10

Let $k$ be a field. Is the group of $k$-automorphisms of $k[[x,y]]$ known? ($k[[x,y]]$ is the ring of formal power series in two variables, see Wikipedia.)

A somewhat relevant question is this question, which deals with $k[[x]]$, with $k$ any commutative ring.

Thanks for any hints and comments.

user26857
  • 52,094
user237522
  • 6,467
  • 1
    I meant for $k[x,y]$ apply to $(x,y,1)$ the invertible $k$-linear maps $k^2 \to k^2$. For $k[[x,y]]$ you need to find which elements are invertible, – reuns Apr 09 '17 at 04:11
  • I know that in $k[[x]]$ an element $\sum_{i=0} a_i x^i$ is invertible iff its constant coefficient $a_0$ is invertible in $k$ (=namely in $k-0$). I guess a 'similar' result holds in $k[[x,y]]$. – user237522 Apr 09 '17 at 04:17
  • In $k[[x,y]]$: Let $A=\sum a_{ij}x^iy^j$. For $A$ to be invertible in $k[[x,y]]$, it is clear that it is necessary that $a_{00}$ is invertible in $k$. Is $a_{00}$ invertible a sufficient condition for invertibility of $A$? – user237522 Apr 09 '17 at 04:26
  • @user237522 Yes, a power series with invertible constant term (you may as well make it 1) is invertible. It's essentially the geometric series: $(1 - f)^{-1} = 1 + f + f^2 + f^3 + ... $ – user399601 Apr 09 '17 at 04:29
  • A recursive formula for the coefficients of the inverse $B$ can be obtained similarly to the one variable case? – user237522 Apr 09 '17 at 04:31
  • @user237522 Yes, if $f(0) = 0$ then any monomial $x_1^{a_1} ... x_n^{a_n}$ can only occur in finitely many powers $f,f^2,f^3,...$ – user399601 Apr 09 '17 at 04:33
  • @user399601, thanks for your comments! – user237522 Apr 09 '17 at 04:40

1 Answers1

14

It is simple to describe all automorphisms of $R=k[[x,y]]$ as a $k$-algebra. First, note that $R$ is a local ring with maximal ideal $m=(x,y)$, so any automorphism must map $m$ to itself and thus be continuous in the $m$-adic topology. Since $k[x,y]$ is dense in $R$ in the $m$-adic topology, this means that an automorphism is determined by where it sends $x$ and $y$. Moreover, given $f,g\in m$, there is a unique continuous $k$-algebra homomorphism $\varphi:R\to R$ such that $\varphi(x)=f$ and $\varphi(y)=g$ (given by "evaluating" power series by plugging in $f$ for $x$ and $g$ for $y$). So the only question is what conditions on $f$ and $g$ guarantee that this $\varphi$ is an automorphism.

The answer is simple: you just need the images of $f$ and $g$ in the $k$-vector space $m/m^2$ to be linearly independent. Concretely, this just means that the linear homogeneous parts of $f$ and $g$ are linearly independent. Clearly this condition is necessary, since the images of $x$ and $y$ in $m/m^2$ are linearly independent. Conversely, if the linear homogeneous parts of $f$ and $g$ are linearly independent, it is easy to show $\varphi$ is surjective (for any homogeneous polynomial $p$ of degree $n$, you can find a homogeneous polynomial $q$ of degree $n$ such that $p$ is the degree $n$ part of $\varphi(q)$, and then you can use this to build a power series whose image under $\varphi$ has any desired value, one degree at a time). A surjective homomorphism from a Noetherian ring to itself is automatically injective, so it follows that $\varphi$ is an automorphism.

So automorphisms of $R$ are in bijection with pairs of power series with no constant term whose linear parts are linearly independent. Beware that the group operation is very complicated from this description--to compose two automorphisms, you need to compose the power series (that is, substitute the power series of one automorphism for the variables $x$ and $y$ in the power series of the other automorphism).

Eric Wofsey
  • 330,363
  • Thank you very much! Your answer seems what I was looking for (I have to check some details in your answer that are not clear to me yet). – user237522 Apr 09 '17 at 04:47
  • I guess it is possible to replace $x,y$ by $x_1,\ldots,x_n$ and get a similar answer to the above? – user237522 Apr 09 '17 at 04:52
  • Yeah, essentially the same story holds for power series in any number of variables. – Eric Wofsey Apr 09 '17 at 05:28
  • Please, there is something that I do not understand: According to your answer $(x,y) \mapsto (x+1,y+2)$ is not an automorphism, but it is (since it is surjective). What am I missing? Thanks. – user237522 Apr 12 '17 at 02:15
  • That isn't even a well-defined map. To be able to evaluate a power series with $f$ and $g$ substituted for $x$ and $y$, $f$ and $g$ must be in $m$. For instance, try defining where your map would send $\sum_n x^n$. – Eric Wofsey Apr 12 '17 at 02:17
  • Oh, you are right...thanks!! – user237522 Apr 12 '17 at 02:22
  • Please, how one shows that $y$ is in the image of $(x,y) \mapsto (x,y+y^2)$? – user237522 Apr 12 '17 at 02:56
  • Just solve for the terms one by one in increasing order of degree. You want to start with $y$ to get the linear term of $y$. Then you'll need to subtract $y^2$ to cancel out the quadratic term. Then you'll need to add $2y^3$ to cancel out the cubic term. And so on. – Eric Wofsey Apr 12 '17 at 03:06
  • Thanks for the clarification! Indeed, $y= 1(y+y^2)-1(y+y^2)^2+2(y+y^2)^3-5(y+y^2)^4+\ldots$. – user237522 Apr 12 '17 at 03:20