8
$\begingroup$

Let $C$ be a $\mathbb{Z}$-linear category, such that $C(x,y)$ is a free abelian group with finite rank, for every $x,y\in\mathrm {Ob}(C)$. Given a commutative ring with identity $R$, let $RC$ denote the category with the same objects of $C$, and morphisms $RC(x,y):=R\otimes_{\mathbb{Z}} C(x,y)$.

Does any isomorphism in $\mathbb{F}_pC(x,y)$ lift to an isomorphism in $\mathbb{Z}^{\wedge}_pC(x,y)$?

$\endgroup$
0

2 Answers 2

4
$\begingroup$

It suffices to show that you can lift isomorphisms along $AC\to BC$ whenever $A\to B$ is a square zero extension. (EDIT : here I'm using finite generation of $C(y,x), C(x,y)$ to obtain that $\mathbb Z_p\otimes C(x,y)$ is $p$-adically complete)

So let $\sum_i b_i\otimes f_i \in BC(x,y)$ be an isomorphism, with inverse $\sum_j d_j\otimes g_j$.

Choose lifts $\tilde b_i, \tilde d_j$ in $A$ of $b_i, d_j\in B $. Then $(\sum_i \tilde b_i\otimes f_i)\circ (\sum_j \tilde d_j \otimes g_j) = id_y + \epsilon$, where $\epsilon \in \ker\otimes C(y,y)$, where $\ker$ is the kernel of $A\to B$. Indeed, because $C(y,y)$ is flat, $\ker\otimes C(y,y)$ is in particular the kernel of $A\otimes C(y,y)\to B\otimes C(y,y)$ (EDIT: as R. van Dobben de Bruyn pointed out, I'm not using flatness here, just right exactness of the tensor product).

In particular, because $\ker^2 = 0$, you find that $\epsilon\circ \epsilon = 0$, and so $id_y + \epsilon$ is invertible in $A\otimes C(y,y)$. In particular, $\sum_i \tilde b_i \otimes f_i$ has a right inverse, and $\sum_j \tilde d_j \otimes g_j$ has a left inverse.

Now (with the same lifts !) reasoning symmetrically (in $C(x,x)$ ) shows that they each have an inverse on the other side, so they are both isomorphisms. Either one of them is a lift, as desired.

In particular, if you were doing this over another base ring than $\mathbb Z$, really the only thing you would need is for an analogue of $\mathbb Z_p\otimes C(x,y)$ (resp. $(y,x)$) being $p$-adically complete.

$\endgroup$
4
$\begingroup$

Here's a simplification of Maxime Ramzi's answer. Given an isomorphism $\bar f \in \mathbf F_p\mathscr C(x,y)$, the claim is that any lift $f \in \mathbf Z_p\mathscr C(x,y)$ of $\bar f$ is an isomorphism.

Indeed, if $g \in \mathbf Z_p\mathscr C(y,x)$ is a lift of $\bar g = \big(\bar f\big)^{-1}$, we know that $gf = 1 - p\phi$ for some $\phi \in \mathbf Z_p\mathscr C(x,x)$. The series $\psi = \sum_{i=0}^\infty p^i\phi^i$ converges in $\mathbf Z_p\mathscr C(x,x)$ and gives a two-sided inverse of $gf$, so $f$ has a left inverse $\psi g$ (and $g$ has a right inverse $f\psi$). Swapping the roles of $f$ and $g$ shows that $f$ has a right inverse as well, so $f$ is invertible. $\square$


Remark. A note on the hypotheses: flatness is never used; the only role it plays in Ramzi's answer is uniqueness of $\phi$ such that $gf = 1-p\phi$, but this is inconsequential. On the other hand, it is crucial that the $\mathscr C(x,y)$ are finitely generated, as we need $\mathbf Z_p\mathscr C(x,y)$ to be $p$-adically complete [Tag 00MA(3)].

$\endgroup$
1
  • $\begingroup$ You're absolutely right, on both accounts ! I forgot to mention this $p$-completeness thing at the beginning, and I didn't realize that I wasn't actually using flatness :) $\endgroup$ Apr 7, 2022 at 16:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.