Let δ>0 be an integer, x an element of qδ, ξ its class in grδ(A)=qδ/qδ+1 and φ the multiplication by ξ in the gr(A)-module gr(M).
a) The dimension of the A-module M/xM is equal to dimA(M) or to dimA(M)−1. In the second case, one has eq(M/xM)≥δeq(M).
b) Suppose that one has dimA(M)≥1 and that the kernel of φ is of finite length over A/q. Then one has dimA(M/xM)=dimA(M)−1. Moreover: (i) If dimA(M)>1, one has eq(M/xM)=δeq(M). (ii) If dimA(M)=1, one has for every integer n≥0
nδeq(M)≤longA(M/xnM)≤nδeq(M)+longA/q(Ker φn), where
φn is the n-th iterate of the endomorphism
φ, and
δeq(M)=exA(M)≤longA(M/xM).
Put M′′=M/xM; consider the Hilbert-Samuel series HM=HM,q and HM′′=HM′′,q, together with the Poincaré series P(T)=∑n≥0longA/q(Ker φn)⋅Tn. By § 4, No. 3, th. 2 and No. 4, remark 2, one has
HM(T)=(1−T)−dRM(T),HM′′(T)=(1−T)−d′′RM′′(T),
with d=dimA(M), d′′=dimA(M′′), RM and RM′′ in Z[T], RM(1)=eq(M), RM′′(1)=eq(M′′). By lemma 6 of § 4, No. 5, one has in Z((T)) the inequalities
(1−Tδ)HM(1)≤HM′′(1)≤(1−Tδ)HM(1)+TδP(1).
Putting R(T)=(1−Tδ)/(1−T)=1+T+⋯+Tδ−1, this may also be written
(1)
(1−T)−dR(T)RM(T)≤(1−T)−d′′−1RM′′(T)≤ ≤(1−T)−dR(T)RM(T)+(1−T)−1TδP(T).
By Lemma 2 of § 4, No. 1, the first inequality (1) implies either d′′≥d, or d′′=d−1 and R(1)RM(1)≤RM′′(1), that is to say δeq(M)≤eq(M′′). This proves a), since d′′≤d.
Under the assumption of b), one has P(T)∈Z[T] and P(1)=longA(Kerφ). The second inequality (1) is written
(1−T)−d′′−1RM′′(T)≤(1−T)−d(R(T)RM(T)+Tδ(1−T)d−1P(T)).
Suppose that one has d>1; then Lemma 2 of § 4, No. 1 yields d′′+1≤d, whence d′′=d−1 by part a) of the proof; one then has
RM′′(1)≤R(1).RM(1)
(loc. cit.), whence (i).
Now suppose d=1. By loc. cit., one has d′′=0 and
RM′′(1)≤R(1).RM(1)+P(1).
Consequently, M′′ is of finite length equal to eq(M′′)=RM′′(1), and one obtains
(2)
δeq(M)≤longA(M/xM)≤δeq(M)+longA(Kerφ).
Let n≥1 be an integer. Replace x by xn in (2); one therefore has
(3)
nδeq(M)≤longA(M/xnM)≤nδeq(M)+longA(Kerφn).
It is immediate that the submodules Kerφn of the noetherian gr(A)-module gr(M) form an increasing sequence and hence are stationary, and that each of them is of finite length over A/q. Dividing by n≥1 in inequality (3) and making n tend to +∞, one finds exA(M)=δeq(M) by definition of exA(M).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.
Nothing.