9

Let $R$ be a commutative ring with unit, $R[[t]]$ the ring of formal power series over $R$ and $R((t))$ the ring of formal Laurent series of $R$.

It is easy to see (and well known) that the group of units $R[[t]]^{\times}$ equals $R^{\times}+tR[[t]]$. On the other hand, the group of units $R((t))^{\times }$ is a less familiar object. In this paper (page 7 example 2.9) it is stated without proof that

$$ R((t))^\times = \{\sum a_n t^n \mid \exists n_0 \in \mathbb{N}, a_{n_0}\in R^\times, a_n \mbox{ nilpotent for } n<n_0 \} $$

While it is clear that all the elements on the RHS are units, I don't see why all units are of this form. Actually, it seems to me just wrong. Over $R=R_1 \times R_2$ we have

$$ ((1,0)\cdot 1+(0,1)\cdot t)((0,1)\cdot t^{-1}+(1,0)\cdot 1)=1 $$

Which seems to be a counter example.

Question 1: Is this indeed wrong, or am I missing something? If it is correct, how to prove it? If not, How can one describe explicitly $R((t))^\times?$

Right after that, there is a claim about $R((t))^{\times}/R[[t]]^\times$, but it also seems incorrect or at least I don't see why it is true.

Question 2: Is there a nice description of the group $R((t))^{\times}/R[[t]]^\times$?

KotelKanim
  • 2,436
  • The cited unit in $R_1 \times R_2$ can be written as $\sum a_i t^i$ with the stated condition having $n_0 =1$. What is wrong here? –  Oct 20 '13 at 10:45
  • @Doldrums, the element $(0,1)$ is neither a nilpotent nor a unit in $R$. – KotelKanim Oct 20 '13 at 10:56
  • "It is easy to see (and well known) that the group of units $R[[t]]^\times$ equals $R^\times[[t]]$" -- You don't mean $R^\times[[t]]$, right? The group of units is $R^\times + (t)$. – Martin Brandenburg Oct 20 '13 at 11:50
  • 1
    Your counterexample is correct. Perhaps the paper only needs the description locally. – Martin Brandenburg Oct 20 '13 at 11:57
  • @MartinBrandenburg, thanks for the correction! this was a typo. About your other comment, is it true locally? (by which I guess you mean Zariski-locally, or maybe something more fancy like fpqc-locally?) – KotelKanim Oct 20 '13 at 12:15
  • Maybe you can contact Ulrich Görtz directly. – Martin Brandenburg Oct 20 '13 at 12:49
  • I guess I can, but I am more interested in these specific questions, while for the paper it is a point of relatively minor importance. I think the joint powers of this community have no lesser chances to provide a good answer. – KotelKanim Oct 20 '13 at 13:03
  • This is a nice question. U. Görtz's statement is correct if $R$ has no non-trivial idempotent elements (that is, $\mathrm{Spec}(R)$ is connected). So your counterexample is in some sense the only possible. I will write more details later. – Cantlog Oct 21 '13 at 08:11
  • 1
    $(a_0+a_1t+\cdots )/t^N$ is invertible if and only if there exists a finite partition of $\mathrm{Spec}(R)$ by principal open subsets $U_0, \dots, U_n$ such that $a_0$ is invertible on $U_0$ and nilpotent on $U_1\cup ... \cup U_n$, $a_1$ is invertible on $U_1$ and nilpotent on $U_2\cup ... \cup U_n$ etc, and $a_n$ is invertible on $U_n$. – Cantlog Oct 21 '13 at 08:16
  • @Cantlog, it would be great if you could expand on this in an answer (when you have time). If you have insights to the second question, it would be great too! thank you. – KotelKanim Oct 21 '13 at 08:24
  • 1
    I don't have a clear picture for question (2). You should probably post it in a separate message. – Cantlog Oct 21 '13 at 23:33
  • 3
    In fact there is no mistake in Görtz's paper : his claimed description of $R((t))^\times$ assumes that $\text{Spec}(R)$ is connected -- this is written in the very sentence describing $R((t))^\times$. – Matthieu Romagny Mar 30 '15 at 18:06

1 Answers1

5

Here is the statement concerning question (1):

A Laurent series $u=(a_0+a_1t+\cdots)/t^N$ is a unit if and only if there exists a finite partition of $\mathrm{Spec}(R)$ by principal open subsets $U_0, \dots,U_n$ such that:

  • $a_0$ is invertible on $U_0$ and nilpotent on $U_1\sqcup \cdots \sqcup U_n$;

  • $a_1$ is invertible on $U_1$ and nilpotent on $U_2\sqcup \cdots \sqcup U_n$...

  • $a_n$ is invertible on $U_n$.

In particular, on each $U_k$, $u$ is of the form described in Görtz's paper.

Proof. We have $R((t))=R[[t]][1/t]$. So an element $a(t)/t^N$ with $a(t)=a_0+a_1t+\cdots\in R[[t]]$ is invertible in $R((t))$ if and only if there exists $b(t)=b_0+b_1t+\cdots\in R[[t]]$ such that $a(t)b(t)=t^m$ for some $m\ge 0$.

Developping the product $a(t)b(t)$, we get $$a_0b_0=0, \ a_1b_0+a_0b_1=0, ..., \ a_{m-1}b_0+\cdots +a_0b_{m-1}=0; \ a_mb_0+\cdots +a_0b_m=1.$$ Multiply these equalities by suitable powers of $a_0$ and we obtain $a_0^{k+1}b_k=0$ for all $k\le m-1$ and $a_0^{m+1}b_m=a_0^m$. Therefore $$ a_0^{m}(1-a_0b_m)=0.$$ Let $U_0=D(a_0)$ be the principal open subset of $\mathrm{Spec}(R)$ defined by $a_0$. Then $a_0$ is invertible on $U_0$. Let $V_0=D(1-a_0b_m)$. Then $\mathrm{Spec}(R)$ is the disjoint union of $U_0$ and $V_0$, and $a_0$ is nilpotent on $V_0$. As $t^{n-1}(1-a_0b(t)t^{-n})=(a_1+a_2t+\cdots)b(t)$ and $a_0$ is nilpotent on $V_0$, we see that $a_1+a_2t+\cdots$ is invertible in $O(V_0)((t))$ (see KotelKanim's comment). So we can start again with $a_1+a_2t+\cdots$ to find that $V_0=(D(a_1)\cap V_0)\sqcup V_1$ for some principal open subset $V_1$ of $V_0$ such that $a_1$ is nilpotent on $V_1$. As $D(a_1)\cap V_1\subseteq D(a_1)\cap V_0\cap V_1=\emptyset$, we have $$\mathrm{Spec}(R)=(D(a_0)\cup D(a_1))\sqcup V_1.$$ By induction, we find $V_k=(D(a_{k+1})\cap V_k)\sqcup V_{k+1}$ with $a_{k+1}$ nilpotent on $V_{k+1}$ and $$\mathrm{Spec}(R)=(D(a_0)\cup ... \cup D(a_{k+1}))\sqcup V_{k+1}.$$ Note that the equality $a_0b_m+\cdots+a_mb_0=1$ implies that $\mathrm{Spec}(R)=\cup_{0\le i\le m} D(a_i)$. So the above construction must stop at some step $n\le m$, we find the partition with $U_1=D(a_1)\cap V_0$, ..., $U_n=D(a_n)\cap V_{n-1}$.

It remains to shwo that if $a(t)$ satisfies the condition of partition, then $a(t)$ is a unit in $R((t))$. But it is enough to notice that on each of $U_i$, it is a unit.

Cantlog
  • 2,962
  • Excellent! It is not important, but I don't understand why $t^{n-1}(1-a_0 b(t)t^{1-n})=a_1 +a_2 t+...=a'(t)$. I do understand though why $a'(t)$ is invertible on $O(V_0)((t))$, just because $a'(t)b(t)=t^n-a_0 b(t)$ and this is a sum of a unit and a nilpotent, hence unit. – KotelKanim Oct 22 '13 at 08:06
  • @KotelKanim: sorry, this is typo. My intention was to show "explicitely" that $a_1+a_2t+\cdots$ is invertible. – Cantlog Oct 22 '13 at 21:46
  • No problem, your answer is great. thanks again. And I will probably post the second question in a different post as you suggested. – KotelKanim Oct 23 '13 at 06:44