Honors Algebra 4, MATH 371 Winter 2010

Assignment 3
Due Friday, February 5 at 08:35

1. Let R 6= 0 be a commutative ring with 1 and let S ⊆ R be the subset of nonzero elements
which are not zero divisors.
(a) Show that S is multiplicatively closed.
(b) By definition, the total ring of fractions of R is the ring Frac(R) := S −1 R; it is a ring
equipped with a canonical ring homomorphism R → S −1 R. If T is any multiplicatively
closed subset of R that is contained in S, show that there is a canonical injective ring
homomorphism T −1 R → Frac(R), and conclude that T −1 R is isomorphic to a subring
of Frac(R).
(c) If R is a domain, prove that Frac(R) is a field and hence that T −1 R is a domain for any
T as above.
Solution:
(a) If a, b are nonzero and are not zero-divisors, then ab can’t be zero on the one hand, and
can’t be a zero divisor on the other since if sab = 0 then (sa)b = 0 which forces sa = 0
as b is not a zero divisor, and this forces s = 0 as a is not a zero divisor.
(b) Because T ⊆ S, under the canonical map ϕ : R → S −1 R, every element of T maps to a
unit. Thus, this map uniquely factors as the composite of the canonical map R → T −1 R
with a unique ring homomorphism ψ : T −1 R → S −1 R. If r/t maps to zero, then there
exists s ∈ S with sr = 0. But s must be nonzero and not a zero divisor, whence we must
have r = 0 and hence r/t = 0. We conclude that ψ is injective, hence an isomorphism
onto its image, which is a subring of S −1 R.
(c) If R is a domain, then S = R \ 0 and every nonzero element of S −1 R is invertible (If
r/s 6= 0 then in particular r 6= 0 and hence r ∈ S so s/r ∈ S −1 R and is the inverse
of r/s). Thus, S −1 R is a field. Since any subring of a field is necessarily a domain, we
conclude as desired.
2. Let R be a commutative ring with 1.
(a) Let S ⊆ R be a multiplicatively closed subset. Prove that the prime ideals of S −1 R
are in bijective correspondence with the prime ideals of R whose intersection with S is
empty.
(b) If p is an ideal of R, show that S := R \ p is a multiplicatively closed subset if and
only if p is a prime ideal. Writing Rp for the ring of fractions S −1 R, show that Rp has
a unique maximal ideal, and that this ideal is the image of p under the canonical ring
homomorphism R → Rp . (In other words, the localization of R at p is a local ring).
 (c) Let r ∈ R be arbitrary. Show that the following are equivalent:
i. r = 0
ii. The image of r in Rp is zero for all prime ideals p of R.
iii. The image of r in Rp is zero for all maximal ideals p of R.

Solution:

(a) Denote by ϕ : R → S −1 R the canonical map. If p is a prime ideal of R not meeting S,

we claim that
S −1 p := {x/s : x ∈ p, s ∈ S}
is a prime ideal of S −1 R. Indeed, if (r1 /s1 )(r2 /s2 ) = x/s ∈ S −1 p then there exists t ∈ S
with
t(sr1 r2 − s1 s2 x) = 0
in R. Since x and 0 lie in p, we conclude that tsr1 r2 ∈ p. Since S ∩ p = ∅, it follows that
r1 r2 ∈ p whence r1 ∈ p or r2 ∈ p as p is prime. It follows that S −1 p is prime.
Conversely, if p is a prime ideal of S −1 R then ϕ−1 p is a prime ideal of R by a previous
homework, and it remains to show that for any prime ideal p of R, we have

ϕ−1 (S −1 p).

If ϕ(r) = x/s lies in S −1 p then t(rs − x) = 0 for some t ∈ S. Arguing as above, we

conclude that r ∈ p.
(b) The first part follows immediately from the definition of prime. The second follows easily
from (2a) since the prime ideals of R not meeting S := R \ p are exactly the prime ideals
of R contained in p.
(c) Clearly (i) =⇒ (ii) =⇒ (iii). If R is the zero ring then the equivalence is obvious, so we
may assume that R is nonzero. Let x ∈ R and denote by

ann(x) := {r ∈ R : rx = 0}

the annihilator of x in R; it is easily seen to be an ideal of R. Suppose that the image

of x in Rp is zero for all maximal ideals p If ann(x) is not the unit ideal, then there
exists a maximal ideal p0 containing ann(x). However, our hypothesis on x implies that
then there exists s ∈ R \ p0 with sx = 0, i.e. ann(x) is not contained in p0 which is a
contradiction. It follows that 1 ∈ ann(x) and hence that x = 0.

3. Do exercises 8–11 in §7.6 of Dummit and Foote (inductive and projective limits).
Solution: This is important stuff, but extremely tedious to write up in TEX. If you have
any questions about it, I’ll be more than happy to discuss.
4. A Bézout domain is an integral domain in which every finitely generated ideal is principal.

(a) Show that a Bézout domain is a PID if and only if it is noetherian.

(b) Let R be an integral domain. Prove that R is a Bezout domain if and only if every pair
of elements a, b ∈ R has a GCD d ∈ R that can be written as an R-linear combination
of a and b, i.e. such that there exist x, y ∈ R with d = ax + by.
(c) Prove that a ring R is a PID if and only if it is a Bézout domain that is also a UFD.
(d) Let R be the quotient ring of the polynomial ring Q[x0 , x1 , . . .] over Q in countably
many variables by the ideal I generated by the set {xi − x2i+1 }i≥0 . Show that R is a
Bézout domain which is not a PID (Hint: have a look at Dummit and Foote, §9.2 # 12).
Remark: The above example of a Bézout domain which is not a PID is somewhat
artificial. More natural examples include the “ring of algebraic integers” (i.e. the set
of all roots of monic irreducible polynomials in one variable over Z) and the ring of
holomorphic functions on the complex plane. The proofs that these are B’ezout domains
is, as far as I know, difficult. For example, in the case of the algebraic integers, one needs
the theory of class groups).

Solution:

(a) Easy unravelling of definitions.

(b) If R is a Bézout domain then the finitely generated ideal (a, b) is principal, say with
generator d, whence there exist x and y with ax + by = d. Clearly d is a GCD of a and
b. Conversely, suppose R has a GCD algorithm of the type described, and that I is an
ideal of R generated by a1 , . . . , an . Let d be a gcd of a1 and a2 . Then by definition of
GCD, we have a1 ∈ (d) and a2 ∈ (d) whence I ⊆ (d, a3 , · · · , an ). Since also we have
d = a1 x + a2 y, we get the reverse inclusion and R can be generated by n − 1 elements.
By descent on n, we deduce that I is principal and hence that R is Bézout.
(c) We have seen that PID implies UFD and Bézout. Conversely, suppose that R is a Bézout
UFD and let I be a nonzero ideal of R. For each irreducible element r of R, denote by
er the minimal exponent of r occurring in the unique factorizations of nonzero elements
of I and write br for any element of I realizing this exponent of r. Clearly er = 0 for all
but finitely many r, say for r1 , . . . , rn .
An easy induction using (4b) shows that there exists a GCD d of br1 , . . . , brn which may
be expressed as an R-linear combination

d = xi br1 + · · · + xn brn ,

so d ∈ I. On the other hand, the exponent of ri in d is at most eri since d|bri whence
it must be exactly ri by minimality. If a ∈ I is not divisible by d then there is some i
for which the exponent of ri in the unique factorization of a is strictly less than eri , a
 contradiction to the minimality of the eri . Thus I ⊆ (d) and we must then have I = (d)
is principal.
(d) For each i, we have an injective Q-algebra homomorphism ϕi : Q[xi ] → R given by
sending xi to xi . We write ψi : Q[xi ] → Q[xi+1 ] for the Q-algebra himomorphism taking
xi to x2i+1 , so that
ϕi ◦ ψi = ϕi+1
and note that ψi is injective. Write Ri := im(φi ); it is a subring of R that is isomorphic
to Q[xi ] and is hence a PID. Moreover, the mappings ψi give ring inclusions Ri ⊆ Ri+1
and it is easy to see from the very definition of R that R = ∪∞ i=1 Ri . We conclude
immediately that R is Bézout: indeed, any finitely generated ideal of R is contained in
some Ri (as this is the case for each of its generators, and the Ri form a chain) and
each Ri is principal. I claim that the ideal M generated by all xi can not be finitely
generated. There are probably a billion ways to see this, so I’ll just pick one that comes
i
to mind: For each i, denote by 21/2 the unique positive 2i th root of 2 in Q and consider
i
the Q-algebra homomorphism Q[x0 , x1 , . . .] → Q sending xi to 21/2 . Clearly, I is in
the kernel of this map so we get a homomorphism of Q-algebras Ψ : R → Q. If M
were finitely generated, the image of Ψ would be a finitely generated Q-subalgebra of
Q and in particular would be Q-vector space of finite dimension d < ∞. It follows that
any element of the image would satisfy a polynomial with rational coefficients having
i
degree at most d. But this image contains 21/2 , which can not satisfy a polynomial
i
with Q-coefficients of degree less than 2i . Indeed, on the one hand 21/2 is a root of
i
Fi := T 2 − 2, which is irreducible over Q be Gauss’s Lemma and Eisenstein’s criterion
applied with p = 2. On the other hand, if g is any nonzero polynomial of minimal degree
i
satisfied by 21/2 , then by the division algorithm we have Fi = gq + r for some rational
polynomials q and r with deg r < deg g whence r = 0 by minimality and Fi = gq. As Fi
is irreducible, we conclude that q is a unit and hence an element of Q× so deg(g) = 2i .
5. Let R = Z[i] := Z[X]/(X 2 + 1) be the ring of Gaussian integers.
(a) Let N : R → Z≥0 be the field norm, that is
N (a + bi) := (a + bi)(a − bi) = a2 + b2 .
Prove that R is a Euclidean domain with this norm. Hint: there is a proof in the book
on pg. 272, but you should try to find a different proof by thinking geometrically.
(b) Show that N is multiplicative, i.e. N (xy) = N (x)N (y) and deduce that u ∈ R is a unit
if and only if N (u) = 1. Conclude that R× is a cyclic group of order 4, with generator
±i.
(c) Let p ∈ Z be a (positive) prime number. If p ≡ 3 mod 4, show that p is prime in Z[i]
and that Z[i]/(p) is a finite field of characteristic p which, as a vector space over Fp , has
dimension 2.
 If p = 2 or p ≡ 1 mod 4, prove that p is not prime in Z[i], but is the norm of a prime
p ∈ Z[i] with Z[i]/(p) isomorphic to the finite field Fp . Conclude that p ∈ Z can be
written as the sum of two integer squares if and only if p = 2 or p ≡ 1 mod 4.

Solution:

(a) In the complex plane, Z[i] corresponds to the integer lattice consisting of all points (a, b)
with integral coordinates. The norm of an element a + bi is precisely the square of the
Euclidean distance from the origin to the point (a, b) corresponding to a + bi. Suppose
now that x = c + di and y = a + bi are Gaussian integers with y 6= 0. The quotient
x/y (as complex numbers) is located inside (or on the perimeter of) a unit square in
the complex plane whose vertices have integral coordinates. The minimal (Euclidean)
distance
√ from x/y to a vertex of this square is at most half the diagonal of the square, or
2/2. We conclude that there exist a Gaussian integer q (a vertex of minimal distance)
with √
x 2 2
N ( − q) ≤ ( )
y 2
or in other words, there exist Gaussian integers q and r := x − yq with
1
x = yq + r and N (r) ≤ N (y) < N (y)
2
so we indeed have a division algorithm and Z[i] is Euclidean.
(b) The multiplicativity of N is a straightforward (albeit tedious) calculation. By definition
u ∈ Z[i] is a unit if there exists v ∈ Z[i] with uv = 1; taking norms gives N (u)N (v) = 1
so since N is nonnegative we conclude that N (u) = 1. Conversely, if u = a + bi satisfies
N (u) = 1, then
1 = N (u) = (a + bi)(a − bi)
so u is a unit. It’s easy to see that the only integer solutions to a2 + b2 = 1 are the 4
points (a, b) = (±1, 0), (0, ±1) corresponding to ±1, ±i. Since (±i)2 = −1 we conclude
that Z[i]× is cyclic of order 4 generated by ±i.
(c) Let p be a prime of Z. If
p = (a + bi)(c + di)
then taking norms gives p2 = (a2 +b2 )(c2 +d2 ) so if neither a+bi nor c+di is a unit then we
deduce that p = a2 + b2 for integers a and b. If p ≡ 3 mod 4 then reducing this equation
modulo 4 implies that 3 = a2 + b2 has a solution in Z/4Z which it obviously doesn’t
(by inspection, sums of squares in Z/4Z can be 0, 1, 2 only). Thus any factorization of
p ≡ 3 mod 4 in Z[i] as above has one of the two factors a unit; we conclude that p is
irreducible and hence prime and hence maximal (we’re in a Euclidean domain after all).
The quotient Z[i]/(p) is therefore a field which is also an Fp -vector space. Using the
isomorphism Z[i] ' Z[X]/(X 2 + 1) and the third isomorphism theorem for rings, we
have
Z[i]/(p) ' (Z[X]/(p))/(x2 + 1) = Fp [X]/(X 2 + 1)
which as an Fp -vector space has basis 1, X so is of dimension 2.
If p ≡ 1 mod 4 we claim that −1 is a square modulo p. Indeed, the group of units F× p is
cyclic of order p − 1 (proof?) so for any generator u we have up−1 = 1 in Fp . It follows
that u(p−1)/2 = ±1 and we must have the negative sign since u is a generator. Since
p ≡ 1 mod 4 so (p − 1)/2 is even, we conclude that −1 is a square mod p. Thus, the
equation
x2 + 1 = py
has a solution for integers x, y. If p were prime in Z[i] then we would have

p|(x − i)(x + i)

which would force p|(x − i) or p|(x + i) both of which are absurd. Thus, p is not prime
in Z[i] and we have a factorication

p = (a + bi)(c + di)

with the norm of each factor strictly bigger than 1. Taking norms, we conclude that
p = N (a + bi). Moreover, p := (a + bi) must be prime in Z[i] as is easily seen by taking
norms. Similarly, (a − bi) is prime.
It is easy to see that the two prime ideals (a + bi) and (a − bi) are co-maximal as the
ideals (a) and (b) of Z must be comaximal (why?) and hence by the Chinese Remainder
Theorem we have
Z[i]/(p) = Z[i]/(a + bi) × Z[i]/(a − bi).
The ring Z[i]/(p) ' Z[X]/(p, x2 + 1) ' Fp [X]/(X 2 + 1) is a vector space of dimension 2
over Fp and hence has cardinality p2 . It follows easily from this that Z[i]/(a + bi) and
Z[i]/(a − bi) each have cardinality p and so the canonical map of rings Fp → Z[i]/(a + bi)
must be an isomorphism.
To handle 2, we argue as above using that 2 = (1 + i)(1 − i). We conclude that p is a
sum of integer squares if and only if p is 2 or p ≡ 1 mod 4.