24

For any topological space $X$, we can define $C(X)$ to be the commutative ring of continuous functions $f\,:\,X\rightarrow \mathbb{R}$ under pointwise addition and multiplication. Then $C(-)$ becomes a contravariant functor $C(-)\,:\,\bf{Top}\rightarrow \text{ComRing}$.

A theorem due to Gelfand and Kolmogorov states the following:

Let $X$ and $Y$ be compact Hausdorff spaces. If $C(X)$ and $C(Y)$ are isomorphic as rings, then $X$ and $Y$ are homeomorphic.

I encountered this theorem as an example in a book on homological algebra, without proof. I have searched for the proof, but have been unable to find it.

If anyone has an idea of how to prove this, or a reference to a proof, I would appreciate it greatly.

commenter
  • 1,706
Espen Nielsen
  • 3,983
  • 18
  • 32
  • 1
    You might already know this, but there is a similar statement which is known as Gelfand duality. It deals with functions to $\mathbb{C}$ and commutative $C^*$-algebras though. – Sebastian Nov 01 '12 at 12:37
  • 2
    Hist. ref.: И. М. Гельфанд, А. Н. Колмогоров, “О кольцах непрерывных функций на топологических пространствах”, Докл. АН СССР, 22:1 (1939), 11–15 – Grigory M Nov 01 '12 at 12:39
  • 2
    An english translation of (part of) the paper referenced by @Grigory's (I hope...). – Raymond Manzoni Nov 01 '12 at 12:47

4 Answers4

11

I have found a proof in "Variations on the Banach-Stone theorem" by M.I. Garrido and J.A. Jaramillo (Extracta Mathematicae, 17(3), 2002, pp. 351-383). See Theorem 18. They also have historical references.

Alex M.
  • 35,927
Nate Eldredge
  • 101,664
  • 1
    Note that the theorem in those notes uses the $\mathbb{R}$-algebra structure on $C(X)$, not just the ring structure. – Eric Wofsey Aug 06 '19 at 05:54
  • @EricWofsey: The R-algebra structure can be canonically reconstructed from the ring structure in this case. – Dmitri P. Nov 24 '19 at 16:22
  • 1
    @DmitriPavlov: I'm aware of that; just pointing out that the reference given in this answer does not fully answer the question. – Eric Wofsey Nov 24 '19 at 16:27
11

The key to the proof is the following fact.

Lemma: Let $X$ be a compact space and let $\varphi:C(X)\to\mathbb{R}$ be a ring-homomorphism. Then there exists $x\in X$ such that $\varphi(f)=f(x)$ for all $f\in C(X)$.

Proof: Note that if $f\in C(X)$ is such that $f\geq 0$ everywhere, then $\varphi(f)=\varphi(\sqrt{f})^2\geq 0$. I claim furthermore that $\varphi$ is an $\mathbb{R}$-algebra homomorphism, so $\varphi(r)=r$ for any $r\in\mathbb{R}$ (thinking of it in $C(X)$ as a constant function on $X$). Indeed, we know this must be true if $r\in\mathbb{Q}$; for arbitrary $r\in\mathbb{R}$, now use the fact that $\varphi(r-q)\geq0$ if $q\leq r$ and $q\in\mathbb{Q}$ and $\varphi(q-r)\geq 0$ if $q\geq r$ and $q\in\mathbb{Q}$.

Now suppose that $\varphi$ is not given by evaluation at any point. Then for each $x\in X$, there is a function $f_x\in C(X)$ such that $\varphi(f_x)\neq f_x(x)$. Letting $r=\varphi(f_x)$ and replacing $f_x$ with $f_x-r$, we may assume $f_x(x)\neq 0$ and $\varphi(f_x)=0$. Replacing $f_x$ with its square, we may further assume that $f_x\geq 0$ everywhere. By compactness of $X$, finitely many of the sets $\{y:f_x(y)>0\}$ cover $X$, and so adding together the corresponding $f_x$'s, we get a function $f\in C(X)$ such that $f>0$ everywhere and $\varphi(f)=0$. But then $1/f$ is continuous so $f$ is a unit and so $\varphi(f)$ cannot be $0$, so this is a contradiction.

Given this fact, the result you ask for follows easily. If $X$ is compact Hausdorff, then we can recover the set $X$ from $C(X)$ (up to canonical bijection) as the set of ring-homomorphisms $C(X)\to\mathbb{R}$. We can moreover recover the topology on $X$ since it is the coarsest topology that makes each element of $C(X)$ continuous, by Urysohn's lemma. (Here if we are identifying $X$ with homomorphisms $C(X)\to\mathbb{R}$, we can think of an element of $C(X)$ as a function on $X$ by evaluation.) So we can recover the space $X$ up to homeomorphism from the ring $C(X)$.

(In fact, it similarly follows from the Lemma that if $X$ and $Y$ are compact Hausdorff, then ring-homomorphisms $C(X)\to C(Y)$ are naturally in bijection with continuous maps $Y\to X$, and this preserves composition. So this gives a contravariant equivalence of categories between compact Hausdorff spaces and rings of the form $C(X)$.)

Eric Wofsey
  • 342,377
  • Hey Eric, can this approach be made to work when one considers complex-valued functions instead? – Martin Argerami Nov 17 '20 at 01:55
  • To prove what result, exactly? That if $C(X,\mathbb{C})\cong C(Y,\mathbb{C})$ as rings then $X\cong Y$ (for compact Hausdorff spaces $X,Y$)? – Eric Wofsey Nov 17 '20 at 02:08
  • 1
    The issue with that is that the direct analogue of my Lemma is horribly false over $\mathbb{C}$ (for instance, even in the case where $X$ is a single point). But you can replace the Lemma with a different Lemma that says every maximal ideal is the set of functions that vanish at some point, and the proof is very similar. – Eric Wofsey Nov 17 '20 at 02:16
  • 1
    The remainder of the proof using the Lemma is then similar, except that you have to determine the topology on $X$ (identified with the maximal ideals of $C(X,\mathbb{C})$) differently, since you don't know what the evaluation maps are. However, the closed sets of a compact Hausdorff space are generated by the zero sets of elements of $C(X,\mathbb{C})$, so you can recover $X$ topologically as the maximal spectrum of $C(X,\mathbb{C})$ with the Zariski topology. – Eric Wofsey Nov 17 '20 at 02:18
  • That's great, thanks! – Martin Argerami Nov 17 '20 at 02:23
  • Yes. It was only today that I was trying to write this carefully and I run into the problem of having additivity and not linearity. Then I noticed that Dugundji only considers the real case, and then I ended up here :D – Martin Argerami Nov 17 '20 at 02:32
  • Alternatively, if you talk about homomorphisms of $\mathbb{C}$-algebras rather than just of rings, the Lemma remains true and you still have the nice equivalence of categories. The proof of the Lemma needs to be changed though (unless you restrict further to $*$-homomorphisms in which case it can stay basically the same): you have to first prove that every maximal ideal corresponds to a point as I mentioned above, and then use the fact that the kernel of an algebra homomorphism to $\mathbb{C}$ will be a maximal ideal. – Eric Wofsey Nov 17 '20 at 02:35
9

Dugundji's Topology has a very short, readable proof.

you can find the proof on page 289. its very readable.

Ameyah
  • 57
6

Gillman-Jerison, Rings of continuous functions, Theorem 7.3.