(S, p) ,→P (hS, Oi, p) hS, Oi,→ Sβ 0 ∧ hS, Oi ∈ V env ∧ S ∈ Vctr ∧ p0 = δ(p, β) (hS, Oi, p),→β P (S0, p0) hS, Oi ,→ v ∧ hS, Oi ∈ Venv ∧ v ∈ Vterm ∧ p ∈ P (v, p) ,→P (v, p)
Figure 7.1: Rules for the edge relation ,→P of the product game, where ,→ is the edge
relation as in Figure 5.1.
7.2
The complete-information ω-regular game
Given a transition system(A, Fair[A]) with the additional information about visibility and controllability as before, let G = (V, ,→) be the complete-information game obtained from A# as in Section 5.1 and let P = (P, Obs, δ, p
0, Acc, Accstop, Accdiv) be the de-
terministic, observation-based Streett automaton for the objective, with acceptance pairs Acc ={(R1, G1), . . . , (Rk, Gk)}, i.e., a Streett condition with k acceptance pairs.
We perform a product construction between the gameG and P to obtain the complete- information gameGP = (VP, ,→P) whereP serves as a monitor to determine acceptance
of the plays in the game. The set of vertices in the product game isVP =VctrP∪ VenvP ∪ VtermP ,
withVP
ctr = Vctr× P , VenvP = Venv× P and VtermP = Vterm× P , i.e., each vertex in the
complete-information game consists of a vertex of the underlying gameG and a state of P. The edge relation ,→P in the complete-information game is given by the three rules in
Figure 7.1, where,→ is the edge relation as in Figure 5.1.
The first two rules ensure that the state ofP embedded in the vertices of VP tracks the ob-
servables occurring in the plays. The third rule introduces self-loops on each of the terminal vertices of the underlying game. This ensures that all maximal plays inGP are infinite and
simplifies the handling of these terminal states later on.
The initial vertex for plays inGP is the vertex([Q0]∗, p0)∈ VctrP, where[Q0]∗ is the initial
vertex ofG and p0 is the initial state ofP. As P is total and deterministic, the product
construction preserves the structure of the underlying game G and there exists a one-to- one relation between maximal plays inG and GP by adding or stripping the state ofP and
extending or contracting the plays reaching vertices inVterm.
We can now treatGP as a standard two-player turn-based game, with the verticesVctrP be-
longing to player 1, while the vertices of player 2 consist ofVP
env. TheVtermP can either be
treated as belonging to player 1 or to player 2, as they only have a single self-looping edge and thus provide only a single option to the player.
7.2. The complete-information ω-regular game Chapter 7. Most general controllers for ω-regular objectives
We now consider how to encode the constraints on decision functions that enforce the ob- jective into a winning condition for the gameGP, i.e., a Streett condition over the vertices of
the game. The winning condition encodes three constraints for player 1. The first condition, ΩP, ensures that the infinite observations of winning plays in the game are accepted by the
automatonP. The second condition, Ωterm, ensures that plays that reach a vertex inVtermP
are correctly classified as winning or losing depending onAccdivandAccstopas well as the
failure to ensure admissibility. The third Streett condition,Ωfair, ensures that winning plays
have to satisfy the fairness conditionFair[A] of the systemA. The first part of the Streett winning conditionΩP =
(RP1, GP1), . . . , (RPk, GPk) for the gameGP is obtained by lifting the acceptance conditionAcc =
(R1, G1), . . . , (Rk, Gk)
ofP from states of P to the corresponding vertices in VP
ctrofGP, i.e., for16 i 6 k,
RP
i = {(S, p) ∈ VctrP : p∈ Ri},
GPi = {(S, p) ∈ VP
ctr : p∈ Gi}.
The second part of the Streett winning condition for the game, Ωterm = (Rterm, Gterm) is
obtained fromAccstopandAccdiv inP, the accepting states in case of termination and di-
vergence, with
Rterm = (stop(s), p)∈ VP
term : p /∈ Accstop
∪ (div(s), p)∈ VtermP : p /∈ Accdiv
∪ (fail(s), p)∈ VtermP and
Gterm = ∅
Hence Rterm contains exactly the vertices from VP
term that represent either non-accepting
termination, non-accepting divergence or failure to ensure admissibility.
Let Fair[A] = {F1, . . . , F`} be the fairness condition of the system A. As in the pre-
vious chapter, we require that Fair[A] is history-independent (cf. Definition 5.1). Re- call that this allows us to relate the vertices S in the game to the corresponding fair choices as specified byFair[A]. The third part of the Streett winning condition for GP,
Ωfair =
(Rfair
1 , Gfair1 ), . . . , (Rfair` , Gfair` )
is then obtained by lifting the fairness condition to the corresponding vertices, i.e., for16 i 6 `,
Rfair i = (S, p)∈ VctrP : Fi(S|A)6= ∅ , Gfair i = (hS, Oi, p) ∈ VP
env : β∈ O implies β ∈ O0for someO0 ∈ Fi(S|A) .
A vertexv ∈ VP
ctr ofGP is thus contained inRfairi if the observations that can be used to
reachv require a fair choice according to FiofFair[A]. Similarly, a vertex v ∈ VenvP ofGP
is contained inGfair
i if the choiceO is a fair choice according to FiofFair[A].
The Streett winning condition for GP is then given by Ω
GP = ΩP ∪ Ωterm ∪ Ωfair, the
conjunction ofΩP,ΩtermandΩfair.
The construction ofGP ensures that there are no vertices without outgoing edges, thus all