I am working on the following problem from Bergman's supplement to Rudin PMA:
Given a set $E$, let $L(E)$ denote its limit points, $L^n(E) = L(\dots(L(E)))$, and $L^0(E) = E$. Construct for every $n \in \mathbb{Z}^+$ a subset $E \subseteq \mathbb{R}$ such that $L^{n - 1}(E) \ne \emptyset$, but $L^n(E) = \emptyset$.
Apparently the construction should be given recursively, and so far my idea is to fill the gaps between each element in the previous iteration with a convergent sequence, which should get eliminated upon taking the limit points.
Let $E_1 = \{0\}$, $E_2 = \{\frac{1}{n} : n \in \mathbb{Z}^+\}$. Having constructed countable $E_3, \dots, E_{n - 1}$, we arrange each $E_m$ into a decreasing sequence $(e^m_i)$, and let $$E_n = \bigcup_{i = 1}^\infty \{e^{n - 1}_i + \frac{1}{n} : n \in \mathbb{Z}^+ \land \frac{1}{n} < e^{n - 1}_{i + 1} - e^{n - 1}_i\}, $$ which is countable as a countable union of countable sets. We claim that $L(E_n) = E_{n - 1}$, in which case we would be done, as $L(E_1) = \emptyset$.
The main problem is that I don't know how to make "arranging into a decreasing sequence" a rigorous argument, especially since we are restricted to the canonical ordering.