10
$\begingroup$

We work in constructive mathematics.

The sets and functions in the foundations form a Grothendieck topos, which means that all colimits exist, and in particular, that all sequential colimits exist. Given a sequence of sets $A_n$ and a sequence of functions $f_n:A_n \to A_{n + 1}$, the sequential colimit of the diagram

$$A_0 \overset{f_0}\to A_1 \overset{f_1}\to A_2 \overset{f_2}\to \ldots$$

is the set $A_\infty$ with a sequence of functions $i_n:A_n \to A_\infty$ where $i_{n + 1}(f_{n}(e)) = i_{n}(e)$ for all elements $e \in A_n$, such that for any other set $B$ and sequence of functions $j_n:A_n \to B$ where $j_{n + 1}(f_{n}(e)) = j_{n}(e)$ for all elements $e \in A_n$, there is a unique function $u_B:A_\infty \to B$ such that for all natural numbers $n$ and for all elemenets $e \in A_n$, $u_B(i_n(e)) = j_n(e)$.

Let $F$ be an Archimedean ordered field. $F$ is Cauchy complete if every Cauchy sequence in $F$ converges.

Now, let us begin with the set of rational numbers $\mathbb{Q}$. By taking the quotient set of equivalent Cauchy sequences of rational numbers, one gets the Cauchy real numbers $\mathbb{R}_C$ with a unique injective field homomorphism $\mathbb{Q} \hookrightarrow \mathbb{R}_C$. The Cauchy real numbers, constructed in this manner, cannot be proven to be Cauchy complete in constructive mathematics, however.

Now, instead of the rational numbers $\mathbb{Q}$, one could take any Archimedean ordered field $F$, and perform the same construction as above: by taking the quotient set of equivalent Cauchy sequences of elements of $F$, one gets another Archimedean ordered field which is denoted here as $\mathrm{QuotCauchy}(F)$, with a unique injective field homomorphism $F \hookrightarrow \mathrm{QuotCauchy}(F)$. If $F$ is already Cauchy complete, then the field homomorphism is a field isomorphism, but this usually cannot be proven for arbitrary Archimedean ordered fields $F$ in constructive mathematics.

This implies that one could repeatedly construct the quotient set of Cauchy sequences starting from the rational numbers, yielding a sequence of unique injective field homomorphisms between Archimedean ordered fields

$$\mathbb{Q} \hookrightarrow \mathrm{QuotCauchy}(\mathbb{Q}) \hookrightarrow \mathrm{QuotCauchy}^2(\mathbb{Q}) = \mathrm{QuotCauchy}(\mathrm{QuotCauchy}(\mathbb{Q})) \hookrightarrow \ldots$$

Is the sequential colimit of this diagram above a Cauchy complete Archimedean ordered field? If it is, is it the initial Cauchy complete Archimedean ordered field?

$\endgroup$
2
  • 5
    $\begingroup$ Great question! I would suspect that this $\omega$-long colimit is still too short: Given a sequence in the colimit, every term can be situated (non-functionally) in one of the finite stages, but there might not be a finite supremum. We probably need to iterate for a total length of $\omega_1$. However it's unclear to me which notion of $\omega_1$ we should use. Unlike with the algebraic closure, where $\omega^2$ steps suffice, perhaps there is no sequential description of the Cauchy closure (but impredicatively we can intersect all Cauchy complete subfields of the Dedekind closure). $\endgroup$ Commented Oct 3, 2023 at 23:12
  • $\begingroup$ Closely related: Does CZF prove there is a minimal cauchy completion of the rationals? $\endgroup$ Commented Jul 12, 2024 at 20:18

1 Answer 1

0
$\begingroup$

$\mathrm{QuotCauchy}$ defined above is an endofunctor on the category of Archimedean ordered fields. Thus, the question above is tantamount to asking whether the sequential limit $\mathrm{QuotCauchy}^\infty(\mathbb{Q})$ is the initial $\mathrm{QuotCauchy}$-algebra.

Mike Shulman on the HoTT Zulipchat says:

Yes, it is frequently the case that you can construct an initial $F$-algebra for an endofunctor $F$ by iterating $F$ on the empty set, or more generally the free $F$-algebra on a type $A$ by iterating $F$ on $A$, but this only works when $F$ preserves sequential colimits, so that $$F(F^\infty(A)) = F(\mathrm{colim}_n F^n(A)) = \mathrm{colim}_n F(F^n(A)) = \mathrm{colim}_n F^{n+1}(A) = F^\infty(A).$$ This is the case for $F(X)=X+1$, so you can construct the natural numbers this way. But it's not the case for $F = \mathrm{QuotCauchy}$, essentially because the definition of $\mathrm{QuotCauchy}$ involves infinite sequences of elements. That is, if you have a sequential colimit diagram $X_0 \to X_1 \to X_2 \to\cdots$ with colimit $X_\infty$, then you could have an element of $\mathrm{QuotCauchy}(X_\infty)$ arising from a sequence $(x_n)$ for which (informally) $x_n \in X_n\setminus X_{n-1}$, and then that element of $\mathrm{QuotCauchy}(X_\infty)$ won't have a limit in $X_\infty$.

In the same thread, David Warn writes:

Just to give a perhaps more down-to-earth explanation of what's going on here: let's suppose we have an ambient Cauchy complete extension of $\mathbb Q$ like the Dedekind or HoTT book reals, written $\mathbb R$. Given a subset $X$ of $\mathbb R$, we can consider the subset $FX$ of $\mathbb R$ consisting of limits of Cauchy sequences of elements of $X$. I believe this is compatible with the $F$ defined above, but this way it's perhaps easier to understand the limit, where we consider the subset $F^\infty X := \bigcup_{n=0}^\infty F^n X$ of $\mathbb R$. OK, how would we show that $F^\infty X$ is Cauchy complete? Say we have a Cauchy sequence $x_n$ in $F^\infty X$. For each $n$, we know there merely exists $i$ such that $x_n \in F^i X$. But in the absence of countable choice, it seems hopeless to look for a choice of $i$ for each $n$. This is our first problem. The second, equally severe problem is that even given a choice of $i_n$ for each $n$, it can happen that the sequence $i_n$ is unbounded. We still have to produce a single $N$ such that the limit $x_n$ is in $F^N X$, which seems hopeless.

So taking the sequential limit of the diagram $$\mathbb{Q} \hookrightarrow \mathrm{QuotCauchy}(\mathbb{Q}) \hookrightarrow \mathrm{QuotCauchy}^2(\mathbb{Q}) = \mathrm{QuotCauchy}(\mathrm{QuotCauchy}(\mathbb{Q})) \hookrightarrow \ldots$$ is not sufficient to get Cauchy completion.

$\endgroup$
6
  • $\begingroup$ Yes but iterating over $\omega_1$ (the first uncountable ordinal) should prevent this problem, right? $\endgroup$ Commented Oct 15, 2023 at 18:15
  • $\begingroup$ @MartinBrandenburg How would this resolve the choice issues raised above? Something about $L(\mathbb{R})$ modeling choice or some such? $\endgroup$ Commented Oct 15, 2023 at 18:22
  • 2
    $\begingroup$ @ Madeleine: The discussion you include here is very nice and apposite, but your conclusion is a bit stronger than it supports. I agree it certainly seems very likely that the sequential colimit is insufficient here — but actually proving that would need some further argument, e.g. a countermodel. There’s a model by Andreas Blass, showing that different countable theory has no free algebra, in Words, free algebras, and coequalisers — perhaps that might give a countermodel for this? $\endgroup$ Commented Oct 15, 2023 at 19:18
  • $\begingroup$ Has this question been resolved on the HoTT Zulip since this answer was posted? $\endgroup$ Commented Jul 7 at 13:50
  • $\begingroup$ @JamesEHanson No it hasn't. Last post on that thread in the HoTT Zulip was from October 2023. We've all moved on to different topics since then, so anybody who is reading this and interested can pick up and research this topic for themselves. $\endgroup$ Commented Aug 12 at 4:26

You must log in to answer this question.