Let $ S = \operatorname{Spec}A $ be an affine scheme, $ f : E \to S $ an elliptic curve and $\mathscr{I}$ the ideal sheaf of the $0$-section. (This is invertible since the section defines the effective relative Cartier divisor.)
Assume that $f_* \Omega_{E/S}, f_*\mathscr{I}^n$ are free over $\mathscr{O}_S$. ($n = 1, \cdots , 6$)
I want to show $\hat{E} \cong \operatorname{Spf} A[[T]]$. And I don't understand the formal expansion, of a basis $\omega$ of $f_* \Omega_{E/S}$ and a basis of $f_*\mathscr{I}^n$, along the $0$-section.

Here is what I have tried: Since the $0$-section is a regular immersion, for any $x \in S$, there exists affine opens $ x \in V \subseteq S$, $ 0(x) \in U \subseteq E$ s.t. $0(V) \subseteq U$ and the diagram

$$\require{AMScd} \begin{CD} S @>{0}>> E \\ @VV{1}V @VV{f}V \\ S @>{1}>> S \end{CD}$$

corresponds to

$$\require{AMScd} \begin{CD} C @<{0}<< B \\ @A{\text{localization by one element}}AA @AAA \\ A @<{1}<< A, \end{CD}$$

where the kernel $I$ of $B \to C$ is generated by $t \in B$, a regular element.

I showed that $\hat{B}$ (the completion of $B$ along the kernel $I$) $\cong C[[t]]$. And $\Omega_{B/A} \otimes_B \hat{B} = dt \hat{B} = dt C[[t]].$
That is, I can show $ \hat{E} \cong \operatorname{Spf}A[[t]]$ locally, and I can expand $\omega$ locally.

How can I extend these operation globally?
I also showed that the isomorphism $\hat{B} \cong C[[t]]$ is compatible with localization. So I think intuitively that these isomorphisms (at any points) are glued together, and we have $ \hat{E} \cong \operatorname{Spf}A[[t]]$.

Any help will be much appreciated!


Yes, what you are saying is true, at least over an algebraically closed base $k$ of characteristic $0$. In fact, all you need is that $S$ is affine and that the normal bundle $I/I^2$ is trivial (as a bundle over $S$), which in your case is equivalent to $f_*\Omega(E/S)$ being free. The rest follows essentially from deformation theory.

There is a standard way to reduce questions about formal thickenings to questions in deformation theory, and this is called deformation to the normal cone. Here's the idea. Let $R = O(\widehat{E})$ be the algebra of function on the formal thickening, viewed as a formal ring over $A$. Then this algebra is filtered by $R_n : = I^n,$ with associated graded ring $$R_{gr}:=\bigoplus_n \frac{I^n}{I^{n-1}}\cong A[I/I^2].$$ Since $I/I^2$ is trivial and one-dimensional over $A$, we have $R_{gr}\cong A[[t]].$ Now the deformation to the normal cone is an algebra over $\text{Spf} (k[[s]])$ interpolating between $R$ and $R_{gr}$, given by the Rees construction $$R_{\text{rees}} : = R[s^{-1}\cdot I]^{\wedge} = \bigoplus_n^\wedge s^{-n}I^n,$$ where the completion is taken along (positive) powers of $s$. The key thing to observe here is that the fiber over $s=0$ of $R_{\text{rees}}$ is $\bigoplus s^{-n}(I^n/I^{n+1})$ (not completed), which is $R_{gr}$ and the generic fiber over $k((s))$ is isomorphic to $R\otimes k((s))$.

Now a standard result in the deformation theory of schemes tells us that flat deformations of a smooth scheme $X$ up to isomorphism (as well as of a smooth formal scheme) are classified by (what is noncanonically isomorphic to) a subset of $H^1(X,T)^\infty,$ i.e. infinite sequences of vectors in the first homology of the tangent bundle (deformations of order $n$ would be certain sequences of size $n$). In particular, if $X$ is affine (in your case — $X = \text{Spf}A[[t]],$ the special fiber of the Rees construction), the relevant $H^1$ is trivial, so there is only one deformation — the trivial one. Equivalently, $R_{\text{rees}}\cong R_{gr}[[s]]$.

Now you're done! Basechanging to the generic point gives $R\otimes k((s)) \cong A[[t]]\otimes k((s)).$ But since you are over an algebraically closed base of characteristic zero, such an isomorphism implies an isomorphism $R\cong A[[t]]$ over $k$.

  • $\begingroup$ Thank you for your comment! I have a question: Why is the freeness of $\mathscr{I/I^2}$ equivalent to one of $f_* \Omega$? My text (Katz-Mazur) says that the freeness of $f_* \Omega$ is also equivalent to one of $f_* \mathscr{I}^{-n}$. I think these two statements follow from the same argument. $\endgroup$ – k.j. Jul 23 at 9:58
  • 1
    $\begingroup$ Since $E/S$ is an elliptic curve, its (co)tangent bundle is constant on each fiber. So its pushforward is the same as the (co)tangent at the zero section. $\endgroup$ – Dmitry Vaintrob Jul 23 at 17:54

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.