next up previous
Next: The Klein-Millman Theorem Up: Applications of Weak Topologies Previous: Sobcyzk's Theorem

The Stone-Cech Compactification

In general, a compactification of a topological space $ X$ is a pair $ (K,i)$ where $ K$ is a compact topological space and $ i:X \rightarrow K$ is a homeomorphism onto its range. The one point compactification is the smallest. The Stone-Cech compactification is, in a suitable sense, the largest. Its defining property is that for all $ f \in C_b(X)$, then $ f$-extends to $ g \in C(K)$.

Example: Consider $ X = \mathbb{R}$. The one point compactification is homeomorphic to $ S^1$.
Consider $ \arctan$ in $ \mathbb{R}$, a bounded continuous function. It does not extend to a continuous function on the one-point compactification since $ \lim_{x \rightarrow -\infty} \arctan(x) \neq \lim_{x \rightarrow +\infty} \arctan(x)$. To extend $ \arctan$, we need at least a two point compactification, which is homeomorphic to $ [0,1]$.
Consider $ \sin$ on $ \mathbb{R}$. Call $ g$ the extension of $ \sin$ to the Stone-Cech compactification. Since $ \sin$ vanishes on $ \{n \pi : n \in \mathbb{N}\}$, $ g^{-1}(0)$ is a compact subset of the compactification and so $ (n \pi)_{n \in \mathbb{N}}$ has a limit point, $ x$.
Consider $ ((2n+\frac12)\pi)_{n \in \mathbb{N}} \subset \sin^{-1}(1)$. Similarly, this has a limit point $ y$. Since

$\displaystyle g(x) = \lim_{n \rightarrow \infty} g(n\pi) $

$\displaystyle g(y) = \lim_{n \rightarrow \infty} g((2n+\frac12)\pi) = 1 $

and $ x \neq y$. In this manner, we can create countably many new limit points ``at infinity." So instead of just adding one new point, we are adding at least a copy of $ [-1,1]$.
Consider $ \sin(x^2), \sin(x^3), \ldots$.
So, the Stone-Cech compactification will be very large, even for ``natural" spaces like $ \mathbb{R}$. However, there are a few examples where the one-point compactification and the Stone-Cech compactification are the same.

We construct the Stone-Cech compactification using a weak topology on part of the dual of $ C_b(X)$.

Let $ (X,\tau)$ be a topological space. For $ x \in X$, define $ \delta_x \in C(X)^*$ by $ \delta_X(f) = f(x)$. Clearly, $ \Vert\delta_x\Vert = 1$, $ \delta_x$ is linear. Define $ \Delta:X \rightarrow C_b(X)^*$ by $ \Delta(x) = \delta_x$.

Claim: $ \Delta$ is $ (\tau,$   wk $ ^*)$-continuous. If $ x_i \rightarrow x_0$ in $ X$, then $ f(x_i) \rightarrow f(x_0)$ for all $ f \in C_b(X)$. That is, $ \delta_{x_i}(f) \rightarrow \delta_{x_0}(f)$ for all $ f$. That is, $ \delta_{x_i} \rightarrow \delta_{x_0}$ in the wk $ ^*$ topology.

Proposition 3.2.1   $ \Delta$ is a homeomorphism onto $ \Delta(X)$ iff $ X$ is completely regular.

Proof.
$ \Leftarrow$
$ \Delta$ is injective.
If $ x_1 \neq x_2$, then % latex2html id marker 14584
$ \exists f \in C_b(X)$ such that $ f(x_1) \neq f(x_2)$. Hence, $ \delta_{x_1}(f) \neq \delta_{x_2}(f)$, so $ \delta_{x_1} \neq \delta_{x_2}$.
$ \Delta$ is an open map
Let $ U \subset X$ be open, $ x_0 \in U$. It suffices to find $ M \subset \Delta(U)$ with $ \Delta(x_0) \in M$ and $ M$ open. By complete regularity, % latex2html id marker 14604
$ \exists f \in C_b(X)$ such that $ f(x_0) = 1$, supp $ f \subset \bar U$. Let $ M_1 = \{ \mu \in C_b(X)^* : \mu(f) > 0 \}$, a wk $ ^*$-open set. To see this, observe

$\displaystyle M_1 = \hat f ( (0,+\infty) ) $

in $ C_b(X)^{**}$. Since $ \hat f$ is wk $ ^*$-cts, $ M_1$ is wk $ ^*$-open.
Let $ M = M_1 \cap \Delta(X)$, a relatively wk $ ^*$-open set. This set shows the desired properties noted above, so $ \Delta$ is an open map.
$ \Rightarrow$
As a compact Hausdorff space, $ ($ball $ C_b(X)^*,$   wk $ ^*)$ is normal and normal implies completely regular.
Since supsets of completely regular topological spaces (with the relative topology) are completely regular, $ \Delta(X)$ is completely regular. Since $ \Delta$ is a homeomorphism, $ X$ is completely regular.
$ \qedsymbol$

Theorem 3.2.2   Stone-Cech Compactification
If $ X$ is completely regular, then there is a unique (up to homeomorphism) compactification of $ X$, $ \beta X$ such that every $ f \in C_b(X)$ extends to an element of $ C(\beta X)$.

Proof. Existence
Let $ \beta X =$   wk $ ^*-$   cl $ \Delta(X)$. By Alaoglu, $ \beta X$ is wk $ ^*$-compact. By the proposition, $ \Delta :X \rightarrow \Delta(X)$ is a homeomorphism. Since $ \beta X$ is the closure of $ \Delta(X)$, the image of $ X$ is dense in $ \beta X$.
Let $ f \in C_b(X)$. As $ \beta X \subset C_b(X)^*$, define $ g \in C(\beta X)$ by $ g(\phi) = \phi(f)$. If $ \phi_i \rightarrow \phi$ wk $ ^*$, $ \phi_i(f) \rightarrow \phi(f)$, so $ g(\phi_i) \rightarrow g(\phi)$, so $ g$ is continuous. If $ \phi = \delta_x$, then $ g(\delta_x) = \delta_x(f) = f(x)$, so $ g$ extends $ f$.
Uniqueness
Let $ \Omega$ be a compactification of $ X$, i.e., $ \Omega$ is compact, there is $ \Pi:X \rightarrow \Omega$ a homeomorphism onto $ \Pi(X)$, and $ \Pi(X)$ is dense in $ \Omega$ such that $ \forall f \in C_b(X)$, there exists $ \tilde f \in C(\Omega)$ such that $ \tilde f \circ \Pi = f$.
Define $ g: \Delta(X) \rightarrow \Pi(X)$ by $ g = \Pi \circ \Delta^{-1}$, a homeomorphism.
Claim:
$ g$ extends to a homeomorphism $ G: \beta X \rightarrow \Omega$. The text proves this directly. We'll prove it using the Banach-Stone theorem, which says, in part, if $ X, Y$ are compact Hausdorff spaces and there is an isometric isomorphism $ \Phi$ from $ C(X)$ to $ C(Y)$, then $ X$ is homeomorphic to $ Y$ and there is a homeomorphism $ \phi : X \rightarrow Y$ such that

$\displaystyle \Phi(f) = \alpha f \circ \phi^{-1}, \forall f \in C(X) $

where $ \alpha : Y \rightarrow \mathbb{T}$ is continuous ( $ \mathbb{T}:= \{ z \in \mathbb{F}: \vert z\vert = 1 \}$).
$ g$ gives an isometric isomorphism $ h:C(\Delta(X)) \rightarrow C(\Pi(X))$ by $ h(f) = f \circ g^{-1}$ for all $ f \in C(\Delta(X))$. By density, for $ f \in C(\beta X)$, $ f\vert _{\Delta(X)}$ determines $ f$. I.e., given $ f_1, f_2 \in C(\beta X)$ such that $ f_1\vert _{\Delta(X)} = f_2\vert _{\Delta(X)}$ then $ f_1 = f_2$ Similarly for $ f_1, f_2 \in C(\Omega)$ and $ f_1\vert _{\Pi(X)} = f_2\vert _{\Pi(X)}$.
Thus we have isometric isomorphisms:

$\displaystyle \Phi : C(\beta X) \rightarrow C_b( \Delta(X) ) $

$\displaystyle \Psi : C(\Omega) \rightarrow C_b( \Pi(X) ) $

To use Banach-Stone, we need an isomorphism between $ C(\beta X)$ and $ C(\Omega)$. (Commutative diagram). Thus, $ \Lambda = \Psi^{-1} \circ h \circ \Phi$ is the isometric isomorphism we desire. By Banach-Stone, $ \beta X$ and $ \Omega$ are homeomorphic.
If you use the specific form of the homeomorphism from the Banach-Stone theorem, it is possible to show that the homeomorphism extends $ g$.
$ \qedsymbol$

Aside: For a completely regular topological space $ X$, $ C_b(X)$ is separable iff $ X$ is a compact metric space.


next up previous
Next: The Klein-Millman Theorem Up: Applications of Weak Topologies Previous: Sobcyzk's Theorem
Brian Bockelman 2006-04-21