Let $X$ be a manifold. We say $\pi: Y \longrightarrow X$ is an rank $n$ affine bundle if there is an open cover $\{ U_\alpha \}$ of $X$ such that $ Y \big|_{U_\alpha} \cong U_\alpha \times \mathbb{R}^n $ and the transition function from $U_\alpha$ to $U_\beta$ is given by $$ (x,v) \mapsto (x, \rho_{\beta \alpha }(x) v + u_{ \beta \alpha} (x)) $$ satisfying the cocycle condition $ \rho_{\gamma \alpha} (x) = \rho_{\gamma \beta} (x) \rho_{\beta \alpha } (x) $ and $u_{\gamma \alpha}(x) = \rho_{\gamma \beta} (x) u_{\beta \alpha} (x) + u_{\gamma \beta}(x)$.

Wikipedia claims that an affine bundle has a global section so it can be identified with the vector bundle glued by the cocycles $\{ \rho_{\gamma \alpha} \}$ in a non-canonical way. How can we construct one exactly? Someone claimed that local sections exist so one can glue them to a global one by standard partition of unity argument. Since multiplying by constant doesn't make sense for affine bundle, I cannot see why this is obvious.


Multiplication by functions doesn’t in general make sense, but affine combinations of sections do.

Specifically, if $\sigma,\tau\in\Gamma(\pi)$ are sections and $f, g\in\mathrm C^\infty(X)$ are smooth functions on the base summing to unity, then the obvious definition of $f\sigma + g\tau$ is independent of trivialization if (!) $f(x)+g(x) = 1$. This defines affine combinations only for a finite number of terms, but you can of course also do that for an arbitrary one, as long as only a finite number of coefficients are non-zero at any given point.

Now choose a (locally finite) trivializing open cover of the bundle, a partition of unity corresponding to that cover and an arbitrary local section over every neighbourhood. Then the affine combination of these sections with coefficients given by this partition is a well-defined global section.


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.