10.107 Epimorphisms of rings
In any category there is a notion of an epimorphism. Some of this material is taken from [Autour] and [Mazet].
Lemma 10.107.1. Let R \to S be a ring map. The following are equivalent
R \to S is an epimorphism,
the two ring maps S \to S \otimes _ R S are equal,
either of the ring maps S \to S \otimes _ R S is an isomorphism, and
the ring map S \otimes _ R S \to S is an isomorphism.
Proof.
Omitted.
\square
Lemma 10.107.2. The composition of two epimorphisms of rings is an epimorphism.
Proof.
Omitted. Hint: This is true in any category.
\square
Lemma 10.107.3. If R \to S is an epimorphism of rings and R \to R' is any ring map, then R' \to R' \otimes _ R S is an epimorphism.
Proof.
Omitted. Hint: True in any category with pushouts.
\square
Lemma 10.107.4. If A \to B \to C are ring maps and A \to C is an epimorphism, so is B \to C.
Proof.
Omitted. Hint: This is true in any category.
\square
This means in particular, that if R \to S is an epimorphism with image \overline{R} \subset S, then \overline{R} \to S is an epimorphism. Hence while proving results for epimorphisms we may often assume the map is injective. The following lemma means in particular that every localization is an epimorphism.
Lemma 10.107.5. Let R \to S be a ring map. The following are equivalent:
R \to S is an epimorphism, and
R_{\mathfrak p} \to S_{\mathfrak p} is an epimorphism for each prime \mathfrak p of R.
Proof.
Since S_{\mathfrak p} = R_{\mathfrak p} \otimes _ R S (see Lemma 10.12.15) we see that (1) implies (2) by Lemma 10.107.3. Conversely, assume that (2) holds. Let a, b : S \to A be two ring maps from S to a ring A equalizing the map R \to S. By assumption we see that for every prime \mathfrak p of R the induced maps a_{\mathfrak p}, b_{\mathfrak p} : S_{\mathfrak p} \to A_{\mathfrak p} are the same. Hence a = b as A \subset \prod _{\mathfrak p} A_{\mathfrak p}, see Lemma 10.23.1.
\square
Lemma 10.107.6.slogan Let R \to S be a ring map. The following are equivalent
R \to S is an epimorphism and finite, and
R \to S is surjective.
Proof.
(This lemma seems to have been reproved many times in the literature, and has many different proofs.) It is clear that a surjective ring map is an epimorphism. Suppose that R \to S is a finite ring map such that S \otimes _ R S \to S is an isomorphism. Our goal is to show that R \to S is surjective. Assume S/R is not zero. The exact sequence R \to S \to S/R \to 0 leads to an exact sequence
R \otimes _ R S \to S \otimes _ R S \to S/R \otimes _ R S \to 0.
Our assumption implies that the first arrow is an isomorphism, hence we conclude that S/R \otimes _ R S = 0. Hence also S/R \otimes _ R S/R = 0. By Lemma 10.5.4 there exists a surjection of R-modules S/R \to R/I for some proper ideal I \subset R. Hence there exists a surjection S/R \otimes _ R S/R \to R/I \otimes _ R R/I = R/I \not= 0, contradiction.
\square
Lemma 10.107.7. A faithfully flat epimorphism is an isomorphism.
Proof.
This is clear from Lemma 10.107.1 part (3) as the map S \to S \otimes _ R S is the map R \to S tensored with S.
\square
Lemma 10.107.8. If k \to S is an epimorphism and k is a field, then S = k or S = 0.
Proof.
This is clear from the result of Lemma 10.107.7 (as any nonzero algebra over k is faithfully flat), or by arguing directly that R \to R \otimes _ k R cannot be surjective unless \dim _ k(R) \leq 1.
\square
Lemma 10.107.9. Let R \to S be an epimorphism of rings. Then
\mathop{\mathrm{Spec}}(S) \to \mathop{\mathrm{Spec}}(R) is injective, and
for \mathfrak q \subset S lying over \mathfrak p \subset R we have \kappa (\mathfrak p) = \kappa (\mathfrak q).
Proof.
Let \mathfrak p be a prime of R. The fibre of the map is the spectrum of the fibre ring S \otimes _ R \kappa (\mathfrak p). By Lemma 10.107.3 the map \kappa (\mathfrak p) \to S \otimes _ R \kappa (\mathfrak p) is an epimorphism, and hence by Lemma 10.107.8 we have either S \otimes _ R \kappa (\mathfrak p) = 0 or S \otimes _ R \kappa (\mathfrak p) = \kappa (\mathfrak p) which proves (1) and (2).
\square
Lemma 10.107.10. Let R be a ring. Let M, N be R-modules. Let \{ x_ i\} _{i \in I} be a set of generators of M. Let \{ y_ j\} _{j \in J} be a set of generators of N. Let \{ m_ j\} _{j \in J} be a family of elements of M with m_ j = 0 for all but finitely many j. Then
\sum \nolimits _{j \in J} m_ j \otimes y_ j = 0 \text{ in } M \otimes _ R N
is equivalent to the following: There exist a_{i, j} \in R with a_{i, j} = 0 for all but finitely many pairs (i, j) such that
\begin{align*} m_ j & = \sum \nolimits _{i \in I} a_{i, j} x_ i \quad \text{for all } j \in J, \\ 0 & = \sum \nolimits _{j \in J} a_{i, j} y_ j \quad \text{for all } i \in I. \end{align*}
Proof.
The sufficiency is immediate. Suppose that \sum _{j \in J} m_ j \otimes y_ j = 0. Consider the short exact sequence
0 \to K \to \bigoplus \nolimits _{j \in J} R \to N \to 0
where the jth basis vector of \bigoplus \nolimits _{j \in J} R maps to y_ j. Tensor this with M to get the exact sequence
K \otimes _ R M \to \bigoplus \nolimits _{j \in J} M \to N \otimes _ R M \to 0.
The assumption implies that there exist elements k_ i \in K such that \sum k_ i \otimes x_ i maps to the element (m_ j)_{j \in J} of the middle. Writing k_ i = (a_{i, j})_{j \in J} and we obtain what we want.
\square
Lemma 10.107.11. Let \varphi : R \to S be a ring map. Let g \in S. The following are equivalent:
g \otimes 1 = 1 \otimes g in S \otimes _ R S, and
there exist n \geq 0 and elements y_ i, z_ j \in S and x_{i, j} \in R for 1 \leq i, j \leq n such that
g = \sum _{i, j \leq n} x_{i, j} y_ i z_ j,
for each j we have \sum x_{i, j}y_ i \in \varphi (R), and
for each i we have \sum x_{i, j}z_ j \in \varphi (R).
Proof.
It is clear that (2) implies (1). Conversely, suppose that g \otimes 1 = 1 \otimes g. Choose generators \{ s_ i\} _{i \in I} of S as an R-module with 0, 1 \in I and s_0 = 1 and s_1 = g. Apply Lemma 10.107.10 to the relation g \otimes s_0 + (-1) \otimes s_1 = 0. We see that there exist a_{i, j} \in R such that g = \sum _ i a_{i, 0} s_ i, -1 = \sum _ i a_{i, 1} s_ i, and for j \not= 0, 1 we have 0 = \sum _ i a_{i, j} s_ i, and moreover for all i we have \sum _ j a_{i, j}s_ j = 0. Then we have
\sum \nolimits _{i, j \not= 0} a_{i, j} s_ i s_ j = -g + a_{0, 0}
and for each j \not= 0 we have \sum _{i \not= 0} a_{i, j}s_ i \in R. This proves that -g + a_{0, 0} can be written as in (2). It follows that g can be written as in (2). Details omitted. Hint: Show that the set of elements of S which have an expression as in (2) form an R-subalgebra of S.
\square
Lemma 10.107.13. Let R \to S be an epimorphism of rings. Then the cardinality of S is at most the cardinality of R. In a formula: |S| \leq |R|.
Proof.
The condition that R \to S is an epimorphism means that each g \in S satisfies g \otimes 1 = 1 \otimes g, see Lemma 10.107.1. We are going to use the notation introduced in Remark 10.107.12. Suppose that g, g' \in S and suppose that (P, U, V) is an n-triple which is associated to both g and g'. Then we claim that g = g'. Namely, write (P, U, V) = (X, YX, XZ) for a matrix factorization (g) = YXZ of g and write (P, U, V) = (X', Y'X', X'Z') for a matrix factorization (g') = Y'X'Z' of g'. Then we see that
(g) = YXZ = UZ = Y'X'Z = Y'PZ = Y'XZ = Y'V = Y'X'Z' = (g')
and hence g = g'. This implies that the cardinality of S is bounded by the number of possible triples, which has cardinality at most \sup _{n \in \mathbf{N}} |R|^ n. If R is infinite then this is at most |R|, see [Ch. I, 10.13, Kunen].
If R is a finite ring then the argument above only proves that S is at worst countable. In fact in this case R is Artinian and the map R \to S is surjective. We omit the proof of this case.
\square
Lemma 10.107.14. Let R \to S be an epimorphism of rings. Let N_1, N_2 be S-modules. Then \mathop{\mathrm{Hom}}\nolimits _ S(N_1, N_2) = \mathop{\mathrm{Hom}}\nolimits _ R(N_1, N_2). In other words, the restriction functor \text{Mod}_ S \to \text{Mod}_ R is fully faithful.
Proof.
Let \varphi : N_1 \to N_2 be an R-linear map. For any x \in N_1 consider the map S \otimes _ R S \to N_2 defined by the rule g \otimes g' \mapsto g\varphi (g'x). Since both maps S \to S \otimes _ R S are isomorphisms (Lemma 10.107.1), we conclude that g \varphi (g'x) = gg'\varphi (x) = \varphi (gg' x). Thus \varphi is S-linear.
\square
Comments (0)