220 likes | 235 Views
LTL Model Checking. Radu Iosif (iosif@cis.ksu.edu). Linear Temporal Logic (LTL) Not exclusively for model checking Also meant for deduction ( Manna, Pnueli) So, there must be some equations involving LTL terms. Kripke Structures AP = {p, q, r, … } is a set of atomic propositions
E N D
LTL Model Checking Radu Iosif (iosif@cis.ksu.edu)
Linear Temporal Logic (LTL) • Not exclusively for model checking • Also meant for deduction(Manna, Pnueli) • So, there must be some equationsinvolving LTL terms
Kripke Structures • AP = {p, q, r, … } is a set of atomic propositions • K = <S, R, L> is a K-structure, where: • S is a finite set of states • R S x S is a transition relation • L : S P(AP)is a labeling function • w=x0, x1, … = s0, s1, … such that xi L(si) for all i 0
LTL Syntax • p AP is a formula • true is a formula • if f, g are formulae, then: • f • f g • X f • f U g are formulae
LTL Semantics Defined on Kripke structures K=(S, R, L): • K, = true always • K, = p iff = s0,s1,…and p L(s0) • K, = f iff not K, = f • K, = fg iff K, = f or K, = g • K, = X f iff = s0,s1,s2, …and K, s1,s2, … = f • K, = f U g iff k 0 . K, = g and 0 i < k K, = f
LTL Syntactic Sugar We write: • false true • fg (f g) • Fg true U g • Gf F (f) • f W g (Gf ) (f U g) (weak until) • f V g (f U g) (release)
LTL equations f U g = g (f X(f U g)) f V g = g (f X(f V g)) = (g f) (g X(f V g)) • hold for every K, assuming that is an infinite path
LTL model checking The model checking problem: • find whether a path generated by a Kripke structure K is a model for a LTL formula f (notation K, = f) To model check an LTL formula f: • first negate it then derive the negation normal form • Then build an automaton [A f] out of the negated formula • The problem is reduced to finding out whether L(A f) L(K) =
Negation normal form: example ((A U (B U C)) D) = (A U (B U C)) D = (A V (B U C)) D = (A V (B V C)) D
TABLEAU A tableau is a proof process represented by a graph, in which edges represents actually steps taken by the prover, and nodes intermediate states in the proof A node in the tableau consists of: • name = unique name of the node • incoming = set of ancestors • new = current proof obligation • old = already met proof obligation • next = proof obligation in the next state
Tableau for p U q name = Node1 incoming = {init} new = {p U q} old = {} next = {} Nodes = {}
Tableau for p U q name = Node1 incoming = {init} new = {p U q} old = {} next = {} Nodes = {} name = Node2 incoming = {init} new = {q} old = {p U q} next = {} name = Node3 incoming = {init} new = {p} old = {p U q} next = {p U q}
Tableau for p U q name = Node1 incoming = {init} new = {p U q} old = {} next = {} Nodes = {} name = Node2 incoming = {init} new = {q} old = {p U q} next = {} name = Node3 incoming = {init} new = {p} old = {p U q} next = {p U q} name = Node2’ incoming = {init} new = {} old = {q, p U q} next = {}
Tableau for p U q Nodes ={2’} name = Node2’ incoming = {init} new = {} old = {q, p U q} next = {} name = Node2’’ incoming = {Node2’} new = {} old = {} next = {}
Tableau for p U q Nodes ={2’, 2’’} name = Node2’ incoming = {init} new = {} old = {q, p U q} next = {} name = Node2’’ incoming = {Node2’, Node2’’} new = {} old = {} next = {} name = Node2’’’ incoming = {Node2’’} new = {} old = {} next = {}
Tableau for p U q name = Node1 incoming = {init} new = {p U q} old = {} next = {} Nodes = {2’, 2’’} name = Node2 incoming = {init} new = {q} old = {p U q} next = {} name = Node3 incoming = {init} new = {p} old = {p U q} next = {p U q} name = Node3’ incoming = {init} new = {} old = {p, p U q} next = {p U q}
Tableau for p U q name = Node3 incoming = {init} new = {p} old = {p U q} next = {p U q} Nodes ={2’, 2’’, 3’} name = Node3’ incoming = {init} new = {} old = {p, p U q} next = {p U q} name = Node3’’ incoming = {Node3’} new = {p U q} old = {} next = {}
name = Node3’’ incoming = {Node3’} new = {p U q} old = {} next = {} name = Node4 incoming = {Node3’} new = {q} old = {pUq} next = {} name = Node5 incoming = {Node3’} new = {p} old = {pUq} next = {pUq} Tableau for p U q Nodes ={2’, 2’’, 3’}
incoming(2’) = {init, Node3’} Tableau for p U q name = Node3’’ incoming = {Node3’} new = {p U q} old = {} next = {} name = Node4 incoming = {Node3’} new = {q} old = {pUq} next = {} name = Node5 incoming = {Node3’} new = {p} old = {pUq} next = {pUq} name = Node4’ incoming = {Node3’} new = {} old = {q, pUq} next = {}
incoming(3’) = {init, Node3’} Tableau for p U q name = Node3’’ incoming = {Node3’} new = {p U q} old = {} next = {} name = Node4 incoming = {Node3’} new = {q} old = {pUq} next = {} name = Node5 incoming = {Node3’} new = {p} old = {pUq} next = {pUq} name = Node5 incoming = {Node3’} new = {} old = {p, pUq} next = {pUq}
Resulting automaton init {p} {q} Node2’ {q} Node3’ {p} {} = true Node2’’ An LTL formula f is satisfied iff there exists an infinite path in Af containing an acceptance state infinitely often {} = true
Automata-Theoretic model checking • Invented by Vardi and Wolper in the 80’s • Implemented in SPIN in the 90’s • Language intersection problem L(A f) L(K) = is reduced to: • computing the synchronous product (A f ) x K • checking whether the synchronous product contains an acceptance cycle • if so, there exists a violation of f on some execution path of K • the model checker will show us the counterexample