Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Solutions to Commutative Algebra with a View Toward

Algebraic Geometry∗
Steven V Sam

January 20, 2008

Contents
1 Roots of Commutative Algebra 1

2 Localization 5

3 Associated Primes and Primary Decomposition 8

4 Integral Dependence and the Nullstellensatz 12

6 Flat Families 13

7 Completions and Hensel’s Lemma 14

9 Fundamental Definitions of Dimension Theory 15

10 The Principal Ideal Theorem and Systems of Parameters 16

15 Gröbner Bases 17

16 Modules of Differentials 17

17 Regular Sequences and the Koszul Complex 17

1 Roots of Commutative Algebra


1. First assume that M is Noetherian, that is, every submodule of M is finitely generated. Let
M0 ⊆ M1 ⊆ · · · be an ascending chain of submodules of M . If each inclusion is strict, choose
fi ∈ Mi \ Mi−1 . Let N be the submodule generated by the fi . By hypothesis, it has a finite
generating set {g1 , . . . , gr }. Each gi is a finite linear combination of the fj , so actually a finite
generating set of N can be taken from a subset of the fi . But this contradicts that each inclusion
in the chain is strict.
Now assume that every ascending chain of submodules of M terminates, and let {Mα } be some
set of submodules of M . Pick M1 in this set. If M1 is not maximal, then some M2 strictly

by David Eisenbud

1
contains it. If M2 is not maximal, we can find M3 strictly containing it, and so on. Since this
gives an ascending chain, it must eventually terminate, so we can find a maximal submodule
with respect to inclusion.
Now suppose every collection of submodules of M has a maximal element. Let f1 , f2 , . . . , ∈ M
be a sequence of elements. Let Mi be the submodule generated by f1 , . . . , fi . Then there is a
maximal submodule in this collection, call it Mr . This is generated by f1 , . . . , fr , so for every
n > r, we can write fn = a1 f1 + · · · + ar fr for some ai ∈ R.
Finally, assume that given any sequence of f1 , f2 , · · · ∈ M , there is a number m for which n > m
implies the existence of an expression fn = a1 f1 +· · ·+am fm with ai ∈ R. Let N be a submodule
of M and suppose N is not finitely generated. Pick f1 ∈ N . Inductively, we can find fr that is
not in the submodule generated by f1 , . . . , fr−1 . But then this contradicts the assumption.
Hence the equivalence of the four statements.

2. Let R be a Noetherian ring. Suppose I ⊂ R is an ideal such that there are infinitely many prime
ideals containing I minimal with respect to inclusion. Among the collection of all such I, there
is one maximal with this property by (Ex. 1.1). Obviously I is not prime, so there exist f, g ∈
/I
such that f g ∈ I. Let P be a prime minimal over I. Then either f ∈ I or g ∈ P , so either P is
minimal over (I, f ) or (I, g). By assumption, there are only finitely many primes minimal over
(I, f ) and (I, g), which contradicts the existence of I.

3. If M is Noetherian, then it is immediate that submodules are Noetherian. Also, the images of
a generating set of a submodule N ⊆ M under the map M → M/M 0 is a generating set for
N/(N ∩ M 0 ), so M/M 0 is also Noetherian.
Now suppose that both M 0 and M/M 0 are Noetherian. Let N be a submodule of M . It is
enough to show that N is finitely generated, and both N ∩ M 0 and M 0 /(N ∩ M 0 ) are Noetherian,
so it really is enough to show that M is finitely generated. Let {f1 , . . . , fn } be a generating
set for M/M 0 . Every element of M/M 0 is of the form m + M 0 where m ∈ M , so write it as
r1 f1 + · · · + rn fn . Fix some coset representative mi for each fi . Then each fi is of the form
mi + M 0 , so there exist elements k, k1 , . . . , kn ∈ M 0 such that

m + k = r1 (m1 + k1 ) + · · · + rn (mn + kn ) .

If {g1 , . . . , gn0 } is a generating set for M 0 , then we have shown that m is contained in the module
generated by the set {m1 , . . . , mn , g1 , . . . , gn0 }, which means that M is contained in this module,
and hence finitely generated.

4. (1)⇒(2): Define R+ = R1 ⊕ R2 ⊕ · · · . Since R is Noetherian, every ideal is finitely generated.


In particular, R+ is an ideal, so is finitely generated. Also note that R/R+ ∼ = R0 . For any
element r ∈ R, we may write r = fi1 + · · · + fin where fij ∈ Rij . Then under the canonical map
R → R/R+ , every fij goes to 0 if j 6= 0, so r is mapped to the image of R0 in R/R+ . However,
note that R0 is disjoint from R+ , and is an abelian group, so if r1 , r2 ∈ R0 , then r1 − r2 ∈ R0 ,
so is not in the kernel of R → R/R+ , which means different elements of R0 are mapped to
different elements of R/R+ , which shows the isomorphism. Thus, R0 is a homomorphic image
of a Noetherian ring and is Noetherian.
(2)⇒(3): Let {x1 , . . . , xk } be a generating set for R+ . If the xi are not homogeneous, then write
each as a finite sum of homogeneous elements, and replace xi with its summands, so without
loss of generality, assume each xi is homogeneous. Let R0 be the R0 -algebra generated by this
set. Any element r ∈ R is the finite sum of homogeneous elements, so to show that R = R0 , it

2
is enough to show that every homogeneous element is contained in R0 . We do this by induction
on the degree of r. If deg r = 0, then r ∈ R0 , and r = rx01 · · · x0k , so r ∈ R0 . Now suppose all
homogeneous elements of degree < n are in R0 and let r be a homogeneous element of degree
n > 0. Then r ∈ Rn , so r ∈ R+ , which means we can write r = a1 x1 + · · · + ak xk where ai ∈ R.
We may write each ai as a sum of homogeneous elements. Let a0i denote the homogeneous part
of ai of degree n − deg xi (we don’t require that a0i 6= 0). Then write

r = a01 x1 + · · · + a0k xk + (a1 − a01 )x1 + · · · + (ak − a0k )xk .

However, r is homogeneous of degree n and each a0i xi is homogeneous of degree n, so the other
terms sum to 0 by definition of the a0i . Since deg xi > 0 for all i, we have that deg a0i < n, so
a0i ∈ R0 , which means r ∈ R0 . So {x1 , . . . , xk } is a generating set for R as an R0 -algebra, which
means it is finitely generated, as desired.
(3)⇒(1): This is exactly the statement of Corollary 1.3 in Eisenbud.

5. Suppose R ⊂ S is a subring which is a direct summand as an R-module and that S is Noetherian.


Then there exists a R-module homomorphism π : S → R which fixes every element of R. Then
R is the quotient of S by the kernel of π, which shows that R is Noetherian.

6. (a) Write A = xm mr n1 nr
1 · · · xr . If A > B = x1 · · · xr , then either deg B < deg A, or deg B = deg A
1

and mi − ni > 0 for the first index i for which mi 6= ni . For d ≥ 0, there are d+r−1

d
monomials of degree d. In particular, this is finite, so the set of all monomials of degree at
most deg A is finite. This set contains all monomials B such that B < A, so there are only
finitely many such B.
(b) Let p be invariant under Σ, and suppose that the initial term of p is kxm mr and
1 · · · xr
1

that m1 ≥ · · · ≥ mr does not hold. Let i be the smallest value such that mi−1 < mi .
Let σ ∈ Σ be the transposition that switches i and i + 1 and fixes all other elements of
mi−1 mi mr to B = kxm1 · · · xmi xmi−1 · · · xmr .
{1, . . . , r}. Then σ maps A = kxm1 · · · xi−1 xi · · · xr
1
1 i−1 i r
Note that deg A = deg B, and that the exponents match for xj where j = 1, . . . , i − 2.
However, mi > mi−1 , so B > A. Since p is invariant under σ, B is a monomial occurring
in p, so this contradicts that A is the leading term of p. Thus, the initial term must satisfy
m1 ≥ · · · ≥ mr .
(c) Let p =Pf1µ1 · · · frµr Every monomial of fi has degree i, so any monomial of p will have
degree ri=1 iµi . Let xm 1 · · · xr
1 mr be the leading term of p. Then m must be maximized
1
among all exponents of x1 of monomials in p. In particular, each monomial of p can be
constructed by choosing one monomial from each term fi and takingPtheir product. Every
fi contains a monomial with x1 occurring in it, so the maximal mi is rj=1 µj . Then among
those monomials of p with maximal m1 , we wish to maximize m2 . Again every fi has a
monomial with m2 occurring in it. However, in f1 , we have already chosen the monomial x1
µ1 times. In the polynomials f2 , . . . , fr , there are monomials containing the product
Pr x1 x2 ,
so the best one can do is adjust the chosen monomials in f2 , . . . , fr to get m2 = j=2 µj .
In general, once we have maximized m1 , and then m2 , . . . , and then mi−1 , we wish to
maximize mi . Since f1 , . . . , fi−1 contain only monomials of degree < i, there is no freedom
to change which monomials one chooses to maximize mi . But the polynomials fP i , . . . , fr
have monomials containing the product x1 x2 · · · xi , so the best one can do is mi = rj=i µj .
Finally, note that
Xr X r Xr r
X
mi = µi = iµi ,
i=1 i=1 j=i i=1

3
so in fact our algorithm produces a monomial of p, and it is the initial term.
(d) Let ϕ : Zr → Zr be defined by (µ1 , . . . , µr ) 7→ (m1 , . . . , mr ) with mi = rj=i µj . Note that
P
ϕ is an additive map because of commutativity of addition in Z, and ϕ preserves the action
of Z on Zr because of the distributive law, so ϕ is a homomorphism of Z-modules. Now
suppose that ϕ maps both (µ1 , . . . , µr ) and (µ01 , . . . , µ0r ) to (m1 , . . . , mr ). We show that
µ0i = µi for all i by induction on r − i. If r − i = 0, then i = r and we have mr = µr = µ0r
by definition. Now assume for all j < i, we have µ0r−j = µr−j . Then
r
X
mr−i = µr−i + µj .
j=r−i+1

By the induction hypothesis, this is equal to µr−i + rj=r−i+1 µ0j . But also by definition of
P
ϕ, we have
X r
mr−i = µ0r−i + µ0j ,
j=r−i+1

so combining the last two equalities gives µ0r−i = µr−i , so in fact ϕ is an injection, so is a
monomorphism.
Given a monomial xm mr
1 · · · xr , let µr = mr and for 0 < i < r, let µi = mi − mi+1 . We have
1

µi ≥ 0 by the requirement that m1 ≥ m2 ≥ · · · ≥ mr . Then the leading term of f1µ1 · · · frµr


µ0 µ0
mr . If the leading term of both f µ1 · · · f µr and f 1 · · · f r is xm1 · · · xmr , then by
is xm
1 ···x
1
r 1 r 1 r 1 r
(c), mi = rj=i µj = rj=i µ0j , which from above implies that µ0j = µj . Thus each monomial
P P
with m1 ≥ · · · ≥ mr is the leading term of a unique product of f1 , . . . , fr .
(e) Choose f ∈ S Σ with initial term c1 xm 1 · · · xr
1 mr with c ∈ k. From (b), m ≥ · · · ≥ m ,
1 1 r
so from (d), there is a unique product p1 = f1µ1 · · · frµr such that the initial term of p1 is
xm mr
1 · · · xr . Then f − c1 p1 is of strictly smaller degree than f , so repeat this process to find
1

a unique product p2 whose initial term is the initial term of f − c1 p1 up to multiplication by


a constant. By (a), there are a finite number of monomials less than c1 xm mr
1 · · · xr , so this
1

process ends and we have f = c1 p1 + · · · + cn pn for some n. This process implies uniqueness
because if f = c01 p01 + · · · + c0m p0m , with c0i 6= 0 then the leading term of the right side has to
be the leading term of one of the p0i because of (d), say p01 without loss of generality. Then
p01 = p1 , and of course c1 = c01 , so we have c2 p2 + · · · + cn pn = c02 p02 + · · · + c0m pm . Repeating
this reasoning shows that n = m and in fact after reordering of indices, we have ci pi = c0i p0i .
We conclude that f can be written uniquely as a polynomial in f1 , . . . , fr .

10. The first figure is represented by C[X, Y ]/((X 2 + Y 2 − 2Y ) ∩ (X 2 − 4Y )) and the second by
C[X, Y ]/((X − Y ) ∩ (X + Y )).

18. First note that (x, y)∩(z, w) = (xz, xw, yz, yw). Let H(d) be the Hilbert function of k[x, y, z, w]/(x, y)∩
(z, w). Then H(d) is the number of monomials of degree d of the form xa y d−a and z a wd−a , so
H(d) = 2d + 2. Let H 0 (d) be the Hilbert function of one projective line, i.e., H 0 (d) is the number
of monomials xa y d−a , which is d + 1.

21. (a) Since G(n) = F (n + 1) − F (n), we get F (n + 1) = G(n) + F (n), which can be expanded to

F (n + 1) = G(n) + G(n − 1) + F (n − 1),

and further to
F (n + 1) = G(n) + G(n − 1) + · · · + G(0) + F (0),

4
so write F (n) = F (0)+ n−1
P
i=0 G(i). If F (n) ∈ Q[n], then G(n) is the sum of two polynomials
from the definition of G(n) so is in Q[n]. Conversely, if G(n) ∈ Q[n], then F (n) is a
 rational polynomials from our last derivation, so F (n) ∈ Q[n]. If we define
finite sum of
Fk (n) = nk , Fk is a polynomial in n of degree k, and the Fk form a basis for Q[n] as a
Q-vector space. SoPif n−1
P
i=0 Fk (i) is a polynomial in n of degree k + 1, we can conclude
n−1
that the degree
P ofi  i=0 nG(i)  is of degree 1 more than deg G, so that deg F = 1 + deg G.
Note that n−1 i=0 k = k+1 since the right hand side counts the number of ways one may
choose subsets of size k + 1 from {1, . . . , n}, while the left hand side counts the same thing
if we consider each ki as the number of subsets of {1,  . . . , n} that must contain i + 1 and
n n(n − 1) · · · (n − k)
have k elements from {1, . . . , i} whenever i ≥ k. Since = is a
k+1 k!
polynomial in n of degree k + 1, we get the desired result.
(b) Choose F ∈ T . We show that F is generated by {Fk } by induction on deg F . If deg F = 0,
then F is an integer, and F0 = 1, so we’re done. Now assume the statement true for all
polynomials in T with degree < deg F . Define G(n) = F (n + 1) − F (n) for all n. Then
G ∈ Q[n], and since F has integral Pvalues for sufficiently large n, the same is true for
n−1
G. We can rewrite F (n) = F (0) + i=0 G(i). By the induction hypothesis, each G(i) is
generated by {Fk }, and thus we have that F (n) is generated by {Fk }. Finally, note that
{Fk } is linearly independent. For if we have

a0 F0 + · · · + am Fm = 0 , ai ∈ Z ,

then the left side is a polynomial of degree m, and the right side is a polynomial of degree
0, so am = 0. Repeating this argument shows that all ai = 0. Thus, {Fk } is a basis for T ,
and we conclude that T is a free abelian group.
(c) First note that for k > 1, we have that for
 
n n(n − 1) · · · (n − k + 1)
= ,
k k!
the constant term is 0 and the coefficient of n is
k−1
(−1)k−1 X (−1)k−1 k(k − 1) (−1)k−1
i= = .
k! k! 2 2(k − 2)!
i=1

We claim that the ascending chain of ideals given by

(F2 ) ⊆ (F2 , F3 ) ⊆ · · · ⊆ (F2 , . . . , Fk ) ⊆ · · ·

is strictly increasing. Suppose that there exists polynomials a2 , . . . , ak−1 ∈ T such that

a2 F2 + · · · + ak−1 Fk−1 = Fk .

Let ci be the constant term of ai . Since the constant coefficient of each Fj is 0, we get that
i−1 c
the coefficient of n of ai Fi is (−1) i
2(i−2)! , so focusing on equality of the coefficient of n of both
sides, we need
c2 (−1)k−2 ck−1 (−1)k−1
+ ··· + = .
2 2(k − 3)! 2(k − 2)!
But since each ai is a unique linear combination of Fd , . . . , F0 , we have ci ∈ Z, so the above
equality cannot be satisfied since the sum on the left side will combine to a fraction with

5
denominator less than 2(k − 2)! when reduced. Thus, Fk is not in the ideal (F2 , . . . , Fk−1 ),
so we have an infinite ascending chain of ideals, which means T is not Noetherian. Since
Z is a Noetherian ring, this implies by Corollary 1.3 that T is not a finitely generated
Z-algebra.

22. Let M be a finitely generated R-module. Then we can find a surjection Rm → M for some m.
The kernel of this map is a submodule of Rm and hence also free, say Rn by the structure theorem
of finitely generated modules over a PID. Hence we have the following finite free resolution of
M:
0 / Rm / Rn /M /0.

2 Localization
4. A map Z/n → Z/m is defined by the image of 1, call it r. Then nr = 0, which means that
m divides nr. The number of such r modulo m is k = gcd(m, n), so HomZ (Z/n, Z, m) = Z/k
generated by the map 1 7→ m/k.
A k[x]-module homomorphism k[x]/(xn ) → k[x]/(xm ) is also determined by the image of 1, call
it ar xr + · · · + a0 with r < m. Then xn 7→ ar xr+n + · · · + a0 xn , which must be divisible by xm .
In the case n ≥ m any such polynomial will work. If n < m, then we must have ai = 0 for
i = 0, . . . , m − n − 1. So we have
(
k[x]/(xm ), n ≥ m
Homk[x] (k[x]/(xn ), k[x]/(xm )) = = k[x]/(xmax(m,n) ).
k[x]/(xn ) n < m

As for (b), we have Z/n ⊗Z Z/m ∼ = Z/(m, n) = Z/ gcd(m, n) and k[x]/(xn ) ⊗k[x] k[x]/(xm ) ∼
=
k[x]/(xm , xn ). Finally, for (c), we have k[x] ⊗k k[x] ∼
= k[x, y].

6. (a) An element r ∈ R is in ker ϕ if and only if r maps toT 0 under each projection R → R/Qi ,
which happens if and only if r ∈ Qi for all i, or r ∈ i Qi .
(b) Let m be any maximal ideal of R. If Qi is not contained in m, then m does not contain
the annihilator of R/Qi , so m ∈ / Supp(R/Qi ). Note that R/Qi is a finitely generated R-
module because the projection of 1 generates it, so (R/Qi )m = 0. Suppose that Qi , Qj are
contained in m and i 6= j. Since m is closed under addition, m contains all sums qi + qj
where qi ∈ Qi and qj ∈ Qj . However, Qi + Qj = R implies R Q ⊆ m, contradicting
Q its
maximality. So at most one Qi is contained in m, which means ( i R/Qi )m = i (R/Qi )m
(localization preserves direct sums, and finite direct products are directQsums) is a product
of zero rings except maybe at one entry. Then the map ϕm : Rm → ( i R/Qi )m is either
zero or the localized projection map Rm → (R/Qi )m , both of which are surjective. Thus,
ϕ isTlocally surjective for all maximal ideals, so must be surjective. We conclude that
R/( i Qi ) ∼
Q
= i R/Qi .

13. Let β be the map B → C in (∗).

(a) If (∗) is split, then there exists τ : C → B such that βτ is the identity on C. Then
given ϕ ∈ HomR (C, C), we have τ ϕ is a map C → B, so τ ϕ ∈ HomR (C, B). Then the
induced map by β sends τ ϕ to βτ ϕ = ϕ, so it is surjective. The induced map is clearly a
homomorphism, so it is an epimorphism.

6
Conversely, if the map HomR (C, B) → HomR (C, C) induced by β is an epimorphism, then
there exists τ ∈ HomR (C, B) that gets mapped to the identity map in HomR (C, C). In
particular, this means that βτ is the identity map on C, so (∗) is split.
(b) Let P be a maximal ideal of R. Then

0 → AP → B P → C P → 0

is exact and split by assumption, which by (a) means that the induced map HomRP (CP , BP ) →
HomRP (CP , CP ) given by composition with βP is surjective. Since C is finitely pre-
sented, there are isomorphisms HomRP (CP , BP ) ∼ = HomR (C, B)P and HomRP (CP , CP ) ∼ =
HomR (C, C)P , which induces an epimorphism HomR (C, B)P → HomR (C, C)P given by
composition by βP (that this is the map that is induced by these isomorphisms follows
from Lemma 2.4 and Proposition 2.10). Then composition by β is locally an epimorphism
for all maximal ideals, which means it is an epimorphism HomR (C, B) → HomR (C, C), so
by (a), we have that (∗) is split.

15. Write R = · · · R−1 ⊕ R0 ⊕ R1 ⊕ · · · .

(a) For nonzero r ∈ rad I, we can write r = f1 + · · · + fm where each fi is homogeneous,


nonzero, and has degree di where d1 < d2 < · · · < dm since R is a direct sum of the Ri .
For some n, (f1 + · · · + fm )n ∈ I. Expanding the product, fm n is a summand with degree

ndm since R is a Z-graded ring, so fm ∈ Rdm implies fm n ∈ R


ndm . Since every other term
in the product has degree < ndm , we conclude that fm n ∈ I because I is generated by

homogeneous elements, and one cannot generate (f1 + · · · + fm )n as a sum of homogeneous


elements of degree 6= ndm . Thus fm ∈ rad I, and so is f1 + · · · + fm−1 . Applying the same
reasoning shows that fm−1 ∈ rad I, etc., so r is generated by homogeneous elements in
rad I. Since r was arbitrary, we conclude that rad I is generated by homogeneous elements
and thus is homogeneous. Of course, r = 0 is trivial, since any set of generators for an ideal
will generate 0.
(b) For f ∈ (I : J), we can write f = f1 + · · · + fm where each fi is homogeneous and
deg f1 < · · · < deg fm . There exists j ∈ J such that f j = r for some r ∈ I. Write
j = j1 + · · · + jm0 and r = r1 + · · · + rn where ji , ri are homogeneous, ji ∈ J and ri ∈ I
since J, I are homogeneous ideals, and deg j1 < · · · < deg jm0 and deg r1 < · · · < deg rn .
Then we have (f1 + · · · + fm )(j1 + · · · + jm0 ) = r1 + · · · + rn , and the highest degree term on
the left side is fm jm0 and the highest degree term on the right side is rn , so they must be
equal because R is a direct sum of the Ri . Then fm ∈ (I : J) by definition, which implies
f1 + · · · + fm−1 is also in (I : J). Applying the same reasoning gives fm−1 ∈ (I : J), etc., so
all fi ∈ (I : J), which means every element of (I : J) is generated by homogeneous elements
in (I : J), and thus (I : J) is homogeneous.
(c) Let f, g ∈ R such that f g = r ∈ I. Then write f = f1 + · · · + fm , g = g1 + · · · + gm0 ,
and r = r1 + · · · + rn where fi , gi , ri are homogeneous elements, ri ∈ I, and we have
deg f1 < · · · < deg fm and similarly for the gi , ri . Then the highest degree term in f g is
fm gm0 , which must equal the highest degree term of r, which is rn because R is a direct
sum of the Ri . Thus fm gm0 ∈ I, which means either fm ∈ I or gm0 ∈ I. If it’s the former,
then fm g ∈ I, so (f − fm )g = r − fm g ∈ I, so we repeat the same reasoning with the
product (f1 + · · · + fm−1 )(g1 + · · · + gm0 ). Something similar happens if instead gm0 ∈ I.
At each step, we either conclude that the highest degree term fi ∈ I, or the highest degree
term gi ∈ I, but eventually we will be able to conclude that either f1 (g1 + · · · + gk ) ∈ I or

7
(f1 + · · · + fk )g1 ∈ I for some k. These may give that f1 ∈ I or g1 ∈ I, or further reduce
to concluding that f1 g1 ∈ I, so either all fi ∈ I or all gi ∈ I, which means either f ∈ I or
g ∈ I, so I is prime.
18. If R is a nonzero ring and has no nonzero homogeneous prime ideals, then R0 has no nonzero
prime ideals. So R0 is a field. We claim that dimR0 R1 ≤ 1. Suppose otherwise, and choose
linearly independent x, y ∈ R1 . Then x + y is a unit, else it generates a homogeneous prime
ideal, which means that (x + y)−1 has degree −1. But x/(x + y) is not an element of R0 . If
it were equal to say a, then a−1 x and (x + y)−1 are multiplicative inverses, which means that
a−1 x = x + y, contradicting that x and y are linearly independent.
If dimR0 R1 = 0, then R = R0 , and if dimR0 R1 = 1, then R = R0 [x, x−1 ].
26. If R = R1 × · · · × Rm where each Ri 6= 0 and m > 1, then the element (11 , . . . , 1m ) where
each 1i is the multiplicative identity of Ri is the multiplicative identity of R1 × · · · × Rm , and
of course (01 , . . . , 0m ) is the additive identity where 0i is the additive identity of Ri . However,
(11 , 02 , . . . , 0m ) is not equal to either element since m > 1 and Ri 6= 0, so is nontrivial, but is
idempotent.
Conversely, let e be a nontrivial idempotent of R. Then Re × R(1 − e) is a subset of R × R closed
under addition and multiplication since (re, s(1 − e)) + (r0 e, s0 (1 − e)) = ((r + r0 )e, (s + s0 )(1 − e))
and (re, s(1−e))(r0 e, s0 (1−e)) = (rr0 e2 , ss0 (1−e)2 ), and e2 = e, and (1−e)2 = 1−2e+e2 = 1−e.
Thus Re × R(1 − e) inherits associativity and commutativity from R × R. It has multiplicative
identity (e, 1 − e) and additive identity (0, 0) with (−re, −s(1 − e)) being the additive inverse of
(re, s(1−e), so Re×R(1−e) is a ring. Define a map ϕ : R → Re×R(1−e) by r 7→ (re, r(1−e)). It
is a homomorphism because ϕ(r + s) = ((r + s)e, (r + s)(1 − e)) = (re, r(1 − e)) + (se, s(1 − e)) =
ϕ(r) + ϕ(s), ϕ(rs) = (rse, rs(1 − e)) = (re, r(1 − e))(se, s(1 − e)) = ϕ(r)ϕ(s), and finally,
ϕ(1) = (e, 1 − e). To see it is injective, suppose (re, r(1 − e)) = (0, 0). Then re = 0 and
r − re = 0, which combine to say r = 0. For surjectivity, pick (re, s(1 − e)) ∈ Re × R(1 − e).
Then
ϕ(s + re − se) = (se + re2 − se, s(1 − e) + re(1 − e) − se(1 − e))
= (re, s − se + re − re2 − se + se2 )
= (re, s(1 − e)).
Thus ϕ is an isomorphism, so R = Re × R(1 − e). Also since e 6= 0 and e 6= 1, both Re and
R(1 − e) are nonzero rings.
Now define ϕ : Re → R[e−1 ] by re 7→ re 1 . Note that the multiplicative identity of Re is e
and the identity for R[e−1 ] is 11 . The multiplicatively closed subset of R containing products
of {e} is {1, e}. Given re, se ∈ Re, we have ϕ(rese) = ϕ(rse) = rse re se
1 = 1 1 = ϕ(re)ϕ(se),
re+se re se e
ϕ(re + se) = 1 = 1 + 1 = ϕ(re) + ϕ(se), and ϕ(e) = 1 . But we have e(e − 1) = e − e = 0,
so 1e = 11 , and ϕ is a homomorphism. For injectivity, if re 0
1 = 1 , then either re − 0 = 0 or
e(re − 0) = 0, both of which give re = 0. For surjectivity, all elements are of the form re or 1r .
Then e(re2 − r) = 0 and e(re − r) = 0, so re r re r
1 = e and 1 = 1 . Thus ϕ is an isomorphism, and
we conclude that Re may be realized as the localization R[e−1 ].

3 Associated Primes and Primary Decomposition


2. Since M 0 = M1 ∩M2 is the kernel of the projection M → M/M1 ⊕M/M2 , M/M 0 is isomorphic to
some submodule of M/M1 ⊕M/M2 , call it N1 ⊕N2 where N1 ⊆ M/M1 and N2 ⊆ M/M2 . Then we

8
know that Ass M/M 0 ⊆ Ass N1 ∪ Ass N2 . It is clear from the definition that Ass N1 ⊆ Ass M/M1
and Ass N2 ⊆ Ass M/M2 , so we get the desired statement Ass M/M 0 ⊆ Ass M/M1 ∪ Ass M/M2 .

3. For any prime P ⊆ R, P ∈ AssR HomR (M, N ) if and only if PP ∈ AssRP (HomR (M, N ))P , and
similar comments apply to Supp M ∩ Ass N . Note that the preimage of a prime in RP is a
prime of R not meeting R − P , so is contained in P . Since the preimage of the localization
map preserves inclusions, any maximal ideal of RP must be contained in PP , so RP is local with
maximal ideal PP . Without loss of generality, we assume that R is a local ring with maximal
ideal P .
If P ∈ Supp M , then P contains annR M , so RM 6= 0, which means M 6= 0. By Nakayama’s
Lemma, this means P M 6= M , so M/P M 6= 0. Moreover, M/P M is a module over R/P . Since
P is maximal, R/P is a field, so M/P M is a nonzero vector space, which means there is a
surjective projection M/P M → R/P . Composing this with the projection M → M/P M gives
a surjection M → R/P . If P ∈ Ass N , then there is a bijection between R/P and a submodule
of N . In particular, there is an injection R/P → N . So if P ∈ Supp M ∩ Ass N , compose the
above surjection with the above injection to get a map ϕ : M → N . It is clear that P annihilates
ϕ. Also, R/P 6= 0, so ϕ 6= 0. Thus, the annihilator of ϕ is a proper ideal containing P , so must
be P , and this gives P ∈ Ass HomR (M, N ).
Conversely, if P ∈ Ass HomR (M, N ), then there exists ϕ ∈ HomR (M, N ) with ϕ 6= 0, so M 6= 0
and N 6= 0. Thus, annR M is a proper ideal of R, so P ⊇ annR M implies P ∈ Supp M . Since
P annihilates ϕ, it annihilates ϕ(M ) ⊆ N . Since ϕ 6= 0, ϕ(M ) 6= 0, so there is some nonzero
n ∈ N annihilated by P . Since n 6= 0, the annihilator of n is a proper ideal of R containing P ,
so must be P by maximality, and P ∈ Ass N . Therefore, Ass HomR (M, N ) = Supp M ∩ Ass N .
We define a map ψ : HomR (R/I, N ) → (0 :N I) by ϕ 7→ ϕ(1), where by 1, we mean the
projection of 1 ∈ R onto R/I. To see this is well-defined, note that ϕ lifted to a map R → N
will contain I in its kernel, so for any r ∈ I, we have rϕ(1) = ϕ(r) = 0, and so ϕ(1) ∈ (0 :N I).
Every map is determined by where it sends 1, so ψ is injective. Given any n ∈ (0 :N I), we can
define a map ϕ0 : R → N by 1 7→ n, and for all r ∈ I, ϕ0 (r) = rϕ0 (1) = rn = 0, so I ⊆ ker ϕ0 ,
which means this maps factors as R → R/I → R/ ker ϕ0 → N , where the first two maps are
projections. Thus, ψ is surjective. It is clear that ψ is a homomorphism of R-modules, so ψ is an
isomorphism and we get the equality HomR (R/I, N ) = (0 :N I). Any prime ideal in Supp R/I
must contain the annihilator of R/I, which is I, so Supp R/I is the set of prime ideals containing
I. Using the above results, we conclude that

Ass(0 :N I) = Ass N ∩ {P ⊆ R | P is a prime ideal and I ⊆ P }.

5. If I is homogeneous, then choose r ∈ I. We may write r = r1 + · · · + rn for some nonzero


homogeneous elements ri since R is a direct sum of the Rγ as abelian groups (n might be 0).
However, I is generated by homogeneous elements, so we may write r = f1 + · · · + fm where
the fi are part of a generating set of I. Since the sum of r is unique, we conclude that m = n
and up to reordering of indices, ri = fi , so ri ∈ I, and I contains the homogeneous components
of each of its elements. Conversely, if I contains the homogeneous components of each of its
elements, then of course each element can be generated by homogeneous elements of I, so I is
homogeneous.

(a) Let P = ann m be a prime ideal. Choose nonzero f ∈ P and write f = f1 + · · · + fs as a


unique sum of nonzero homogeneous elements of R with deg f1 < · · · < deg fs where deg fi
is γ ∈ Γ such that fi ∈ Rγ . To show P is homogeneous, it is enough to show that f1 ∈ P

9
since by induction that will give f2 ∈ P , etc. Now write m = m1 + · · · + mt as a unique
sum of nonzero homogeneous elements of M with deg m1 < · · · < deg mt where deg mi is
λ ∈ Λ such that mi ∈ Mλ . We induct on t to show that P is homogeneous. We know that
f m = 0, so f m = f1 m1 + (terms of higher degree) since the action of Γ on Λ is compatible
with the orders. Then f1 m1 = 0 since each fi mj is a homogeneous element. If t = 1, then
f1 ∈ P , so we’re done, and this takes care of the base case for the induction. Otherwise, we
know that f1 m = f1 m2 + · · · + f1 mt , so let I = ann f1 m. Then P ⊆ I because p ∈ P gives
p(f1 m) = f1 (pm) = f1 0 = 0. If P = I, then P is the annihilator for an element of M with
less than t terms, so is homogeneous by induction. Otherwise, there exists g ∈ I such that
g∈ / P , so gf1 m = 0, which implies gf1 ∈ P . Since P is prime, this means f1 ∈ P . Thus,
all homogeneous components of f are contained in P , which means P is homogeneous.
Let m = m1 + · · · + mt as above, and let h ∈ P be any homogeneous element. Then
0 = hm = hm1 + · · · + hmt , and each hmi has different degree, so they are all 0. Thus,
since any element f ∈ P is a sum of homogeneous elements, we conclude T f mi = 0 for
i = 1, . . . , t, so P m
Ti = 0. Then P ⊆ ann m i , so we have P = ann m ⊇ i (ann mi ) ⊇ P ,
which means P = i (ann mi ).QIf we pick elements T p i ∈ ann mi , then p1 · · · p t will annihilate
each mi by commutativity, so i (ann mi ) ⊆ i (ann mi ) = P . Now suppose that there does
not exist i so that ann mi ⊆ P . Then choose ni ∈ mi − P . But then n1 · · · nt ∈ P , and P is
prime, so one of the ni ∈ P , a contradiction. Thus, some ann mi ⊆ P , and we have equality
because P mi = 0 as seen above, and we have shown that P annihilates some homogeneous
element of M .
(b) If P ∈ Ass M , then (a) shows that P is homogeneous, so all ideals of Ass M are homoge-
neous.
Define a homogeneous submodule N ⊆ M to be homogeneously irreducible if it cannot
be written as a finite intersection of two properly larger homogeneous submodules. If
some homogeneous submodule is not homogeneously irreducible, we may take a maximal
such submodule, which will be written as the intersection of two properly larger homoge-
neous submodules. These two must be a finite intersection of homogeneously irreducible
submodules, a contradiction, so in particular, 0 can be written as a finite intersection of
homogeneous submodules. Suppose some homogeneously irreducible submodule N ⊆ M
is not primary. Then M/N has an induced grading from M (if M = ⊕λ∈Λ Mλ , then
M/N = ⊕λ∈Λ Mλ /(Mλ ∩ N ) gives the grading) and has two different associated primes P
and Q, each of which is homogeneous and annihilates homogeneous elements m1 and m2 ,
respectively, of M/N by (a). Thus, R/P ∼ = Rm1 and R/Q ∼ = Rm2 , both of which are
homogeneous submodules of M/N . The annihilator of every nonzero element of Rm1 is P
and the annihilator of every nonzero element of Rm2 is Q, so Rm1 ∩ Rm2 = 0, and thus 0 is
homogeneously reducible. Lifting these submodules to M gives homogeneous submodules
(since N is homogeneous, the homogeneous generators nj of Rmi can be lifted to M as
the sum nj + generators of N , but the lift of Rmi contains N , so will also be generated by
homogeneous elements and thus is homogeneous) that intersect in N , contradicting that
N is homogeneously irreducible. In particular, 0 is a homogeneous submodule, so has a
primary decomposition into homogeneous submodules.
Now we obtain a filtration

0 = M0 ⊆ M1 ⊆ · · · ⊆ M n = M

with each Mi homogeneous and Mi+1 /Mi ∼ = R/Pi for some homogeneous prime ideal Pi by
induction on i. Since R is Noetherian, there is a prime P0 ∈ Ass(M ), which is homogeneous

10
by (a), and annihilates a homogeneous element m1 . Thus, R/P0 ∼= Rm1 =: M1 , which is a
homogeneous submodule. In general, given Mi 6= M homogeneous, there is a prime Pi ∈
Ass(M/Mi ). Since M/Mi has an induced grading from M , Pi annihilates a homogeneous
element mi+1 ∈ M/Mi , so R/Pi ∼ = Rmi+1 =: Mi+1 , which is a homogeneous submodule
of M/Mi . Since Mi is homogeneous, the homogeneous generators nj of Mi+1 can be lifted
to M as the sum nj + generators of Mi , but the lift of Mi+1 contains Mi , so will also be
generated by homogeneous elements and thus is homogeneous. The process stops since M
is a Noetherian module, so some M = Mn for some n.
(c) Let M0 and M1 be the images of R0 and R1 , respectively, of the projection R → M . It
is clear that M0 + M1 = M . Suppose m0 ∈ R0 and m1 ∈ R1 have the same image in M 0 .
Then there is some polynomial p such that m0 − m1 = (x2 − y 2 )p. Since the terms of m0
all have even degree in y and the terms of m1 all have odd degree in y, we conclude that
none of the terms in m0 − m1 cancel, and that m0 is the product of (x2 − y 2 ) and the sum
of the terms of p with even degree in y, and m1 is the product of (x2 − y 2 ) and the sum
of the terms of p with odd degree in y, so both m0 and m1 are divisible by x2 − y 2 and
are thus mapped to 0, so M0 ∩ M1 = 0, and we have M = M0 ⊕ M1 . The fact that M is
graded by Γ follows from the fact that R is graded by Γ.
The ideal P = (x−y) is prime because x−y is an irreducible polynomial, and P annihilates
x + y ∈ M . Multiplying any monomial in M by x will not change its degree in y, but
multiplying a monomial in M by −y will, so multiplying x − y with any polynomial will
yield one with terms with odd and even degrees in y, so cannot be homogeneous. Since
every element in P is the product of x − y by a polynomial, P cannot be generated by
homogeneous elements, so is not homogeneous. Let ≤ be a total ordering on Z/(2) and
suppose that the action of Z/(2) on itself is compatible with ≤. Then the ordering is either
defined by 0 ≤ 1 or 1 ≤ 0. In the first case, we also have 1 ≤ 1, so 0 + 1 ≤ 1 + 1, which
means 1 ≤ 0, a contradiction. In the second case, we will get 1 + 1 ≤ 0 + 1, which means
0 ≤ 1, also a contradiction, so Z/(2) cannot be ordered so that the action of Z/(2) on itself
is compatible with the order.
We claim that the only associated primes of M are (x − y), which annihilates x + y, and
(x + y), which annihilates x − y. No other element divides x2 − y 2 , so if a nonzero element
m is annihilated by p in M , then mp ∈ (x2 − y 2 ) ⊆ R, so either m is divisible by x + y, in
which case its annihilator is (x − y), or m is divisible by x − y, in which case its annihilator
is (x + y), or x2 − y 2 ∈ annR m. In the last case, annR m must contain T x + y and x − y to
be prime, but no nonzero element of M is annihilated by both. Let ni=1 Mi be a primary
decomposition of 0 ⊆ M . Without loss of generality, this intersection is irredundant, so
M/Mi is either (x + y)-coprimary or (x − y)-coprimary. Then either (x + y) or (x − y)
is minimal over annR (M/Mi ), which contains Mi . Neither of (x + y) or (x − y) contains
homogeneous elements, so none of the Mi are homogeneous. Thus, 0 ⊆ M does not have a
primary decomposition into homogeneous submodules.

10. (a) Every element in R can be written in the form ca + p(b) for c ∈ k and p a polynomial in
b. To see why, note that the element am is equivalent to either 0 or a, and similarly, any
0
element am bm is equivalent to 0 for m, m0 > 0. The annihilator of R/(bn ) contains (bn ) and
is contained in (a, b). Then (a, b) is minimal among the primes over (bn ) because any prime
ideal containing (bn ) must contain b and must contain a since a2 = 0. The complement
of (a, b) consists of nonzero constants in a field k, so are nonzerodivisors. Thus, R/(bn ) is
(a, b)-coprimary, so (bn ) is (a, b)-primary in R. Similarly, (a) is (a)-primary because R/(a)

11
consists of polynomials in b, so has annihilator (a), which is prime, and the complement of
(a) contains no zerodivisors on R/(a). Finally, 0 is not a primary submodule of R. If it
were, then Ass R would consist of one element, but the annihilator of a is (a, b), and the
annihilator of b is (a), both of which are prime. Thus, a primary decomposition of 0 cannot
be the intersection of one ideal, so 0 = (a) ∩ (bn ) is a minimal primary decomposition.
(b) Note that (a + λbn ) contains bλ−1 (a + λbn ) = bn+1 . All elements of R are of the form
ca + p(b) with c ∈ k and p a polynomial in b, so (ca + p(b))(a + λbn+1 ) = λbn+1 p(b),
and thus a ∈ / (a + λbn ). Then (a + λbn ) is contained in the annihilator of R/(a + λbn ),
which is contained in (a, b). A prime minimal over (a + λbn ) must contain b because
bn+1 ∈ (a + λbn ) and must contain a since a2 = 0. Since (a, b) is prime, it is minimal over
(a + λbn ), and its complement contains no zerodivisors, so a + λbn is (a, b)-primary. We’ve
shown that a ∈ / (a + λbn ), so ca ∈ / (a + λbn ) for nonzero c. Since all nonzero elements of
(a) are of the form ca, we get 0 = (a) ∩ (a + λbn ). Any proper ideal I properly containing
(a + λbn ) contains an element of the form f = λa a + λ0 + λ1 b + · · · + λd bd . If d > n,
then multiply a + λbn by λ−1 λd bd−n and subtract it from f to get an element of lower
degree in I. If d ≤ n, then multiplying f by λ−1 d λb
n−d and subtracting a + λbn gives
−1 n−d n−1
λd λ(−a + λ0 b + · · · + λd−1 b ) ∈ I. We repeat these two steps as necessary to end
up with some multiple of nonzero a in I, so (a + λbn ) is maximal among ideals with zero
intersection with (a). Thus we get the following chain for R/(a + λbn ),

0 = (a + λbn ) $ (a, bn ) $ (a, bn−1 ) $ · · · $ (a, b) $ R/(a + λbn ) ,

which has length at least n, so the rings R/J where J is a maximal (a, b)-primary component
of 0 have arbitrarily large length.
(c) Note that (a) is prime and is the annihilator of S/(a) because the only elements annihilating
constants are multiples of a. All elements not in (a) are nonzerodivisors because c is an
algebraically independent element, and b only annihilates a in S, but a = 0 in S/(a), so (a)
is (a)-primary in S. A prime minimal over the annihilator of S/(acn−1 +λbn ) must contain a
since a2 = 0, and must contain b because λ−1 b(acn−1 +λbn ) = bn+1 , so bn+1 ∈ (acn−1 +λbn ).
Since constants in k and polynomials in c do not annihilate any elements of S/(acn−1 +λbn ),
neither do sums of these with other elements of S, so the complement of (a, b) in S contains
no zerodivisors, and we conclude that (acn−1 + λbn ) is (a, b)-primary, and is generated by
a homogeneous element of degree n. The product of any element in S with acn−1 + λbn
is not a multiple of a since multiplying it with a polynomial in S gives 0 or a polynomial
with terms with just b and c, so 0 = (a) ∩ (acn−1 + λbn ). If a homogeneous ideal I properly
contains (acn−1 + λbn ), then without loss of generality, let

f = ap−1 (c) + p0 (c) + bp1 (c) + · · · + bd pd (c)

where each pi (c) is a polynomial in c and f is homogeneous, and suppose I = (acn−1 +


λbn , f ). If d > n, then multiply acn−1 + λbn by λ−1 pd (c)bd−n and subtract from f to get a
polynomial of lower degree. If d ≤ n, then multiply f by bn−d and subtract λ−1 pd (c)(acn−1 +
λbn ) from it to be in the first case. We may repeat this process to show that some nonzero
multiple of a is in I. So I ∩ (a) 6= 0, and thus (acn−1 + λbn ) is maximal among homogeneous
ideals that can be used as primary components.
(d) There is a maximal degree d such that all nonzero homogeneous elements in (0 : (x1 , . . . , xr ))
have degree ≤ d. If not, then there are monomials m of arbitrarily large degree such that
xi m ∈ I for i = 1, . . . , r. Then at some degree d0 , all monomials of degree > d0 will be

12
divisible by these xi mj . Thus, polynomials in (0 : (x1 , . . . , xr )) of degree > d0 are contained
in I and hence are zero in R. Let J1 ⊇ I be a (x1 , . . . , xr )-primary homogeneous ideal.
If J1 does not contain some homogeneous polynomial of degree ≥ d, then let J be the
ideal generated by this homogeneous polynomial and J1 . Then (x1 , . . . , xr ) is a minimal
prime over the annihilator of R/J, and the elements in the complement of (x1 , . . . , xr ) are
nonhomogeneous polynomials, so their product with any other polynomial in R will be
nonhomogeneous, and hence will not lie in J, so the complement of (x1 , . . . , xr ) contains no
zerodivisors of R/J, and thus J is (x1 , . . . , xr )-primary. We conclude that if J1 is maximal
among the (x1 , . . . , xr )-primary homogeneous ideals containing I, then J1 must contain all
homogeneous polynomials of degree ≥ d. Thus, R/J1 consists of polynomials in R with
degree < d. Note that R/J1 is Noetherian since it is a homomorphic image of R, which is a
homomorphic image of k[x1 , . . . , xr ], which is Noetherian. Also, it is Artinian because for
any descending chain of ideals, the degree of the generators will increase, and eventually
become d, so will terminate. Thus, R/J1 has finite length. In particular, it has shorter
length than the homomorphic image of R that includes all polynomials with degree < d,
since R/J1 can be considered either a homomorphic image of it, or a submodule of it, so
we have an upper bound on the length of R/J1 that does not rely on our choice of J1 , as
d depends only on I.

4 Integral Dependence and the Nullstellensatz


20. (a) Given an integer n, let m be the square-free part of n such that n = s2 m. To see that
√ √
Z[ n] is a domain, note that it is a subring of C. Every element of Z[ n] is a polynomial
√ √ √ 2
in n, so can be written as a + b n where a, b ∈ Z√ since n ∈ Z. The field of√fractions
√ √
of Z[ n] is Q[ n] because given any fraction a+b √n , one can multiply by c−d√n to get
c+d n c−d n
1 √ √ √ √ √
c2 −d 2 (ac − bdn + (bc − ad) n) ∈ Q[ n]. Now note that Q[ n] = Q[s m] = Q[ m], so
√ √
Z[ n] and Z[ m] have the same field of fractions.
√ √ √
The element m is integral over Z[ n] using the polynomial x2 − m, so Z[ m] is an
√ √
integral extension of Z[ n]. If N is the integral closure of Z[ m], then we claim that N
√ √
is also the integral closure of Z[ n]. Let N 0 be the integral closure of Z[ n]. Clearly,
N 0 ⊆ N . For the other inclusion, it is enough to show that any element of N is integral

over Z[ n]. Let α ∈ N . Then there is a monic polynomial xn + an−1 xn−1 + · · · + a0 with
√ √
α as a root such that ai ∈ Z[ m]. The ring R = Z[ n][a0 , . . . , an−1 ] is a finitely generated

algebra over Z[ n] by integral elements, so is a finitely generated module by Corollary 4.5.
Also, R[α] is a finitely generated algebra over R by integral elements, so again is a finitely

generated module over R. This gives that R[α] is a finitely generated Z[ n]-module, so

R[α] is integral over Z[ n], as desired.

Thus, it is enough to study the square-free case Z[ m], since it has the same integral

closure as Z[ n].

(b) Following the above comments, the integral closure of Z in Q[ n] is the same as the integral
√ √
closure of Z[ n] in Q[ n]. Since x2 − Trace(α)x + Norm(α) divides any monic polynomial
that has α as a root, by Proposition 4.11, Trace(α) and Norm(α) are integral over Z.
However, Trace(α), Norm(α) ∈ Q, and Z is normal, so α ∈ R if and only if Trace(α) and
Norm(α) are integers.
(c) Since Norm(α) = 2a is an integer, it is immediate that a ∈ 21 Z.

13
p
If a = 0, then Norm(α) = −b2 n is an integer. Suppose b = q in reduced form. Then
2
−b2 n = − pq2n . Since n is square-free, this is an integer if and only if q = ±1, so b ∈ Z.
1 1 p
If a = 2 and α ∈ R, then Norm(α) = 2 − b2 n is an integer. Again, write b = q in reduced
p2 n q 2 −2p2 n 2
form. Then 21 − = q2
so 1 − 2q 2
, 2 pq2n
is an even integer. Again, q 2 does not divide
n since n is square-free, and q does not divide p, so we conclude that q divides 2, which
means that b ∈ 12 Z.

Finally, given α = 21 + b n ∈ R, we may assume that b = 0 or b = 21 by subtracting a

multiple of n from α and keeping in mind that integral closures are closed under addition.
However, the case b = 0 implies that 21 ∈ R, which contradicts that Z is normal.

(d) If α = a + b n and a = 0, then it is clear that α ∈ R. Otherwise, for α ∈ R, we have shown
in (c) that a ∈ 21 Z, so subtracting an appropriate integer amount, suppose that a = 12 since
√ √
Z ⊆ R. Then 12 +b n ∈ R if and only if 12 + 21 n ∈ R if and only ifNorm(α) = 41 − 14 n = 1−n
4
is an integer. This is equivalent to n ≡ 1 (mod 4). Thus, if n 6≡ 1 (mod 4), then α ∈ R if

and only if a ∈ Z. This translates to saying that the integral closure of Z[ n] is itself if
n 6≡ 1 (mod 4). On the other hand, if n ≡ 1 (mod 4), then the integral closure in this case
√ √ √ √
is Z[ 21 + 12 n] since 21 + 12 n, 1, and n are integral over Z[ n], and any other integral
element is a sum of these three elements as shown above.

6 Flat Families
1. (a) Let N be any R-module. LetL I be an indexing set with one element for each element of N
so that we have a surjection I R → N . Letting K be the kernel of this map, we have a
short exact sequence
/K / /N /0.
L
0 IR

The long exact sequence gives from this

/ TorR (M, N ) / M ⊗R K / / M ⊗R N /0,


L
0 1 I M

so that TorR
L
1 (M, N ) is the kernel of the map M ⊗R K → I M . Hence if M is flat, then
TorR1 (M, N ) = 0 for all R-modules N . Conversely, suppose that TorR
1 (M, N ) = 0 for all
R-modules N . Let A → B be any injective map of R-modules. Then the short exact
sequence
0 /A /B / A/B /0

gives the exact sequence on homology

··· / TorR (M, A/B) / M ⊗R A / M ⊗R B / ···


1

Since TorR
1 (M, A/B) = 0, M ⊗R A → M ⊗R B is injective, so M is flat. The other statement
follows by symmetry of Tor.
(b) Take any projective resolution of N . If M is flat, then tensoring with M is an exact functor,
so the resulting homology of the projective resolution is trivial, which means TorR
i (M, N ) =
0 for all i > 0. The converse statement is by (a).

14
7 Completions and Hensel’s Lemma
5. (a) We use Theorem 7.2a, which says that if M is a finitely generated R-module, then R̂⊗R M = ∼

M̂ . Then we have HomR̂ (M̂ , N̂ ) = HomR̂ (R̂⊗R M, R̂⊗R N ). By Theorem 7.2b, R̂ is a flat R-
module, so Proposition 2.10 applies, giving HomR̂ (R̂ ⊗R M, R̂ ⊗R N ) ∼
= R̂ ⊗R HomR (M, N ).
Note that HomR (M, N ) is a finitely generated R-module since M and N are both finitely
generated. To see why, fix a generating set of M and fix a generating set of N . For each
maximal linearly independent set in the generating set of M , we may map each element
in the set to elements of the generating set of N . There are finitely many such functions
for each maximal linearly independent set, and the values on the linearly independent
set determine (not necessarily uniquely) values for other elements of M . There are finitely
many maximal linearly independent subsets of our generating set of M , and these functions
will generate a module (which will be Noetherian) that has HomR (M, N ) as a submodule.
Then we use Theorem 7.2a once more to get R̂ ⊗R HomR (M, N ) ∼ = HomR (M, N )∧ , and
combining the previous isomorphisms gives the desired one.
(b) Let ϕ ∈ P̂ HomR̂ (M̂ , N̂ ). Then ϕ = pψ for some ψ ∈ HomR̂ (M̂ , N̂ ) and p ∈ P̂ . For any
x ∈ M̂ , we have ϕ(x) = pψ(x) ∈ pN̂ . Thus, ϕ(M̂ ) ⊆ pN̂ ⊆ P̂ N̂ , so P̂ HomR̂ (M̂ , N̂ )
consists of maps that take M̂ to P̂ N̂ .
(c) Let Q = N̂ /ϕ0 (M̂ ). Then Q/P̂ Q = N̂ /(P̂ N̂ + ϕ0 (M̂ )). Note that ϕ0 differs from an
isomorphism of M̂ to N̂ by an element of P̂ HomR̂ (M̂ , N̂ ), and by part (b), P̂ HomR̂ (M̂ , N̂ )
consists of maps from M̂ to P̂ N̂ , so P̂ N̂ + ϕ0 (M̂ ) = N̂ . This gives Q/P̂ Q = N̂ /N̂ = 0,
which means Q = P̂ Q. Since Q is a homomorphic image of a Noetherian module, it is
finitely generated, so by Nakayama’s Lemma, Q = 0, which means N̂ = ϕ0 (M̂ ), and thus
ϕ0 is an epimorphism.
(d) Let ϕ ∈ HomR̂ (M̂ , N̂ ) be an isomorphism. From the identity R̂ ⊗R HomR (M, N ) ∼ =
P
HomR̂ (M̂ , N̂ ), we can write ϕ = si ⊗R ϕi where si ∈ R̂ and ϕi ∈ HomR (M, N ). There
is a projection R → R̂/P̂ since R̂/P̂ = R/P , so from the projection R̂ → R̂/P̂ , can
write si = ri + pi with ri ∈ R and pi ∈ P̂ . Now define ϕ0 =
P
ri ⊗R ϕi , which differs
from ϕ by an element in P̂ HomR̂ (M̂ , N̂ ), so is an epimorphism M̂ → N̂ . Noting that
ri ϕi and M̂ = R̂ ⊗R M gives that ϕ0 also gives a map
P P P
ri ⊗R ϕi = ri (1 ⊗R ϕi ) =
M → N . To see that this is an epimorphism, let f1 , . . . , fr ∈ M be generators for M̂ over
R̂. Then hϕ0 (f1 ), . . . , ϕ0 (fr )i = N̂ over R̂, and define L = hϕ0 (f1 ), . . . , ϕ0 (fr )i ⊆ N over R.
We get
0 = N/L ⊗R R̂/P̂ = N/L ⊗R R/P = (N/L)/(P N/L) .
The first equality comes since N/L⊗R R̂ = N̂ /L̂ = 0, and the second from R̂/P̂ = R/P . So
N/L = P N/L, and by Nakayama’s Lemma, N/L = 0, which implies N = L, and thus ϕ0 is
an epimorphism. Switching the roles of M and N gives an epimorphism ϕ00 ∈ HomR (N, M ).
Then ϕ0 ϕ00 is an epimorphism N → N and ϕ00 ϕ0 is an epimorphism M → M , so by Corollary
4.4a, they are both isomorphisms. Thus, ϕ0 and ϕ00 are inverses to one another, and hence
are isomorphisms themselves, which means M ∼ = N.

9 Fundamental Definitions of Dimension Theory


1. (a) Let m be the maximal ideal of R generated by a. Suppose that not all ideals of R are
principal. Since R is Noetherian, there exists an ideal I maximal among those that are

15
not principal, so write I = (f1 , . . . , fr ). Since I is contained in m (R is principal, so I is
proper), we can write fi = afi0 for i = 1, . . . , r. Consider the ideal I 0 = (f10 , . . . , fr0 ). It is
clear that I ⊆ I 0 . If the inclusion is strict, then I 0 is principal by maximality of I, generated
by say b. Then we can write fi0 = di b for i = 1, . . . , r, so fi = di ab. But there is a relation
c1 f10 + · · · + cr fr0 = b, which translates to c1 f1 + · · · + cr fr = ab, so ab ∈ I and generates I,
which is a contradiction. Thus I 0 = I. However, we can again write fi0 = afi00 and get an
ideal I 00 , which is equal to I. In this way, we get an infinite chain of ideals

(f1 ) ⊆ (f10 ) ⊆ (f100 ) ⊆ · · · .


(i) (i+1)
Since R is Noetherian, there is some i such that f1 = af1 , and such that they generate
the same ideal, which means a is a unit. This contradicts the fact that m is a maximal
ideal, so we conclude that all ideals of R are principal.
Now let I be a nonzero proper ideal. We know that I is generated by some element f . We
can do the same thing as above and write f = af 0 , and f 0 = af 00 , etc. Since a is not a unit,
we get strict inclusions
(f ) $ (f 0 ) $ (f 00 ) $ · · ·
which eventually terminates at f (i) since R is Noetherian. Thus the element f (i) must be
a unit, otherwise there would be a strict inclusion (f (i) ) $ (f (i+1) ). Then I = (ai f (i) ) =
(ai ) = mi , so every nonzero ideal is a power of the maximal ideal.
(b) We use the fact that dim R = supm codim m where m ranges over all maximal ideals of R,
and that codim m = dim Rm . Since Rm is a Noetherian local ring, and mm is principal,
we use part (a) to conclude that every nonzero ideal of Rm is a power of mm . Let mm be
generated by a. If I = mjm , then I = (aj ), and does not contain a if j > 1, so is not prime.
Thus, any chain of prime ideals in Rm cannot contain any ideal other than 0 or mm , so
dim Rm ≤ 1. This gives that dim R ≤ 1.

6. Any chain of prime ideals in S is contained in some Pm [U −1 ], so

dim S = sup{codim(Pm [U −1 ]) | 1 ≤ m ≤ ∞} .

We can use Krull’s Principal Ideal Theorem to show that codim(Pm [U −1 ]) = d(m) − d(m − 1),
which gives the desired equality. However, even without explicitly computing dim S, it can be
more directly shown that S is infinite dimensional if d(m) − d(m − 1) is unbounded by saying
that
(xd(m−1)+1 ) $ (xd(m−1)+1 , xd(m−1)+2 ) $ · · · $ (xd(m−1)+1 , . . . , xd(m) )
is a chain of prime ideals in S for all m.
To prove the lemma, let I ⊆ S be an ideal containing s. For each maximal ideal m containing s,
there is a finite set of generators Bm for Im since Sm is Noetherian. Since I ⊆ m
S in each case, the
generators correspond to elements of I, so let J be the ideal of S generated by Bm and s. Then
J is finitely generated since there are finitely many such m, and there is an inclusion ϕ : J → I.
The localized map ϕm for each maximal ideal m that contains I is surjective by construction. If
m does not contain I, then Jm and Im both have invertible elements and become Rm , so again
ϕm is surjective. By Corollary 2.9, this means ϕ is surjective, which gives that J = I, so each
ideal of S is finitely generated, and hence S is Noetherian.
To see that S satisfies the hypotheses of the lemma, note that the only maximal ideals of S are
the ideals Pm [U −1 ]. Upon localizing S at one of these ideals, all of the variables not contained

16
in Pm become invertible. Hence the localization looks like a localization of a polynomial ring in
the variables xd(m−1)+1 , . . . , xd(m) over a field, which is Noetherian. For the second condition,
any nonzero s ∈ S contains finitely many variables, so can only be contained in the maximal
ideals Pm [U −1 ] that contain at least one of these variables. So S satisfies the conditions of the
lemma, and is therefore Noetherian.

10 The Principal Ideal Theorem and Systems of Parameters


3. Consider the ideal generated by (x, 1). It is prime, and contains the ideal generated by (0, 1),
which is also prime. By the principal ideal theorem, the codimension of the ideal is ≤ 1, so we
have shown that it is equal to 1.

4. Consider the ring R[b/a], where b/a is a fraction such that a · b/a = b. Every element of
R[b/a] is a polynomial in b/a, so suppose that (c0 + c1 (b/a) + · · · + cn (b/a)n )x = 0 for some
n and x ∈ R[b/a]. We claim that R[b/a] is a domain. Multiplying both sides by an gives
(c0 an +c1 ban−1 +· · ·+cn bn )x = 0, which is a relation in R, so either x = 0 or c0 an +· · ·+cn bn = 0.
If x = 0, we’re done, so assume c0 an + · · · + cn bn = 0. Then c0 + c1 (b/a) + · · · + cn (b/a)n = 0
because a is not a zero divisor, so R[b/a] is a domain. Let R[x] be a polynomial ring over R,
and consider the map ϕ : R[x] → R[b/a] defined by x 7→ b/a. Since a, b is a regular sequence,
b/a is not an element of R. If it were, then b/a = c for some c ∈ R, which means b = ac, and
thus a is a zerodivisor of R/(b). The kernel of ϕ consists of those polynomials in R[x] that have
b/a as a root. Any multiple of ax − b is in the kernel, and any polynomial in the kernel must
be divisible by ax − b since x − b/a ∈ / R[x], so ker ϕ = (ax − b), which gives an isomorphism of
domains R[x]/(ax − b) ∼ = R[b/a]. This means that (ax − b) is a prime ideal, so ax − b is a prime
element of R[x].

15 Gröbner Bases
4. (a) The first equality is obvious, and the second follows because counting the number of mono-
mials in r − s variables of degree ν is the same as counting the number of ways to pick ν
objects from a choice of r − s objects with repetition allowed.
(b) The map ϕ is given by multiplication by n. This is a degree d map, so the shift fixes that.
Also multiplication of n is an injective map S/J → S/I because if nf ∈ I, then f ∈ J by
definition. The second map S/I → S/(I, n) is given by projection which is surjective. It
is clear that the image of ϕ is in the kernel of the projection. The projection map can be
described by writing a polynomial f ∈ S/I as a sum f 0 + gn where n does not divide f 0
and mapping it to f 0 . It is clear from this description that the sequence is exact at S/I.

14. (a) We need to show that if in(g1 ), . . . , in(gs ) generate in(M ), then g1 , . . . , gs generate M .
Take any f ∈ M . Then there is a linear combination a1 in(g1 ) + · · · + as in(gs ) = in(f ), so
f 0 = f − (a1 g1 + · · · + as gs ) < f . In particular, f 0 has less terms than f , so we proceed by
induction to see that f is generated by g1 , . . . , gs .
(b) We may always obtain a Gröbner basis where (i) holds because k is a field. To get
condition (ii), we appeal to part (a). That is, suppose in(hi ) divides a term of hj for
i 6= j. If it is the initial term, then {in(h1 ), . . . , in(hs )} \ {in(hj )} also generates in(M ), so
h1 , . . . , hj−1 , hj+1 , . . . , hs still generates M by (a). If it is not the initial term, then we may

17
subtract the offending term from hj to get h0j such that in(hj ) = in(h0j ), and again by (a),
{h1 , . . . , h0j , . . . , hs } generates M .
It is clear that given such h1 , . . . , hs , the in(hi ) are a minimal set of generators of in(M ).
Let g1 , . . . , gs be a Gröbner basis that also satisfies (i) and (ii). For each in(gi ), we can
write it as a linear combination of the in(hj ), and vice versa. Putting these together means
that we can write in(gi ) as a linear combination a1 in(g1 ) + · · · + as in(gs ). This implies that
ai = 1 and aj = 0 for j 6= i, so gi = hj for some j. Therefore, {gi } = {hi }.

16 Modules of Differentials
1. Let b ∈ S be an idempotent and d : S → M be a derivation. Then db = db2 = bdb + bdb = 2bdb,
so (2b − 1)db = 0. However, (2b − 1)(2b − 1) = 4b − 2b − 2b + 1 = 1, so 2b − 1 is a unit, which
means that db = 0.

17 Regular Sequences and the Koszul Complex


1. Let D = ad − bc, e = D−1 , w1 = ax + by, and w2 = cx + dy. We need to construct isomorphisms
f , g, and h such that the following diagram commutes:

K(x, y) : 0 /R /R⊕R /R

f g h
  
K(w1 , w2 ) : 0 /R /R⊕R /R

   
d c ae −ce
Let f be the identity, g = , and h be multiplication by D. Then g −1 =
b a −be de
and h−1 is multiplication by e.

18

You might also like