I am trying to prove that the functor \begin{align*} \mathrm{Top} &\longrightarrow \mathrm{Cond}(\mathrm{Set}) \\ X &\longmapsto \underline{X} \end{align*} admits a left adjoint and it is the functor $T \mapsto T(*)$ where $T(*)$ has the quotient topology of the map $\bigsqcup_{\underline{S} \rightarrow T} S \rightarrow T(*)$ where the disjoint union runs over all profinite sets $S$ with a map to $T$.

To prove this I am trying to construct the following bijection. We have the map \begin{align*} \phi: \mathrm{Hom}(T, \underline{X}) &\longrightarrow \mathrm{Hom}(T(*), X) \\ f &\longmapsto f_* \ . \end{align*} On the other hand, if $g \in \mathrm{Hom}(T(*), X)$, then we can consider the map $\underline{g}: \underline{T(*)} \rightarrow \underline{X}$. I am trying to construct a map $i: T \rightarrow \underline{T(*)}$ to have an inverse for $\phi$: \begin{align*} \psi: \mathrm{Hom}(T(*), X) &\longrightarrow \mathrm{Hom}(T, \underline{X}) \\ g &\longmapsto \underline{g} \circ i \ . \end{align*} Is there a way to construct $i$? How to prove this adjunction?

Edit: I have managed to construct $i$, but I still could not prove that $\phi$ and $\psi$ are mutually inverses.

I have constructed $i$ as the following: For each $S$ profinite we have the equivalence $\mathrm{Hom}(\underline{S},T) = T(S)$ thanks to the Yoneda Lemma. We can define

\begin{align*} i_S: \mathrm{Hom}(\underline{S},T) &\longrightarrow \mathrm{Hom}(S, T(*)) \\ \eta &\longmapsto \eta_* \ . \end{align*}

It is easy to prove that for $g \in \mathrm{Hom}(T(*), X)$ we have $(\underline{g} \circ i)_* = g$, but have not managed to prove that for $f \in \mathrm{Hom}(T, \underline{X})$ we have $f = \underline{f_*} \circ i$. So it only remains to prove that for $\eta \in \mathrm{Hom}(\underline{S},T)$ we have

\begin{align*} f_S (\eta) = f_* \circ \eta_* \end{align*}

I suspect this is due to the Yoneda Lemma but I could not prove it.

  • 5
    $\begingroup$ Note that for set-theoretic reasons (see Warning 2.14 in Scholze's notes), the functor $\underline T$ is not a condensed set unless $T$ is a T1 topological space. $\endgroup$
    – Wojowu
    Apr 10, 2022 at 13:39
  • 2
    $\begingroup$ An element of $T(S)$ is a map $\underline{S} \to T$. Given such a map we then get a map $S \to \coprod_{\underline{S'}\to T} S' \to T(*)$, which gives an element of $\underline{T(*)}(S)$. That's your counit $T \to \underline{T(*)}$. $\endgroup$ Apr 10, 2022 at 16:56
  • 1
    $\begingroup$ @Wojowu although this problem goes away by working with pyknotic sets instead (after Barwick and Haine), at the expense of assuming Grothendieck universes. $\endgroup$ Apr 10, 2022 at 17:20
  • $\begingroup$ Thank you, @DylanWilson! I managed to do it! $\endgroup$ Apr 10, 2022 at 19:36
  • $\begingroup$ @DylanWilson I managed to do it, I have constructed $i$ (the counit). The maps $\phi$ and $\psi$ are inverses? $\endgroup$ Apr 10, 2022 at 19:51

1 Answer 1


As Wojowu noted in the comments, one should really look at $T_1$ topological spaces. Consider the functors \begin{align*} G\!: \mathbf{Top}_{T_1} &\leftrightarrows \mathbf{Cond}_\kappa:\!F\\ X &\mapsto \big(\underline X \colon S \mapsto \operatorname{Cont}(S,X)\big)\\ T(*) &\leftarrow\!\shortmid T, \end{align*} where $T(*)$ is topologised with the quotient topology via the surjection $$\pi \colon \coprod_{(S,f \in T(S))} S \to T(*)\tag{1}\label{1}$$ given on the component $f \in T(S)$ by $f \colon S \to T(*)$. By the Yoneda lemma, this really means that $f \colon h_S \to T$ is a morphism from the representable sheaf $h_S$ to $T$, and the map $S \to T(*)$ is the set-theoretic map $f_{\{*\}} \colon h_S(*) \to T(*)$. But there is the consistent abuse of notation to denote $h_S = \underline S$ as $S$. To see that (\ref{1}) is surjective, use $S = \{*\}$.

As Dylan Wilson noted in the comments, the unit $\eta \colon 1 \to GF$ is given by the natural transformation \begin{align*} (\eta_T)_S \colon T(S) &\to \operatorname{Cont}(S,T(*)) \\ f &\mapsto \pi \circ \iota_f, \end{align*} where $\iota_f \colon S \to \displaystyle\coprod_{(S',f')} S'$ is the insertion of the coordinate corresponding to $(S,f)$. This gives the maps \begin{align*} \phi\!: \operatorname{Hom}_{\mathbf{Cond}}(T,\underline X) &\leftrightarrows \operatorname{Cont}(T(*),X) :\!\psi \\ f &\mapsto f_{\{*\}} \\ g \circ \eta_T &\leftarrow\!\shortmid g. \end{align*} It remains to check that these are inverses. If $g \colon T(*) \to X$ is continuous, then $\phi(\psi(g)) \colon T(*) \to X$ is given by $\psi(g)_{\{*\}}$, i.e. the map taking $t \in T(*)$ to $g \circ \pi \circ \iota_f \colon \{*\} \to X$. But $\pi \circ g \colon \{*\} \to T(*)$ is just (the constant map with value) the point $t$ (this is how we checked surjectivity of $\pi$ earlier!), so $\psi(g)_{\{*\}}$ takes $t$ to $g(t)$, i.e. $\phi(\psi(g)) = g$.

Conversely, given a natural transformation $f \colon T \to \underline X$, we get another natural transformation $\psi(\phi(f)) \colon T \to \underline X$. Write $g \colon T(*) \to X$ for $\phi(f) = f_{\{*\}}$. If $S$ is extremally disconnected and $h \in T(S)$, then $\psi(g)_S$ takes $h$ to the composition $$S \overset{\iota_h}\to \coprod_{(S',h')} S' \overset\pi\to T(*) \overset g\to X.$$ By definition, the composition $\pi \circ \iota_h \colon S \to T(*)$ is the map $h_{\{*\}} \colon h_S(*) \to T(*)$. Thus $\psi(g)_S(h)$ is the composition $$h_S(*) \overset{h_{\{*\}}}\longrightarrow T(*) \overset{f_{\{*\}}}\longrightarrow X.$$ This is the same thing as $(fh)_{\{*\}} \colon h_S(*) \to \underline X(*)$, which is the continuous map $S \to X$ given by $f_S(h) \in \underline X(S)$. We conclude that $\psi(\phi(f))_S(h) = f_S(h)$, and since $S$ and $h$ are arbitrary that $\psi(\phi(f)) = f$. $\square$

Remark. Morally what's going on here is the following: since $T(*)$ has the quotient topology, a map $g \colon T(*) \to X$ is continuous if and only if each of the compositions $gf \colon S \to X$ are continuous with $S$ extremally disconnected and $f \colon S \to T$ an $S$-point of $T$. Thus, a continuous map $T(*) \to X$ is the same as continuous maps $S \to X$ for every $S \to T$, which is roughly what a natural transformation $T \to \underline X$ is.

  • $\begingroup$ Where, in this argument, is the T1 separation argument used? $\endgroup$ Aug 12, 2022 at 13:42
  • $\begingroup$ My understanding is that you don't need it when $\kappa$ is fixed, but a condensed set in Clausen–Scholze is a thing living in the colimit of $\mathbf{Cond}_\kappa$ over 'all' cardinals $\kappa$ (within some universe?), where the transition maps are given by left Kan extension. If I understand correctly, the claim is then that for cardinals $\kappa \leq \lambda$, the left Kan extension of $\underline X \in \mathbf{Cond}_\kappa$ to $\mathbf{Cond}_\lambda$ agrees with $\underline X \in \mathbf{Cond}_\lambda$ when $X$ is $T_1$, but not in general. $\endgroup$ Aug 12, 2022 at 16:10
  • $\begingroup$ I think it would be useful to make explicit what is the problem with the Kan extension in the particular case when $X=\{s,\eta\}$ is the $2$-point non-$T_1$ space in which only the point $s$ is closed. I don't yet see why the Kan extension disagrees with $\underline{X}$ in this particular case. $\endgroup$ Aug 12, 2022 at 16:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.