# Bourbaki: Quantification Expansion

I'm reading Theory of Sets where existential quantification is defined:

a) $$\exists x R = (\tau_x(R)|x)R$$

and I have questions regarding the expansion for an example relation:

b) $$\exists y \exists x \in x y$$

using the definition to expand. Link bars are placed over and between symbols $$\tau_{\alpha} \rightarrow \square_{\alpha}$$. $$\alpha$$ is only used in this post to identify the initial and final point of the link bar. If the link bars are added on top of the symbols, $$\alpha$$ can be removed.

c) $$\exists y (\in (\tau_x \in \square_x y) y)$$ Use definition (a) for $$\exists y$$: $$y_{sub} = \tau_y \in (\tau_x \in \square_x \square_y) \square_y$$ and the resulting expression:

d) $$(\in (\tau_{x_1} \in \square_{x_1} (\tau_{y_1} \in (\tau_{x_2} \in \square_{x_2} \square_{y_1}) \square_{y_1})) (\tau_{y_2} \in (\tau_{x_3} \in \square_{x_3} \square_{y_2}) \square_{y_2}))$$

$$\tau_x(R)$$ represents a fully qualified object such that $$R$$ holds. If there is no such object, then $$\tau_x(R)$$ represents an arbitrary object. In the latter case, such an arbitrary object would cause $$R$$ to be false.

Question Does the final form (d) look correct? I have doubts because the outer most choice operators $$\tau_{x_1}$$ and $$\tau_{y_2}$$ may result in different values which may not hold for the expression $$\in x y$$. For instance, the sub-expression starting with $$\tau_{y_1}$$ and $$\tau_{y_2}$$ is the same, but the link bars are self contained to each sub-expression, meaning $$y$$ may hold for multiple values, which may not correspond to $$x$$.

I think either form (d) is incorrect or Bourbaki intended that duplicate $$\tau_x$$ linked expressions all choose the same value. In the later case, it seems one would have to search the expression for duplicates somehow.

Thanks

UPDATE The form (d) above is correct and if there are duplicate $$\tau$$ expressions the choice operator must choose the same for $$\square$$.

• Is this using Polish notation? And what is "$\Box_\alpha$"? – Noah Schweber Mar 15 at 0:06
• The first of the two expressions under (c) is fine, but I don't get the second one. Where did '$y=$" come from (and where did it go in (d))? In (d), I can see why $x$ has been split into $x_1$ and $x_2$, since each of the two $y$'s gets a $\tau_x$ and you need to keep them distinct. But there should be only one $\tau_y$, coming from the $\exists y$ in (c). – Andreas Blass Mar 15 at 2:57
• @Nick Where did I "mention only a single $y$"? The closest thing to that in my comment is "only one $\tau_y$". – Andreas Blass Mar 16 at 2:38
• @Nick OK, I get it. You generate a single $\tau$ expression for $y$ but then you have to substitute that for both of the $y$'s, and your subscript convention leads to different subscripts in those two places (where Bourbaki's connecting lines would look identical in both places). – Andreas Blass Mar 16 at 16:02

Hyper long comment

IMO, Bourbaki's cumbersome notation (squares and links) is only needed in order to define the usual concepts of scope and binding of a quantifier.

The $$\tau$$ operator is exactly like the set-builder one: $$\{ x \mid \varphi(x) \}$$; when applied to a formula with $$x$$ free produces a term, i.e. the name of an object.

In the same way, we have that [page 20]:

if $$B$$ is an assertion and $$x$$ a letter, then $$\tau_x(B)$$ is an object. [When used with] an assertion $$B$$ that expresses a property of an object $$x$$, it represents a distinguished object which has that property, if such an object exists.

Bourbaki uses it to define the quantifiers, and specifically: $$\exists x A := (\tau_x (A) \mid x)A$$, which we can write as: $$A[\tau_x(A) /x]$$.

What happens with a double quantification $$\exists y \exists x R(x,y)$$?

We have to keep track of the formula to which we apply the $$\tau$$ operator; thus, with the first choice, the operator is not relative to formula $$R(x,y)$$ but to formula $$S(y)=\exists x R(x,y)$$, where only $$y$$ is free.

In order to keep track of the formula $$A$$ to which the $$\tau$$ operator applies, I'll use $$\tau_x^A$$.

Thus, the first step will produce: $$\exists x R(x,\tau_y^S)$$, and the second step will produce:

$$R(\tau_x^R,\tau_y^S)$$.

In conclusion, IMO we have to consider not different variables $$x_1,x_2$$, but different formulas.

Now, going back to your example, we have that $$R$$ is $$\in$$ and thus: $$\in(\tau_x^R,\tau_y^S)$$, but what $$\tau_x^R$$ and $$\tau_y^S$$ look like?

$$\tau_y^S$$ will be $$\tau_y(\exists x \in(x,y))$$, and thus: $$(\tau_y \exists x \in (x, \square_y)$$. Does it work? I think so, because the formula $$\exists x \in (x, y)$$ has only $$y$$ free, and thus we can correctly apply the $$\tau$$ operator (that will bind it).

Up to now, we have "reduced" the original formula to $$\in (\tau_x^R, (\tau_y \exists x \in (x, \square_y))$$.

The second step will apply $$\tau$$ operator to $$\exists x R(x,\tau_y^S)$$, i.e. to $$\exists x \in (x,(\tau_y \exists x \in (x, \square_y))$$.

Formula $$R$$ is $$\in (x,(\tau_y \exists x \in (x, \square_y)))$$ and again it has only $$x$$ free: the second and third occurrences of $$x$$ are bound and thus we cannot use them in the substitution process (we can rewrite them with $$z$$ and nothing will change).

Thus, again, we have the term $$\tau_x[\in (\square_x,(\tau_y \exists x \in (x, \square_y)))]$$ that we have to replace in place of $$x$$ into $$R$$ to get:

$$\in (\tau_x[\in (\square_x,(\tau_y \exists x \in (x, \square_y)))],(\tau_y \exists x \in (x, \square_y)))$$.

Now, we have to try to read it...

According to the semantics of the $$\tau$$ operator, the term $$\tau_y \exists x \in (x, \square_y)$$ represents a distinguished objcet that has property $$\exists x \in (x, y)$$, if any. But this amounts to saying that it represents a non-empty set.

Assuming a "naive" set universe with only two sets: $$\emptyset$$ and $$\mathcal P(\emptyset)$$, the formula $$\exists x \in (x, y)$$ is satisfied by $$\mathcal P(\emptyset)$$, and thus we may choose it as value for $$\tau_y \exists x \in (x, \square_y)$$.

The second step is obvious: $$\tau_x[\in (\square_x,(\tau_y \exists x \in (x, \square_y)))]$$ represents a distinguished object that satisfy formula $$\in (x, \mathcal P(\emptyset))$$ and thus we may choose $$\emptyset$$ as value for it.

In conclusion, in our universe formula $$\exists y \exists x (x \in y)$$ is true because: $$\emptyset \in \mathcal P (\emptyset)$$.

• I had time to read your answer. Our final forms agree (see d). I derived mine from the innermost expressions as I understood the domain of $\tau$ to be the basic signs of the theory vs symbols like $\exists$. You mention $\{x:\phi(x)\}$ and $\tau_x$ which are slightly different. It's my understanding the choice op $\tau$ will represent any object if there is no such object that satisfies $R$, while the former set builder will evaluate to $\emptyset$. This is ok for $\tau$ because the logic value of $\exists x$ depends on what was chosen by $\tau$ substituted in relation $R$. – Nick Apr 11 at 14:03
• My question: In our final form $\in (\tau_x[\in (\square_x,(\tau_y \exists x \in (x, \square_y)))],(\tau_y \exists x \in (x, \square_y)))$ my confusion is the choice operation for the two duplicate forms $Z = (\tau_y \exists x \in (x, \square_y))$. Both must evaluate to the same value for the final expression to be true. Do we just assume $Z$ is the same mathematical expression with the same choice made? If so, I wonder how this works at larger scale. Does one perform comparisons of the symbols to substitute in the same choice for duplicate expressions everywhere? – Nick Apr 11 at 14:12
• @Nick - IMO, yes, the two "occurrences" must be evaluated at the same time to get the correct result. And NO, IMO this does not work at larger scale. Significantly, there is no use at all in the maybe thousands of pages of Bourbaki's works. – Mauro ALLEGRANZA Apr 11 at 15:19
• It appears a way to evaluate is to find every sequence which $\tau_x$ appears and form an equivalence class of positions. The choice operator $\tau_x$ can then choose a value for each equivalence class. The value is then substituted back into each position. Based on my example, I was unsure if there is an optimization to only search if $\in \tau (R_0) \tau (R_1)$. I'm puzzled why Bourbaki left this out, unless they planned to use higher level math for it later on. – Nick Apr 11 at 22:39