Proves various first-order logic properties using SBV. The properties we prove all come from https://en.wikipedia.org/wiki/First-order_logic

>>> -- For doctest purposes only, ignore.
>>> import Data.SBV
>>> :set -XDataKinds

\(\lnot \forall x\,P(x)\Leftrightarrow \exists x\,\lnot P(x)\)

>>> let p = uninterpret "P" :: SU -> SBool
>>> prove $ sNot (qe (\(Forall x) -> p x)) .<=> qe (\(Exists x) -> sNot (p x))
Q.E.D.

\(\lnot \exists x\,P(x)\Leftrightarrow \forall x\,\lnot P(x)\)

>>> let p = uninterpret "P" :: SU -> SBool
>>> prove $ sNot (qe (\(Exists x) -> p x)) .<=> qe (\(Forall x) -> sNot (p x))
Q.E.D.

\(\forall x\,\forall y\,P(x,y)\Leftrightarrow \forall y\,\forall x\,P(x,y)\)

>>> let p = uninterpret "P" :: (SU, SV) -> SBool
>>> prove $ qe (\(Forall x) (Forall y) -> p (x, y)) .<=> qe (\(Forall y) (Forall x) -> p (x, y))
Q.E.D.

\(\exists x\,\exists y\,P(x,y)\Leftrightarrow \exists y\,\exists x\,P(x,y)\)

>>> let p = uninterpret "P" :: (SU, SV) -> SBool
>>> prove $ qe (\(Exists x) (Exists y) -> p (x, y)) .<=> qe (\(Exists y) (Exists x) -> p (x, y))
Q.E.D.

\(\forall x\,P(x)\land \forall x\,Q(x)\Leftrightarrow \forall x\,(P(x)\land Q(x))\)

>>> let p = uninterpret "P" :: SU -> SBool
>>> let q = uninterpret "Q" :: SU -> SBool
>>> prove $ (qe (\(Forall x) -> p x) .&& qe (\(Forall x) -> q x)) .<=> qe (\(Forall x) -> p x .&& q x)
Q.E.D.

\(\exists x\,P(x)\lor \exists x\,Q(x)\Leftrightarrow \exists x\,(P(x)\lor Q(x))\)

>>> let p = uninterpret "P" :: SU -> SBool
>>> let q = uninterpret "Q" :: SU -> SBool
>>> prove $ (qe (\(Exists x) -> p x) .|| qe (\(Exists x) -> q x)) .<=> qe (\(Exists x) -> p x .|| q x)
Q.E.D.

Provided \(x\) is not free in \(P\): \(P\land \exists x\,Q(x)\Leftrightarrow \exists x\,(P\land Q(x))\)

>>> let p = uninterpret "P" :: SBool
>>> let q = uninterpret "Q" :: SU -> SBool
>>> prove $ (p .&& qe (\(Exists x) -> q x)) .<=> qe (\(Exists x) -> p .&& q x)
Q.E.D.

Provided \(x\) is not free in \(P\): \(P\lor \forall x\,Q(x)\Leftrightarrow \forall x\,(P\lor Q(x))\)

>>> let p = uninterpret "P" :: SBool
>>> let q = uninterpret "Q" :: SU -> SBool
>>> prove $ (p .|| qe (\(Forall x) -> q x)) .<=> qe (\(Forall x) -> p .|| q x)
Q.E.D.

It's instructive to look at an example where the proof actually fails. Consider, for instance, an example of a merging quantifiers like we did above, except when the equality doesn't hold. That is, we try to prove the "correct" sounding, but incorrect conjecture:

\(\forall x\,P(x)\lor \forall x\,Q(x)\Leftrightarrow \forall x\,(P(x)\lor Q(x))\)

We have:

>>> let p = uninterpret "P" :: SU -> SBool
>>> let q = uninterpret "Q" :: SU -> SBool
>>> prove $ (qe (\(Forall x) -> p x) .|| qe (\(Forall x) -> q x)) .<=> qe (\(Forall x) -> p x .|| q x)
Falsifiable. Counter-example:
  P :: U -> Bool
  P U!val!2 = True
  P U!val!0 = True
  P _       = False

  Q :: U -> Bool
  Q U!val!2 = False
  Q U!val!0 = False
  Q _       = True

The solver found us a falsifying instance: Pick a domain with at least three elements. We'll call the first element U!val!2, and the second element U!val!0, without naming the others. (Unfortunately the solver picks nonintuitive names, but you can substitute better names if you like. They're just names of two distinct objects that belong to the domain \(U\) with no other meaning.)

Arrange so that \(P\) is true on U!val!2 and U!val!0, but false for everything else. Also arrange so that \(Q\) is false on these two elements, but true for everything else.

With this assignment, the right hand side of our conjecture is true no matter which element you pick, because either \(P\) or \(Q\) is true on any given element. (Actually, only one will be true on any element, but that is tangential.) But left-hand-side is not a tautology: Clearly neither \(P\) nor \(Q\) are true for all elements, and hence both disjuncts are false. Thus, the alleged conjecture is not an equivalence in first order logic.

We can use the ExistsUnique constructor to indicate a value must exists uniquely. For instance, we can prove that there is an element in E that's less than C, but it's not unique. However, there's a unique element that's less than all the elements in E:

>>> prove $ \(Exists       (me :: SE)) -> me .<= sC
Q.E.D.
>>> prove $ \(ExistsUnique (me :: SE)) -> me .<= sC
Falsifiable
>>> prove $ \(ExistsUnique (me :: SE)) (Forall e) -> me .<= e
Q.E.D.

Given a formula, skolemization produces an equisatisfiable formula that has no existential quantifiers. Instead, the existentials are replaced by uninterpreted functions. Skolemization is useful when we want to see the instantiation of nested existential variables. Interpretation for such variables will be functions of the enclosing universals.

A partial order is a reflexive, antisymmetic, and a transitive relation. We can prove these properties for relations that are checked by the isPartialOrder predicate in SBV:

\(\forall x\,R(x,x)\)

\(\forall x\,\forall y\, R(x, y) \land R(y, x) \Rightarrow x = y\)

\(\forall x\,\forall y\, \forall z\, R(x, y) \land R(y, z) \Rightarrow R(x, z)\)

>>> let r         = uninterpret "R" :: Relation U
>>> let isPartial = isPartialOrder "poR" r
>>> prove $ \(Forall x) -> isPartial .=> r (x, x)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) -> isPartial .=> (r (x, y) .&& r (y, x) .=> x .== y)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isPartial .=> (r (x, y) .&& r (y, z) .=> r (x, z))
Q.E.D.

A linear order, ensured by the predicate isLinearOrder, satisfies the following axioms:

\(\forall x\,R(x,x)\)

\(\forall x\,\forall y\, R(x, y) \land R(y, x) \Rightarrow x = y\)

\(\forall x\,\forall y\, \forall z\, R(x, y) \land R(y, z) \Rightarrow R(x, z)\)

\(\forall x\,\forall y\, R(x, y) \lor R(y, x)\)

>>> let r        = uninterpret "R" :: Relation U
>>> let isLinear = isLinearOrder "loR" r
>>> prove $ \(Forall x) -> isLinear .=> r (x, x)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) -> isLinear .=> (r (x, y) .&& r (y, x) .=> x .== y)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isLinear .=> (r (x, y) .&& r (y, z) .=> r (x, z))
Q.E.D.
>>> prove $ \(Forall x) (Forall y) -> isLinear .=> (r (x, y) .|| r (y, x))
Q.E.D.

A tree order, ensured by the predicate isTreeOrder, satisfies the following axioms:

\(\forall x\,R(x,x)\)

\(\forall x\,\forall y\, R(x, y) \land R(y, x) \Rightarrow x = y\)

\(\forall x\,\forall y\, \forall z\, R(x, y) \land R(y, z) \Rightarrow R(x, z)\)

\(\forall x\,\forall y\,\forall z\, (R(y, x) \land R(z, z)) \Rightarrow (R (y, z) \lor R (z, y))\)

>>> let r      = uninterpret "R" :: Relation U
>>> let isTree = isTreeOrder "toR" r
>>> prove $ \(Forall x) -> isTree .=> r (x, x)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) -> isTree .=> (r (x, y) .&& r (y, x) .=> x .== y)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isTree .=> (r (x, y) .&& r (y, z) .=> r (x, z))
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isTree .=> ((r (y, x) .&& r (z, x)) .=> (r (y, z) .|| r (z, y)))
Q.E.D.

A piecewise linear order, ensured by the predicate isPiecewiseLinearOrder, satisfies the following axioms:

\(\forall x\,R(x,x)\)

\(\forall x\,\forall y\, R(x, y) \land R(y, x) \Rightarrow x = y\)

\(\forall x\,\forall y\, \forall z\, R(x, y) \land R(y, z) \Rightarrow R(x, z)\)

\(\forall x\,\forall y\,\forall z\, (R(x, y) \land R(x, z)) \Rightarrow (R (y, z) \lor R (z, y))\)

\(\forall x\,\forall y\,\forall z\, (R(y, x) \land R(z, x)) \Rightarrow (R (y, z) \lor R (z, y))\)

>>> let r           = uninterpret "R" :: Relation U
>>> let isPiecewise = isPiecewiseLinearOrder "plR" r
>>> prove $ \(Forall x) -> isPiecewise .=> r (x, x)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) -> isPiecewise .=> (r (x, y) .&& r (y, x) .=> x .== y)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isPiecewise .=> (r (x, y) .&& r (y, z) .=> r (x, z))
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isPiecewise .=> ((r (x, y) .&& r (x, z)) .=> (r (y, z) .|| r (z, y)))
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> isPiecewise .=> ((r (y, x) .&& r (z, x)) .=> (r (y, z) .|| r (z, y)))
Q.E.D.

The transitive closure of a relation can be created using mkTransitiveClosure. Transitive closures are not first-order axiomatizable. That is, we cannot write first-order formulas to uniquely describe them. However, we can check some of the expected properties:

>>> let r   = uninterpret "R" :: Relation U
>>> let tcR = mkTransitiveClosure "tcR" r
>>> prove $ \(Forall x) (Forall y) -> r (x, y) .=> tcR (x, y)
Q.E.D.
>>> prove $ \(Forall x) (Forall y) (Forall z) -> r (x, y) .&& r (y, z) .=> tcR (x, z)
Q.E.D.

What's missing here is the check that if the transitive closure relates two elements, then they are connected transitively in the original relation. This requirement is not axiomatizable in first order logic.