26 Aug 2022 -

There is an isomorphism theorem of Gelfand, detailed in the next paragraph. It ends up giving an (anti)-equivalence of categories between CC^\ast-algebras and the category of locally compact haussdorf spaces. It gives rise to the philosophy that CC^\ast algebras are really the study of "non-commutative topology". One of the ideas here is that phenomena in topology should have their counterparts in CC^\ast-algebras. One of the most important examples of this is the extension of KK-theory to CC^\ast-algebras. Today, we do something simpler but still very much striking. We talk about compactifications and their analogues, and how we can use this interplay between topology and CC^\ast-algebras to learn more about compactifications.

The Gelfand IsomorphismPermalink

Recall a Banach space is a normed C\bb{C}-vector space complete under the induced metric. A Banach algebra in turn is a Banach space with a multiplication, such that  ⁣ab ⁣ ⁣a ⁣ ⁣b ⁣\norm{ab}\leq \norm{a}\norm{b}. The simplest such objects to study are the abelian Banach algebras. For each unital abelian Banach algebra AA, take Spec(A)\Spec(A) to be the space of characters, i.e homomorphisms AγCA\lra{\gamma}\bb{C}. We can give Spec(A)\Spec(A) the weak*-topology and its easy to check due after applying the Banach-Alaoglu theorem that this is infact a compact space. Now we get a representation Γ:AC(Spec(A)),Γ(a)(γ)=γ(a)\Gamma: A \lra{} C(\Spec(A)),\quad \Gamma(a)(\gamma) = \gamma(a) The norm on C(X)C(X) for compact haussdorf XX, the space of continous functions XCX\lra{}\bb{C}, is  ⁣f ⁣=supxXf(x)\norm{f}_\infty= \mathrm{sup}_{x\in X} |f(x)|. We call this the Gelfand representation. Note we could have asked for AA to be non-unital, and then the character space would have been locally compact haussdorf. We would have to use C0(Spec(A))C_0(\Spec(A)) then, the space of continous functions like before but they go to zero at infinity (think about infinity by thinking about the one point compactification.).

A CC^\ast algebra is an algebra with an involution \ast such that  ⁣aa ⁣= ⁣a ⁣2\norm{a^\ast a} = \norm{a}^2. Note that if pointwise complex conjugation is the involution, both C(Y)C(Y) and C0(X)C_0(X) are CC^\ast-algebras.Now the big theorem due to Gelfand is that for commutative CC^\ast algebras, the Gelfand representation is an isomorphism. This tells us each non-unital commutative CC^\ast-algebra is dual to some LCH space, and that every unital commutative is dual to some compact haussdorf space. A lot of CC^\ast theory now is to see how much of this isomorphism theorem we can translate over to the non-commutative case. So in a sense CC^\ast-algebra is the study of non-commutative topology. With that being said, we expect concepts in topology to carry over to CC^\ast-algebras and vice versa. We will see that for compactification today.

I should mention that this really is an (anti-)equivalence of categories between commutative CC^\ast algebras and the category of LCH spaces with continuous maps. Send a space to its C0C_0 and a commutative algebra to its spectrum.

Unitizations and CompactificationsPermalink

We shall say compact YY is a compactification of LCH XX (non-compact) if

  • There is an embedding ι:XY\iota: X \hookrightarrow Y
  • The image ι(X)\iota(X) is dense in YY
  • The easiest example of this is αX\opc{X}, the one point compactification. This is just adding a single point to XX, call it \infty, to make it compact. We can also write it by its universal property, that any compactification of XX maps into αX\opc{X}, i.e αX\opc{X} is the "smallest" compactificaiton.

    As mentioned before, the philosophy is that every topological concept should have an analogue in operator theory. So lets look at C(αX)C(\opc{X}) towards that end. Notice that we have a map C(αX)C0(X)CC(\opc{X})\lra{\cong} C_0(X) \oplus \bb{C} sending f((ff())X,f())f\mapsto ((f-f(\infty))\mid_{X},f(\infty)). It is easy to see this is an isomorphism. So adjoining a unit like this is the analogue of the one-point compactification. Lets try to find the analogues for compactifications in general. There is a unique norm on each CC^\ast-algebra that satisfies the CC^\ast identity, I leave it as an exercise to figure out the norm on C0(X)CC_0(X)\oplus \bb{C} making it a CC^\ast-algebra.

    Let XιYX\lra{\iota}Y be a compactification of XX. Consider fC0(X)f\in C_0(X), then it extends to a αfC(αX)\opc{f} \in C(\opc{X}). We can combine the map YαXY\lra{}\opc{X} given by the universal property of αX\opc{X} and αf\opc{f} to get an element of C(Y)C(Y).

    Hence we have defined an embedding C0(X)C(Y)C_0(X) \lra{} C(Y). Note that this entire arguement only required YY to be compact, i.e we did not use the density of XX in YY. Here is how we translate that, take an ideal 0IC(Y)0\neq I\subset C(Y), then notice that C0(X)C(Y)0C_0(X)\cap C(Y)\neq 0. This is because if gC(Y)g\in C(Y), then gC0(X)=0gC_0(X)=0 would mean gg is 00 on XX and hence by density, on YY. So we collect these to get the dual concept of compactification for CC^\ast algebras.

    A unitization of a CC^\ast-algebra AA is:

  • An embedding into an unital CC^\ast-algebra BB, i.e ABA\lra{}B
  • the image of AA in BB is a (two-sided) essential ideal, i.e for every ideal IBI\subset B, AI0A\cap I \neq 0.
  • I have already shown compactifications lead to unitizations.I leave it as an exercise to show the converse.

    Multiplier Algebras.Permalink

    The Stone-Čech compactification of a LCH space XX, βX\beta X, is the biggest compactification of XX. As we saw above, we expect this to correspond to the biggest unitization of a non-unital CC^\ast-algebra AA. The advantage of thinking in these terms is that for CC^\ast algebras this is quite explicit. Indeed, the biggest unitization of AA, called M(A)M(A)(its multiplier algebra) is given as follows:

    M(A)M(A) is the set of double centralizers (L,R)(L,R) on AA. This means:

  • L,RL,R are bounded linear maps AAA\lra{} A
  • aL(b)=R(a)baL(b) = R(a)b
  • λ(L,R)+(L,R)=(λL+L,λR+R)\lambda(L,R) +(L',R') = (\lambda L+L', \lambda R+R') for λC\lambda\in \bb{C}
  • (L,R)(L,R)=(LL,RR)(L,R)(L',R')=(LL',R'R)
  • Its unit is (1A,1A)(1_A,1_A)
  • One can check from this that under the operator norm,  ⁣L ⁣= ⁣R ⁣\norm{L}=\norm{R}, so naturally we define the norm  ⁣(L,R) ⁣= ⁣L ⁣\norm{(L,R)} = \norm{L}
  • (L,R)=(R,L)(L,R)^\ast= (R^\ast, L^\ast), where T(a)=T(a)T^\ast(a) = T(a^\ast)^\ast for any bounded operator TT on AA and aAa\in A.
  • For cAc\in A, we can define the double centralizer (Lc,Rc)(L_c,R_c), where LcL_c is left translation by cc and RcR_c right translation. This gives an isometric embedding AM(A)A\lra{} M(A). I leave it as an exercise to show that AA is essential in M(A)M(A).

    Suppose BB contains AA as an ideal. Then notice we have a map ψ:BM(A)\psi:B\lra{}M(A), by sending c(Lc,Rc)c\mapsto (L_c,R_c) again and using the fact that AA is an ideal. This extends the cannonical map AM(A)A\lra{} M(A). This is infact unique, for say ϕ:BM(A)\phi:B\lra{} M(A) also extended it. Then for bB,aAb\in B, a\in A, we would have ψ(b)(La,Ra)=ψ(ba)=(Lba,Rba)=ϕ(ba)=ϕ(b)(La,Ra)\psi(b) (L_a,R_a) = \psi(ba) = (L_{ba},R_{ba}) =\phi(ba) = \phi(b) (L_a,R_a) So that (ϕ(a)ψ(a))A=0 (\phi(a)-\psi(a)) A =0 in M(A)M(A), but AA is essential in M(A)M(A) so ϕ(a)=ψ(a)\phi(a)=\psi(a). Finally suppose AA is essential in BB, then if bker(ψ)b\in \ker(\psi) then Lb(A)=bA=0L_b(A) = bA=0, so that b=0b=0. Hence the map is injective.

    To summarize, whenever BB contains AA as an essential ideal, BB has a unique embedding into M(A)M(A). So M(A)M(A) is the biggest unitization as promised. We will get to the commutative case in the next section, but a good example for now is that if A=K(H)A=K(H), compact operators on some hilbert space, then M(A)=B(H)M(A) = B(H), bounded operators on that hilbert space.

    Stone-Čech CompactificationPermalink

    The Stone-Čech Compactification, βX\beta X,is the biggest compactification of a space, i.e if YY is a compactification than βX\beta X uniquely maps into YY making the obvious diagram commute. This is literally the dual condition of the multiplier algebra, so we just need to figure out what M(C0(X))M(C_0(X)) is. Luckily for us, the answer is easy: Cb(X)C_b(X), bounded continuous functions on XX with sup norm. One can easily check that C0(X)C_0(X) is essential in Cb(X)C_b(X), so that there is a injective map ψ:Cb(X)M(C0(X))\psi: C_b(X) \lra{} M(C_0(X)). All we need to do is show surjection, which is easily done but the proof I know is a bit annoying and uses some other tools (like approximate units). I will hence refer the user to Gerard Murphy's book on the topic, page 83.

    By the Gelfand Isomorphism, we know that Cb(X)C(βX)C_b(X)\cong C(\beta X) for some βX\beta X, and by the universal property discussed before, this βX\beta X is indeed the Stone-Čech compactification of XX. This proves that such a compactification exists in the first place but also gives a very nice explicit form for what the space should be.

    Updated:

    Comments