Content uploaded by Aniello Murano
Author content
All content in this area was uploaded by Aniello Murano on Nov 08, 2018
Content may be subject to copyright.
Decidability results for ATL with imperfect
information and perfect recall∗
Raphaël Berthon1, Bastien Maubert2, and Aniello Murano3
1 École Normale Supérieure de Rennes, France
raphael.berthon@ensrennes.fr
2 Università degli Studi di Napoli Federico II, Italy
bastien.maubert@gmail.com
3 Università degli Studi di Napoli Federico II, Italy
murano@na.infn.it
Abstract
Alternatingtime Temporal Logic (ATL∗) is a central logic for multiagent systems. It has been
extended in various ways, notably with imperfect information (ATL∗
i). Since the modelchecking
problem against ATL∗
ifor agents with perfect recall is undecidable, studies have mostly focused
either on agents without memory, or on alternative semantics to retrieve decidability. In this
work, we establish new, strong decidability results for agents with perfect recall. We ﬁrst prove a
metatheorem that allows the transfer of decidability results for classes of multiplayer games with
imperfect information, such as games with hierarchical observation, to the modelchecking prob
lem for ATL∗
i. We also establish that model checking ATL∗with strategy context and imperfect
information for hierarchical instances is decidable.
1 Introduction
In formal system veriﬁcation, model checking is a wellestablished method to automatically
check the correctness of a system [
8
,
31
,
9
]. It consists in modelling the system as a
mathematical structure, expressing its desired behaviour as a formula from some suitable
logic, and checking whether the model satisﬁes the formula. In the nineties, interest has
arisen in the veriﬁcation of multiagent systems (MAS), in which various entities (the agents)
interact and can form coalitions to attain certain objectives. This led to the development of
logics that allow reasoning about strategic abilities in MAS [2, 27, 19, 35, 1, 7].
Alternatingtime Temporal Logic (
ATL∗
), introduced by Alur, Henzinger, and Kupfer
man [
2
], plays a central role in this line of work. This logic, interpreted on concurrent game
structures, extends
CTL∗
with strategic modalities. These modalities allow one to reason
about the existence of strategies for coalitions of agents to force the system’s behaviour to
satisfy certain temporal properties.
ATL∗
has been extended in many ways, and among these
extensions an important one is
ATL∗
with strategy context [
6
,
25
]. In
ATL∗
, strategies of all
agents are forgotten at each new strategic modality. In
ATL∗
with strategy context (
ATL∗
sc
)
instead they are stored in a strategy context, and are forgotten only when replaced by a new
strategy or when the formula explicitly unbinds the agent from her strategy. Thanks to this
additional expressive power,
ATL∗
sc
can express important game theoretic concepts such as
the existence of Nash Equilibria [25].
∗
This project has received funding from the European Union’s Horizon 2020 research and innovation
programme under the Marie SklodowskaCurie grant agreement No 709188.
arXiv:1805.12582v2 [cs.LO] 3 Sep 2018
2 Decidability results for ATL with imperfect information and perfect recall
In many reallife scenarios, such as poker, agents do not always know precisely what is
the current state of the system. Instead, they have a partial view, or observation, of the state.
This fundamental feature of MAS is called imperfect information, and it is known to quickly
bring about undecidability when involved in strategic problems, especially when agents have
perfect recall of the past, which is a usual and important assumption in games with imperfect
information and epistemic temporal logics [
13
]. For instance solving multiplayer games
with imperfect information and perfect recall, i.e., deciding the existence of a distributed
winning strategy in such games, is already undecidable for reachability objective, as proven by
Peterson, Reif and Azhar [
28
]. Since such games are easily captured by
ATL∗
with imperfect
information (ATL∗
i), model checking ATL∗
iwith perfect recall is also undecidable [2].
However it is known that restricting attention to cases where some sort of hierarchy exists
on the diﬀerent agents’ information yields decidability for several problems related to the
existence of strategies. Synthesis of distributed systems, which implicitly uses perfect recall
and is undecidable in general [
30
], is decidable for hierarchical architectures [
23
]. Actually, for
branchingtime speciﬁcations, distributed synthesis is decidable exactly on architectures free
from information forks, for which the problem can be reduced to the hierarchical case [
14
]. For
richer speciﬁcations from alternatingtime logics, being free of information forks is no longer
suﬃcient, but distributed synthesis is decidable precisely on hierarchical architectures [
32
].
Similarly, solving multiplayer games with imperfect information and perfect recall, i.e.,
checking for the existence of winning distributed strategies, is decidable for
ω
regular winning
conditions when there is a hierarchy among players, each one observing more than those
below [
29
,
23
]. Recently, it has been proven that this assumption can be relaxed while
maintaining decidability: the problem remains decidable if the hierarchy can change along a
play, or even if transient phases without such a hierarchy are allowed [5].
Our contribution.
In this work we establish several decidability results for model
checking
ATL∗
i
with perfect recall, with and without strategy context, all related to notions
of hierarchy. Our ﬁrst result is a theorem that allows the transfer of decidability results for
classes of multiplayer games with imperfect information, such as those mentioned above,
to the modelchecking problem for
ATL∗
i
. This theorem essentially states that if solving
multiplayer games with imperfect information, perfect recall and omegaregular objectives
is decidable on some class of concurrent game structures, then model checking
ATL∗
i
with
perfect recall is also decidable on this class of models (a simple bottomup algorithm that
evaluates innermost strategic modalities in every state of the model suﬃces). As a direct
consequence we easily obtain new decidability results for the model checking of
ATL∗
i
on
several classes of concurrent game structures.
Our second contribution considers
ATL∗
with imperfect information and strategy context
(
ATL∗
sc,i
). Because there are in general inﬁnitely many possible strategy contexts, the
bottomup approach used for
ATL∗
i
cannot be used here. Instead we build upon the proof
presented in [
25
] to establish the decidability of model checking
ATL∗
sc
, by reduction to
the modelchecking problem for Quantiﬁed
CTL∗
(
QCTL∗
). The latter extends
CTL∗
with
secondorder quantiﬁcation on atomic propositions, and it has been studied in a number
of works [
34
,
21
,
22
,
15
,
24
].
QCTL∗
i
, an imperfectinformation extension of
QCTL∗
, has
recently been introduced, and its modelchecking problem was proven decidable for the class
of hierarchical formulas [
4
]. In this paper we deﬁne a notion of hierarchical instances for the
ATL∗
sc,i
modelchecking problem: informally, an
ATL∗
sc,i
formula
ϕ
together with a concurrent
game structure
G
is a hierarchical instance if outermost strategic modalities in
ϕ
concern
agents who observe less in
G
. We adapt the proof from [
25
] and reduce the modelchecking
problem for
ATL∗
sc,i
on hierarchical instances to the modelchecking problem for hierarchical
R. Berthon, B. Maubert and A. Murano 3
QCTL∗
i
formulas. We obtain that model checking hierarchical instances of
ATL∗
sc,i
with
perfect recall is decidable.
Related work.
The modelchecking problem for
ATL∗
i
is known to be decidable when
agents have no memory [
33
], and the case of agents with bounded memory reduces to that of
no memory. Another way to retrieve decidability is to assume that all agents in a coalition
have the same information, either because their observations of the system are the same, or
because they can communicate and share their observations [
16
,
11
,
17
,
20
]. This idea was
also used recently to establish a decidability result for
ATL∗
sc,i
[
26
] when all agents have the
same observation of the game.
The results we establish here thus strictly extend previously known results on the
decidability of model checking
ATL∗
i
and
ATL∗
sc,i
with perfect recall and standard semantics,
and they hold for vast, natural classes of instances, that all rely on notions of hierarchy,
which seems to be inherent to all decidable cases of strategic problems for multiple entities
with imperfect information and perfect recall.
Outline.
After setting some basic deﬁnitions in Section 2, we present our transfer theorem
and its various corollaries concerning the model checking problem for
ATL∗
i
in Section 3. In
Section 4 we prove that when restricted to hierarchical instances, model checking
ATL∗
sc,i
is
decidable, and we conclude in Section 5.
2 Preleminaries
Let Σbe an alphabet. A ﬁnite (resp. inﬁnite)word over Σis an element of Σ
∗
(resp. Σ
ω
).
The empty word is noted
, and Σ
+
= Σ
∗\ {}
. The length of a word is
w
:= 0 if
w
is the
empty word
, if
w
=
w0w1. . . wn
is a ﬁnite nonempty word then
w
:=
n
+ 1, and for an
inﬁnite word
w
we let
w
:=
ω
. Given a word
w
and 0
≤i, j ≤ w −
1, we let
wi
be the
letter at position
i
in
w
and
w
[
i, j
]be the subword of
w
that starts at position
i
and ends at
position
j
. For
n∈N
we let [
n
] :=
{
1
, . . . , n}
. Finally, let us ﬁx a countably inﬁnite set of
atomic propositions AP and let AP ⊂ AP be some ﬁnite subset of atomic propositions.
2.1 Kripke structures
AKripke structure over
AP
is a tuple
S
= (
S, R, `
)where
S
is a set of states,
R⊆S×S
is a
lefttotal1transition relation and `:S→2AP is a labelling function.
Apointed Kripke structure is a pair (
S, s
)where
s∈ S
. A path in a structure
S
= (
S, R, `
)
is an inﬁnite word
λ
over
S
such that for all
i∈N
,(
λi, λi+1
)
∈R
. For
s∈S
,
Paths
(
s
)is
the set of all paths that start in s.
2.2 Inﬁnite trees
Let Xbe a ﬁnite set. An Xtree τis a nonempty set of words τ⊆X+such that
there exists r∈X, called the root of τ, such that each u∈τstarts with r;
if u·x∈τand u6=, then u∈τ, and
if u∈τthen there exists x∈Xsuch that u·x∈τ.
The elements of a tree
τ
are called nodes. If
u·x∈τ
, we say that
u·x
is a child of
u
.
Similarly to Kripke structures, a path is an inﬁnite sequence of nodes
λ
=
u0u1. . .
such that
for all
i
,
ui+1
is a child of
ui
, and
P aths
(
u
)is the set of paths that start in node
u
. An
1i.e., for all s∈S, there exists s0such that (s, s0)∈R.
4 Decidability results for ATL with imperfect information and perfect recall
AP
labelled
X
tree, or (
AP, X
)tree for short, is a pair
t
= (
τ, `
), where
τ
is an
X
tree called
the domain of tand `:τ→2AP is a labelling.
IDeﬁnition 1
(Tree unfoldings)
.
Let
S
= (
S, R, `
)be a Kripke structure over
AP
, and let
s∈S
. The treeunfolding of
S
from
s
is the (
AP, S
)tree
tS
(
s
) = (
τ, `0
), where
τ
is the set
of all ﬁnite paths that start in s, and for every u∈τ,`0(u) = `(last(u)).
3ATL∗with imperfect information
In this section we recall the syntax and semantics of
ATL∗
with imperfect information and
synchronous perfectrecall semantics, or
ATL∗
i
for short, and establish a metatheorem on
the decidability of its modelchecking problem.
3.1 Deﬁnitions
We ﬁrst introduce the models of the logics we study. For the rest of the paper, let us ﬁx a
nonempty ﬁnite set of agents Ag and a nonempty ﬁnite set of moves M.
IDeﬁnition 2.
Aconcurrent game structure with imperfect information (or
CGSi
for short)
over
AP
is a tuple
G
= (
V, E , `, {∼a}a∈Ag
)where
V
is a nonempty ﬁnite set of positions,
E
:
V×MAg →V
is a transition function,
`
:
V→
2
AP
is a labelling function and for each
agent a∈Ag, ∼a⊆V×Vis an equivalence relation.
In a position
v∈V
, each agent
a
chooses a move
ma∈M
, and the game proceeds to
position
E
(
v, m
), where
m∈MAg
stands for the joint move (
ma
)
a∈Ag
(note that we assume
E
(
v, m
)to be deﬁned for all
v
and
m2
). For each position
v∈V
,
`
(
v
)is the ﬁnite set of
atomic propositions that hold in
v
, and for
a∈Ag
, equivalence relation
∼a
represents the
observation of agent
a
: for two positions
v, v0∈V
,
v∼av0
means that agent
a
cannot tell
the diﬀerence between
v
and
v0
. We may write
v∈ G
for
v∈V
. A pointed
CGSi
(
G, v
)is a
CGSiGtogether with a position v∈ G.
In Section 3.2 we also use nondeterministic
CGSi
, which are as in Deﬁnition 2 except
that they have a transition relation
E⊆V×MAg ×V
instead of a transition function. In a
position
v
, after every agent has chosen a move, forming a joint move
m∈MAg
, a special
agent called Nature (not in
Ag
) chooses a next position
v0
such that (
v, m, v0
)
∈E
(see [
5
]
for detail). In the following, unless explicitly speciﬁed,
CGSi
always refers to deterministic
CGSi
. The following deﬁnitions also concern deterministic
CGSi
, but they can be adapted
to nondeterministic ones in an obvious way.
Aﬁnite (resp. inﬁnite)play is a ﬁnite (resp. inﬁnite) word
ρ
=
v0. . . vn
(resp.
π
=
v0v1. . .
)
such that for all
i
with 0
≤i < ρ −
1(resp.
i≥
0), there exists a joint move
m
such that
E
(
vi,m
) =
vi+1
. A ﬁnite (resp. inﬁnite) play
ρ
(resp.
π
)starts in a position
v
if
ρ0
=
v
(resp.
π0
=
v
). We let
Plays
(
G, v
)be the set of plays, either ﬁnite or inﬁnite, that start in
v
.
In this work we consider agents with synchronous perfect recall, meaning that the
observational equivalence relation for each agent
a
is extended to ﬁnite plays the following
way:
ρ∼aρ0
if
ρ
=
ρ0
and
ρi∼aρ0
i
for every
i∈ {
0
,...,ρ −
1
}
. A strategy for agent a
is a function
σ
:
V+→M
such that
σ
(
ρ
) =
σ
(
ρ0
)whenever
ρ∼aρ0
. The latter constraint
captures the essence of imperfect information, which is that agents can base their strategic
2
This assumption, as well as the choice of a unique set of moves for all agents, is made to ease presentation.
All the results presented here also hold when the set of available moves depends on the agent and the
position.
R. Berthon, B. Maubert and A. Murano 5
choices only on the information available to them, and removing this constraint yields the
semantics of classic ATL with perfect information.
Astrategy proﬁle for a coalition
A⊆Ag
is a mapping
σA
that assigns a strategy to each
agent
a∈A
; for
a∈A
, we may write
σa
instead of
σA
(
a
). An inﬁnite play
π
follows a
strategy proﬁle
σA
for a coalition
A
if for all
i≥
0, there exists a joint move
m
such that
E
(
πi,m
) =
πi+1
and for each
a∈A
,
ma
=
σa
(
π
[0
, i
]). For a strategy proﬁle
σA
and a
position
v∈V
, we deﬁne the outcome
Out
(
v, σA
)of
σA
in
v
as the set of inﬁnite plays that
start in vand follow σA.
The syntax of
ATL∗
i
is the same as that of
ATL∗
, and is given by the following grammar:
ϕ::= p ¬ϕϕ∨ϕ hAiϕXϕϕUϕ,
where p∈ AP and A⊆Ag.
X
and
U
are the classic next and until operators, respectively, while the strategic operator
hAiquantiﬁes on strategy proﬁles for coalition A.
The semantics of
ATL∗
i
is deﬁned with regards to a
CGSiG
= (
V, E , `, {∼a}a∈Ag
), an
inﬁnite play πand a position i≥0along this play, by induction on formulas:
G, π, i =pif p∈`(πi)
G, π, i =¬ϕif G, π, i 6=ϕ
G, π, i =ϕ∨ϕ0if G, π, i =ϕor G, π, i =ϕ0
G, π, i =hAiϕif there exists a strategy proﬁle σAs.t.
for all π0∈Out(πi, σA),G, π0,0=ϕ
G, π, i =Xϕif G, π, i + 1 =ϕ
G, π, i =ϕUϕ0if there exists j≥is.t. G, π, j =ϕ0and,
for all ks.t. i≤k < j,G, π, k =ϕ.
An
ATL∗
i
formula
ϕ
is closed if every temporal operator (
X
or
U
) in
ϕ
is in the scope of
a strategic operator
hAi
. Since the semantics of a closed formula
ϕ
does not depend on the
future, we may write G, v =ϕif G, π, 0=ϕfor any inﬁnite play πthat starts in v.
The modelchecking problem for
ATL∗
i
consists in deciding, given a closed
ATL∗
i
formula
ϕand a ﬁnite pointed CGSi(G, v), whether G, v =ϕ.
3.2 Model checking ATL∗
i
It is well known that the modelchecking problem for
ATL∗
i
is undecidable for agents with
perfect recall [
2
], as it can easily express the existence of distributed winning strategies for
multiplayer reachability games with imperfect information and perfect recall, which was
proved undecidable by Peterson, Reif and Azhar [
28
]. A direct proof of this undecidability
result for
ATL∗
i
is also presented in [
12
]. However, there are classes of multiplayer games with
imperfect information that are decidable. For many years, the only known decidable case
was that of hierarchical games, in which there is a total preorder among players, each player
observing at least as much as those below her in this preorder [
29
,
23
]. Recently, this result
has been extended by relaxing the assumption of hierarchical observation. In particular,
it has been shown that the problem remains decidable if the hierarchy can change along a
play, or if transient phases without such a hierarchy are allowed [
5
]. We establish that these
results transfer to the modelchecking problem for ATL∗
i.
We remind that a concurrent game with imperfect information is a pair ((
G, v
)
, W
)where
(
G, v
)is a pointed nondeterministic
CGSi
and
W
is a property of inﬁnite plays called the
winning condition. The strategy problem is, given such a game, to decide whether there exists
6 Decidability results for ATL with imperfect information and perfect recall
a strategy proﬁle for the grand coalition
Ag
to enforce the winning condition against Nature
(for more details see, e.g., [5]).
Before stating our transfer theorem we need to introduce a couple of additional notions.
First we introduce a notion of abstraction over a group of agents. Informally, abstracting a
CGSiG
over an agent consists in erasing her from the group of agents and letting Nature
play for her in G.
IDeﬁnition 3
(Abstraction)
.
Let
A⊆Ag
be a group of agents and let
G
= (
V, E , `, {∼a}a∈Ag
)
be a
CGSi
. The abstraction of
G
from
A
is the nondeterministic
CGSi
over set of agents
Ag \Adeﬁned as G ↑A:= (V, E 0, `, {∼a}a∈Ag\A), where for every v∈Vand m∈MAg\A,
(v, m, v0)∈E0if ∃m0∈MAs.t. E(v , (m,m0)) = v0.
Thanks to this notion we can deﬁne the following problem:
IDeﬁnition 4
(
A
strategy problem)
.
The
A
strategy problem takes as input a pointed
CGSi
(
G, v
), a set
A⊆Ag
of agents and a winning condition
W
, and returns the answer to the
strategy problem for the game ((G ↑Ag\A, v), W ).
The
A
strategy problem for (
G, v
)with winning condition
W
thus consists in deciding whether
there is a strategy proﬁle for agents in Ato enforce Wagainst everybody else.
Finally we introduce the following notion, which simply captures the change of initial
position in a game from a position vto another position v0reachable from v:
IDeﬁnition 5
(Initial shifting)
.
Let
G
be a
CGSi
and let
v, v0∈ G
. The pointed
CGSi
(
G, v0
)
is an initial shifting of (G, v)if v0is reachable from vin G.
We are now ready to state our ﬁrst result.
ITheorem 6.
If
C
is a class of pointed
CGSi
closed under initial shifting and such that the
A
strategy problem with
ω
regular objective is decidable on
C
, then model checking
ATL∗
i
is
decidable on C.
Proof.
Let
C
be such a class of pointed
CGSi
, and let (
ϕ,
(
G, v
)) be an instance of the
modelchecking problem for
ATL∗
i
on
C
. A bottomup algorithm consists in evaluating each
innermost subformula of
ϕ
of the form
hAiϕ0
, where
ϕ0
is thus an LTL formula, on each
position
v0
of
G
reachable from
v
. Evaluating
hAiϕ0
on
v0
amounts to solving an instance of
the
A
strategy problem
3
with
ω
regular objective (recall that LTL properties are
ω
regular).
By assumption (
G, v
)
∈ C
, and because
C
is closed by initial shifting and
v0
is reachable
from
v
, we have that (
G, v0
)
∈ C
. Also by assumption, the
A
strategy problem for
ω
regular
winning conditions is decidable on
C
. We thus have an algorithm to evaluate each
hAiϕ0
on
each
v0
. One can then mark positions of the game with fresh atomic propositions indicating
where these formulas hold, and repeat the procedure until all strategic operators have been
eliminated. It then remains to evaluate a boolean formula in the initial position v.J
Let us recall for which classes of nondeterministic
CGSi
the strategy problem is known
to be decidable. A (nondeterministic or deterministic)
CGSiG
has hierarchical observation
if there exists a total preorder
4
over
Ag
such that if
a4b
and
v∼av0
, then
v∼bv0
.
3
Observe that if
A
=
Ag
then
G ↑Ag\A
=
G
, and Nature thus does not do anything. This is coherent with
the fact that for agents with perfect recall
hAgiϕ≡Eϕ
, where
E
is the CTL path quantiﬁer, even for
imperfect information.
R. Berthon, B. Maubert and A. Murano 7
This notion was reﬁned in [
5
] to take into account the agents’ memory, using the notion
of information set: for a ﬁnite play
ρ∈Plays
(
G, v
)and an agent
a
, the information set of
agent
a
after
ρ
is
Ia
(
ρ
) :=
{ρ0∈Plays
(
G, v
)
ρ∼aρ0}
. A ﬁnite play
ρ
yields hierarchical
information if there is a total preorder
4
over
Ag
such that if
a4b
, then
Ia
(
ρ
)
⊆Ib
(
ρ
). If
all ﬁnite plays in
Plays
(
G, v
)yield hierarchical information for the same preorder over agents,
(
G, v
)yields static hierarchical information. If this preorder can vary depending on the play,
(
G, v
)yields dynamic hierarchical information. The last generalisation consists in allowing for
transient phases without hierarchy: if every inﬁnite play in
Plays
(
G, v
)has inﬁnitely many
preﬁxes that yield hierarchical information, (G, v)yields recurring hierarchical information.
IProposition 7.
Hierarchical observation as well as static, dynamic and recurring hierarch
ical information are preserved by abstraction.
Proof.
Abstraction removes agents without aﬀecting observations of remaining ones. The
result thus follows from the respective deﬁnitions of hierarchical observation and of static,
dynamic and recurring hierarchical information. J
IProposition 8.
Hierarchical observation as well as static, dynamic and recurring hierarch
ical information are preserved by initial shifting.
This is obvious for hierarchical observation. For the other cases we establish Lemma 9
below. It is then easy to check that Proposition 8 holds.
ILemma 9.
If a ﬁnite play
v·ρ·v0·ρ0
yields hierarchical information in (
G, v
), so does
v0·ρ0in (G, v0), with the same preorder among agents.
Proof.
Assume that
v·ρ·v0·ρ0
yields hierarchical information in (
G, v
)with preorder
4
over
Ag
. Suppose towards a contradiction that there are agents
a, b ∈Ag
such that
a4b
but
Ia
(
v0·ρ0
)
6⊆ Ib
(
v0·ρ0
). This means that there is
v0·ρ00 ∈Plays
(
G, v0
)such that
v0·ρ0∼av0·ρ00
but
v0·ρ06∼bv0·ρ00
. By deﬁnition of synchronous perfect recall relations we
then have that
v·ρ·v0·ρ0∼av·ρ·v0·ρ00
and
v·ρ·v0·ρ06∼bv·ρ·v0·ρ00
. This implies that
Ia
(
v·ρ·v0·ρ0
)
6⊆ Ib
(
v·ρ·v0·ρ0
), which contradicts the fact that
a4b
. Therefore for all
agents
a, b
such that
a4b
we have
Ia
(
v0·ρ0
)
⊆Ib
(
v0·ρ0
), and thus
v0·ρ0
yields hierarchical
information with preorder 4.J
Let
Cobs
(resp.
Cstat
,
Cdyn
,
Crec
) be the class of pointed
CGSi
with hierarchical observation
(resp. static, dynamic, recurring hierarchical information). We instantiate Theorem 6 to
obtain three decidability results for ATL∗
i.
ITheorem 10.
Model checking
ATL∗
i
is decidable on the class of
CGSi
with hierarchical
observation.
Proof.
By Proposition 8,
Cobs
is closed under initial shifting. It is proven in [
23
] that the
strategy problem is decidable for games with hierarchical observation and
ω
regular objectives.
Since, by Proposition 7, all pointed nondeterministic
CGSi
obtained by abstracting agents
from
CGSi
in
Cobs
also yield hierarchical observation, we get that the
A
strategy problem
with
ω
regular objectives is decidable on
Cobs
. We can therefore apply Theorem 6 on
Cobs
.
J
It is proven in [
5
] that the strategy problem with
ω
regular objectives is also decidable
for games with static hierarchical information and for games with dynamic hierarchical
information. Since Proposition 7 and Proposition 8 also hold for
Cstat
and
Cdyn
, with the
same argument as in the proof of Theorem 10, we obtain the following results as consequences
of Theorem 6:
8 Decidability results for ATL with imperfect information and perfect recall
ITheorem 11.
Model checking
ATL∗
i
is decidable on the class of
CGSi
with static hierarchical
information.
ITheorem 12.
Model checking
ATL∗
i
is decidable on the class of
CGSi
with dynamic
hierarchical information.
Note that in fact, since
Cobs ⊂ Cstat ⊂ Cdyn
, Theorem 10 and Theorem 11 are also obtained
as corollaries of Theorem 12, but we wanted to illustrate how Theorem 6 can be applied to
obtain decidability results for diﬀerent classes of CGSi.
I
Remark. The last result in [
5
] establishes that the strategy problem is decidable for games
with recurring hierarchical information, but only for observable
ω
regular winning conditions,
i.e., when all agents can tell whether a play is winning or not. Now considering
ATL∗
i
on
Cdyn
we could require atomic propositions to be observable for all agents; in that case we could
evaluate the innermost strategy quantiﬁers using the abovementioned result. But then the
fresh atomic propositions that mark positions where these subformulas hold (see the proof of
Theorem 6) would not, in general, be observable by all agents. So on
Crec
we could obtain a
decision procedure for the fragment of
ATL∗
i
without nested nontrivial strategy quantiﬁers,
where nontrivial means for coalitions other than the empty coalition or the one made of all
agents (which, we recall, are simply the CTL path quantiﬁers). We do not state it explicitly
because it does not seem of much interest.
Concerning complexity, the strategy problem for games with imperfect information and
hierarchical observation is already nonelementary [30, 28], hence the following result:
ICorollary 13.
Model checking
ATL∗
i
is nonelementary on games with hierarchical observa
tion, hence also for games with static or dynamic hierarchical information.
We now turn to ATL with imperfect information and strategy context, and study its
modelchecking problem.
4ATLiwith strategy context
While in ATL strategies for all agents are forgotten each time a new strategy quantiﬁer is
met, in ATL with strategy context (
ATLsc
) [
6
,
10
,
25
] agents keep using the same strategy as
long as the formula does not say otherwise. In this section we consider
ATLsc
with imperfect
information (
ATLsc,i
). As far as we know, the only existing work on this logic is [
26
], which
proved its modelchecking problem to be decidable in the case where all agents have the
same observation of the game. We extend signiﬁcantly this result by establishing that the
modelchecking problem is decidable as long as strategy quantiﬁcation is hierarchical, in the
sense that if there is a strategy quantiﬁcation for agent
a
nested in a strategy quantiﬁcation
for agent
b
, then
b
should observe no more than
a
. In other terms, innermost strategic
quantiﬁcations should concern agents who observe more.
4.1 Syntax and semantics
The models are still
CGSi
. To remember which agents are currently bound to a strategy, and
what these strategies are, the semantics uses strategy contexts. Formally, a strategy context for
a set of agents
B⊆Ag
is a strategy proﬁle
σB
. We deﬁne the composition of strategy contexts
as follows. If
σB
is a strategy context for
B
and
σA
is a new strategy proﬁle for coalition
A
,
we let
σA◦σB
be the strategy context for
A∪B
deﬁned as
σA∪B
:
a7→ (σA(a)if a∈A,
σB(a)otherwise
.
R. Berthon, B. Maubert and A. Murano 9
So if
a
is assigned a strategy by
σA
, her strategy in
σA◦σB
is
σA
(
a
). If she is not assigned
a strategy by σAher strategy remains the one given by σB, if any.
Also, given a strategy context
σB
and a set of agents
A⊆Ag
, we let (
σB
)
\A
be the
strategy context obtained by restricting σBto the domain B\A.
Finally, because agents who do not change their strategy keep playing the one they were
assigned, if any, we cannot forget the past at each strategy quantiﬁer, as in the semantics of
ATL∗
i
(see Section 3.1). We thus deﬁne the outcome of a strategy proﬁle
σA
after a ﬁnite
play
ρ
, written
Out
(
ρ, σA
), as the set of inﬁnite plays
π
that start with
ρ
and then follow
σA
:
π∈Out
(
ρ, σA
)if
π
=
ρ·π0
for some
π0
, and for all
i≥ ρ −
1, there exists a joint move
m∈MAg such that E(πi,m) = πi+1 and for each a∈A,ma=σa(π[0, i]).
To diﬀerentiate from
ATL∗
, in
ATL∗
sc
the strategy quantiﬁer for a coalition
A
is written
h·A·i
instead of
hAi
.
ATL∗
sc
also has an additional operator, (
A
), that releases agents in
A
from their current strategy, if they have one. The syntax of
ATL∗
sc,i
is the same as that of
ATL∗
sc and is thus given by the following grammar:
ϕ::= p ¬ϕϕ∨ϕ h·A·iϕ(A)ϕXϕϕUϕ,
where p∈ AP and A⊆Ag.
I
Remark. In [
25
] the syntax of
ATL∗
sc
contains in addition operators
h·A·i
and (
A
)for
complement coalitions. While they add expressivity when the set of agents is not ﬁxed, and
are thus of interest when considering expressivity or satisﬁability, they are redundant if we
consider model checking, which is our case in this work. To simplify presentation we thus
choose not to consider them here.
The semantics of
ATL∗
sc,i
is deﬁned with regards to a
CGSiG
= (
V, E , `, {∼a}a∈Ag
), an
inﬁnite play
π
, a position
i∈N
along this play, and a strategy context
σB
. The semantics is
deﬁned by induction on formulas:
G, π, i =σBpif p∈`(πi)
G, π, i =σB¬ϕif G, π , i 6=σBϕ
G, π, i =σBϕ∨ϕ0if G, π , i =σBϕor G, π, i =σBϕ0
G, π, i =σBh·A·iϕif there exists a strategy proﬁle σAs.t.
for all π0∈Out(π[0, i], σA◦σB),G, π0, i =σA◦σBϕ
G, π, i =σB(A)ϕif G, π , i =(σB)\Aϕ
G, π, i =σBXϕif G, π , i + 1 =σBϕ
G, π, i =σBϕUϕ0if there exists j≥is.t. G, π , j =σBϕ0
and, for all ksuch that i≤k < j,G, π , k =σBϕ.
The notion of closed formula is as deﬁned in Section 3.1 and once more, the semantics
of a closed formula
ϕ
being independent from the future, we may write
G, v 
=
σBϕ
instead
of
G, π,
0

=
σBϕ
for any inﬁnite play
π
that starts in position
v
. We also write
G, v 
=
ϕ
if
G, v =σ∅ϕ, that is if ϕholds in vwith the empty strategy context.
The modelchecking problem for
ATL∗
sc,i
consists in deciding, given a closed
ATL∗
sc,i
formula
ϕand a ﬁnite pointed CGSi(G, v), whether G, v =ϕ.
We now present
QCTL∗
with imperfect information, or
QCTL∗
i
for short, before proving our
main result on the modelchecking problem for
ATL∗
sc,i
by reducing it to the modelchecking
problem for a decidable fragment of QCTL∗
i.
4.2 QCTL∗with imperfect information
Quantiﬁed
CTL∗
, or
QCTL∗
for short, is an extension of
CTL∗
with secondorder quantiﬁers
on atomic propositions that has been well studied [
34
,
21
,
22
,
24
]. It has recently been
10 Decidability results for ATL with imperfect information and perfect recall
further extended to take into account imperfect information, resulting in the logic called
QCTL∗
with imperfect information, or
QCTL∗
i
[
3
,
4
]. We brieﬂy present this logic, as well as
a decidability result on its modelchecking problem proved in [
3
,
4
] and that we rely on to
establish our result on the model checking of ATL∗
sc,i.
Imperfect information is incorporated into
QCTL∗
by considering Kripke models with
internal structure in the form of local states, like in distributed systems (see for instance [
18
]),
and then parameterising quantiﬁers on atomic propositions with observations that deﬁne
what portions of the states a quantiﬁer can “observe”. The semantics is then adapted to
capture the idea of quantiﬁcation on atomic propositions being made with partial observation.
Let us ﬁx a collection
{Li}i∈[n]
of
n
disjoint ﬁnite sets of local states. We also let
Xn=L1×. . . ×Ln.
IDeﬁnition 14.
Acompound Kripke structure (CKS) over
AP
is a Kripke structure
S
=
(S, R, `)such that S⊆Xn.
The syntax of
QCTL∗
i
is that of
QCTL∗
, except that quantiﬁers over atomic propositions
are parameterised by a set of indices that deﬁnes what local states the quantiﬁer can “observe”.
It is thus deﬁned by the following grammar:
ϕ:= p ¬ϕϕ∨ϕEϕ ∃op. ϕ XϕϕUϕ
where
p∈ AP
and
o⊂N
is a ﬁnite set of indices. We use standard abbreviations:
>
:=
p∨¬p
,
⊥:= ¬>,Fϕ:= >Uϕ,Gϕ:= ¬F¬ϕand Aϕ:= ¬E¬ϕ.
A ﬁnite set
o⊂N
is called an observation, and two states
s
= (
l1, . . . , ln
)and
s0
=
(l0
1, . . . , l0
n)are oindistinguishable, written s≈os0, if for all i∈[n]∩o, it holds that li=l0
i.
The intuition is that a quantiﬁer with observation
o
must choose the valuation of atomic
propositions uniformly with respect to
o
. Note that in [
3
], two semantics are considered for
QCTL∗
i
, just like in [
24
] for
QCTL∗
: the structure semantics and the tree semantics. In the
former, formulas are evaluated directly on the structure, while in the latter the structure is
ﬁrst unfolded into an inﬁnite tree. Here we only present the tree semantics, as it is this one
that allows us to capture agents with perfect recall. But we ﬁrst need a few more deﬁnitions.
For
p∈ AP
, two labelled trees
t
= (
τ, `
)and
t0
= (
τ0, `0
)are equivalent modulo
p
, written
t≡pt0
, if
τ
=
τ0
and for each node
u∈τ
,
`
(
u
)
\ {p}
=
`0
(
u
)
\ {p}
. So
t≡pt0
if they are the
same trees, except for the labelling of proposition p.
This notion of equivalence modulo
p
is the one used to deﬁne quantiﬁcation on atomic
propositions in
QCTL∗
: intuitively, an existential quantiﬁcation over
p
chooses a new labelling
for valuation
p
, all else remaining the same, and the evaluation of the formula continues from
the current node with the new labelling. For imperfect information we need to express the
fact that this new labelling for a proposition is done uniformly with regards to the quantiﬁer’s
observation.
First, we deﬁne the notion of indistinguishability between two nodes in the unfolding of a
CKS. Let
o
be an observation, let
τ
be an
Xn
tree (which may be obtained by unfolding
some pointed CKS), and let
u
=
s0. . . si
and
u0
=
s0
0. . . s0
j
be two nodes in
τ
. The nodes
u
and
u0
are
o
indistinguishable, written
u≈ou0
, if
i
=
j
and for all
k∈ {
0
, . . . , i}
, we have
sk≈os0
k
. Observe that this deﬁnition corresponds to the notion of synchronous perfect
recall in
CGSi
(see Section 3.1). We now deﬁne what it means for the labelling of an atomic
proposition to be uniform with regards to an observation.
IDeﬁnition 15.
Let
t
= (
τ, `
)be a labelled
Xn
tree, let
p∈ AP
be an atomic proposition
and
o⊂N
an observation. Tree
t
is
o
uniform in
p
if for every pair of nodes
u, u0∈τ
such
that u≈ou0, we have p∈`(u)iﬀ p∈`(u0).
R. Berthon, B. Maubert and A. Murano 11
The satisfaction relation

=
t
(
t
is for tree semantics) is now deﬁned as follows, where
t= (τ, `)is a labelled Xntree, λis a path in τand i∈Na position along that branch:
t, λ, i =tpif p∈`(λi)
t, λ, i =t¬ϕif t, λ, i 6=tϕ
t, λ, i =tϕ∨ϕ0if t, λ, i =tϕor t, λ, i =tϕ0
t, λ, i =tEϕif there exists λ0∈P aths(λi)such that t, λ0,0=tϕ
t, λ, i =t∃op. ϕ if there exists t0≡ptsuch that t0is ouniform in pand t0, λ, i =tϕ
t, λ, i =tXϕif t, λ, i + 1 =tϕ
t, λ, i =tϕUϕ0if there exists j≥isuch that t, λ, j =tϕ0and for i≤k < j, t, λ, j =tϕ
Similarly to
ATL∗
sc,i
, we say that a
QCTL∗
i
formula is closed if all temporal operators are
in the scope of a path quantiﬁer. The semantics of such formulas depending only on the
current node, for a closed formula
ϕ
we may write
t
=
tϕ
for
t, r 
=
tϕ
, where
r
is the root of
t
, and given a pointed CKS (
S, s
)and a
QCTL∗
i
formula
ϕ
, we write
S, s 
=
tϕ
if
tS
(
s
)

=
tϕ
.
I
Remark. In [
3
] the syntax is presented with path formulas distinguished from state formulas,
and the semantics is deﬁned accordingly. To make the presentation more uniform with that
of ATLsc,i we chose here a diﬀerent, but equivalent, presentation.
IRemark. Note that when nis ﬁxed, the propositional quantiﬁer with perfect information
from
QCTL∗
is equivalent to the
QCTL∗
i
quantiﬁer that observes all the components, i.e.,
the quantiﬁer parameterised with observation [n].
The modelchecking problem for QCTL∗
iis the following: given a closed QCTL∗
iformula
ϕand a ﬁnite pointed CKS (S, s), decide whether S, s =tϕ.
We now deﬁne the class of
QCTL∗
i
formulas for which the modelchecking problem is
known to be decidable with the tree semantics.
IDeﬁnition 16.
A
QCTL∗
i
formula
ϕ
is hierarchical if for all subformulas
ϕ1, ϕ2
of the form
ϕ1=∃o1p1. ϕ0
1and ϕ2=∃o2p2. ϕ0
2where ϕ2is a subformula of ϕ0
1, we have o1⊆o2.
The following result is proved in [
3
], where
QCTL∗
i,⊂
is the set of hierarchical
QCTL∗
i
formulas:
ITheorem 17 ([3]).Model checking QCTL∗
i,⊂with tree semantics is decidable.
4.3 Model checking ATL∗
sc,i
We establish that model checking
ATL∗
sc,i
is decidable on a class of instances whose deﬁnition
relies on the notion of hierarchical observation.
IDeﬁnition 18.
Let
G
= (
V, E , `, {∼a}a∈Ag
)be a
CGSi
, and let
a, b ∈Ag
be two agents.
Agent
a
observes no more than agent
b
in
G
, written
a4Gb
, if for every pair of positions
v, v0∈V
,
v∼bv0
implies
v∼av0
. We say that
A⊆Ag
is hierarchical in
G
if
4G
is a total
preorder on A.
If a set of agents
A
is hierarchical in a
CGSiG
, we thus may talk about maximal and
minimal agents in A, referring to maximal and minimal elements of Afor the relation 4G.
The essence of the requirement that makes the problem decidable is the same as for the
decidability result on
QCTL∗
i
(Theorem 17): nesting of quantiﬁers (here, strategy quantiﬁers)
should be hierarchical, with those observing more inside those observing less. However, unlike
in
QCTL∗
i
, in
ATL∗
sc,i
observations are not part of formulas, but rather they are given by the
models. We thus deﬁne the notion of hierarchical ATL∗
sc,i formula with respect to a CGSi:
12 Decidability results for ATL with imperfect information and perfect recall
IDeﬁnition 19.
Let Φbe an
ATL∗
sc,i
formula and let
G
be a
CGSi
. We say that Φis
hierarchical in Gif:
for every subformula ϕof the form ϕ=h·A·iϕ0,Ais hierarchical in G, and
for all subformulas
ϕ1, ϕ2
of the form
ϕ1
=
h·A1·iϕ0
1
and
ϕ2
=
h·A2·iϕ0
2
where
ϕ2
is a
subformula of ϕ0
1, maximal agents of A1observe no more than minimal agents of A2.
An instance (Φ
,
(
G, v
)) of the modelchecking problem for
ATL∗
sc,i
is hierarchical if Φis
hierarchical in G.
In the rest of the section we establish the following:
ITheorem 20. Model checking ATL∗
sc,i is decidable on the class of hierarchical instances.
We build upon the proof in [
25
] that establishes the decidability of the modelchecking
problem for
ATL∗
sc
by reduction to the modelchecking problem for
QCTL∗
. The main diﬀer
ence is that we reduce to the modelchecking problem for
QCTL∗
i
instead, using quantiﬁers
parameterised with observations corresponding to agents’ observations. We also need to
make a couple of adjustments to obtain formulas in the decidable fragment QCTL∗
i,⊂.
Let (Φ
,
(
G, vι
)) be a hierarchical instance of the
ATL∗
sc,i
modelchecking problem, where
G
= (
V, E , `, {∼a}a∈Ag
)is a
CGSi
over
AP
. In the reduction we will transform Φinto an
equivalent
QCTL∗
i
formula Φ
0
in which we need to refer to the current position in the model
G
, and also to talk about moves taken by agents. To do so, we consider the additional sets of
atomic propositions
APv
:=
{pvv∈V}
and
APm
:=
{pa
ma∈Ag and m∈M}
, that we
take disjoint from AP.
First we deﬁne the CKS
SG
on which Φ
0
will be evaluated. Since the models of the
two logics use diﬀerent ways to represent imperfect information (equivalence relations on
positions for
CGSi
and local states for CKS) this requires a bit of work. First, for each
v∈V
and
a∈Ag
, let us deﬁne [
v
]
a
as the equivalence class of
v
for relation
∼a
. Now, noting
Ag
=
{a1, . . . , an}
, we deﬁne for each
i∈
[
n
]the set
Li
:=
{
[
v
]
aiv∈V}
of local states for
agent
ai
. Since we need to know the actual position of the
CGSi
to deﬁne the dynamics, we
also let
Ln+1
:=
V
. States of
SG
will thus be tuples in
L1×. . . ×Ln×Ln+1
. For each
v∈ G
,
let sv:= ([v]a1,...,[v]an, v)be its corresponding state in SG.
We can now deﬁne SG:= (S, R, `0), where
S:= {svv∈V},
R:= {(sv, sv0) ∃m∈MAg s.t. E(v, m) = v0}, and
`0(sv) := `(v)∪ {pv}.
To make the connection between ﬁnite plays in
G
and nodes in tree unfoldings of
SG
,
let us deﬁne, for every ﬁnite play
ρ
=
v0. . . vk
, the node
uρ
:=
sv0. . . svk
in
tSG
(
sv0
)(which
exists, by deﬁnition of
SG
and of tree unfoldings). Observe that the mapping
ρ7→ uρ
is in
fact a bijection between the set of ﬁnite plays starting in a given position
v
and the set of
nodes in tSG(sv).
Now it should be clear that giving to a propositional quantiﬁer in
QCTL∗
i
observation
oi
:=
{i}
, for
i∈
[
n
], amounts to giving him the same observation as agent
ai
. Formally,
one can prove the following lemma, simply by applying the deﬁnitions of observational
equivalence in the two frameworks:
ILemma 21. For all ﬁnite plays ρ, ρ0starting in v,ρ∼aiρ0iﬀ uρ≈oiuρ0in tSG(sv).
R. Berthon, B. Maubert and A. Murano 13
We now describe the translation
4
from
ATLsc,i
formulas to
QCTL∗
i
formulas. First we
recall the translation from [25] for the perfectinformation case.
The translation from
ATLsc
to
QCTL∗
is parameterised by a coalition
B⊂Ag
, that
conveys the set of agents who are currently bound to a strategy. It is deﬁned by induction
on Φas follows:
pB:= p¬ϕB:= ¬ϕB
ϕ∨ϕ0B:= ϕB∨ϕ0B(A)ϕB:= ϕB\A
XϕB:= XϕBϕUϕ0B:= ϕBUϕ0B
The only nontrivial case is for formulas of the form
h·A·iϕ
. For the rest of the section, we let
M={m1, . . . , ml}. Now, if A={ai1, . . . , aik}, we deﬁne
h·A·iϕB:= ∃mai1
1. . . mai1
l. . . maik
1. . . maik
lpout.
Φstrat(A)∧Φout (A∪B)∧A(Gpout →ϕA∪B),
where
Φstrat(A) := ^
a∈A
AG _
m∈M
(ma∧^
m06=m
¬m0a)
and
Φout(A) := pout ∧AG [¬pout →AX¬pout]∧
AG
pout →_
v∈V_
m∈MA
pv∧pm∧AX
_
v0∈E(v,m)
pv0↔pout
.
In Φ
out
(
A
), for
m
= (
ma
)
a∈A∈MA
, notation
pm
stands for the propositional formula
Va∈Ama
a
which characterizes the joint move
m
that agents in
A
play in
v
. Also,
E
(
v, m
)is
the set of possible next positions when the current one is vand agents in Aplay m, and it
is deﬁned as E(v, m) := {E(v, (m,m0)) m0∈MAg\A}.
The idea of this translation is the following: ﬁrst, for each agent
a∈A
and each possible
move
m∈M
, an existential quantiﬁcation on the atomic proposition
ma
“chooses” for each
ﬁnite play
ρ
of (
G, vι
)(or, equivalently, for each node
uρ
of the
tSG
(
svι
)) whether agent
a
plays move
m
in
ρ
or not. Formula Φ
strat
(
A
)ensures that each agent
a
chooses exactly one
move in each ﬁnite play, and thus that atomic propositions
ma
characterise a strategy for her.
An atomic proposition
pout
is then used to mark the paths that follow the currently ﬁxed
strategies: formula Φ
out
(
A∪B
)states that
pout
marks exactly the outcome of strategies just
chosen for agents in
A
, as well as those of agents in
B
, that were chosen previously by a
strategy quantiﬁer “higher” in Φ.
Note that we simpliﬁed slightly Φ
strat
(
A
)and Φ
out
(
A
), using the fact that unlike in [
25
],
we have assumed in our deﬁnition of
CGSi
that the set of available moves is the same for all
agents in all positions (see Footnote 2).
It is proven in [
25
] that this translation is correct, in the sense that for every
ATLsc
closed
formula
ϕ
and pointed perfectinformation concurrent game structure (
G, v
), letting
SG
be
4
Here we abuse language: the construction depends on the model
G
and is therefore not a translation in
the usual sense.
14 Decidability results for ATL with imperfect information and perfect recall
as described above but removing the local states for all agents and keeping only the
Ln+1
component, we have:
G, v =ϕiﬀ tSG(sv)=tϕ∅.
We now explain how to adapt this translation to the case of imperfect information. Observe
that the only diﬀerence between
ATL∗
sc
and
ATL∗
sc,i
is that in the latter, strategies must be
deﬁned uniformly over indistinguishable ﬁnite plays, i.e., a strategy
σ
for an agent
a
must
be such that if
ρ∼aρ0
, then
σ
(
ρ
) =
σ
(
ρ0
). To enforce that the strategies coded by atomic
propositions
ma
in
h·A·iϕB
are uniform, we use the propositional quantiﬁers with partial
observation of
QCTL∗
i
. Formally, we deﬁne a translation
fB
from
ATL∗
sc,i
to
QCTL∗
i
. It is
deﬁned exactly as the one from ATL∗
sc to QCTL∗, except for the following inductive case.
If A={ai1, . . . , aik}we let
^
h·A·iϕ
B:= ∃oi1mai1
1. . . mai1
l. . . ∃oikmaik
1. . . maik
l∃pout.
Φstrat(A)∧Φout (A∪B)∧A(Gpout →eϕA∪B),
where Φ
strat
(
A
)and Φ
out
(
A
)are deﬁned as before, and
∃pout
is a macro for
∃{1,...,n+1}pout
(see Remark 4.2).
So the only diﬀerence from the previous translation is that now, the labelling of each
atomic proposition
mai
must be
oi
uniform. This means that if two nodes
u
and
u0
in
tSG
(
svι
)
are
oi
indistinguishable, then
u
is labelled with
mai
if and only if
u0
also is. In other words,
in the strategy coded by atomic propositions
mai
, agent
ai
plays
m
in
u
if and only if she
also plays it in
u0
, and thus this strategy is uniform (recall that, by Lemma 21, observation
oi
correctly reﬂects agent
ai
’s observation in
tSG
(
svι
)). It is then clear that this translation
is correct:
G, vι= Φ iﬀ tSG(svι)=te
Φ∅.(1)
However, even if we have taken (Φ
,
(
G, vι
)) to be a hierarchical instance,
e
Φ∅
is not in the
decidable fragment
QCTL∗
i,⊂
. Indeed, with the current deﬁnition of observations
{oi}i∈[n]
,
hierarchical observation in
G
does not imply hierarchical observation in
SG
: since
oi
=
{i}
,
for
i6
=
j
it is never the case that
oi⊆oj
. Still, we note that if agent
aj
observes no more
than agent
ai
, then letting
ai
see also what agent
aj
sees does not increase her knowledge of
the situation:
ILemma 22.
If
aj4Gai
, then for all ﬁnite plays
ρ, ρ0
that start in the same position,
uρ≈oiuρ0iﬀ uρ≈oi∪ojuρ0.
Proof. Assume that aj4Gai. It is enough to see that for every pair of states sv, sv0in SG,
we have
sv≈oisv0
iﬀ
sv≈oi∪ojsv0
. The righttoleft implication is obvious: if two states
have the same
i
th and
j
th components, in particular they have the same
i
th component.
For the other direction, assume that
sv≈oisv0
. This means that [
v
]
ai
= [
v0
]
ai
, and thus
that
v∼aiv0
. Since
aj4Gai
, we also have that
v∼ajv0
, and thus that [
v
]
aj
= [
v0
]
aj
, and it
follows that sv≈oi∪ojsv0.J
In the light of this Lemma 22, we can safely redeﬁne observations as follows: for each
i∈[n], we let
o0
i:= [
jaj4Gai
oj.
R. Berthon, B. Maubert and A. Murano 15
Observe that in fact
o0
i
=
{jaj4Gai}
. Informally, a quantiﬁer with observation
o0
i
sees
what agent
ai
observes (note that
4G
is reﬂexive), as well as what agents that see no more
than aiobserve.
Let us deﬁne a new version of the translation
fB
. First, Φbeing hierarchical in
G
, for
each subformula of Φof the form
h·A·iϕ
we have that
A
is hierarchical in
G
. It is thus possible
to choose for agents in
A
an indexing
A
=
{ai1, . . . , aik}
such that for all 1
≤c < d ≤k
, we
have aic4Gaid.
Now the translation remains the same as before except for the following inductive case:
If A={ai1, . . . , aik}, where for all 1≤c < d ≤k, we have aic4Gaid, we let
^
h·A·iϕ
B:= ∃o0
i1mai1
1. . . mai1
l. . . ∃o0
ikmaik
1. . . maik
l∃pout.
Φstrat(A)∧Φout (A∪B)∧A(Gpout →eϕA∪B),
where Φstrat(A)and Φout (A)are deﬁned as before.
From Lemma 22 we have that this new translation is still correct in the sense of Equa
tion (1). In addition, for all 1≤c<d≤kwe have o0
ic⊆o0
id.
Now consider formula
e
Φ∅
. Because Φis hierarchical in
G
, for every pair of subformulas
ϕ1, ϕ2
of the form
ϕ1
=
h·A1·iϕ0
1
and
ϕ2
=
h·A2·iϕ0
2
where
ϕ2
is a subformula of
ϕ0
1
, maximal
agents of
A1
observe no more than minimal agents of
A2
. It is then easy to see that
e
Φ∅
would be hierarchical if there were not the perfectinformation quantiﬁcations on atomic
proposition
pout
that break the monotony of observations along subformulas when there are
nested strategic quantiﬁers. We explain how to remedy this last problem.
We remove altogether proposition
pout
, and we use instead the formula
ψout
(
A
)deﬁned
below to characterise which paths are in the outcome of the currentlyﬁxed strategies:
ψout(A) := G
^
v∈V^
m∈MA
pv∧pm→X_
v0∈E(v,m)
pv0
.
Clearly, this formula holds in a path
λ
of
tSG
(
svι
)marked with propositions
ma
charac
terising strategies for agents in
A
, if at each point along
λ
corresponding to some position
v
,
the next point in
λ
corresponds to a position
v0
that can be attained from
v
when agents in
A
each play the move prescribed by their current strategy. The last modiﬁcation to
fB
is
thus the following:
If A={ai1, . . . , aik}, where for all 1≤c < d ≤k, we have aic4Gaid, we let
^
h·A·iϕ
B:= ∃o0
i1mai1
1. . . mai1
l. . . ∃o0
ikmaik
1. . . maik
l.Φstrat(A)∧Aψout (A∪B)→eϕA∪B,
where Φstrat(A)is deﬁned as before.
It follows from the above considerations that this translation is still correct in the sense
of Equation (1), and one can check that
e
Φ∅
is a hierarchical
QCTL∗
i
formula. We conclude
the proof by recalling that by Theorem 17, model checking QCTL∗
i,⊂is decidable.
Concerning complexity, model checking
ATLsc
being already nonelementary [
25
], so is it
for ATLsc,i.
5 Conclusion
In this work we established new decidability results for the modelchecking problem of
ATL∗
with imperfect information and perfect recall as well as its extension with strategy context.
16 Decidability results for ATL with imperfect information and perfect recall
Should new decidable classes of multiplayer games with imperfect information be discovered,
and assuming the reasonable property of closure under initial shifting, our transfer theorem
(Theorem 6) would entail new decidability results also for
ATL∗
i
. As for
ATL∗
sc,i
, it would be
interesting to investigate whether a meaningful notion of hierarchical instances based on,
e.g., dynamic or recurring hierarchical information instead of hierarchical observation as here,
could lead to stronger decidability results.
References
1T. Agotnes, V. Goranko, and W. Jamroga. AlternatingTime Temporal Logics with Irre
vocable Strategies. In TARK, pages 15–24, 2007.
2R. Alur, T.A. Henzinger, and O. Kupferman. AlternatingTime Temporal Logic. Journal
of the ACM, 49(5):672–713, 2002.
3Raphaël Berthon, Bastien Maubert, and Aniello Murano. Quantiﬁed CTL with imperfect
information. CoRR, abs/1611.03524, 2016. arXiv:https://arxiv.org/abs/1611.03524.
4Raphael Berthon, Bastien Maubert, Aniello Murano, Sasha Rubin, and Moshe Y. Vardi.
Strategy logic with imperfect information. In LICS’17, pages 1–12. IEEE, 2017. doi:
10.1109/LICS.2017.8005136.
5Dietmar Berwanger, Anup Basil Mathew, and Marie van den Bogaard. Hierarchical in
formation patterns and distributed strategy synthesis. In ATVA’15, pages 378–393, 2015.
doi:10.1007/978331924953 7\_28.
6Thomas Brihaye, Arnaud Da Costa Lopes, François Laroussinie, and Nicolas Markey. ATL
with strategy contexts and bounded memory. In LFCS’09, pages 92–106, 2009. URL: http:
//dx.doi.org/10.1007/978354092687 0_7,doi:10.1007/9783540 926870_7.
7N. Bulling and W. Jamroga. Comparing variants of strategic ability: how uncertainty
and memory inﬂuence general properties of games. Autonomous Agents and MultiAgent
Systems, 28(3):474–518, 2014.
8E.M. Clarke and E.A. Emerson. Design and Synthesis of Synchronization Skeletons Using
BranchingTime Temporal Logic. In 81, LNCS 131, pages 52–71, 1981.
9E.M. Clarke, O. Grumberg, and D.A. Peled. Model Checking. 2002.
10 A. Da Costa, F. Laroussinie, and N. Markey. ATL with Strategy Contexts: Expressiveness
and Model Checking. In FSTTCS’10, LIPIcs 8, pages 120–132, 2010.
11 Catalin Dima, Constantin Enea, and Dimitar P. Guelev. Modelchecking an alternating
time temporal logic with knowledge, imperfect information, perfect recall and communic
ating coalitions. In GANDALF’10, pages 103–117, 2010. URL: http://dx.doi.org/10.
4204/EPTCS.25.12,doi:10.4204/EPTCS.25.12.
12 Catalin Dima and Ferucio Laurentiu Tiplea. Modelchecking ATL under imperfect in
formation and perfect recall semantics is undecidable. CoRR, abs/1102.4225, 2011. URL:
http://arxiv.org/abs/1102.4225.
13 Ronald Fagin, Joseph Y. Halpern, Yoram Moses, and Moshe Y. Vardi. Reasoning about
knowledge, volume 4. MIT press Cambridge, 1995.
14 Bernd Finkbeiner and Sven Schewe. Uniform distributed synthesis. In LICS’05, pages 321–
330, 2005. URL: http://dx.doi.org/10.1109/LICS.2005.53,doi:10.1109/LICS.2005.
53.
15 Tim French. Decidability of quantifed propositional branching time logics. In Australian
Joint Conference on Artiﬁcial Intelligence, pages 165–176. Springer, 2001.
16 Dimitar P. Guelev and Catalin Dima. Modelchecking strategic ability and knowledge
of the past of communicating coalitions. In DALT’08, pages 75–90, 2008. URL: http:
//dx.doi.org/10.1007/978354093920 7_6,doi:10.1007/9783540 939207_6.
R. Berthon, B. Maubert and A. Murano 17
17 Dimitar P. Guelev, Catalin Dima, and Constantin Enea. An alternatingtime temporal
logic with knowledge, perfect recall and past: axiomatisation and modelchecking. Journal
of Applied NonClassical Logics, 21(1):93–131, 2011. URL: http://dx.doi.org/10.3166/
jancl.21.93131,doi:10.3166/jancl.21.93131.
18 Joseph Y Halpern and Moshe Y Vardi. The complexity of reasoning about knowledge and
time. i. lower bounds. Journal of Computer and System Sciences, 38(1):195–237, 1989.
19 W. Jamroga and W. van der Hoek. Agents that Know How to Play. 63(23):185–219, 2004.
20 Piotr Kazmierczak, Thomas Ågotnes, and Wojciech Jamroga. Multiagency is coordination
and (limited) communication. In PRIMA’14, pages 91–106, 2014. URL: http://dx.doi.
org/10.1007/978331913191 7_8,doi:10.1007/9783319 131917_8.
21 O. Kupferman. Augmenting branching temporal logics with existential quantiﬁcation over
atomic propositions. In CAV’95, LNCS 939, pages 325–338. Springer, 1995.
22 O. Kupferman, P. Madhusudan, P. S. Thiagarajan, and M. Y. Vardi. Open systems in
reactive environments: Control and synthesis. In CONCUR’00, LNCS 1877, pages 92–107.
Springer, 2000.
23 Orna Kupferman and Moshe Y. Vardi. Synthesizing distributed systems. In LICS’01,
pages 389–398, 2001. URL: http://dx.doi.org/10.1109/LICS.2001.932514,doi:10.
1109/LICS.2001.932514.
24 François Laroussinie and Nicolas Markey. Quantiﬁed CTL: expressiveness and complexity.
Logical Methods in Computer Science, 10(4), 2014. URL: http://dx.doi.org/10.2168/
LMCS10(4:17)2014,doi:10.2168/LMCS10(4:17)2014.
25 François Laroussinie and Nicolas Markey. Augmenting ATL with strategy contexts. Inf.
Comput., 245:98–123, 2015. URL: http://dx.doi.org/10.1016/j.ic.2014.12.020,doi:
10.1016/j.ic.2014.12.020.
26 François Laroussinie, Nicolas Markey, and Arnaud Sangnier. ATLsc with partial observa
tion. In GandALF’15, pages 43–57, 2015. URL: http://dx.doi.org/10.4204/EPTCS.193.
4,doi:10.4204/EPTCS.193.4.
27 M. Pauly. A Modal Logic for Coalitional Power in Games. 12(1):149–166, 2002.
28 Gary Peterson, John Reif, and Salman Azhar. Lower bounds for multiplayer noncooperative
games of incomplete information. Computers & Mathematics with Applications, 41(7):957–
992, 2001.
29 Gary Peterson, John Reif, and Salman Azhar. Decision algorithms for multiplayer nonco
operative games of incomplete information. Computers & Mathematics with Applications,
43(1):179–206, 2002.
30 Amir Pnueli and Roni Rosner. Distributed reactive systems are hard to synthesize. In
FOCS’90, pages 746–757, 1990. URL: http://dx.doi.org/10.1109/FSCS.1990.89597,
doi:10.1109/FSCS.1990.89597.
31 J.P. Queille and J. Sifakis. Speciﬁcation and Veriﬁcation of Concurrent Programs in Cesar.
In 81, LNCS 137, pages 337–351, 1981.
32 Sven Schewe and Bernd Finkbeiner. Distributed synthesis for alternatingtime lo
gics. In ATVA’07, pages 268–283, 2007. URL: http://dx.doi.org/10.1007/
978354075596 8_20,doi:10.1007/9783540 755968_20.
33 P.Y. Schobbens. AlternatingTime Logic with Imperfect Recall. ENTCS, 85(2):82–93, 2004.
34 A.P. Sistla. Theoretical Issues in the Design and Ceriﬁcation of Distributed Systems. PhD
thesis, Harvard University, Cambridge, MA, USA, 1983.
35 Dirk Walther, Wiebe van der Hoek, and Michael Wooldridge. AlternatingTime Temporal
Logic with Explicit Strategies. In TARK, pages 269–278, 2007.