6:53 AM
@Craig: a heuristic model for a cohomology theory associated to a formal group law of height n is that it should look a bit like bundles of n-vector spaces (e.g. for n = 2 this is baas-dundes-rognes and for general n this is redshift). here is in turn a heuristic definition of n-vector space: a 1-vector space is a vector space. this defines a symmetric monoidal (topological) category Vect. a 2-vector space is a Vect-module category.
inductively, if you know what n-Vect is as a symmetric monoidal (infinity) category, then an (n+1)-vector space is an (n-Vect)-module (n-category)
(although really I just want dualizable objects all around)
whatever that inductive definition means, it should in particular imply that endomorphisms of the "free module of rank 1" over n-Vect is n-Vect itself; in other words, (n-Vect)-Mod is a nonconnective delooping of n-Vect
now, the invertible objects in Vect itself can naturally be identified with BC* = B^2 Z (here I mean complex vector spaces). by iterating the above construction, there's a natural map from B^{n+1} Z to the invertible objects in n-Vect; this is the space that classifies "n-line bundles"
for n = 2 there are a couple of concrete models you can take for what a "2-line bundle" should be. there is in particular a natural map from bundles of clifford algebras to 2-line bundles given by taking some fiberwise version of the category of modules over a clifford algebra (these are, I believe, precisely the invertible Vect-modules over R or C).
in particular, start with any real vector bundle V, give it a riemannian metric, take fiberwise clifford algebras, and then complexify. the resulting bundle of complex clifford algebras has a characteristic class in H^3(-, Z) which turns out to be precisely W_3(V) = \beta w_2(V), where \beta is a bockstein
unfortunately, bundles like these only give torsion elements of H^3(-, Z), and to get the rest of them one has to do various other things, e.g. take bundles of copies of the C*-algebra K of compact operators on an infinite-dimensional hilbert space
in this case the corresponding class in H^3(-, Z) is called the dixmier-douady class, and taking global sections of the corresponding bundle of Ks is a standardish construction in C*-algebras; it classifies a particular kind of them called the continuous trace C*-algebras