1 1
c2 3
1 3
1 1 4/3 1 4/3
xi (t + 1) = Mf xi−2 xi−1 xi xi+1 xi+2 , i = 1, 2, 3, 4, 5, c3 1 1 0 0 1 1 3/2 3/2
(19) c4 1 1
2
0 1 1
2
0 1 2
c5 1 1 0 3/2 1 1 0 3/2
where f1 2 2 2 2 2 2 2 2
Mf = δ2 [1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 f2 ( 23 , 13 ) 2 2 2 2 2 2 2
(20) ( 12 , 12 )
1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]. f3 1 1 2 2 2 2 2
f4 1 ( 12 , 12 ) 2 2 ( 21 , 12 ) 2 2 2
Note that we use (mod 5) notation. That is, x0 = x5 , f5 1 2 ( 12 , 12 ) 2 1 2 2 2
x−1 = x4 and so on.
Example 13. Assume network graph is Fig. 1 (b); FNG is
A-2 with A = H = 2, B = G = 1, E = F = C = D = 0 M3 = δ2 [1, 1, 1, ( 21 , 12 ), 1, ( 12 , 21 ), ( 12 , 12 ), 2, 1, 1, 1, 2, 2, 2, 2, 2
(the game of Battle of The Sexes). SUR is Fermi Rule 1, 1, 1, ( 12 , 12 ), 1, 2, ( 12 , 21 ), 2, 1, 1, ( 12 , 12 ), 2, 2, 2, 2, 2],
(FM). Since the game is not symmetric, the FEEs for (24)
individual nodes are different. We need to work out them
one by one.
M4 = δ2 [1, 1, 1, 2, 1, 1, 2, 2, 1, ( 21 , 21 ), 2, 2, ( 12 , 12 ), 2, 2, 2
xi (t + 1) = Mi x1 x2 x3 x4 x5 := Mi x, i = 1, 2, 3, 4, 5,
(21) 1, 1, 1, 2, 1, 1, 2, 2, 1, ( 12 , 21 ), 2, 2, ( 12 , 12 ), 2, 2, 2],
5
(25)
where x = ni=1 xi and Mi will be calculated in the
following two steps:
M5 = δ2 [1, 1, 1, ( 12 , 12 ), 1, ( 12 , 21 ), 1, 2, 1, 2, 1, 2, 1, 2, 2, 2
(i) If the profile is known, then the payment for each 1, 1, 1, ( 21 , 12 ), 1, 2, ( 12 , 21 ), 2, 1, 2, ( 12 , 12 ), 2, 1, 2, 2, 2].
player is known. For instance, if the profile
(26)
(x1 , x2 , x3 , x4 , x5 ) = (1 1 2 2 2),
Then it is easy to calculate that Example 14. Assume network graph is Fig. 1 (c); FNG
is S-3 with A = F = I = 0, B = E = G = 1, C =
c1 = 2,
D = H = −1 (the Game of Rock-Scissor-Paper), SUR is
1 1 MBRA (It is easy to check that for this example MBRA-1
c2 = (1 + 0 + 0) = ,
3 3 and MBRA-2 lead to the same FEE). We emphasize only
1 1 one thing: based on the definition, the neighborhood for a
c3 = (0 + 1) = ,
2 2 player to play with maybe different from the neighborhood
1 from which he can get of information. For instance, player
c4 = (2 + 2) = 2,
2 1 plays with players 2, 3, 4, and 5, but he can only get the
1 1 information from 3 and 4. So
c5 = (0 + 1) = .
2 2 (i) if (x3 (t) = 2)∩(x4 (t) = 2) or (x3 (t) = 2)∩(x4 (t) = 1)
(ii) Comparing c1 with c2 , we have f1 = x1 = 1. As for or (x3 (t) = 1) ∩ (x4 (t) = 2), then
f2 we have three choices:
j = 1 ⇒ f2 = x1 = 1, x1 (t + 1) = 1;
j = 3 ⇒ f2 = x3 = 2, (ii) if (x3 (t) = 3)∩(x4 (t) = 3) or (x3 (t) = 3)∩(x4 (t) = 2)
j = 5 ⇒ f2 = x5 = 2. or (x3 (t) = 2) ∩ (x4 (t) = 3), then
Hence f2 = 1 with probability 13 and f2 = 2 with x1 (t + 1) = 2;
probability 32 . We briefly express this by f2 = ( 13 , 23 ).
Similarly, we can calculate fi as in Table 5. (iii) if (x3 (t) = 1)∩(x4 (t) = 1) or (x3 (t) = 1)∩(x4 (t) = 3)
or (x3 (t) = 3) ∩ (x4 (t) = 1), then
Finally, we have
M1 = δ2 [1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 1, 2, 2, 2, 2, 2 x1 (t + 1) = 3.
(22)
1, 1, 1, 1, 1, 2, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2], We skip the detailed computation process and present the
results as follows.
M2 = δ2 [1, 1, 1, 1, 1, 1, 1, ( 13 , 23 ), ( 23 , 31 ), ( 13 , 23 ), 2, 2, ( 13 , 32 ), 2,
fi (t + 1) = Mi x(t), i = 1, 2, 3, 4, 5, (27)
2, 2, 1, 1, 1, 1, 1, 1, 1, ( 13 , 23 ), ( 23 , 13 ), 2, 2, 2, 2, 2, 2, 2],
(23) where

279
19th IFAC World Congress
Cape Town, South Africa. August 24-29, 2014

M1 = δ3 [ 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3, 3, 2, 2, 2, 2 D. Cheng, F. He, H. Qi, T. Xu (Preprint2013). Mod-

2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3, 3, 2, 2
eling, analysis and control of networked evolutionary
2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3, 3
2, 2, 2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1, 2, 2, 2, 3 games, http://lsc.amss.ac.cn/~dcheng/preprint/
3, 3, 2, 2, 2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1, 2, 2 NTGAME02.pdf (IEEE TAC, provisionally accepted).
2, 3, 3, 3, 2, 2, 2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1, 1, 1 E. Fornasini, M.E. Valcher (2013b). Observability, recon-
2, 2, 2, 3, 3, 3, 2, 2, 2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1, 1, 1
structibility and state observers of Boolean control net-
1, 1, 2, 2, 2, 3, 3, 3, 2, 2, 2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3, 1, 1
1, 1, 1, 1, 2, 2, 2, 3, 3, 3, 2, 2, 2, 2, 2, 2, 3, 3, 3, 1, 1, 1, 3, 3, 3 works, IEEE Trans. Aut. Contr., vol. 58, no. 6, pp. 1390-
1, 1, 1, 1, 1, 1, 2, 2, 2, 3, 3, 3, 2, 2, 2, 2, 2, 2]; 1401.
M2 = δ3 [ 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3
G. Hochma, M. Margaliot, E. Fornasini, M.E. Valcher
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3 (2013). Symbolic dynamics of Boolean control networks,
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3 Automatica, vol. 49, no. 8, pp. 2525-2530. .
3, 3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 D. Laschov, M. Margaliot (2012). Controllability of
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
Boolean control networks via the Perron-Frobenius the-
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 ory, Automatica, vol. 48, no. 6, pp. 1218-1223.
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 F. Li, J. Sun (2011a). Controllability of Boolean control
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 networks with time delays in states, Automatica, vol.
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2];
47, no. 3, pp. 603-607.
M3 = δ3 [ 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3 R. Li, M. Yang, T. Chu (2013b). State feedback stabiliza-
3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 tion for Boolean control networks, IEEE Trans. Aut.
1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2
2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3 Contr., vol. 58, no. 7, pp. 1853-1857.
3, 3, 3, 3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 Z. Liu, Y. Wang (2012). Disturbance decoupling of mix-
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2 valued logical networks via the semi-tensor product
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3 method, Automatica, vol. 48, no. 8, pp. 1839-1844.
3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 2, 2 L. Ljung, T. Söderström (1982). Theory and Practice of
2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2]; Recursive Identification, MIT Press.
M4 = δ3 [ 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3
M.A. Nowak, R.M. May (1992). Evolutionary games and
1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1 spatial chaos, Nature, vol. 359, pp. 826-829..
2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2 H. Ohtsuki, C. Hauert, E. Lieberman, M.A. Nowak (2006).
3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3 A simple rule for the evolution of cooperation on graphs
1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1
2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2
and social networks, Nature, vol. 441, pp. 502-505.
3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3 E. Rasmusen (2007). Games and Information: An In-
1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1 troduction to Game Theory, 4th Ed., Basil Blackwell,
2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2 Oxford.
3, 2, 3, 1, 1, 1, 1, 2, 2, 3, 2, 3, 1, 1, 3, 2, 2, 2, 3, 2, 3, 1, 1, 3, 2
2, 2, 3, 2, 3, 1, 1, 3, 2, 2, 2, 3, 2, 3, 1, 1, 3, 2, 2, 2, 3, 2, 3, 1, 1
F.C. Santos, M.D. Santos, J.M. Pacheco (2008). Social
3, 2, 2, 2, 3, 2, 3, 1, 1, 3, 2, 2, 2, 3, 2, 3, 1, 1, 3, 2, 2, 2, 3, 2, 3 diversity promotes the emergence of cooperation in
3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2, 3, 1, 2]; public goods games, Nature, vol. 454, pp. 213-216.
M5 = M2 . J.M. Smith (1982). Evolution and the Theorem of Games,
Cambridge Univ. Press, Cambridge.
R. Sugden (1986). The Economics of Rights, Cooperation
4. CONCLUSION
and Welfre, Blackwwell, Oxford.
G. Szabo, C. Toke (1998). Evolutionary prisoner’s dilemma
This paper gives a comprehensive introduction for the game on a square lattice, Phys. Rev. E, vol. 58, pp. 69.
modeling of networked evolutionary games. After an intro- Pl Taylor, L. Jonker (1978). Evolutionary stable strategies
duction to NEGs and to semi-tensor product, the NEG is and game dynamics, Math. Biosci., vol. 16, pp. 76-83.
formulated as a triplet: (i) network graph; (ii) fundamental A. Traulsen, M.A. Nowak, J.M. Pacheco (2006). Stochastic
network game; (iii) strategy updating rule. A detailed dynamics of invasion and fixation, Phys. Rev., E, vol. 74,
discussion with some illustrative examples are presented. pp. 011909.
Then the fundamental evolutionary equation (FEE) is Y. Wang, C. Zhang, Z. Liu (2012). A matrix approach to
proposed, which determines the overall network dynamics. graph maximum stable set and coloring problems with
The formulas to calculate FEE according to the triplet of application to multi-agent systems, vol. 48, no. 7, pp.
NEG are obtained. 1227-1236.
X. Xu, Y. Hong (2013). Matrix approach to model match-
REFERENCES ing of asynchrollous sequential machines, IEEE Trans.
Aut. Contr., vol. 58, no. 11, pp. 2974-2979.
J.P. Benoit, V. Krishna (1985). Finitely repeated games, H.P. Young (1993). The evolution of conventions, Econo-
Econometrica, vol. 17, no. 4, pp. 317-320. metrica, vol. 61, pp. 57-84.
E.L. Charnov (1982). The Theory of Sex Allocation, L. Zhang, K. Zhang (2013). L2 stability, H∞ control
Princiton Univ. Press, New Jersey. of switched homogeneous nonlinear systems and their
D. Cheng, H. Qi, Z. Li (2011). Analysis and Control of semi-tensor product of matrices representation, Int. J.
Boolean Networks: A Semi-tensor Product Approach, Robus. Nonli. Contr., vol. 23, pp. 638-652.
Springer, London, 2011. Y. Zhao, J. Kim, M. Filippone (2013). Aggregation al-
D. Cheng, H. Qi, Y. Zhao (2012a). An Introduction to gorithm towards large-scale Boolean network analysis,
Semi-tensor Product of Matrices and Its Applications, IEEE Trans. Aut. Contr., vol. 58, no. 8, pp. 1976-1985.
World Scientific, Singapore.

280