3

I've studied closed graph theorem. In particular I first saw the open mapping theorem, then the bounded inverse theorem and finally the closed graph. I saw the proof that if a linear operator is closed then is bounded. Then my notes state ($X,Y$ Banach spaces): $$ T:X\to Y \text{ linear, then } \\ T \text{ closed} \iff \textit{G}(T)\text{ closed }$$ Now I've some doubts. This statement looks very trivial to me but when I try to prove this I have necessarily to define a norm in $X \times Y$, e.g. $ ||\cdot||_{X\times Y}=||\cdot||_X+||\cdot||_Y$ and prove it, but this is not a general proof. I give definitions: \begin{align} & T \text{ is closed if } x_n\to x \text{ and } Tx_n \to y \implies y=Tx \\& G(T)=\{ (x,y)\in X \times Y: Tx=y \} \end{align} So the point is this, my notes proved: $$ T \text{ closed} \iff T \text{ bounded}$$ Then they say $$ T \text{ closed} \iff G(T) \text{ closed}$$ Hence the first theorem can be restated as $$ T \text{ bounded} \iff G(T) \text{ closed}$$ I want to clarify the second $\iff$. Moreover it concludes by saying that: $$ \text{Open mapping } \iff \text{Bounded inverse }\iff \text{Closed graph} $$ Im not sure about this. I tryed to prove that$$ \text{Closed graph} \implies \text{Open mapping }$$ But I only managed to prove that if $C$ is closed then $T(C)$ is closed. But$$ T(C^c) \ne T(C)^c$$ unless T is injective, so that I cannot conclude my proof.

Peanojr
  • 312
  • I didn't read through everything yet, but stumbled over "A closed linear Operator is bounded". That is not true, is it? Since boundedness implies continuity. The converse is true, though. – hal4math Sep 22 '19 at 18:15
  • Yes it is, $X,Y$ are Banach space. You define the norm $ ||x||{*}=||x||_X+||Tx||_Y$, and prove X is complete with respect this norm. Hence, since $||\cdot||_X \le ||\cdot||$, the two norms are equivalent by bounded inverse theorem, so $||Tx||Y \le ||x|| \le M||x||_X$, which implies continuity. – Peanojr Sep 22 '19 at 18:29
  • Ah, okay, I (miss)read the first sentence then, because you only mean that for Banach spaces. In general a closed operator needs not to be bounded. – hal4math Sep 22 '19 at 18:39
  • So, what are you not sure about when it comes to prove $T$ closed iff $G(T)$ closed? I think your defined norm on $G(T)$ is exactly the right one, btw? Is there a problem? Because I don't think that this statement holds for any random norm nor is it meant like that. – hal4math Sep 22 '19 at 18:50
  • If you state $ T \text{ bounded} \iff G(T) \text{ closed} $, what would this mean? What does it mean $G(T) \text{ closed} $? From what you say it looks the statement should be $ T \text{ bounded} \iff G(T) \text{ closed w.r.t. the norm I defined above} $. – Peanojr Sep 22 '19 at 19:58
  • Yes, I think that is precisely what it means. And yes I also think that statement is quite straight forward to proof. One has $G(T) \subset X \times Y$ and I think your choice of a norm for $X\times Y$ is the canonical one. There are of course equivalent ones. – hal4math Sep 22 '19 at 20:22
  • With that norm I think I can prove the theorem, my doubt was indeed about the generality of my proof. I keep thinking it's a strange way to state a theorem, if the thoerem referes to a particular norm it should say that, while everywhere I whatch there is no mention to a particular norm. Besides I think there are other norms for which the theorem holds. – Peanojr Sep 22 '19 at 21:06
  • Quite likely. But I still think this product topology is the most commonly (even only?!) used one, simply because it just really is so natural. Maybe see here under "In functional analysis". There the topology is also referenced. As a counterexample: I think $|\cdot|_{X\times Y} : = |\cdot|_X$ should also be a norm. And in this topology I don't think these equivalency hold anymore. – hal4math Sep 22 '19 at 21:16
  • If someone is interested, I think I managed to prove the last part. It can be proved that Closed graph $\implies $ Bounded inverse, by proving $T^{-1}$ is closed, using $T$ bijective and $T$ continuous. This implies $T^{-1}$ is continuous by closed graph theorem. $T^{-1}$ continuous implies open mapping. – Peanojr Sep 22 '19 at 21:39
  • @hal4math $|(x,y)|{X \times Y} = |x|_X$ is not a norm on the product since e.g. $|(0,y)|{X \times Y} = 0$ for $y \neq 0$. – Rhys Steele Sep 23 '19 at 10:47
  • The statement does indeed mean that $G(T)$ is closed in the product topology. This was probably not stated because usually by the time you get this far it's been stated somewhere that there are a bunch of equivalent natural norms on the product of Banach spaces that make that product again a Banach space e.g. $|(x,y)|{X \times Y} = (|x|_X^p + |y|_Y^p)^\frac{1}{p}$ for $p \in [1,\infty)$ or $|(x,y)|{X \times Y} = \max{|x|_X,|y|_Y}$. These all induce the product topology so this is always assumed in this context. – Rhys Steele Sep 23 '19 at 10:50
  • @Peanojr I am struggling to tell from the question what you are actually asking. Can you comment exactly which implications are left that you are struggling with? – Rhys Steele Sep 23 '19 at 10:53
  • @RhysSteele I managed to do the second part. What I need to prove is that $ T \text{closed} \iff \text{Graph}(T) closed$. I didnt understand your comment. You mean that this statement holds for all the norms that make $X \times Y$ a Banach space? Or equivalently that all the norms that a make $X \times Y$ a Banach space are equivalent to $||\cdot||_{X \times Y} = ||\cdot||_X +||\cdot||_Y$? – Peanojr Sep 23 '19 at 12:04
  • @RhysSteele Of course, sorry, that was too naive. I actually didn't thought about "bad" norm too much yet. You know one in which case either direction is false? – hal4math Sep 23 '19 at 12:29
  • 1
    @hal4math Off the top of my head I don't. If you want the equivalency $T$ closed iff $G(T)$ closed to fail then you at least need to pick a norm that doesn't induce the product toplogy and a reasonable constraint to add is that $X \times Y$ should still be a Banach space. If these are the only contraints you place then the problem should be hard. If I recall correctly (though I don't have a reference) it is consistent with ZF that there are no real vector spaces with inequivalent complete norms so you'll need to use choice somewhere to get a potential counterexample. – Rhys Steele Sep 23 '19 at 13:02
  • 1
    @hal4math If you additionally add the other reasonable assumption that the new norm you put on $X \times Y$ should products of open sets open then there is no such norm since a Banach space topology satisfying this assumption is the product topology (actually by an application of closed graph theorem). This means that any "reasonable" Banach space on the product space makes it true that $T$ is closed iff $G(T)$ is closed. – Rhys Steele Sep 23 '19 at 13:04
  • @RhysSteele thanks a lot for your inside! – hal4math Sep 23 '19 at 13:08

1 Answers1

3

This is an answer for the part which you say in the comments you are yet to prove. It sounds like your main source of confusion is the topology/choice of norm on the product space so let me spend some time clarifying that first.

Given topological spaces $X$ and $Y$, there is a natural topology $X \times Y$ on the product space called the product topology. In the case of normed spaces, we would like to have that this product space $X \times Y$ with the product topology is again a normed space for some norm (i.e. there is a norm which induces the product topology). There are lots of norms that do this. As I mentioned in the comments $\|(x,y)\|_p = (\|x\|^p + \|y\|^p)^{\frac{1}{p}}$ for $p \in [1, \infty)$ and $\|(x,y)\|_\infty = \max\{\|x\|,\|y\|\}$ are all examples of norms on the product space that induce the product topology (in particular, they are equivalent norms). In fact, when $X$ and $Y$ are Banach spaces then these norms also make $X \times Y$ a Banach space.

As a result of this, it is normal that when you are talking about Banach spaces and you want to define the product space $X \times Y$, then you just assume that it comes with any one of the norms I've mentioned unless the choice for some reason matters (a lot of the time it doesn't). In this case, since closedness of $G(T)$ only depends on the topology of $X \times Y$, which norm we pick of the ones mentioned definitely doesn't matter.

So what this means is that you are being asked to prove that $T$ is closed if and only if $G(T)$ is closed in the product space for any one of the norms $\|\cdot \|_p$, $p \in [1, \infty]$ on $X \times Y$. For definiteness, you can just assume that we have decided to work with $p = \infty$.

The trick is to prove the following lemma, whose proof I'll leave as an exercise here.

Lemma: $(x_n,y_n) \to (x,y)$ in $X \times Y$ if and only if $x_n \to x$ in $X$ and $y_n \to y$ in $Y$.

Now assume $T$ is closed. We want to show that $G(T)$ is closed which means we want to show that if $(x_n, Tx_n) \to (x,y)$ then $(x,y) \in G(T)$. By the Lemma, $x_n \to x$ in $X$ and $Tx_n \to y$ in $Y$ and so since $T$ is closed $y = Tx$ which means $(x,y) \in G(T)$ as desired.

Conversely, assume $G(T)$ is closed. To check that $T$ is closed we need to take a sequence $x_n$ such that $x_n \to x$ and $Tx_n \to y$ and show that $Tx = y$. By the Lemma, we know that $(x_n, Tx_n) \to (x,y)$ in $X \times Y$. But $(x_n, Tx_n) \in G(T)$ and $G(T)$ is closed so that $(x,y) \in G(T)$ also. This means exactly that $Tx = y$ as desired.

Rhys Steele
  • 19,671
  • 1
  • 19
  • 50