9
$\begingroup$

I tried looking up a question regarding graphs of continuous functions on this site, but all the ones I found consider functions from $\mathbb{R}$ into $\mathbb{R}$. I have been pondering the following question: given a general topological spaces $X, Y$, and a function $f: X\to Y$, when does $Graph(f)$ closed in $X\times Y$ imply that $f$ is continuous. By the closed graph theorem, this is true whenever $X$ and $Y$ are both Banach spaces.

Also, it is fairly easy to prove that whenever $Y$ is a Hausdorff space and $f$ is continuous, then $Graph(f)$ is closed, but I do not think that the converse is true, so I am trying to find an example where $X$ is some topological space, $Y$ a Hausdorff space, $f: X\to Y$ a function with a closed graph in $X\times Y$, but who fails to be continuous. As of yet I have not been able to find such a counterexample, partially because I have no clue where to look for such a counterexample. I would really appreciate getting some directions to go in.

$\endgroup$
  • 1
    $\begingroup$ Hint: consider the derivative as a function of $C^1[0, 1]$ to $C^0[0, 1]$. This is a linear, unbounded function (hence not continuous), but you can show that its graph is closed. $\endgroup$ – student Jan 12 '12 at 18:12
  • $\begingroup$ @Leandro: what norm are you using on $C^1[0,1]$? If it's $\lVert f'\rVert + \lVert f\rVert$, then $(f_n)$ converges in $C^1[0,1]$ iff both $(f_n)$ and $(f_n')$ converge uniformly on $[0,1]$ $\endgroup$ – kahen Jan 12 '12 at 18:19
  • $\begingroup$ @kahen: the norm on $C^1[0, 1]$ should be just $\|f\|$ (the supremum norm) for this example. $\endgroup$ – student Jan 12 '12 at 18:21
  • 3
    $\begingroup$ Your Banach space example needs the hypothesis of "linear" also. $\endgroup$ – GEdgar Jan 12 '12 at 18:22
24
$\begingroup$

$f\colon\mathbb{R}\to\mathbb{R}$ defined by $$ f(x)=\frac1x\quad\text{if}\quad x\ne0,\quad f(0)=0. $$

$\endgroup$
  • $\begingroup$ What if we want $X$ to be compact? ie, I need a discontinuous function $f:X \longrightarrow \mathbb{R} $, whose graph is closed. $\endgroup$ – Janson A.J Nov 17 '15 at 16:55
  • 1
    $\begingroup$ Ohh.. We just have to take the same function on $[-1,1]$.. $\endgroup$ – Janson A.J Nov 17 '15 at 17:03
  • $\begingroup$ @JansonA.J $X=[0,1]$, same $f$. $\endgroup$ – Julián Aguirre Nov 17 '15 at 17:05
  • $\begingroup$ @JuliánAguirre, how do you know the graph is closed? Can you prove that? $\endgroup$ – Kamil Aug 1 '16 at 8:41
  • $\begingroup$ Why is it necessary that $f(0)=0$? Does $f(x)=1/x$ not work as a counter-example on its own? $\endgroup$ – jackson5 Jan 24 '17 at 7:37
3
$\begingroup$

I'd like to answer the comment about why the graph is closed (particularly as it also took me a while to see it).

To see that the graph of $f\colon\mathbb{R}\to\mathbb{R}$ defined by $$ f(x)=\frac1x\quad\text{if}\quad x\ne0,\quad f(0)=0. $$ is closed consider the equivalent definition of a closed graph in a metric space. $G(f)$ is closed if whenever $(x_n) \to x$ and $(f(x_n)) \to y$ then $y = f(x)$. (This follows since in a metric space closure is equivalent to sequential closure and convergence in the product space is equivalent to convergence of the components)

Now for $(x_n) \to 0$ then $(f(x_n))$ is not convergent so $(x_n) \to x$ and $(f(x_n)) \to y$ is not satisfied, while for every other sequence $(x_n) \to x \ne 0$ then $(f(x_n)) \to f(x)$.

So it is true that whenever $(x_n) \to x$ and $(f(x_n)) \to y$ then $y = f(x)$ and therefore $G(f) $ is closed.

$\endgroup$
  • $\begingroup$ @Kamil Here's a proof. $\endgroup$ – Tom Collinge Aug 22 '16 at 10:36
  • 1
    $\begingroup$ Alternatively: The set in question is a finite union of closed sets: The origin (a singleton) and the level set $xy = 1$ (which is clearly closed since $f(x, y) = xy$ is continuous). :) $\endgroup$ – Andrew D. Hwang Aug 22 '16 at 10:44
  • $\begingroup$ @AndrewD.Hwang. Thanks. I'm a little confused about $f(x, y) = xy = 1$ since it isn't defined on all of $X \times Y$, i.e. not defined at $x = 0$ (or at $y = 0$). $\endgroup$ – Tom Collinge Aug 22 '16 at 11:58
  • 1
    $\begingroup$ The function $f(x, y) = xy$ is a quadratic polynomial in two variables, hence defined (and continuous) everywhere. You're perfectly correct that the level set $f(x, y) = 1$ doesn't touch the coordinate axes, but that's not the same as saying $f$ is undefined along the axes. ;) $\endgroup$ – Andrew D. Hwang Aug 22 '16 at 12:07

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.