Math 121 Homework 5: Notes on Selected Problems

12.1.2. Let M be a module over the integral domain R.

(a) Assume that M has rank n and that x1 , . . . , xn is any maxi-
mal set of linearly independent elements of M. Prove that N
is isomorphic to R n and that the quotient M/N is a torsion R-
module.
(b) Prove conversely that if M contains a submodule N that is free
of rank n (i.e., N › R n ) such that the quotient M/N is a torsion
R-module then M has rank n.

Solution.
(a) Write e1 , . . . , en for the standard basis of R n . The R-linear map
R n → N given by ei , xi is injective by independence of the
set x1 , . . . , xn and surjective by definition of N. For any x in M,
x, x1 , . . . , xn is R-linearly dependent by maximality so we can
write an R-linear dependence
ax + a1 x1 + · · · + an xn = 0
for a, a1 , . . . , an in R not all zero. Since x1 , . . . , xn are R-linearly
independent, a is nonzero. Therefore the image of x in M/N is
R-torsion, being killed by the nonzero element a of R.
(b) If N is free on x1 , . . . , xn , then x1 , . . . , xn is an R-linearly in-
dependent set of elements of M so the rank of M is at least
n. Let y1 , . . . , yn+1 be elements of M. Since M/N is torsion,
for each i there exists a nonzero ai in R such that ai yi is in
N. Then since N is free of rank n, there exists a nontrivial
R-linear dependence among a1 y1 , . . . , an+1 yn+1 , which gives a
nontrivial R-linear dependence among y1 , . . . , yn+1 since the ai
are nonzero and hence not zero divisors as R is an integral
domain. Therefore the rank of M is precisely n. (Instead of ap-
pealing to the fact that n + 1 elements in a free module of rank
n over a commutative ring with identity must be dependent—
proved in the text in the case that R is a domain—one may also,
as indicated in the hint, imitate/repeat either of the proofs.) 

12.1.5. Let R = Z[x] and let M = (2, x) be the ideal generated by 2 and
x, considered as a submodule of R. Show that {2, x} is not a basis of
M. Show that the rank of M is 1, but that M is not free of rank 1.

Solution. The set {2, x} is not a basis of M because it is not Z[x]-

linearly independent: (−x) · 2 + 2 · x = 0 is a nontrivial Z[x]-linear
1
 2

dependence. In fact, for any p(x) and q(x) in M, either both of p(x)
and q(x) are zero or (−q(x)) · p(x) + p(x) · q(x) = 0 is a nontrivial
Z[x]-linear dependence. Therefore M does not contain any indepen-
dent sets with more than one element. Any nonzero element in M is
non-torsion, however, so M has rank at least 1. Therefore M has rank
precisely 1.
To show that M is not free of rank 1 it suffices to show that no
element of M principally generates M. Let r (x) be an element of M
so that r (x)Z[x] contains 2, that is r (x) divides 2. By degree con-
siderations, r (x) is a constant polynomial, which must be equal to
some integral multiple of 2. Then x is not divisible by r (x). Therefore
r (x)Z[x] does not contain x and so is a proper Z[x]-submodule of M.
Finally, M is not a principal Z[x]-module and hence not free of rank
1. 

12.1.20. Let R be an integral domain with quotient field F and let M be

any R-module. Prove that the rank of M equals the dimension of the
vector space F ⊗R M over F .
Proof. We shall assume that the kernel of M → F ⊗R M is the torsion
submodule of M. This is proved in several ways in the comment and
note below.
We first show that every element of F ⊗R M is an F -multiple of a
simple tensor in the image of M → F ⊗R M. This is true of simple
a a
tensors in F ⊗R M: an arbitrary simple tensor b ⊗ m either has b = 0
so can be written as 0(1 ⊗ m) or otherwise can be written as
a
b
⊗ m = a( b1 ⊗ m) = ab b( b1 ⊗ m) = ab (1 ⊗ m).
The set of F -multiples of elements in the image of M → F ⊗R M is also
closed under addition
a1 a2 1
b1
(1 ⊗ m1 ) + b2
(1 ⊗ m2 ) = b1 b2
(b2 a1 (1 ⊗ m1 ) + b1 a2 (1 ⊗ m2 )
1
= b1 b2
(1 ⊗ (b2 a1 m1 + b1 a2 m2 ))
so the general case follows. Now consider an F -linearly independent
set in F ⊗R M, which by the above we may write as {fi (1 ⊗ mi )}. Then
{1 ⊗ mi } is also an F -linearly independent set in F ⊗R M. We claim that
{mi } is an R-linearly independent set in M. Assume that for each i
P
there is ri in R, all but finitely many zero, such that ri mi = 0. Then
P
ri (1 ⊗ mi ) = 0. By independent of {1 ⊗ mi }, ri = 0 for each i. This
proves that {mi } is R-linearly independent.
Conversely, let {mi } be an R-linearly independent subset of M. We
claim that {1 ⊗ mi } is an F -linearly independent subset of F ⊗R M.
 3

Assume that for each i there is fi in F , all but finitely many zero, such
P
that fi (1 ⊗ mi ) = 0. Multiplying out denominators for the finitely
P
many nonzero coefficients gives that ri (1 ⊗ mi ) = 0 where each ri is
P
a nonzero multiple of fi . Therefore 1 ⊗ ri mi = 0. Since the kernel
of M → F ⊗R M consists of torsion elements, there is a nonzero r in R
P
such that r ri mi = 0. By R-linear independence, for each i we have
r ri = 0, but r is nonzero so ri = 0. It follows that each fi is a zero
divisor and hence equal to zero. This proves F -linear independence of
{1 ⊗ mi }.
Given an R-linearly independent subset of M, we demonstrated an
F -linearly independent subset of F ⊗R M of the same cardinality, and
the other way around. Therefore the R-rank of M is the same as the
F -dimension of F ⊗R M. 
Comment. The harder part is showing that the F -dimension of F ⊗R M
is at least the rank of M. The main difficulty is showing that 1 ⊗ m = 0
in F ⊗R M implies that m in M is R-torsion. We illustrate a proof of
this.
Let N 0 > N be an inclusion of R-modules. We show that any R-
bilinear map ϕ0 : F ×N 0 → F extends to an R-bilinear map ϕ : F ×N → F .
Assume first that N is generated over N 0 by a single element n. If
Rn ∩ N 0 = 0 so N = N 0 ⊕ Rn, we set ϕ(f , n) = 0 for all f in F and
extend linearly. If Rn ∩ N 0 6= 0, then there is a nonzero r in R and n0
in N 0 so that r n = n0 , and then we set ϕ(f , n) = r −1 ϕ0 (f , n0 ) for all
f in F and extend linearly. Inductively we have proved the existence of
extensions when N is finitely generated over N 0 , and the general case
follows from Zorn’s lemma.
Now consider a non-torsion element m of M. Then Rm is a free R-
module on {m}, and Rm⊗R F is a free F -module on {m⊗1}. Since m⊗1
is not zero in Rm ⊗R F , there exists an R-bilinear map Rm × F → F that
does not kill (m, 1). An extension to an R-bilinear map of M × F → F
exists, and by virtue of being an extension also does not kill (m, 1).
Therefore m ⊗ 1 is nonzero in M ⊗R F . 
Note. Alternatively, one may use the “module of fractions”, which
is defined as an R-module homomorphism M → MF with codomain
some F -module MF that is universal, meaning that any R-module ho-
momorphism M → Z with codomain an F -module factors uniquely
through M → MF . (Every F -module is an R-module via “restriction of
scalars” R → F .) A realization of MF is as follows: As a set MF is the
set of equivalence classes of M × (R \ 0) with the equivalence rela-
tion (m, s) ∼ (m0 , s 0 ) if and only if there exists s 00 in R \ 0 such that
s 00 (s 0 m − sm0 ) = 0. Denoting the equivalence class of (m, s) by m s
may
 4

help in understanding the definition; we will simply write [(m, s)] for
the equivalence class of (m, s). The set map M → MF sending m to the
equivalence class of (m, 1) will be an R-module homomorphism when
MF has the following F -module structure: the zero element is [(0, 1)]
and addition is given by

[(m, s)] + [(m0 , s 0 )] = [(s 0 m + sm0 , ss 0 )]

and the action of F is given by

r0
s0
[(m, s)] = [(r 0 m, s 0 s)].

Having discussed the module of fractions MF , we now return to the

original question by showing that 1⊗m is zero in F ⊗R M only if m is R-
torsion. An R-bilinear map M × F → MF is given as the composition of
F × M → F × MF that is the identity in the second component followed
by the F -action, which is a map F × MF → MF . The bilinear map induces
an R-linear map M ⊗R F → MF (which is in fact an isomorphism) taking
m ⊗ 1 to [(m, 1)]. By definition of the equivalence relation, (m, 1) ∼
(0, 1) if and only if there exists a nonzero r in R such that r m = 0,
that is if and only if m is torsion. In particular, m ⊗ 1 has nonzero
image in MF when m is not torsion. Therefore m ⊗ 1 is not zero when
m is not torsion.

Note. Another alternative to the first proof, which extended bilinear

maps, is to use the adjoint property of the tensor product. We prove
the following lemma (which was implicitly proved above).

Lemma. Let R be an integral domain and let F be the fraction field of

R. Then for any inclusion of R-modules N 0 > N, the corresponding
map N 0 ⊗R F → N ⊗R F is injective.

Proof. Any R-linear map N 0 → F can be extended to an R-linear map

N → F as follows. For any n in N, either Rn ∩ N 0 = 0 in which case
the extension may send n anywhere, or there is a nonzero r in R such
that r n = n0 for some n0 in N 0 , in which case we send n to 1/r times
the image of n0 . This allows us to create the extension in steps; when
N is not finitely generated over N 0 we invoke Zorn’s lemma. Hence the
natural R-linear map HomR (N, F ) → HomR (N 0 , F ) is surjective.
We now show that the natural F -linear map

HomF (N ⊗R F , F ) → HomF (N 0 ⊗R F , F )
 5

is surjective.1 Let ϕ : N 0 ⊗R F → F be an F -linear map. Then an R-

linear map N 0 → F is given by n0 , ϕ(n0 ⊗ 1). Choose an extension
α to an R-linear map N → F . The R-bilinear map N × F → F given by
(n, f ) , α(n)f induces an R-linear map N ⊗R F → F whose image in
HomR (N 0 ⊗R F , F ) satisfies
a a a a

n0 ⊗ b , α(n0 ) b = ϕ(n0 ⊗ 1) b = ϕ (n0 ⊗ 1) b
= ϕ (n0 ⊗ ab b) b1 = ϕ(n0 ⊗ ab )

for n0 in N 0 and a in R and nonzero b in R, and therefore equals ϕ.

We have proved that every element of HomF (N 0 ⊗R F , F ) comes from
an element of HomF (N ⊗R F , F ).
If x in N ⊗R F is nonzero, there exists an F -linear map N ⊗R F → F
that does not kill x, and hence there exists an F -linear map N 0 ⊗R F → F
not killing the image of x in N 0 ⊗R F . Consequently the natural map
N 0 ⊗R F → N ⊗R F is injective. 

Problem 2. Find, with proof, a free basis for the subgroup of Z4 defined
by the constraint {(x, y, z, w) ∈ Z4 : 11x + 8y + 3z − 9w = 0}.
Solution. Define a Z-linear map ρ : Z4 → Z by
ρ(x, y, z, w) = 11x + 8y + 3z − 9w.
Using the Euclidean algorithm we may find a preimage of 1 under ρ, for
example ρ(3, −4, 0, 0) = 1. Then we may define a section ι : Z → Z4 of ρ
as the unique Z-linear map sending 1 to (3, −4, 0, 0). If ι(x) is in ker ρ,
then x = ρ ◦ ι(x) = 0 so ι(x) = 0. Therefore ker ρ and im ι intersect
trivially. Any x in Z4 can be written as x = (x − ι ◦ ρ(x)) + ι ◦ ρ(x), but
x − ι ◦ ρ(x) is in ker ρ and ι ◦ ρ(x) is in im ι. Thus Z4 is the internal
direct sum of ker ρ and im ι. Then
im(1Z4 − ι ◦ ρ) = ker ρ and ker(1Z4 − ι ◦ ρ) = im ι.
Hence ker ρ has the following presentation: generators
v1 = (1Z4 − ι ◦ ρ)(1, 0, 0, 0) = (−32, 44, 0, 0)
v2 = (1Z4 − ι ◦ ρ)(0, 1, 0, 0) = (−24, 33, 0, 0)
v3 = (1Z4 − ι ◦ ρ)(0, 0, 1, 0) = (−9, 12, 1, 0)
v4 = (1Z4 − ι ◦ ρ)(0, 0, 0, 1) = (27, −36, 0, 1)

1
This follows directly from the natural F -linear isomorphism between F and
HomF (F , F ) and adjointness: for any R-module E there is a natural bijection
HomR (E, HomF (F , F )) ↔ HomF (E ⊗R F , F ),
but we will argue directly.
 6

with the only relations among the vi being multiples of 3v1 − 4v2 = 0
(which is obtained from the choice of section ι). In other words, ker ρ
has presentation matrix
v1 v2 v3 v4
.
1st relation 3 −4 0 0
Replacing v1 with v1 − v2 gives the presentation matrix
v1 − v2 v2 v3 v4
.
1st relation 3 −1 0 0
Replacing v2 with v2 − 3(v1 − v2 ) gives the presentation matrix
v1 − v2 −3v1 + 4v2 v3 v4
1st relation 0 −1 0 0
or equivalently
3v1 − 4v2 v1 − v2 v3 v4
.
1st relation 1 0 0 0
Finally, a free basis for ker ρ is given by
v1 − v2 = (−8, 11, 0, 0)
v3 = (−9, 12, 1, 0)
v4 = (27, −36, 0, 1). 
Note. One method to find integer kernels in general is (to use Hermite
normal form) has follows. We iuse elementary integer column operations
to transform 11 8 3 −9 to its Hermite normal form:
h i
11 8 3 −9
1 0 0 0
 
h i h i 0 1 0 0
-→ 11 −1 3 −9 = 11 8 3 −9 
 
0 −3 1 0

0 0 0 1
1 0 0 0
 
h i h i 11 1 3 −9
-→ 0 −1 0 0 = 11 −1 3 −9 
 
0 0 1 0 

0 0 0 1
0 1 0 0
 
h i h i −1 0 0 0
-→ 1 0 0 0 = 0 −1 0 0  ,
 
 0 0 1 0
0 0 0 1
 7

that is
0 1 0 0
 
h i h i −1 11 3 −9
1 0 0 0 = 11 8 3 −9 
 
 3 −33 −8 27 

0 0 0 1
where
0 1 0 0
 
−1 11 3 −9
U =
 
 3 −33 −8 27 

0 0 0 1
is unimodular because it is a composition of matrices corresponding
to elementary integer operations. In particular we see that the list

1
    
0 0
 11   3  −9
B =   ,   ,  
     
−33 −8  27 
0 0 1
of three elements of Z4 , which corresponds hto the last 3 columns
i
of U , consists of elements in the kernel of 11 8 3 −9 . Con-
h itranspose
versely assume that a1 a2 a3 a4 in Z4 is in the kernel of
h i
11 8 3 −9 . Then
−1  
0 1 0 0 a1

−1 11 3 −9 a 
   2
 3 −33 −8 27  a3 
   
0 0 0 1 a4
h i
is in the kernel of 1 0 0 0 . Thus we may write
−1    
0 1 0 0 a1 0

−1 11 3 −9  a  b 
  2  2
  = 

 3 −33 −8 27  a3  b3 

0 0 0 1 a4 b4
where necessarily b1 , b3 , b4 are integers since U −1 has integral coeffi-
cients. Therefore
a1 1 0 0
       
a   11  3 −9
 2
  = b2   + b3   + b4  
     
a 3  −33 −8  27 
a4 0 0 1
is in the Z-span
h of B . Finally
i we have proved that B is a basis for the
kernel of 11 8 3 −9 .
 8

In general, to find a basis for the kernel of an integer matrix M, we

find a unimodular matrix U such that MU is in Hermite normal form,
whose kernel is the span of the elementary basis vectors correspond-
ing to ‘non-pivot columns’. The images under left multiplication by U
of these elementary basis vectors is then a basis for the kernel of M.

Problem 3. In this exercise we give a direct proof that a submodule of

a free module is free in the case of the ring Z.
Let K ⊆ Zm . Let KQ be the Q-subspace spanned by K; let r be the
dimension of KQ , and let ω be an alternating r -form on KQ .
• Prove that there exists a rational number q so that ω(v1 , . . . , vr )
is in qZ whenever v1 , . . . , vr ∈ K.
• Let x1 , . . . , xr ∈ K be chosen to minimize |ω(x1 , . . . , xr )| among
nonzero values. (In words, the volume of the parallelepiped
that the xi span is as small as possible.) Prove that K is free on
x1 , . . . , x r .

Solution. The space of alternating r -forms on the r -dimensional vec-

tor space KQ has dimension 1 so we may choose ω to be a nonzero
alternating r -form on KQ . Write ei for the ith standard basis vector in
Zn that has 1 in the ith component and zeros elsewhere and set
E = {e1 , . . . , en }.
Then E ×r is a finite set so ω(E ×r ) is a finite subset of Q and hence
there exists a rational number q so that ω(E ×r ) ⊆ qZ. Then by linearity
ω((Zn )×r ) ⊆ qZ, but K ×r ⊆ (Zn )×r so ω(v1 , . . . , vr ) is in qZ whenever
v1 , . . . , vr are in K.
Now choose x1 , . . . , xr in K so that ω(x1 , . . . , xr ) has minimal ab-
solute value among nonzero values of ω restricted to K × · · · × K.
That ω(x1 , . . . , xr ) is nonzero immediately implies that x1 , . . . , xr are
Q-linearly independent, hence a Q-basis for KQ , and in particular in-
dependent over Z. Let λi be the Q-linear functional on KQ that sends
x in KQ to the value obtained by replacing xi by x in the expression
ω(x1 , . . . , xr ), which is just ω(x1 , . . . , xr ) times the dual basis element
corresponding to xi . Therefore the map
X λi (x)
x, xi
ω(x1 , . . . , xr )
is the identity of KQ . For d an integer such that ω(K ×r ) ⊆ d1 Z, we may
write ω(x1 , . . . , xr ) = a/d for some integer a. Consider an index i
and assume x satisfies λi (x) = b/d. Then there exist integers r and
s so that r a + sb = gcd(a, b) and so λ(r xi + sx) = gcd(a, b)/d. By
 9

minimality, it follows that |a| ≤ gcd(a, b) and so a divides b. This

proves that λi (K) ⊆ ω(x1 , . . . , xr )Z. Consequently for x in K,
X λi (x)
x= xi
ω(x1 , . . . , xr )
is an expression for x as a Z-linear combination of x1 , . . . , xr . There-
fore x1 , . . . , xr is a free Z-basis for K. 

Problem 4. Let v = (x, y, z) and v 0 = (x 0 , y 0 , z0 ) be elements of Z3 .

Show that v, v 0 are be extended to a basis if and only if
gcd(xy 0 − yx 0 , xz0 − zx 0 , yz0 − zy 0 ) = 1.

Solution. The vector v 00 = (x 00 , y 00 , z00 ) extends v, v 0 to a basis if and

only if
 
x y z
det  x 0 y 0 z0 
 
x 00 y 00 z00
is a unit in Z. The units in Z are ±1 so, by the cofactor expansion, the
latter is equivalent the existence of x 00 , y 00 , z00 in Z such that
x 00 (yz0 − zy 0 ) − y 00 (xz0 − zx 0 ) + z00 (xy 0 − yx 0 ) = ±1,
which is the statement that gcd(xy 0 − yx 0 , xz0 − zx 0 , yz0 − zy 0 ) = 1.
The result follows. 

Problem 5. How many subgroups K ⊆ Z3 have the property that Z3 /K

is isomorphic to (Z/2Z)3 ? For how many K is Z3 /K isomorphic to
Z/2Z × Z/4Z × Z/8Z?

Solution. The group GL(Zn ) of group automorphisms of Zn acts on

subgroups of Zn . If K is a subgroup of Zn and φ is an automorphism
of Zn , then Zn /K is isomorphic to Zn /φ(K). Conversely, assume that
K1 and K2 are subgroups of Zn such that Zn /K1 and Zn /K2 are iso-
morphic to each other. For each i there is a basis vi,1 , . . . , vi,n of Zn
and integers ai,1 , . . . , ai,`i (for some `i with 0 ≤ `i ≤ n) such that
ai,1 vi,1 , . . . , ai,`i vi,`i is basis of Ki . Then `1 = `2 and after reorder-
ing if necessary a1,j = a2,j for each j. The automorphism v1,j , v2,j
of Zn sends K1 to K2 , and in particular K1 and K2 are conjugates under
the action of GL(Zn ). We have proved that for any subgroup K0 of Zn ,
GL(Zn ) acts transitively on the collection of subgroups K of Zn such
that Zn /K is isomorphic to Zn /K0 . Therefore the set of such subgroups
is in bijective correspondence with the collection of cosets in GL(Zn )
of the stabilizer of K0 .
 10

Let K0 be the subgroup 2Z3 of Z3 , which satisfies Z3 /K0 › (Z/2Z)3 .

Note that K0 is a characteristic subgroup (because it is fully invariant)
so the stabilizer of K0 is all of GL(Z3 ). Hence there is precisely one
subgroup K of Z3 such that Z3 /K › (Z/2Z)3 .
Let K0 be the subgroup of Z3 generated by (2, 0, 0), (0, 4, 0), and
(0, 0, 8), which satisfies Z3 /K0 › Z/2Z × Z/4Z × Z/8Z. If the matrix
 
c11 c12 c13
c21 c22 c23 
 
c31 c32 c33

represents an automorphism φ of Z3 with respect to the standard ba-

sis, then the condition that φ(K0 ) = K0 (equivalently φ(K0 ) ⊆ K0 ) is
that        
c11 c12 c13 2Z
2 c21  , 4 c22  , 8 c23  ∈ 4Z ,
       
c31 c32 c33 8Z
that is 2 | c21 , 2 | c32 , and 4 | c31 . Thus the subgroup H of G = GL3 (Z)
whose lower left coefficients satisfy the above divisibility conditions is
the isotropy subgroup of K0 under the action of G. We wish to calculate
the order of G/H. The kernel of the surjective2 group homomorphism
GL3 (Z) → GL3 (Z/4Z),
is a normal subgroup of GL3 (Z) contained in H. The above homomor-
phism then identifies G/H as the quotient of GL3 (Z/4Z) by the image
H of H in GL3 (Z/4Z):
 
c11 c12 c13
c21 c22 c23 
 
0 c32 c33
in GL3 (Z/4Z) with 2 | c21 and 2 | c32 . One can check that H is pre-
cisely determined by the conditions c21 , c32 ∈ {0, 2} and c11 , c22 , c33 ∈
{−1, +1}. Therefore the cardinality of H is 25 · 43 = 211 .
The surjective3 homomorphism
GL3 (Z/4Z) → GL3 (Z/2Z)

2
Surjectivity follows, for example, by factoring into a unimodular times a diago-
nal times a unimodular, and then noting that every unit in Z/4Z has a lift to a unit
of Z, something that is only true for Z/2a 3b Z with a ≤ 2 and b ≤ 1 and at least one
inequality strict. In general the image under reduction modulo n is the subgroup
of diagonal matrices with diagonal entries in (Z/nZ)× along with all possible left
and right translates by unimodular matrices.
3
Reduction Z/p r +1 Z → Z/p r Z always induces a surjective map on invertible
matrices because every lift of a unit in Z/p r Z is a unit in Z/p r +1 Z.
 11

has kernel the set of

   
1 0 0 c11 c12 c13
0 1 0 + c21 c22 c23 
   
0 0 1 c31 c32 c33

for which 2 | cij for all i and j, which has order 29 . Since GL3 (Z/2Z) has
order (23 − 20 )(23 − 21 )(23 − 22 ) = 23 · 3 · 7, it follows that GL3 (Z/4Z)
has order 212 · 3 · 7. Consequently the index of H in GL3 (Z/4Z) is
212 · 3 · 7/211 = 2 · 3 · 7.
Finally, the number of subgroups K of Z3 such that Z3 /K is isomor-
phic to Z/2Z × Z/4Z × Z/8Z is 2 · 3 · 7 = 42. 
Note. The task is to compute the number of Hermite normal forms
for a given Smith normal form, which has been computed in general
when the invariant factors are powers of a single prime. For any sub-
group L of Zn of finite index [Zn : L], [Zn : L]Zn ⊆ L by Lagrange’s
theorem applied to the quotient Zn /L. In particular, by the third iso-
morphism theorem, the set of subgroups K such that Z3 /K is iso-
morphic to Z/2Z × Z/4Z × Z/8Z is in bijection with the set of sub-
groups H of Z3 /26 Z3 = (Z/26 Z)3 such that (Z/26 Z)3 /H is isomorphic
to Z/2Z × Z/4Z × Z/8Z via K , K/26 Z3 ⊆ Z3 /26 Z3 = (Z/26 Z)3 .
We now show that for finite abelian groups, the number of sub-
groups of a certain cotype (isomorphism class of cokernel) equals the
number of subgroups of that type (isomorphism class). For a finite
abelian group G, write G∨ = Hom(G, T) for the dual group where T is
the multiplicative group of complex numbers of modulus 1. To each
subgroup H of the finite abelian group G associate H 0 , the annihilator
of H in G, which is the subgroup (G/H)∨ → G∨ . Then H 0 › G/H and
G∨ /H 0 › H so H , H 0 takes a group H of type K1 and cotype K2 to a
group H 0 of type K2 and cotype K1 . By Pontryagin duality,

H 00 = (G∨ /(G/H)∨ )∨ = H ∨∨ = H

so the association is a bijection from the subgroups of G of type K1

and cotype K2 with subgroups of G∨ of type K2 and cotype K1 , but
G∨ is (non-canonically) isomorphic to G so this implies the desired
equality of the number of subgroups of given type and the number of
subgroups of the same cotype.
We must now compute the number of subgroups of (Z/26 Z)3 iso-
morphic to Z/2Z × Z/4Z × Z/8Z. Represent an element of (Z/26 Z)3 as
a row vector with 3 components each in (0, 26 ]. Then a subgroup can
be represented as a matrix of width 3 whose rows represent the gen-
erators. The p 4 + 2p 3 + 2p 2 + p subgroups of (Z/p 6 Z)3 isomorphic to
 12

Z/pZ × Z/p 2 Z × Z/p 3 Z are represented by the following matrices.

p3 (0, p]p 3 (0, p 2 ]p 3 p 3 (0, p 2 ]p 3 (0, p]p 3

   
 6
p4 (0, p]p 4  ,
 6
p p p5 p4  ,
 
p6 p6 p5 p 6
p5 p6
 4
p 3 (0, p 2 ]p 3 (0, p]p 4 p 3 (0, p]p 3
  
p
 4
p p 6 (0, p]p 4  ,  p5 p6 p4  ,
  
p6 p6 p5 p5 p6 p6
 4
(0, p]p 4 p 3 (0, p]p 4 p 4 p3
  
p
 4
p (0, p]p 4 p 6  ,  p5 p4 p6  .
  
p6 p 5
p 6
p 5
p6 p 6

Another possibility is to write down the possible Hermite normal

forms and determine which have the desired Smith normal form. A
subgroup of Zn generated by at most m elements can be represented
as the row space of an m × n matrix, but two matrices represent the
same subgroup when they are left equivalent (in the same orbit under
the action of the n × n unimodular matrices). Therefore subgroups
of Zn correspond to orbits under the action of unimodular matrices,
that is equivalence classes of left equivalent matrices. There is exactly
one Hermite normal form in each equivalence class so subgroups of
Zn generated by at most m elements are in bijective correspondence
with m × n Hermite normal forms.
In particular each subgroup K ⊆ Z3 is the row space of a unique
3 × 3 matrix in Hermite normal form. When Z3 /K is finite, its order is
the product of the diagonal entries of the Hermite normal form matrix
corresponding to K. Therefore the index 2 · 4 · 8 = 26 subgroups of Z3
are each uniquely expressable as the row space of a matrix of the form
 
c11 0 0
c21 c22 0 
 
c31 c32 c33

where each cij is nonnegative, c11 c22 c33 = 26 , c21 < c22 , c31 , c32 < c33 .
By considering the diagonalized form, we see that if di is the greatest
common divisor of the set of i × i minor matrix determinants, then
Z3 /K is isomorphic to Z/d1 × Z/d−1 −1
1 d2 Z × Z/d2 d3 Z. We want d1 = 2,
d2 = 23 , d3 = 26 . In particular 2 | d1 so each cij must be even and then
the possible diagonal entries are permutations of 2, 2, 16 or 2, 4, 8 or
4, 4, 4. We already see that 2, 2, 16 is impossible for in that case a
2 × 2 minor matrix would have determinant 4, but we want 8 | d2 .
In the case of matrices with diagonal entries 4, 4, 4 we find that the
 13

possibilities for the lower left are

" # " # " # " # " #
0 2 0 0 2
, , , , .
0 2 0 0 2 0 2 2 2 0

Finally we consider the six possible diagonals with entries 2, 4, 8. In

this case the only condition on the Hermite normal forms is that the
entries in the lower left are even (this ensures d1 = 2 since 2 already
appears on the diagonal) and every 2 × 2 minor matrix has determi-
nant determinant a multiple of 8 (which is sufficient to ensure d2 = 8
since the triangular minor matrix with diagonal entries 2 and 4 has
determinant 8).
• Order: 2, 4, 8. The possibilities for the lower left are 00 z for
 

z ∈ {0, 2}, but we must have 8 | 2z so z = 0. The 1 possibility

is 00 0 .
 

• Order: 2, 8, 4. The possibilities for the lower left are 00 z for

 

z ∈ {0, 2, 4, 6}, but we must have 8 | 2z, that is 4 | z. Thus in

this case the possibilities are 00 0 and 00 4 .
   

• Order: 4, 2, 8. The possibilities for the lower left are of the

x 
form y 0 so in order that the lower left minor matrix have
determinant divisible by 8 the lower left entry y cannot be 2
and must be 0. Thus the possibilities for the lower left are
x 
0 0 for x ∈ {0, 2} so there are 2 in this case.
x
• Order: 4, 8, 2. If the lower left is y z then 2x must be di-
visible by 8 so x = 0, but otherwise there are no further h i re-
0
strictions. Thus the possibilities for the lower left are y z for
x ∈ {0, 2, 4, 6} and y ∈ {0, 2} so there are 8 in this case.
x 
• Order: 8, 2, 4. If the lower left is y 0 then y ∈ {0, 4} and
x ∈ {0, 2, 4, 6} so there are 8 in this case.
x 
• Order: 8, 4, 2. If the lower left is y z then 8 | 2x so x ∈ {0, 4},
y ∈ {0, 2, 4, 6}, and z ∈ {0, 2} so there are 16 in this case.
We have found the 42 sublattices and found generators for them.

Problem 6. Let M be an n×m matrix with entries in the PID R. Let K be

the submodule of R n spanned by the columns, and S the submodule
of R m spanned by rows. Prove that the torsion submodules of R n /K
and R m /S are isomorphic.

Solution. Let A be an element of GLn (R). Then AM is obtained from

M by elementary row operations so the row spans of M and AM are
the same. Write ϕA for the element of GL(R n ) corresponding to A
with respect to the standard basis. We have R n /K › ϕA (R n )/ϕA (K) =
 14

R n /K 0 where K 0 is the span of the columns of AM. Therefore the result

is invariant under left multiplication by an element of GLn (R).
Let B be an element of GLm (R). Then MB is obtained from M by
elementary column operations so the column spans of M and MB are
the same. Write ψB for the element of GL(R m ) corresponding to A
with respect to the standard basis. We have R m /S › ψA (R m )/ψA (S) =
R m /S 0 where S 0 is the span of the columns of MB. Therefore the result
is invariant under left multiplication by an element of GLm (R).
We now show that there exist matrices A in GLn (R) and B in GLm (R)
such that AMB is zero except for the diagonal entries (which can be
chosen so that each diagonal entry is divisible by the one before it).4
This is equivalent to performing elementary row and column opera-
tions on M to obtain the desired form. Consider a fixed column (resp.
row). If the ith and jth entries are a and b, then it is possible to per-
form elementary operations to rows (resp. columns) i and j to trans-
form the ith entry to a generator of the ideal generated by a and b
and transform the jth entry to 0. With these operations we may make
every entry in the first row or the first column to be zero, except pos-
sibly the (i, j) entries for i + j ≤ 3. Along with induction, this reduces
to the case where m = n = 2. Row operations can transform a 2 × 2
matrix to the form ∗0 ∗


∗
and column operations can transform a 2 × 2
0
matrix to the form ∗ ∗ ∗ . Alternately apply the two operations above.
Each time, the upper left entry is replaced by a divisor of itself. Since
increasing sequences of ideals must stabilize (consider a generator of
the union of the sequence), the ideal generated by the upper left entry
eventually stabilizes. Then the upper left entry divides both the lower
left and upper right entries and so the matrix can be made diagonal.
When M has all nondiagonal entries equal to 0, the result is clear, so
this completes the proof. 

4
I believe this was proved in lecture.