Let ${f : E \rightarrow S}$ be a geometric morphism (between toposes).

For $s$ in $S$ and $x$ in $E$ let ${\pi : f^* s \times x \rightarrow x}$ be the obvious projection in $E$.

Let ${u \rightarrow f^* s \times x}$ be a complemented subobject of ${f^* s \times x}$.

Is the image of $u$ along $\pi$ complemented as a subobject of $x$?

(See also Images of complemented subobjects in hyperconnected toposes over Boolean bases)


No, not even if $E=S$, $f$ is the identity morphism, and $x=1$. In that special case, your question asks whether $\forall z\in s\,\big((z\in u)\lor \neg(z\in u)\big)$ (in the internal language of $S$) implies $(\exists z\in s\,z\in u)\lor\neg(\exists z\in s\,z\in u)$. When $s$ is $\mathbb N$, this is the limited principle of omniscience, which is not intuitionistically valid.

  • $\begingroup$ Thanks very much for your reply, @Andreas Blass. I'd be happy to mark my question as answered; but I think it would be more interesting to edit it. What if we assume that $S$ is Boolean or satisfies Choice? (Shouldn't $\pi$ behave as a codiagonal?) $\endgroup$ – Mendieta Jun 22 at 23:43
  • 3
    $\begingroup$ @Mendieta If $S$ is Boolean (which is weaker than having AC), every subobject of every object is complemented. $\endgroup$ – მამუკა ჯიბლაძე Jun 23 at 6:46
  • 2
    $\begingroup$ @მამუკაჯიბლაძე What you say is true, but the question still makes sense because E need not be Boolean. $\endgroup$ – Mendieta Jun 23 at 17:29
  • $\begingroup$ @Mendieta Sorry you are right. What I said does not matter. What matters is that one can modify the answer to include any non-Boolean topos $E$ that admits a geometric morphism to a Boolean topos $S$. Take $u$ any non-complemented subobject of some object $x$ in $E$, and the rest as in the answer: take $s$ the terminal object of $S$ and $\pi$ the identity morphism of $x$. $\endgroup$ – მამუკა ჯიბლაძე Jun 23 at 19:21
  • $\begingroup$ @მამუკაჯიბლაძე, the subobject $u$ is complemented by hypothesis. $\endgroup$ – Mendieta Jun 23 at 19:30

A different flavour of counterexample from Andreas Blass’s answer, showing this can fail when $S$ is Boolean and satisfies choice: Take $S$ to be sets, and $E = Sh(2^{\mathbb{N}})$, where $2^{\mathbb{N}}$ is the Cantor space (and $f$ is the unique geometric morphism $(\Gamma,\Delta) : Sh(2^{\mathbb{N}}) \to S$).

Now for $n \in \mathbb{N}$, we have $U_n \subseteq 1$ in $E$ corresponding to the clopen $\{ x \in 2^\mathbb{N}\ |\ x_n = 1 \}$. Now each $U_n \subseteq 1$ is complemented, so $\coprod_n U_n \subseteq \Delta(n) \times 1$ is complemented; but $\pi_!(\coprod_n U_n) = \bigcup_n U_n \subseteq 1$ is not complemented, since it’s dense but not equal to $1$.

Generally, sheaf toposes give many counterexamples to the principle “set-indexed unions of complemented subobjects are complemented”.

  • $\begingroup$ please confirm: you mean ${\coprod_n U_n \subseteq \Delta(\mathbb{N}) \times 1}$ ? $\endgroup$ – Mendieta Jun 23 at 22:13
  • $\begingroup$ @Mendieta: Yes — sorry, am on mobile so editing/proofreading the maths is tricky. $\endgroup$ – Peter LeFanu Lumsdaine Jun 23 at 22:21
  • 1
    $\begingroup$ ok. Good example, many thanks. Yet, it is the sort of thing I wanted to avoid with hyperconnectedness. I am not sure how MO works so I hope the following is appropriate: I'll arrow up your answer, consider the question answered by Blass and ask a new question with the further hypotheses. $\endgroup$ – Mendieta Jun 23 at 22:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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