I ran into the following sanity check. Is the following statement true?

Every smooth fiber bundle (with compact fiber) over $S^2$ can be extended to a smooth fiber bundle over $\mathbb{C}P^\infty$ (or just $\mathbb{C}P^n$).

For the case of $\mathbb{C}P^2$, if the bundle is a principal bundle, then it can be extended, since $\pi_3(BG)=\pi_2(G)=0$. So I thought possible counterexamples might come from a fiber $F$, with $\pi_2(\text{Diff}(F))\ne 0$, for example $\text{Diff}(S^1\times S^2)\simeq O(2)\times O(3)\times \Omega SO(3)$, but seems like the fiber bundle can still be extended, since $S^3\rightarrow S^2\rightarrow B(\text{Diff}(S^1\times S^2))$ is still nullhomotopic.

Update: The counterexample is provided in Oscar's post, when the fiber is $(D^7,\partial D^7)$. Now my question is, if the fiber is a closed manifold, is the following statment true?

Every smooth fiber bundle(with closed fiber) over $S^2$ can be extended to a smooth fiber bundle over $\mathbb{C}P^\infty$ (or just $\mathbb{C}P^n$).

In fact, this should be my original question, but somehow I wrote "compact" but with "closed" in mind. I am really sorry about that.

  • $\begingroup$ If you distinguish between a bundle and a principal bundle, it's better to define a bundle first. $\endgroup$ – Alex Degtyarev Oct 21 '16 at 8:25

No it isn't, but I had to dig quite deep to get a counterexample. Let us look at smooth $(D^7, \partial D^7)$-bundles over $S^2$, i.e $D^7 \to E \overset{\pi} \to S^2$ with an identification $\partial E \cong S^2 \times \partial D^7$. These are classified by a map $$f : S^2 \to BDiff_\partial(D^7)$$ to the classifying space of the group of diffeomorphisms of $D^7$ relative to the boundary. I will show that there exists an $f$ which does not extend to $\mathbb{CP}^2$.

Smoothing theory gives an identification $$BDiff_\partial(D^7) \simeq \Omega^7\left(\frac{TOP(7)}{O(7)}\right)$$ and by page 246 of Kirby--Siebenmann the map $$\Omega^7\left(\frac{TOP(7)}{O(7)}\right) \to \Omega^7\left(\frac{TOP}{O}\right)$$ is 2-connected, so there is a surjection $$\pi_2(BDiff_\partial(D^7)) \to \pi_9(\tfrac{TOP}{O}) \quad \quad(*).$$ To find an example of such a bundle which does not extend over $\mathbb{CP}^2$, it is therefore enough to show that the map $$- \circ \eta : \pi_9(\tfrac{TOP}{O}) \to \pi_{10}(\tfrac{TOP}{O}) \quad \quad(**)$$ given by precomposition with the (suitably suspended) Hopf map is not trivial.

By surgery theory $\pi_n(\tfrac{TOP}{O}) \cong \Theta_n$, the group of homotopy $n$-spheres, and a homotopy $n$-sphere has a unique Spin structure, determining a $\pi_n(\tfrac{TOP}{O}) \cong \Theta_n \to \Omega^{Spin}_n$, which we can further compose with the Atiyah--Bott--Shapiro map $$\alpha : \Omega^{Spin}_n \to KO_n.$$ Applying this construction to both sides, the map $(**)$ above becomes the map $$- \circ \eta : KO_9 = \mathbb{Z}/2 \to KO_{10} = \mathbb{Z}/2,$$ which is an isomorphism. Finally, it is well-known that $$\pi_n(\tfrac{TOP}{O}) \cong \Theta_n \to \Omega^{Spin}_n \to KO_n$$ is surjective for $n \equiv 1,2 \mod 8$.

So choose a $f' \in \pi_9(\tfrac{TOP}{O}) \cong \Theta_9$ which is non-trivial under the Atiyah--Bott--Shapiro map, and use that $(*)$ is surjective to lift it to a $f \in \pi_2(BDiff_\partial(D^7))$. This has the desired property.


The following gambit can be used to get a closed fibre example. The product of the natural maps $SO(n+1) \to Diff^+(S^n)$ and $Diff_\partial(D^n) \to Diff^+(S^n)$ gives a map $$SO(n+1) \times Diff_\partial(D^n) \to Diff^+(S^n)$$ which is a weak homotopy equivalence (though not a homomorphism). Thus the map $$BDiff_\partial(D^7) \to BDiff^+(S^7)$$ is (split) injective on all homotopy groups, so the example above also gives an oriented $S^7$-bundle over $S^2$ which cannot extend to $\mathbb{CP}^2$.

  • 2
    $\begingroup$ Wow, thank you for the counterexample. Is it possible to find a counterexample with closed fiber? Actually that should be my original question, but somehow I wrote "compact", sorry about that. $\endgroup$ – ZZY Oct 21 '16 at 16:31

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.