Self-checked by the writing AI.

Profinite spaces and finite-stage maps

Written by GPT-6.1 Sol (OpenAI), reasoning effort Ultra, September 2026. Self-checked; no independent review. Public domain (CC0).

A finite observation of an infinite sequence depends on finitely much information. For inverse systems of finite sets, this idea has a categorical form: continuous maps between their limits are exactly formal pro-morphisms. A transition need not be surjective. Some points at a finite stage can disappear from every compatible infinite family, and equality of maps must account for those disappearing points.

Read Ind-objects through their elements, Section 5 for the opposite-category convention, Formal colimits and test objects, Section 2 for formal Hom and composition, and Change an index, Section 2 for cofinal reindexing with its full cocone factor. We retain two proved dependencies: Stacks, Lemma 4.21.7, on nonempty limits of finite nonempty sets, and Stacks, Lemma 5.22.2, on the topological characterization and finite-partition presentation of profinite spaces. The proof below supplies the morphism comparison, including systems with non-surjective maps and empty stages.

1. Formal systems and their topological realization

Fix an indexing universe. Let \(\mathsf{FinSet}\) be its finite sets and \(\mathsf{Stone}\) its compact Hausdorff totally disconnected spaces, with continuous maps. The empty space is allowed. Give each finite set the discrete topology.

By Stacks, Lemma 4.21.5, any small cofiltered diagram has a cofinal presentation as an inverse system over a small directed poset. This replacement preserves limits in topological spaces and preserves the formal pro-object. For the latter assertion, apply ordinary cofinality to the filtered colimits \(\operatorname{colim}_i\operatorname{Hom}(X_i,T)\) describing its covariant functor, then take the opposite category. We can therefore work with directed posets without restricting which pro-objects occur.

Write an inverse system as \((X_i,p_{ki})\), where

\[ p_{ki}:X_k\longrightarrow X_i\quad(k\geq i), \qquad p_{ii}=1, \qquad p_{\ell i}=p_{ki}p_{\ell k}. \]

Its realization is the topological limit

\[ L_X=\lim_i X_i, \qquad \pi_i:L_X\longrightarrow X_i. \tag{1.1} \]

The retained topological theorem puts \(L_X\) in \(\mathsf{Stone}\). Its topology is the subspace topology from \(\prod_i X_i\). In particular finite-coordinate cylinders give a neighborhood basis. Directedness lets us replace any finite collection of coordinate conditions by a condition at one common later coordinate. Consequently the sets \(\pi_i^{-1}(\{a\})\) form a clopen basis, after empty members are discarded.

The formal Hom formula is

\[ \operatorname{Hom}_{\operatorname{Pro}(\mathsf{FinSet})}(X,Y) =\lim_j\operatorname{colim}_i\operatorname{Map}(X_i,Y_j). \tag{1.2} \]

In the inner colimit, moving to a later source stage means precomposing with \(p_{ki}\). The outer limit imposes compatibility with the target transitions. We use this existing formula and its composition rule; Section 3 will identify it with continuous maps.

2. Which finite-stage points survive?

Define the stable image at stage \(i\) by

\[ S_i=\bigcap_{k\geq i}p_{ki}(X_k)\subset X_i. \tag{2.1} \]

Lemma 2.1. The projection image \(\pi_i(L_X)\) is \(S_i\). For each \(i\), there is a \(k\geq i\) such that \(p_{ki}(X_k)=S_i\).

Proof. A compatible family has its \(i\)-coordinate in every later image, giving \(\pi_i(L_X)\subset S_i\). Conversely, take \(a\in S_i\). On the directed tail \(k\geq i\), the sets

\[ F_k=\{x\in X_k:p_{ki}(x)=a\} \]

are finite and nonempty. Their transitions restrict to an inverse system, since \(p_{\ell i}=p_{ki}p_{\ell k}\). The retained nonempty-limit lemma gives a compatible family in these fibers. The tail is cofinal, so it determines a point of \(L_X\) with \(i\)-coordinate \(a\). This proves the first assertion without assuming surjective transitions in the original system.

For the second, every \(a\in X_i\setminus S_i\) is absent from some image \(p_{k_a i}(X_{k_a})\). There are only finitely many such points. Choose one common upper bound \(k\) of \(i\) and these \(k_a\). Its image contains no point outside \(S_i\), while the intersection (2.1) is contained in that image. They are equal. If there are no excluded points, take \(k=i\). \(\square\)

This is stabilization of the image at a fixed stage. It does not say the objects or transitions eventually become constant.

We will also use the empty case of the retained nonempty-limit lemma:

\[ \begin{gathered} L_X=\varnothing\\ \Longleftrightarrow\\ X_i=\varnothing\text{ for some }i. \end{gathered} \tag{2.2} \]

The reverse implication follows from projection to that stage. For the forward implication, if every finite stage were nonempty, the cited lemma would give a point of the limit. Once an empty stage occurs, every stage in its directed tail is empty, because there is no map from a nonempty set to the empty set.

3. A finite target detects a finite stage

For a finite discrete set \(T\), stage maps yield continuous maps by composition with \(\pi_i\). They give a canonical comparison

\[ \begin{aligned} \gamma_T:\ &\operatorname{colim}_i\operatorname{Map}(X_i,T)\\ &\longrightarrow\operatorname{Cont}(L_X,T). \end{aligned} \tag{3.1} \]

Theorem 3.1. The comparison (3.1) is bijective, naturally in \(T\). This includes \(T=\varnothing\) and does not require the source transitions to be surjective.

Proof: surjectivity. First suppose \(L_X\) and \(T\) are nonempty, and let \(f:L_X\to T\) be continuous. Each point has a single-coordinate cylinder neighborhood on which \(f\) is constant: use openness of its fiber, the cylinder basis, and a common coordinate for the finitely many conditions defining a basic neighborhood. Compactness gives finitely many such cylinders covering \(L_X\). Choose a common later index \(k\) for their coordinates.

If two points have the same \(k\)-coordinate, they belong to the same members of this cover. One such cylinder contains the first point, hence also the second, and \(f\) has the same value at both. Thus \(f\) factors through a map

\[ \bar f:S_k=\pi_k(L_X)\longrightarrow T. \]

Extend \(\bar f\) to \(X_k\) by assigning any fixed element of \(T\) to the points outside \(S_k\). This stage map represents \(f\) in (3.1). The extension is used only for existence; injectivity below will remove its arbitrary choices.

If \(L_X\) is empty and \(T\) is nonempty, any constant stage map represents its unique map to \(T\). If \(T\) is empty, a continuous map exists only when \(L_X\) is empty. Equation (2.2) then supplies an empty stage, whose unique map to \(T\) represents it. This handles every case in which the right side has an element.

Proof: injectivity. Suppose \(a:X_i\to T\) and \(b:X_j\to T\) give the same continuous map. Move them to a common stage \(k\), obtaining \(a p_{ki}\) and \(b p_{kj}\). Equality on \(L_X\) says that these maps agree on \(S_k\), by Lemma 2.1. Choose \(\ell\geq k\) with \(p_{\ell k}(X_\ell)=S_k\). Their composites from \(X_\ell\) therefore agree on the whole set. This is exactly equality of their classes in the filtered colimit. The argument also applies to empty stages and targets whenever representatives exist.

Postcomposition with a map of finite targets commutes with the construction of (3.1), proving naturality. \(\square\)

There is no assertion that a representative is uniquely determined on all of \(X_k\). It is determined on the surviving points, and a further transition makes those the only points relevant to its class.

4. The category equivalence

For a second inverse system \((Y_j)\), the topological limit property gives

\[ \operatorname{Cont}(L_X,L_Y) \simeq\lim_j\operatorname{Cont}(L_X,Y_j). \tag{4.1} \]

Indeed a compatible family of coordinate maps defines a unique set map into the limit. It is continuous because every coordinate map is continuous and the limit has the subspace topology from the product. This includes empty spaces: the usual limit property has no nonemptiness hypothesis.

Theorem 4.1. Taking topological limits defines an equivalence

\[ R:\operatorname{Pro}(\mathsf{FinSet})\simeq\mathsf{Stone}. \tag{4.2} \]

Proof. Combining (1.2), Theorem 3.1 and (4.1) gives bijections

\[ \begin{aligned} \operatorname{Hom}_{\operatorname{Pro}(\mathsf{FinSet})}(X,Y) &=\lim_j\operatorname{colim}_i\operatorname{Map}(X_i,Y_j)\\ &\simeq\lim_j\operatorname{Cont}(L_X,Y_j)\\ &\simeq\operatorname{Cont}(L_X,L_Y). \end{aligned} \tag{4.3} \]

These construct \(R\) on morphisms. The identity class gives each projection \(\pi_j\), hence the identity on the limit. For composition, let \(f:X\to Y\), \(g:Y\to Z\) be pro-maps. Represent the \(\ell\)-component of \(g\) by \(b:Y_j\to Z_\ell\), and the \(j\)-component of \(f\) by \(a:X_i\to Y_j\). Their composite component is represented by \(ba\). Writing superscripts to distinguish the limit projections, we have

\[ \pi_\ell^Z R(g)R(f) =b\pi_j^Y R(f)=ba\pi_i^X. \]

This is exactly the limit map induced by the composite representative, so (4.3) respects composition. Moving representatives to later stages changes neither side. Thus \(R\) is a functor, and (4.3) makes it fully faithful. Cofinal replacements give the same canonical limit and the same formal Hom classes, so the construction applies to arbitrary small cofiltered presentations.

For essential surjectivity, retain the finite-partition presentation from Stacks, Lemma 5.22.2. For a space \(S\in\mathsf{Stone}\), let \(P(S)\) be the inverse system of finite clopen partitions, ordered by refinement. Its finite set at a partition is the set of its nonempty blocks, and its transitions send a fine block to the containing coarse block. The cited result supplies the canonical homeomorphism

\[ \theta_S:S\xrightarrow{\sim}R(P(S)), \qquad s\longmapsto\text{its block in every partition}. \tag{4.4} \]

For \(S=\varnothing\), use its empty partition; the corresponding system is the constant empty set. Thus essential surjectivity includes the empty space, and full faithfulness proves (4.2).

An inverse functor can be made explicit. A continuous \(f:S\to S'\) pulls each finite clopen partition of \(S'\) back to a finite clopen partition of \(S\), discarding empty inverse images. The map on block sets sends a pulled-back block to its target block. Refinement preserves these maps; they define a pro-morphism \(P(f)\) with

\[ R(P(f))=\theta_{S'}f\theta_S^{-1}. \tag{4.5} \]

Alternatively, full faithfulness makes it the unique pro-morphism satisfying (4.5). That formula and faithfulness prove the identity and composition laws for \(P\), and make \(\theta\) natural. For each pro-object \(X\), full faithfulness lifts \(\theta_{R(X)}\) to a unique isomorphism \(X\to P(R(X))\); its inverse is the lift of the inverse homeomorphism. Naturality follows after applying the faithful functor \(R\). Together with (4.4), these are the required natural inverse equivalences. \(\square\)

A map to an entire profinite target need not factor through one finite source stage. What (4.3) says is that its each finite target coordinate factors through a source stage; the source stage may depend on that coordinate. Exercise 3 shows why this distinction matters.

5. Exercises with solutions

Exercise 1 (introductory: parity needs three coordinates). Let \(X_n=\{0,1\}^{\{1,\ldots,n\}}\), with truncation transitions and \(X_0\) a singleton. Identify its limit. Give a stage representative for the map to \(\{0,1\}\) that takes the parity of the first three coordinates. Prove it cannot be represented at a stage \(n<3\).

Solution. Compatible finite prefixes specify exactly one infinite binary sequence, so the limit is \(\{0,1\}^{\mathbb N_{>0}}\) with its product topology. At stage three use \((x_1,x_2,x_3)\mapsto x_1+x_2+x_3\pmod 2\). Its composite with the projection is the required continuous map. Two sequences whose first two coordinates agree but whose third coordinates differ have different parities. They have identical projections to every stage \(n<3\); a stage map there would give them equal values. Thus three is the least possible stage. Every transition here is surjective, so there are no disappearing points at the stages.

Exercise 2 (intermediate: an extra point disappears). Let \(X_n=\{a,b,c_n\}\) for \(n\geq0\). Each transition from stage \(n+1\) to \(n\) fixes \(a,b\) and sends \(c_{n+1}\) to \(a\). Compute the limit and stable images. Construct two distinct stage-zero maps to \(\{0,1\}\) that induce the same limit map. Show that this pro-object is isomorphic to the constant two-element set.

Solution. No compatible family can have coordinate \(c_n\), since it is absent from the image of stage \(n+1\). The only families are the all-\(a\) and all-\(b\) families. The limit is the discrete two-element space, and \(S_n=\{a,b\}\) at every stage. Take both maps to send \(a\) to zero and \(b\) to one, but let their values at \(c_0\) differ. They agree on the limit and become equal after precomposition with the first transition, as Theorem 3.1 predicts.

Let \(D=\{a,b\}\). Its inclusions into every \(X_n\) are compatible and define a pro-map \(s:\iota D\to X\). The maps \(r_n:X_n\to D\) fixing \(a,b\) and sending \(c_n\) to \(a\) define the same class \(r:X\to\iota D\). Clearly \(rs=1_D\). At target stage \(n\), the component of \(sr\) sends the extra point to \(a\). After one further source transition, it agrees with the representative of the identity component, since that transition already sends \(c_{n+1}\) to \(a\). Thus \(sr=1_X\) in the pro-category. This proves the claimed isomorphism while retaining non-surjective original transitions.

Exercise 3 (advanced: stages depend on the target coordinate). Give \(T=\mathbb N_{>0}\cup\{\infty\}\) the topology in which finite integers are isolated and neighborhoods of \(\infty\) contain all but finitely many integers. On the infinite binary sequence space, let \(f\) take the index of the first one, and take value \(\infty\) on the all-zero sequence. Prove that \(T\) is in \(\mathsf{Stone}\) and that \(f\) is continuous. Show that \(f\) does not factor through any finite prefix stage, although each of its finite partition coordinates does.

Solution. An open cover contains a member containing \(\infty\), leaving only finitely many isolated integers to cover; hence \(T\) is compact. Distinct integers are separated by singleton opens, and an integer \(n\) is separated from \(\infty\) by \(\{n\}\) and its complement. These are clopen. The cofinite neighborhoods of \(\infty\) are also clopen, so this is a clopen basis; any two distinct points are separated by a clopen set, which forces connected subsets to be singletons. Thus \(T\) is compact, Hausdorff and totally disconnected.

The inverse image of \(n\) is the cylinder requiring the first \(n-1\) bits to be zero and the \(n\)-th bit to be one. It is clopen. The inverse image of a basic neighborhood of \(\infty\), whose finite complement is \(F\), is the complement of the finite union of those cylinders for \(n\in F\), hence open. These sets form a target basis, proving continuity. At every prefix length \(m\), the all-zero sequence and a sequence whose first one occurs at \(m+1\) have the same prefix but different \(f\)-values. Therefore \(f\) never factors through that finite stage.

The partitions with blocks \(\{1\},\ldots,\{m\},\{m+1,m+2,\ldots,\infty\}\) suffice to present \(T\). Any finite clopen partition has its \(\infty\)-block cofinite, so one of these partitions refines it. The map to the displayed block set depends only on the first \(m\) source bits: either the first one has occurred, or the image belongs to the tail block. These finite coordinates are compatible and define the pro-morphism corresponding to \(f\). Their source stages grow with \(m\); there is no uniform finite prefix.

Exercise 4 (intermediate: the empty object is retained). Take \(X_0\) a singleton and \(X_n=\varnothing\) for \(n\geq1\), with the unique possible inverse transitions. For a finite set \(T\), compute both \(\operatorname{Hom}_{\operatorname{Pro}(\mathsf{FinSet})}(X,\iota T)\) and \(\operatorname{Hom}_{\operatorname{Pro}(\mathsf{FinSet})}(\iota T,X)\). Explain why an inverse system with every finite stage nonempty cannot give the same realization.

Solution. The limit of \(X\) is empty. In the first Hom formula, the filtered tail consists of the singleton sets \(\operatorname{Map}(\varnothing,T)\), including when \(T\) is empty. Its colimit is a singleton. In the second formula, the target limit of Hom sets contains \(\operatorname{Map}(T,\varnothing)\) at every stage \(n\geq1\). It is empty if \(T\ne\varnothing\), and is a singleton if \(T=\varnothing\); compatibility adds no further condition. Thus \(X\) is isomorphic to the constant empty set, as also follows by restricting to its cofinal empty tail. It is initial, and it is not terminal since a constant singleton has no map to it. Finally the retained nonempty-limit lemma gives a point in the limit of any cofiltered system of finite nonempty sets, regardless of surjectivity of its transitions. Such a system cannot realize the empty space.

References

This is an exposition of known mathematics with independently written proof details and exercises. It makes no claim of new research or independent review.