10

Define an inverse system of polynomial rings over a commutative ring $k$ by the canonical projection $k[x_1,...,x_n] \to k[x_1,...,x_m]\;(m< n)$.

Question: What is the projective limit $\varprojlim_n k[x_1,...,x_n]$ ?

Tomasz
  • 156
  • Looks like $k[x_1,x_2,\ldots]$ to me... – Andrew Sep 30 '12 at 23:57
  • I think you are mixing up projective limit and direct limit: $k[x_1,...]$ is the direct limit of $k[x_1,...,x_n] \hookrightarrow k[x_1,...,x_m];(n < m)$. – Tomasz Oct 01 '12 at 00:28
  • I also thought about this power series ring. But have have no idea on how to define the projections $k[[x_1,x_2,...]] \to k[x_1,...,x_n]$ ? – Tomasz Oct 01 '12 at 00:52
  • Ups, the previous comment concerns a comment of John Stalfos who suggested the projective limit in question is $k[[x_1,x_2,...]]$ (and was deleted while I was typing my comment). – Tomasz Oct 01 '12 at 00:56
  • What was the problem with $k[x_1,x_2,...],$ again? There are projections onto each element of your inverse system which commute with the maps in the system, though I haven't tried to check universality. – Kevin Carlson Oct 01 '12 at 01:24
  • 1
    No element of $k[x_1, x_2, \cdots]$ (with the canonical projection) is capable of mapping to $\sum_{i=1}^n x_i \in k[x_1, \cdots, x_n]$ for every $n$... but there is a diagram of maps from $k[y]$ that does. (I assume the "canonical projection" mentioned in the question sends $x_n \to 0$, although that really isn't canonical) –  Oct 01 '12 at 01:32
  • @Kevin: The sequence $x=(x_1+...+x_n)_{n\ge 1}$ is an element in the projective limit (considered as subring of $\prod_n k[x_1,...,x_n]$) which I think has no counterpart in $k[x_1,x_2,...]$. In analogy with p-adic numbers I would interpret $x$ as $x_1 + x_2 + ...$ which is an element in $k[[x_1,x_2,...]]$. But I don't know if this a correct. – Tomasz Oct 01 '12 at 01:34
  • Ah, yes, ignore what I said, $k[x_1,\ldots]$ is a direct limit. – Andrew Oct 01 '12 at 03:00

3 Answers3

5

Note that we can write, as sets:

$$ k[x_1, x_2, \cdots, x_n] \cong k[x_1] \times x_2 k[x_1, x_2] \times \cdots \times x_n k[x_1, x_2, \cdots, x_n]$$

(all but the first component of the product are ideals) An element of the product is interpreted as the sum of all of its components. Furthermore, the canonical projection (assuming you mean to send $x_n$ to $0$) is just the projections onto the first $m$ components.

Therefore,

$$ \lim_n k[x_1, x_2, \cdots, x_n] \cong k[x_1] \times x_2 k[x_1, x_2] \times x_3 k[x_1, x_2, x_3] \times \cdots $$

What addition is should be straightforward. Working with multiplication should be similar in flavor to working in power series rings.

The elements should probably be thought of as infinite sums; given a finite set of variables, each such sum should only have finitely many monomials that involve only those variables.

  • 1
    Dear Hurkyl, what you write seems correct but I think your answer would be clearer if you described your ring simply as a subring of $k[[x_1, x_2,x_3,...]]$. Meanwhile, +1. – Georges Elencwajg Oct 02 '12 at 16:25
1

This question was recently asked again here, so I thought I'd give an alternative viewpoint, presenting the ring as a ring of formal sums.

A larger ring

Let $I$ range over sequences $(i_1,\ldots,i_n,\ldots)$ of natural numbers with only finitely many nonzero terms. Call this set of sequences $\newcommand\N{\Bbb{N}}\N^{(\N)}$. Consider formal sums over this set of sequences, $$ \sum_{I} a_I x^I, $$ with the $a_I\in R$, and where $x^I$ is a formal element we think of as representing the (finite) monomial $$\prod_{n=1}^\infty x_n^{i_n}.$$

With no restrictions on the $a_I$, the collection of all such formal sums form a ring, with the product the usual convolution product, $$ \newcommand\of[1]{\left({#1}\right)} \of{\sum_I a_I x^I} \of{\sum_J b_J x^J} = \sum_K c_K x^K, $$ where $$c_K = \sum_{I+J=K} a_Ib_J,$$ noting that the sum is finite, since $K$ has only finitely many nonzero terms.

Also note that the set of formal sums can be identified with the set of functions, $\mathbf{Set}(\N^{(\N)},R)$.

Some comments

Note that this ring is too big to be our limit, since it contains power series rings, things like $\sum_{n=0}^\infty x_1^n$.

In fact, this is variant (iii) of a formal power series ring in infinitely many variables $R[[x_1,\ldots,x_n,\ldots]]$ here.

The limit ring as a subring

Let $S$ be the ring of all formal sums defined above. We want to produce a subring $T$ which will be our limit ring. The key problem is that $S$ contains terms of infinite degree in a fixed variable, which is not allowed in our limit.

Let $\pi_n : S\to R[[x_1,\ldots,x_n]]$ be the projection onto this subset of the variables, defined by noticing that we have an inclusion $\iota_n: \N^n\hookrightarrow \N^{(\N)}$ as the first $n$ coordinates, so precomposition induces a map $$\iota_n^* : \mathbf{Set}(\N^{(\N)},R)\to \mathbf{Set}(\N^n,R)$$ that respects the ring structure.

We have that $$ \pi_n\of{\sum_I a_Ix^I}= \sum_J b_Jx^J, $$ where $J$ ranges over multi-indices $J=(j_1,\ldots,j_n)$ and $b_J = a_{\iota_n(J)}$.

Then define $$ T = \bigcap_n \pi_n^{-1}(R[x_1,\ldots,x_n]).$$

$T$ is our limit.

By definition of $T$, the maps $\pi_n: S\to R[[x_1,\ldots,x_n]]$ restrict to maps $\pi_n : T\to R[x_1,\ldots,x_n]$. It's not hard to check that such a definition gives us a cone from $T$ to our sequence in $R$-algebras.

On the other hand, given a cone from another $R$-algebra $T'$ to each polynomial ring, $\psi_n : T'\to R[x_1,\ldots,x_n]$, then define $\tilde{\psi} : T'\to T$ by $\tilde{\psi}(t) = \sum_I a_I x^I$, with $a_I$ the coefficient of $\pi_N(x^I)$ in $\psi_N(t)$, where $N$ is large enough that $i_m = 0$ for $m> N$, since $I$ has only finitely many nonzero terms. It's not hard to check that this is well defined and unique.

Edit note

Originally I defined $T$ in the following manner:

If $a=\sum_I a_I x^I$ is a formal sum, then define the degree of $x_n$ in $a$, denoted $[a]_n$ to be $$ \sup \{ k : \text{there exists $I$ with $a_I\ne 0$ and $i_n=k$}\}, $$ where the sup is taken in $\Bbb{N}\cup \{\infty\}$.

Then define $$T=\{a\in S : [a]_n <\infty\text{ for all $n$}.\}$$

But this is too strict, it excludes things like $$\sum_{n=0}^\infty x_1^nx_n.$$

jgon
  • 28,469
  • Thank you for such a nice answer! When you say that $\sum_{n=0}^{\infty}x_{1}^{n}$ doesn't live in the limit, do you really mean that things like $(x_1,x_1+x_{1}^{2},x_1+x_{1}^{2}+x_{1}^{3},x_1+x_{1}^{2}+x_{1}^{3}+x_{1}^{4},\dots)$ don't live in the limit? – billy192 Mar 29 '20 at 15:01
  • Hi @jgon. I really can't quite see why, for example, $\displaystyle \sum_{n=1}^{\infty} x_{1}^{n}$ , is not in $T$. Would you mind explaining? – amator2357 Apr 28 '20 at 23:07
  • @amator2357 Slightly informally, the reason is that when you set all but a finite number of variables to 0, you should get a polynomial. $\sum_{n=1}^\infty x_1^n$ is still a proper power series after setting all variables except $x_1$ to $0$. – jgon Apr 28 '20 at 23:21
  • I guess I phrased my question incorrectly. I understand why it shouldn't be. But if I understand it correctly, the elements $x_{1},x_{1}^2,\dots,x_1^{\infty}$ are in $T$ which is a subring of the ring of formal power series with operations inherited from it. What is stopping me from just adding them all inside $T$? – amator2357 May 13 '20 at 16:19
  • 1
    @amator2357 $x_1^\infty$ is not an element of this ring, nor of any power series or polynomial ring. You are correct that $x_1,x_1^2,\ldots,x_1^n,\ldots$ are all inside $T$, but you cannot add infinitely many elements together in a normal ring without a topological structure or something similar. After all, all of these elements are in $k[x_1]$ as well, but we can't add them all together insider $k[x_1]$ either. – jgon May 14 '20 at 04:49
  • @jgon thank you. I guess this point at infinity is what confuses me, among other things. I can't quite wrap my head around why is $\displaystyle \sum_{n=1}^{\infty} x_{1}^{n}$ not in $T$ but $\displaystyle \sum_{n=1}^{\infty} x_{1}^{n}x_n$ is. I mean again, I understand why one should be and the other one shouldn't be in there, but I can't quite see how this is achieved b defining $T$ in the way you did. – amator2357 May 14 '20 at 07:28
-1

I can answer your question. Call a linear combination of monomials $\prod_i x_i^{m_i}$ such that $m_i\ge 0, \sum_i m_i=m$ an $m$-form. Then as a subring $$\varprojlim_{n \to \infty} k[x_1,...,x_n] = \lbrace f_1 + \cdots + f_m \mid m \ge 0,\;\; f_i\; i\text{-form}\rbrace \le k[[x_1,x_2,...]]$$

For the proof note that there are canonical projections $p_n: k[[x_1,x_2,...]] \to k[[x_1,...,x_n]]$ and the restriction to $S = \lbrace f_1 + \cdots + f_m \mid m \ge 0,\;\; f_i\; i\text{-form}\rbrace$ has image in $k[x_1,...,x_n]$. Hence $p= \prod_n p_n: S \to \prod_n k[x_1,...,x_n]$ is an embedding of rings whose image is exactly $\varprojlim_{n \to \infty} k[x_1,...,x_n]$.

tj_
  • 2,141
  • Dear tj_, I don't think what you write is correct because you can't account for $x_1+x_2^2+x_3^3+...+x_r^r+...$, which is in the projective limit. – Georges Elencwajg Oct 02 '12 at 16:19