During the proof of the Stone-Weierstrass Theorem there is a lemma given:

Lemma: Let $g_{1}$ and $g_{2}$ be in $\mathcal{A}$ (an algebra of real-valued continuous functions which strongly separates points in an arbitrary compact metric space $M$). Then for every error $1/m$ there exist functions $g_{3}$ and $g_{4}$ in $\mathcal{A}$ such that $|g_{3}-\max(g_{1},g_{2})| \le 1/m$ and $|g_{4}-\min(g_{1},g_{2})| \le 1/m$ for all points of $M$.

The proof of the lemma starts of by saying that it suffices to prove the following: if $g$ is in $\mathcal{A}$ then there exists $g_{1}$ in $\mathcal{A}$ with $|g_{1}-|g|| \le 1/m$. I'm sure this is quite trivial but I can't justify why it's sufficient to prove this statement. Could someone explain why?


You have $\max(f,g)=\frac12(f+g+|f-g|)$ and $\min(f,g)=\frac12(f+g-|f-g|)$. So if you can approximate $|f-g|$ well, you can approximate $\min$ and $\max$.

  • $\begingroup$ Is the absolute value of g unnecessary in the inequality then? $\endgroup$ – Twiss013 Jul 16 '18 at 16:24
  • $\begingroup$ It is necessary. You start with $h_1,h_2\in\cal A$, then you find $H\in \cal A$ such that $|H-|h_1-h_2||<\frac1m$, then you have for $\min(h_1,h_2)$ that $|\frac12(h_1+h_2+H)-\min(h_1,h_2)|=\frac12|H-|h_1-h_2||<\frac{1}{2m}$ and (as $\cal A$ is a vector space) $\frac12(h+1+h_2+H)\in \cal A$. $\endgroup$ – Kusma Jul 16 '18 at 16:36

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.