236 R. Schwieger and H. Siebert
In addition to the (local) interaction graph(s), we attribute tofa second
graph, which describes the dynamics of the components. Here, we use an asyn-
chronous update scheme attributing tofanasynchronous state transitiongraph
Gasync(f)=(Vasync(f),Easync(f)) withVasync(f):={ 0 , 1 }nand
Easync(f)=
{
(s, t)∈Vasync(f)×Vasync(f)|
(
diff(s, t)={i}andfi(s)=ti
)
ors=t=f(s)
}
.
Example 1.Letf(x 1 ,x 2 ,x 3 ,x 4 )=(1−x 4 ,x 1 ,x 2 ·(1−x 4 ), 1 −x 3 ) be a Boolean
function. Its global interaction graph and asynchronous state transition graph is
depicted in Fig. 1. The global interaction graph is given by the adjacency matrix
Σ=
⎛
⎜
⎜
⎝
000 −
+000
0+0−
00 − 0
⎞
⎟
⎟
⎠,
x1
x2
x3
x4
0000
0001
1000
0010
1010
0011
0100
0101
0110
1100
1110
0111
1001
1101
1011
1111
Fig. 1.Left: Global interaction graph of the running example. Right: ASTG of the
running example.
2.2 Monotonic Ensembles and Qualitative Differential Equations
The motivation of our results on sets of Boolean models comes from approaches
to analyze families of ODE models ̇x=f(x) which share some qualitative prop-
erties. Mainly, these are sign constraints on the Jacobi matrix of the right hand
sides of the ODE-System. Instead of the solutionsx(·) of the ODE-systems,
so-called “abstractions” are considered. In the context of this paper, these
abstractions are sequences of sign vectors of the derivatives of the solutions.
A state transition graph on the sign vectors can be constructed based on
the sign matrix, which captures restrictions on the behavior of the solutions.