Theorem. Let XX be a normed vector space and let ΩX\Omega\subset X be with int(Ω)\mathrm{int}(\Omega)\neq\emptyset. Then:

  1. int(Ω)=Ω\overline{\mathrm{int}(\Omega)}=\overline{\Omega}.
  2. int(Ω)=int(Ω)\mathrm{int}(\overline{\Omega})=\mathrm{int}(\Omega).
Remarks

Context. For convex sets with nonempty interior, "taking closure" and "taking interior" are tightly compatible. This is special to convexity and can fail for arbitrary sets.

Proof sketch.

  1. The inclusion int(Ω)Ω\overline{\mathrm{int}(\Omega)}\subset\overline{\Omega} is immediate. For the reverse, it suffices to show Ωint(Ω)\Omega\subset\overline{\mathrm{int}(\Omega)}: fix bΩb\in\Omega and aint(Ω)a\in\mathrm{int}(\Omega); by , points (1/k)a+(11/k)b(1/k)a+(1-1/k)b lie in int(Ω)\mathrm{int}(\Omega) and converge to bb.
  2. The inclusion int(Ω)int(Ω)\mathrm{int}(\Omega)\subset\mathrm{int}(\overline{\Omega}) is obvious. For the converse, use a "push-out" argument: given bint(Ω)b\in \mathrm{int}(\overline{\Omega}) and aint(Ω)a\in\mathrm{int}(\Omega), move slightly past bb along the ray from aa through bb to find a point cΩc\in\overline{\Omega}, then apply the interior-segment lemma to conclude bint(Ω)b\in\mathrm{int}(\Omega).