TITLE: Central extensions of loop groups
QUESTION [8 upvotes]: Let $LG=\operatorname{Maps}(S^1,G)$ be the loop group of a compact Lie group $G$.  I should add some adjectives to $G$, but for sake of simplicity let's just take $G=SU(2)$.
There is a central extension
$$1\to S^1\to\widetilde{LG}\to LG\to 1$$
(these are classified by "level" in $H^3(G)$, but we may as well restrict attention to the "universal" such extension corresponding to a generator of this group).  The constructions of this central extension that I have found so far (e.g. the one in Pressley--Segal) all go via first defining a closed $2$-form on $LG$, arguing it defines a unique $S^1$-bundle, and then putting a group structure on this bundle.

Is there a more intrinsic definition of $\widetilde{LG}$?

In other words, given a loop $\gamma:S^1\to G$, I would like to have an intrinsically defined principal $S^1$ homogeneous space (or, equivalently, a $1$-dimensional complex vector space).
For example, here is an answer "up to homotopy".  Since $\pi_1(G)=\pi_2(G)=0$ and $\pi_3(G)=\mathbb Z$, given any loop $\gamma:S^1\to G$, the space of extensions $\bar\gamma:D^2\to G$ (i.e. $\bar\gamma|_{\partial D^2=S^1}=\gamma$) is homotopy equivalent to $\Omega^2G$ which is connected with fundamental group $\mathbb Z$.  If we take the $1$-truncation of this space (add cells to kill all higher homotopy groups), we get $S^1$ (up to homotopy).
This gives an "intrinsically defined" space homotopy equivalent to $S^1$ defined in terms of a given loop $\gamma:S^1\to G$ (although it's not very explicit, and has questionable meaning/use).  What about an honest one-dimensional complex vector space? (with a natural meaning).  Even better, can we define intrinsically a holomorphic line bundle over the complexified loop group $LG_{\mathbb C}$?

REPLY [6 votes]: Fix a cocycle $\omega\in C^3(G, \mathbb R)$ such that $\omega(H_3(G, \mathbb Z)) = \mathbb Z$.  (For $G = SU(2)$, we can take $\omega$ to be the standard volume form.)  Fix a loop $L$ in $G$, and let $\partial^{-1}(L) \subset C_2(G, \mathbb Z)$ denote the set of 2-chains whose boundary is $L$.  For $x,y \in \partial^{-1}(L)$, define an equivalence relation $x \sim y$ if $\omega(\partial^{-1}(x - y)) \in \mathbb Z$.  Then $\partial^{-1}(L)/\sim$ is an $S^1$ torsor.
The above paragraph defines an $S^1$-bundle over $LG$, but it doesn't tell you how to multiply two elements of this bundle.  For this, we use the fact that $\pi_2(G)$ is trivial and redefine $\partial^{-1}(L)$ to be maps of $D^2$ into $G$ which restrict to $L$ on the boundary.  We can multiply the maps $D^2 \to G$ pointwise.  This group structure will extend to the quotient if we choose $\omega$ to be invariant under left and right multiplication by elements of $G$.  (I have not thought about this last claim as carefully as I should, so be skeptical here.)
Assuming I did not make a stupid error in the previous paragraph (time constraints!), we can summarize as follows: Elements of $\widetilde{LG}$ are represented by maps to $D^2$ to $G$, and two such maps are considered equivalent if the $\omega$-volume they cobound is an integer.