nLab Dedekind completion

Idea

The Dedekind completion of a linear order $L$ is a new strict linear order $\widebar{L}$ that contains suprema for all inhabited bounded subsets, and such that a supremum in $L$ is still a supremum in $\widebar{L}$.

While Dedekind completeness was traditionally described in the context of the real numbers, it can be stated for any strict linear order, although it really works best for dense? and unbounded (without top or bottom) strict linear orders. Intuitively, a strict linear order is Dedekind complete if Dedekind cuts don’t give any ‘new’ elements.

Definitions

Let $S$ be a set equipped with the structure of a dense linear order without top or bottom elements (endpoints).

Definition

A cut in $S$ is a pair of subsets $L, U \subset S$ of $S$ that satisfy the following eight properties:

1. $L$ is inhabited;
2. Dually, $U$ is inhabited;
3. If $x \lt y \in L$, then $x \in L$;
4. Dually, if $x \gt y \in U$, then $x \in U$;
5. If $x \in L$, then $x \lt y \in L$ for some $y$;
6. Dually, $x \in U$, then $x \gt y \in U$ for some $y$;
7. If $x \lt y$, then $x \in L$ or $y \in U$;
8. If $x \in L$ and $y \in U$, then $x \lt y$.

Definition

The linearly ordered set $S$ is Dedekind complete if every cut $(L,U)$ is of the form

$$L = \{x \;| \;x \lt a\},\; U = \{x \;|\; x \gt a\}$$

for some unique $a \in S$.

Definition

If $T$ is also an unbounded dense strict linear order, $S$ is Dedekind complete, and we have a universal arrow $u\colon T \to S$ in the category of strict linear orders, then $S$ (equipped with $u$) is the Dedekind completion of $T$.

The set of Dedekind cuts of rational numbers –the set of real numbers– is Dedekind complete. In fact, starting with any unbounded dense linearly ordered set $S$, the set of Dedekind cuts is isomorphic to the reals as long as $S$ is a countably infinite set.

The operation of forming the set of Dedekind cuts is idempotent, so the Dedekind completion can be constructed as the set of Dedekind cuts. More precisely, the Dedekind-complete strict linear orders form a reflective subcategory of the category of dense unbounded strict linear orders, so that Dedekind completion is a kind of completion in the abstract categorial sense.

Using open intervals

A set $F$ with a strict linear order is Dedekind complete if

• For all elements $a \in F$ and $b \in F$, the open interval $(a,b)$ is inhabited.

• For all elements $a \in F$, the upper unbounded open interval $(a,\infty)$ is inhabited.

• For all elements $a \in F$, the lower unbounded open interval $(-\infty,a)$ is inhabited.

• For all elements $a \in F$ and $b \in F$, $a \lt b$ if and only if $(b,\infty)$ is a subinterval of $(a,\infty)$

• For all elements $a \in F$ and $b \in F$, $b \lt a$ if and only if $(-\infty,b)$ is a subinterval of $(-\infty,a)$

• For all elements $a \in F$ and $b \in F$, if $a \lt b$, then $F$ is a subinterval of the union of $(a, \infty)$ and $(-\infty, b)$

• For all elements $a \in F$ and $b \in F$, the intersection of $(a,\infty)$ and $(-\infty,b)$ is a subinterval of the open interval $(a,b)$

Generalisations

In predicative constructive mathematics

In predicative constructive mathematics, subsets are large and therefore are proper classes. Thus, any Dedekind completion of a dense linear order, and in particular the rational numbers, is a proper class. Instead, predicative constructive mathematicians typically use pairs of functions $L, U: \mathbb{Q} to \Sigma$ representing the open subsets of $\mathbb{Q}$ to define predicative cuts, where $\Sigma$ is a small $\sigma$-frame with an embedding $\Sigma \to \mathrm{Prop}_\mathcal{U}$ into the large set of propositions for a universe $\mathcal{U}$. The definitions then become

Definition

Given a small $\sigma$-frame $\Sigma$ that embeds into the large set of propositions $\mathrm{Prop}_\mathcal{U}$ for a universe $\mathcal{U}$ and a linearly ordered set $S$, a cut in $S$ is a pair of functions $L, U: S \to \Sigma$ that satisfy the following eight properties:

1. There is an element $a \in S$ such that $L(a) = \top$.
2. There is an element $b \in S$ such that $U(b) = \top$.
3. If $a \lt b$ are elements of $S$ with $L(b) = \top$, then $L(a) = \top$.
4. If $a \lt b$ are elements of $S$ with $U(a) = \top$, then $U(b) = \top$.
5. If $L(a) = \top$, then there is an element of $S$ $a \lt b$ such that $L(b) = \top$.
6. If $U(b) = \top$, then there is an element of $S$ $a \lt b$ for some $U(a) = \top$.
7. If $a \lt b$ are elements of $S$, then $L(a) = \top$ or $U(b) = \top$.
8. If $L(a) = \top$ and $U(b) = \top$, then $a \lt b$.

Definition

The linearly ordered set $S$ is $\Sigma$-Dedekind complete if for every cut $(L,U)$ there exists a unique element $a \in S$ such that for all elements $x \in S$, $x \lt a$ implies $L(x) = \top$ and $x \gt a$ implies $U(x) = \top$.

Given any linearly ordered set $S$, for any two $\sigma$-frames $\Sigma$ and $\Sigma^{'}$ that embed into $\mathrm{Prop}_\mathcal{U}$ such that $\Sigma \subseteq \Sigma^{'}$, if $A$ is the $\Sigma$-Dedekind completion of $S$ and $B$ is the $\Sigma^{'}$-Dedekind completion of $S$, then $A \subseteq B$.

Strict preorders

Any paragraph containing the string ‘duisp’ is original research (although lower duisps at least are known in domain theory).

One can generalise Dedekind completion from strict linear orders to strict preorders.

Definition

A duisp (dense unbounded inhabited strict preorder) is a strictly preordered set $S$ such that, given finite (here always meaning Kuratowski-finite) subsets $F$ and $G$ of $S$ such that $x \lt z$ whenever $x \in F$ and $z \in G$, we have some $y$ in $S$ such that $x \lt y \lt z$ whenever $x \in F$ and $z \in G$.

Note that, for $S$ a strict linear order, $S$ is a duisp iff $S$ is dense, unbounded, and inhabited, hence the term ‘duisp’. (Using linearity, we may assume that $F$ and $G$ are subsingletons; then two singleton subsets is denseness, one singleton subset and one empty subset is unboundedness, and two empty subsets is inhabitedness.)

Definition

Given a duisp $S$, a cut is a pair $(L,U)$ of subsets such that:

1. $L$ is inhabited (which is a special case of (5) for $F$ the empty subset);
2. Dually, $U$ is inhabited (a special case of (6));
3. If $x \lt y \in L$, then $x \in L$;
4. Dually, if $x \gt y \in U$, then $x \in U$;
5. If $F$ is a finite subset of $L$ (which we may assume inhabited if we include (1)), then for some $x \in L$, every $y \in F$ satisfies $y \lt x$;
6. Dually, if $F$ is a finite (inhabited) subset of $U$, then for some $x \in U$, every $y \in F$ satisfies $y \gt x$;
7. If $L \lt x \lt U$ and $L \lt y \lt U$, then $x = y$;
8. If $x \in L$ and $y \in U$, then $x \lt y$.

We then define Dedekind-complete duisps and Dedekind completions of duisps the same as for dense linear orders, using this notion of cut.

A good example of a duisp is the set of rational-valued functions on any set $X$; the Dedekind completion is the set of real-valued functions on $X$.

Sections 4.31–39 of HAF do things in even more generality, but I don't really understand it yet.

One-sided Dedekind completions

At least in classical mathematics, considering only $L$ (for a lower cut) or $U$ (for an upper cut) doesn't really give us anything new for strict linear orders; we have only the technicality that $\infty \coloneqq (S,\empty)$ or $-\infty \coloneqq (\empty,S)$ is a cut (depending on the side), and we can rule even these out by simply requiring that $L$ have an upper bound or that $U$ have a lower bound.

In constructive mathematics, one-sided cuts are more general; see one-sided real number for a discussion of the case where $S$ is the strict linear order of rational numbers.

Even in classical mathematics, one-sided cuts do give us something new for strict preorders. Here, we have first more general one-sided notions of duisps: a lower duisp need only satisfy the condition of a duisp for $G$ a singleton, and an upper duisp need only satisfy the condition for $F$ a singleton. Then the lower Dedekind completion of a lower duisp is its set of lower cuts, and the upper Dedekind completion of an upper duisp is its set of upper cuts.

For example, let $X$ be a compactum and let $S$ be the strictly preordered set of continuous real-valued functions on $X$. Then $S$ is a duisp, hence both a lower and upper duisp. Its lower Dedekind completion is the set of lower semicontinuous functions on $X$ taking values in the lower reals (which classically are all either real or $\infty$); and dually on the upper side. Even working classically and ignoring the technicality of $\infty$, semicontinuous functions are much more general than continuous ones.

References

• Paul Taylor's page on Dedekind cuts

• Page 249&250 of Continuous Lattices and Domains (Google book) covers lower Dedekind completions of duisps (although not under that name) in the context of domain theory.

• Univalent Foundations Project, Homotopy Type Theory – Univalent Foundations of Mathematics (2013)

• Auke B. Booij, Extensional constructive real analysis via locators, (arxiv:1805.06781)

• Steve Vickers, “Localic Completion Of Generalized Metric Spaces I”, TAC

• Davorin Lešnik, Synthetic Topology and Constructive Metric Spaces, (arxiv:2104.10399)