Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

When is it that we can plug in the limit point into the function to evaluate a limit?

I believe for real limit points, we can do this when the function is continuous at the point. But what about for limits at $ \infty $ and $ -\infty $? Is there a general statement that we can make about this?

Example: It is easy to believe that the following limit evaluates to $ 0 $ $$ \lim_{n\to\infty}{\sqrt{n^2+n}-n} $$

How can we not fall into these traps?

EDIT: Yes, the above limit is simple to evaluate correctly (it's $ 1/2 $). But what I'm asking is for a rigorous condition under which we can know whether or not we can naively plug in the value to get the correct answer.

share|cite|improve this question
There is a number system called "Riemann sphere" where "plug in $\infty$" can be done. But even in that case, $\infty - \infty$ is undefined. Or (as they say in calculus) indeterminate. – GEdgar Jul 25 '11 at 14:46
@tskuzzy: The problem here is not "plugging in" or "not plugging in"; the problem here is that you get an indeterminate form, namely $\infty-\infty$. If you don't get an indeterminate form, then you're fine. For example, $\lim\limits_{n\to\infty}\sqrt{1+\frac{1}{n}} - n$ does evaluate to "$1-\infty=-\infty$", because $c-\infty$ is not indeterminate. – Arturo Magidin Jul 25 '11 at 16:44
@Arturo: What about the case where the indeterminate form is encountered "at the end"? For example, the limit laws apply to $$\lim_{n\to\infty}\left({n\sqrt{1+\frac{1}{n}}-n}\right)$$ which equals $\infty\cdot 1 - \infty$. Why does one not conclude that the limit does not exist in this case? Does an indeterminate form always mean that another approach needs to be taken? – raxacoricofallapatorius Sep 20 '11 at 22:12
@raxacoricofallapatorius: No, the limit law that says that the limit of a difference is the difference of the limits when they both exist most definitely does not apply to that limit. What makes you think it does? – Arturo Magidin Sep 21 '11 at 4:15
@Arturo: Working outward: $1/n\to0$, so $\sqrt{{1}+{1/n}}\to 1$, so $n\sqrt{{1}+{1/n}}\to \infty \cdot 1$. Where am I going wrong? Is it because $L=\infty$ isn't really a limit in the sense of the limit law? – raxacoricofallapatorius Sep 21 '11 at 4:32
up vote 13 down vote accepted

First, I don't think you actually asked what you meant to ask. I'll answer what you actually asked, and then comment on what I think you meant to ask.

You are correct that if $\lim\limits_{x\to a}f(x) = k\in\mathbb{R}$ and $g$ is continuous at $k$, then $$\lim_{x\to a}\;g(f(x)) = g\left(\lim_{x\to a}f(x)\right) = g(k);$$ and this also holds if $a$ is replaced by $\infty$ or $-\infty$.

It gets a bit trickier if the limit of $f$ is $\infty$ or $-\infty$; for one thing, it no longer makes sense to write $g(\lim\limits_{x\to a}f(x))$, so instead we must talk about $\lim\limits_{t\to\infty}g(t)$. But even with that correction, we do not have an easy criteria for when it makes sense to "plug in"; here are two propositions and one example along those lines.

Proposition 1. If $\lim\limits_{x\to a}f(x) = \infty$, and $\lim\limits_{t\to\infty}g(t) = k\in\mathbb{R}\cup\{\infty,-\infty\}$, then $\lim\limits_{x\to a}\;g(f(x))=k$.

Proof. Suppose $\lim\limits_{t\to\infty}g(t) = k$. First, take $k\in\mathbb{R}$. We show that $\lim\limits_{x\to a}\;g(f(x)) = k$. Let $\epsilon\gt 0$. Then there exists $M\gt 0$ such that if $t\gt M$, then $|g(t)-k|\lt \epsilon$. Since $\lim\limits_{x\to a}f(x) = \infty$, there exists $\delta\gt 0$ such that if $0\lt |x-a|\lt \delta$, then $f(x)\gt M$. Thus, if $0\lt |x-a|\lt \delta$, then $f(x)\gt M$, hence $|g(f(x))-k|\lt\epsilon$. Therefore, $\lim\limits_{x\to a}\;g(f(x)) = k$, as claimed.

Now consider the case $k=\infty$. Let $N\gt 0$; then there exists $M\gt 0$ such that if $t\gt M$, then $g(t)\gt N$. Since $\lim\limits_{x\to a}f(x)=\infty$, there exists $\delta\gt 0$ such that if $0\lt |x-a|\lt \delta$, then $f(x)\gt M$; hence $g(f(x))\gt N$ whenever $0\lt |x-a|\lt \delta$, proving that $\lim\limits_{x\to a}\;g(f(x))=\infty$. The same argument holds if $k=-\infty$. QED

The result also holds if $a$ is replaced with $\infty$ or $-\infty$.

The converse doesn't work in general:

Example: Functions $f$ and $g$, with $g$ continuous everywhere (in fact, uniformly continuous), $\lim\limits_{x\to a}f(x)=\infty$, $\lim\limits_{x\to a} \;g(f(x))$ exists, and $\lim\limits_{t\to\infty}g(t)$ does not exist.

Let $f(x) = \lfloor x\rfloor\pi$, $g(t)=\sin(t)$, $a=\infty$. Then $\lim\limits_{x\to \infty}\;g(f(x))=0$, but $\lim\limits_{t\to\infty}g(t)$ does not exist, nor is it equal to $\infty$ or $-\infty$. It is not hard to come up with similar examples with $a\in\mathbb{R}$. QED

On the other hand, with a bit more information, we have:

Proposition 2. Let $f$ and $g$ be functions, and assume that:

  1. $\lim\limits_{x\to a}f(x)=\infty$; and
  2. For every $\delta\gt 0$ there exists $M_{\delta}\gt 0$ such that $(M_{\delta},\infty)\subseteq f\Bigl( (a-\delta,a)\cup(a,a+\delta)\Bigr)$.

If $\lim\limits_{x\to a}\;g(f(x))=k\in\mathbb{R}\cup\{\infty,-\infty\}$, then $\lim\limits_{t\to\infty}g(t) = k$.

Proof. We do the case $k\in\mathbb{R}$; the other two cases are similar. Let $\epsilon\gt 0$. Then there exists $\delta\gt 0$ such that if $0\lt|x-a|\lt \delta$, then $|g(f(x))-k|\lt\epsilon$.

By (2), there exists $M_{\delta}$ such if $t\gt M_{\delta}$, then there exists $x\in (a-\delta,a)\cup(a,a+\delta)$ with $f(x)=t$.

Therefore, if $t\gt M$, then there exists $x$, $0\lt |x-a|\lt \delta$ such that $g(t) = g(f(x))$, hence $|g(t)-k| = |g(f(x))-k|\lt \epsilon$, with the last inequality by choice of $\delta$. Therefore, $\lim\limits_{t\to\infty}g(t)=k$, as claimed. QED

Similar results hold if we replace $a$ with $\infty$ or $-\infty$.

Corollary. Let $f$ and $g$ be functions with $\lim\limits_{x\to a}f(x) =\infty$ and $\lim\limits_{x\to a}\;g(f(x))=k\in\mathbb{R}\cup\{\infty,-\infty\}$. If $f$ is continuous on an open punctured neighborhood of $a$, then $\lim\limits_{t\to\infty}\;g(t)=k$.

Proof. If $f$ is continuous on an open punctured neighborhood of $a$, then the condition that $f(x)\to\infty$ as $x\to a$ ensures that $f$ satisfies condition 2 of Proposition 2. QED

In fact, it's enough that $f$ be continuous on either an interval of the form $(a-\delta_f,a)$, or on an interval of the form $(a,a+\delta_f)$.

Continuity can even be weakened further:

Corollary. Let $f$ and $g$ be functions with $\lim\limits_{x\to a}f(x)=\infty$ and $\lim\limits_{x\to a}\;g(f(x))=k\in\mathbb{R}\cup\{\infty,-\infty\}$. If there exists $\delta_f\gt 0$ such that $f$ has the intermediate value property on either $(a-\delta_f,a)$ or on $(a,a+\delta_f)$, then $\lim\limits_{t\to \infty} g(t) = k$.

Proof. Assume that $f$ has the intermediate value property on $(a-\delta_f,a)$; the other case is similar. We prove that $f$ satisfies condition (2) of Proposition 2.

Let $\delta\gt 0$; we may assume that $\delta\leq\delta_f$.

Let $M=f(a-\frac{\delta}{2})$; let $z\gt M$. Since $\lim\limits_{x\to a}\;f(x)=\infty$, there exist $x_1$, $a-\frac{\delta}{2}\lt x_1 \lt a$ such that $f(x_1)\gt z$. Since $f$ has the intermediate value property on $(a-\delta,a)$ and $f(a-\frac{\delta}{2})\lt z\lt f(x_1)$, we can find $x_2\in (a-\frac{\delta}{2},x_1)$ such that $f(x_2)= z$; thus, $z\in f(a-\delta,a)$. This proves that for all $z\gt M$, $z\in f(a-\delta,a)$, hence $(M,\infty)$ is contained in the image and Condition (2) of Proposition 2 holds; conclusion of the corollary now follows from Proposition 2. QED

Note that in my example above, $f$ does not have the intermediate value property, which is what leads to problems.

That said: the real issue with your example lies in the indeterminate form $\infty-\infty$, rather than a problem with "plugging in", I think.

Instead, what you are dealing with here is an attempt at applying the limit laws that say, for example, that the limit of a difference is the difference of the limits:

If $\lim\limits_{x\to a}f(x) = L$ and $\lim\limits_{x\to a}g(x)=M$, then $\lim\limits_{x\to a}\Bigl(f(x)-g(x)\Bigr) = L-M$.

If we try to extend this (and related results on sums, products, and quotients) to the cases where limits are equal to $\infty$ or $-\infty$, then you have to deal with the extended reals, $\mathbb{R}\cup\{\infty,-\infty\}$. The problem here is that not every operation among extended reals is well-defined. Some operations are well defined and the limits laws work for them:

  • $\infty+a = \infty$ for all real numbers $a$;
  • $-\infty+a = -\infty$ for all real numbers $a$;
  • $\pm\infty\times a =\pm\infty$ if $a\gt 0$ or $a=\infty$;
  • $\pm\infty\times a = \mp\infty$ if $a\lt 0$ or $a=-\infty$;
  • $\frac{\pm\infty}{a} = \pm\infty$ if $a\gt 0$;
  • $\frac{\pm\infty}{a} = \mp\infty$ if $a\lt 0$;
  • $\frac{a}{\pm\infty} = 0$.

But other expressions are indeterminate forms: the limit laws don't apply to them. They don't apply to expressions like $\infty-\infty$ or $\frac{\infty}{\infty}$ (just like they don't apply in the real numbers to expressions like $\frac{a}{0}$).

So the problem is not one of "plugging in", but rather that the mistake you mention lies in trying to apply the limit laws and forgetting that $\infty-\infty$ is an indeterminate form and not an operation that can be performed.

share|cite|improve this answer
(+1) for actually answering the question! – The Chaz 2.0 Jul 26 '11 at 1:14
+1 yet another amazing answer. Wish I could do +more, for what it's worth ($\epsilon $) – Ross Millikan Jul 26 '11 at 4:19
@The Chaz: (and Arturo) Sorry, I menat the rep is worth $\epsilon$. The answer is worth much more. – Ross Millikan Jul 26 '11 at 13:04
I keep seeing these excerpts of the book "Everything I Wished I was Taught in Math" by Arturo Magidin, yet I can't seem to find the book anywhere... when is it coming out, and if it does come out, where can I buy a copy? – J. M. Jul 27 '11 at 3:10
@J.M. Well, as my just-completed edit shows, it needs to go through several rounds of refereeing... – Arturo Magidin Jul 27 '11 at 3:12

For $x\to\pm\infty$, consider the change of variables $u=1/x$ and let $u\to 0^{\pm}$.

share|cite|improve this answer
That's a method of evaluating such limits. But how can we use this to know that we can simply plug in $ 0^\pm $? – tskuzzy Jul 25 '11 at 14:20
@tskuzzy, you can when the function $f(u)$ is continuous at 0. – lhf Jul 25 '11 at 14:50
Are you trying to say "never"? Because $ f(u) $ is always undefined at 0 and thus discontinuous. – tskuzzy Jul 25 '11 at 14:52
The function is continuous at 0 in this case is a shorthand phrase for: there exists a continuous extension to 0. – Florian Jul 25 '11 at 14:58
@tskuzzy: What Ihf means is that even if the function is not defined at zero, thus discontinuous, the disctontinuity might be removable...Keep in mind that after you do some algebraic manipulations/simplifications sometimes you obtain expressions which are defined at 0. – N. S. Jul 25 '11 at 15:14

In this case its rather easy to see if you notice that

$$\sqrt{n^2+n}-n=\sqrt{n^2+n+\frac{1}{4}-\frac{1}{4}}-n = \sqrt{(n+\frac{1}{2})^2-\frac{1}{4}}-n$$

In general, I think the best thing to do is to expand it into a polynomial, and look at what the resulting structure looks like.

share|cite|improve this answer
I know the correct way of evaluating the above limit. But how can we avoid evaluating it the wrong way? Under what conditions can we rigorously show that a limit can/cannot be evaluated by plugging in the value? – tskuzzy Jul 25 '11 at 14:24
I sort of mentioned it later in my statement. You would want to expand it and see what the terms are. In this case you would have something like $n+\frac{1}{2}+... - n$ – picakhu Jul 25 '11 at 14:26
To me, that still seems more like an algorithm or a heuristic to evaluate it. But what if I gave you the limit $ \frac{n+0.5}{n} $? Clearly it's 0 via plugging in $ \infty $. In general, why is it that we can do this for certain classes of limits? – tskuzzy Jul 25 '11 at 14:29
you cannot in general "plug in" $\infty$, you would have you apply l'hopitals in that case to know (if its a fraction) – picakhu Jul 25 '11 at 14:32
@tskuzzy: I imagine you meant two comments ago that the limit was clearly $1$. "Plugging in" $\infty$ will almost never work. One could come up with a small number of types of problem in which it will work, but they will almost never be useful, and when they are useful, correct methods are very fast also. – André Nicolas Jul 25 '11 at 15:04

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.