website/content/research/atp/truth-functional-expansion.md

3 KiB

title math
Finding Counter-Models through Truth Functional Expansions true

The best way to disprove a statement is to find a counter-example. For first order formulas, the quantifiers are over some universe of discourse \mathcal{D}.

For example, \forall x, P(x) says that for every x within the universe of discourse, x has property P. Let's say that our universe of discourse consists of two elements c_1 and c_2. Then the statement \forall x, P(x) is equivalent to P(c_1) \wedge P(c_2). That is, both elements in the universe of discourse hold property P. This equivalence is called the truth-functional expansion.

Let \mathcal{D} = \\{ c_1, \dots, c_n \\} be the universe of discourse. Then the following equivalences hold: $$ \forall x P(x) \iff P(c_1) \wedge \dots \wedge P(c_n) $$ $$ \exists x P(x) \iff P(c_1) \vee \dots \vee P(c_n)

To find a counter-example, you'll need to find a model in which the formula does not hold. A model consists of a universe of discourse and whether or not each possible predicate holds. For example, let's take a binary predicate F. For a domain of discourse of size two, the following predicates may hold: F(c_1, c_1), F(c_1, c_2), F(c_2, c_1), F(c_2, c_2). For any given model, each of those instantiations can hold or not, amounting to 2^4 = 16 possible models for a domain of discourse of size 2 with one binary predicate.

Now let's find a counter-example for an invalid formula. Consider \exists x F(x, x) \implies \forall x F(x, x). To perform a truth functional expansion, we need to first convert to prenex normal form. This works out to be \forall x \forall y (F(x, x) \implies F(y, y))

{{<details "Details">}} $$ \begin{align*} \exists x F(x, x) \implies \forall x F(x, x) &\iff \forall x (F(x, x) \implies \forall y F(y, y)) \\ &\iff \forall x \forall y (F(x, x) \implies F(y, y)) \end{align*} $$ {{}}

Let \mathcal{D} = \\{ c_1 \\}. The truth functional expansion is F(c_1, c_1) \implies F(c_1, c_1). This is a tuatlogy so the formula is valid for domain of size 1.

Now let \mathcal{D} = \\{ c_1, c_2 \\}. After removing the tautologies, the truth functional expansion works out to be (F(c_1, c_1) \implies F(c_2, c_2)) \wedge (F(c_2, c_2) \implies F(c_1, c_1))

{{<details "Details">}} $$ \forall x \forall y (F(x, x) \implies F(y, y)) \\ (\forall y (F(c_1, c_1) \implies F(y, y))) \wedge (\forall y (F(c_2, c_2) \implies F(y, y))) \\ ((F(c_1, c_1) \implies F(c_1, c_1)) \wedge (F(c_1, c_1) \implies F(c_2, c_2))) \wedge ((F(c_2, c_2) \implies F(c_1, c_1)) \wedge (F(c_2, c_2) \implies F(c_2, c_2))) \\ (F(c_1, c_1) \implies F(c_2, c_2)) \wedge (F(c_2, c_2) \implies F(c_1, c_1))

{{}}

Consider the model \\{ F(c_1, c_1), \neg F(c_1, c_2), \neg F(c_2, c_1), \neg F(c_2, c_2) \\}. This causes the left hand side of the conjunction to fail. Hence the formula is invalid and we have found a counterexample.