This question arises from one of the Berkeley Problems in Mathematics, fall 1981. It asks me to find a counterexample to the following claim:

Suppose $\lim\limits_{x\rightarrow x_{0}}f(x)=a$, $\lim\limits_{t\rightarrow a}g(t)=b$, prove or find a counterexample for the claim $\lim\limits_{x\rightarrow x_{0}}g(f(x))=b$.

I feel this should not be true, for we can select functions with infinite 'singularities' near the origin like $x\sin[1/x]$. Then for each value $x=\frac{2}{k\pi}$ we should have the value to be equal to $x$. But this is not a counterexample. Looking up the book the authors give the following example:

$f=g=1$ when $t\not=0$, $f=g=0$ when $t=0$.

The authors then claimed that $\lim_{x\rightarrow 0}g(f(x))=1$ while $\lim_{t\rightarrow 0}g=0$. Unless I am confused with the definition of the limit, that there exists $\lambda$ such that for any $\epsilon$, if $|x-x_{0}|\le \lambda$, then $f(x)-A|\le \epsilon$, I do not see why $\lim_{t\rightarrow 0}g=0$ since $t=0$ is a removable singularity for $g$. So assuming I am wrong somewhere, I want to ask:

1): why the limit of $g$ at 0 is 0?

2): is there better counterexamples?

  • 3
    $\begingroup$ Actually, it looks like the authors define $$f(t) = g(t) = \begin{cases} 0 \text{ for } t \neq 0 \\ 1 \text{ for } t = 0, \end{cases}$$ and then state that $\lim_{t \to 0} f(t) = \lim_{t \to 0} g(t) = 0$, and $\lim_{t \to 0}f(g(t)) = 1$. $\endgroup$ – Jesse Madnick Jul 22 '12 at 5:52
  • $\begingroup$ yeah, I think I copied it wrong. I see. $\endgroup$ – Bombyx mori Jul 22 '12 at 5:56

I think you just made a careless error. In the example provided, the limit of $g$ at 0 is not 0; it is 1, as you said. Here $x_0 = 0$, so that $a = 1$. So lim$_{t\rightarrow a}{g(t)}$ = lim$_{t\rightarrow 1}{g(t)} = 0$. And lim$_{x\rightarrow x_0}{g(f(x))} = $ lim$_{x\rightarrow 0}{g(f(x))} = 1$. This is probably as simple a counterexample as one could hope for.


Another counter-example that would be more familiar to a college or high-school calculus student would be the following. consider $f,g: \mathbb{R} \to \mathbb{R}$ defined

$$ \begin{array}{ll} g(x) = \frac{x^2 -1}{x-1} = \frac{(x+1)(x-1)}{x-1} \text{and}\\ f(x) = 1. \end{array} $$

Then $$ \lim_{x \to 1} g(x) = \lim_{\epsilon \to 0} \frac{(1 + \epsilon +1)(1 + \epsilon-1)}{1+ \epsilon -1} = \lim_{\epsilon \to 0} \frac{(1 + \epsilon +1)(\epsilon)}{\epsilon} = 2. $$

However $$ \lim_{x \to 1}g(f(x)) = \lim_{x \to 1}g(1) = \frac{0}{0}, $$ which is undefined.

So we need that $f(x) \neq \lim_{x \to x_0} = a$ on a small interval $(x_0 -\epsilon, x_0 + \epsilon)$ containing $x_0$ to avoid problems (How to prove this limit composition theorem?).

The counter example in the text likely meant to say $f$ is identically 0 and $g(x)= 1$ for $x \neq 0$ and $g(0) = 0$. Hopefully you can see the parallel between this example and my own. When $f(x)$ is constant around $x_0$ then the limit of $g$ is forced to equal $g(f(x_0))$.

There may be a typo in your textbook. This is very common, and it is important to do a common sense check on math problems and "read" what the author meant to say rather than the exact words on the page.


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.