1

Having trouble showing this lemma and its corollary within Stone-Weierstrass. My instructor said this was easy to show, so he did not show it in class. But I cannot seem to do it.

Lemma: Let $\epsilon >0$ and $0 < c < 1$. There exists $n,m \geq 1$ s.t. $p(x) = 1 - (1 - x^n)^m$ satisfying $p(\frac{c}{2}) < \epsilon$ and $p(c) > 1 - \epsilon$.

Corollary: Let $p(x)$ be given as in the above lemma. Then $0 \leq p(x) < \epsilon$ where $x \in [0,\frac{c}{2}]$ and $p(x) \geq 1 - \epsilon$ where $x \in [c,1]$.

I feel like the lemma is more of a statement, and not a lemma. It seems like he is just giving us a function, $p(x)$, and telling us how it is defined. I looked back through my notes, and asked him, and he said that is all we need. So I've kind of ignored that, but I'm just not getting anywhere with the corollary. If I plug in $0$, I get $0$, the lower bound. So I'm assuming by plugging in $c/2$, I have to show the function has its maximum between $0$ and $\epsilon$ at $c/2$.

Similar for the next one, I have to show that at $c$ the function is equal to $1 - \epsilon$ and then I know it's maximum is $1$ from the definition in the lemma. I'm just not sure how to do it, or if my thoughts are on the right track.

Arctic Char
  • 16,972
Nolan P
  • 1,136

1 Answers1

1

Corollary:

For the corollary notice that for all $x \in [0,1]$ $$ p'(x)=mnx^{n-1}(1-x^n)^{m-1} \geq 0. $$ So $p$ is non-decreasing in $[0,1]$. Notice that $p(0)=0<p(\frac{c}{2})<\epsilon$. Hence $0 \leq p(x)<\epsilon$ for all $x \in [0,\frac{c}{2}]$. And notice that $p(c) \geq 1- \epsilon$. Hence $p(x) \geq 1- \epsilon$ for all $x \in [c,1]$.

Without using differantiation:

Suppose $x<y \in [0,1]$. Then $x^n<y^n$ and $-x^n>-y^n$. Hence $1-x^n>1-y^n$ and $(1-x^n)^m>(1-y^n)^m$. So $-(1-x^n)^m<-(1-y^n)^m$ and $1-(1-x^n)^m<1-(1-y^n)^m$. This implies that $p(x)<p(y)$ and $p$ is increasing.

Lemma: This is my attempt on proving the Lemma where I used Bernoulli's inequality which is a standard result and with an easy proof. Hope it helps and if there is something wrong with the proof please let me know.

Edit: Bernoulli's inequality states that $$ (1+x)^r \geq 1+rx $$ for any integer $r \geq 0$ and for every real number $x > -1$. For a proof of Bernoulli's inequality take a look at Proof by induction of Bernoulli's inequality $ (1+x)^n \ge 1+nx$. It is a simple application of induction.

Notice that we can suppose $\epsilon<1$. Now, since $0<c<1$, we know that $\frac{1}{c} > 1$ and we can choose $k$ the smallest positive integer such that $k>\frac{1}{c}$. Then we have $$ k-1 \leq \frac{1}{c} \implies k \leq \frac{1}{c}+1<\frac{2}{c}. $$ Also notice that $-\frac{c^n}{2^n}> -1$ for any $n \in \mathbb{N}$. Taking $x=-\frac{c^n}{2^n}$ and $r=k^n$ we can apply Bernoulli's inequality $$ \left(1+\left(-\frac{c^n}{2^n}\right) \right)^{k^n} \geq 1-k^n\frac{c^n}{2^n}=1-\left(k\frac{c}{2} \right)^n $$ Then follows from $k<\frac{2}{c}$ that $\delta_0=(k\frac{c}{2}) < 1$ and \begin{equation} \left(1-\frac{c^n}{2^n} \right)^{k^n} \geq 1-{\delta_0}^n. \end{equation} Now let's use Bernoulli's inequality again using that $c^n > -1$ for any $n \in \mathbb{N}$. This time take $x=c^n$ and $r=k^n$ and apply the inequality $$ (1-c^n)^{k^n} = \frac{1}{(kc)^n}(1-c^n)^{k^n}k^nc^n \leq \frac{1}{(kc)^n}(1-c^n)^{k^n}(1+k^nc^n) $$ $$ (1-c^n)^{k^n} \leq \frac{1}{(kc)^n}(1-c^n)^{k^n}(1+c^n)^{k^n}=\frac{1}{(kc)^n}(1-c^{2n})^{k^n}<\frac{1}{(kc)^n}. $$ First equality holds because we multiplied and divided by $k^nc^n$ and first equality holds because $k^nc^n<1+k^nc^n$. Then we just used Bernoulli's inequality on $1+k^nc^n$ and used the fact that $0<1-c^{2n}<1$. Now using that $k>\frac{1}{c}$ we have $\delta_1=\frac{1}{kc}<1$. Then \begin{equation} (1-c^n)^{k^n}<{\delta_1}^n. \end{equation} Notice that ${\delta_0}^n \to 0$ and ${\delta_1}^n \to 0$ as $n \to \infty$ (because $0<\delta_0,\delta_1<1$). Hence, using the definition of limit, we can choose $n_0$ such that ${\delta_0}^{n_0}<\epsilon$ and ${\delta_1}^{n_0}<\epsilon$. From what we already proved we have $$ \left(1-\frac{c^{n_0}}{2^{n_0}} \right)^{k^{n_0}} \geq 1-{\delta_0}^n>1-\epsilon $$ $$ (1-c^{n_0})^{k^{n_0}} < {\delta_1}^n<\epsilon. $$ By choosing $n=n_0$ and $m=k^{n_0}$ we have $$ p \left(\frac{c}{2} \right)=1-\left(1-\frac{c^{n_0}}{2^{n_0}}\right)^{k^{n_0}}<1-(1-\epsilon)=\epsilon $$ $$ p(c)=1-(1-c^{n_0})^{k^{n_0}}>1-\epsilon. $$

PAM1499
  • 1,177
  • We have not covered differentiation yet in this course. We cannot use it to prove this. – Nolan P Nov 03 '20 at 12:32
  • Does this help? – PAM1499 Nov 03 '20 at 12:43
  • It makes sense, but I don't understand how this shows what I'm being asked. We get that p is increasing. Ok great, how does that imply p(c/2) < epsilon and p(c) > 1 - epsilon? I don't see how this relates. – Nolan P Nov 04 '20 at 21:56
  • Oh no... I am just answering how to prove the Corollary. So I am assuming the Lemma and showing how you can use it to prove the Corollary. – PAM1499 Nov 04 '20 at 22:02
  • Just looked back at this now. Sorry had a busy few days. A few things I'm not sure about. Where did the -c^n/2^n >= -1 come from? I don't see where that was implied from the proof, it seems like you just wrote it. Similar in the next part, where did c^n >= -1 come from? I know it's true, but I don't quite see where you thought to do that step. I don't think those are mistakes though, just things I want to clarify for myself. Overall though, it looks good. I will say, if it is right, that is definitely the most complicated proof in our class so far. There's a lot going on. – Nolan P Nov 08 '20 at 13:48
  • By hypothesis $0<c<1$. This implies that $-1<0<c^n<1$. And notice that $0<\frac{c^n}{2^n}<1$. Hence $0>-\frac{c^n}{2^n}>-1$. This inequalities are needed in order to use Bernoulli's inequality. I honestly do not know what is the "easy" way of doing this. When my professor presented us Stone-Weierstrass I do not remember seeing this Lemma. So this might end up being overcomplicated. If It is the case, please let me know ;) – PAM1499 Nov 08 '20 at 17:33
  • Ah I see now. My professor actually used three lemmas in his proof for Stone Weirestrass. This is the only one he did not prove, and I can kind of see why now. I will say, it is quite complicated for me to follow being that it's my first time seeing Stone Weirestrass. Maybe he proved Stone Weirestrass in a more complicated way which is what caused this to come up. I have no idea in all honesty. – Nolan P Nov 09 '20 at 15:12
  • If you believe you can modify it to possibly make it a bit more clear for a first-time analysis student, let me know. If not I'll go ahead and accept it. I believe it does answer the question, and it'll likely be helpful for future students! – Nolan P Nov 09 '20 at 17:28
  • Honestly I think it is accessible for a first time analysis student but I can try to make some parts clearer. Are you familiar with Bernoulli's inequality? There is any concept or part of the proof there is still not that clear? Knowing this is important so that I can modify it. – PAM1499 Nov 09 '20 at 17:49
  • I think that could be what's causing the confusion. I've heard of Bernoulli's inequality, but I've never used it nor proved it. Other than that, I think if it sits with me I'll get it. It may not be as much of the content that's getting me, it just seems maybe a bit overwhelming I guess? If that makes sense. – Nolan P Nov 09 '20 at 18:16
  • Ok! I'll try to modify and explain a bit better how I used the inequality. – PAM1499 Nov 09 '20 at 18:20
  • Try to see now how Bernoulli's inequality was applied. About being overhelming... Well, I guess once you have done some proofs of this kind you end up getting used to it. Just take your time and I recommend that you ask your instructor/colleagues how they would actually prove this lemma. They might be able to give a proof better suited to your context. I hope it helps ;) – PAM1499 Nov 09 '20 at 18:48
  • It definitely does. Thank you for the insight! – Nolan P Nov 09 '20 at 19:05
  • You are welcome! ;) – PAM1499 Nov 09 '20 at 19:06