Recall that complex topological $K$-theory is representable on reasonable spaces by the space $BU \times \mathbb{Z}$ (where $BU$ is a colimit of various infinite Grassmannians), and that the total Chern class provides a natural map $\mathrm{Vect}(B) \to H^*(B)$ for every such space $B$. By the multiplicativity property, this map factors through the K-group and leads to a natural transformation $K(B) \to H^*(B)$. $H^*(B)$ is also representable by a product of Eilenberg-MacLane spaces $K(\mathbb{Z}, n)$ over all $n$. There is thus a map, unique up to homotopy $$BU \times\mathbb{Z} \to \prod_n K(\mathbb{Z},n).$$

What is this map?

As Mariano observes, one can simply define the individual Chern classes on the $K$-group as well, albeit not immediately through the universal property, so the question reduces to the determination of the (homotopy class of the) map $BU \times \mathbb{Z} \to K(\mathbb{Z},n)$ induced by each Chern class.

  • $\begingroup$ It will be the product of all maps $BU\times\mathbb Z\to K(\mathbb Z,n)$ representing the individual Chern classes, so there is little gain in considering the whole product :) $\endgroup$ – Mariano Suárez-Álvarez Dec 17 '10 at 15:11
  • 2
    $\begingroup$ I am a little confused. Are you asking essentially for an element of $H^*(BU;\mathbb{Z})$? this we know. They are the universal chern classes right? I guess you could say that the map is the universal chern class, but I am not sure if this is what you are after. $\endgroup$ – Sean Tilson Dec 17 '10 at 15:47
  • 2
    $\begingroup$ Well, there are reasonably nice models for $BU$ and for $K(\mathbb Z,n)$, so one can wish for a nice description of the actual maps between those models. math.stanford.edu/~ralph/charact.ps seems to do precisely that in the algebraic case. $\endgroup$ – Mariano Suárez-Álvarez Dec 17 '10 at 16:43
  • 2
    $\begingroup$ @Akhil: I don't understand how Sean's answer answers the question! I though you were asking, in essence "Give me a model of $BU$, a model of $K(\mathbb Z, n)$, and a map between them which represents the $n$th Chern class. Sean's nice answer does not answer that. If you follow the link in my earlier comment, you'll see it done in the category of algebraic varieties: howpefully someomne can give us the details in the case of spaces, which should be considerably easier. $\endgroup$ – Mariano Suárez-Álvarez Dec 17 '10 at 23:35
  • 1
    $\begingroup$ @Mariano: Dear Mariano, Thanks for your comment and link, which directly address my initial question (in a positive direction, which is much better than the other direction!). $\endgroup$ – Matt E Dec 18 '10 at 5:25

One way, my favorite, of defining the chern classes goes as follows: first compute $H^*(BU; \mathbb{Z})=\mathbb{Z}[c_1, c_2, ...]$ by using your favorite method. I know two methods, one is using some cellular description of the grassmanians (see Milnor and Stasheff), the other is to first compute the cohomology of the lie groups $U(n)$ and then use a path-loop space fibration and the Serre SS to get at the cohomology of $H^*(BU(n);\mathbb{Z})$ (see Homology and Euler characteristics of the classical Lie groups). Next you notice that $BU$ and $BU(n)$ have the same cell structure through a range and so you essentially take the limit (colimit, and there are some subtleties here, the place I recall seeing these is in Jacob Lurie's survey on elliptic cohomology, at the beginning).

Next suppose you have a complex vector bundle $\xi : E \to B$ then it is classified by a map $f: B \to BU(n)$ where n is the dimension of $\xi$. You can get a map on all of $BU$ by just adding on trivial bundles to get a map out of $BU(k)$ for all $k$ larger than $n$. This gives a map out of BU (although technically you don't really need this, $BU(n)$ works fine for defining the chern classes of an $n$ dimensional complex vector bundle). Now define $c_n(\xi):=f^*(c_n)$.

From this perspective we started with the universal case, so maybe it is a bit of a cheat. For me this even clarifies how I should think about characteristic classes in general. Suppose you want to look at $G$-bundles and see what you can tell about them from $E$-theory ($E$ some ring spectrum and $G$ some compact lie group or whatever you need for $BG$ to be nice, I am not sure if there are other restrictions for this to work). Now compute $E^*BG$ and use the fact that $G$-bundles over $X$ are classified by homotopy classes $X \to BG$. You should check out some of the threads on MO about characteristic classes, I think I can learn something from each of Rezk's answers.

please let me know if I can make some of the above clearer or if there are any mistakes.

  • $\begingroup$ Thanks! (And extra characters.) $\endgroup$ – Akhil Mathew Dec 17 '10 at 17:28
  • $\begingroup$ I agree that this is definitely the best way of constructing Chern classes, it's so much nicer than talking about curvature forms and all that. One time we spent a whole lecture in my symplectic geometry class trying to show why $c_1(L)\in H^2(X;\mathbb{Z})$ classified line bundles! $\endgroup$ – Aaron Mazel-Gee Dec 17 '10 at 22:57
  • 2
    $\begingroup$ @Aaron: Dear Aaron, I just want to remark that there are lots of reasons for considering the definition of Chern classes via curvature (or via obstruction theory, or via $K$-theory, or via lots of other points of view). So while I agree that the picture in terms of cohomology of the classifying space is very beautiful and satisfying, it's good to be open-minded about other points of view too! $\endgroup$ – Matt E Dec 18 '10 at 5:27
  • $\begingroup$ @Aaron: Milnor-Stasheff construct the Stiefel-Whitney classes via Steenrod squaring the fundamental class of the bundle and applying the Thom isomorphism. I guess that doesn't work for Chern classes anymore since they're over $\mathbb{Z}$, but is there another way via different cohomology operations? $\endgroup$ – Akhil Mathew Dec 18 '10 at 5:51
  • $\begingroup$ M&S use the Gysin Sequence (a consequence of the Thom isomorphism) and the euler class to construct the chern classes inductively. $\endgroup$ – Sean Tilson Dec 18 '10 at 18:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.