Given an LPS:
a formula of the form
is generated for each of the summands, where \(inv()\) is the expression passed using the option --invariant. This expression is an invariant of the LPS if it holds in the initial state and all the generated formulas are tautologies.
The invariant is used to eliminate summands as follows. A formula of the form
is generated for each of the summands or for the summand indicated using the option --summand only. The tool uses a BDD based prover for expressions of sort Bool to see if the generated formula is a contradiction. If the formula is a contradiction for some summand, this summand will be eliminated from the LPS. If the formula is not a contradiction, the summand remains unchanged unless the option --simplify-all is used.
The option --simplify-all will replace the conditions of all summands by the equivalent BDD of the condition in conjunction with the invariant passed using the option --invariant. This may enable other tools, like lpsconstelm and lpsparelm, to simplify the LPS even further.
In some cases it may be useful to use an SMT solver to assist the prover. The SMT solver can further reduce BDDs by removing inconsistent paths. A specific SMT solver can be chosen using the option --smt-solver=SOLVER`. Either the SMT solver Ario or CVC3 can be used. To use one of these solvers, the directory containing the corresponding executable must be in the path.
Without using the option --no-check, lpsinvelm will first check if the given expression is an invariant of the LPS. If this is not the case, no elimination or simplification will be done. In some cases the invariant may hold even though the prover is unable to determine this fact. In cases where an expression is an invariant of the LPS, but the prover is unable to determine this, the option --no-check can be used to eliminate or simplify summands anyway. Note that this also makes it possible to eliminate or simplify summands using an expression that is not an invariant of the LPS.
The option --verbose gives insight into what the prover is doing and can be used to see if rewrite rules have to be added to the specification, in order to enable the prover to determine the invariance of an expression.
Consider a linear process specification
act a:Nat; b,c; act a, b, c; proc X(b1,b2:Bool) = b1 -> a.X(!b1,b2) + b2 ->b.X(true,b2 && b1) + (b1 && b2)->c.X(false,false); init X(false,true);
If the lineariser is applied to this process using:
$ mcrl22lps -D infile.mcrl2 outfile.lps
the resulting LPS looks like
act c,b,a; proc P(b1_X,b2_X: Bool) = b1_X -> a . P(b1_X = !b1_X) + b2_X -> b . P(b1_X = true, b2_X = b2_X && b1_X) + (b1_X && b2_X) -> c . P(b1_X = false, b2_X = false) + delta; init P(false, true);
Inspection of this linear process shows that b1_X and b2_X cannot both be true at the same time. So, we can define this in a file invariant.inv. This linear process specification has as an invariant that
!(b1_X && b2_X)
See below for a detailed definition of an invariant.
$ lpsinvelm -v -iinvariant.inv outfile.lps outfile1.lps
it is possible to check the invariant. Moreover, by default the summand with conditions that in conjunction with the invariant are false are removed. In the example above, the summand with action c is removed. Using the -l flag, the invariant is put into conjunction with the condition of each summand, and the resulting condition is simplified using the eq-BDD prover. So, applying:
$ lpsinvelm -v -l -iinvariant.inv outfile.lps outfile1.lps
yields the following:
act c,b,a; proc P(b1_X,b2_X: Bool) = if(b1_X, if(b2_X, false, true), false) -> a . P(b1_X = !b1_X) + if(b1_X, false, if(b2_X, true, false)) -> b . P(b1_X = true, b2_X = b2_X && b1_X) + if(b1_X, if(b2_X, false, true), true) -> delta; init P(false, true);
Note that the conditions now have an if-then-else structure, due to the eq-BDD prover. Also note that the summand with action c has been removed.
Sometimes, this result is unreadable or the simplifications of the conditions in conjunction with the invariant is extremely time consuming. This is for instance the case if many non-boolean data types are used. In such a case the application of the tool lpsbinary can be helpful, by replacing finite data domains by boolean data domains. Using the -e flag it is possible to add the invariants to the summands, without simplifying the summands. So, by applying:
$ lpsinvelm -v -e -iinvariant.inv outfile.lps outfile1.lps
the result becomes
act c,b,a; proc P(b1_X,b2_X: Bool) = (!(b1_X && b2_X) && b1_X) -> a . P(b1_X = !b1_X) + (!(b1_X && b2_X) && b2_X) -> b . P(b1_X = true, b2_X = b2_X && b1_X) + (!(b1_X && b2_X) && b1_X && b2_X) -> c . P(b1_X = false, b2_X = false) + !(b1_X && b2_X) -> delta; init P(false, true);
Note that the c summand is now still present.
When an invariant is being checked, but turns out to be false, then counterexamples are very helpful (use the -c flag. Counterexamples can also be presented in dot format.
When the data types that are used in a process are complex, the prover is not able to prove that the invariant is actually an invariant. This for instance happens when inequalities are used. In such a case, the flag -n can be used to skip the check that the invariant indeed satisfies the invariant properties.