 Research
 Open Access
Distinguishing extended finite state machine configurations using predicate abstraction
 Khaled ElFakih^{1}Email author,
 Nina Yevtushenko^{2},
 Marius Bozga^{3} and
 Saddek Bensalem^{3}
https://doi.org/10.1186/s4041101600274
© ElFakih et al. 2016
 Received: 27 August 2015
 Accepted: 14 March 2016
 Published: 31 March 2016
Abstract
Background
Extended Finite State Machines (EFSMs) provide a powerful model for the derivation of functional tests for software systems and protocols. Many EFSM based testing problems, such as mutation testing, fault diagnosis, and test derivation involve the derivation of input sequences that distinguish configurations of a given EFSM specification.
Method and Results
In this paper, a method is proposed for the derivation of a distinguishing sequence for two explicitly given or symbolically represented, possibly infinite, sets of EFSM configurations using a corresponding FSM abstraction that is derived based on finite sets of predicates and parameterized inputs. An abstraction function that maps configurations and transitions of the EFSM specification to corresponding abstract states and transitions of the abstraction FSM is proposed. Properties of the abstraction are established along with a discussion on a proper selection of the sets of predicates and parameterized inputs used while deriving an FSM abstraction. If no distinguishing sequence is found for the current FSM abstraction then a refined abstraction is constructed by extending the sets of predicates and parameterized inputs. Simple heuristics for the selection of additional predicates are discussed and application examples are provided.
Conclusions
The work can be applied in various domains such as EFSMbased test derivation, mutation testing, and fault diagnosis.
Keywords
 Model based conformance testing
 Mutation testing
 Model transformation
 Extended finite state machines
 Predicate abstraction
 Distinguishing sequences
1 Introduction, background and related work
2 Preliminaries
In this paper, we study the distinguishability problem for configurations of an Extended Finite State Machine (EFSM) using a corresponding abstraction Finite State Machine (FSM). Accordingly, in the following, a description of these models with related notions used in the paper is provided.
2.1 Extended FSM model
Let X and Y be finite sets of inputs and outputs, R and V be finite disjoint sets of input parameters and context variables. For x ∈ X, R _{ x } ⊆ R denotes the set of input parameters and D _{ Rx } denotes the set of valuations of the parameters over the set R _{ x }. The set D _{ V } denotes the set of context variable valuations. A context variable valuation is denoted as vector v, where v ∈ D _{ V }.
The sets R _{ x } and D _{ V } can be infinite. Definitions 1 to 4 given below are related to EFSMs and they are mostly adopted from (Petrenko et al. 2004).
Definition 1: An EFSM M over X, Y, R, V with the associated valuation domains is a pair (S, T) where S is a finite nonempty set of states, including the designated initial state, and T is the set of transitions between states in S, such that each transition t ∈ T is a tuple (s, x, P, y, up, s′), where:
s and s′ are the start and final states of the transition t,
x ∈ X is the input of transition t,
y ∈ Y is the output of transition t,
P and up are functions defined over context variables V and input parameters as follows:
P : D _{ Rx } × D _{ V } → {True (or 1), False (or 0)} is the predicate (or the guard) of the transition t,
up : D _{ Rx } × D _{ V } → D _{ V } is the context update function of transition t.
A transition can have the trivial guard that is always True; in this case, we say that the transition has no guard. An input x can have no parameters; in this case, R _{ x } = ∅, D _{ Rx } = {⊥}, and the input (x, ⊥) is simply denoted by (x) or x. If x is nonparameterized then corresponding predicates are defined only over D _{ V }.
Definition 2: A context variable valuation v ∈ D _{ V } is called a context of M. A configuration of M is a tuple (s, v) where s is a state and vector v is a context.
An EFSM operates as follows. Assume that EFSM is at a current configuration (s, v) and the machine receives an input (x, p _{ x }) such that (v, p _{ x }) satisfies the guard P of an outgoing transition t = (s, x, P, op, y, up, s′). Then the machine being at (s, v), upon receiving the input (x, p _{ x }), executes the update statements of t; produces the (parameterized) output where parameter values are provided by the output parameter function op, and moves to configuration (s′, v′), where v′ = up(p _{ x }, v). Thus, a transition can be represented as (s, v)  (x, p_{ x })/(y, p_{ y }) → (s′, v′), where op(p _{ x }, v) = (y, p _{ y }). Such a transition can also be written as ((s, v), (x, p _{ x }), (y, p _{ y }), (s′, v′)). The machine usually starts from a designated initial configuration which contains the designated initial state and initial context variable valuation.

predicate complete if, for each transition t, every element in D _{ Rx } × D _{ V } evaluates to True at least one predicate of the set of all predicates guarding transitions with the same start state and input as t;

input complete if, for each pair (s, x) in S × X, there exists at least one transition leaving state s with input x;

deterministic if any two transitions outgoing from the same state with the same input have mutually exclusive predicates.
Hereafter, we consider input complete, predicate complete and deterministic EFSMs. This class of EFSMs covers almost all the known types of EFSMs studied in the context of testing (Petrenko et al. 2004).
Definition 4: Given input x and the possibly empty set D _{ Rx } of input parameter valuations, a parameterized input (or an input) is a tuple (x, p _{ x }), where p _{ x } ∈ D _{ Rx }. Recall in case R _{ x } = ∅, D _{ Rx } = {⊥}, and the input (x, ⊥) is denoted as x. A sequence of parameterized inputs is called a (parameterized) input sequence. An (parameterized) output sequence can be defined in a similar way.
Consider the machine in Fig. 1 with the set {a, b} of inputs, where a is parameterized and the domain of its input parameter D _{ Ra } is the set of all nonnegative integers and b is nonparameterized, D _{ Rb } = {⊥}. The sequence (a, 3) b b (a, 5) is an input sequence of the machine.
By definition, for a deterministic predicate and input complete EFSM M, for every state s ∈ S, input x ∈ X, (possibly empty) input parameter valuation p _{ x } ∈ D _{ Rx }, and context valuation v ∈ D _{ V }, there exists one and only one guard P defined at outgoing transitions of s under the input x, such that P(p _{ x } , v) is True. Thus, for each configuration (s, v) and each (parameterized) input (x, p _{ x }), there is only one transition from the configuration (s, v) under (x, p _{ x }) and for each (parameterized) input sequence α there exists exactly one output sequence β such that α/β is a trace at (s, v).
Definition 5: Given a parameterized input sequence that sometimes is simply called an input sequence α, configurations c of EFSM M and c′ of EFSM N are distinguishable by α if the output sequences produced by M and N at configurations c and c′ in response to α are different. The sequence α is called a sequence distinguishing c and c′.
Given the EFSM M _{1} in Fig. 1, the two configurations (2, z = 3) and (2, z = 4) are distinguishable by the input sequence (a, 1). Actually, they are distinguishable by any input sequence (a, i) with any value of i.
Definition 6: Two sets of configurations C of the EFSM M and C′ of EFSM N are distinguishable if there exists an input sequence α that distinguishes each pair (c, c′), c ∈ C and c′ ∈ C′; in this case, we say that the sequence α distinguishes sets C and C′.
Given EFSM M _{1} (Fig. 1) and the two (infinite) sets of configurations {(1, z): z is odd} and {(1, z): z is even}, any two configurations of different sets are distinguishable by the input sequence (a, 5) (a, 1). As another example, consider the sets of configurations {(s, z): z is odd} and {(s, z): z is even}, s = {1, 2}. The sequence (a, 5) (a, 1) (a, 1) distinguishes any two configurations of these two sets.
Definition 7: Given a configuration (s _{1}, v) of EFSM M, an Input/Output (I/O) sequence σ = (x _{1} , p _{ x1} )/y _{1}….. (x _{ l }, p _{ xl })/y _{ l } is a trace at configuration (s _{1}, v), written (s _{1}, v) σ >, if there exists a (s _{ l }, v _{ l }) such that σ takes the M from (s _{1}, v) to (s _{ l }, v _{ l }), written (s _{1}, v) σ  > (s _{ l }, v _{ l }). In other words, in M there exists a sequence of transitions (s _{1}, v _{1})  (x _{1} , p _{ x1} )/(y _{1}, p _{ y1})  (s _{2}, v _{2})  > (s _{2}, v _{2})  (x _{2} , p _{ x2} )/(y _{2}, p _{ y2})  (s _{3}, v _{3}) … > (s _{ l1}, v _{ l1})  (x _{ l } , p _{ xl } )/(y _{ l }, p _{ yl })  (s _{ l }, v _{ l }). The (parameterized) input projection α of a trace σ is α = σ_{↓X } = (x _{1} , p _{ x1} )…. (x _{ l } , p _{ xl } ) and the output projection of σ is the output sequence (y _{1}, p _{ y1})(y _{2}, p _{ y2}) ….. (y _{ l }, p _{ yl }).
Given EFSM M _{1} (Fig. 1), there is the sequence of transitions (1, z = 0) – b/1  (1, z = 0)  > (1, z = 0)  (a, 3)/1  (1, z = 3)  > (1, z = 3) – (a, 7)/0  (1, z = 4) at configuration (1, z = 0); thus, b/1 (a, 3)/1 (a, 7)/0 is a trace at the configuration (1, z = 0) with the input projection b (a, 3) (a, 7) and the output projection 1 1 0.
2.2 FSM model
A finite state machine (FSM) or simply a machine, is a 5tuple 〈S, s _{1}, I, O, h _{ S }〉, where S is a finite nonempty set of states with s _{1} as the initial state; I and O are finite input and output alphabets; and h _{ S } ⊆ S × I × O × S is a behavior relation. The behavior relation defines all possible transitions of the machine. Given a current state s _{ j } and an input symbol i, a 4tuple (s _{ j }, i, o, s _{ k }) ∈ h _{ S } represents a possible transition from state s _{ j } under the input i to the next state s _{ k } with the output o. A machine is deterministic if for each pair (s,i) ∈ S × I there exists at most one pair (o, s′)∈O × S such that (s, i, o, s′) ∈ h _{ S }; otherwise, the machine is nondeterministic. If for each pair (s, i) ∈ S × I there exists (o, s′) ∈ O × S such that (s, i, o, s′) ∈ h _{ S } then FSM S is said to be complete; otherwise, the machine is called partial. FSMs considered in the paper are complete but can be nondeterministic.
In a usual way, the behavior relation is extended to input sequences I* and output sequences O*. Given states s, s′ ∈ S, an input sequence α = i _{1} i _{2}…i _{ k } ∈ I* and an output sequence β = o _{1} o _{2}…o _{ k } ∈ O*, there is a transition (s, α, β, s′)∈h _{ S } if there exist states s _{1} = s, s _{2}, …, s _{ k }, s _{ k+1} = s′ such that (s _{ i }, i _{ i }, o _{ i }, s _{ i+1}) ∈ h _{ S }, i = 1, …, k. In this case, the input sequence α is a defined input sequence at state s. Given states s and s′, the defined input sequence α can take (or simply takes) the FSM S from state s to state s′ if there exists an output sequence β such that (s, α, β, s′) ∈ h _{ S } . The set out(s, α) denotes the set of all output sequences (responses) that the FSM S can produce at state s in response to a defined input sequence α, i.e. out _{ S }(s, α) = {β : ∃ s′∈S [(s, α, β, s′) ∈ h _{ S }]}. The pair α/β, β ∈ out _{ S }(s, α), is an Input/Output (I/O) sequence (or a trace) at state s. Given states s and s′, the I/O sequence α/β can take (or simply takes) the FSM S from state s to state s′ if (s, α, β, s′)∈h _{ S } . Given an I/O sequence σ = α/β, the input projection of σ, written σ _{↓I}, is α. Given two complete FSMs S = 〈S, I, O, h _{ S }〉 and R = 〈R, I, O, h _{ R }〉, state s of S and state r of R are nonseparable if for each input sequence α ∈ I* it holds that out _{ S }(s, α) ∩ out _{ R }(r, α) ≠ ∅, otherwise, states s and r are separable. For separable states s and r, there exists an input sequence α ∈ I* such that out _{ S }(s, α) ∩ out _{ R }(r, α) = ∅, i.e., the sets of output responses of FSMs S and R at states s and r to the input sequence α are disjoint. In this case, α is a separating sequence of states s and r or simply α separates s and r. Given complete FSMs S = 〈S, I, O, h 〉 and R = 〈R, I, O, h 〉, subsets S′ ⊆ S and R′ ⊆ R are separable if there exists an input sequence α that separates each pair (s, r), s ∈ S′ and r′ ∈ R′. The sequence α is a separating sequence for the sets S′ and R′.
Consider FSM M _{2} ^{D} in Fig. 5 with the set {a, b} of inputs and the set {0, 1} of outputs. The two states (1, B, P) and (1, B, ~P) are separable by the input sequence a a. The two sets of states {(0, B, P), (0, B, ~P)} and {(0, ~B, P), (0, ~B, ~P)} are separable by the sequence a a a, as in response to a a a at any state of the former set the FSM produces the output sequence 0 0 0 while at any state of the latter set the FSM produces the output sequence 0 0 1. We note that states (0, ~B, P) and (0, ~B, ~P) are nonseparable and the reader can intuitively check that by observing that under the input a the FSM reaches the same state (1, B, P) from both states producing the same output 0 and under the input b states are reachable from each other with the same output 1.
3 A method for distinguishing configurations of an EFSM using an FSM abstraction
In this section, the problem of distinguishing two disjoint sets of EFSM configurations is introduced and an FSM predicate abstraction function that maps an EFSM into an abstract FSM using given finite sets of predicates and parameterized inputs is presented. This section also includes a discussion how such sets can be selected with some application examples.
3.1 Distinguishing sets of configurations: problem definition
In this paper, we tackle the problem of distinguishing two sets of EFSM configurations where these sets can be either explicitly given or symbolically represented by predicates. As mentioned before, the latter case cannot be handled by existing related work.
Problem Definition: Given an EFSM M = (S, T) over sets X, Y, R, V, D _{ Rx }, D _{ V }, and two disjoint sets of configurations C′, C′′ ⊆ S × D _{ V } of M derive an input sequence that distinguishes the sets of configurations C′ and C′′. A particular case of this problem is deriving a distinguishing sequence for two sets where all configurations of the two sets have the same state. In this case, for disjoint sets C′ = {(s, v): v ∈ V _{ r }} and C′′ = {(s, v′): v′ ∈ V _{ t }}, it holds that the sets V _{ r } are V _{ t } are also disjoint, i.e., V _{ r } ∩ V _{ t } = ∅.
In this paper, we attempt to derive a distinguishing sequence for two disjoint sets C′ and C′′ of configurations of an EFSM M using an abstraction FSM M ^{B} of M. In the following subsection we show how such an FSM abstraction can be derived. We also establish properties of the abstraction FSM; namely, we show that a separating sequence of the sets of abstract states (when it exists) of an abstraction FSM M ^{B} that corresponds to the given sets of configurations C′ and C′′ of the EFSM M is a distinguishing sequence for the sets C′ and C′′, i.e., it distinguishes every two configurations c′ and c′′, c′ ∈ C′ and c′′ ∈ C′′.
3.2 FSM predicate abstraction of an EFSM
In this section, an FSM predicate abstraction that suites the distinguishability problem tackled in this paper is proposed. In particular, given an EFSM M and finite sets of predicates B and parameterized inputs P _{ x }, an FSM abstraction is defined using M and the given sets B and P _{ x } and some properties of the abstraction useful for the distinguishing problem are established. We also discuss how sets B and P _{ x } can be derived.
3.2.1 Defining an abstraction FSM
Definition 8: Given an EFSM M = (S, T) over X, Y, R, V, D _{ Rx }, D _{ V } and a parameterized input x, select a set P _{ x } = {p _{ x1} , p _{ x2} … p _{ xl }} such that for each state s, each configuration (s, v) and each outgoing transition from state s under the parameterized input x with the predicate P, there exists p _{ x } ∈ P _{ x } such that (v, p _{ x } ) satisfies P, written (v, p _{ x } )  = P. For EFSM M, we call the parameterized inputs (x, p _{ x }), p _{ x } ∈ P _{ x }, selected (parameterized) inputs and other (parameterized) inputs (x, p _{ x }), p _{ x } ∈ D _{ Rx }\P _{ x }, are nonselected (parameterized) inputs. In Section 3.2.3, several criteria for a proper selection of a set P _{ x } are discussed.
According to Definition 3, for each input sequence α that contains nonparameterized inputs and/or parameterized input (x, p _{ x }) such that p _{ x } ∈ P_{ x } and each configuration (s, v), there exists exactly one trace with the input projection α at the configuration (s, v).
Predicate Abstraction: Given a finite set B of k predicates B _{1}, …, B _{ k } defined over the set of variables W, we define the finite Babstraction a _{B}: D _{ W } → {(0, 1)}^{ k } with a _{B}(w) = (b _{1},…, b _{ k }), w ∈ D _{ W }, b _{ i } ∈ {0, 1}, i = 1 .. k. By definition, b _{ i } = 1 iff B _{ i }(w) is True, i.e., iff w satisfies B _{ i } (written w  = B _{ i }); correspondingly, b _{ i } = 0 iff B _{ i }(w) is False, i.e., iff valuation w does not satisfy B _{ i } (written w  ≠ B _{ i }). We note that the abstraction a _{B} is finite, since the number of Boolean vectors of length k is finite. In Section 3.2.1, several ways that can be used for the appropriate selection of the set B are suggested.
Given a set of parameterized inputs P _{ x } and a set of predicates B, an FSM abstraction M ^{B} of the given EFSM specification M is derived as follows:
Definition 9: FSM Predicate Abstraction with Selected Set of Parameterized Inputs: Given an EFSM M = (S, T) over X, Y, R, V, D _{ Rx }, D _{ V } and a finite set B of k predicates B _{1}, …, B _{ k } defined over the set V of variables, and for each (parameterized) input x a selected set of (parameterized) inputs P _{ x }, a predicate abstraction FSM M ^{B} is defined as the FSM (S ^{B}, X ^{B}, Y, h ^{B}) where
S ^{B} = S × ({0, 1}^{ k }) is the set of abstract states of the FSM,
X ^{B} = {(x, p _{ x })  x ∈ X, R _{ x } ≠ ∅, p _{ x } ∈ P _{ x }}∪{x  x ∈ X, R _{ x } = ∅} is the set of abstract inputs, and
 1.
For every configuration (s, v) and the predicate P of the outgoing transition (s, x, P, y, up, s′) from state s under a nonparameterized input x, such that v  = P and up(v) = v′, i.e., if M has a transition ((s, v), x, y, (s′, v′)), h ^{B} includes the (abstract) transition ((s, a), x, y, (s′, a′)), a = a _{B}(v) and a′ = a _{B}(v′).
 2.
Given a parameterized input x, for every p _{ x } ∈ P _{ x }, every configuration (s, v) and the predicate P of the outgoing transition (s, x, P, y, up, s′) from state s under the parameterized input x, such that (v, p _{ x })  = P and up(v) = v′, i.e. if M has a transition ((s, v), (x, p _{ x }), y, (s′, v′)), h ^{B} includes the (abstract) transition ((s, a), (x, p _{ x }), y, (s′, a′)), where a = a _{B}(v) and a′ = a _{B}(v′).
Now, also at state 0 of M _{2}, for the valuations w = 0 or w =1, with the corresponding abstract state (0, B), under the input a, the transition (0, a, (w ≤ 2), 0, up, 1), where up contains w := 1, is enabled. Executing this transition leads to state 1 where the value of w is updated to 1 and thus satisfies B. Accordingly in the abstraction FSM we have the transition ((0, B), a, 0, (0, B)). At state 0, when w = 2, which satisfies ~ B, the configuration with w = 2 maps to the abstract state (0, ~B), and a transition (0, a, (w ≤ 2), 0, up,1) under a where up contains w := 1, is enabled. Executing this transition leads to state 1 with output 0 and w is updated to 1 satisfying B. Therefore, the corresponding abstraction FSM has the transition ((0, ~B), a, 0, (1, B)). Configurations with w > 2 satisfy ~ B, and thus for the corresponding abstract state (0, ~B), under the input a the machine executes the transition (0, a, (w > 2), 0, up, 1), where up contains w := 1. Correspondingly, the abstraction FSM moves to state 1 producing the output 0. Thus, in the abstraction FSM we have the transition ((0, ~B), a, 0, (1, B)).
Now, consider state 1 of M _{2} and a corresponding abstract state (1, B). All transitions from state 2 to 1 of M _{2} do not update the value of w, and all other outgoing transitions from state 2 do not update w. Thus, all configurations from state 2 under the input a take the machine to state 1 and lead to valuations of w that still satisfy B. Thus, for state 1, there is the only abstract state (1, B) in the abstraction FSM. At state 1 of M _{2} for the transition under the input b we have the corresponding abstract transition ((1, B), b, 1, (1, B)). At state 1, under the input a, the machine produces 0 while moving to state 2 without updating the value of w, and thus, for all valuations of w, we have the abstract transition ((1, B), a, 0, (2, B). We also observe that for state 2 of M _{2}, there is the only corresponding abstract state (2, B). Moreover, at state 2 under input a, if the value of variable z is even, the machine produces 0 while moving to 2 without updating w. Thus, we have the corresponding abstract transition ((2, B), a, 0, (2, B)). When z is odd the machine produces 1 while moving to state 1 without updating w; thus we have the transition ((2, B), a, 1, (1, B)).
Given EFSM M and its abstraction M ^{B}, let (s, v) be a configuration of EFSM M. We denote A ^{B}(s, v) the abstract state (s, a) of FSM M ^{B} such that a = (B _{1}(v),…, B _{ k }(v)) (also written A (s, v) when it is not necessary to indicate the use of the set B of predicates) while using A ^{1}(s, a) (or \( {\mathrm{A}}^{{\mathrm{B}}^{\hbox{} 1}} \)(s, a)) to denote the inverse of (s, a), i.e., the set of all configurations (s, v) such that a = (B _{1}(v), …, B _{ k }(v)).
For the set of configurations C = {(s, v): s ∈ S, v ∈ D _{ V }}, we denote A(C) the set of all abstract states (s, a), ∃ (s, v) ∈ C, A(s, v) = (s, a). We write A ^{B}(C) when necessary to indicate the use of the set B. The notion, A ^{1}(C) (also written A ^{B1}(C) when necessary) for the set C ^{B} = {(s, a); a ∈ {0, 1}^{B}} denotes the union of the inverse of the union of the inverse over all (s, a) ∈ C ^{B}, i.e., the subset of configurations (s, v) of M such that A(s, v) ∈ C ^{B}.
An FSM abstraction M ^{B} is not unique as it depends on selected sets of parameterized inputs P _{ x }. However, if the set P _{ x } is set then by construction, an obtained FSM abstraction M ^{B} is complete under nonparameterized inputs and selected parameterized inputs of P _{ x } but M ^{B} can be nondeterministic. In fact, the following properties hold for an obtained FSM abstraction M ^{B}.
 (1)
For each (s, v) and (s′, v′) such that there exists a transition (s, x, P, y, up, s′), where x is a nonparameterized input, v  = P, v′ = up(v), there exists a corresponding abstract transition ((s, a _{B}(v)), x, y, (s′, a _{B}(v′))) in M ^{B}.
 (2)
For each configuration (s, v) and each p _{ x } ∈ P _{ x }, such that there exists a transition (s, x, P, y, up, s′) with parameterized input x, where (v, p _{ x } )  = P, v′ = up(v, p _{ x }), there exists a corresponding (abstract) transition ((s, a _{B}(v)), (x, p _{ x }), y, (s′, a _{B}(v′)) in h ^{B}.
In fact, for each nonparameterized input x there exists a corresponding (abstract) transition ((s, a _{B}(v)), x, y, (s′, a _{B}(v′)) in h ^{B} since the EFSM M is input complete. Furthermore, for each parameterized input x, there exists a corresponding (abstract) transition ((s, a _{B}(v)), (x, p_{ x }), y, (s′, a _{B}(v′)) in h ^{B} as M is input complete and according to the choice of P_{ x } (Definition 8). Using the induction on the trace length, the following statement can be established about the relationship between traces of the initial EFSM M and its abstraction FSM M ^{B}.
 (1)
Let σ = (x _{1} , p _{ x1} )/y _{1}… > (x _{ l } , p _{ xl } )/y _{ l } be a trace of EFSM M at configuration (s _{1}, v), i.e. (s, v) σ  > .
A trace over nonparameterized and/or some selected parameterized inputs at configuration (s _{1}, v) of EFSM M is also a trace at the corresponding abstract state A(s _{1}, v) of FSM M ^{B}. However, a trace at the configuration (s _{1}, v) that has some nonselected parameterized inputs is not a trace at the corresponding abstract state of M ^{B}, i.e., the following (a), (b), and (c) statements hold: a.
Given a trace σ at the configuration (s _{1}, v), if the input projection α = x _{1} … x _{ l } of σ is defined only over nonparameterized inputs, i.e. x _{1} ,.., x _{ l } are nonparameterized, then σ is also a trace at state A(s _{1}, v) of an FSM M ^{B},
 b.
Given a trace σ at the configuration (s _{1}, v), if the input projection α of σ is defined over selected parameterized and possibly over some nonparameterized inputs, then σ is a trace at A(s _{1}, v) of the FSM M ^{B}.
 c.
Given a trace σ at the configuration (s _{1}, v), if the input projection α of σ is defined over some nonselected parameterized and possibly over some nonparameterized inputs, then σ is not a trace at state A(s _{1}, v) of M ^{B}.
 a.
 (2)
Given a trace σ = (x _{1} , p _{ x1} )/y _{1} (x _{ l } , p _{ xl } )/y _{ l } of the abstraction FSM M ^{B} at abstract state (s _{1}, a), it can happen that for a configuration (s _{1}, v) ∈ A ^{1}(s _{1}, a) of EFSM M, σ is not a trace at (s _{1}, v).
 (3)
If σ is a trace at an abstract state (s, a) of FSM M ^{B} then at each configuration (s _{1}, v) ∈ A ^{1}(s _{1}, a) of EFSM M, there exists a single trace σ′ with the input projection α. That is, given a state (s _{1}, a) of an abstraction FSM M ^{B}, let σ = (x _{1}, p_{ x1})/y _{1}… >  (x _{ l }, p_{ xl })/y _{ l } be a trace at state (s _{1}, a) of FSM M ^{B} with the input projection α. Then at each configuration (s _{1}, v) ∈ A ^{1}(s _{1}, a) of EFSM M, there exists a trace σ′ such that the input projection of σ′ is α.
Proposition 3: Given two disjoint sets of configurations C′ and C′′ of EFSM M and their corresponding sets of abstract states S′ = A(C′) and S′′ = A(C′′) of an abstraction FSM M ^{B}. If α is a separating sequence for the sets S′ and S′′, then α is a distinguishing sequence for the sets C′ and C′′ of M. That is, α is a distinguishing sequence for each pair of configurations (s, v _{ r }) and (s′, v _{ u }) of M such that A(s, v _{ r }) ∈ S′, A(s′, v _{ u }) ∈ S′′.
Proof. Given two disjoint sets of configurations C′ and C′′ of EFSM M, let α be a separating sequence for corresponding sets of abstract states S′ = A(C′) and S′′ = A(C′′) of an abstraction FSM M ^{B}. Consider two configurations (s′_{1}, v′) ∈ C′ and (s′′_{1}, v′′) ∈ C′′. Also consider the sets Tr _{1} of all traces at state (s′_{1}, a′) = A(s′_{1}, v′) and Tr _{2} at state (s′′_{1}, a′′) = A(s′′_{1}, v′′) with the input projection α. The sets of output projections of traces of Tr _{1} and Tr _{2} are disjoint, since α is a separating sequence for states (s′_{1}, a′) and (s′′_{1}, a′′) of M ^{B}. Due to the definition of the abstraction FSM, α can have only nonparameterized inputs and/or parameterized inputs (x, p _{ x }) where p _{ x } ∈ P _{ x }. According to Definition 8, for the sets of traces Tr _{ M1} at state (s′_{1}, v′) ∈ C′ and Tr _{ M2} at (s′′_{1}, v′′) ∈ C′′ with the input projection α it holds that Tr _{ M1} ⊆ Tr _{1} and Tr _{ M2} ⊆ Tr _{2}. Since the sets Tr _{1} and Tr _{2} do not intersect, the sets of output responses at states (s′_{1}, v′) and (s′′_{1}, v′′) to α are different, and thus α is a distinguishing sequence of these two configurations.
By definition of a separating sequence, if the sets S′and S′′ intersect then there is no a separating sequence for these sets. We further describe how the set of predicates B is derived using the initial sets of configurations of C′ and C′′ of EFSM M in such a way that the corresponding abstract states S′ and S′′ are disjoint. In fact, a distinguishing sequence for C′ and C′′ derived using an FSM abstraction M ^{B} can also distinguish many other configurations of M as illustrated in Proposition 3.
3.2.2 On selecting the initial set of predicates
Given EFSM M and the sets of configurations C′ and C′′, below we describe how an initial set of predicates B can be constructed when the sets of configurations are either explicitly given or symbolically specified by predicates.
 (i)
∀ v ∈ V _{ r }\(V _{ r } ∩ V _{ t }), v  = B _{1} ∧ v  ≠ B _{2},
 (ii)
∀ v ∈ V _{ t } \ (V _{ r } ∩ V _{ t }), v  = B _{2} ∧ v  ≠ B _{1},
 (iii)
∀ v ∈ D _{ V }\(V _{ r } ∪ V _{ t }), v  ≠ B _{1} ∧ v  ≠ B _{2},
 (iv)
∀ v ∈ (V _{ r } ∩ V _{ t }), v  = B _{1} ∧ v  = B _{2}.
By construction, for every configuration v ∈ V _{ r }\(V _{ r } ∩ V _{ t }), it holds that a _{B}(s _{ p } , v) = (s _{ p }, B _{1}, ~B _{2}), for every v ∈ V _{ t }\(V _{ r } ∩ V _{ t }) it holds that a _{B}(s _{ q } , v) = (s _{ q }, ~B _{1}, B _{2}).
Moreover, for every v ∈ (V _{ r } ∩ V _{ t }), it holds that a _{B}((s, v)) = (s, B _{1}, B _{2}), and for every v ∈ D _{ V }\{V _{ r } ∪ V _{ t }} it holds that a _{B}((s, v)) = (s, ~B _{1}, ~B _{2}). Therefore, for the above predicates B _{1} and B _{2} and disjoint sets of configurations C′ ∪ C′′, the sets S′ = A(C′) and S′′ = A(C′′) are disjoint. In this case, a separating sequence for the two sets of abstract states S′and S′′ (when such a sequence exists), is a distinguishing sequence for every two configurations c′ and c′′, c′ ∈ C′, c′′ ∈ C′′.
When all the explicitly specified configurations of the sets C′ and C′′ have the same control state s, i.e., C′ = {(s, v) : v ∈ V _{ r } ⊂ D _{ V }} and C′′ = {(s, v) : v ∈ V _{ t } ⊂ D _{ V }} and V _{ r } ∩ V _{ t } = ∅, rule (iv) is not applicable. In this case, for every configuration p ∈ C′, a _{B}(p) = (s, B _{1}, ~B _{2}), and for every configuration q ∈ C′′, a _{B}(q) = (s, ~B _{1}, B _{2}), and for every configuration (s, v) in the set {(s, v) : v ∈ D _{ V }\{V _{ r } ∪ V _{ t }}}, it holds that a _{B}((s, v)) = (s, ~B _{1}, ~B _{2}). In this case, a separating sequence for the two abstract states (s, B _{1}, ~B _{2}) and (s, ~B _{1}, B _{2}) (when such a sequence exists), is a distinguishing sequence for every two configurations c′ and c′′, c′ ∈ C′, c′′ ∈ C′′.
Our work can also be applied to distinguishing two disjoint symbolically described sets of configurations C′ and C′′. In this case, the sets are not necessary finite and can be described using sets of assertions.
As a simple example, consider EFSM M _{1} in Fig. 1 and the sets C′ = {(s, z) : s ∈ S and z is odd} and C′′ = {(s, z) : z is even} where these two sets are infinite. These two sets can be represented using the predicates B and ~ B, where B is the predicate (odd(z)), i.e., B(z) = 1 if z is odd, and ~ B is the predicate (even (z)), i.e., B(z) = 0 if z is even. An FSM abstraction M _{1} ^{B} of M _{1} derived using B = {B} and a selected set of parameterized inputs is shown in Fig. 4.
Representing valuations by predicates
V _{ r } has each valuation v such that  V _{ t } has each valuation v such that  Related Predicates 

w(v) = kn  w(v) ≠ kn  B: B(v) is True iff w(v) = kn 
~B: B(v) is True iff w(v) ≠ kn  
w(v) = k  w(v) ≠ k  B: B(v) is True iff w(v) = k 
~B: B(v) is True iff w(v) ≠ k  
w(v) + z(v) = k  w(v) + z(v) ≠ k  B: B(v) is True iff w(v) + z(v) = k 
~B: B(v) is True iff w(v) + z(v) ≠ k  
w(v) = z(v)  w(v) ≠ z(v)  B: B(v) is True iff w(v) = z(v) 
~B: B(v) is True iff w(v) ≠ z(v) 
Representing Boolean valuations by predicates
V _{ r } has each valuation v such that  V _{ t } has each valuation v such that  Related Predicates 

w(v) ⊕ z(v) = 1  w(v) ⊕ z(v) = 0  B: B(v) is True iff w(v) ≠ z(v) 
~B: B(v) is True iff w(v) = z(v)  
w(v) ∨ z(v) = 1  ~w(v) ∧ ~ z(v) = 0  B: B(v) is True iff w(v) ∨ z(v) = 1 
~B: B(v) is True iff ~ w((v) ∨ z(v)) ≠ 1 
3.2.3 On selecting a set of parameterized inputs P_{ x }
An abstraction FSM M ^{B} depends on the set of predicates B as well as on selected set of parameterized inputs P _{ x }. In Section 4, we discuss how a set B can be constructed in order to increase the chance of separating designated states in the abstraction FSM M ^{B} and here we briefly discuss some criteria for selecting a set P _{ x } of parameterized inputs such that the chance of separating states in an FSM abstraction M ^{B} is increased. This can be done, for example, by reducing the degree of nondeterminism through reducing the number of nondeterministic transitions of the machine, and thus increasing the chances of having separating sequences. That is, by reducing nondeterminism of an abstraction FSM, the number of output sequences produced at two (or more) states in response to an input sequence usually decreases; and thus the possibility of producing disjoint output responses, i.e. separating sequences, in response to the input sequence increases.
 (1)
One way is to define the degree of nondeterminism as the average number of outgoing transitions under a selected input (x, p _{ x }) at an abstract state (or at a set of abstract states). Given a state s of M and a set of predicates B, an example of reducing nondeterminism is when the valuations of M are canonically partitioned (using B) into m disjoint subsets: valuations of each subset have the same abstract image while valuations of different subsets are mapped into different abstract states. A parameterized input (x, p _{ x }) is selected in such a way that the average number of outgoing transitions under (x, p _{ x }) at the m abstract states is minimized.
 (2)
One also can think of reducing nondeterminism by selecting (x, p _{ x }) such that for a maximum number m′ of the m abstract states, the number of outgoing transitions under (x, p _{ x }) at each state of the m′ states is minimized.
 (3)
Another criterion for increasing the chance of separating states in an FSM abstraction is to choose an input (x, p _{ x }) such that a large number of the m abstract states are separable by (x, p _{ x }).
In fact, one can think of applying the above criteria for the selection of (parameterized) inputs considering all states of M or only to a (properly selected) subset of these states.
Example 4. In Example 3, given EFSM M _{1} (Fig. 1), a corresponding FSM abstraction M _{1} ^{B} (Fig. 4) is derived using B = {B = odd(z)} and the set P _{ a } = {i = 1, i = 5, i = 6}. Below, we informally describe a way that can be used for the selection of a set P _{ a } based on appropriate partitioning of valuations.
The abstraction a _{B}(s, z) of all configurations (1, z) where z is odd is the abstract state (1, B) and abstraction a _{B}(s, z) of all configurations (1, z) where z is even is the abstract state (1, ~B).
Consider state 1 of M _{1} and all valuations of z that satisfy ~ B, i.e., z is even, the corresponding abstract state of all these valuations is the state (1, ~B).
Consider the outgoing transition (1, a, (1 ≤ a.i ≤ 4), 0, z = z + 1, 2) at state 1 defined under input a(i), all values a.i that satisfy the guard (1 ≤ a.i ≤ 4) of this transition, i.e., a.i = 1, 2, 3, 4, move M _{1} from state 1 to state 2 where the update statement z := z + 1 results in the value of z that satisfies B (z is odd). Thus, select one of these values, here select a.i = 1, include 1 into (initially empty) set P _{ a }. Correspondingly, the FSM abstraction has the transition ((1, ~B), (a,1), 0, (2, B)).
Consider the outgoing transition (1, a, (a.i ≥ 5), 1, z = z + a.i, 1) at state 1 defined under input a(i). All values of a.i that satisfy the guard a.i ≥ 5 are partitioned into two disjoint sets where the value a.i is odd or a.i is even. All inputs of the former set move M _{1} to state 1 with odd z value (i.e. z satisfies B) and all inputs of the latter set move M _{1} to state 1 with even z value (i.e. z satisfies ~ B). Correspondingly, from each subset, we select one value of a parameterized input and add the value to the set P _{ a }. In particular, for the former set, select a.i = 5, add 5 to P _{ a }, and from the latter set select a.i = 6, add 6 to P _{ a } and obtain P _{ a } = {1, 5, 6}. In this case, the FSM abstraction includes the transitions ((1, ~B), (a, 5), 1, (1, B)) and ((1, ~B), (a, 6), 0, (1, ~B)).
Considering all other states of M _{1} in the same way we obtain P _{ a } = {1, 5, 6} and the FSM abstraction shown in Fig. 4.
We recall that in this paper we derive an FSM abstraction using for every input x the same set P _{ x } (of selected inputs) over all states s _{1}, … s _{ n } of the given EFSM. Another way of deriving an abstraction can be carried out using different (not necessarily disjoint) sets say P ^{ s1} _{ x }, P ^{ s2} _{ x }, P ^{ sn } _{ x } for the states s _{1}, … s _{ n }. However, in this case, an FSM abstraction can be partial under the union of these sets. Thus, a trace defined over the union of these sets may not be a trace of an obtained FSM abstraction, and correspondingly, a distinguishing sequence for two sets of configurations of M may not be defined at the corresponding sets of abstract states.
4 A method for distinguishing sets of configurations of an EFSM
4.1 Method overview
 1.
Derive the set of predicates
 2.Derive an FSM abstraction of M

Consider the sets of abstract states S′ and S′′ that correspond to the sets of configurations C′ and C′′, respectively.

If S′ and S′′ are separable in the abstraction FSM by a separating sequence α

then Return α


Else

Refine the set of predicates and repeat (1) and (2) as long the (given) maximum number of allowed refinements is not reached.


 3.
Return a message “a distinguishing sequence is not determined”
4.2 Algorithm for distinguishing two sets of EFSM configurations
Due to Proposition 3, given two disjoint sets of configurations C′ and C′′ of EFSM M and their corresponding sets of abstract states S′ = A(C′) and S′′ = A(C′′) of an abstraction FSM M ^{B}, if the algorithm returns a separating sequence α for the sets S′ and S′′, then α is a distinguishing sequence for the sets C′ and C′′ of M. Thus, the following statement holds.
Theorem 1: A separating sequence α derived using Algorithm 1 (when such a sequence exists) is a distinguishing sequence for the sets C′ and C′′ of EFSM M.
An application example of Algorithm 1 is given below. More related application examples are provided in the following sections.
Example 5: As an example of Algorithm 1, consider an EFSM in Fig. 1 and two infinite sets C′ and C′′ of configuration, C′ = {(s, z): z is odd} and C′′ = {(s, z): z is even}. An FSM abstraction derived at Step 2 over the nonparameterized input b does not distinguish the sets C′ and C′′. At Step 3, the FSM abstraction M _{1} ^{B} (shown in Fig. 4) is derived using B = {B = odd(z)} and the set P _{a} for a.i = 1, a.i = 5, a.i = 6 as illustrated in Example 4. The sets of abstract states corresponding to the sets and C′ and C′′ are S′ = A ^{B}(C′) = {(1, B), (2, B)} and S′′ = A ^{B}(C′′) = {(1, ~B), (2, ~B)}. The sequence (a, 5) (a, 1) (a, 1) is a separating sequence for the sets S′ and S′′ of abstract states, and thus this sequence is a distinguishing sequence for the sets C′ and C′′.
The complexity of deriving a FSM abstraction from an EFSM depends on the types and operations on data variables used in the EFSM. In general, for a given set of predicates, the construction of the abstract FSM amounts to the computation of a finite number of transitions, where for every transition, an existential quantifier elimination involving data variables from the EFSM is performed (Graf & Saidi 1997; Bensalem et al. 1998). For instance, the general case of integer data and addition only correspond to Presburger arithmetic and has worstcase superexponential complexity (Fischer & Rabin 1974). Complexity decreases however to polynomial if only difference/octagonal constraints appear in transitions and predicates e.g., of the form ± x ± y ≤ c. If multiplication is allowed, quantifier elimination becomes undecidable and therefore automatic derivation of an FSM abstraction is not feasible in general. Obviously, if all EFSM data are interpreted on finite domains, quantifier elimination is decidable. For Booleans, quantifier elimination reduces to a SAT problem, which is known to be NPcomplete. Furthermore, the problem of deriving a separating sequence for the set of states of a deterministic FSM is PSCOMPLETE (Lee and Yannakakis (Lee & Yannakakis 1994) for complete machines, Hierons and Türker for partial machines (Hierons & Türker 2014)). For nondeterministic FSMs, Kushik and Yevtushenko (Kushik & Yevtushenko 2015) have shown that the problem of deriving a separating sequence for the set of states of a nondeterministic FSM is PSCOMPLETE. It is also known that in the worst case the length of a separating sequence even for two states is exponential w.r.t. the number of FSM states (Spitsyna et al. 2007). However, our experiments show that if a separating sequence exists then this sequence usually is very short (Spitsyna et al. 2007).
5 Predicate refinement
Given two disjoint sets of configurations C′ and C′′ and their corresponding disjoint sets of abstract states, a separating sequence (if it exists) for the abstract states distinguishes all pairs of configurations c′ and c′′, c′ ∈ C′, c′′ ∈ C′′. However, if there is no such separating sequence we can refine the abstraction, for example, by considering more predicates D = B ∪ B′ and deriving an abstraction A ^{D}. Given an EFSM M and an abstraction M ^{B} for B and P _{ x }, the predicate refinement of M ^{B} is carried out by expanding the set of predicates B to a set D and deriving an abstraction M ^{D} using the same, i.e., the unchanged, set P _{ x }. A refined FSM abstraction M ^{D} has more abstract states than M ^{B} and this usually reduces the degree of nondeterminism in M ^{D} compared with M ^{B} and thus increases the chance of the existence of a separating sequence.
5.1 Predicate refinement methods
 (1)
Predicate derived based on a selected set of valuations
Consider the sets of configurations C′ and C′′ with their corresponding sets of valuations V _{ t } and V _{ r }, respectively. Let M ^{B} be an FSM abstraction for a selected set of parameters P _{ x } and a set of predicates B = {B _{1}, B _{2}} derived, for example, as illustrated in Section 3.2.2. The initial set of predicates B = {B _{1}, B _{2}} can be refined by adding to B a predicate B ′ ∉ B that partitions valuation of some/all the sets V _{ r }, V _{ t }, D _{ V } \{V _{ r } ∪ V _{ t }}, and V _{ r } ∩ V _{ t }.
Let V ′ ⊂ D _{ V } be a selected set of valuations, construct a predicate B ′ ∉ B such that for every valuation v ∈ V ′, v  = B ′ and for every v ∈ D _{ V }\V ′, v  ≠ B ′, and derive an FSM abstraction M ^{D} using D = {B _{1}, B _{2}} ∪ {B ′}. For every configuration (s, v), s ∈ S, v ∈ D _{ V }, let a _{B}(s, v) be the abstract state (s, a), where a = (B _{1}(v),…, B _{ k }(v)), then by construction, a _{D}(s, v) = (s, a, B ′) if v  = B ′ and a _{D}(s, v) = (s, a, ~B ′) if v  ≠ B ′. As the sets of abstract states A ^{B}(C ′) and A ^{B}(C ′′) are disjoint by construction, the sets A ^{D}(C ′) and A ^{D}(C ′′) are also disjoint.
By construction, the added predicate results in an FSM abstraction M ^{D} with more abstract states than M ^{B}, and this can reduce nondeterminism in the abstraction FSM and thus, can enhance chances for deriving a separating sequence for sets A ^{D}(C ′) and A ^{D}(C ′′) that is a distinguishing sequence for C ′ and C ′′. Usually, splitting states of a current abstraction can be effective when V ′ is selected as a subset of one of the sets V _{ r }, V _{ t }, D _{ V } \{V _{ r } ∪ V _{ t }}, or V _{ r } ∩ V _{ t }.
 (2)
Predicate derived based on states of an abstraction FSM M ^{B}
Splitting valuations can also be done based on selected state(s) of an abstraction FSM M ^{B}.
Given M ^{B}, select an abstract state r = (s _{ r } , a) such that r has many outgoing transitions under a (parameterized) input (x, p _{x}). Then, considering the prototype configurations of r, i.e. the set (s _{ r }, v _{ r }) ∈ A ^{1}(s, a), construct a predicate B ′ such that valuations of the set A ^{1}(s, a) is split into disjoint sets V _{ i } and V _{ j } where in a refined FSM abstraction M ^{D}, D = B ∪ {B′}, configurations (s _{ r }, v), v ∈ V _{ i }, are mapped into an abstract state (s _{ r }, a′), and configurations (s _{ r }, v), v ∈ V _{ j }, are mapped into another abstract state (s _{ r }, a′′) and the sets of outputs of outgoing transitions under (x, p _{ x }) at (s _{ r }, a′) and at (s _{ r }, a′′) are disjoint, or the number of outgoing transitions under (x, p _{x}) at (s _{ r }, a′) and at (s _{ r }, a′′) of M ^{D} are less than those at (s _{ r }, a) of the FSM abstraction M ^{B}.
 (3)
Predicate derived based on guards of the specification EFSM
Another way of predicate refinement is to add new predicates into a given set of predicates, for example, the guards of the given EFSM specification that are defined only over context variables. These guards are special as they are part of the specification and including these guards into an abstraction predicates splits naturally (as in the specification) states in a refined abstraction and thus, reduces nondeterminism in a refined abstraction.
Consider an EFSM M where P = {P _{1},.. P _{ m }} is the set of all guards of M defined only over context variables and its abstraction M ^{B}. Consider the set D = B ∪ P of (k + m) predicates and derive a predicate abstraction M ^{D} using the same set P _{ x } that is used for deriving M ^{B}. We note that if the set P of guards is empty, we do not consider guard refinement.
Example 6. As an example of Algorithm 1 and guard refinement, consider the two configurations p = (0, w = 0) and q = (0, w = 2) of M _{2} in Fig. 2. Let B be the predicate (w < 2) and B = {B}. The corresponding FSM abstraction M _{2} ^{B} is shown in Fig. 3.
Example 7. As another example, consider the FSM M _{2} in Fig. 2 with context variables V = {w, z}. Let V _{ r } be the valuations {(w = 0, z ≥ 0), (w = 1, z ≥ 0)}, and V _{ t } be the valuations {(w = 4, z ≥ 0), (w = 6, z ≥ 0)}. Let C′ = {(0, v) : v ∈ V _{ r }} and C′′ = {(0, v) : v ∈ V _{ t }}. Consider B = {B _{1}, B _{2}} such that configurations in C′ are mapped into the abstract state S′ = (0, B _{1}, ~B _{2}) of the FSM abstraction and configurations in C′′ are mapped into the abstract S′′ = (0, ~B _{1}, B _{2}). The abstract states S′ and S′′ are nonseparable in such FSM abstraction. Refining the abstraction by considering D = B ∪ P where P is the guard even(z) of M _{2}, thus, ~P if z is odd, we obtain an FSM where the configurations in C′ are mapped into abstract states of the set S′ = {(0, B _{1}, ~B _{2}, P), (0, B _{1}, ~B _{2}, ~P)} and the configurations in C′′ are mapped into the abstract states of the set S′′ = {(0, ~B _{1}, B _{2}, P), (0, ~B _{1}, B _{2}, ~P)}. The sequence a a a is a separating sequence for the sets S′ and S′′. The obtained sequence distinguishes any two configurations c′ and c′′, c′ ∈ C′, c′′ ∈ C′′.
In general, the complexity of the refinement is tightly related to the data and operations used in EFSM. Predicate refinement is at the heart of CEGARbased (CounterExampleGuided Abstraction Refinement) automated verification (Clarke et al. 2003). In recent years, this approach has been thoroughly investigated in the context of software verification. In particular, interpolation has been proposed as a general effective technique to extract information from spurious counterexamples in order to refine the abstract model (see e.g., (Henzinger et al. 2004) for an introduction). In our context, if we consider predicate abstraction when the domains of variables are finite we get polynomial complexity by selecting appropriately the predicates of the machine for the refinement when a predicate abstraction is based on adding guards of the specification. Also, when selecting predicates for reducing the nondeterminism of the abstraction FSM, one can appropriately select a partition over the set of given valuations. We note that checking the existence of a separating sequence is based on the derivation of a socalled FSM successor tree and this tree can be considered as a counterexample when a separating sequence does not exist. For example, the tree can be used as a guide when selecting predicates for reducing the nondeterminism of an FSM abstraction. In all cases, investigating and elaborating heuristics for reducing the complexity of refining a predicate abstraction for the considered problem is an interesting direction for future research.
5.2 Relationship between distinguishing configurations of an abstraction and a refined predicate abstraction
The following propositions establish the correspondence between states of an abstraction FSM and their corresponding states in a refined abstraction FSM. The predicate refinement step of Algorithm 1 (Step3) relies on this correspondence for deriving separating sequences against a refined FSM abstraction.
Proposition 4. Given an EFSM M, two sets of configurations C′ and C′′ of EFSM M and an abstraction FSM M ^{B} of M, let an abstract FSM M ^{D} be a predicate refinement of M ^{B}. If states A ^{B}(C′) and A ^{B}(C′′) of FSM M ^{B} can be separated by an input sequence α then states A ^{D}(C′) and A ^{D}(C′′) of the abstract FSM M ^{D} can also be separated by α.
Definition 10: Given a Boolean vector a of length k, a Boolean vector a′ of length n, n ≥ k, is an extension of a if a′ = ac where c is a Boolean vector of length n  k.
 (1)
Given states (s, a′) of M ^{D} and (s, a) of M ^{B}, the set of configurations that are mapped by D into abstract state (s, a′) is a subset of the set of configurations mapped by B into abstract state (s, a) where a′ is the extension of a. That is, if a′ is an extension of a then \( {\mathrm{A}}^{{\mathrm{D}}^{\hbox{} 1}} \)(s, a′) ⊆ \( {\mathrm{A}}^{{\mathrm{B}}^{\hbox{} 1}} \)(s, a).
 (2)
The union of configurations that lead to abstract states (s, a′) where a′ is an extension of a is actually the set of configurations that lead to the abstract state (s, a). That is for each state (s, a) of M ^{B}, it holds that the set of configurations \( {\mathrm{A}}^{{\mathrm{B}}^{\hbox{} 1}} \)(s, a) is the union of the sets of configurations \( {\mathrm{A}}^{{\mathrm{D}}^{\hbox{} 1}} \)(s, a′) over all (s, a′) such that a′ is an extension of a.
6 Discussion: Limitations of the proposed work
In general, an EFSM can have parameterized outputs and the work can be adapted to EFSMs with output parameters. In this case, the values of parameterized outputs obtained when deriving an FSM abstraction are included into the FSM abstraction outputs, i.e., an abstract FSM output is the concatenation of the original EFSM output symbol and the obtained valuation of output parameters.
In practice, an EFSM specification can be partial. The work presented in this paper can be used for partial EFSM specifications that are completed by adding selfloop transitions with null outputs. However, it is worth to extend the proposed work to partial specifications considering only the defined partial behavior of the specification EFSM. We think that this can be done by using the work presented in (Kushik et al. 2014) that handles the derivation of a separating sequence using only the defined inputs of the machine.
Though this paper establishes the theoretic framework for solving the considered problem; however, there is a need to empirically assess the proposed method using many application examples. This requires; for instance, assessing the proposed work using several randomly generated EFSM specifications or EFSMs of realistic application examples (case study). For this purpose there is a need for the development and assessment of software tool that derives randomly generated EFSM specifications considering the various attributes of the EFSM model. In addition, this also requires the development of a software tool that implements the proposed method and then the use of the tool in the empirical assessment. The tool has to handle the derivation of (a) an abstraction FSM from the given EFSM specification, (b) a separating sequence for the considered sets, and (c) a refined abstraction. We think that the known verification IF tool (Bozga et al. 2004; Mounier et al. 2002) can be easily tailored to dealing with (a) when the designer provides the initial sets of predicates and parameterized inputs used by the proposed method and when predicate abstraction is carried out using the guards of the specification machine. However, in order to deal with the automatic derivation of the sets of predicates and selected inputs and to refine the abstraction based on valuations or states of an abstraction, then there is a need to implement and incorporate the heuristics given in this paper into the IF tool. The tool we have used in (Spitsyna et al. 2007) can be used for the derivation of separating sequences.
7 Conclusions and future work
Given two disjoint sets, specified possibly over (infinite) sets of Extended Finite State Machine (EFSM) configurations, a method for deriving an input sequence that distinguishes each pair of configurations of these sets is proposed. The method is based on deriving such a sequence for designated sets of states of an FSM abstraction. The FSM abstraction is derived from the given EFSM specification using selected sets of predicates and parameterized inputs. If a distinguishing sequence for the corresponding two sets of abstract states is not found using the current FSM abstraction, the abstraction is refined. A refined FSM abstraction is derived by adding more predicates or more inputs to the selected sets of predicates and inputs in order to reduce nondeterminism in the abstraction FSM. Some refinement strategies are discussed. The work can be applied in various domains such as EFSMbased test derivation, mutation testing, and fault diagnosis.
Finally, in EFSM fault diagnosis (ElFakih et al. 2003), several EFSMs, socalled fault functions, are constructed using different types of EFSM faults. In particular, from a given EFSM specification and a selected type of fault, an EFSM with the selected type of faults is derived. For instance, common types of EFSM faults (Bochmann & Petrenko 1994; ElFakih et al. 2003) include output, transfer, predicate, and assignment faults. Then, fault diagnosis is carried out by deriving tests distinguishing between these EFSM fault functions. The fault functions are specified as complete nondeterministic EFSMs. Thus, the proposed work can be used to derive diagnostic tests provided the work is extended to handle nondeterministic EFSM specifications. Our preliminary investigation shows that the work presented in this paper can be used for distinguishing two configurations of a nondeterministic EFSM. In fact, in this case, in the abstraction function (Definition 9), for a given configuration and parameterized input, there could be many enabled transitions at a state with the same parameterized input and the abstraction function takes into consideration all such transitions. However, a thorough investigation of adapting the presented work to nondeterministic EFSMs would be interesting and to the best of our knowledge no work has been done on distinguishing configurations of nondeterministic EFSMs.
Declarations
Acknowledgments
The authors would like to thank Dr. Paulo Borba and the anonymous reviewers for their help in improving the paper.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 Andrade WL, Machado PDL (2013) Generating test cases for realtime systems based on symbolic models. IEEE TSE 39(9):1216–1229Google Scholar
 Bensalem S, Bouajjani A, Loiseaux C, and Sifakis J (1992) Propertypreserving simulations. In Proc. of the 4th International Conference on Computer Aided Verification CAV ‘92, LNCS vol. 663. SpringerGoogle Scholar
 Bensalem S, Lakhnech Y and Owre S (1998) Computing abstractions of infinite state systems automatically and compositionally. In Proc. of Computer Aided VerificationCAV’98. LNCS vol. 1427, pp. 319331. SpringerGoogle Scholar
 Bochmann GV, Petrenko A (1994) Protocol testing: review of methods and relevance for software testing. In: Proc. of International Symposium on Software Testing and Analysis, Seattle., pp 109–123Google Scholar
 Bourhfir C, Abdoulhamid E, Khendek F, Dssouli R (2001) Test cases selection from SDL specifications. Comput Netw 35(6):693–708View ArticleGoogle Scholar
 Bozga M, Graf S, Ober I, and Sifakis J (2004) The IF Toolset. Revised Lectures of the 4th International School on Formal Methods for the Design of Computer, Communication and Software Systems: Real Time SFMRT 2004. LNCS vol. 3185, pp. 237267. SpringerGoogle Scholar
 Cavalli A, Gervy C, Prokopenko S (2003) New approaches for passive testing using an extended finite state machine specification. Inf Softw Technol 45:837–852View ArticleGoogle Scholar
 Cho H, Hachtel G, Jeong SW, Plessier B, Schwarz E, Somenzi F (1991) Results on the interface between formal verification and ATPG. Proc. Workshop in ComputerAided Verification Conf.(CAV’90), DIMACS Series in Discrete Mathematics and Theoretical Computer Science, vol. 3, pp. 615627Google Scholar
 Clarke E, Grumberg O, Long D (1994) Verification tools for finitestate concurrent systems. In: A Decade of Concurrency  Reflections and Perspectives. LNCS 803Google Scholar
 Clarke EM, Grumberg O, Jha S, Lu Y, Veith H (2003) Counterexampleguided abstraction refinement for symbolic model checking. J ACM 50(5):752–794MathSciNetView ArticleMATHGoogle Scholar
 Clatin M, Groz R, Phalippou M, Thummel R (1995) Two approaches linking a test Generation tool with verification techniques. Proc. IFIP Eighth Int’l Workshop Protocol Test SystemsGoogle Scholar
 Dams DR, Grumberg O, and Gerth R (1994) Abstract interpretation of reactive systems: abstractions preserving ∀CTL*, ∃CTL*, CTL*. In Proc. Of the IFIP Working Conference on Programming Concepts, Methods and CalculiGoogle Scholar
 Das S (2003) Predicate abstraction. Ph.D. Thesis. Department of Electrical Engineering, Stanford UniversityGoogle Scholar
 ElFakih K, Prokopenko S, Yevtushenko N, Bochmann VG (2003) Fault diagnosis in extended finite state machines. In Proc. of the 15th International Conference on Testing of Communicating Systems, LNCS vol. 2644, pp.197210. SpringerGoogle Scholar
 ElFakih K, Kolomeez A, Prokopenko S, Yevtushenko N (2008) Extended finite state machine based test derivation driven by user defined faults. In Proc. of the IEEE International Conference on Software Testing, Verification, and Validation ICST 2008, Lillehammer, Norway: pp. 308317Google Scholar
 Fernandez JC, Jard C, Je´ron T, Viho C (1996) An experiment in automatic generation of test suites for protocols with verification technology. Sci Comput Program 29:123–146View ArticleGoogle Scholar
 Fischer MJ, Rabin MO (1974) Superexponential complexity of Presburger arithmetic. In: Proc. of the SIAMAMS Symposium on Applied Mathematics, vol 7., pp 27–41Google Scholar
 Ghedamsi A, Bochmann G, Dssouli R (1993) Multiple fault diagnosis for finite state machines. In: Proc. of IEEE INFOCOM’93., pp 782–791Google Scholar
 Graf S, and Saidi H (1997) Construction of abstract state graphs with PVS. In Proc. of ComputerAided VerificationCAV’97. LNCS vol. 1254, pp. 7283 SpringerGoogle Scholar
 Harel D, Naamad A (1996) The STATEMATE Semantics of Statecharts. ACM Trans Softw Eng Methodol 5(4):293–333View ArticleGoogle Scholar
 Henzinger TA, Jhala R, Majumdar R, McMillan K (2004) Abstractions from Proofs. In: Proc. of the ACM SIGPLANSIGACT symposium on Principles of programming languages., pp 232–244Google Scholar
 Hierons RM, and Türker UC (2014) Distinguishing sequences for partially specified FSMs. In Proc. of NASA Formal Methods (NFM 2014). LNCS vol. 8430, pp. 6276. SpringerGoogle Scholar
 ITUT (1994) Recommendation Z.100Specification and Description Language (SDL), GenevaGoogle Scholar
 Jia Y, Harman M (2011) An anaylsis and survey of the development of mutation testing. IEEE TSE 37(5):649–678Google Scholar
 Keum C, Kang S, Ko IY, Baik J, Choi Y (2006) Generating test cases for web services using extended finite state machine. In: Proc. of Testing of Communicating Systems. LNCS vol. 3964, pp. 103–117. SpringerGoogle Scholar
 Kushik N and Yevtushenko N (2015) Describing homing and distinguishing sequences for nondeterministic finite state machines via synchronizing automata. In Proc. of International Conference Implementation and Application of Automata. LNCS vol. 9223, pp. 188198. SpringerGoogle Scholar
 Kushik N, Yevtushenko N, Cavalli AR (2014) On Testing against partial nonobservable specifications. In: Proc. of Quality of Information and Communications Technology (QUATIC 2014)., pp 230–233Google Scholar
 Lee D, Yannakakis M (1994) Testing finitestate machines: state identification and verification. IEEE Trans Comput 43(3):306–320MathSciNetView ArticleGoogle Scholar
 Long D (1993) Model Checking, Abstraction and Compositional Verification. Ph.D. Thesis. School of Computer Science, Carnegie Mellon University, CMUCS93178Google Scholar
 Merayo M, Nunez M, Rodriguez I (2008) Extending EFSMs to specify and test timed systems with action durations and timeouts. IEEE Trans Comput 57(6):835–844MathSciNetView ArticleMATHGoogle Scholar
 Mounier L, Bozga M, Graf S (2002) IF2.0: A validation environment for componentbased realtime systems. In Proc. of the Int. Conf. on ComputerAided Verification. LNCS vol. 2404, pp. 343348. SpringerGoogle Scholar
 Okun V, Black PE, Yesha Y (2002) Testing with model checker: insuring fault visibility. Proc. Int. Conf. System Science, Applied Mathematics and Computer Science, and Power Eng. Systems, pp. 1351 1356Google Scholar
 OMG (2002) Unified Modelling Language, version 1.4. OMG Standard, Nov.Google Scholar
 Petrenko A, Boroday S, Groz R (2004) Confirming configurations in EFSM Testing. IEEE Trans Softw Eng 30(1):29–42View ArticleMATHGoogle Scholar
 Ramalingom T, Thulasiraman K, Das A (2003) Context independent unique state identification sequences for testing communication protocols modelled as extended finite state machines. Comput Commun 26:1622–1633View ArticleGoogle Scholar
 Rusu V, du Bousquet L, and Jéron T (2000) An approach to symbolic test generation. In Proc. Integrated Formal Methods, pp. 338357Google Scholar
 Spitsyna N, ElFakih K, Yevtushenko N (2007) Studying the separability relation between finite state machines. Softw Test Verification Reliability 17(4):227–241View ArticleGoogle Scholar
 Starke P (1972) Abstract automata, American Elsevier, 3–419Google Scholar
 Sugeta T, Maldonado JC, Wong E (2004) Mutation testing applied to validate SDL specifications. In Proc. of the 16th International Conference on Testing of Communicating Systems. Lecture Notes in Computer Science 2978: pp. 193208Google Scholar
 Wang CJ, Koh LS, Liu MT (1994) Protocol validation tools as test case generators, Proc. Seventh Int. Workshop Protocol Test Systems, pp. 155170Google Scholar
 Wong WE, Restrepo A, Choi B (2009) Validation of SDL specifications using EFSM based test generation. Inf Softw Technol 51(11):1505–1519View ArticleGoogle Scholar