8

Let $R$ be a commutative Noetherian ring, and let $M$ be a finitely generated $R$-module with a fixed automorphism $f$. Can one lift $f$ to an automorphism of a finitely generated free module $R^n$? In other words, is there an automorphism $\tilde f$ of some $R^n$ together with a surjection $\pi:R^n\to M$ such that $\pi\circ\tilde f = f\circ\pi$?

Here is my progress. If we are given a surjection $\pi:R^n\to M$, it is possible to lift $f$ to an endomorphism $f':R^n\to R^n$ making the relevant diagram commute. Furthermore, we can lift $f^{-1}$ to $f'':R^n\to R^n$ such that $f'\circ f''$ is the identity map "up to relations of $M$". My thought is that one could adjoin generators to $R^n$ that "represent the relations of $M$" in order to make an actual invertible map.

As an explicit example, it is impossible to lift the automorphism given by multiplication by $3$ on $\mathbb Z/8$ to an automorphism on $\mathbb Z$; however, you can lift it to an automorphism on $\mathbb Z^2$. Namely, if the generators of $\mathbb Z^2$ are $x$ and $y$, then the map $\mathbb Z^2\to \mathbb Z/8$ is $x\mapsto 1$ and $y\mapsto 0$, and the automorphism of $\mathbb Z^2$ is $(x,y)\mapsto (3x+8y,-2x-5y)$. Here, $y$ is the new generator that represents the relation of $\mathbb Z/8$.

Andrea B.
  • 833

1 Answers1

3

It is true, and your idea works (we lift the obstruction).

Proposition. Let $R$ be a ring and let $M$ be a finitely presented $R$-module. If $f : M \to M$ is an automorphism, then there is an epimorphism $\pi : G \to M$ from a finite free $R$-module $G$ and an automorphism $\widetilde{f~} : G \to G$ such that $$\begin{array}{ccc} G & \xrightarrow{\widetilde{~f~}} & G \\ \pi \downarrow ~~ && ~~\downarrow \pi \\ M & \xrightarrow{f} & M \end{array}$$ commutes.

Partial proof. Choose a presentation $$K \xrightarrow{i} F \xrightarrow{p} M \to 0$$ where $K,F$ are finite free. We will prove that $f$ lifts to an automorphism of $G := F \oplus K$.

In the following, I will also assume that $i$ is a monomorphism. I am pretty sure that the proof can be fixed without doing that assumption.

The rest of the proof works in an arbitrary abelian category, we will just need the presentation and that $K,F$ are projective.

Let $g := f^{-1}$. Let $K' := \mathrm{im}(i)= \ker(p)$. Since $F$ is projective, there are homomorphisms $\overline{f},\overline{g} : F \to F$ with $$p \overline{f} = fp, \quad p \overline{g} = g p.$$ The homomorphisms $\overline{f}, \overline{g} : F \to F$ don't have to be inverse to each other, but we can calculate the obstruction: We compute $$p(\overline{f} \overline{g}-1) = p\overline{f} \overline{g}-p = fp \overline{g} - p = f g p - p = 0.$$ Thus, $\overline{f} \overline{g}-1$ lands inside $K'$. Since $F$ is projective and $K \to K'$ is an epimorphism, there is a homomorphism $\alpha : F \to K$ with $$\overline{f} \overline{g} - 1 = i \alpha.$$ Similarly, there is a homomorphism $\beta : F \to K$ with $$\overline{g} \overline{f} - 1 = i \beta.$$ Next, since $p \overline{f} i = f p i = 0$ and $K$ is projective, there is a homomorphism $\gamma : K \to K$ with $$\overline{f} i = i \gamma.$$ Similarly, there is a homomorphism $\delta : K \to K$ with $$\overline{g} i = i \delta.$$

Recall that endomorphisms of a direct sum can be represented as matrices: $$\mathrm{End}(F \oplus K) = \begin{pmatrix} \mathrm{Hom}(F,F) & \mathrm{Hom}(K,F) \\ \mathrm{Hom}(F,K) & \mathrm{Hom}(K,K) \end{pmatrix}$$ Also, composition of endomorphisms is the usual matrix multiplication.

We consider the endomorphisms $$\widetilde{f~} := \begin{pmatrix} \overline{f} & -i \\ \beta & -\delta \end{pmatrix}, \quad \widetilde{g~} := \begin{pmatrix} \overline{g} & -i \\ \alpha & -\gamma \end{pmatrix}.$$ Let us compute the products. $$\widetilde{f~} \widetilde{g~} = \begin{pmatrix} \overline{f} \overline{g} - i \alpha & -\overline{f} i + i \gamma \\ \beta \overline{g} - \delta \alpha & -\beta i + \delta \gamma \end{pmatrix}$$ The entries in the first row are $1,0$ by construction. To compute the other entries, we use the assumption that $i$ is a monomorphism. Then we may compose with $i$. So to prove $\beta \overline{g} - \delta \alpha = 0$, it suffices to compute $$i(\beta \overline{g} - \delta \alpha)=i \beta \overline{g} - i \delta \alpha = (\overline{g} \overline{f}-1) \overline{g} - \overline{g} i \alpha=(\overline{g } \overline{f}-1)\overline{g} - \overline{g}(\overline{f} \overline{g}-1) = 0.$$ And to prove $-\beta i + \delta \gamma = 1$, it suffices to compute $$i(-\beta i + \delta \gamma) = -(\overline{g} \overline{f}-1)i + (\overline{g}i)\gamma = -(\overline{g} \overline{f}-1)i + \overline{g}(\overline{f}i) = i.$$ Thus, $$\widetilde{f~} \widetilde{g~} = \begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix} = 1.$$ The same proof shows $\widetilde{g~} \widetilde{f~} = 1$. Thus, $\widetilde{f~}$ is an automorphism of $F \oplus K$.

For $\mathrm{pr}_F : F \oplus K \to F$ we have $$\mathrm{pr}_F \widetilde{f~} = \overline{f} \, \mathrm{pr}_F - i \, \mathrm{pr}_K$$ by construction of $\widetilde{f~}$. Hence, the composition $\pi = p \circ \mathrm{pr}_F : F \oplus K \to M$ now satisfies $$\pi \widetilde{f~} = p \, \mathrm{pr}_F \widetilde{f~} = p (\overline{f} \, \mathrm{pr}_F - i \, \mathrm{pr}_K) = p \overline{f} \, \mathrm{pr}_F = f p \, \mathrm{pr}_F = f \pi,$$ and we are finally done.

  • If $i$ is not a monomorphism, probably we have another obstruction that we can lift *waves-hands* – Martin Brandenburg Sep 07 '24 at 20:01
  • Okay, I finally went through the details. I haven't thought about dealing with i not monic yet. I noticed two minor typos by the way: at one point you have fpb - p instead of fp(g-bar) - p. And towards the end, one should have - i pr_K instead of + i pr_K (this typo appears twice) – Andrea B. Sep 13 '24 at 20:59
  • $\widetilde f(x,y)=(\overline f x - i y, \text{blah})$ so $\mathrm{pr}_F \widetilde f (x,y) = \overline f x - i y = \overline f \mathrm{pr}_F (x,y) - i \mathrm{pr}_K(x,y)$ – Andrea B. Sep 13 '24 at 21:13
  • And for the first typo, it's in the display line after you say "don't have to be inverse to each other, but we can calculate the obstruction: We compute". There's an $fpb-p$, but $b$ is undefined. It should be $fp\overline g-p$ – Andrea B. Sep 13 '24 at 21:15