The splitting principle is as follows.

Given a vector bundle $E \to X$ with $X$ compact Hausdorff, there is a compact Hausdorff space $F(E)$ and a map $p: F(E) \to X$ such that the induced map $p^*: K^*(X) \to K^*(F(E))$ is injective and $p^*(E)$ splits as the sum of line bundles.

My question is, what is the idea/intuition behind the proof of the splitting principle?

  • 8
    $\begingroup$ Grassmanians map to products of projective spaces, where the universal bundles split. $\endgroup$ – Alex Degtyarev Feb 4 '16 at 21:38

In the topological language you are using, $F(E)$ is the space of "orthogonal splittings". That is to say, $p^{-1}(x)$ is the space of all ways to write the fiber $E_x$ as an orthogonal sum of one dimensional spaces. Since it is the "space of splittings", there is a tautological splitting over it. "$\square$"

Remark on alternative versions you may have seen: It is more common to describe $F(E)$ as the space of flags. A (complete) flag $F_{\bullet}$ in a vector space $V$ is a chain of subspaces $F_1 \subset F_2 \subset \cdots \subset F_d = E$ where $\dim F_k = k$. When $E$ is equipped with a positive definite symmetric or Hermitian form, this is the same as a splitting; the summands of the splitting are $F_k \cap F_{k-1}^{\perp}$.

The flag formulation works better when working with holomorphic vector bundles, in which case the statement is that the vector bundle has a filtration with one dimensional filtered pieces, not necessarily a splitting.

| cite | improve this answer | |
  • 2
    $\begingroup$ I guess that the miracle here is less that there is an universal splitting (as you say it is quite trivial) and more that the map in K-theory (and integral cohomology) is injective. $\endgroup$ – Denis Nardin Feb 4 '16 at 22:15
  • $\begingroup$ Good point. I don't have an immediate intuition for that but I'll think about it $\endgroup$ – David E Speyer Feb 4 '16 at 22:17
  • $\begingroup$ Well, in the flag formulation, $F(E)$ is an iterated projective bundle, which explains the injectivity on cohomology... $\endgroup$ – Daniel Litt Feb 4 '16 at 22:31
  • 5
    $\begingroup$ @DanielLitt I don't think it explains it. I know the proof but in the end it boils down to a special property of the cohomology of projective space (that it is a power series ring in one variable) plus the Leray-Hirsch theorem. I find this quite mysterious in fact but I'd welcome a good intuition $\endgroup$ – Denis Nardin Feb 5 '16 at 0:02

Perhaps my very short (4 pages plus bibliography) paper ``A note on the splitting principle'' http://www.math.uchicago.edu/~may/PAPERS/Split.pdf may be illuminating. It shows that the splitting principle can be viewed as a statement about the reduction of the structural group of a $G$-bundle $\xi$ from $G$ to a maximal torus $T$, where $G$ is a compact Lie group. It applies more generally than in just the usual examples. One starts with the bundle $BT\to BG$ with fiber $G/T$. For a $G$-bundle over $X$ classified by $f\colon X\to BG$, one has a pullback bundle $q\colon Y\to X$ with fiber $G/T$ together with a reduction of the structure group of $q^*\xi$ to $T$. When $H^*(BG;R)$ is concentrated in even degrees, $q^*\colon H^*(X;R)\to H^*(Y;R)$ is a monomorphism. That is easily seen to imply the splitting theorem as usually stated, and many variants thereof. As stated and explained briefly in the paper, the argument adapts to $K$-theory.

| cite | improve this answer | |

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.