Skip to content

The category of uniform spaces is (probably) coregular #317

Description

@ScriptRaccoon

Two properties of the category of uniform spaces remain open after #312: if it is regular or coregular. Probably the category is coregular (but not regular). The proof of coregularity cannot be the same as for Top and Meas since the uniform structure on a quotient is not as easy. Gemini has produced the following proof that Unif is coregular. This needs to be checked and then added to the database if it is correct. I will paste the proof here so that it doesn't get lost. Warning. AI can (and will) make errors.


The underlying set of the pushout in $\mathbf{Unif}$ is just the pushout in $\mathbf{Set}$, and the complexity lies entirely in the entourages. Because the composition axiom requires a sequence of entourages satisfying $W_{n+1} \circ W_{n+1} \subseteq W_n$, we cannot just map a single entourage forward. We have to close it under finite chains.

We can prove this stability entirely relationally—without ever defining a real-valued pseudometric—by using a combinatorial "word length" rule on sequences of entourages.

Here is the direct proof.

1. The Setup

Let $i: A \to B$ be a regular mono (uniform embedding) and $f: A \to C$ be any uniformly continuous map. Let $P = B \sqcup_A C$ be the pushout in $\mathbf{Set}$, with canonical maps $g: B \to P$ and $j: C \to P$.

We need to show that $j: C \to P$ is a uniform embedding.

  1. Injectivity: Because $i$ is injective, pushout properties in $\mathbf{Set}$ guarantee that $j$ is injective.
  2. Uniformity: We need to show that the subspace uniformity induced on $C$ by $P$ is exactly the original uniformity on $C$. Since $j$ is uniformly continuous, we only need to prove that for every entourage $U \in E_C$, there is an entourage $W \in E_P$ such that $(j \times j)^{-1}(W) \subseteq U$.

2. Matching the Sequences

Let $U \in E_C$ be given. We first construct the sequence you mentioned on $C$: choose a sequence of symmetric entourages $U_n \in E_C$ such that $U_0 = U$ and $U_{n+1} \circ U_{n+1} \subseteq U_n$ for all $n \ge 0$.

Because $f$ is uniformly continuous, the preimage $(f \times f)^{-1}(U_n)$ is an entourage on $A$.
Because $i: A \to B$ is a uniform embedding, its pullbacks generate the exact uniformity on $A$. Therefore, for each $n$, there exists a symmetric entourage $V_n \in E_B$ such that:

$$(i \times i)^{-1}(V_n) \subseteq (f \times f)^{-1}(U_n)$$

By standard uniform space operations, we can recursively shrink the $V_n$ sequence to also guarantee the composition rule $V_{n+1} \circ V_{n+1} \subseteq V_n$ on $B$.

3. Constructing Pushout Entourages Combinatorially

Now we build the entourages on $P$. Let $Q_n$ be the raw, uncomposed image of the $n$-th entourages:

$$Q_n = (j \times j)(U_n) \cup (g \times g)(V_n) \cup \Delta_P$$

Because $Q_{n+1} \circ Q_{n+1}$ is not necessarily contained in $Q_n$ (one might step from $C$ to $B$ across the glued subspace $A$), we define $W_n \subseteq P \times P$ by taking the closure over "admissible" finite chains.

We say a chain of points $z_0, z_1, \dots, z_k$ in $P$ is $n$-admissible if each step $(z_{l-1}, z_l) \in Q_{m_l}$ for some index $m_l$, and the indices satisfy the dyadic weight condition:

$$\sum_{l=1}^k 2^{-m_l} \le 2^{-n}$$

(Note: We are using fractions purely as a combinatorial rule for concatenating relations, completely avoiding metric functions on the spaces).

Let $W_n$ be the set of all pairs $(x, y) \in P \times P$ connected by an $n$-admissible chain.
This sequence immediately satisfies $W_{n+1} \circ W_{n+1} \subseteq W_n$ (since concatenating an $(n+1)$-admissible chain with another $(n+1)$-admissible chain yields a sum of weights $\le 2^{-(n+1)} + 2^{-(n+1)} = 2^{-n}$, making it $n$-admissible). Since $W_n$ contains $Q_n$, it pulls back to entourages in $B$ and $C$, meaning $W_n$ are valid entourages for the final uniformity on $P$.

4. The Projection Argument (No Short-Circuiting)

We want to prove that $(j \times j)^{-1}(W_0) \subseteq U_0 = U$.

Suppose $(j(c_1), j(c_2)) \in W_0$. This means there is a $0$-admissible chain in $P$ from $j(c_1)$ to $j(c_2)$. The chain alternates between steps in $j(C)$ and steps in $g(B)$.

Consider any contiguous "excursion" the chain makes into $g(B \setminus A)$. Such an excursion must start and end at the boundary where $B$ and $C$ meet—which is exactly the image of $A$.
Therefore, a $B$-excursion looks like a sequence of points in $g(B)$ starting at some $g(i(a_p)) = j(f(a_p))$ and ending at $g(i(a_q)) = j(f(a_q))$, taking steps entirely in $V_{m_l}$.

Let the sum of the weights $2^{-m_l}$ of this specific $B$-excursion be bounded by $2^{-M}$ for some integer $M$.
By the standard relational chaining lemma (which holds because $V_{k+1} \circ V_{k+1} \subseteq V_k$), this sequence of steps in $B$ composes to a single pair in $V_M$:

$$(i(a_p), i(a_q)) \in V_M$$

Here is the magic step: By our earlier choice of $V_M$, the fact that $(i(a_p), i(a_q)) \in V_M$ implies that:

$$(a_p, a_q) \in (i \times i)^{-1}(V_M) \subseteq (f \times f)^{-1}(U_M)$$

Which means:

$$(f(a_p), f(a_q)) \in U_M$$

This tells us that any excursion into the glued space $B$ can be canonically replaced by a single step in $C$ using $U_M$, and this replacement has a weight $2^{-M}$ which is less than or equal to the sum of the weights of the $B$-steps it replaced.

5. Conclusion

By applying this projection to every $B$-excursion, our original $0$-admissible chain from $j(c_1)$ to $j(c_2)$ in $P$ collapses into a chain lying entirely in $C$, taking steps only in $U$-entourages.

Crucially, because the weights of the replaced steps only decreased or stayed the same, the new $C$-chain is still $0$-admissible! By applying the relational chaining lemma one last time to $C$ (using $U_{n+1} \circ U_{n+1} \subseteq U_n$), the entire $C$-chain composes into $U_0$.

Thus, $(c_1, c_2) \in U_0 = U$. We have successfully shown that $(j \times j)^{-1}(W_0) \subseteq U$, proving that $j$ is a uniform embedding. Regular monos are stable under pushouts, and $\mathbf{Unif}$ is a coregular category.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions