12
$\begingroup$

Let us define the infinitely-many-variable formal power series ring

$$ K[[X_1,\ldots]] \colon= \underset{m \geq 1}{\varprojlim}\,K[[X_1,\ldots,X_m]]. $$

$K[[X_1,\ldots]]$ is known to be a UFD by a theorem of Nishimura (c.f. On the unique factorisation theorem for formal power series, Journal Math. Kyoto. Univ. Vol 7. No 2. 1967, 151-160).

Now let us choose an irreducible element $f \in K[[X_1,\ldots]]$ and consider the image $f_m \in K[[X_1,\ldots,X_m]]$ of $f$ by the natural quotient ring homomorphism $K[[X_1,\ldots]] \twoheadrightarrow K[[X_1,\ldots,X_m]]$.

Q. Is $f_m$ also irreducible for $m \gg 0$?

$\endgroup$
1
  • $\begingroup$ I notice that you have never marked a question as "answered". If you feel that an answer satisfactorily addresses a question, you can click the check box next to it to communicate this. $\endgroup$ Sep 27, 2018 at 18:55

3 Answers 3

5
$\begingroup$

Permit me to make the following bibliographic remark: the very same article of Nishimura which was cited by OP, already contains an affirmative answer to the OP's question: (1) on page 157 of Nishimura's 1967 article one reads

enter image description here

Nishimura's proof, which seems self-contained and recommendable reading, uses too many preliminary results to be conveniently summarizable (by me). I tried to write an exposition, but that attempt foundered on my not understanding Nishiguro's argument in few places (which does imply anything for Nishimura's proof of course).

${}$___________________________

(1) So the OP either did not read or did not trust Nishigura's article in its entirety (and thought it more polite to not mention any issues that there may or may not be with Ishimura's argument); of course there isn't anything wrong with that; it's perfectly fine to quote Nishigura's article for the purpose of giving a reference for the UFD-claim only. I simply think it should be pointed out for completeness that Nishimura's article seems to already contain an answer.

$\endgroup$
4
  • 2
    $\begingroup$ Just out of curiosity, who are Ishimura, Nishigura and Nishiguro in this story? :-) $\endgroup$ Feb 20, 2018 at 20:58
  • $\begingroup$ Thanks for Peter's kindness. I all agree with you. Rinmyaku $\endgroup$
    – Pierre
    Feb 21, 2018 at 4:39
  • $\begingroup$ @IgorKhavkine: thanks for catching that; these are simply typos made when writing in a hurry; the spelling "Nishimura" is the only correct one. (That is, it is the transliteration into Latin characters which appears in Nishimura's publications.) I will only correct this when I have something more substantial to edit-it, which currently I haven't. $\endgroup$ Feb 21, 2018 at 6:58
  • 1
    $\begingroup$ Thanks Peter Heinig, and I agree with you. Pierre $\endgroup$
    – Pierre
    Mar 2, 2019 at 14:19
1
$\begingroup$

(I assume that $K$ is a field)

Decompose each $f_m$ into a product of irreducible series $f_m=g^1_m\dots g^{i_m}_m$. For each $m$ we get $f_m=\overline{g^{1}_{m+1}}\dots \overline{g^{i_{m+1}}_{m+1}}$ where overline denotes the reduction map $K[[x_1,\dots, x_m,x_{m+1}]]\to K[[x_1,\dots, x_m]]$. Since $K[[x_1,\dots,x_m]]$ is a UFD, $\overline{g^j_{m+1}}$ is a product of some elements $g^{k_1},\dots ,g^{k_{n_{j,m+1}}}$ up to a unit. More precisely, decompositions of $f_m$s give a sequence of partitions of the set $\{1,\dots, i_1\}$ such that the $m$-th partition is a refinement of the $(m+1)$-th for every $m$.

If there are no $N$ such that for every $m>N$ the series $f_m$ is irreducible, all these partitions consist of at least two elements. However, the sequence of positive integers $i_1\geq i_2\geq \dots$ must stabilize eventually, so there is a number $N$ such that for every $m\geq N$ we get (maybe after permuting the irreducible factors) $\overline{g^k_{m+1}}=g^k_mu^k_m$ where $u^k_m$ are units such that $u^1_{m}\dots u^{i_m}_{m}=1$. We will now modify the decompositions to obtain a decomposition of $f$. Namely for each $m\geq {N-1}$, put $h^k_{m+1}=g^{k}_{m+1}\cdot (\iota(u^k_N)\iota(u^{k}_{N+1})\dots \iota(u^{k}_{m}))^{-1}$ where $\iota^{m+1}_{l}:K[[x_1,\dots, x_l]]\to K[[x_1,\dots, x_{m+1}]]$ are the embeddings.

We've arranged things so that $f_{m}=h^{1}_m\dots h^{i}_m$ for all $m\geq N$ ($i$ is the stabilizing value of the sequence $(i_m)$) and $\overline{h^k_{m+1}}=h^k_m$. Hence, we get non-invertible elements $h^k$ of $K[[x_1,\dots]]$ such that $f=h^1\dots h^i$ so $f$ is reducible.

$\endgroup$
-1
$\begingroup$

I suppose the natural quotient ring homomorphism is the one that sets $X_k = 0$ for $k>m$.

For small $m$, I think that $X_1^2 - X_2^2 + X_3^2 \twoheadrightarrow (X_1 - X_2) (X_1 + X_2)$, with $m=2$, is a counter example. But for any $f \in K[[X_1,\ldots]]$ there is a sufficiently large $m$ such that $f \in K[[X_1,\ldots,X_m]]$ (by the structure of the inductive limit), so that $f_m = f$. But then, reducibility in $K[[X_1,\ldots,X_m]]$ trivially implies reducibility in $K[[X_1,\ldots]]$. The the answer to your Q seems to be Yes.

$\endgroup$
10
  • 6
    $\begingroup$ Wait, are we talking about the inductive or the projective limit of the finite-variable power series rings? The question says projective (I didn't look up Nishimura's result to check). $\endgroup$
    – Gro-Tsen
    Feb 19, 2018 at 12:49
  • $\begingroup$ @Gro-Tsen Aha, it's quite possible that I mixed them up. Even though I should know better! The inductive limit is easy to understand: it consists of the power series that depend on finitely many variables (without a limit on the number of variables). The projective limit seems rather weirder. What is its pedestrian description? $\endgroup$ Feb 19, 2018 at 22:54
  • $\begingroup$ I suppose that we are engaged in Projective limit. So such an element as X_1 + X_2 + ... ∈ K[[X_1,...]]. K[[X_1,...]] is much larger than the inductive limit of K[[X_1,...,X_m]]. $\endgroup$
    – Pierre
    Feb 20, 2018 at 3:38
  • $\begingroup$ @IgorKhavkine the projective limit is the set of power series such that, if you set sufficiently high-index variables all to zero, then you're left with a polynomial in the remaining variables. (This is assuming the projection maps are defined by $x_i = 0$ at each step, which I believe is standard.) $\endgroup$ Feb 20, 2018 at 4:34
  • 2
    $\begingroup$ @HarryRichman, hmm, I think what you wrote can't be strictly correct, since it corresponds to $\varprojlim_{m\to\infty} K[X_1,\ldots,X_m]$, which would exclude elements like $(1+X_1+X_1^2+\cdots) + X_2$. Deciphering Bourbaki, as per abx's reference, I think it's more accurate to say that any element of the projective limit is a sum of infinitely many homogeneous components, where each homogeneous component is a polynomial after setting all but finitely many variables to zero. $\endgroup$ Feb 20, 2018 at 9:38

Your Answer

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