Theorem: If $S$ is a complete, separable metric space, then each probability measure on it is inner regular.
Proof: Since $S$ is separable, for each $n \in \mathbb{N}$ there exist countably many balls of radius $1/n$ such that $S = \bigcup\limits_{j = 1}^\infty {{K_{j,n}}} = \bigcup\limits_{j = 1}^\infty {{{\overline K }_{j,n}}} $. Let $\varepsilon > 0$. Since $\mathbb{P}$ is continuous with respect to a rising sequence of events, for each $n \in \mathbb{N}$ there exists ${k_n} \in \mathbb{N}$ such that $\mathbb{P}\left( {\bigcup\limits_{j = 1}^{{k_n}} {{{\bar K}_{j,n}}} } \right) > 1 - \frac{\varepsilon }{{{2^n}}}$.
Now comes the part which is unclear to me:
Define ${K_\varepsilon } = \bigcap\limits_{n = 1}^\infty {\left( {\bigcup\limits_{j = 1}^{{k_n}} {{{\overline K }_{j,n}}} } \right)} $; ${K_\varepsilon }$ is a closed set (obviously) and for each $n \in \mathbb{N}$ we have ${K_\varepsilon } \subseteq \bigcup\limits_{j = 1}^{{k_n}} {{{\overline K }_{j,n}}} $, by applying Frechet-Hausdorff theorem(J.L.Kelley, no page reference) it follows that ${K_\varepsilon }$ is compact.
Furthermore, $1 - \mathbb{P}\left( {{K_\varepsilon }} \right) = \mathbb{P}\left( {K_\varepsilon ^C} \right)\sum\limits_{n = 1}^\infty {\mathbb{P}\left( {{{\left( {\bigcup\limits_{j = 1}^{{k_n}} {{{\bar K}_{j,n}}} } \right)}^C}} \right)} < \sum\limits_{n = 1}^\infty {\frac{\varepsilon }{{{2^n}}}} = \varepsilon $ is inner regular.
My question: How do I conclude that ${K_\varepsilon }$ is compact? What does the referenced theorem state?
EDIT: For definition of inner-regularity, see here.
Since $K_\epsilon$ is closed it is complete (as a subspace of $S$).
Choose some $n$. Then the ${{{\overline K }_{j,n}}}$ ($j=1,...,k_n$) cover $K_\epsilon$. Select $x_j \in K_\epsilon \cap {{{\overline K }_{j,n}}}$. Then the $x_j$ form a ${2 \over n}$-net for $K_\epsilon$. Hence for any $\delta>0$ we can find a finite $\delta$-net for $K_\epsilon$ and so it is totally bounded.
Hence $K_\epsilon$ is compact.