Avoiding new parameter substitution when doing $\exists$ elimination.

54 Views Asked by At

I am trying to understand natural deduction in first order logic, but am confused about the $\forall$ and $\exists$ elimination rules.

My understanding:

If we know $\forall x \phi$ then we can substitute all free occurrences of $x$ in $\phi$ with a term $x_0$ because irrespective of the value of $x$ it must be that $\phi[x_0 / x]$ holds.

If we know $\exists x \phi$ then for some value of $x$ it must be that $\phi$ holds. However, we do not know which value of $x$ makes $\phi$ hold. So we substitute $x_0$ as a generic value which cannot appear anywhere before in the proof (because we can't assume anything about $x_0$). Essentially, we are giving a name to something we know must exist, $x_0$.

My Question:

After seeing the quantifier equivalences below,

$$\neg\forall x \phi \dashv \vdash \exists x \neg \phi $$ $$\neg\exists x \phi \dashv \vdash \forall x \neg \phi $$

it seems as though we can avoid substituting a new parameter when doing $\exists$ elimination by converting $\exists x \phi$ to $\neg\forall x \neg\phi$ and applying $\forall$ elimination to it.

So using this conversion we could avoid going from $\exists x \phi$ to $\phi[c / x]$ where $c$ is a new parameter, by going from $\exists x \phi$ to $\neg\forall x \neg\phi$ to $\neg(\neg\phi[x_0 / x]) \equiv \phi[x_0 / x]$ for any term $x_0$.

How can this be allowed?

2

There are 2 best solutions below

2
On BEST ANSWER

So $\exists$-elimination is used when you have $\exists x \, \phi$ available as an assumption to prove something. For example in the proof $\exists x \, \phi \vdash \alpha$ you can use it to get $\phi[t / x] \vdash \alpha$ for some term $t$.

Indeed, you can instead use the equivalence to get $\neg \forall x \, \neg\phi \vdash \alpha$, but here the leftmost negation prevents you from using the $\forall$-elimination.

0
On

Inference rules can only be applied to statements that have the relevant connective as its main connective.

This is why you cannot apply $\land \ Elim$ to a statement like $(P \land Q) \to R$ and hope to get $P \to R$ ... the main connective of $(P \land Q) \to R$ is the conditional, not the conjunction. Good thing too, because $(P \land Q) \to R$ does not logically imply $P \to R$