Local structure of étale algebra

Thierry Coquand · 2026

Read Local structure of étale algebra by Thierry Coquand (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-04143.txt.

Introduction

Thierry Coquand, writing alone, gives a constructive version of the proof for the local structure of étale algebras. Étale algebras are a central class in algebra, valued because their local behaviour can be described in a standard form. A constructive proof, unlike a classical one, builds its objects and avoids steps that rely on pure existence claims.

Catalog record

CREATORThierry Coquand
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE9 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-04143

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Local structure of étale algebra Thierry Coquand Abstract Introduction The goal of this note is to provide a constructive version of the proof of local structure of étale algebra, attributed to Chevalley in [ Gr , 18.4.6 (ii)] , in the setting of constructive mathematics [ LQ ] . As in [ Gr , Ray ] , the main tool is Zariski Main Theorem, that was already proved constructively in [ ACL ] . 1 Statement of the local result Let R be a residually discrete local ring with maximal ideal \mathfrak{m} and residual field k=R/\mathfrak{m} . Let E be a (finitely presented) étale R -algebra. Theorem 1 . There exists s_{1},\dots,s_{m} in E that are comaximal mod. \mathfrak{m} , i.e. their images generate the unit ideal in E/\mathfrak{m}E , and each E[1/s_{i}] is a standard étale R -algebra. Note that E/\mathfrak{m}E is a (finitely presented) unramified k -algebra and so, following [ LQ , L ] , is a finite product of monogene separable extensions of k . Using constructive Zariski Main Theorem [ ACL ] we get that there exists a finite R -subalgebra B of E and elements s_{i}\in B , comaximal in E , such that B_{i}=B[1/{s_{i}}]\cong E[1/s_{i}] for all i . Some B_{i}/\mathfrak{m}B_{i} may be the trivial algebra. We discard them, which explains why in the end result, we only get a family which is comaximal residually . Each B_{i}/\mathfrak{m}B_{i} is a corresponding product of monogene separable extensions. By refining s_{i} , we can assume that each B_{i}/\mathfrak{m}B_{i} is a itself a monogene separable extension of k . 2 Some lemmas about B/\mathfrak{m}B Consider now A=B/\mathfrak{m}B which is a finite k -algebra. We still write s_{i} for its class in A . Then A[1/s_{i}] is an extension of k of the form k[X]/(p) for some separable polynomial p . Lemma 1 . We have u and N such that s_{i}^{N}(1-s_{i}u)=0 . If we write e_{i}=(s_{i}u)^{N} , we have an idempotent e_{i} of A such that A[1/e_{i}]=A[1/s_{i}] . Proof. Since A is finite, s_{i} is integral and satisfies a polynomial equation. Since A[1/s_{i}] is non trivial, we can write this polynomial on the form X^{N}(1-XP) . We then have s_{i}^{N}=s_{i}^{N}(s_{i}u)^{N} and so e_{i}=e_{i}^{2} for e_{i}=(s_{i}u)^{N} . ∎ A[1/e_{i}] is a extension of the form k[X]/(p) . We can always assume that the constant coefficient of p is \neq 0 . (Since p is separable, the only problem is in the case p=X , but then we replace it by p=X-1 .) We can choose x in B such that x generates A[1/e_{i}] . We then have k[x]e_{i}=Ae_{i} . If we take y=xe_{i} we still have Ae_{i}=k[y]e_{i} but now, since we have P(x)=0 and the constant coefficient of P is \neq 0 , it is in xk[x] , and so e_{i} is in yk[y] and so Ae_{i}\subseteq k[y] . 3 Lifting to B B is finite over R , and,from the previous section, we have y in B and e_{i} in R[y] such that e_{i}B\subseteq R[y]+\mathfrak{m}B . By Nakayama’s Lemma, we get p_{i} in R[y] such that p_{i}B\subseteq R[y] 1 1 1 Let b_{1},\dots,b_{m} be generators of the R -module B . We can write e_{i}b_{j}=q_{j}+\Sigma_{l}\mu_{jl}b_{l} with q_{j} in R[y] and \mu_{jl} in \mathfrak{m} . We let p_{i} be the determinant of the matrix e_{i}\delta_{jl}-\mu_{jl} . . We then have B[1/p_{i}]\cong R[y][1/p_{i}] . To summarize, we have found t_{1},\dots,t_{n} in B , comaximal modulo \mathfrak{m} , such that each E[1/t_{i}] is of the form R[y][1/g] for some y in B and g in R[y] . 4 A candidate for a standard étale presentation We now consider the étale algebra E[1/t_{i}] and rewrite it as E=R[y][1/g] . We look at the R -algebra R[y] . We know that R[y]/\mathfrak{m}R[y] is a monogene finite k -algebra k[y] , but we don’t have access to its rank constructively in general. We also know that k[y][1/g] is a monogene separable algebra k[X]/(p) for some polynomial p . This implies that we have g^{n}p(y)=0 in k[y] for some n . Let G in R[X] such that G(y)=g . Since R[y] is finite, y satisfies a monic equation Q_{0}(y)=0 for some monic Q_{0} in R[X] . We define recursively the sequence of monic polynomials Q_{l} of decreasing degree d(Q_{l}) in R[X] , all satisfying Q_{l}(y)=0 in R[y] , starting with Q_{0} and stopping when d(Q_{l+1})=d(Q_{l}) . Let P be gcd(Q_{l},G^{n}p) in k[X] . We have P(y)=0 in k[y] since both Q_{l}(y)=0 and G(y)^{n}p(y)=0 . Let d=d(P)\leqslant d(Q_{l}) be the degree of P . We have that 1,y,\dots,y^{d-1} generate k[y]=R[y]/\mathfrak{m}R[y] , and so, using Nakayama’s Lemma, also generate R[y] . It follows that y also satisfies a monic equation Q_{l+1}(y)=0 for some monic Q_{l+1} in R[X] of degree d . So after a finite number of steps, we obtain Q=Q_{l} monic in R[X] such that Q(y)=0 in k[y] and Q divides G^{n}p in k[X] . Consider now R[a]=R[X]/(Q) , and S=R[a][1/G(a)] . We have G^{n}(a)p(a)=0 in S/\mathfrak{m}S and G(a) invertible in S , so p(a)=0 in S/\mathfrak{m}S . Also, if q(a)=0 in S/\mathfrak{m}S for some q in k[X] , we have that Q divides G^{m}q for some m in k[X] and so q(y)=0 in k[y][1/g] and so d(p)\leqslant d(q) . It follows that the canonical map u:S\rightarrow R[y][1/g]=E is a surjection such that S/\mathfrak{m}S\rightarrow E/\mathfrak{m}E is an isomorphism 2 2 2 Note that, contrary to Raynaud [ Ray ] , we cannot be sure to have the canonical map R[a]\rightarrow R[y] to be an isomorphism residually, since this would amount to be able to compute the rank of k[y] ; but it becomes a bijection residually when we localise. . All this works so far only assuming E (finitely presented) unramified. 5 What happens if E is étale We have a surjection u:S\rightarrow E , and S and E are finitely presented, so the kernel I is an ideal of S which is finitely generated. Furthermore S/\mathfrak{m}S=E/\mathfrak{m}E=k[y][1/g] . We have the exact sequence 0\rightarrow I/I^{2}\rightarrow S/I^{2}\rightarrow E\rightarrow 0 , and, since E is étale and I/I^{2} of square 0 , this exact sequence is split (in the sense of existence of a section). Furthermore, if an exact sequence 0\rightarrow A\rightarrow B\rightarrow C\rightarrow 0 is split, then the sequence 0\rightarrow A/\mathfrak{m}A\rightarrow B/\mathfrak{m}B\rightarrow C/\mathfrak{m}C\rightarrow 0 is still exact. Since S/I^{2}\rightarrow E is an isomorphism mod. \mathfrak{m} , it follows that I/I^{2} is 0 mod. \mathfrak{m} , and so, since I is finitely generated, we have I=I^{2} by Nakayama. This implies that I is generated by an idempotent element e=1-f . Thus we get f in S such that I[1/f]=0 and u(f)=1 . Then S[1/f] is isomorphic to E which is standard étale. 6 A global version We now take R an arbitrary ring and E a finitely presented étale R -algebra. Theorem 2 . There exists s_{1},\dots,s_{m} in E that are comaximal and f_{1},\dots,f_{m} in R comaximal in E such that each E[1/s_{i}] is isomorphic to a standard étale R[1/f_{i}] -algebra. Proof. We introduce a generic (decidable) ideal prime q of E , and p=q\cap R . We applied the arguments in the previous sections to E_{p} which is étale on R_{p} , which is a local ring. We then get s not in q and

Read the complete text