Recall that give a map $f : (X,\mathcal{O}_X) \to (Y,\mathcal{O}_Y)$ of ringed spaces and a sheaf $\mathcal{F}$ on $Y$ we can form the pullback $f^\ast \mathcal{F} := f^{-1}\mathcal{F}\otimes_{f^{-1}\mathcal{O}_Y} \mathcal{O}_X$.

Now the pullback of sheaves appears in both chapters 2.5 and 2.7 of Hartshorne. In particular, it is used to define a correspodence between line bundles $\mathcal{L}$ on a scheme $(X \to \operatorname{Spec} A)$ and maps to projective space $\Bbb{P}^n_A$. More explicitly, for any morphism $\varphi : X\to \Bbb{P}^n_A$ we have $\varphi^\ast \mathcal{O}(1)$ being a line bundle generated by the linear polynomials $x_0,\ldots,x_n$. Conversely given $n+1$ global sections $s_0,\ldots,s_n$ of a line bundle $\mathcal{L}$ on $X$, we have a unique morphism $\varphi : X \to \Bbb{P}^n_A$ such that $\varphi^\ast(x_i) = s_i$.

Problem 1: My first problem I have with $\varphi^\ast$ is say I have the $x_i$'s as above. What are the "canonical induced sections" $s_i \in \varphi^\ast\mathcal{O}(1)$? All I can say now is that each $x_i$ gives a map from $\mathcal{O}_{\Bbb{P}^n_A} \to \mathcal{O}(1)$ but how does this give us our $s_i$? Is there a more concrete description of what these actually are?

Problem 3: Can we see explicitly "in coordinates" what the map $\varphi : X\to \Bbb{P}^n_A$ determined by these sections is?

Problem 2: My second problem with the definition of this $\varphi^\ast$ is getting hold of its global sections. From the definition I gave in the beginning of the pullback, it seems that definition is not much help in practical situations! Say I want to compute global sections of $\varphi^\ast \mathcal{O}(1)$ where $\varphi$ is the closed immersion of the conic $ V(xy - z^2)$ in $\Bbb{P}^2$. How can I do this? What about computing global sections of the bullback bundle by say $\varphi: \Bbb{P}^n\setminus V(x_0,x_1,x_2) \to \Bbb{P}^2$?

Last and finally, is there a way to think of this pullback operation that can significantly enhance my feel for it? Let me give an example of what I mean by this. When I first learned about the tensor product of sheaves I was very fearful because I didn't know how the sheaf behaved (for example what on earth are the global sections?). However if we tensor two quasi-coherent sheaves, then on an affine I know exactly what the tensor product is: it behaves just like a module!

  • $\begingroup$ The details of this bijection are given very explicitly in (EGA, II, 4.2.3). $\endgroup$ – user314 Oct 25 '13 at 12:51
  • 2
    $\begingroup$ For Problem 1, one concrete way to understand the pullback might be via rational functions. There is a non-canonical way to identify sections of a bundle with rational functions having certain prescribed poles (explained e.g. in Volume 2 of Shafarevich's book). Given this, you can pull back sections of a bundle just by the usual pullback of rational functions. Of course this has certain drawbacks --- it's not canonical, and when $\varphi$ isn't dominant you need to worry about indeterminacy. Still, I think it helps to give some insight. $\endgroup$ – user64687 Oct 25 '13 at 14:18
  • 7
    $\begingroup$ Forget about the explicit construction of the pullback, it isn't useful at all. Pullback = left adjoint to direct image. Locally the pullback just tensors with the corresponding ring extension. This is how one works with pullbacks. P1: What do you mean by concrete? In what terms do you want to describe them? We have sections $x_i : \mathcal{O} \to \mathcal{O}(1)$, so we can pull them back to $\phi^* x_i : \mathcal{O} \to \phi^* \mathcal{O}(1)$. So this is just functoriality. $\endgroup$ – Martin Brandenburg Oct 26 '13 at 8:57
  • 3
    $\begingroup$ P3: What do you mean by coordinates? Perhaps $k$-valued points? Then the map is given by $x \mapsto [s_0(x):\dotsc:s_n(x)]$. By this one actually means $[s_0(x)/s_i(x) : \dotsc : s_n(x)/s_i(x)]$ when $s_i(x)$ generates $\mathcal{L}$, and $s_j(x)/s_i(x)$ is the unique element of $k$ with $s_j(x)/s_i(x) \cdot s_i(x) = s_j(x)$ in $\mathcal{L}$. $\endgroup$ – Martin Brandenburg Oct 26 '13 at 9:00
  • 3
    $\begingroup$ An off topic note: In my opinion, it isn't appropriate to use as a pseudonym the name and image of an actual living mathematician. $\endgroup$ – David E Speyer Oct 30 '13 at 21:02

Pullback of invertible sheaves corresponds to pull-back of line bundles in the geometric sense.

In usual topology, a (complex) line bundle is a family of one-dimensional $\mathbb C$-vector spaces varying over a topological space $X$. More precisely, it is a topological space $V$, equipped with a continuous map $V \to X$, a zero section $X \to V$, a scalar multiplication action $\mathbb C \times V \to V$ compatible with the projections to $X$, an addition map $V \times_X V \to V$, again compatible with the projections to $X$, all satisfying some evident axioms, which more or less amount to requiring that these structure endow each fibre $V_x$ of $V$ over a point $x$ of $X$ with the structure of a one-dimensional $\mathbb C$-vector space, in such a way that $V$ is locally trivial.

Now if $\varphi: Y \to X$ is a continuous map, we may form the fibre product $\varphi^*V := V\times_X Y,$ and this is a line bundle over $Y$.

If you just think about how fibre product is defined, you see that the fibre of $\varphi^*V$ over a point $y \in Y$ is canonically identified with the fibre of $V$ over $\varphi(y)$. So if we think of $V$ as being the family $V_x$ of lines parameterized by the points $x \in X$, then $\varphi^*V$ is the family of lines $V_{\phi(y)}$ parameterized by $y \in Y$.

We may form the sheaf of sections $\mathcal L$ of $V$; this is a locally free sheaf of rank one over the sheaf of continuous $\mathbb C$-valued functions on $X$. The sheaf of sections of $\varphi^*V$ is then $\varphi^*\mathcal L$.

There is an exercise somewhere in Hartshorne which describes the analogue of the above theory of vector bundles in the algebro-geometric setting; everything goes over more-or-less the same way. I am now going to work in the algebro-geometric setting.

It is not so easy to relate the sections of $V$ to those of $\varphi^*V$; it depends very much on the space $Y$ and the morphism $\varphi$. For example, if $Y$ is just a point $x \in X$, then $\varphi^*V$ is just the line $V_x$, and its space of sections is one-dimensional (indepdently of what $V$ is). But it might be that $V$ has no sections that are non-zero at $x$.

More generally, a non-trivial bundle on $X$ might pull-back to something trivial on $Y$.

So you have to analyze this question on a case-by-case basis. Cohomology is one of the basic tools available here, but I am guessing that you're not at the point of using cohomology yet.

E.g. if $Y = \mathbb P^1$ and $\varphi: Y \to \mathbb P^2$ embeds $Y$ as a plane conic, then this image is a degree two curve, so $\mathcal O(1)$ pulls back to a degree two sheaf on $Y$, i.e. to $\mathcal O(2)$. Thus its space of global sections is three dimensional. So in this case $\varphi^*$ induces an isomorphism on spaces of global sections.

E.g. If $Y = \mathbb P^1$ but $\varphi: Y \to \mathbb P^2$ embeds it as a line, then $\varphi^*\mathcal O(1)$ is just $\mathcal O(1)$ on $\mathbb P^1$ (a line is a degree one curve), and so the map on sections is surjective, with a one-dimensional kernel (which is precisely the linear form cutting out the image of $\varphi$).

As Martin Brandenburg indicates in an answer, the sections $s_i$ give a projective embedding in the following concrete way: locally we may trivialize $V$, and so regard the $s_i$ as simply functions; we thus obtain a morphism $x \mapsto [s_0(x): \cdots : s_n(x)].$ Note that if we change trivialization, then the interpretation of the $s_i$ as functions changes by a nowhere zero function (the same function for all the $s_i$), which means that the point in $\mathbb P^n$ doesn't change. Thus the map is well-defined independent of the trivialization.

Finally, it might be helpful to know that in topology, the complex line bundles are classified (up to isomorphism) by homotopy classes of maps to $\mathbb C P^{\infty}$. Again, the map is given by pulling back $\mathcal O(1)$.

The situation in algebraic geometry is analogous to this, but more rigid: topological isomorphism is much less rigid than algebraic isomorphism (hence only the homotopy class of $\varphi$ matters), and all line bundles have lots of sections. In algebraic geometry, only sufficiently positive (i.e. sufficiently ample) bundles arise from maps to projective space.


Your Answer

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