46
$\begingroup$

I was sitting in analysis yesterday and, naturally, we took the limit of some expression. It occurred to me that "taking the limit" of some expression abides the rules of a linear transformation $$\lim_{x \rightarrow k}\ c(f(x)+g(x)) = c \lim_{x \rightarrow k} f(x) + c\ \lim_{x \rightarrow k} g(x),$$ and (my group theory is virtually non existent) appears also to be a homomorphism: $$\lim_{x \rightarrow k} (fg)(x) = \lim_{x \rightarrow k} f(x)g(x), $$ etc.

Anyway, my real question is, what mathematical construct is the limit?

$\endgroup$
  • 15
    $\begingroup$ I would avoid defining $\lim$ as a function/relation/etc. but rather $\lim_{x\to k} f(x)=a$ as a notational convention or abbreviation for a much more complicated statement. You might as well ask whether $\dim$ is a function from vector spaces to natural numbers, but you will have trouble defining the "set of all vector spaces". Instead writing $\dim(V)=n$ is an abbreviation of a much longer statement about bases, linear indendence, etc. $\endgroup$ – M. Winter Feb 2 at 20:13
  • 3
    $\begingroup$ This is exactly what I'm wondering. Of course I understand that it is a convention of notation, but it's not only a convention of notation. It can't be "turtles all the way down". There has to be some foundational notion at the bottom of the abstraction pile which lends itself to a thorough understanding of what's happening when one takes a limit. That was my question. @M.Winter $\endgroup$ – solidstatejake Feb 3 at 14:37
  • 2
    $\begingroup$ Similar question (old, but it turned up on the front page recently): What kind of "mathematical object" are limits? $\endgroup$ – Calum Gilhooley Feb 3 at 20:26
51
$\begingroup$

In general, let $X, Y$ be topological spaces, and $x_0$ a non-isolated point of $X$. Then strictly speaking, "$\lim_{x\to x_0} f(x) = L$" is a relation between functions $f : X \to Y$ and points $L \in Y$ (the equality notation being misleading in general).

Now, if $Y$ is a Hausdorff topological space, it happens that this relation is what is known as a partial function: for any $f : X \to Y$, there is at most one $L \in Y$ such that $\lim_{x\to x_0} f(x) = L$. Now, for any relation $R \subseteq (X \to Y) \times Y$ which is a partial function, we can define a corresponding function $\{ f \in (X \to Y) \mid \exists y \in Y, (f, y) \in R \} \to Y$ by sending $f$ satisfying this condition to the unique $y$ with $(f, y) \in R$. Then that somewhat justifies the "equality" in the notation $\lim_{x\to x_0} f(x) = L$, though you still need to keep in mind that it is a partial function where $\lim_{x\to x_0} f(x)$ is not defined for all $f$. (This part relates to the answer by José Carlos Santos.)

Building on top of this, in the special case of $Y = \mathbb{R}$, we can put a ring structure on $X \to Y$ by pointwise addition, pointwise multiplication, etc. Then $\{ f : X \to \mathbb{R} \mid \exists L \in \mathbb{R}, \lim_{x\to x_0} f(x) = L \}$ turns out to be a subring of $X \to \mathbb{R}$, and the induced function from this subring to $\mathbb{R}$ is a ring homomorphism. (More generally, this will work if $Y$ is a topological ring. Similarly, if $Y$ is a topological vector space, then the set of $f$ with a limit at $x_0$ is a linear subspace of $X \to Y$ and the limit gives a linear transformation; if $Y$ is a topological group, you get a subgroup of $X \to Y$ and a group homomorphism; and so on.)

$\endgroup$
  • 6
    $\begingroup$ +1, though I don't think your hesitation with respect to the equals sign is justified; $\frac{6}{x}$ is not defined when $x = 0$, but this certainly doesn't stop us from writing $\frac{6}{2} = 3$. $\endgroup$ – ruakh Feb 2 at 1:34
  • 1
    $\begingroup$ Perhaps Daniel is thinking along these lines: quantamagazine.org/… $\endgroup$ – Lawnmower Man Feb 2 at 5:07
  • 6
    $\begingroup$ That comment was in the context of the general case (not necessarily Hausdorff $Y$) in which you do not have uniqueness of the limit. (So, roughly the same objection that many people have to the $f = O(g)$ notation.) $\endgroup$ – Daniel Schepler Feb 2 at 7:00
  • $\begingroup$ You don't need a space to create Lim, but limits on spaces are just mostly known. $\endgroup$ – sanaris Feb 2 at 15:29
  • $\begingroup$ @sanaris How would you take a limit without a space? $\endgroup$ – Jam Feb 2 at 21:50
20
$\begingroup$

Let $I\subset\mathbb R$ be a subset of $\mathbb R$ such that $k$ is an accumulation point of $I$. Let$$R=\left\{f\colon I\longrightarrow\mathbb R\,\middle|\,\lim_{x\to k}f(x)\text{ exists}\right\}.$$Then $(R,+,\times)$ is a ring and the map$$\begin{array}{ccc}R&\longrightarrow&\mathbb R\\f&\mapsto&\lim_{x\to k}f(x)\end{array}$$is a ring homomorphism.

$\endgroup$
  • $\begingroup$ what if $f$ is unbounded around $k$ or am I missing something? E.g assume $0\in I$, then $\frac{1}{x^2} \xrightarrow[x\to 0]{} \infty$. $\endgroup$ – Alvin Lepik Feb 1 at 16:49
  • $\begingroup$ Such a function $f$ will not belong to $R$; in the words, the limit $\lim_{x\to k}f(x)$ will not exist then. $\endgroup$ – José Carlos Santos Feb 1 at 16:54
  • $\begingroup$ Take $I=(0,1)$. Then $f\in R$ and its limit is infinity as $x\to 0$, where $0$ is an accumulation point of $I$. The limit does exist. $\endgroup$ – Alvin Lepik Feb 1 at 16:55
  • $\begingroup$ No, it doesn't. I was thinking only about the case in which $\lim_{x\to k}f(x)$ exists in $\mathbb R$. $\endgroup$ – José Carlos Santos Feb 1 at 16:57
  • $\begingroup$ Ah, right, my bad. Would be helpful to say limit exists and is finite or some such. $\endgroup$ – Alvin Lepik Feb 1 at 16:58
14
$\begingroup$

To be more precise, "taking $\limsup$ of a sequence" and "taking $\liminf$ of a sequence" are both functionals, i.e. linear functions from a vector space whose elements are sequences to its underlying field of real numbers. (You can also consider it a functional in a more advanced way, by putting a topology on the vector space for which the operation is continuous.)

This captures just the additivity. To capture multiplicativity, you need to consider the space of sequences not just as a vector space, but as a ring (as before, you can put a topology on this ring, in which case it becomes natural to consider a space of sequences satisfying enough conditions to become a Banach algebra). Then $\limsup$ and $\liminf$ become ring homomorphisms from a ring whose elements are sequences to the real numbers.

$\endgroup$
  • $\begingroup$ Aren't $\limsup/\liminf$ of a sequence different that the limit of a function? $\endgroup$ – solidstatejake Feb 1 at 16:16
  • 2
    $\begingroup$ Not when the limit exists. The reason I mention these two is that they are well-defined for all sequences, whereas $\lim$ is not. In fact, if you try to extend the definition of limit to all sequences, all possible extensions interpolate between liminf (the smallest extension) and limsup (the largest). $\endgroup$ – pre-kidney Feb 1 at 16:16
  • 5
    $\begingroup$ Umm, $\liminf$ and $\limsup$ aren't additive in general - nor does either one respect multiplication by negative scalars. (Plus, you would need to restrict to the space of bounded functions in order to avoid having values of $+\infty$ or $-\infty$.) $\endgroup$ – Daniel Schepler Feb 1 at 16:23
  • $\begingroup$ @DanielSchepler yes, which is why I was careful to avoid saying that the vector space consists of all sequences in my answer. But thanks for pointing it out explicitly. You do raise a good point though that we are not gaining much by going from lim to limsup/liminf since we will need to consider restrictions either way. $\endgroup$ – pre-kidney Feb 1 at 16:33

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.