I know that the tangent bundle $TS^2$ is stably trivial: if $\nu$ denotes the normal bundle of the embedding of $S^2$ in $\mathbb{R}^3$ as the unit sphere, then $\nu$ is a trivial line bundle. But then the sum of these is trivial, since it just gives the restriction of the tangent bundle on $\mathbb{R}^3$. If I wanted, I could just tack on an extra trivial line bundle to have $TS^2 \oplus \varepsilon^2 \cong \varepsilon^4$.

Now $S^2 \cong \mathbb{CP}^1$ is also a complex manifold, so $TS^2$ is a complex line bundle. If this complex bundle were stably trivial, then the product theorem for Chern classes would yield that the total Chern class of $TS^2$ is trivial. However, $$c(TS^2) = 1 + c_1(TS^2) = 1+e(TS^2)$$ where the Euler class $e(TS^2)$ is twice a generator of $H^2(S^2) \cong \mathbb{Z}$, a contradiction. Doesn't this now contradict the last line of the previous paragraph, since rank-$2k$ real trivial bundles are the same as rank $k$-complex trivial bundles?


1 Answer 1


The difference between a rank $2k$ real bundle and a rank $k$ complex bundle is that in a complex bundle one is using $\mathbb C$ as the scalar field, whereas in a real bundle one is only using $\mathbb R$ as the scalar field.

Now let's suppose we are given a complex $n$-dimensonal bundle $E \mapsto X$.

It is true that if this bundle is trivial over $\mathbb C$, i.e. if there is a $\mathbb C$-linear isomorphism of bundles $E \mapsto X \times \mathbb C^n$ over $X$, then one immediately obtains an $\mathbb R$ linear isomorphism of bundles $E \mapsto X \times \mathbb R^{2n}$ over $X$: just restrict the scalar field from $\mathbb C$ to $\mathbb R$.

But the converse does not hold: if there is an $\mathbb R$ linear isomorphism of bundles $E \mapsto X \times \mathbb R^{2n}$ over $X$, you don't know how to extend this to a $\mathbb C$-linear isomorphism; roughly speaking, you don't know how to define multiplication by $i$ in $X \times \mathbb R^{2n}$ in such a way that makes the given map of bundles into a $\mathbb C$-linear isomorphism.

  • $\begingroup$ Ah I see. So even though on the top $TS^2$, $\varepsilon^2$, and $\varepsilon^4$ all have some pretty obvious complex structures, the identification $TS^2 \oplus \varepsilon^2 \cong \varepsilon^4$ doesn't respect those complex structures. I suppose the (real) isomorphism really does have to be mixing up $TS^2$ and the first factor of $\varepsilon^2$, while leaving the second factor of $\varepsilon^2$ totally untouched. $\endgroup$ Apr 9, 2020 at 21:13
  • 1
    $\begingroup$ In general if you try to use the Hermitian metric to split up a vector bundle into subbundles (which is what you are trying to do with your exact sequence) you are going to have a bad time if you expect it to give a holomorphic decomposition. $\endgroup$
    – Sempliner
    Apr 9, 2020 at 21:19

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.