For any collection $\mathscr C$ of sets, write $\Upsilon(\mathscr C)$ for the collection of arbitrary unions in $\mathscr C$. Now, I ask the innocent question of idempotency of $\Upsilon$:
Is $\Upsilon(\Upsilon(\mathscr C)) = \Upsilon(\mathscr C)$?
"$\supseteq$" is clearly true. So, let's analyze the other direction: Let $\mathscr A\subseteq \Upsilon(\mathscr C)$. For each $A\in\mathscr A$, choose a $\mathcal C_A\subseteq\mathscr C$ such that $A = \bigcup\mathcal C_A$. Then $\bigcup\mathscr A = \bigcup_{A\in\mathscr A} \bigl(\bigcup \mathcal C_A\bigr) = \bigcup\bigl(\bigcup_{A\in\mathscr A} \mathcal C_A\bigr)\in \Upsilon(\mathscr C)$.
This usage of the full power of AC raises the following questions:
- Is a "better" proof known that uses milder versions of choice?
- Does the idempotence of $\Upsilon$ (for all $\mathscr C$'s) imply any form choice?