Let X be a locally compact noncompact Hausdorff space and consider the C$^*$-Algebra $C_0(X)$ of continuous functions vanishing at infinity. I want to construct an injective *-represenatation of $C_0(X)$ to give an example of Gelfand-Naimark. I want to construct the representation with the GNS-construction. My beginning:
First of all i need the states of $C_0(X)$ and if i consider http://en.wikipedia.org/wiki/Riesz%E2%80%93Markov%E2%80%93Kakutani_representation_theorem i would say that the positive linear functionals are $\phi(\mu)(\cdot)=\int_X \cdot d\mu$ with $\mu:X\to\mathbb{C}$ is a positive regular borel-measure (therefore $\mu$ has only real values) and its total variation has to be 1 because it has to be state. The next step of the GNS-construction is to adjoin a unit to $C_0(X)$ ( i write $C_0(X)^1$ for the unitalisized C$^*$-algebra) and extend $\phi(\mu)$ to $C_0(X)^1$. I obtain $\tilde{\phi(\mu)}((f,\lambda))=\int_X f d\mu +\lambda\|\phi(\mu)\|$, $\tilde{\phi(\mu)}:C_0(X)^1\to \mathbb{C}$. Because $\phi(\mu)$ is a state, i have $\tilde{\phi(\mu)}((f,\lambda))=\int_X f d\mu +\lambda 1$.
Now, i have lots of problems to continue the constraction:
Every state give me a scalar product $\tilde{\phi(\mu)}((g,\eta)^*(f,\lambda)):=<(f,\lambda),(g,\eta)>_{\tilde{\phi(\mu)}}$ (im not sure if it is correct and how to write out $<(f,\lambda),(g,\eta)>=...$.
The next step is to define $N_{\tilde{\phi(\mu)}}:=\{(f,\lambda) \in C_0(X)^1: <(f,\lambda),(f,\lambda)>=0\}$. The $<\cdot,\cdot>_{\tilde{\phi(\mu)}}$ has to be a scalar product on the quotient $C_0(X)/N_{\tilde{\phi(\mu)}}$ and i have to take the Completion of $C_0(X)/N_{\tilde{\phi(\mu)}}$. The GNS-construction is not finished with this step. But the things will be more complicated in every step. For example, which space is $C_0(X)/N_{\tilde{\phi(\mu)}}$ ? Could you help me to construct an injective *-representation of $C_0(X)$ ant to coninue here maybe? I'm not sure if i do this in a correct way. Regards
Edit: If you need some definitions, i will add them.
I don't think you can get too far with your approach, because you want to deal with the set of all measures on $X$, and there is nothing explicit about it.
So you want to use fewer states.
1) Following on what Phoenix87 said, here is an example of a faithful representation (denomination way more common in the literature than "injective"). It is based on using the pure states, which are nothing but the point evaluations.
Let $H$ be a Hilbert space with basis indexed by $X$, i.e. $\{e_x\}_{x\in X}$. Then you map $f$ to the diagonal operator $\{f(x)\}_{x\in X}$. Namely, $$ \pi(f)e_x=f(x)e_x. $$
2) Another way it to use a single state, but such that it is faithul. You can get such a state if you can construct a probability measure $\mu$ on $X$ such that $\int_Xf\,d\mu=0$ implies $f=0$ for any $f\in C_0(X)$. This is for example the case with $\mathbb R$ and the Lebesgue measure, but I don't think you can have something like this for arbitrary $X$.
If you do have such $\mu$, then you can do GNS for the state $f\longmapsto\int_Xf\,d\mu$. If you push this through, you'll find that your representation is the map $f\longmapsto M_f$, where $M_f$ is the multiplication operator by $f$ on $L^2(X,\mu)$.