I'm trying to do the following exercise for my Real Analysis class:

Let $p$ be a given natural number. Give an example of a sequence $\left(x_{n}\right)$ that is not a Cauchy sequence, but that satisfies $\lim \left|x_{n+p}-x_{n}\right|=0$

However, I am in constant doubt in regard to the "let" word in math texts. Can I choose, say $p = 1$, or when one says "let" I am supposed to stick with $p \in \mathbb{N}$ and nothing more? What kind of "control" do I have over $p$?

For instance, when one says "Given $\epsilon \gt 0$", I usually see things like: Let $\epsilon = \frac{\epsilon}{2}$ so one can finish a certain argument.

Can someone help me? I need to solve this doubt once and for all, it bothers me very often.

  • 5
    $\begingroup$ You have no control over $p$. The author has introduced $p$ and told you that it is a natural number, and has given you no additional information about $p$. $\endgroup$ – littleO Oct 30 '20 at 4:54
  • 18
    $\begingroup$ Does Bartle actually say "Let $\epsilon=\epsilon /2$" after stating that $\epsilon>0$? $\endgroup$ – DanielWainfleet Oct 30 '20 at 5:04
  • 3
    $\begingroup$ @DanielWainfleet: If Bartle does that, then Bartle is wrong. That's just nonsense. Please, Lucas, clarify whether Bartle really does that. $\endgroup$ – user21820 Oct 30 '20 at 16:13
  • 1
    $\begingroup$ And I strongly suggest you read carefully through this post about "let". $\endgroup$ – user21820 Oct 30 '20 at 16:14
  • 1
    $\begingroup$ Bartle does not do that @user21820 I got confused because sometimes $\frac{\epsilon}{2}$ just pops up. However, that does not happen like i said. Edited my question in that matter $\endgroup$ – Lucas Oct 30 '20 at 16:44

To attempt an answer to your question, any time you read an instance of "Let $x$ be an element of $S$," you are being asked to consider an element $x$ whose only salient property as far as you are concerned is that it is a member of $S$. Until further assumptions are made about the specific nature of $x$, you should only use properties of $x$ that it has as an element of $S$ (and any supersets that $S$ is contained in, depending on the context). Some examples:

  1. Let $x$ be a real number. (Or "Let $x$ be an element of $\mathbb R$" to match the language I used above.)
  • This means to only consider the properties that $x$ has as a real number, or possibly also as a complex number, for example depending on the situation.
  1. Let $n$ be a given natural number.
  • This means that for our discussion, $n$ is a natural number, but that is all we know about it. Depending on what we talk about next, we may or may not also use other properties of $n$, like that it is a rational number, or that it is a complex number, for instance.
  1. Let $A$ be a subset of $\mathbb R$.
  • This is similar to the first example. All we want to do at this point is consider the properties of $A$ we know it has as a subset of $\mathbb R$. It may be the collection of natural numbers, it may be the empty set, it may be $\{\pi, \pi^2\}$, and so on.

Part of the point of making statements like this is that if we then go on to prove something involving this arbitrary element $x$, then we can substitute the value of $x$ with any specific element of $S$ and obtain a "new" theorem. For your example, if you can prove the statement you are given assuming nothing other than $p$ is a natural number, then you know, for example that there is a sequence $\{x_n\}$ in $\mathbb R$ that is not Cauchy, with the property that $$ \lim_{n\to\infty}(x_{n+314159}-x_n) = 0, $$ which looks pretty impressive, but is really nothing other than a special case where $p = 314159$ of the theorem stated in terms of the otherwise arbitrary $p$. We get to know that this "$314159$ theorem" is true for free if the only property of $p$ we use in our proof is that $p$ is a natural number since $314159$ is also a natural number.

Edit: "The $\epsilon$ issue." In analysis, we often want to prove theorems that have the form "For all $\epsilon > 0$, $P(\epsilon)$ is true." Where $P(\epsilon)$ is a statement involving $\epsilon$. For example,

  1. $P(\epsilon) = $ there exists $\delta > 0$ so that $|x^2-100| < \epsilon$ if $|x-10| < \delta$.

  2. $P(\epsilon) = $ there exists $N\in\mathbb N$ so that for all $n,m\ge N$, $|x_n-x_m| < \epsilon$.

When you think about these theorems in this way, I think it becomes a little bit clearer why in order to establish the truth of $P(\epsilon)$ for all $\epsilon > 0$, it suffices to establish the truth of $P(\epsilon/2)$ for all $\epsilon > 0$. Essentially, $\epsilon$ is a place-holder for "all the positive real numbers," and for that purpose, if $C$ is any constant like $2$ or $\frac12$, then $C\epsilon$ is also a place-holder for all the positive real numbers if $\epsilon$ is a place-holder for all positive real numbers. So it suffices to show that for all $\epsilon > 0$, $P(C\epsilon)$ is true. (It is critical however that $C$ does not depend on $\epsilon$! For instance, if we tried to let $C$ depend on $\epsilon$, then we could "cheat" and prove $P(\frac1\epsilon \epsilon)$ is true for all $\epsilon > 0$, but this only shows that $P(1)$ is true...)

Edit to the edit: I wrote in parentheses at the end of my last edit that $C$ is not allowed to depend on $\epsilon$, but this is not quite correct. So long as $C\epsilon$ is a place-holder for all the positive real numbers, $C$ can itself be a function of $\epsilon$. Here are two examples and one non-example of when $C\epsilon$ is a place-holder for all the positive real numbers when $\epsilon$ is itself a place-holder for all the positive real numbers (you can try to think of more examples for fun!):

  • $C = (1+\epsilon^2)$. Then $C\epsilon = \epsilon + \epsilon^3$ is still a place-holder for all the positive real numbers.

  • $C = \frac{1}{\sqrt{\epsilon}}$. Then $C\epsilon = \sqrt{\epsilon}$ is still a place-holder for all the positive real numbers.

  • $C = \frac{1}{\epsilon}$. Then $C\epsilon = 1$ is not a place-holder for all the positive real numbers. This was the kind of "malicious" dependence on $\epsilon$ that I had in mind when I said that $C$ should not depend on $\epsilon$ in the parenthetical sentence at the end of my first edit.

Sometimes it can be convenient to allow $C$ to depend on $\epsilon$. I know of at least one example where the bound I had by the end of the proof was $\epsilon + \epsilon^3$, which motivates my first example I wrote here where $C = 1 + \epsilon^2$.

  • $\begingroup$ Interesting and super clear. What about the $\epsilon$ thing i wrote about? It really makes me uncomfortable. I tend to think that the claim should be proved for $\epsilon > 0$ and that's it. I don't get why one has the right to choose a certain $\epsilon$ $\endgroup$ – Lucas Oct 30 '20 at 4:41
  • $\begingroup$ @Lucas: I updated my answer a little bit to furnish more examples and also address "the epsilon issue." Hopefully it helps! $\endgroup$ – Alex Ortiz Oct 30 '20 at 4:54
  • 4
    $\begingroup$ Well, this comment box right here tells me i should not use it only to say "Thanks". However, there is just no way i could not thank you for such a detailed explanation!!! I am convinced now. My experience here at mathstack exchange has been great! Thank you again! Cheers from Brazil! $\endgroup$ – Lucas Oct 30 '20 at 5:04

The short answer is: It depends on the situation.

Here, you can't choose $p=1$, as it is given that $p$ is a fixed natural number.

However, when it comes to working with $\epsilon$'s, you are required to show a result to be true for all $\epsilon > 0 $. Hence, given an $\epsilon > 0$, its a convention for authors to use, "choose $\epsilon = \frac{\epsilon}{2}$ ", when they actually mean to say, they are trying to prove the result for this fixed $\epsilon$, by choosing some ${\epsilon}' = \frac{\epsilon}{2} $, and thus, they are proving the result for all $\epsilon > 0$, as this fixed $\epsilon$ was arbitrary.


No, you cannot take $p=1$, in order to prove the statement you need to show that it holds for arbitrary $p \in \mathbb{N}$. Taking $\epsilon' = \frac{\epsilon}{2}$ is allowed because this still works for an arbitrary choice of $\epsilon$. If I take $\epsilon = 0.1$ or if take it to be $\epsilon =0.9$, the proof still works since $\epsilon'$ is defined relatively, but if you only prove something for say $p=1,2$, then you have not proven the claim generally, and your proof might break down for $p=4$ say

  • $\begingroup$ But what does it mean to work for an arbitrary choice of $\epsilon$? It always looks to me as if we are proving the claim only for $\frac{\epsilon}{2}$ only $\endgroup$ – Lucas Oct 30 '20 at 4:39
  • 1
    $\begingroup$ why does this make a difference though? If i said take $\epsilon=0.1$, then it would be an issue, but if take $\epsilon'=\epsilon/2$, then the proof still works for any choice of $\epsilon>0$ right? As in, for any $\epsilon>0$ you choose, the proof will always work out because you can always also find a corresponding $\epsilon'$ $\endgroup$ – WeakLearner Oct 30 '20 at 4:43
  • $\begingroup$ So can i take, for instance, $p^{\prime} = \frac{p}{2}$ and prove the statement? $\endgroup$ – Lucas Oct 30 '20 at 4:50
  • 1
    $\begingroup$ well p/2 might no longer be an integer, so this might not be useful.. but you will see some proofs where they might break it into two cases, $p=2k$ and $p=2k+1$ for some integer $k$ (even and odd cases).. $\endgroup$ – WeakLearner Oct 30 '20 at 4:59

Well, usually the word "let" it's use for defining the context of a variable and also for letting fixed its value. However, it's usual to redefine the value of certain variable as in your example "$\epsilon = \frac{\epsilon}{2}$" strictly speaking, it shouldn't happen, but it's a convention, in order to avoid choosing from the beginning the value $\epsilon/2$ or others, which works but which could be seen as non intuitive, and to avoid rewrite the whole proof from the start.

All you know about $p$, it's that is a natural number, could be any, but you cannot choose one at your convenience, unless you will go into all of them, for example considering $p$ as an odd number, then as an even number, and so on...


You have no choice about $p$: you have to find an answer that works no matter what $p$ you are given. For example, $$x_n=\cos\frac{2\pi n}p$$would work for any given nonzero $p$, while $x_n=n$ would do in the trivial case $p=0$.


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.