2
$\begingroup$

Suppose $R$ is a Noetherian ring and $I$ a nontrivial ideal of $R$, and $A_0\to B_0$ a finite faithfully flat lci map of smooth $R_0 := R/I$-algebras.

We fix a smooth $R$-algebra $A$ lifting $A_0$ and assume $A$ and $A_0$ integral.

Assume $R$ is $I$-adically complete.

Does there exist a smooth $R$-algebra $B$ with a finite $R$-map $A\to B$ lifting $A_0\to B_0$?

If a finite $R$-map $A\to B$ exists, then it is finite lci and faithfully flat, because it is an integral extension and so it is a finite surjective lci on spectra.

I'm trying to use this and variations on the theme. I've been able to show that there is a quasi-finite syntomic map $A\to B$ with $B$ not necessarily smooth.

$\endgroup$
2
  • $\begingroup$ Sorry when you say $A$ and $A_0$ are integral do you mean they are integral domains? $\endgroup$ Commented Aug 10, 2021 at 5:46
  • $\begingroup$ Yes, I do. They’re domains $\endgroup$
    – user315884
    Commented Aug 10, 2021 at 6:08

1 Answer 1

1
$\begingroup$

No, I don't think so.

Lemma: Given $R$, $I$, $R_0 = R/I$, $A$, $A_0 = A/I$. Assume $2$ is invertible in $R_0$. If the answer to the question is "yes" then the image of the map $Pic(A) \to Pic(A_0)$ contains all $2$-torsion of $Pic(A_0)$.

Proof: Namely, suppose $L_0$ is an invertible $A_0$-module of order $2$ in the Picard group of $A_0$. Choosing an isomorphism $\psi : L_0 \otimes_{A_0} L_0 \to A_0$ we can set $B_0 = A_0 \oplus L_0$ with multiplication defined by $\psi$. Then $A_0 \to B_0$ is finite etale because $2$ is invertible in $A_0$ by the branched covering trick, whence $B_0$ is smooth. If we can find $B$ as desired, then we see that $\det_A(B) \in Pic(A)$ lifts $L_0$. QED

So now we just have to find $R$ and $A$ where the $2$-torsion in $Pic(A_0)$ does not lift. Read on if you want to see what I came up with. To understand it, you might have to recall some facts about the deformation to the normal cone. See example at the end for an explicit case.

Say $R$ is a complete discrete valuation ring with uniformizer $t$ and residue field $k$. Let $n \geq 2$ and let $C \subset \mathbf{A}^n_k$ be an irreducible closed smooth subscheme such that $Pic(C)$ has lots of $2$-torsion. Denote $$ b : T \longrightarrow \mathbf{A}^n_R $$ the blowing up of $C$ viewed as a closed subscheme by identifying $\mathbf{A}^n_k$ with a closed subscheme of $\mathbf{A}^n_R$. Then we have $$ V(t) = E + X $$ where $E$ is the exceptional fibre of $b$ and $X$ is the strict transform of $\mathbf{A}^n_k$, i.e., $X$ is the blowing up of $C$ in $\mathbf{A}^n_k$. Since $E$ is an anti-ample divisor, we see that $X$ is an ample divisor on $T$. Hence we see that $$ U = T \setminus X = \text{Spec}(A) $$ is an affine scheme (complement of an ample divisor on a scheme projective over an affine scheme is affine). It is smooth over $R$ as the non-smooth locus of $b$ is $E \cap X$. The special fibre $U_0 = \text{Spec}(A_0)$ is an $\mathbf{A}^c$-bundle over $C$ where $c$ is the codimension of $C$ in $\mathbf{A}^n_k$. Thus we see that $Pic(A_0) = Pic(U_0) = Pic(C)$ has nonzero $2$-torsion elements.

Finally, the generic fibre of $\text{Spec}(A)$ is $\mathbf{A}^n_K$ where $K$ is the fraction field of $R$. Since $t$ is a prime element of $A$ this implies that $\text{Pic}(A)$ is trivial.

Example: If $C$ is a hypersurface $C = V(\overline{f})$, then I think you just get $A = R[x_1, \ldots, x_n, y]/(ty - f)$ where $f$ is a lift of $\overline{f}$. You can verify all the properties of this directly, I guess.

$\endgroup$
2
  • $\begingroup$ I understand. This is great. Thanks $\endgroup$
    – user315884
    Commented Aug 12, 2021 at 18:03
  • $\begingroup$ And thanks for taking the time to write this! $\endgroup$
    – user315884
    Commented Aug 12, 2021 at 18:12

You must log in to answer this question.