Bourbaki

4. Lifting in complete rings

1. Strongly relatively prime polynomials

Let RR be a commutative ring. Two elements x,yx, y of RR are called strongly relatively prime if the principal ideals RxRx and RyRy are relatively prime, in other words (Chapter II, § 1, no. 2) if Rx+Ry=RR x + R y = R; it amounts to the same to say that there exist two elements a,ba, b of RR such that ax+by=1ax + by = 1.

Lemma 1 ("Euclid's Lemma"). Let x,yx, y be two strongly relatively prime elements of RR; if z∈Rz \in R is such that xx divides yzyz, then xx divides zz.

If 1=ax+by1 = ax + by, then z=x(az)+(yz)bz = x(az) + (yz)b.

If xx and yy are strongly relatively prime in RR, then

Rxy=(Rx)∩(Ry) R_{xy} = (Rx) \cap (Ry)
(Chapter II, § 1, no. 2, Proposition 5); if RR is an integral domain, two strongly relatively prime elements then have an lcm equal to their product (Algebra, Chapter VI, § 1, no. 8) and are therefore relatively prime in the sense of Algebra, Chapter VI, § 1, no. 12. Conversely, if RR is a principal ideal domain, two relatively prime elements are also strongly relatively prime, as follows from Bezout's identity (Algebra, Chapter VII, § 1, no. 2, Theorem 1).

For polynomial rings there is the following result:

Proposition 1

Let A\mathbf{A} be a commutative ring and PP and P′P' two strongly relatively prime polynomials in A[X]\mathbf{A}[X]. Suppose that PP is monic and of degree ss. Then every polynomial TT in A[X]\mathbf{A}[X] may be written uniquely in the form*

T=PQ+P′Q′ T = PQ + P'Q'
*where Q∈A[X]Q \in \mathbf{A}[X], Q′∈A[X]Q' \in \mathbf{A}[X] and deg⁡(Q′)<s\deg(Q') < s. If further deg⁡(T)≤t\deg(T) \leq t and deg⁡(P′)≤t−s\deg(P') \leq t - s, then deg⁡(Q)≤t−s\deg(Q) \leq t - s.

As PP is monic, PR≠0PR \neq 0 for every polynomial R≠0R \neq 0 of A[X]\mathbf{A}[X] and in this case deg⁡(PR)=s+deg⁡(R)\deg(PR) = s + \deg(R).

Let TT be any polynomial in A[X]\mathbf{A}[X]. As the ideal generated by PP and P′P' is the whole of A[X]\mathbf{A}[X], there exist polynomials Q1Q_1 and Q′Q' such that

T=PQ1+P1′Q1′; T = PQ_1 + P'_1Q'_1;
as PP is monic of degree ss, Euclidean division (Algebra, Chapter IV, § 1, no. 5) shows that there exist two polynomials Q′,Q′′Q', Q'' such that Q1′=PQ′′+Q′Q'_1 = PQ'' + Q' where deg⁡(Q′)<s\deg(Q') < s; then we deduce that
T=PQ1+P′(PQ′′+Q′)=PQ+P′Q′ T = PQ_1 + P'(PQ'' + Q') = PQ + P'Q'
where Q=Q1+P′Q′′Q = Q_1 + P'Q''. To show the uniqueness of formula (1), it is sufficient to prove that the relations
0=PQ+P′Q′,deg⁡(Q′)<s 0 = PQ + P'Q', \quad \deg(Q') < s
imply Q=Q′=0Q = Q' = 0. Now, if (2) holds, PP divides −PQ=P′Q′-PQ = P'Q' and, as PP and

P' are strongly relatively prime, P divides Q' by Lemma 1; if Q' ≠ 0, there would exist a polynomial S ≠ 0 such that Q' = PS, whence

deg⁡(Q′)=s+deg⁡(S)≥s, \deg(Q') = s + \deg(S) \geq s,

which is a contradiction. We conclude that Q' = 0, whence PQ = 0 and finally Q = 0 by the remark at the beginning.

Finally, suppose that deg⁡(T)≤t\deg(T) \leq t and deg⁡(P′)≤t−s\deg(P') \leq t - s; with the polynomial T in the form (1),

deg⁡(P′Q′)≤deg⁡(P′)+deg⁡(Q′)<s+deg⁡(P′)≤t \deg(P'Q') \leq \deg(P') + \deg(Q') < s + \deg(P') \leq t

and therefore

s+deg⁡(Q)=deg⁡(PQ)=deg⁡(T−P′Q′)≤t s + \deg(Q) = \deg(PQ) = \deg(T - P'Q') \leq t

whence deg⁡(Q)≤t−s\deg(Q) \leq t - s.

Example

For a polynomial P∈A[X]P \in \mathbf{A}[X] to be strongly relatively prime to X−aX - a (where a∈Aa \in \mathbf{A}), it is necessary and sufficient that P(a)P(a) be invertible in A\mathbf{A}. For if P and X−aX - a are strongly relatively prime, it follows from Proposition 1 that there exist c∈Ac \in \mathbf{A} and a polynomial Q∈A[X]Q \in \mathbf{A}[X] such that cP+(X−a)Q=1cP + (X - a)Q = 1, whence cP(a)=1cP(a) = 1 and P(a)P(a) is invertible. Conversely, by Euclidean division

P=(X−a)R+P(a) P = (X - a)R + P(a)

and, if P(a)=b−1P(a) = b^{-1}, where b∈Ab \in \mathbf{A}, we deduce that 1=bP−b(X−a)R1 = bP - b(X - a)R, which shows that P and X−aX - a are strongly relatively prime.

Let A\mathbf{A} and B\mathbf{B} be two commutative rings and f:A→Bf : \mathbf{A} \to \mathbf{B} a ring homomorphism. If P=∑i≥0aiXiP = \sum_{i \geq 0} a_i X^i is a formal power series in A[[X]]\mathbf{A}[[X]], let fˉ(P)\bar{f}(P) denote the formal power series ∑i≥0f(ai)Xi\sum_{i \geq 0} f(a_i) X^i in B[[X]]\mathbf{B}[[X]]. If P is a polynomial, so is fˉ(P)\bar{f}(P) and, if further P is monic, then fˉ(P)\bar{f}(P) is monic of the same degree as P. Finally, P↦fˉ(P)P \mapsto \bar{f}(P) is clearly a homomorphism of A[[X]]\mathbf{A}[[X]] to B[[X]]\mathbf{B}[[X]] which extends ff and maps X to X. *The notation will be constantly used in this sense in the rest of this paragraph.

Proposition 2

Let A\mathbf{A} and B\mathbf{B} be two commutative rings, ff a homomorphism from A\mathbf{A} to B\mathbf{B} and P,P′P, P' two polynomials in A[X]\mathbf{A}[X]. If PP and P′P' are strongly relatively prime in A[X]\mathbf{A}[X], then fˉ(P)\bar{f}(P) and fˉ(P′)\bar{f}(P') are strongly relatively prime in B[X]\mathbf{B}[X]. The converse is true iff is surjective, if its kernel is contained in the Jacobson radical of A\mathbf{A} and if PP is monic.

Suppose that P and P' are strongly relatively prime; then there exist polynomials Q, Q' in A[X]\mathbf{A}[X] such that PQ+P′Q′=1PQ + P'Q' = 1; we deduce that

fˉ(P)fˉ(Q)+fˉ(P′)fˉ(Q′)=1, \bar{f}(P)\bar{f}(Q) + \bar{f}(P')\bar{f}(Q') = 1,

whence the first assertion. To show the second, let a denote the kernel of ff;

let E=A[X]E = A[X] and FF the ideal of EE generated by PP and P′P'; as ff is surjective and fˉ(P)\bar{f}(P) monic, Proposition 1 shows that for every polynomial T∈A[X]T \in A[X] there exist two polynomials Q,Q′Q, Q' in A[X]A[X] such that

fˉ(T)=fˉ(P)fˉ(Q)+fˉ(P′)fˉ(Q′), \bar{f}(T) = \bar{f}(P) \bar{f}(Q) + \bar{f}(P') \bar{f}(Q'),
whence the relation E=F+aEE = F + aE. Now, E/FE/F is a finitely generated AA-module, for every polynomial is congruent mod. PP to a polynomial of degree <deg⁡(P)< \deg(P), PP being monic. As E/F=a(E/F)E/F = a(E/F) and aa is contained in the Jacobson radical of AA, Nakayama's Lemma shows that E/F=0E/F = 0 (Algebra, Chapter VIII, § 6, no. 3, Corollary 2 to Proposition 6 ), which means that PP and P′P' are strongly relatively prime.

2. Restricted formal power series

Definition 1

A commutative topological ring AA is said to be linearly topologized (and its topology is said to be linear) if there exists a fundamental system B\mathscr{B} of neighbourhoods o\mathfrak{o} consisting of ideals o⊂A\mathfrak{o} \subset A.

Note that in such a ring, the ideals 3∈B3 \in \mathscr{B} are open and closed (General Topology, Chapter 111, § 2, no. 1, Corollary to Proposition 4 ). For all 3∈B3 \in \mathscr{B}, the quotient topological ring A/JA/\mathfrak{J} is then discrete; for 8∈B8 \in \mathscr{B}, 8′∈B8' \in \mathscr{B}, J′⊂8\mathfrak{J}' \subset 8, let

hJJ′:A/J′→A/J h_{\mathfrak{J}\mathfrak{J}'} : A/\mathfrak{J}' \to A/\mathfrak{J}
be the canonical mapping. We know (General Topology, Chapter III, § 7, no. 3) that (A/J,hJJ′)(A/\mathfrak{J}, h_{\mathfrak{J}\mathfrak{J}'}) is an inverse system of discrete rings (relative to the indexing set B\mathscr{B} which is ordered by ⊃\supset and directed), whose inverse limit is a complete Hausdorff linearly topologized ring A~\tilde{A}; further (loc. cit., Proposition 2 ), a strict morphism i:A→A~i : A \to \tilde{A} is defined, whose kernel is the closure of {0}\{0\} in AA and whose image is everywhere dense in A~\tilde{A}, so that A~\tilde{A} is canonically identified with the Hausdorff completion of AA.

Definition 2

Given a commutative topological ring AA, a formal power series

T=∑(ni)cn1n2…npX1n1X2n2…Xpnp T = \sum_{(n_i)} c_{n_1 n_2 \ldots n_p} X_1^{n_1} X_2^{n_2} \ldots X_p^{n_p}
in the ring A[[X1,…,Xp]]A[[X_1, \ldots, X_p]] is called restricted if, for every neighbourhood VV of 0 in AA, there is only a finite number of coefficients cn1n2…npc_{n_1 n_2 \ldots n_p} not belonging to VV (in other words, the family (cn1n2…np)(c_{n_1 n_2 \ldots n_p}) tends to 0 in AA with respect to the filter of complements of finite subsets of Np\mathbf{N}^p).

If AA is linearly topologized, the restricted formal power series in A[[X1,…,Xp]]A[[X_1, \ldots, X_p]] form a subring of A[[X1,…,Xp]]A[[X_1, \ldots, X_p]], denoted by A{X1,…,Xp}A\{X_1, \ldots, X_p\}: for if

T=∑(ni)cn1…npX1n1…Xpnp,T′=∑(ni)cn1…np′X1n1…Xpnp T = \sum_{(n_i)} c_{n_1 \ldots n_p} X_1^{n_1} \ldots X_p^{n_p}, \quad T' = \sum_{(n_i)} c'_{n_1 \ldots n_p} X_1^{n_1} \ldots X_p^{n_p}
are two restricted formal power series and J\mathfrak{J} a neighbourhood of 0 in AA which is an ideal of AA, there exists an integer mm such that cn1…np∈zc_{n_1 \ldots n_p} \in \mathfrak{z} and cn1…np′∈gc'_{n_1 \ldots n_p} \in \mathfrak{g} for every system (n1,…,np)(n_1, \ldots, n_p) such that nk≥mn_k \geq m for at least one index kk; now, if

T′′=TT′=∑(ni)cn1…np′′X1n1…Xpnp, T'' = TT' = \sum_{(n_i)} c''_{n_1 \ldots n_p} X_1^{n_1} \ldots X_p^{n_p},

then cn1…np′′=∑cr1…rpcs1…sp′c''_{n_1 \ldots n_p} = \sum c_{r_1 \ldots r_p} c'_{s_1 \ldots s_p} for all systems (rk),(sk)(r_k), (s_k) such that rk+sk=nkr_k + s_k = n_k for 1≤k≤p1 \leq k \leq p; we conclude that if nk≥2mn_k \geq 2m, then rk≥mr_k \geq m or sk≥ms_k \geq m and hence, since z\mathfrak{z} is an ideal, cn1…np′′∈gc''_{n_1 \ldots n_p} \in \mathfrak{g} so long as nk≥2mn_k \geq 2m for at least one kk, which establishes our assertion. Moreover, every derivative ∂T/∂Xi\partial T / \partial X_i (1≤i≤p1 \leq i \leq p) of a restricted formal power series is restricted, as follows immediately from the definition and the fact that the neighbourhoods z∈B\mathfrak{z} \in \mathscr{B} are additive subgroups of A\mathbf{A}.

If A\mathbf{A} is discrete, the ring of restricted formal power series is just the polynomial ring A[X1,…,Xn]\mathbf{A}[X_1, \ldots, X_n].

Let us always assume that A\mathbf{A} is linearly topologized and let B\mathscr{B} be a fundamental system of neighbourhoods of 0 in A\mathbf{A} consisting of ideals of A\mathbf{A}; for all g∈B\mathfrak{g} \in \mathscr{B}, let pg:A→A/gp_\mathfrak{g}: \mathbf{A} \to \mathbf{A}/\mathfrak{g} be the canonical homomorphism. By definition, for every restricted formal power series T∈A{X1,…,Xn}T \in \mathbf{A}\{X_1, \ldots, X_n\},

pˉg(T)∈(A/g)[X1,…,Xp]. \bar{p}_\mathfrak{g}(T) \in (\mathbf{A}/\mathfrak{g})[X_1, \ldots, X_p].

Clearly

((A/g)[X1,…,Xp],hˉgg′) ((\mathbf{A}/\mathfrak{g})[X_1, \ldots, X_p], \bar{h}_{\mathfrak{g}\mathfrak{g}'})

is an inverse system of rings (relative to the directed indexing set B\mathscr{B}) and (pˉg)(\bar{p}_\mathfrak{g}) is an inverse system of homomorphisms A{X1,…,Xn}→(A/g)[X1,…,Xp]\mathbf{A}\{X_1, \ldots, X_n\} \to (\mathbf{A}/\mathfrak{g})[X_1, \ldots, X_p]; as every polynomial is a restricted formal power series, pˉg\bar{p}_\mathfrak{g} is surjective; its kernel NgN_\mathfrak{g} is the ideal of A{X1,…,Xn}\mathbf{A}\{X_1, \ldots, X_n\} consisting of the restricted formal power series all of whose coefficients belong to g\mathfrak{g}; we shall give A{X1,…,Xn}\mathbf{A}\{X_1, \ldots, X_n\} the (linear) topology for which the NgN_\mathfrak{g} (for g∈B\mathfrak{g} \in \mathscr{B}) form a fundamental system of neighbourhoods of 0 (a topology which obviously depends only on that on A\mathbf{A}). Then it follows from General Topology, Chapter III, § 7, no. 3, Proposition 2 that

π=lim⁡←pˉg:A{X1,…,Xn}→lim⁡→(A/g)[X1,…,Xp] \pi = \lim_{\leftarrow} \bar{p}_\mathfrak{g}: \mathbf{A}\{X_1, \ldots, X_n\} \to \lim_{\rightarrow} (\mathbf{A}/\mathfrak{g})[X_1, \ldots, X_p]

is a strict morphism whose kernel is the closure of {0}\{0\} in A{X1,…,Xn}\mathbf{A}\{X_1, \ldots, X_n\} and whose image is dense in

A′=lim⁡→(A/g)[X1,…,Xn]. \mathbf{A}' = \lim_{\rightarrow} (\mathbf{A}/\mathfrak{g})[X_1, \ldots, X_n].

Proposition 3

If the linearly topologized commutative ring A\mathbf{A} is Hausdorff and complete, the canonical homomorphism π\pi is a topological ring isomorphism.

For all (n1,…,n)∈Np(n_1, \ldots, n) \in \mathbf{N}^p and all 3∈B3 \in \mathscr{B}, let ϕn1…np3\phi_{n_1 \ldots n_p}^3 be the mapping (A/J)[X1,…,Xp]→A/J(A/\mathfrak{J})[X_1, \ldots, X_p] \to A/\mathfrak{J} which maps every polynomial to the coefficient of X1n1…XpnpX_1^{n_1} \ldots X_p^{n_p} in this polynomial; clearly the ϕn1…np3\phi_{n_1 \ldots n_p}^3 form an inverse system of (A/J)(A/\mathfrak{J})-module homomorphisms (relative to the ordered set B\mathscr{B}) and, as AA is canonically identified with lim⁡←(A/J)\lim_{\leftarrow} (A/\mathfrak{J}) by hypothesis, ϕn1…np=lim⁡←ϕn1…np3\phi_{n_1 \ldots n_p} = \lim_{\leftarrow} \phi_{n_1 \ldots n_p}^3 is a continuous AA-homomorphism from A′A' to AA. For every element S=(SJ)J∈BS = (S_\mathfrak{J})_{\mathfrak{J} \in \mathscr{B}} of A′A', we shall see that the formal power series T=∑(ni)ϕn1…np(S)X1n1…XpnpT = \sum_{(n_i)} \phi_{n_1 \ldots n_p}(S) X_1^{n_1} \ldots X_p^{n_p} is restricted and satisfies π(T)=S\pi(T) = S. For all 3∈B3 \in \mathscr{B} and all J′∈B\mathfrak{J}' \in \mathscr{B} such that J′⊂J\mathfrak{J}' \subset \mathfrak{J}, the relation ϕn1…npJ′(SJ′)=0\phi_{n_1 \ldots n_p}^{\mathfrak{J}'}(S_{\mathfrak{J}'}) = 0 implies

ϕn1…npJ′(SJ′)∈J/J′ \phi_{n_1 \ldots n_p}^{\mathfrak{J}'}(S_{\mathfrak{J}'}) \in \mathfrak{J}/\mathfrak{J}'
as SJS_{\mathfrak{J}} is a polynomial, we see that ϕn1…np(S)∈J\phi_{n_1 \ldots n_p}(S) \in \mathfrak{J} except for those (n1,…,np)(n_1, \ldots, n_p), finite in number, such that ϕn1…npJ(SJ)≠0\phi_{n_1 \ldots n_p}^{\mathfrak{J}}(S_{\mathfrak{J}}) \neq 0, which proves our first assertion; the second follows from the definitions. As AA is Hausdorff, the intersection of the NJ\mathbf{N}_\mathfrak{J} reduces to 0 and hence π\pi is bijective, which completes the proof, since π\pi is a strict morphism.

Proposition 4

Let A,BA, B be two linearly topologized commutative rings, BB being Hausdorff and complete, and u:A→Bu : A \to B a continuous homomorphism. For every family b=(bi)1≤i≤pb = (b_i)_{1 \leq i \leq p} of elements of BB, there exists a unique continuous homomorphism*

u~:A{X1,…,Xp}→B \tilde{u} : A\{X_1, \ldots, X_p\} \to B
*such that u~(a)=u(a)\tilde{u}(a) = u(a) for all a∈Aa \in A and u~(Xi)=bi\tilde{u}(X_i) = b_i for 1≤i≤p1 \leq i \leq p.

There exists a unique homomorphism v:A[X1,…,Xp]→Bv : A[X_1, \ldots, X_p] \to B such that v(a)=u(a)v(a) = u(a) for a∈Aa \in A and v(Xi)=biv(X_i) = b_i for 1≤i≤p1 \leq i \leq p. Moreover, if J\mathfrak{J} is a neighbourhood of 0 in BB which is an ideal, u‾−1(J)=3\overline{u}^{-1}(\mathfrak{J}) = 3 is an ideal of AA which is a neighbourhood of 0 and, for every polynomial P∈NJP \in \mathbf{N}_\mathfrak{J}, clearly v(P)∈Jv(P) \in \mathfrak{J} and hence vv is continuous. As A[X1,…,Xp]A[X_1, \ldots, X_p] is dense in A{X1,…,Xp}A\{X_1, \ldots, X_p\}, the existence and uniqueness of u~\tilde{u} follow from General Topology, Chapter 111,§ 3, no. 3, Proposition 5 and the principle of extension of identities.

In the special case where A=BA = B and uu is the identity mapping we shall write f(b1,…,bp)f(b_1, \ldots, b_p) or f(b)f(b) for the value of u~(f)\tilde{u}(f) for every restricted formal power series f∈A{X1,…,Xp}f \in A\{X_1, \ldots, X_p\}.

Remarks

Remark 1

Proposition 4 proves that for every closed ideal aa in a ring AA which is assumed to be Hausdorff and complete, the relations bi∈ab_i \in a for 1≤i≤p1 \leq i \leq p imply f(b1,…,bp)∈af(b_1, \ldots, b_p) \in a for every restricted formal power series f∈A{X1,…,Xp}f \in A\{X_1, \ldots, X_p\}.

Remark 2

Suppose that AA is linearly topologized; let rr be an integer such that 1 ≤r≤p\leq r \leq p and let the ring A{X1,…,Xr}A\{X_1, \ldots, X_r\} be given the topology defined above. Then the topological ring A{X1,…,Xr}A\{X_1, \ldots, X_r\} is identified with the ring of restricted formal power series

(A{X1,…,Xr}){Xr+1,…,Xp} (A\{X_1, \ldots, X_r\})\{X_{r+1}, \ldots, X_p\}

as follows immediately from the definitions.

Remark 3

With the notation of Remark 2, suppose further that AA is Hausdorff and complete and let us write every restricted formal power series f∈A{X1,…,Xr}f \in A\{X_1, \ldots, X_r\} in the form

f=∑(ni)cnr+1…np(X1,…,Xr)Xr+1nr+1…Xpnp f = \sum_{(n_i)} c_{n_{r+1} \ldots n_p}(X_1, \ldots, X_r) X_{r+1}^{n_{r+1}} \ldots X_p^{n_p}

where the cnr+1…npc_{n_{r+1} \ldots n_p} are restricted formal power series. For every system x=(x1,…,xr)x = (x_1, \ldots, x_r) of elements of AA, let

bnr+1…np=cnr+1…np(x1,…,xp). b_{n_{r+1} \ldots n_p} = c_{n_{r+1} \ldots n_p}(x_1, \ldots, x_p).

It follows immediately from Remark 1 that ∑(ni)bnr+1…npXr+1nr+1…Xpnp\sum_{(n_i)} b_{n_{r+1} \ldots n_p} X_{r+1}^{n_{r+1}} \ldots X_p^{n_p} is a restricted formal power series denoted by f(x1,…,xr,Xr+1,…,Xp)f(x_1, \ldots, x_r, X_{r+1}, \ldots, X_p); it is said to be obtained by substituting the xix_i for the XiX_i for 1≤i≤r1 \leq i \leq r in ff.

3. Hensel's lemma

In a topological ring AA, an element xx is called topologically nilpotent if OO is a limit of the sequence (xn)n≥0(x^n)_{n \geq 0}. If AA is a linearly topologized commutative ring, to say that xx is topologically nilpotent means that for every open ideal J\mathfrak{J} of AA the canonical image of xx in A/JA/\mathfrak{J} is a nilpotent element of that ring. If rg\mathfrak{r}_g is the nil-radical of A/JA/\mathfrak{J}, clearly (rg)(\mathfrak{r}_g) is an inverse system of subsets and the set tt of topological nilpotent elements of AA is the inverse image of r=lim⁡rg\mathfrak{r} = \lim \mathfrak{r}_g under the canonical homomorphism A→lim⁡A/JA \to \lim A/\mathfrak{J}; it is therefore a closed ideal of AA. If also AA is Hausdorff and complete, this ideal is contained in the Jacobson radical of AA and, for an element x∈Ax \in A to be invertible, it is necessary and sufficient that its class mod. tt be invertible in A/tA/t (§ 2, no. 13, Lemma 3).

Note that if AA is a ring and mm a two-sided ideal of AA, the elements of mm are topologically nilpotent with respect to the m-adic topology.

Theorem 1 (Hensel). Let AA be a complete Hausdorff linearly topologized commutative ring. Let mm be a closed ideal of AA whose elements are topologically nilpotent. Let B=A/mB = A/m be the quotient topological ring and φ:A→B\varphi : A \to B the canonical mapping. Let RR be a restricted formal power series in A{X}A\{X\}, P‾\overline{P} a monic polynomial in B[X]B[X] and Q‾\overline{Q} a restricted formal power series in B{X}B\{X\}. Suppose that φ‾(R)=P‾⋅Q‾\overline{\varphi}(R) = \overline{P} \cdot \overline{Q} and that P‾\overline{P} and Q‾\overline{Q} are strongly relatively prime in B{X}B\{X\}. Then there exists a unique ordered pair (P,Q)(P, Q) consisting of a monic polynomial P∈A[X]P \in A[X] and a restricted formal power series Q∈A{X}Q \in A\{X\} such that (4)

R=P.Q,φˉ(P)=P,φˉ(Q)=Q‾. R = P.Q, \quad \bar{\varphi}(P) = P, \quad \bar{\varphi}(Q) = \overline{Q}.
Moreover, PP and QQ are strongly relatively prime in A{X}A\{X\} and, if RR is a polynomial, so is QQ.

The proof is divided into four steps. In the first three we assume that AA is discrete, in which case RR and Q‾\overline{Q} are polynomials.

(I) m2=0m^2 = 0 Let S,TS, T be two polynomials of A[X]A[X] such that SS is monic and φˉ(S)=P‾\bar{\varphi}(S) = \overline{P}, φˉ(T)=Q‾\bar{\varphi}(T) = \overline{Q}; Proposition 2 of no. 1 shows that SS and TT are strongly relatively prime; hence (no. 1, Proposition 1) there exists a unique ordered pair of polynomials (S′,T′)(S', T') of A[X]A[X] such that (5)

R−ST=ST′+TS′anddeg⁡(S′)<deg⁡(S)=deg⁡(P‾). R - ST = ST' + TS' \quad \text{and} \quad \deg(S') < \deg(S) = \deg(\overline{P}).
The polynomials P=S+TS′P = S + TS', Q=T+T′Q = T + T' are then solutions to the problem; then in fact (6)
P‾⋅φˉ(T′)+Q‾⋅φˉ(S′)=φˉ(ST′+TS′)=φˉ(R−ST)=0. \overline{P} \cdot \bar{\varphi}(T') + \overline{Q} \cdot \bar{\varphi}(S') = \bar{\varphi}(ST' + TS') = \bar{\varphi}(R - ST) = 0.
As P‾\overline{P} is monic, P‾\overline{P} and Q‾\overline{Q} strongly relatively prime and
deg⁡(φˉ(S′))<deg⁡(P‾), \deg(\bar{\varphi}(S')) < \deg(\overline{P}),
Proposition 1 of no. 1 shows that φˉ(S′)=φˉ(T′)=0\bar{\varphi}(S') = \bar{\varphi}(T') = 0, in other words the coefficients of S′S' and T′T' belong to mm and the relation m2=0m^2 = 0 gives
PQ=ST+ST′+TS′=R, PQ = ST + ST' + TS' = R,
which satisfies relation (4). Since φˉ(P)=P‾\bar{\varphi}(P) = \overline{P} and φˉ(Q)=Q‾\bar{\varphi}(Q) = \overline{Q}, PP and QQ are strongly relatively prime (no. 1, Proposition 2); finally, if P1P_1 and Q1Q_1 are two other polynomials in A[X]A[X] satisfying (4) and such that P1P_1 is monic, then necessarily, setting S1′=P1−S, T1′=Q1−T, deg⁡(S1′)<deg⁡(S)S'_1 = P_1 - S, \ T'_1 = Q_1 - T, \ \deg(S'_1) < \deg(S) and R−ST=ST1′+TS1′R - ST = ST'_1 + TS'_1 since S1′S'_1 and T1′T'_1 have their coefficients in mm; but Proposition 1 then proves that S′=S1′S' = S'_1 and T′=T1′T' = T'_1, which proves the uniqueness of the ordered pair (P,Q)(P, Q).

(2) mm is nilpotent Let nn be the smallest integer such that mn=0m^n = 0 and let us argue by induction on n>2n > 2, the theorem having been shown for n=2n = 2. Let A=A/mn−1A = A/m^{n-1}, m′=m/mn−1m' = m/m^{n-1}; as m′n−1=0{m'}^{n-1} = 0, there exists a unique ordered pair (P′,Q′)(P', Q') of polynomials in A′[X]A'[X] such that P′P' is monic, R′=P′Q′R' = P'Q', ψ(P′)=P‾\psi(P') = \overline{P} and ψ(Q′)=Q‾\psi(Q') = \overline{Q}, where ψ\psi denotes the canonical homomorphism A′→A′/m′=BA' \to A'/m' = B, θ\theta the canonical homomorphism A→A′A \to A' and R′=θ~(R)R' = \tilde{\theta}(R). On the other hand, as (mn−1)2=0(m^{n-1})^2 = 0, there exists a unique ordered pair (P,Q)(P, Q) of polynomials in A[X]A[X] such that PP is monic and R=PQR = PQ, θˉ(P)=P‾\bar{\theta}(P) = \overline{P}, θˉ(Q)=Q‾\bar{\theta}(Q) = \overline{Q}; as ϕ=ψ∘θ\phi = \psi \circ \theta, this shows the existence and uniqueness of PP and QQ satisfying (4); moreover P′P' and Q′Q' are strongly relatively prime by the induction hypothesis and hence so are PP and QQ.

(3) A is discrete

Note that in this case mm is no longer necessarily nilpotent, but it is always a nilideal by hypothesis. Let P0P_0, Q0Q_0 be two polynomials of A[X]A[X] such that φˉ(P0)=P‾\bar{\varphi}(P_0) = \overline{P}, φˉ(Q0)=Q‾\bar{\varphi}(Q_0) = \overline{Q} and P0P_0 is monic. Let us consider the ideal nn of AA generated by the coefficients of R−P0Q0R - P_0 Q_0; it is finitely generated and contained in mm, hence it is nilpotent (Chapter 11, § 2, no. 6, Proposition 15) and by definition, if ψ:A→A/n\psi : A \to A/n is the canonical mapping, then ψˉ(R)=ψˉ(P0)ψˉ(Q0)\bar{\psi}(R) = \bar{\psi}(P_0) \bar{\psi}(Q_0). Moreover, ψˉ(P0)\bar{\psi}(P_0) and ψˉ(Q0)\bar{\psi}(Q_0) are strongly relatively prime, as follows from the hypothesis on P‾\overline{P} and Q‾\overline{Q} and Proposition 2 of no. 1 applied to the canonical homomorphism A/n→A/mA/n \to A/m. By virtue of case (2), there therefore exists an ordered pair (P,Q)(P, Q) of polynomials in A[X]A[X] such that PP is monic and relations (4) hold. The fact that P‾\overline{P} and Q‾\overline{Q} are strongly relatively prime implies also here that PP and QQ are strongly relatively prime in A[X]A[X] by virtue of no. 1, Proposition 2, for mm is contained in the Jacobson radical of AA. Suppose finally that P1P_1, Q1Q_1 are two polynomials in A[X]A[X] satisfying (4) and such that P1P_1 is monic and let n1n_1 be the finitely generated ideal of AA generated by the coefficients of P−P1P - P_1 and the coefficientsof Q−Q1Q - Q_1; as n1n_1 is contained in mm, it is nilpotent and, if ψ1:A→A/n1\psi_1 : A \to A/n_1 is the canonical mapping, ψˉ1(P)=ψˉ1(P1)\bar{\psi}_1(P) = \bar{\psi}_1(P_1) and

ψˉ1(Q)=ψˉ1(Q1); \bar{\psi}_1(Q) = \bar{\psi}_1(Q_1);
the uniqueness property for case (2) therefore implies P=P1, Q=Q1P = P_1,\ Q = Q_1.

(4) General case

Let B\mathscr{B} be a fundamental system of neighbourhoods of 0 in AA consisting of ideals of AA. For all s∈B\mathfrak{s} \in \mathscr{B}, let fsf_{\mathfrak{s}} be the canonical mapping A→A/sA \to A/\mathfrak{s}, ϕs\phi_{\mathfrak{s}} the canonical mapping

A/s↦(A/s)/((m+s)/s)=A/(m+s), A/\mathfrak{s} \mapsto (A/\mathfrak{s})/((m + \mathfrak{s})/\mathfrak{s}) = A/(m + \mathfrak{s}),
gsg_{\mathfrak{s}} the canonical mapping B=A/m→A(m+s)B = A/m \to A(m + \mathfrak{s}) and write Rs=fs(R)R_{\mathfrak{s}} = f_{\mathfrak{s}}(R), Ps=gs(P‾)P_{\mathfrak{s}} = g_{\mathfrak{s}}(\overline{P}), Q‾s=gs(Q‾)\overline{Q}_{\mathfrak{s}} = g_{\mathfrak{s}}(\overline{Q}). As each ring A/sA/\mathfrak{s} is discrete, case (3) can be applied to it and we see that there exists a unique ordered pair (Ps,Qs)(P_{\mathfrak{s}}, Q_{\mathfrak{s}}) of polynomials in (A/s)[X](A/\mathfrak{s})[X] such that PsP_{\mathfrak{s}} is monic and Rs=PsQsR_{\mathfrak{s}} = P_{\mathfrak{s}} Q_{\mathfrak{s}}, φˉs(Ps)=P‾s\bar{\varphi}_{\mathfrak{s}}(P_{\mathfrak{s}}) = \overline{P}_{\mathfrak{s}}, φˉs(Qs)=Q‾s\bar{\varphi}_{\mathfrak{s}}(Q_{\mathfrak{s}}) = \overline{Q}_{\mathfrak{s}}. The uniqueness of this ordered pair implies that, if s′⊂s\mathfrak{s}' \subset \mathfrak{s}, s′∈B\mathfrak{s}' \in \mathscr{B} and fs′s:A/s′→A/sf_{\mathfrak{s}'\mathfrak{s}} : A/\mathfrak{s}' \to A/\mathfrak{s} is the canonical mapping, then Ps=fs′s(Ps′)P_{\mathfrak{s}} = f_{\mathfrak{s}'\mathfrak{s}}(P_{\mathfrak{s}'}), Qs=fs′s(Qs′)Q_{\mathfrak{s}} = f_{\mathfrak{s}'\mathfrak{s}}(Q_{\mathfrak{s}'}). Then it follows from the canonical identification of A[X]A[X] with lim⁡(A/s)[X]\lim(A/\mathfrak{s})[X] (no. 2, Proposition 3) that there exists P∈A{X}P \in A\{X\} and Q∈A{X}Q \in A\{X\} such that R=PQR = PQ and fˉs(P)=Ps,fˉs(Q)=Qs\bar{f}_s(P) = P_s, \bar{f}_s(Q) = Q_s for all s∈Bs \in \mathscr{B}. Moreover,
gˉs(P‾−φ‾(P))=0,gˉs(Q‾−φ‾(Q))=0 \bar{g}_s(\overline{P} - \overline{\varphi}(P)) = 0, \quad \bar{g}_s(\overline{Q} - \overline{\varphi}(Q)) = 0
for all s∈Bs \in \mathscr{B}, which means that for all J∈B\mathfrak{J} \in \mathscr{B} the coefficients of P‾−φ‾(P)\overline{P} - \overline{\varphi}(P) and Q‾−φ‾(Q)\overline{Q} - \overline{\varphi}(Q) all belong to (m+J)/m(m + \mathfrak{J})/m. But, as mm is closed in AA, ⋂s(m+s)=m\bigcap_s (m + s) = m, whence P‾=φ‾(P),Q‾=φ‾(Q)\overline{P} = \overline{\varphi}(P), \overline{Q} = \overline{\varphi}(Q) and PP and QQ then certainly satisfy (4); moreover, as the PsP_s are monic and of the same degree, the restricted formal power series PP is a monic polynomial. If (P′,Q′)(P', Q') were another ordered pair satisfying (4) and such that P′P' is a monic polynomial, we would deduce that Rs=fˉs(P′)fˉs(Q′),φˉs(fˉs(P′))=P‾sR_s = \bar{f}_s(P') \bar{f}_s(Q'), \bar{\varphi}_s(\bar{f}_s(P')) = \overline{P}_s and φˉs(fˉs(Q′))=Q‾s\bar{\varphi}_s(\bar{f}_s(Q')) = \overline{Q}_s and by the uniqueness in case (3) fˉs(P′)=Ps,fˉs(Q′)=Qs\bar{f}_s(P') = P_s, \bar{f}_s(Q') = Q_s for all s∈Bs \in \mathscr{B}, which implies that P=P′P = P' and Q=Q′Q = Q'. Let us show finally that PP and QQ are strongly relatively prime; by virtue of case (3) and Proposition 1 of no. 1, for all s∈Bs \in \mathscr{B}, there exists a unique ordered pair (Ss,Ts)(S_s, T_s) of polynomials in (A/J)[X](A/\mathfrak{J})[X] such that
1=PsSs+QsTsanddeg⁡(Ts)<deg⁡(Ps)=deg⁡(P‾). 1 = P_s S_s + Q_s T_s \quad \text{and} \quad \deg(T_s) < \deg(P_s) = \deg(\overline{P}).
The uniqueness of this ordered pair shows immediately that, for 3′∈B3' \in \mathscr{B}, J′⊂J,S3′=fˉ33′(S3),T3′=fˉ33′(T3);\mathfrak{J}' \subset \mathfrak{J}, S_{3'} = \bar{f}_{33'}(S_{3}), T_{3'} = \bar{f}_{33'}(T_{3}); taking account of no. 2, Proposition 3, we conclude that there exist two restricted formal power series S,TS, T of A{X}A\{X\} such that Ss=fˉs(S),Ts=fˉs(T)S_s = \bar{f}_s(S), T_s = \bar{f}_s(T) and 1=PS+QT1 = PS + QT.

It remains to verify that, if RR is a polynomial, so is QQ. Now, the QsQ_s are polynomials by construction and, as PsP_s is monic, the relation Rs=PsQsR_s = P_s Q_s implies

deg⁡(Qs)≤deg⁡(Rs)≤deg⁡(R) \deg(Q_s) \leq \deg(R_s) \leq \deg(R)
for all s∈Bs \in \mathscr{B}; whence immediately the required result by definition of QQ.

4. Composition of systems of formal power series

Let AA be a commutative ring; we shall say that a system

f=(f1,…,fp)∈(A[[X1,…,Xq]])p \mathbf{f} = (f_1, \ldots, f_p) \in (A[[X_1, \ldots, X_q]])^p
of formal power series in the XjX_j (1≤j≤q1 \leq j \leq q), with coefficients in AA, is without constant term if this is true of all the fjf_j. For every system (8) of formal power series and every system
g=(g1,…,gq)∈(A[[X1,…,Xr]])q \mathbf{g} = (g_1, \ldots, g_q) \in (A[[X_1, \ldots, X_r]])^q
of qq formal power series without constant term, we shall denote by f∘g\mathbf{f} \circ \mathbf{g} (or f(g)\mathbf{f}(\mathbf{g})) the system of formal power series fj(g1,…,gq)f_j(g_1, \ldots, g_q) (1≤j≤p1 \leq j \leq p) in
(A[[X1,…,Xr]])p (A[[X_1, \ldots, X_r]])^p

(Algebra, Chapter IV, § 5, no. 5). If

h=(h1,…,hr)∈(A[[X1,…,Xs]])r \mathbf{h} = (h_1, \ldots, h_r) \in (\mathbf{A}[[X_1, \ldots, X_s]])^r
is a third system without constant term, then
(10)(10)
(f∘g)∘h=f∘(g∘h).(f \circ g) \circ h = f \circ (g \circ h).

For every integer mm,

(f(m)∘g(m))∘h(m)=f(m)∘(g(m)∘h(m))(f^{(m)} \circ g^{(m)}) \circ h^{(m)} = f^{(m)} \circ (g^{(m)} \circ h^{(m)})
where f(m),g(m),h(m)f^{(m)}, g^{(m)}, h^{(m)} denote the systems of polynomials consisting of terms of total degree ≤m\leq m in the systems of formal power series f,g,hf, g, h. But clearly the terms of total degree ≤m\leq m in the series of (f∘g)∘h(f \circ g) \circ h (resp. f∘(g∘h)f \circ (g \circ h)) are the same as in (f(m)∘g(m))∘h(m)(f^{(m)} \circ g^{(m)}) \circ h^{(m)} (resp. f(m)∘(g(m)∘h(m))f^{(m)} \circ (g^{(m)} \circ h^{(m)})), whence our assertion.

For every system (8), we shall denote by MfM_f or Mf(X)M_f(\mathbf{X}), the Jacobian matrix (∂fi/∂Xj)(\partial f_i / \partial X_j) (1≤i≤p,1≤j≤q1 \leq i \leq p, 1 \leq j \leq q) where ii is the index of the rows and jj that of the columns; for two systems (8) and (9), where gg is without constant term,

(11)(11)
Mf∘g=(Mf(g)).MgM_{f \circ g} = (M_f(g)). M_g
where Mf(g)M_f(g) is the matrix whose elements are obtained by substituting gg, for X\mathbf{X}, (1≤j≤q1 \leq j \leq q) in each series element of MfM_f; this formula is just a reformulation of formula (9) of Algebra, Chapter IV, § 5, no. 8. We shall denote by Mf(0)M_f(0) the matrix of constant terms of the elements of MfM_f; then we deduce from (11) that
(12)(12)
Mf∘g(0)=Mf(0).Mg(0).M_{f \circ g}(0) = M_f(0). M_g(0).

Given an integer n>0n > 0, we shall write

(13)(13)
1n=X=(X1,…,Xn)∈(A[[X1,…,Xn]])n,\mathbf{1}_n = \mathbf{X} = (X_1, \ldots, X_n) \in (\mathbf{A}[[X_1, \ldots, X_n]])^n,
which will be considered as a matrix with a single column.

For every system f=(f1,…,fn)∈(A[[X1,…,Xn]])n\mathbf{f} = (f_1, \ldots, f_n) \in (\mathbf{A}[[X_1, \ldots, X_n]])^n, MfM_f is a square matrix of order nn; we shall denote by JfJ_f or Jf(X)J_f(\mathbf{X}) its determinant and by Jf(0)J_f(0) the constant term of JfJ_f, equal to det⁡(Mf(0))\det(M_f(0)); if g=(g1,…,gn)\mathbf{g} = (g_1, \ldots, g_n) is a system without constant term in (A[[X1,…,Xn]])n(\mathbf{A}[[X_1, \ldots, X_n]])^n, then, by (11) and (12),

(14)(14)
Jf∘g=Jf(g)⋅JgJ_{f \circ g} = J_f(g) \cdot J_g
(15)(15)
Jf∘g(0)=Jf(g)Jg(0).J_{f \circ g}(0) = J_f(g) J_g(0).

Proposition 5

Let A\mathbf{A} be a commutative ring and f=(f1,…,fn)\mathbf{f} = (f_1, \ldots, f_n) a system without constant term of nn series in A[[X1,…,Xn]]\mathbf{A}[[X_1, \ldots, X_n]]. Suppose that Jf(0)J_f(0) is invertible in A\mathbf{A}. Then there exists a system without constant term g=(g1,…,gn)\mathbf{g} = (g_1, \ldots, g_n) of nn series in A[[X1,…,Xn]]\mathbf{A}[[X_1, \ldots, X_n]] such that*

(16)(16)
f∘g=1n.f \circ g = \mathbf{1}_n.

This system is unique and

(17) g∘f=11g \circ f = 1_1.

The existence and uniqueness of gg follows from Algebra, Chapter IV, § 5, no. 9, Proposition 10, applied to the nn formal power series

fi(Y1,…,Yn)−Xi(1≤i≤n). f_i(Y_1, \ldots, Y_n) - X_i \quad (1 \leq i \leq n).

It follows from (15) and (16) that Jf(0)Jg(0)=1J_f(0)J_g(0) = 1 and hence Jg(0)J_g(0) is also invertible. We conclude that there exists a system h=(h1,…,hn)h = (h_1, \ldots, h_n) of nn series without constant term in A[[X1,…,Xn]]A[[X_1, \ldots, X_n]] such that g∘h=11g \circ h = 1_1; from this relation and (16) it then follows, with the aid of (10), that

h=1n∘h=(f∘g)∘h=f∘(g∘h)=f∘1n=f. h = 1_n \circ h = (f \circ g) \circ h = f \circ (g \circ h) = f \circ 1_n = f.

Proposition 5 and formulae (10) and (15) show that under the law of composition (f,g)↦f∘g(f, g) \mapsto f \circ g the set of systems f=(f1,…,fn)f = (f_1, \ldots, f_n) of nn series without constant term in A[[X1,…,Xn]]A[[X_1, \ldots, X_n]] for which Jf(0)J_f(0) is invertible in AA, is a group.

5. Systems of equations in complete rings

To abbreviate, we shall say in what follows that a ring satisfies Hensel’s conditions if it is commutative, linearly topologized, Hausdorff and complete; given an ideal mm in such a ring, mm (or the ordered pair (A,m)(A, m)) will be said to satisfy Hensel’s conditions if mm is closed in AA and its elements are topologically nilpotent. The ideal tt of AA consisting of all the topologically nilpotent elements satisfies Hensel’s conditions (no. 3).

In particular, if AA is a commutative ring and mm an ideal of AA and AA is Hausdorff and complete with respect to the m-adic topology, the ordered pair (A,m)(A, m) satisfies Hensel’s conditions.

Proposition 6

Let AA be a commutative ring, BB a ring satisfying Hensel’s conditions and u:A→Bu : A \to B a homomorphism. For every family x=(x1,…,xn)x = (x_1, \ldots, x_n) of topologically nilpotent elements of BB, there exists a unique homomorphism u~\tilde{u} from A[[X1,…,Xn]]A[[X_1, \ldots, X_n]] to BB such that u~(a)=u(a)\tilde{u}(a) = u(a) for all a∈Aa \in A and u~(Xi)=xi\tilde{u}(X_i) = x_i for 1≤i≤n1 \leq i \leq n. Moreover, if mm denotes the ideal of series without constant term in A[[X1,…,Xn]]A[[X_1, \ldots, X_n]], u~\tilde{u} is continuous for the m-adic topology.

Let aa be the finitely generated ideal generated in BB by the xix_i (1≤i≤n1 \leq i \leq n); for every open ideal H\mathfrak{H} of BB, the images of the xix_i in B/HB/\mathfrak{H} are nilpotent, hence the ideal (a+H)/H(a + \mathfrak{H})/\mathfrak{H} is nilpotent in B/HB/\mathfrak{H} and there exists an integer kk such that, for ∑i=1npi≥k\sum_{i=1}^n p_i \geq k, x1p1⋯xnpn∈Hx_1^{p_1} \cdots x_n^{p_n} \in \mathfrak{H}. As every element of mkm^k is a finite sum of formal power series of the form X1p1⋯Xnpng(X1,…,Xn)X_1^{p_1} \cdots X_n^{p_n} g(X_1, \ldots, X_n), where ∑i=1npi≥k\sum_{i=1}^n p_i \geq k, it is seen that, if u~\tilde{u} solves the problem, then u~(mk)⊂H\tilde{u}(m^k) \subset \mathfrak{H}, which proves the continuity of u~\tilde{u}. There obviously exists a unique homomorphism

ν:A[X1,…,Xn]→B \nu : A[X_1, \ldots, X_n] \to B

such that ν(a)=u(a)\nu(a) = u(a) for a∈Aa \in A and ν(Xi)=xi\nu(X_i) = x_i for 1≤i≤n1 \leq i \leq n and the above argument shows that ν\nu is continuous with respect to the topology induced on A[X1,…,Xn]A[X_1, \ldots, X_n] by the m-adic topology. As A[X1,…,Xr]A[X_1, \ldots, X_r] is dense in A[[X1,…,Xr]]A[[X_1, \ldots, X_r]] with the m-adic topology and B is Hausdorff and complete, this completes the proof of the existence and uniqueness of u~\tilde{u}.

Note that this proposition gives us again as a special case (i) of Proposition 11 of § 2, no. 9.

If AA itself is linearly topologized, the restriction of u~\tilde{u} to A{X1,…,Xr}A\{X_1, \ldots, X_r\} coincides with the homomorphism derived from uu in Proposition 4 of no. 2. This follows immediately from the fact that A[X1,…,Xr]A[X_1, \ldots, X_r] is dense in A{X1,…,Xr}A\{X_1, \ldots, X_r\} if this ring is given the topology with fundamental system of neighbourhoods of 0 the ideals mk∩N3m^k \cap N_3 (in the notation of no. 2, this topology is the least upper bound of the topology induced on A{X1,…,Xr}A\{X_1, \ldots, X_r\} by the m-adic topology on A[[X1,…,Xn]]A[[X_1, \ldots, X_n]] and the topology defined in no. 2).

If B=AB = A and uu is the identity mapping, we shall write f(x1,…,xr)f(x_1, \ldots, x_r) or f(x)f(x) for the element u~(f)\tilde{u}(f) for every formal power series f∈A[[X1,…,Xr]]f \in A[[X_1, \ldots, X_r]]; for every system f=(f1,…,fr)\mathbf{f} = (f_1, \ldots, f_r) of formal power series of A[[X1,…,Xr]]A[[X_1, \ldots, X_r]], let f(x)\mathbf{f}(x) denote the element (f1(x),…,fr(x))(f_1(x), \ldots, f_r(x)) of ArA^r, then it is said to be obtained by substituting the xix_i for the XiX_i in f\mathbf{f}. If n≤mn \leq m and F is a formal power series of A[[X1,…,Xm]]A[[X_1, \ldots, X_m]], it is possible to consider F as a formal power series in Xn+1,…,XmX_{n+1}, \ldots, X_m with coefficients in A[[X1,…,Xn]]A[[X_1, \ldots, X_n]]; let

F(x1,…,xn,Xn+1,…,Xm) F(x_1, \ldots, x_n, X_{n+1}, \ldots, X_m)

denote the formal power series in A[[Xn+1,…,Xm]]A[[X_{n+1}, \ldots, X_m]] obtained by substituting the xix_i for the XiX_i in the coefficients of F, for 1≤i≤n1 \leq i \leq n.

Let us take BB to be the ring of formal power series A[[X1,…,Xr]]A[[X_1, \ldots, X_r]] and let nn be the ideal of series in BB without constant term, so that (B,n)(B, n) satisfies Hensel's conditions (§2§ 2, no. 6, Corollary to Proposition 6). Proposition 6 may be applied by taking the xi∈Bx_i \in B to be series without constant term; then, for every series f∈A[[X1,…,Xr]]\mathbf{f} \in A[[X_1, \ldots, X_r]], u~(f)\tilde{u}(f) is just the formal power series f(x1,…,xn)f(x_1, \ldots, x_n) defined in Algebra, Chapter IV, §5§ 5, no. 5. This is obvious if ff is a polynomial and it follows from the proposition in the general case by remarking that f↦f(x1,…,xn)f \mapsto f(x_1, \ldots, x_n) is continuous on A[[X1,…,Xn]]A[[X_1, \ldots, X_n]] with respect to the m-adic topology.

Corollary

Let AA be a ring satisfying Hensel's condition and x=(x1,…,xn)x = (x_1, \ldots, x_n) a family of topologically nilpotent elements of AA. Let g=(g1,…,gq)g = (g_1, \ldots, g_q) be a system without constant term of series in A[[X1,…,Xs]]A[[X_1, \ldots, X_s]] and f=(f1,…,fp)f = (f_1, \ldots, f_p) a system of formal power series in A[[X1,…,Xq]]A[[X_1, \ldots, X_q]]. Then g(x)=(g1(x),…,gq(x))g(x) = (g_1(x), \ldots, g_q(x)) is a family of topologically nilpotent elements of AA and

(f∘g)(x)=f(g(x)). (f \circ g)(x) = f(g(x)).

The fact that the gi(x)g_i(x) are topologically nilpotent follows immediately from Proposition 6 and the fact that in AA the ideal of topologically nilpotent elements is closed. Relation (18) is obvious when the fjf_j are polynomials; on the other hand, if mm and m′m' are the ideals of series without constant term in

A[[X1,…,Xq]]andA[[X1,…,Xn]] A[[X_1, \ldots, X_q]] \quad \text{and} \quad A[[X_1, \ldots, X_n]]

respectively, clearly the relation f∈m′kf \in {m'}^k implies f(g1,…,gq)∈m′kf(g_1, \ldots, g_q) \in {m'}^k. The two sides of (18) are therefore continuous functions of ff to (A[[X1,…,Xq]])p(A[[X_1, \ldots, X_q]])^p if A[[X1,…,Xq]]A[[X_1, \ldots, X_q]] is given the mm-adic topology, by virtue of the above remark and Proposition 6; whence relation (18).

In what follows, for a ring AA and an ideal mm of AA we shall denote by mxnm^{x^n} the product set ∏i=1nm\prod_{i=1}^n m, in AnA^n, where mi=mm_i = m for 1≤i≤n1 \leq i \leq n, to avoid ambiguity.

Proposition 7

Let AA be a ring and mm an ideal of AA such that the ordered pair (A,m)(A, m) satisfies Hensel's conditions. Let f=(f1,…,fn)f = (f_1, \ldots, f_n) be a system without constant term of series in A[[X1,…,Xn]]A[[X_1, \ldots, X_n]] such that Jf(0)J_f(0) is invertible in AA. Then, for all x∈mxnx \in m^{x^n}, f(x)∈mxnf(x) \in m^{x^n} and x↦f(x)x \mapsto f(x) is a bijection of mxnm^{x^n} onto itself, the inverse bijection being x↦g(x)x \mapsto g(x), where gg is given by relation (16) in no. 4.

The fact that f(x)∈mxnf(x) \in m^{x^n} is obvious when theft are polynomials and follows in the general case from Proposition 6 and the fact that mm is closed in AA. The other assertions of the proposition are then immediate consequences of (16), (17) and (18).

Corollary

Let qq be a closed ideal of AA contained in mm. Then the relation x≡x′(modqxn)x \equiv x' \pmod{q^{x^n}} is equivalent to f(x)≡f(x′)(modqxn)f(x) \equiv f(x') \pmod{q^{x^n}}.

For every formal power series f∈A[[X1,…,Xn]]f \in A[[X_1, \ldots, X_n]],

f(X1,…,Xn)−f(Y1,…,Yn)=∑i=1n(Xi−Yi)hi(X1,…,Xn,Y1,…,Yn) f(X_1, \ldots, X_n) - f(Y_1, \ldots, Y_n) = \sum_{i=1}^n (X_i - Y_i) h_i(X_1, \ldots, X_n, Y_1, \ldots, Y_n)

where the hih_i belong to A[[X1,…,Xn,Y1,…,Yn]]A[[X_1, \ldots, X_n, Y_1, \ldots, Y_n]] (Algebra, Chapter IV, § 5, no. 8, Proposition 9); it follows immediately that the relation x≡x′(modqxn)x \equiv x' \pmod{q^{x^n}} implies f(x)≡f(x′)(modqxn)f(x) \equiv f(x') \pmod{q^{x^n}}. The converse is obtained by replacing ff by its "inverse" gg.

Theorem 2

Let AA be a ring and mm an ideal of AA such that the ordered pair (A,m)(A, m) satisfies Hensel's conditions. Let f=(f1,…,fn)f = (f_1, \ldots, f_n) be a system of nn elements of

A\{X_1, \ldots, X_n\} and let a∈Ana \in A^n; let us write Jt(a)=eJ_t(a) = e. There exists a system g=(g1,…,gn)g = (g_1, \ldots, g_n) of restricted formal power series without constant term in A{X1,…,Xn}A\{X_1, \ldots, X_n\} such that (i) Mg(0)=IM_g(0) = I, (unit matrix). (ii) For all x∈Anx \in A^n,

(19)(19)
f(a+ex)=f(a)+Mt(a)⋅eg(x).f(a + ex) = f(a) + M_t(a) \cdot eg(x).
(iii) Let h=(h1,…,hn)h = (h_1, \ldots, h_n) be the system of formal power series without constant term (not necessarily restricted) such that g∘h=1ng \circ h = 1_n (Proposition 5). For all y∈mn×ny \in m^{n \times n},
(20)(20)
f(a+eh(y))=f(a)+Mt(a)⋅ey.f(a + eh(y)) = f(a) + M_t(a) \cdot ey.
For every formal power series f∈A[[X1,…,Xn]]f \in A[[X_1, \ldots, X_n]],
(21)(21)
f(X+Y)=f(X)+Mf(X)⋅Y+∑1≤i<j≤nGij(X,Y)YiYj,f(X + Y) = f(X) + M_f(X) \cdot Y + \sum_{1 \leq i < j \leq n} G_{ij}(X, Y)Y_i Y_j,
where the GijG_{ij} are well determined formal power series in
A[[X1,…,Xn,Y1,…,Yn]].A[[X_1, \ldots, X_n, Y_1, \ldots, Y_n]].
Iff is restricted, so are the elements of MfM_f and the GijG_{ij}, for these formal power series are polynomials iff is a polynomial and it follows from their uniqueness that for every open ideal J\mathfrak{J} of AA, denoting by pJ:A→A/Jp_\mathfrak{J}: A \to A/\mathfrak{J} the canonical mapping, the image of GijG_{ij} under pJp_\mathfrak{J} is the coefficient of YiYjY_i Y_j in pJ(F)p_\mathfrak{J}(F) where FF is the formal power series f(X+Y)f(X + Y) in A[[X1,…,Xn,Y1,…,Yn]]A[[X_1, \ldots, X_n, Y_1, \ldots, Y_n]]; whence our assertion.

This being so, writing formula (21) for each seriesft (1≤i≤n)(1 \leq i \leq n), we obtain for all x∈Anx \in A^n (no. 2, Proposition 4),

(22)(22)
f(a+ex)=f(a)+Mt(a)⋅ex+e2r(x)f(a + ex) = f(a) + M_t(a) \cdot ex + e^2 r(x)
where r=(r1,…,rr)r = (r_1, \ldots, r_r) is a system of restricted formal power series each of which is of total order ≥2\geq 2. It follows from formulae (18) of Algebra, Chapter III, § 6, no. 5 that there exists a square matrix M′∈Mn(A)M' \in \mathbf{M}_n(A) such that
(23)(23)
Mt(a)⋅M′=eIn,M_t(a) \cdot M' = eI_n,
whence using this in (22)
(24)(24)
f(a+ex)=f(a)+Mt(a)⋅ex+Mt(a)M′⋅er(x).f(a + ex) = f(a) + M_t(a) \cdot ex + M_t(a) M' \cdot er(x).

Writing g=1,+M′⋅rg = 1, + M' \cdot r, we see that gg satisfies conditions (i) and (ii); then it is sufficient to replace xx by h(y)h(y) to obtain (iii).

Corollary 1

Let AA be a ring and mm an ideal of AA such that the ordered pair (A,m)(A, m) satisfies Hensel's conditions. Let f∈A{X}f \in A\{X\}, a∈Aa \in A and write e=f′(a)e = f'(a). If f(a)≡0f(a) \equiv 0 (mod. e2me^2 m), then there exists b∈Ab \in A such that f(b)=0f(b) = 0 and b≡ab \equiv a (mod. emem). If b' is another element of A\mathbf{A} such that f(b′)=0f(b') = 0 and b′≡a(modm)b' \equiv a \pmod{\mathfrak{m}}, then e(b−b′)=0e(b - b') = 0. In particular, bb is unique if ee is not a divisor of zero in A\mathbf{A}.

Let f(a)=e2cf(a) = e^2 c where c∈mc \in \mathfrak{m}; formula (20) for n=1n = 1 gives

f(a+eh(y))=e2(c+y) f(a + eh(y)) = e^2(c + y)
and it is therefore sufficient to take y=−cy = -c, b=a+eh(−c)b = a + eh(-c). Moreover if b=a+ex,b′=a+ex′,x∈m,x′∈m,f(b)=f(b′)=0b = a + ex, b' = a + ex', x \in \mathfrak{m}, x' \in m, f(b) = f(b') = 0, we deduce from (19) that e2(g(x)−g(x′))=0e^2(g(x) - g(x')) = 0. As g(X)−g(Y)=(X−Y)u(X,Y)g(\mathbf{X}) - g(\mathbf{Y}) = (\mathbf{X} - \mathbf{Y})u(\mathbf{X}, \mathbf{Y}), where uu is restricted and u(0,0)=1u(0, 0) = 1, g(x)−g(x′)=(x−x′)vg(x) - g(x') = (x - x')v, where v∈Av \in \mathbf{A} is invertible, for, since mm is closed, v−1=u(x,x′)−1∈mv - 1 = u(x, x') - 1 \in \mathfrak{m} and mm is contained in the Jacobson radical of A\mathbf{A}; whence the relation e(b−b′)=0e(b - b') = 0.

Remark

The corollary applies notably when ee is invertible in A\mathbf{A}; we can then also deduce the existence of bb from Hensel's Theorem, for the canonical image of f(X)f(\mathbf{X}) in (A/m){X}(\mathbf{A}/\mathfrak{m})\{\mathbf{X}\} is of the form (X−a)f1(X),X−a(\mathbf{X} - a)f_1(\mathbf{X}), \mathbf{X} - a and f1(X)f_1(\mathbf{X}) being strongly relatively prime, for f1(α)=f′(α)f_1(\alpha) = f'(\alpha) is the image of ee (no. 1, Example).

Examples (1) Let pp be a prime number ≠2\neq 2 and nn an integer whose class mod. pp is a square ≠0\neq 0 in the prime field Fp\mathbf{F}_p. If Zp\mathbf{Z}_p is the ring of pp-adic integers (§ 2, no. 12, Example 3), the application of Corollary 1 to the polynomial X2−n\mathbf{X}^2 - n shows that nn is a square in Zp\mathbf{Z}_p; for example 7 is a square in Z3\mathbf{Z}_3. (2) Let A=K[[Y]]\mathbf{A} = \mathbf{K}[[\mathbf{Y}]] be the ring of formal power series in one indeterminate with coefficients in a commutative field K\mathbf{K}; with the (Y)(\mathbf{Y})-adic topology, the ring A\mathbf{A} is Hausdorff and complete (§ 2, no. 6, Corollary to Proposition 6) and the mapping f(Y)↦f(0)f(\mathbf{Y}) \mapsto f(0) defines by passing to the quotient ring an isomorphism of A/(Y)\mathbf{A}/(\mathbf{Y}) onto the field K\mathbf{K}. By Corollary 1, if F(Y,X)F(\mathbf{Y}, \mathbf{X}) is a polynomial in X\mathbf{X} with coefficients in A\mathbf{A} and aa is a simple root of F(0,X)F(0, \mathbf{X}) in K\mathbf{K}, there exists a unique formal power series f(Y)f(\mathbf{Y}) such that f(0)=af(0) = a and F(Y,f(Y))=0F(\mathbf{Y}, f(\mathbf{Y})) = 0.

Corollary 2

Let A\mathbf{A} be a ring and m\mathfrak{m} an ideal of A\mathbf{A} such that the ordered pair (A,m)(\mathbf{A}, m) satisfies Hensel's conditions. Let r,nr, n be integers such that 0≤r<n0 \leq r < n and f=(fr+1,…,fn)\mathbf{f} = (f_{r+1}, \ldots, f_n) is a system of n−rn - r elements of A{X1,…,Xn}\mathbf{A}\{\mathbf{X}_1, \ldots, \mathbf{X}_n\}; let Jf(n−r)(X)J^{(n-r)}_{\mathbf{f}}(\mathbf{X}) denote the minor of Mf(X)M_{\mathbf{f}}(\mathbf{X}) consisting of the columns of index jj such that r+1≤j≤nr + 1 \leq j \leq n. Let a∈Ana \in \mathbf{A}^n be such that Jf(n−r)(a)J^{(n-r)}_{\mathbf{f}}(a) is invertible in A\mathbf{A} and f(a)≡0(modm(n−r))\mathbf{f}(a) \equiv 0 \pmod{\mathfrak{m}^{(n-r)}}. Then there exists a unique x=(x1,…,xn)∈An\mathbf{x} = (x_1, \ldots, x_n) \in \mathbf{A}^n such that xk=akx_k = a_k for 1≤k≤r1 \leq k \leq r, x≡a(modmn)\mathbf{x} \equiv a \pmod{\mathfrak{m}^n} and f(x)=0\mathbf{f}(\mathbf{x}) = 0.

Substituting aa, for XkX_k for 1≤k≤r1 \leq k \leq r in the fif_i (no. 2, Remark 3 ), we see immediately that we may restrict our attention to the case where r=0r = 0 to prove the corollary. Theorem 1 and Proposition 7 show then that f\mathbf{f} defines a bijection of a+mna + \mathfrak{m}^n onto f(a)+mn=mn\mathbf{f}(a) + \mathfrak{m}^n = \mathfrak{m}^n; the corollary follows from the fact that 0∈mn0 \in \mathfrak{m}^n.

Corollary 3

With the notation & Corollary 2, let a∈Ana \in \mathbf{A}^n; let us write e=Jr(n−r)(a)e = J_r^{(n-r)}(\mathbf{a}) (not necessarily invertible in A\mathbf{A}) and suppose that f(a)≡0f(\mathbf{a}) \equiv 0 (mod. e2m×(n−r)e^{2m^{\times(n-r)}}). Then there exist n−rn - r formal power series without constant term ϕi\phi_i (r+1≤i≤nr + 1 \leq i \leq n) in A[[X1,…,Xr]]\mathbf{A}[[X_1, \ldots, X_r]] such that, for all t=(t1,…,tr)∈m×rt = (t_1, \ldots, t_r) \in m^{\times r},

fi(a1+e2t1,…,ar+e2tr,ar+1+eϕr+1(t),…,an+eϕn(t))=0 f_i(a_1 + e^2 t_1, \ldots, a_r + e^2 t_r, a_{r+1} + e \phi_{r+1}(t), \ldots, a_n + e \phi_n(t)) = 0

for r+1≤i≤nr + 1 \leq i \leq n.

*For 1≤i≤r1 \leq i \leq r, let fi(X)=Xi−af_i(X) = X_i - a, and let u=(f1,…,fn)u = (f_1, \ldots, f_n); then Ju(a)=eJ_u(a) = e and Theorem 2 may be applied to the system uu. With the notation of Theorem 2 it follows from the above definitions that gi(X)=Xg_i(X) = X, for 1≤i≤r1 \leq i \leq r, whence hi(X)=Xih_i(X) = X_i for 1≤i≤r1 \leq i \leq r; moreover, if M′∈Mn(A)M' \in \mathbf{M}_n(\mathbf{A}) is such that Mu(a).M′=eInM_u(a) . M' = e I_n, M′M' is of the form

(eIr0∗∗). \begin{pmatrix} e I_r & 0 \\ * & * \end{pmatrix}.

Replacing yy by M′.zM'.z (where z=(z1,…,zn)∈m×nz = (z_1, \ldots, z_n) \in m^{\times n}) in formula (20), we obtain

fi(a1+e2z1,…,ar+e2zr,ar+1+ehr+1(M′.z),…,an+ehn(M′.z))=&(a)+e2zifor 1≤i≤n. f_i(a_1 + e^2 z_1, \ldots, a_r + e^2 z_r, a_{r+1} + e h_{r+1}(M'.z), \ldots, a_n + e h_n(M'.z)) = \& (a) + e^2 z_i \quad \text{for } 1 \leq i \leq n.

By hypothesis, fj(a)=e2bjf_j(a) = e^2 b_j, where bj∈mb_j \in m for r+1≤j≤nr + 1 \leq j \leq n. Let us write ψj(X1,…,Xr)=hj(M′.X)\psi_j(X_1, \ldots, X_r) = h_j(M'.X) and

ϕj(X1,…,Xr)=ψj(X1,…,Xr,−br+1,…,−bn) \phi_j(X_1, \ldots, X_r) = \psi_j(X_1, \ldots, X_r, -b_{r+1}, \ldots, -b_n)

for r+1≤j≤nr + 1 \leq j \leq n. For r+1≤i≤nr + 1 \leq i \leq n, substituting tjt_j for zjz_j for 1≤j≤r1 \leq j \leq r and bjb_j for zjz_j for r+1≤j≤nr + 1 \leq j \leq n in (26), we obtain relations (25) for all t∈m×rt \in m^{\times r}.

6. Application to decompositions of rings

Lemma 2

Let A\mathbf{A} be a ring and mm an ideal of A\mathbf{A} such that the ordered pair (A,m)(\mathbf{A}, m) satisfies Hensel’s conditions. Let B\mathbf{B} be the quotient ring A/m\mathbf{A}/m and π:A→B\pi : \mathbf{A} \to \mathbf{B} be the canonical homomorphism. For every idempotent cc of B\mathbf{B} there exists a unique idempotent ee of A\mathbf{A} such that π(e)=c\pi(e) = c.

Let aa be such that π(a)=c\pi(a) = c; Corollary 1 to Theorem 2 of no. 5 may be applied to the polynomial f(X)=X2−Xf(X) = X^2 - X in A[X]\mathbf{A}[X] and the element a∈Aa \in \mathbf{A}. Then f′(a)=2a−1f'(a) = 2a - 1 and, as π(2a−1)=2c−1\pi(2a - 1) = 2c - 1 and (2c−1)2=1(2c - 1)^2 = 1 in B\mathbf{B}, 2c−12c - 1 is invertible in B\mathbf{B} and hence 2a−12a - 1 is invertible in A\mathbf{A} (§ 2, no. 13, Lemma 3). As f(a)∈mf(a) \in m, Corollary 1 to Theorem 2 of no. 5 immediately gives the existence and uniqueness of ee.

Proposition 8

Let A\mathbf{A} be a ring and mm an ideal of A\mathbf{A} such that the ordered pair (A,m)(\mathbf{A}, m) satisfies Hensel’s conditions. Let B\mathbf{B} be the quotient ring A/m\mathbf{A}/m and π:A→B\pi : \mathbf{A} \to \mathbf{B} the canonical homomorphism. If B is the direct composition of a finite family (bi)i∈I(\mathfrak{b}_i)_{i \in I} of ideals, there exists a unique family (ai)i∈I(\mathfrak{a}_i)_{i \in I} of ideals of A such that π(ai)=bi\pi(\mathfrak{a}_i) = \mathfrak{b}_i for all i∈Ii \in I and A is the direct composition of the family (ai)(\mathfrak{a}_i).

Let I=∑iciI = \sum_i c_i where ci∈bic_i \in \mathfrak{b}_i for all ii; the cic_i are idempotents of BB such that cicj=0c_i c_j = 0 for i≠ji \neq j. By Lemma 2 there therefore exist idempotents eie_i of AA (i∈Ii \in I) such that π(ei)=ci\pi(e_i) = c_i for all ii; as eieje_i e_j is an idempotent such that

π(eiej)=cicj=0 \pi(e_i e_j) = c_i c_j = 0
for i≠ji \neq j, eiej=0e_i e_j = 0 for i≠ji \neq j (Lemma 2); as 1−∑iei1 - \sum_i e_i is an idempotent such that
π(1−∑iei)=1−∑ici=0, \pi\left(1 - \sum_i e_i\right) = 1 - \sum_i c_i = 0,
similarly 1=∑iei1 = \sum_i e_i. It follows that AA is the direct composition of the ideals ai=eiA\mathfrak{a}_i = e_i A and that π(ai)=π(ei)B=bi\pi(\mathfrak{a}_i) = \pi(e_i) B = \mathfrak{b}_i.

It remains to show the uniqueness of such a decomposition. Now, suppose that AA is the direct composition of another family (ai′)i∈I(\mathfrak{a}'_i)_{i \in I} of ideals such that π(ai′)=bi\pi(\mathfrak{a}'_i) = \mathfrak{b}_i for all ii; then 1=∑iei′1 = \sum_i e'_i where ei′∈ai′e'_i \in \mathfrak{a}'_i, whence, in BB, 1=∑iπ(ei′)1 = \sum_i \pi(e'_i) where π(ei′)∈bi\pi(e'_i) \in \mathfrak{b}_i, which implies π(ei′)=ci\pi(e'_i) = c_i; as ei′e'_i and eie_i are idempotents, necessarily ei′=eie'_i = e_i (Lemma 2), which completes the proof.

Remark

Proposition 8 again gives the structure of a semi-local ring on AA which is Hausdorff and complete with the cc-adic topology (c the Jacobson radical of AA), which has already been obtained as a consequence of § 2, no. 13, Corollary to Proposition 19.

Exercises

See the exercises for § 4.

Transcribed from the printed volume. Not translated.