Maps between real numbers are often defined by convergent series. For example, to define the exponential map, we can just prove that series $$ \sum_{n = 0}^{\infty} \frac{x^n}{n!} $$ converges, which implies that there is a map to which it converges. Can we use this approach to define (constructively) localic maps between localic reals? If not, how do we define maps like these?
-
1$\begingroup$ An alternative to my answer that works to define any classical functions is to use uniform continuity and density of the rationals in the locale $\mathbb{R}$. If you have a function $f$ defined on Dedekind real numbers (or just rational number) with value in $\mathbb{R}$ (or more generally in any complete metric locale) which is locally uniformly continuous, then it admit a unique continuous extension to the locale $\mathbb{R}$. The constructive theory of complete metric locale is developed there sciencedirect.com/science/article/pii/S0001870816001006 $\endgroup$– Simon HenryCommented Jul 16, 2022 at 11:45
1 Answer
Here is a fairly general methods for this sort of thing :
Step 1) We give a constructive proof that for each (Dedekind) real $x$, the serie $\sum \frac{x^n}{n!}$ converge. We define $exp(x)$ as the limit.
Step 2) We show that $x \mapsto exp(x)$ is a geometric construction. That is if $f : \mathcal{E} \to \mathcal{S}$ is a geometric morphism and $x$ is a Dedekind real in $\mathcal{S}$ then $exp(f^*x) = f^*exp(x)$.
For limits of series this is relatively easy to do : The key point is that quantification over $\mathbb{N}$ is well behaved when it comes to pullback along a geometric morphism: if $P(n)$ denotes some proposition in $\mathbb{S}$ indexed by the NNO (so essentially a subobject of $\mathbb{N}_S$) then $$f^*(\exists n \in \mathbb{N}, P(n)) = \exists n \in \mathbb{N}, f^*(P(n))$$ and $$f^*(\forall n \in \mathbb{N}, P(n)) \Rightarrow \forall n \in \mathbb{N}, f^*(P(n))$$ and as the definition of limit can be written using only quantification over $\mathbb{N}$ that's enough.
Step 3) Just use the Yoneda lemma : the locale $\mathbb{R}$ represent the functor on the category of locale that sends a locale $X$ to the set of Dedekind real in $Sh(X)$, and the two points above shows that $exp$ is a natural transformation of this functor.
I think Step 2 alway work for things defined as limits of a series or sequence (at least I can't imagine a situation where it doesn't), and step 3) is completely formal, so in general, the only things you need to do to define this kind of functor is Step 1) (though you should quickly check step 2).
-
$\begingroup$ I don't see that $f^*$ respects universal quantification over $\mathbb N$. Suppose $f:\text{Sets}\to\text{Sh}(\mathbb R)$ is the point $0\in\mathbb R$, and suppose the truth value of $P(n)$ is the open interval $(-1/n,1/n)$. Since all these intervals contain 0, each $f^*(P(n))$ is true. But the truth value of $(\forall x\in\mathbb N)\,P(n)$ is the empty subset of $\mathbb R$, and so $f^*$ of it is false. What am I missing here? $\endgroup$ Commented Jul 16, 2022 at 14:57
-
1$\begingroup$ @AndreasBlass : I've only put an implication in the case of universal quantification, not an equality. But unless I've missed something, this is enough to show that $f^*$ preserves convergence (in the sense that if something converge in the base topos then $f^*$ of it also converge, not in the sense that the proposition "$U_n$ converge to $x$" is preserved). $\endgroup$ Commented Jul 16, 2022 at 15:20
-
$\begingroup$ Thanks. I had missed the one-way implication. $\endgroup$ Commented Jul 16, 2022 at 15:22