Bourbaki

§ 2. Dimension of algebras

1. Dimension and flatness

Let ρ:A→B\rho : A \to B be a homomorphism of rings. We denote by (PM) the following condition: (PM) There exists a B-module N faithfully flat over A such that, for every prime ideal q of B, one has N⊗Bκ(q)≠0N \otimes_B \kappa(q) \neq 0.

Remark 1

The condition (PM) is satisfied when there exists a finitely generated B-module, faithfully flat over A and with support equal to Spec(B). This is the case, in particular, if the A-module B is faithfully flat.

Remark 2

The existence of a B-module N faithfully flat over A implies the injectivity of ρ (I, § 3, No. 5, prop. 8), and the surjectivity of the mapping “ρ : Spec(B) → Spec(A) (II, § 2, No. 5, cor. 4 to prop. 11).

Remark 3

Suppose that ρ : A → B is a local homomorphism of local rings and that there exists a B-module N flat over A and such that N ⊗_B κ(q) ≠ 0 for every prime ideal q of B. Then N is faithfully flat over A and ρ therefore has the property (PM): indeed we have N/m_B N = N ⊗_B κ(m_B) ≠ 0, hence N ≠ m_B N and fortiori N ≠ m_A N, and the conclusion follows from prop. 1 of I, § 3, No. 1.

Proposition 1

Let ρ : A → B be a homomorphism of rings satisfying the condition (PM).

a) Let h : A → A′ be a homomorphism of rings. Then the homomorphism ρ′ : A′ → A′ ⊗_A B deduced from ρ satisfies the condition (PM).

b) Let q be a prime ideal of B and p = ρ^{-1}(q). The canonical homomorphism ρ_q : A_p → B_q satisfies the condition (PM).

c) Let q be a prime ideal of B and p = ρ^{-1}(q). For every prime ideal p′ of A contained in p, there exists a prime ideal q′ of B lying above p′ and contained in q.

Let N be a B-module faithfully flat over A and such that N ⊗_B κ(q) ≠ 0 for every prime ideal q of B.

Prove a). The A′-module N′ = A′ ⊗_A N is faithfully flat (I, § 3, No. 3, prop. 5); let q′ be a prime ideal of B′ = A′ ⊗_A B and q its inverse image in B. We have isomorphisms

N′⊗B′κ(q′)→N⊗Bκ(q′)→(N⊗Bκ(q))⊗κ(q)κ(q′); N' \otimes_{B'} \kappa(q') \to N \otimes_B \kappa(q') \to (N \otimes_B \kappa(q)) \otimes_{\kappa(q)} \kappa(q');

as we have N ⊗_B κ(q) ≠ 0, we also have N′ ⊗_{B′} κ(q′) ≠ 0.

Prove b). According to props. 13 and 14 of II, § 3, No. 4, the A_p-module N_q is flat. On the other hand, let b′ be a prime ideal of B_q; it is of the form bB_q where b is a prime ideal of B contained in q (II, § 3, No. 1, prop. 3); we have N ⊗_B κ(b) ≠ 0 in view of the assumption made on N, and since N_q ⊗_{B_q} κ(b′) is isomorphic to N ⊗_B κ(b′), we have N_q ⊗_{B_q} κ(b′) ≠ 0. Remark 3 allows us to conclude.

Prove c). The local homomorphism ρ_q : A_p → B_q deduced from ρ satisfies the condition (PM) according to b). The mapping Spec(B_q) → Spec(A_p) is therefore surjective (Remark 2), which was to be proved.

Corollary

Let F be a closed subset of Spec(A). If Y is an irreducible component of the inverse image of F by the mapping “ρ : Spec(B) → Spec(A), then the closure of “ρ(Y) is an irreducible component of F.

Let indeed a be an ideal of A such that F = V(a) and q the prime ideal of B such that Y = V(q). The inverse image by “ρ of F is the subset V(ρ(a)B) of Spec(B) and the closure of “ρ(Y) is the irreducible closed subset V(ρ^{-1}(q)) of Spec(A).

It is a matter of proving that if q is minimal among the prime ideals of B containing ρ(a)\rho(a), then ρ−1(q)\rho^{-1}(q) is minimal among the prime ideals of A containing a. Otherwise, there would exist a prime ideal p′p' of A with a⊂p′⊂ρ−1(q)a \subset p' \subset \rho^{-1}(q) and p′≠ρ−1(q)p' \neq \rho^{-1}(q); according to prop. 1, c), there would exist a prime ideal q′q' of B such that q′⊂qq' \subset q and p′=ρ−1(q′)p' = \rho^{-1}(q'), whence ρ(a)⊂q′⊂q\rho(a) \subset q' \subset q and q′≠qq' \neq q contrary to the assumption made on q.

Proposition 2

Let ρ:A→B\rho : A \to B be a homomorphism of nonzero rings possessing the property (PM). One has the inequality

(1)

dim⁡(B)≥dim⁡(A)+inf⁡m∈Sdim⁡(B/mB) \dim(B) \geq \dim(A) + \inf_{m \in S} \dim(B/mB)
where S is the set of maximal ideals of A.

We know that dim⁡(A)=sup⁡m∈Sdim⁡(Am)\dim(A) = \sup_{m \in S} \dim(A_m) (§ 1, No. 3, prop. 8). It is therefore enough to establish the inequality

(2)

dim⁡(B)≥dim⁡(Am)+dim⁡(B/mB) \dim(B) \geq \dim(A_m) + \dim(B/mB)
for every maximal ideal m of A. In other words, it is a matter of proving the inequality

(3)

dim⁡(B)≥n+r \dim(B) \geq n + r
if p0⊂...⊂pnp_0 \subset ... \subset p_n is a chain of prime ideals of A contained in m and q‾0⊂...⊂q‾r\overline{q}_0 \subset ... \subset \overline{q}_r is a chain of prime ideals of B/mB. For 0≤i≤r0 \leq i \leq r, there exists a prime ideal qn+iq_{n+i} of B lying above m such that q‾i=qn+i/mB\overline{q}_i = q_{n+i}/mB, and qn⊂...⊂qn+rq_n \subset ... \subset q_{n+r} is a chain of prime ideals of B. Put pi′=pip'_i = p_i for 0≤i≤n−10 \leq i \leq n-1 and pn′=mp'_n = m, so that p0′⊂...⊂pn′p'_0 \subset ... \subset p'_n is a chain of prime ideals of A and q_n lies above pn′p'_n. If qiq_i is a prime ideal of B lying above pi′p'_i (1≤i≤n1 \leq i \leq n), prop. 1, c) proves that there exists a prime ideal qi−1q_{i-1} of B lying above pi−1′p'_{i-1} and contained in qiq_i. By descending induction, we thus construct a chain q0⊂...⊂qnq_0 \subset ... \subset q_n of prime ideals of B such that qiq_i lies above pip_i for 0≤i≤n0 \leq i \leq n. As q0⊂...⊂qn+rq_0 \subset ... \subset q_{n+r} is a chain of prime ideals of B, we have proved the inequality (3).

Remark 4

Let ρ:A→B\rho : A \to B be a local homomorphism of noetherian local rings satisfying condition (PM). We shall see later (§ 3, no 4, prop. 7) that in this case there is equality in (1). In the general case there may be strict inequality (p. 84, exercise 1).

Corollary

For every ideal a of A, one has ht(a)≤ht(ρ(a)B)\mathrm{ht}(a) \leq \mathrm{ht}(\rho(a)B).

Let q be a prime ideal of B containing ρ(a)B\rho(a)B, and p=ρ−1(q)p = \rho^{-1}(q). According to prop. 1, the local homomorphism ρq:Ap→Bq\rho_q : A_p \to B_q deduced from ρ\rho satisfies (PM), and we therefore have dim⁡(Ap)≤dim⁡(Bq)\dim(A_p) \leq \dim(B_q) according to prop. 2. According to prop. 7 of § 7, no 3, we have ht(a)≤dim⁡(Ap)\mathrm{ht}(a) \leq \dim(A_p) since p contains a, whence ht(a)≤dim⁡(Bq)\mathrm{ht}(a) \leq \dim(B_q) for every prime ideal q of B containing ρ(a)B\rho(a)B. The corollary then follows from loc. cit.

Lemma 1

Let ρ:A→B\rho : A \to B be a homomorphism of rings and p a prime ideal of A. The continuous mapping ah:Spec(B⊗Aκ(p))→Spec(B)^a h : \mathrm{Spec}(B \otimes_A \kappa(p)) \to \mathrm{Spec}(B), associated with the canonical homomorphism h:B→B⊗Aκ(p)h : B \to B \otimes_A \kappa(p), induces a homeomorphism of Spec(B⊗Aκ(p))\mathrm{Spec}(B \otimes_A \kappa(p)) onto the subspace (αρ)−1(p)(\alpha\rho)^{-1}(p) of Spec(B)\mathrm{Spec}(B) formed by the prime ideals of BB lying above pp.

The homomorphism hh is composed of the homomorphism of passing to the quotient of BB into B/ρ(p)B/\rho(p) BB and of the canonical homomorphism of B/ρ(p)B/\rho(p) BB into its ring of fractions (ρ(A−p))−1(B/ρ(p) B)(\rho(A - p))^{-1}(B/\rho(p) \ B). According to the remark and the corollary to prop. 13 of II, § 4, no 3, ah^a h therefore induces a homeomorphism of Spec(B⊗Aκ(p))\mathrm{Spec}(B \otimes_A \kappa(p)) onto the subspace of Spec(B)\mathrm{Spec}(B) formed by the prime ideals qq of BB which contain ρ(p)\rho(p) and are disjoint from ρ(A−p)\rho(A - p), that is to say which lie above pp.

Remark 5

According to prop. 2 and Lemma 1, we therefore have, under the hypotheses of prop. 2, the inequality

dim⁡(Spec(B))≥dim⁡(Spec(A))+inf⁡p∈Spec(A)dim⁡(αρ−1(p)). \dim(\mathrm{Spec}(B)) \geq \dim(\mathrm{Spec}(A)) + \inf_{p \in \mathrm{Spec}(A)} \dim(\alpha\rho^{-1}(p)) .

2. Dimension of an algebra of finite type

Proposition 3

Let* ρ:A→B\rho : A \to B be a homomorphism of rings. Put n=sup⁡p∈Spec(A)dim⁡(B⊗Aκ(p))n = \sup_{p \in \mathrm{Spec}(A)} \dim(B \otimes_A \kappa(p)). One has the inequality

dim⁡(B)+1≤(dim⁡(A)+1).(n+1). \dim(B) + 1 \leq (\dim(A) + 1).(n + 1) .

We may suppose dim⁡(A)≠−∞\dim(A) \neq -\infty and n<+∞n < +\infty. Let q0⊂...⊂qmq_0 \subset ... \subset q_m be a chain of prime ideals of B; put pi=ρ−1(qi)p_i = \rho^{-1}(q_i). The sequence of the pip_i is increasing, hence the set of its values has cardinal ≤dim⁡(A)+1\leq \dim(A) + 1. For each p∈Spec(A)p \in \mathrm{Spec}(A), the set of the qjq_j such that pj=pp_j = p is a chain of the subset aρ−1(p)^a\rho^{-1}(p) of Spec(B)\mathrm{Spec}(B), hence has cardinal less than dim⁡(B⊗Aκ(p))+1\dim(B \otimes_A \kappa(p)) + 1 (no 1, Lemma 1), and consequently less than or equal to (n+1)(n + 1). It follows that m+1≤(dim⁡(A)+1)(n+1)m + 1 \leq (\dim(A) + 1)(n + 1), whence the proposition.

Remark 1

If the rings AA and BB are noetherian, we shall see below (§ 3, no 4, cor. 2 to prop. 7) that one has the inequality dim⁡(B)≤dim⁡(A)+n\dim(B) \leq \dim(A) + n, stronger than that of prop. 3.

Corollary 1

Suppose that one has* dim⁡(A)<+∞\dim(A) < +\infty and that there exists an integer nn such that dim⁡(B⊗Aκ(p))≤n\dim(B \otimes_A \kappa(p)) \leq n for every p∈Spec(A)p \in \mathrm{Spec}(A). Then one has dim⁡(B)<+∞\dim(B) < +\infty.

Corollary 2

Let* AA be a ring and B=A[X]B = A[X] the ring of polynomials in one indeterminate with coefficients in AA. One has :

1+dim⁡(A)≤dim⁡(B)≤1+2dim⁡(A). 1 + \dim(A) \leq \dim(B) \leq 1 + 2\dim(A) .

The first inequality has already been proved (§ 1, No. 3, example 4). Let us prove the second one. For every prime ideal pp of AA, the ring B⊗Aκ(p)B \otimes_A \kappa(p), isomorphic to κ(p)[X]\kappa(p)[X], is principal and is not a field, hence is of dimension 1 (§ 1, No. 3, example 2), and the inequality follows from prop. 3.

However, whatever the integers n and q with n + 1 ≤ q ≤ 2n + 1 may be, there exists a ring A of dimension n such that dim(A[X]) = q (see p. 84, exerc. 7).

Corollary 3

If A is of finite dimension, every nonzero finitely generated A-algebra is of finite dimension.

Indeed, one deduces from cor. 2, by induction on n, that the ring A[T₁, ..., Tₙ] is of finite dimension if A is of finite dimension; fortiori, every nonzero quotient of A[T₁, ..., Tₙ] is of finite dimension (§ 1, No. 3, prop. 6).

3. Dimension of an integral algebra

Lemma 2

Let ρ : A → B be a homomorphism of rings such that, for every prime ideal p of A, the κ(p)-algebra B ⊗ₓₐ κ(p) is integral (V, § 1, No. 1, def. 2). Let q and q’ be two prime ideals of B such that q ⊂ q’ and q ≠ q’. Then ρ⁻¹(q) ≠ ρ⁻¹(q’).

Indeed, if q and q’ are over the same prime ideal p of A, one has dim(B ⊗ₓₐ κ(p)) ≥ 1 by Lemma 1 of No. 1, which contradicts the fact that dim(B ⊗ₓₐ κ(p)) ≤ 0 (§ 1, No. 3, example 6).

Theorem 1

Let ρ : A → B be a homomorphism of rings making B an integral A-algebra.

a) Let M be a finitely generated A-module. Then one has dim_B(M ⊗ₓₐ B) ≤ dim_A(M). In particular, one has dim(B) ≤ dim(A). If the mapping “ρ : Spec(B) → Spec(A) is surjective, for example (V, § 2, No. 1, th. 1) if ρ is injective, one has dim_B(M ⊗ₓₐ B) = dim_A(M), and in particular dim(B) = dim(A).

b) Let b be an ideal of B and a = ρ⁻¹(b) its inverse image in A. One has ht(b) ≤ ht(a) and dim(B/b) = dim(A/a). If “ρ : Spec B → Spec A is surjective, one has ht(aB) ≤ ht(a) for every ideal a of A.

c) Suppose that B is finite over A and let N be a finitely generated B-module. Then one has dim_B(N) = dim_A(N). In particular, one has dim(B) = dim_A(B).

Let us prove a). By prop. 5 of V, § 1, No. 1, the κ(p)-algebra B ⊗ₓₐ κ(p) is integral for every prime ideal p of A. Let q₀ ⊂ ... ⊂ qₘ be a chain of prime ideals of B; by Lemma 2, the ideals pᵢ = ρ⁻¹(qᵢ) are pairwise distinct, hence p₀ ⊂ ... ⊂ pₘ is a chain of prime ideals of A, whence m ≤ dim(A). Hence dim(B) ≤ dim(A).

Suppose now that the mapping “ρ is surjective. Let p₀ ⊂ ... ⊂ pₙ be a chain of prime ideals of A; there exists therefore a prime ideal q₀ of B over p₀. By cor. 2 to the first existence theorem (V, § 2, No. 1, th. 1), one can construct, by induction on n, a chain q₀ ⊂ ... ⊂ qₙ of prime ideals of B such that qᵢ is over pᵢ for 0 ≤ i ≤ n. Hence n ≤ dim(B) and consequently dim(A) ≤ dim(B).

This proves aa in the case where M=AM = A. In the general case, let aa be the annihilator of MM, so that the support of MM is identified with Spec⁡(A/a)\operatorname{Spec}(A/a), and one has dim⁡A(M)=dim⁡(A/a)\dim_A(M) = \dim(A/a). By II, § 4, No. 4, prop. 19, the support of M⊗ABM \otimes_A B is the inverse image by aρ^a\rho of the support of MM, hence is identified with Spec⁡(B/ρ(a)B)\operatorname{Spec}(B/\rho(a) B), and one has dim⁡B(M⊗AB)=dim⁡(B/ρ(a)B)\dim_B(M \otimes_A B) = \dim(B/\rho(a) B). It remains to observe that the homomorphism ρ′:A/a→B/ρ(a)B\rho': A/a \to B/\rho(a) B deduced from ρ\rho makes B/ρ(a)BB/\rho(a) B an (A/a)(A/a)-algebra integral, and that “ρ′\rho' is surjective when aρ^a\rho is so.

Let us prove bb. By prop. 7 of § 1, No. 3, it suffices to prove that ht⁡(b)≤dim⁡(Ap)\operatorname{ht}(b) \leq \dim(A_p) for every prime ideal pp of AA containing aa; let pp be such an ideal. By V, § 2, No. 1, cor. 2 to th. 1, there exists a prime ideal qq of BB over pp and containing bb, and one has ht⁡(b)≤dim⁡(Bq)\operatorname{ht}(b) \leq \dim(B_q) by prop. 7 of § 1, No. 3.

Now BqB_q identifies with a ring of fractions of the ApA_p-algebra B⊗AApB \otimes_A A_p, whence

dim⁡(Bq)≤dim⁡(B⊗AAp)≤dim⁡(Ap) \dim(B_q) \leq \dim(B \otimes_A A_p) \leq \dim(A_p)
by Prop. 6 of § 1, No. 3 and the assertion aa above. We have thus proved the inequality ht⁡(b)≤ht⁡(a)\operatorname{ht}(b) \leq \operatorname{ht}(a). Moreover, the homomorphism from A/aA/a into B/bB/b deduced from ρ\rho is injective and makes B/bB/b an (A/a)(A/a)-algebra which is integral; we therefore have dim⁡(B/b)=dim⁡(A/a)\dim(B/b) = \dim(A/a) by aa. Suppose “ρ\rho is surjective and let aa be an ideal of AA and pp a prime ideal of AA containing aa. There exists by assumption a prime ideal qq of BB lying above pp. We have aB⊂qaB \subset q, whence ht⁡(aB)≤ht⁡(q)≤ht⁡(p)\operatorname{ht}(aB) \leq \operatorname{ht}(q) \leq \operatorname{ht}(p) by what precedes. Passing to the greatest lower bound, we obtain ht⁡(aB)≤ht⁡(a)\operatorname{ht}(aB) \leq \operatorname{ht}(a).

Finally, cc follows from bb applied to the annihilator bb of NN.

Theorem 2

Let AA be an integrally closed ring, and BB a ring containing AA, integral over AA. Assume that BB is a torsion-free AA-module. For every ideal aa of AA, one has ht⁡(a)=ht⁡(aB)\operatorname{ht}(a) = \operatorname{ht}(aB). Let bb be an ideal of BB and a=b∩Aa = b \cap A; one then has ht⁡(a)=ht⁡(b)\operatorname{ht}(a) = \operatorname{ht}(b).

Soit ρ\rho l’application canonique de AA dans BB. Soient aa un idéal de AA. Si a=Aa = A, la première égalité est claire. Supposons a≠Aa \neq A. Comme ρ\rho est injectif, aρ^a\rho est surjectif (V, § 2, No. 1, Theorem 1). Par suite aB≠BaB \neq B. Soit alors qq un idéal premier de BB contenant aBaB. Posons p=q∩Ap = q \cap A. On a a⊂pa \subset p, d’où ht⁡(a)≤ht⁡(p)\operatorname{ht}(a) \leq \operatorname{ht}(p). Soit p0⊂...⊂pnp_0 \subset ... \subset p_n une chaîne d’idéaux premiers de AA avec pn=pp_n = p. D’après le deuxième théorème d’existence (V, § 2, No. 4, Theorem 3), on construit par récurrence une chaîne q0⊂...⊂qnq_0 \subset ... \subset q_n d’idéaux premiers de BB telle que qn=qq_n = q et qiq_i soit au-dessus de pip_i pour 0≤i≤n0 \leq i \leq n. On a n≤ht⁡(q)n \leq \operatorname{ht}(q), d’où ht⁡(a)≤ht⁡(q)\operatorname{ht}(a) \leq \operatorname{ht}(q). En passant à la borne inférieure on obtient ht⁡(a)≤ht⁡(aB)\operatorname{ht}(a) \leq \operatorname{ht}(aB) (§1§ 1, No. 3, Prop. 7). L’inégalité ht⁡(aB)≤ht⁡(a)\operatorname{ht}(aB) \leq \operatorname{ht}(a) résulte du Theorem 1, d’où la première equality. Soit bb un idéal de BB. Posons a=ρ−1(b)a = \rho^{-1}(b). On a aB⊂baB \subset b, d’où ht⁡(a)=ht⁡(aB)≤ht⁡(b)\operatorname{ht}(a) = \operatorname{ht}(aB) \leq \operatorname{ht}(b). L’inégalité ht⁡(b)≤ht⁡(a)\operatorname{ht}(b) \leq \operatorname{ht}(a) résulte du Theorem 1, d’où le theorem.

Remark

Soient AA un anneau intègre et BB un anneau contenant AA, entier sur AA. Soit pp un idéal premier de AA tel que la clôture intégrale de ApA_p soit un anneau local. On peut démontrer que, pour tout idéal premier qq de BB au-dessus de pp, on a ht⁡(p)=ht⁡(q)\operatorname{ht}(p) = \operatorname{ht}(q) (p. 85, Exer. 9) lorsque BB est intègre.

4. Algebras of finite type over a field

Dans ce numéro, kk désigne un field.

Lemma 3

Let AA be a kk-algebra of finite type and p0⊂...⊂pmp_0 \subset ... \subset p_m a maximal chain of prime ideals of AA. There exists an integer n≥mn \geq m, a sequence (x1,...,xn)(x_1, ..., x_n) of elements of AA, algebraically free over kk (A, IV, p. 4), and such that : a) AA is entire over the ring B=k[x1,...,xn]B = k[x_1, ..., x_n]; b) for every jj such that 0≤j≤m0 \leq j \leq m, the ideal pj∩Bp_j \cap B is generated by the xkx_k with 1≤k≤n−m+j1 \leq k \leq n - m + j.

According to the normalization lemma (V, § 3, No. 1, th. 1), there exists an integer n≥0n \geq 0, a finite sequence (x1,...,xn)(x_1, ..., x_n) of elements of AA algebraically free and an increasing sequence (h(j))0≤j≤m(h(j))_{0 \leq j \leq m} of integers ≤n\leq n such that the ideal pj∩Bp_j \cap B is equal to the prime ideal qjq_j of BB generated by the xkx_k with 1≤k≤h(j)1 \leq k \leq h(j), and that AA is entire over the ring BB. Let jj be an integer such that 0≤j<m0 \leq j < m. By passing to quotients, one deduces from the canonical injection of BB into AA an injective homomorphism of B/qjB/q_j into A/pjA/p_j which makes A/pjA/p_j a finite (B/qj)(B/q_j)-algebra. Since the ring B/qjB/q_j is isomorphic to an algebra of polynomials in n−h(j)n - h(j) indeterminates over kk, it is integrally closed (V, § 1, No. 3, Cor. 3 of Prop. 13). According to th. 2 of No. 3, one therefore has

1=ht(pj+1/pj)=ht(qj+1/qj)≥h(j+1)−h(j). 1 = \mathrm{ht}(p_{j+1}/p_j) = \mathrm{ht}(q_{j+1}/q_j) \geq h(j+1) - h(j) .

It follows that one has h(j+1)≤h(j)+1h(j+1) \leq h(j) + 1 and qj+1≠qjq_{j+1} \neq q_j, whence h(j+1)=h(j)+1h(j+1) = h(j) + 1. But one has h(m)=nh(m) = n since qmq_m is maximal (V, § 2, No. 1, Prop. 1), whence finally h(j)=n−m+jh(j) = n - m + j.

Theorem 3

Let AA be a kk-algebra of finite type. a) For every minimal prime ideal pp of AA, all maximal chains of prime ideals of AA with origin pp have length the integer deg⁡.trkκ(p)\deg.\mathrm{tr}_k \kappa(p). b) The ring AA is catenary and its dimension is the supremum of the integers deg⁡.trkκ(p)\deg.\mathrm{tr}_k \kappa(p), where pp runs through the minimal prime ideals of AA. c) If AA is integral, then all maximal chains of prime ideals of AA have the same length, and the dimension of AA is the degree of transcendence over kk of the field of fractions of AA.

Suppose AA is integral and consider a maximal chain p0⊂...⊂pmp_0 \subset ... \subset p_m of prime ideals of AA. We have p0=0p_0 = 0. We deduce then from Lemma 3 the existence of an injective homomorphism φ:k[X1,...,Xm]→A\varphi : k[X_1, ..., X_m] \to A of kk-algebras which makes AA a finite k[X1,...,Xm]k[X_1, ..., X_m]-algebra. Consequently, the degree of transcendence over kk of the field of fractions of AA is equal to mm, whence c). Assertion a) follows from assertion c) applied to the ring A/pA/p and assertion b) is an immediate consequence of a) and Prop. 5 of § 1, No. 2.

Corollary 1

Let nn be a positive integer. We have

dim⁡(k[X1,...,Xn])=n. \dim(k[X_1, ..., X_n]) = n .

For a finite type k-algebra A to have dimension n, it is necessary and sufficient that there exist an injective k-homomorphism φ:k[X1,...,Xn]→A\varphi : k[X_1, ..., X_n] \to A making A a finite algebra over k[X1,...,Xn]k[X_1, ..., X_n].

This follows from th. 3, from the normalization lemma (V, § 3, No. 1, th. 1) and from th. 1, a) of No. 3.

Corollary 2

Let A be an integral k-algebra of finite type. For every prime ideal p of A, we have

ht⁡(p)=dim⁡(Ap)=dim⁡(A)−dim⁡(A/p) \operatorname{ht}(p) = \dim(A_p) = \dim(A) - \dim(A/p)
=dim⁡(A)−deg⁡.tr⁡kκ(p). = \dim(A) - \deg.\operatorname{tr}_k\kappa(p).

In particular, we have ht⁡(m)=dim⁡(Am)=dim⁡(A)\operatorname{ht}(m) = \dim(A_m) = \dim(A) for every maximal ideal m of A.

This follows from th. 3 and Remark 4 of § 1, No. 3.

Corollary 3

Let A be a k-algebra of finite type and let f be an element of A which belongs to no minimal prime ideal of A (for example an element of A which is not a divisor of zero, cf. IV, § 1, No. 1, Cor. 3 to Prop. 2 and No. 3, Cor. 1 to Prop. 7). We have dim⁡(A)=dim⁡(Af)\dim(A) = \dim(A_f).

The mapping p↦pAfp \mapsto pA_f is a bijection of the set of minimal prime ideals of A onto the set of minimal prime ideals of AfA_f. Moreover the rings A/pA/p and Af/pAf=(A/p)fA_f/pA_f = (A/p)_f have the same field of fractions. It is therefore sufficient to apply th. 3, b).

Corollary 4

Let A be a k-algebra of finite type and let p be a prime ideal of A.

a) In order that p be maximal, it is necessary and sufficient that the field of fractions of A/pA/p be a finite extension of k.

b) Let f∈A−pf \in A - p; the ideal p is a maximal ideal of A if and only if pAfpA_f is a maximal ideal of AfA_f.

If p is a maximal ideal of A, then A/pA/p is a field, hence a ring of dimension 0; it is a finite type extension of k whose degree of transcendence is 0 (th. 3, c)), it is therefore a finite extension of k. Conversely, if the field of fractions of A/pA/p is a finite extension of k, we have dim⁡(A/p)=0\dim(A/p) = 0 hence p is maximal. Assertion b) follows from assertion a), taking into account that A/pA/p and Af/pAfA_f/pA_f have the same field of fractions.

Assertion a) of Cor. 4 is a form of the theorem of zeros (V, § 3, No. 3, Prop. 1).

Corollary 5

Let A be a k-algebra of finite type, p a prime ideal of A and (pi)i∈I(\mathfrak{p}_i)_{i \in I} the family of minimal prime ideals of A contained in p. We have:

dim⁡p(A)=sup⁡i∈Idim⁡(A/pi) \dim_p(A) = \sup_{i \in I} \dim(A/\mathfrak{p}_i)
=dim⁡(Ap)+dim⁡(A/p) = \dim(A_p) + \dim(A/p)
=dim⁡(Ap)+deg⁡.tr⁡kκ(p). = \dim(A_p) + \deg.\operatorname{tr}_k\kappa(p).

We have dim⁡p(A)=sup⁡i∈Idim⁡p/pi(A/pi)\dim_p(A) = \sup_{i \in I} \dim_{p/p_i}(A/p_i) (§ 1, No. 3, Prop. 6). But, according to Cor. 3, we have dim⁡p/pi(A/pi)=dim⁡(A/pi)\dim_{p/p_i}(A/p_i) = \dim(A/p_i), whence the first equality. According to Cor. 2, we have dim⁡(A/pi)=dim⁡((A/pi)p)+dim⁡(A/p)\dim(A/p_i) = \dim((A/p_i)_p) + \dim(A/p). The second equality of the corollary therefore follows from the fact that dim⁡(Ap)=sup⁡i∈Idim⁡((A/pi)p)\dim(A_p) = \sup_{i \in I} \dim((A/p_i)_p) and the third from th. 3.

Thus dim⁡p(A)\dim_p(A) is the supremum of the lengths of the chains of prime ideals of AA of which pp is an element.

Corollary 6

Let A be a finitely generated k-algebra not reduced to 0, and let n be an integer ⩾0\geqslant 0. The following conditions are equivalent:

a) For every p∈Ass⁡(A)p \in \operatorname{Ass}(A), one has dim⁡(A/p)=n\dim(A/p) = n.

b) Every associated prime ideal of A is minimal and all the irreducible components of Spec(A) have dimension n.

c) There exists an injective k-homomorphism φ:k[X1,...,Xn]→A\varphi : k[X_1, ..., X_n] \to A making A a k[X1,...,Xn]k[X_1, ..., X_n]-module of finite type without torsion.

The equivalence of a) and b) is immediate. Let us show that a) implies c). According to b), the ring A has dimension n and there therefore exists (cor. 1) an injective k-homomorphism φ:k[X1,...,Xn]→A\varphi : k[X_1, ..., X_n] \to A making A a k[X1,...,Xn]k[X_1, ..., X_n]-module of finite type. For every prime ideal p∈Ass⁡(A)p \in \operatorname{Ass}(A), the ring A/pA/p is then integral over k[X1,...,Xn]k[X_1, ..., X_n], and we therefore have n=dim⁡(A/p)=dim⁡(k[X1,...,Xn]/φ−1(p))n = \dim(A/p) = \dim(k[X_1, ..., X_n]/\varphi^{-1}(p)) by th. 1, a) of no. 3, whence φ−1(p)=0\varphi^{-1}(p) = 0. The image by the injective homomorphism φ\varphi of a nonzero element of k[X1,...,Xn]k[X_1, ..., X_n] is not a zero divisor in A (IV, § 1, no. 1, cor. 3 to prop. 2), whence c).

Conversely, suppose that condition c) is satisfied. For every prime ideal p∈Ass⁡(A)p \in \operatorname{Ass}(A), the homomorphism k[X1,...,Xn]→A/pk[X_1, ..., X_n] \to A/p deduced from φ\varphi is injective (loc. cit.). We therefore have dim⁡(A/p)=n\dim(A/p) = n by cor. 1.

Remark 1

By cor. 5, the conditions a), b), c) of cor. 6 imply that one has dim⁡p(A)=dim⁡(A)\dim_p(A) = \dim(A) for every prime ideal pp of A.

Proposition 4

Let A and B be two finitely generated k-algebras and ρ:A→B\rho : A \to B an algebra homomorphism. Suppose that A is integral and that the A-module B is without torsion, and denote by K the field of fractions of A. One has

dim⁡(B)=dim⁡(A)+dim⁡(B⊗AK). \dim(B) = \dim(A) + \dim(B \otimes_A K).

Suppose first that B is integral. The algebra B⊗AKB \otimes_A K is then a ring of fractions of B defined by a multiplicative subset not containing 0. It therefore has as its field of fractions the field of fractions L of B. By th. 3, one has

dim⁡(B)=deg⁡.trkL,dim⁡(A)=deg⁡.trkK, \dim(B) = \deg.\mathrm{tr}_k L,\quad \dim(A) = \deg.\mathrm{tr}_k K,
dim⁡(B⊗AK)=deg⁡.trKL. \dim(B \otimes_A K) = \deg.\mathrm{tr}_K L.

Now, by the corollary of A, V, p. 106

deg⁡.trkL=deg⁡.trKL+deg⁡.trkK, \deg.\mathrm{tr}_k L = \deg.\mathrm{tr}_K L + \deg.\mathrm{tr}_k K,

whence the result in this case.

Let us pass to the general case. Every minimal prime ideal pp of BB is formed of zero divisors in BB, and is therefore above the ideal 0 of AA. It follows that the mapping p↦p.(B⊗AK)p \mapsto p . (B \otimes_A K) is a bijection from the set of minimal prime ideals of BB onto the set of minimal prime ideals of B⊗AKB \otimes_A K. The proposition therefore follows from the first part of the proof and from prop. 6, c) of § 1, no 3.

Corollary

Let ρ:A→B\rho : A \to B be an injective homomorphism of finitely generated k-algebras.* One has dim⁡(A)≤dim⁡(B)\dim(A) \leq \dim(B).

Indeed, let pp be a minimal prime ideal of AA such that dim⁡(A)=dim⁡(A/p)\dim(A) = \dim(A/p). There exists a prime ideal qq of BB above pp (II, § 2, no 6, prop. 16). By prop. 4 applied to A/pA/p and B/qB/q, one has dim⁡(A)=dim⁡(A/p)≤dim⁡(B/q)≤dim⁡(B)\dim(A) = \dim(A/p) \leq \dim(B/q) \leq \dim(B), whence the corollary.

Lemma 4. — Let AA and BB be two integral k-algebras, MM an AA-module without torsion, NN a BB-module without torsion. If the ring A⊗kBA \otimes_k B is integral, then M⊗kNM \otimes_k N is a module without torsion over A⊗kBA \otimes_k B.

Let KK (resp. LL) be the field of fractions of AA (resp. BB). There exists a set II (resp. JJ) such that MM (resp. NN) is isomorphic to a submodule of K(I)K^{(I)} (resp. L(J)L^{(J)}). The (A⊗kB)(A \otimes_k B)-module M⊗kNM \otimes_k N is therefore isomorphic to a submodule of K(I)⊗kL(J)K^{(I)} \otimes_k L^{(J)}, which is isomorphic to (K⊗kL)(I×J)(K \otimes_k L)^{(I \times J)}. Since K⊗kLK \otimes_k L is a ring of fractions of the integral ring A⊗kBA \otimes_k B, it is a torsion-free module over A⊗kBA \otimes_k B, whence the lemma.

*PROPOSITION 5. — Let k′k' be an extension of kk, AA a k-algebra of finite type and BB a k′k'-algebra of finite type.

a) The k′k'-algebra A⊗kBA \otimes_k B is of finite type and one has

dim⁡(A⊗kB)=dim⁡(A)+dim⁡(B). \dim(A \otimes_k B) = \dim(A) + \dim(B).

b) Let rr be a prime ideal of A⊗kBA \otimes_k B; let pp (resp. qq) denote the inverse image of rr in AA (resp. BB). One has

dim⁡r(A⊗kB)=dim⁡p(A)+dim⁡q(B). \dim_r(A \otimes_k B) = \dim_p(A) + \dim_q(B).

Put n=dim⁡(A)n = \dim(A) and m=dim⁡(B)m = \dim(B). By Cor. 1 to Theorem 3 there exist injective algebra homomorphisms φ:k[X1,...,Xn]→A\varphi : k[X_1, ..., X_n] \to A and ψ:k′[Y1,...,Ym]→B\psi : k'[Y_1, ..., Y_m] \to B making AA and BB respectively finite algebras over k[X1,...,Xn]k[X_1, ..., X_n] and k′[Y1,...,Ym]k'[Y_1, ..., Y_m]. The homomorphism φ⊗ψ\varphi \otimes \psi is then injective and makes A⊗kBA \otimes_k B into a finite algebra over the k′k'-algebra k[X1,...,Xn]⊗kk′[Y1,...,Ym]k[X_1, ..., X_n] \otimes_k k'[Y_1, ..., Y_m], which is identified with k′[X1,...,Xn,Y1,...,Ym]k'[X_1, ..., X_n, Y_1, ..., Y_m]. Therefore dim⁡(A⊗kB)=n+m\dim(A \otimes_k B) = n + m by Cor. 1 to Theorem 3, which proves a).

Let us remark that when AA and BB are integral, A⊗kBA \otimes_k B is a torsion-free k′[X1,...,Xn,Y1,...,Ym]k'[X_1, ..., X_n, Y_1, ..., Y_m]-module by Lemma 4, and that therefore one has

dim⁡r(A⊗kB)=n+m=dim⁡(A)+dim⁡(B) \dim_r(A \otimes_k B) = n + m = \dim(A) + \dim(B)
for every prime ideal rr of A⊗kBA \otimes_k B by Remark 1.

Prove now b). Let r0r_0 be a minimal prime ideal of A⊗kBA \otimes_k B contained in rr, and denote by p0p_0 (resp. q0q_0) the inverse image of r0r_0 in AA (resp. BB). The ring

(A \otimes_k B)/r_0 is isomorphic to a quotient of the ring (A/p_0) \otimes_k (B/q_0). We have therefore, by aa,

dim⁡((A⊗kB)/r0)≤dim⁡(A/p0)+dim⁡(B/q0). \dim((A \otimes_k B)/r_0) \leq \dim(A/p_0) + \dim(B/q_0).

Applying cor. 5 to th. 3, we deduce the inequality

dim⁡r(A⊗kB)≤dim⁡p(A)+dim⁡q(B). \dim_r(A \otimes_k B) \leq \dim_p(A) + \dim_q(B).

Conversely, let p0p_0 (resp. q0q_0) be a minimal prime ideal of A (resp. B) contained in pp (resp. q). According to the remark made above, we have

dim⁡(A/p0)+dim⁡(B/q0)=dim⁡r‾((A/p0)⊗k(B/q0)) \dim(A/p_0) + \dim(B/q_0) = \dim_{\overline{r}}((A/p_0) \otimes_k (B/q_0))

where r‾\overline{r} is the image of r by the canonical surjection A⊗kB→(A/p0)⊗k(B/q0)A \otimes_k B \to (A/p_0) \otimes_k (B/q_0). The second member of the preceding equality is obviously less than dim⁡r(A⊗kB)\dim_r(A \otimes_k B). Applying cor. 5 to th. 3, we deduce the inequality

dim⁡p(A)+dim⁡q(B)≤dim⁡r(A⊗kB), \dim_p(A) + \dim_q(B) \leq \dim_r(A \otimes_k B),

which completes the proof.

Corollary

Let A be a finitely generated k-algebra, k' an extension of k, and A' the k'-algebra A \otimes_k k'.

a) We have dim⁡(A′)=dim⁡(A)\dim(A') = \dim(A).

b) Let p' be a prime ideal of A' and p its inverse image in A; we have dim⁡p′(A′)=dim⁡p(A)\dim_{p'}(A') = \dim_p(A).

c) Let p' be a minimal prime ideal of A' and p its inverse image in A. Then p is minimal and we have dim⁡(A′/p′)=dim⁡(A/p)\dim(A'/p') = \dim(A/p).

The assertions a) and b) follow from prop. 5 by taking B=k′B = k'. Let us prove c). The ideal p is minimal (No. 1, prop. 1) and we have

dim⁡(A′/p′)=dim⁡p′(A′)=dim⁡p(A)=dim⁡(A/p). \dim(A'/p') = \dim_{p'}(A') = \dim_p(A) = \dim(A/p).

Remark 2

Suppose that the extension k' of k is radicial. Then the canonical mapping f:Spec(A′)→Spec(A)f : \mathrm{Spec}(A') \to \mathrm{Spec}(A) is a homeomorphism.

Let indeed p∈Spec(A)p \in \mathrm{Spec}(A). By Lemma 1 of No. 1, the space f−1({p})f^{-1}(\{p\}) is homeomorphic to Spec(κ(p)⊗kk′)\mathrm{Spec}(\kappa(p) \otimes_k k'). Now the set a of nilpotent elements of κ(p)⊗kk′\kappa(p) \otimes_k k' is a prime ideal (A, V, p. 134, corollary) and the quotient ring (κ(p)⊗kk′)/a(\kappa(p) \otimes_k k')/a, integral and finite over κ(p)\kappa(p), is a field (A, V, p. 16, cor. 1 and p. 10, prop. 1). Consequently f−1({p})f^{-1}(\{p\}) is reduced to one element. It follows that the mapping ff is an increasing bijection of Spec(A′)\mathrm{Spec}(A') onto Spec(A)\mathrm{Spec}(A), these two sets being ordered by inclusion of the prime ideals, and hence induces a bijection between the irreducible closed subsets of Spec(A)\mathrm{Spec}(A) and those of Spec(A′)\mathrm{Spec}(A'). Since the closed subsets of Spec(A)\mathrm{Spec}(A) (resp. Spec(A′)\mathrm{Spec}(A')) are finite unions of irreducible closed subsets, ff is a homeomorphism.

For a generalization, see exerc. 24, p. 98.

Exercises

See the exercises for § 2.

Machine translation of the English, by gpt-5-6-mini, gpt-5.4, against glossary version 34. Not checked by a person. That model is a cut down version of the one the rest of this language was written by.