TENSOR PRODUCTS

KEITH CONRAD
1. Introduction
Let R be a commutative ring and M and N be R-modules. Formation of their direct
sum M N is an addition operation on modules. We introduce now a product operation,
called the tensor product M
R
N. To start o, we will describe roughly what a tensor
product of modules looks like. The rigorous denitions will come in Section 3.
The tensor product M
R
N is an R-module spanned by symbols m n satisfying
distributive laws:
(1.1) (m+m

) n = mn +m

n, m(n +n

) = mn +mn

.
Also multiplication by any r R is associative with on both sides:
(1.2) r(mn) = (rm) n = m(rn).
(So the notation rmn is unambiguous: it is both (rm) n and r(mn).) The formulas
(1.1) and (1.2) should be contrasted with those for the direct sum M N, where
(m+m

, n) = (m, n) + (m

, 0), r(m, n) = (rm, rn).

In M N, every (m, n) decomposes as (m, 0) + (0, n), but m n in M
R
N does not
break apart in any general way. While every element of M N is a pair (m, n), there are
more elements of M
R
N than the products mn. All elements of M
R
N are R-linear
combinations
1
r
1
(m
1
n
1
) +r
2
(m
2
n
2
) + +r
k
(m
k
n
k
),
where k 1, r
i
R, m
i
M, and n
i
N. Since r
i
(m
i
n
i
) = (r
i
m
i
) n
i
, we can rename
r
i
m
i
as m
i
and write the linear combination as a sum
(1.3) m
1
n
1
+m
2
n
2
+ +m
k
n
k
.
In M N, equality is easy to dene: (m, n) = (m

, n

) if and only if m = m

and n = n

.
But when are two sums of the form (1.3) equal in M
R
N? This is not easy to say in terms
of the description of the tensor product above, except in one special case: M and N are
nite free R-modules with bases e
i
and f
j
. In this case, M
R
N has basis e
i
f
j
:
every element of M
R
N is a unique sum

i,j
c
ij
e
i
f
j
and two sums of this special form
are equal only when the coecients of like terms are equal.
To describe equality in M
R
N when there arent bases, we need to use a universal
mapping property of the tensor product. In fact, the tensor product is the rst fundamental
concept in algebra which can be used (in the most general case) only through its universal
mapping property, which is: M
R
N is the universal object that turns bilinear maps out
of M N into linear maps.
1
Compare with the polynomial ring R[X, Y ], whose terms are not only the products f(X)g(Y ), but sums
of such products like
P
i,j
aijX
i
Y
j
. It turns out that R[X, Y ]

= R[X] R R[Y ], so the comparison is not
merely an analogy but a special case (Example 4.10).
1
2 KEITH CONRAD
After a discussion of bilinear (and multilinear) maps in Section 2, the denition and
construction of tensor products is presented in Section 3. Examples of tensor products are
in Section 4. In Section 5 we will show how the tensor product interacts with some other
constructions on modules. Section 6 describes the important operation of base extension,
which is a process of turning an R-module into an S-module where S is another ring.
2. Bilinear Maps
We described M
R
N as sums (1.3) subject to the rules (1.1) and (1.2). The intention
is that M
R
N be the freest object built out of M and N subject to (1.1) and (1.2).
The essence of (1.1) and (1.2) is bilinearity, which well discuss before getting back to the
tensor product.
A function B: M N P, where M, N, and P are R-modules, is called bilinear when
it is linear in each coordinate with the other one xed:
B(m
1
+m
2
, n) = B(m
1
, n) +B(m
2
, n), B(rm, n) = rB(m, n),
B(m, n
1
+n
2
) = B(m, n
1
) +B(m, n
2
), B(m, rn) = rB(m, n).
So B(, n) is a linear map M P for each n and B(m, ) is a linear map N P for each
m. Here are some examples:
(1) The dot product v w on R
n
is a bilinear function R
n
R
n
R.
(2) For any A M
n
(R), the function 'v, w` = v Aw is a bilinear map R
n
R
n
R.
It reduces to the dot product when A = I
n
.
(3) The cross product v w is a bilinear function R
3
R
3
R
3
.
(4) Multiplication R R R is bilinear.
(5) For an R-module M, scalar multiplication R M M is bilinear.
(6) The determinant det: M
2
(R) R is a bilinear function of the columns.
(7) The dual pairing M

M R given by (, m) (m) is bilinear.

(8) The function M

N Hom
R
(M, N) given by (, n) [x (x)n] is bilinear.
(9) If MN
B
P is bilinear and P
L
Q is linear, the composite MN
LB
Q
is bilinear. (This is a very important example. Check it!)
(10) If f : M Hom
R
(N, P) is linear then B: MN P given by B(m, n) = f(m)(n)
is bilinear.
(11) In our naive description of M
R
N, the expression m n is bilinear in m and n.
That is, the function M N M
R
N given by (m, n) mn is bilinear.
Even though elements of MN and MN are written in the same way, as pairs (m, n),
bilinear functions M N P should not be confused with linear functions M N P.
For example, addition as a function R R R is linear, but as a function R R R
it is not bilinear. Multiplication as a function R R R is bilinear, but as a function
RR R it is not linear. Linear functions are generalized additions and bilinear functions
are generalized multiplications.
An extension of bilinearity is multilinearity. For R-modules M
1
, . . . , M
k
, a function
f : M
1
M
k
M is called multilinear when f(m
1
, . . . , m
k
) is linear in each m
i
with the other coordinates xed. We use the label k-multilinear if the number of factors
has to be mentioned, so 2-multilinear means bilinear.
Here are a few examples of multilinear functions:
(1) The scalar triple product u (v w) is trilinear R
3
R
3
R
3
R.
(2) The function f(u, v, w) = (u v)w is trilinear R
n
R
n
R
n
R
n
.
TENSOR PRODUCTS 3
(3) The function M

M N N given by (, m, n) = (m)n is trilinear.

(4) If B: M N P and B

: P Q T are bilinear then M N Q T by

(m, n, q) B

(B(m, n), q) is trilinear.

(5) Multiplication R R R with k factors is multilinear.
(6) The determinant det: M
n
(R) R, as a function of the columns, is n-multilinear.
(7) If M
1
M
k
f
M is multilinear and M
L
N is linear then the composite
M
1
M
k
Lf
N is multilinear.
The R-linear maps M N are an R-module Hom
R
(M, N) under addition of functions
and R-scaling. The R-bilinear maps M N P form an R-module Bil
R
(M, N; P) in the
same way. However, unlike linear maps, bilinear maps are missing some features:
(1) There is no kernel of a bilinear map M N P since M N is not a module.
(2) The image of a bilinear map M N P need not form a submodule.
Example 2.1. Bilinear maps MN R can be created using M

, N

, and multiplication
in R: for M

and N

, set B
,
: M N R by B
,
(m, n) = (m)(n). Each
B
,
is bilinear. Also B: M

Bil
R
(M, N; R) given by B(, ) = B
,
is bilinear.
(So B is a bilinear map whose values are bilinear maps.) The sum
1
(m)
1
(n)+
2
(m)
2
(n)
is bilinear in m and n but usually not expressible as (m)(n), so B is not surjective. And
B is generally not injective: for any u R

, B(, ) = B(u, (1/u)).

Lets look at the special case that M and N are nonzero nite free R-modules, with re-
spective bases e
1
, . . . , e
k
and f
1
, . . . , f

. Any bilinear map B: MN R is determined

by its values on the basis pairs (e
i
, f
j
) since B(

i
a
i
e
i
,

j
b
j
f
j
) =

i,j
a
i
b
j
B(e
i
, f
j
). Since
B
e

i
,f

j
on MN equals 1 at (e
i
, f
j
) and 0 at other basis pairs, B =

i,j
B(e
i
, f
j
)B
e

i
,f

j
for
every B in Bil
R
(M, N; R): both sides agree at basis pairs, so they agree everywhere since
both sides are bilinear. This shows the functions B
e

i
,f

j
= B(e

i
, f

j
) span Bil
R
(M, N; R).
They are also linearly independent: if

i,j
c
ij
B
e

i
,f

j
= O then evaluating at the basis pair
(e
i
, f
j
) shows c
i

j
= 0, so all coecients are 0. Thus Bil
R
(M, N; R) has basis B
e

i
,f

j
.
Because B sends basis pairs (e

i
, f

j
) in M

to a basis of Bil
R
(M, N; R), we
would like to say B is in some sense an isomorphism, but this makes no sense: the domain
of B is not an R-module, and B is neither injective nor surjective. These defects will
be rectied by the tensor product in Example 4.11: B can be converted into a linear
map L: M

R
N

Bil
R
(M, N; R) which is an isomorphism for nite free M and N.
The non-injective feature B(, ) = B(u, (1/u))), for instance, will wipe out because
= u(1/u) in M

R
N

by (1.2) even though (, ) = (u, (1/u)) in M

.
3. Construction of the Tensor Product
Any bilinear map M N P to an R-module P can be composed with a linear map
P Q to get a map M N Q which is bilinear.
P
linear

M N
bilinear

v
v
v
v
v
v
v
v
v
composite is bilinear!

G
G
G
G
G
G
G
G
G
Q
4 KEITH CONRAD
We will construct the tensor product as a solution to the following universal mapping
problem: nd an R-module T and bilinear map b: M N T such that all bilinear maps
out of M N are composites of the single bilinear map b with all linear maps out of T.
T
linear?

M N
b

w
w
w
w
w
w
w
w
w
bilinear

H
H
H
H
H
H
H
H
H
P
This turns the task of constructing bilinear maps out of MN to that of constructing lin-
ear maps out of T. It is analogous to the universal mapping property of the abelianization
G/[G, G] of a group G: homomorphisms G A with abelian A are the same as homo-
morphisms G/[G, G] A because every homomorphism G
f
A is the composite of the
canonical homomorphism G

G/[G, G] with a unique homomorphism G/[G, G]
e
f
A.
G/[G, G]
e
f

v
v
v
v
v
v
v
v
v
f

I
I
I
I
I
I
I
I
I
I
A
Denition 3.1. The tensor product M
R
N is an R-module equipped with a bilinear map
MN

M
R
N such that for any bilinear map MN
B
P there is a unique linear
map M
R
N
L
P making the following diagram commute.
M
R
N
L

M N

q
q
q
q
q
q
q
q
q
q
B

M
M
M
M
M
M
M
M
M
M
M
M
P
While the functions in the universal mapping property for G/[G, G] are all homomor-
phisms (out of G and out of G/[G, G]), functions in the universal mapping property for
M
R
N are not all of the same type: those out of M N are bilinear and those out of
M
R
N are linear.
The denition of the tensor product involves not just a new module M
R
N, but also
a distinguished bilinear map to it: M N

M
R
N. This is similar to the universal
mapping property for the abelianization G/[G, G], which requires not just G/[G, G] but also
the homomorphism G

G/[G, G] through which all homomorphisms from G to abelian
groups factor. The universal mapping property makes no sense without xing this extra
piece of information.
TENSOR PRODUCTS 5
Before constructing the tensor product, lets show all constructions of it are essentially
the same. Suppose there are R-modules T and T

and bilinear maps M N

b
T and
MN
b

with the universal mapping property of the tensor product. From universality
of M N
b
T, the bilinear map M N
b

factors uniquely through T: there is a

unique linear map f : T T

making
(3.1) T
f

M N
b

v
v
v
v
v
v
v
v
v
b

H
H
H
H
H
H
H
H
H
T

commute. From universality of M N

b

, the bilinear map M N

b
T factors
uniquely through T

: there is a unique linear map f

: T

T making
(3.2)
T

M N
b

v
v
v
v
v
v
v
v
v
b

H
H
H
H
H
H
H
H
H
T
commute. We combine (3.1) and (3.2) into the commutative diagram
T
f

M N
b

H
H
H
H
H
H
H
H
H
b

v
v
v
v
v
v
v
v
v
T

T
Removing the middle, we have the commutative diagram
(3.3) T
f

M N
b

w
w
w
w
w
w
w
w
w
b

G
G
G
G
G
G
G
G
G
T
From universality of (T, b), a unique linear map T T ts in (3.3). The identity map
works, so f

f = id
T
. Similarly, f f

= id
T
by stacking (3.1) and (3.2) together in the
other order. Thus T and T

are isomorphic R-modules by f and also f b = b

, which
means f identies b with b

. So any two tensor products of M and N can be identied with

6 KEITH CONRAD
each other in a unique way compatible
2
with the distinguished bilinear maps to them from
M N.
Theorem 3.2. A tensor product of M and N exists.
Proof. Consider M N simply as a set. We form the free R-module on this set:
F
R
(M N) =

(m,n)MN
R
(m,n)
.
Let D be the submodule of F
R
(M N) spanned by all the elements

(m+m

,n)

(m,n)

(m

,n)
,
(m,n+n

(m,n)

(m,n

)
,
(rm,n)

(m,rn)
,
r
(m,n)

(rm,n)
, r
(m,n)

(m,rn)
.
The quotient module by D will serve as the tensor product: set
M
R
N := F
R
(M N)/D.
We write the coset
(m,n)
+D in M
R
N as mn.
(Notice how large F
R
(M N) can be. If R = M = N = R then F
R
(R R) is a
direct sum of R
2
-many copies of R, even though well see in Theorem 4.8 that the quotient
R-module R
R
R is 1-dimensional!)
We need to write down a bilinear map M N M
R
N and show all bilinear maps
out of M N factor uniquely through this one. From the denition of D, we get relations
in F
R
(M N)/D like

(m+m

,n)

(m,n)
+
(m

,n)
mod D,
which is the same as
(m+m

) n = mn +m

n
in M
R
N. Similarly, m(n + n

) = mn + mn

and r(mn) = rmn = mrn

in M
R
N. This shows the function M N

M
R
N given by (m, n) m n is
bilinear. (No other function M N M
R
N will be considered except this one.)
Now suppose P is any R-module and M N
B
P is a bilinear map. Treating M N
simply as a set, so B is just a function on this set (ignore its bilinearity), the universal
mapping property of free modules extends B from a function M N P to a linear
function : F
R
(M N) P with (
(m,n)
) = B(m, n), so the diagram
F
R
(M N)

M N
(m,n)
(m,n)

p
p
p
p
p
p
p
p
p
p
p
B

O
O
O
O
O
O
O
O
O
O
O
O
O
P
commutes. We want to show makes sense as a function on M
R
N, which means showing
ker contains D. From the bilinearity of B,
B(m+m

, n) = B(m, n) +B(m, n

), B(m, n +n

) = B(m, n) +B(m, n

),
rB(m, n) = B(rm, n) = B(m, rn),
2
The universal mapping property is not about modules T per se, but about pairs (T, b).
TENSOR PRODUCTS 7
so
(
(m+m

,n)
) = (
(m,n)
) +(
(m

,n)
), (
(m,n+n

)
) = (
(m,n)
) +(
(m,n

)
),
r(
(m,n)
) = (
(rm,n)
) = (
(m,rn)
).
Since is linear, these conditions are the same as
(
(m+m

,n)
) = (
(m,n)
+
(m

,n)
), (
(m,n+n

)
) = (
(m,n)
+
(m,n

)
),
(r
(m,n)
) = (
(rm,n)
) = (
(m,rn)
).
Therefore the kernel of contains all the generators of the submodule D, so induces a
linear map L: F
R
(M N)/D P where L(
(m,n)
+ D) = (
(m,n)
) = B(m, n), so the
diagram
F
R
(M N)/D
L

M N
(m,n)m,n+D

n
n
n
n
n
n
n
n
n
n
n
n
B

P
P
P
P
P
P
P
P
P
P
P
P
P
P
P
commutes. Since F
R
(MN)/D = M
R
N and
(m,n)
+D = mn, the above diagram is
(3.4) M
R
N
L

M N

q
q
q
q
q
q
q
q
q
q
B

M
M
M
M
M
M
M
M
M
M
M
M
P
.
So every bilinear map B out of M N comes from a linear map L out of M
R
N such
that L(mn) = B(m, n) for all m M and n N.
It remains to show the linear map M
R
N
L
P in (3.4) is the only one which makes
(3.4) commute. We go back to the denition of M
R
N as a quotient of the free module
F
R
(MN). From the construction of free modules, every element of F
R
(MN) is a nite
sum
r
1

(m
1
,n
1
)
+ +r
k

(m
k
,n
k
)
.
The reduction map F
R
(MN) F
R
(MN)/D = M
R
N is linear, so every element of
M
R
N is a nite sum
(3.5) r
1
(m
1
n
1
) + +r
k
(m
k
n
k
),
This means the particular elements m n in M
R
N span it as an R-module. Therefore
linear maps out of M
R
N are completely determined by their values on all m n. So
there is at most one linear map M
R
N P with the eect mn B(m, n). Since we
have created a linear map out of M
R
N with this eect in (3.4), it is the only one.
8 KEITH CONRAD
Having shown a tensor product of M and N exists,
3
its essential uniqueness lets us
call M
R
N the tensor product rather than a tensor product. Dont forget that the
construction involves not simply the module M
R
N but also the distinguished bilinear
map M N

M
R
N given by (m, n) m n, through which all bilinear maps
out of M N factor. We call this distinguished map the canonical bilinear map to the
tensor product. We call general elements of M
R
N tensors, and will denote them by
the letter t. Tensors of the form m n are called elementary tensors. (Other names for
elementary tensors are simple tensors, decomposable tensors, and pure tensors.) Just as
most elements of the free module F
R
(A) on a set A are not of the form
a
but are linear
combinations of these, most tensors in M
R
N are not elementary tensors but are linear
combinations of elementary tensors. Actually, all tensors are sums of elementary tensors
since r(mn) = (rm) n. This shows all elements of M
R
N have the form (1.3).
The role of elementary tensors among all tensors is like that of separable solutions
f(x)g(y) to a PDE among all solutions. Even if not all solutions to a PDE are separa-
ble, one rst seeks separable solutions and then tries to form the general solution as a sum
(perhaps an innite series; it is analysis) of separable solutions.
From now on forget the explicit construction of M
R
N as the quotient of an enormous
free module F
R
(M N). It will confuse you more than its worth to try to think about
M
R
N in terms of its construction. What is more important to remember is the universal
mapping property of the tensor product.
Lemma 3.3. Let M and N be R-modules with respective spanning sets x
i

iI
and y
j

jJ
.
The tensor product M
R
N is spanned linearly by the elementary tensors x
i
y
j
.
Proof. An elementary tensor in M
R
N has the form m n. Write m =

i
a
i
x
i
and
n =

j
b
j
y
j
, where the a
i
s and b
j
s are 0 for all but nitely many i and j. From the
bilinearity of ,
mn = (

i
a
i
x
i
) (

j
b
j
y
j
) =

i,j
a
ij
x
i
y
j
is a linear combination of the tensors x
i
y
j
. So every elementary tensor is a linear
combination of the elementary tensors x
i
y
j
, which means every tensor is such a linear
combination too.
Example 3.4. Let e
1
, . . . , e
k
be the standard basis of R
k
. The R-module R
k

R
R
k
is
linearly spanned by the k
2
elementary tensors e
i
e
j
. We will see later (Theorem 4.8) that
these elementary tensors are a basis.
Theorem 3.5. In M
R
N, m0 = 0 and 0 n = 0.
Proof. Since m n is linear in n with m xed, m 0 = m (0 + 0) = m 0 + m 0.
Subtracting m 0 from both sides, m 0 = 0. That 0 n = 0 follows by a similar
argument.
3
What happens if R is a noncommutative ring? If M and N are left R-modules and B is bilinear on
MN then for any m M, n N, and r and s in R, rsB(m, n) = rB(m, sn) = B(rm, sn) = sB(rm, n) =
srB(m, n). Usually rs = sr, so asking that rsB(m, n) = srB(m, n) for all m and n puts us in a delicate
situation! The correct tensor product M R N for noncommutative R uses a right R-module M, a left R-
module N, and a middle-linear map B where B(mr, n) = B(m, rn). In fact M R N is not an R-module
but just an abelian group! We will not discuss this further.
TENSOR PRODUCTS 9
Example 3.6. If A is a nite abelian group, Q
Z
A = 0 since every elementary tensor is
0: for a A, let na = 0 for some positive integer n. Then in Q
Z
A, r a = n(r/n) a =
r/n na = r/n 0 = 0. Every tensor is a sum of elementary tensors, so all tensors are
0. In fact we dont need niteness of A in this argument, but rather than each element of
A has nite order. The group Q/Z has this property, so Q
Z
(Q/Z) = 0. (For instance,
(2/5) (1/3) = 0, so we can have mn = 0 without m or n being 0.)
By a similar argument, Q/Z
Z
Q/Z = 0.
Since M
R
N is spanned additively by elementary tensors, any linear (or just additive)
function out of M
R
N is determined at all tensors from its values on elementary tensors.
This is why linear maps on tensor products will in practice only be described by their
values on elementary tensors. It is similar to describing a linear map between nite free
modules using a matrix, which tells you the values of the map only on basis vectors, but
this information is enough to determine the linear map everywhere.
There is a key dierence between elementary tensors and basis vectors: elementary tensors
have lots of linear relations. Using elementary tensors to describe all tensors is like describing
the vectors in R
2
as linear combinations of (1, 0), (2, 3), (8, 4), and (1, 5). A linear
map out of R
2
is determined by its values on these four vectors, but those values are not
independent: they have to satisfy every linear relation the four vectors satisfy because a
linear map preserves linear relations. Using this linearly dependent spanning set requires
keeping track of linear relations to be sure constructions made in terms of this spanning set
are well-dened. Nobody wants to do that, which is why using bases is so convenient. But
you usually cant get away from the zillions of linear relations among elementary tensors,
and that is why a random function on elementary tensors generally does not extend to a
linear map on the tensor product.
For that matter, functions on elementary tensors themselves have a well-denedness issue:
the function f(mn) = m+n, for instance, makes no sense since mn = (m) (n)
but m + n is usually not m n. The only way to create linear maps on M
R
N is
with the universal mapping property of the tensor product (it creates linear maps out of
bilinear maps), because all the linear relations among elementary tensors are built into the
universal mapping property of M
R
N. There will be a lot of practice with this in Section
4. Understanding how the universal mapping property of the tensor product can be used
to compute (nonzero) examples and to prove properties of the tensor product is about the
best way to get used to the tensor product; after all, if youre incapable of writing down
functions out of M
R
N, you dont understand M
R
N.
Lets address a few beginner questions about the tensor product:
Questions
(1) What is mn?
(2) What does it mean to say mn = 0?
(3) What does it mean to say M
R
N = 0?
(4) What does it mean to say m
1
n
1
+ +m
k
n
k
= m

1
n

1
+ +m

?
(5) Is there a way to picture the tensor product?
Answers
(1) A logically correct answer is: m n is the image of (m, n) M N under the
canonical bilinear map M N

M
R
N that is part of the denition of the
tensor product. That may seem like too formal an answer. Heres another answer,
which is not a denition but is more closely aligned with how mn actually occurs
10 KEITH CONRAD
in practice: mn is that element of M
R
N at which the linear map M
R
N P
corresponding to a bilinear map M N
B
P takes the value B(m, n). Review
the proof of Theorem 3.2 and check this property of mn really holds.
(2) We have m n = 0 if and only if every bilinear map out of M N vanishes at
(m, n). Indeed, if mn = 0 then for any bilinear map B: M N P we have a
commutative diagram
M
R
N
L

M N

q
q
q
q
q
q
q
q
q
q
B

M
M
M
M
M
M
M
M
M
M
M
M
P
for some linear map L, so B(m, n) = L(m n) = L(0) = 0. Conversely, if every
bilinear map out of M N sends (m, n) to 0 then the canonical bilinear map
M N M
R
N, which is a particular example, sends (m, n) to 0. Since this
bilinear map actually sends (m, n) to mn, we obtain mn = 0.
To show a particular elementary tensor mn is not 0, nd a bilinear map B out
of M N such that B(m, n) = 0. We will use this idea in Theorem 4.8.
(3) The tensor product M
R
N is 0 if and only if every bilinear map out of M N is
identically 0. First suppose M
R
N = 0. Then every elementary tensor m n is
0, so B(m, n) = 0 for any bilinear map out of M N by the answer to the second
question. Thus B is identically 0. Next suppose every bilinear map out of M N
is identically 0. Then the canonical bilinear map M N

M
R
N, which is a
particular example, is identically 0. Since this function sends (m, n) to m n, we
have mn = 0 for all m and n. Since M
R
N is additively spanned by all mn,
the vanishing of all elementary tensors implies M
R
N = 0.
To show M
R
N = 0, nd a bilinear map on M N which is not identically 0.
(4) We have

k
i=1
m
i
n
i
=

j=1
m

j
n

j
if and only if for all bilinear maps B out of
M N,

k
i=1
B(m
i
, n
i
) =

j=1
B(m

j
, n

j
). The justication is along the lines of
the previous two answers and is left to the reader.
(5) There are objects in physics that are described using tensors (stress, elasticity, elec-
tromagnetic elds), but this doesnt really lead to a picture of a tensor. Tensors
arise in physics as multi-indexed quantities which are subject to multilinear eects
under a change in coordinates.
The tensor product can be extended to allow more than two factors. Given k modules
M
1
, . . . , M
k
, there is a module M
1

R

R
M
k
which is universal for k-multilinear maps:
it admits a k-multilinear map M
1
M
k

M
1

R

R
M
k
and every k-multilinear
map out of M
1
M
k
factors through this by composition with a unique linear map
TENSOR PRODUCTS 11
out of M
1

R

R
M
k
:
M
1

R

R
M
k
unique linear

M
1
M
k

k
k
k
k
k
k
k
k
k
k
k
k
k
k
multilinear

T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
P
The image of (m
1
, . . . , m
k
) in M
1

R

R
M
k
is written m
1
m
k
. This k-fold tensor
product can be constructed as a quotient of the free module F
R
(M
1
M
k
). It can also
be constructed using tensor products of modules two at a time:
( ((M
1

R
M
2
)
R
M
3
)
R
)
R
M
k
.
The canonical k-multilinear map to this R-module from M
1
M
k
is (m
1
, . . . , m
k
)
( ((m
1
m
2
) m
3
) ) m
k
. This is not the same construction of the k-fold tensor
product using F
R
(M
1
M
k
), but it satises the same universal mapping property
and thus can serve the same purpose (all tensors products of M
1
, . . . , M
k
are isomorphic to
each other in a unique way compatible with the distinguished bilinear maps to them from
M
1
M
k
).
As an exercise, check from the universal mapping property that m
1
m
k
= 0 in
M
1

R

R
M
k
if and only if all k-multilinear maps out of M
1
M
k
vanish at
(m
1
, . . . , m
k
). The module M
1

R

R
M
k
is spanned additively by all m
1
m
k
.
Important examples of the k-fold tensor product are tensor powers M
k
:
M
0
= R, M
1
= M, M
2
= M
R
M, M
3
= M
R
M
R
M,
and so on. (The formula M
0
= R is a convention, like a
0
= 1.)
4. Examples of Tensor Products
Theorem 4.1. For positive integers a and b with d = (a, b), Z/aZ
Z
Z/bZ

= Z/dZ as
abelian groups. In particular, Z/aZ
Z
Z/bZ = 0 if and only if (a, b) = 1.
Proof. Since 1 spans Z/aZ and Z/bZ, 1 1 spans Z/aZ
Z
Z/bZ by Lemma 3.3. From
a(1 1) = a 1 = 0 1 = 0 and b(1 1) = 1 b = 1 0 = 0,
the additive order of 1 1 divides a and b, and therefore also d, so #(Z/aZ
Z
Z/bZ) d.
To show Z/aZ
Z
Z/bZ has size at least d, we create a Z-linear map from Z/aZ
Z
Z/bZ
onto Z/dZ. Since d[a and d[b, we can reduce Z/aZ Z/dZ and Z/bZ Z/dZ in the
natural way. Consider the map Z/aZZ/bZ
B
Z/dZ which is reduction mod d in each
factor followed by multiplication: B(x mod a, y mod b) = xy mod d. This is Z-bilinear, so
the universal mapping property of the tensor product says there is a (unique) Z-linear map
12 KEITH CONRAD
f : Z/aZ
Z
Z/bZ Z/dZ making the diagram
Z/aZ
Z
Z/bZ
f

Z/aZ Z/bZ

l
l
l
l
l
l
l
l
l
l
l
l
l
B

R
R
R
R
R
R
R
R
R
R
R
R
R
R
Z/dZ
commute, so f(xy) = xy. In particular, f(x1) = x, so f is onto. Therefore Z/aZ
Z
Z/bZ
has size at least d, so the size is d and were done.
Example 4.2. The abelian group Z/3Z
Z
Z/5Z is 0. Equivalently, every Z-bilinear map
Z/3Z Z/5Z A to an abelian group is identically 0.
In Z/aZ
Z
Z/bZ all tensors are elementary tensors: x y = xy(1 1) and a sum of
multiples of 1 1 is again a multiple, so Z/aZ
Z
Z/bZ = Z(1 1) = x 1 : x Z.
Notice in the proof of Theorem 4.1 how the map f : Z/aZ
Z
Z/bZ Z/dZ was created
from the bilinear map B: Z/aZ Z/bZ Z/dZ and the universal mapping property of
tensor products. Quite generally, to dene a linear map out of M
R
N that sends all
elementary tensors m n to particular places, always back up and start by dening a
bilinear map out of M N sending (m, n) to the place you want m n to go. Make sure
you show the map is bilinear! Then the universal mapping property of the tensor product
gives you a linear map out of M
R
N sending mn to the place where (m, n) goes, which
gives you what you wanted: a (unique) linear map on the tensor product with specied
values on the elementary tensors.
Theorem 4.3. For ideals I and J in R, there is a unique R-module isomorphism
R/I
R
R/J

= R/(I +J)
where x y xy. In particular, taking I = J = 0, R
R
R

= R by x y xy.
For R = Z and nonzero I and J, this is Theorem 4.1.
Proof. Start with the function R/I R/J R/(I + J) given by (x mod I, y mod J)
xy mod I +J. This is well-dened and bilinear, so from the universal mapping property of
the tensor product we get a linear map f : R/I
R
R/J R/(I +J) making the diagram
R/I
R
R/J
f

R/I R/J

n
n
n
n
n
n
n
n
n
n
n
n
(x mod I,y mod J)xy mod I+J

P
P
P
P
P
P
P
P
P
P
P
P
R/(I +J)
commute, so f(x mod I y mod J) = xy mod I + J. To write down the inverse map, let
R R/I
R
R/J by r r(11). This is linear, and when r I the value is r1 = 01 = 0.
Similarly, when r J the value is 0. Therefore I + J is in the kernel, so we get a linear
map g : R/(I +J) R/I
R
R/J by g(r mod I +J) = r(1 1) = r 1 = 1 r.
TENSOR PRODUCTS 13
To check f and g are inverses, a computation in one direction shows
f(g(r mod I +J)) = f(r 1) = r mod I +J.
To show g(f(t)) = t for all t R/I
R
R/J, we show all tensors are scalar multiples of 11.
Any elementary tensor has the form x y = x1 y1 = xy(1 1), which is a multiple of
11, so sums of elementary tensors are multiples of 11 and thus all tensors are multiples
of 1 1. We have
g(f(r(1 1)) = rg(1 mod I +J) = r(1 1) = r(1 1).

Remark 4.4. For two ideals I and J, we know a few operations that produce new ideals:
I +J, I J, and IJ. The intersection I J is the kernel of the linear map R R/I R/J
where r (r, r). Theorem 4.3 tells us I +J is the kernel of the linear map R R/I
R
R/J
where r r(1 1).
Theorem 4.5. For an ideal I in R and R-module M, there is a unique R-module isomor-
phism
(R/I)
R
M

= M/IM
where r m rm. In particular, taking I = (0), R
R
M

= M by r m rm, so
R
R
R

= R as R-modules by r r

rr

.
Proof. We start with the bilinear map (R/I) M M/IM given by (r, m) rm. From
the universal mapping property of the tensor product, we get a linear map f : (R/I)
R
M
M/IM where f(r m) = rm.
(R/I)
R
M
f

(R/I) M

n
n
n
n
n
n
n
n
n
n
n
n
(r,m)rm

P
P
P
P
P
P
P
P
P
P
P
P
M/IM
To create an inverse map, start with the function M (R/I)
R
M given by m 1 m.
This is linear in m (check!) and kills IM (generators for IM are products rm for r I
and m M, and 1 rm = r m = 0 m = 0), so it induces a linear map g : M/IM
(R/I)
R
M given by g(m) = 1 m.
To check f(g(m)) = m and g(f(t)) = t for all m M/IM and t (R/I)
R
M, we do
the rst one by a direct computation:
f(g(m)) = f(1 m) = 1 m = m.
To show g(f(t)) = t for all t M
R
N, we show all tensors in R/I
R
M are elementary.
An elementary tensor looks like r m = 1 rm, and a sum of tensors 1 m
i
is 1

i
m
i
.
Thus all tensors look like 1 m. We have g(f(1 m)) = g(m) = 1 m.
Example 4.6. For any abelian group A, (Z/nZ)
Z
A

= A/nA as abelian groups by
ma ma.
14 KEITH CONRAD
Remark 4.7. Saying R
R
M

= M by r m rm is another way of saying the R-
bilinear maps B out of RM can be identied with the linear maps out of M, and thats
clear because B(r, m) = B(1, rm) when B is bilinear, and B(1, ) is linear in the second
component.
The next theorem justies the discussion in the introduction about bases for tensor
products of free modules.
Theorem 4.8. If F and F

are free R-modules then F

R
F

is a free R-module. If e
i

iI
and e

jJ
are bases of F and F

then e
i
e

(i,j)IJ
is a basis of F
R
F

.
Proof. The result is clear if F or F

is 0, so let them both be nonzero free modules (hence

R = 0 and F and F

have bases). By Lemma 3.3, e

i
e

j
spans F
R
F

as an R-module.
To show this spanning set is linearly independent, suppose

i,j
c
ij
e
i
e

j
= 0, where all
but nitely many c
ij
are 0. We want to show every c
ij
is 0. Pick two basis vectors e
i
0
and
e

j
0
in F and F

. To create a coordinate function for e

i
0
e

j
0
in F
R
F

, dene the
function F F

R by (v, w) a
i
0
b
j
0
, where v =

i
a
i
e
i
and w =

j
b
j
e

j
. This function
is bilinear, so by the universal mapping property of tensor products there is a linear map
f
0
: F
R
F

R such that f
0
(v w) = a
i
0
b
j
0
on all elementary tensors v w.
F
R
F

f
0

F F

r
r
r
r
r
r
r
r
r
r
(v,w)a
i
0
b
j
0

M
M
M
M
M
M
M
M
M
M
M
R
In particular, f
0
(e
i
0
e

j
0
) = 1 and f
0
(e
i
e

j
) = 0 for (i, j) = (i
0
, j
0
). Applying f
0
to the
equation

i,j
c
ij
e
i
e

j
= 0 in F
R
F

tells us c
i
0
j
0
= 0 in R. Since i
0
and j
0
are arbitrary,
all the coecients are 0.
Theorem 4.8 has a concrete meaning in terms of bilinear maps out of F F

. It says
that any bilinear map out of F F

is determined by its values on the pairs (e

i
, e

j
) and
that any assignment of values to these pairs extends in a unique way to a bilinear map.
(The uniqueness of the extension is connected to the linear independence of the elementary
tensors e
i
e

j
.) This is the bilinear analogue of the existence and uniqueness of a linear
extension of a function from a basis of a free module to the whole module.
Example 4.9. Let R = 0 and F = Re
1
Re
2
be free of rank 2. Then F
R
F is free of
rank 4 with basis
e
1
e
1
, e
1
e
2
, e
2
e
1
, e
2
e
2
.
(The tensors e
1
e
2
and e
2
e
1
are not equal, since theyre dierent terms in a basis,
but theres also an explanation of this using bilinear maps: if e
1
e
2
= e
2
e
1
then all
bilinear maps out of F F takes the same value at (e
1
, e
2
) and (e
2
, e
1
), and that is not
true. Consider the bilinear map F F R given by (ae
1
+ be
2
, ce
1
+ de
2
) ad. So
e
1
e
2
= e
2
e
1
.)
The sum e
1
e
2
+ e
2
e
1
in F
R
F is an example of a tensor that is provably not an
elementary tensor. Any elementary tensor in F
R
F has the form
(4.1) (ae
1
+be
2
) (ce
1
+de
2
) = ace
1
e
1
+ade
1
e
2
+bce
2
e
1
+bde
2
e
2
.
TENSOR PRODUCTS 15
If this equals e
1
e
2
+e
2
e
1
then
ac = 0, ad = 1, bc = 1, bd = 0.
Since ad = 1 and bc = 1, a and c are invertible, but that contradicts ac = 0. So e
1
e
2
+e
2
e
1
is not an elementary tensor.
4
Example 4.10. As an R-module, R[X]
R
R[Y ] is free with basis X
i
Y
j

i,j0
, so this
tensor product is isomorphic to R[X, Y ] as R-modules by

c
ij
(X
i
Y
j
)

c
ij
X
i
Y
j
.
More generally, R[X
1
, . . . , X
k
]

= R[X]
k
as R-modules with X
i
corresponding to the tensor
1 X 1 where X is in the ith position. The dierence between ordinary products
and tensor products is like the dierence between multiplying polynomials as f(T)g(T) and
as f(X)g(Y ).
Example 4.11. We return to Example 2.1. For in M

and in N

, B
,
(m, n) =
(m)(n) is a bilinear map MN R. Set B: M

Bil
R
(M, N; R) by B(, ) =
B
,
. This is bilinear, so we get a linear map L: M

R
N

Bil
R
(M, N; R) where
L( ) = B
,
.
When M and N are nite free R-modules, lets show L is an isomorphism. We may sup-
pose M and N are nonzero, with respective bases e
1
, . . . , e
k
and f
1
, . . . , f

. By Theorem
4.8, M

R
N

is free of rank k with basis e

i
f

j
. By Example 2.1, Bil
R
(M, N; R) is
free of rank k with basis B
e

i
,f

j
. Since B
e

i
,f

j
= L(e

i
f

j
), L sends a basis to a basis,
so its an isomorphism when M and N are nite free.
It is not true that L is an isomorphism for all M and N. When p is prime, R = Z/p
2
Z,
and M = Z/pZ, show as an exercise that L: M

R
M

Bil
R
(M, M; R) is identically 0
but M

R
M

and Bil
R
(M, M; R) both have size p.
Theorem 4.12. Let F be a free R-module with basis e
i

iI
. For any k 1, the kth tensor
power F
k
is free with basis e
i
1
e
i
k

(i
1
,...,i
k
)I
k.
Proof. This is similar to the proof of Theorem 4.8.
Theorem 4.13. If M is an R-module and F is a free R-module with basis e
i

iI
, then
every element of M
R
F has a unique representation in the form

iI
m
i
e
i
, where all
but nitely many m
i
equal 0.
Proof. Using M as a spanning set of M and e
i

iI
as a spanning set for F as R-modules,
by Lemma 3.3 every element of M
R
F is a linear combination of elementary tensors
m
i
e
i
, where m
i
M. Since r(m
i
e
i
) = (rm
i
) e
i
, we can write every tensor in M
R
F
as a sum of elementary tensors of the form m
i
e
i
. So we have a surjective linear map
f :

iI
M M
R
F given by f((m
i
)
iI
) =

iI
m
i
e
i
. (All but nitely many m
i
are
0, so the sum makes sense.)
To create an inverse to f, consider the function MF

iI
M where (m,

i
r
i
e
i
)
(r
i
m)
iI
. This function is bilinear (check!), so there is a linear map g : M
R
F

iI
M
where g(m

i
r
i
e
i
) = (r
i
m)
iI
.
To check f(g(t)) = t for all t in M
R
F, we cant expect that all tensors in M
R
F are
elementary (an idea used in the proofs of Theorems 4.3 and 4.5), but we only need to check
f(g(t)) = t when t is an elementary tensor since f and g are additive and the elementary
4
From (4.1), a necessary condition for
P
2
i,j=1
cijei ej to be elementary is that c11c22 = c12c21. When
R = K is a eld this condition is also sucient, so the elementary tensors in K
2
K K
2
are characterized
among all tensors by a polynomial equation of degree 2. For a generalization, see [2].
16 KEITH CONRAD
tensors additively span M
R
F. (We will use this kind of argument a lot to reduce the
proof of an identity involving functions of all tensors to the case of elementary tensors even
though most tensors are not themselves elementary. The point is all tensors are sums of
elementary tensors and the formula we want to prove involves additive functions.) Any
elementary tensor looks like m

i
r
i
e
i
, and
f

iI
r
i
e
i

= f((r
i
m)
iI
)
=

iI
r
i
me
i
=

iI
mr
i
e
i
= m

iI
r
i
e
i
.
These sums have nitely many terms (r
i
= 0 for all but nitely many i), from the denition
of direct sums. Thus f(g(t)) = t for all t M
R
F.
For the composition in the other order,
g(f((m
i
)
iI
)) = g

iI
m
i
e
i

iI
g(m
i
e
i
) =

iI
(. . . , 0, m
i
, 0, . . . ) = (m
i
)
iI
.
Now that we know M
R
F

=

iI
M, with

iI
m
i
e
i
corresponding to (m
i
)
iI
, the
uniqueness of coordinates in the direct sum implies the sum representation

iI
m
i
e
i
is
unique.
Example 4.14. For any ring S R, elements of S
R
R[X] have unique expressions of the
form

n0
s
n
X
n
, so S
R
R[X]

= S[X] as R-modules by

n0
s
n
X
n

n0
s
n
X
n
.
Remark 4.15. Although you can check an additive identity f(g(t)) = t by only checking
it on elementary tensors, it would be really dumb to think you have proved injectivity
of some linear map f : M
R
N P by only looking at elementary tensors.
5
That is, if
f(mn) = 0 mn = 0, there is no reason to believe f(t) = 0 t = 0 for all t M
R
N,
since injectivity of a linear map is not an additive identity.
6
This is the number one reason
that proving that a linear map out of a tensor product is injective can require technique. As
a special case, if you want to prove a linear map out of a tensor product is an isomorphism,
it might be easier to construct an inverse map and check the composite in both orders is
the identity than to show the original map is injective and surjective.
Theorem 4.16. If M is nonzero and nitely generated then M
k
= 0 for all k.
Proof. Write M = Rx
1
+ + Rx
d
, where d 1 is minimal. Set N = Rx
1
+ + Rx
d1
(N = 0 if d = 1), so M = N +Rx
d
and x
d
N. Set I = r R : rx
d
N, so I is an ideal
in R and 1 I, so I is a proper ideal. From the denition of I, the function M
k
R/I
given by
(n
1
+r
1
x
d
, . . . , n
k
+r
k
x
d
) r
1
r
d
mod I
5
Unless every tensor in M R N is elementary, which is usually not the case.
6
Heres an example. Let C
R
C C be the R-linear map with the eect z w zw on elementary
tensors. If z w 0 then z = 0 or w = 0, so z w = 0, but the map is not injective: 1 i i 1 0 but
1 i i 1 = 0 since 1 i and i 1 belong to a basis of C
R
C by Theorem 4.8.
TENSOR PRODUCTS 17
is well-dened. Its multilinear (check!), so there is a linear map M
k
R/I such that
x
d
x
d
. .. .
k terms
1. That shows M
k
= 0.
Tensor powers of non-nitely generated modules could vanish: Q/Z
Z
Q/Z = 0 (Ex-
ample 3.6).
The rest of this section is concerned with properties of tensor products over domains.
Theorem 4.17. Let R be a domain with fraction eld K. There is a unique R-module
isomorphism K
R
K

= K where x y xy. In particular, Q
Z
Q

= Q.
Proof. Multiplication is a function m: K K K. It is R-bilinear, so the universal
mapping property of tensor products says there is an R-linear function f : K
R
K K
where f(x y) = xy on elementary tensors. That says the diagram
K
R
K
f

K K

r
r
r
r
r
r
r
r
r
r
m

M
M
M
M
M
M
M
M
M
M
M
K
commutes. Since f(x 1) = x, f is onto.
To show f is one-to-one, rst we show every tensor in K
R
K looks like 1 y. For an
elementary tensor x y, write x = a/b with a and b in R where b = 0. Then x y =
x by/b = bx y/b = a y/b = 1 ay/b = 1 xy. (We cant say x y = 1 xy from
bilinearity of , as is R-bilinear and x K.) Thus all elementary tensors have the form
1 y, so every tensor is 1 y for some y K since every tensor is a sum of elementary
tensors. Now we can show f has trivial kernel: if f(t) = 0 then, writing t = 1 y, we get
y = 0, so t = 1 0 = 0.
Remark 4.18. Multiplication K K K is also K-bilinear, not just R-bilinear, and
there is a K-vector space isomorphism K
K
K

= K (tensor product over K, not R).
For any R-module M, every tensor in K
R
M is elementary from manipulations with
tensors: any nite set of tensors of the form a
i
/b
i
m
i
can be provided with a common
denominator for the b
i
s, say a
i
/b
i
= a

i
/b, so

i
a
i
/b
i
m
i
=

i
1/b a

i
m
i
= 1/b
(

i
a

i
m
i
), which is an elementary tensor. What we cant easily do by manipulations with
tensors alone is determine when an elementary tensor x m is 0 in K
R
M.
Theorem 4.19. Let R be a domain with fraction eld K and M be any nonzero R-module
in K. There is a unique R-module isomorphism K
R
M

= K where x y xy. In
particular, K
R
I

= K for every nonzero ideal I in R.
Proof. The proof is largely like the previous one, but it needs a few extra steps because the
second module is an R-module in K rather than K itself.
7
Multiplication gives a function
K M K which is R-bilinear, so we get an R-linear map f : K
R
M K where
f(x y) = xy. To show f is onto, we cant look at f(x 1) since 1 is usually not in M.
Instead we can argue with a choice of nonzero m M: for any x K, f(x/mm) = x.
7
Theorem 4.17 is just a special case of Theorem 4.19, but we worked it out separately since some of the
technicalities are simpler.
18 KEITH CONRAD
To show f is injective, suppose f(t) = 0. All tensors in K
R
M are elementary, so we
can write t = x y. Then xy = 0 in K, so x = 0 or y = 0, so t = x y = 0.
Theorem 4.20. Let R be a domain and F and F

be free R-modules. If x and x

are
nonzero in F and F

, then x x

= 0 in F
R
F

.
Proof. Write x =

i
a
i
e
i
and x

=

j
a

j
e

j
using bases for F and F

. Then x x

i,j
a
i
a

j
e
i
e

j
in F
R
F

. Since x and x

are nonzero, they each have a nonzero coecient,

say a
i
and a

j
. Then a
i
a

j
= 0 since R is a domain, so xx

has a nonzero coordinate in the

basis e
i
e

j
of F
R
F

(Theorem 4.8). Thus x x

= 0.
Theorem 4.21. Let R be a domain with fraction eld K.
(1) For any R-module M, K
R
M

= K
R
(M/M
tor
) as R-modules, where M
tor
is the
torsion submodule of M.
(2) If M is a torsion R-module then K
R
M = 0 and if M is not a torsion module
then K
R
M = 0.
(3) If N M is a submodule such that M/N is a torsion module then K
R
N

= K
R
M
as R-modules by x n x n.
Proof. (1) The map K M K
R
(M/M
tor
) given by (x, m) x m is R-bilinear, so
there is a linear map f : K
R
M K
R
(M/M
tor
) where f(x m) = x m.
To go the other way, the canonical bilinear map KM

K
R
M vanishes at (x, m)
if m M
tor
: when rm = 0 for r = 0, x m = r(x/r) m = x/r rm = x/r 0 = 0
(Theorem 3.5). Therefore we get an induced bilinear map K(M/M
tor
) K
R
M given
by (x, m) xm. (The point is that an elementary tensor xm in K
R
M only depends
on m through its coset mod M
tor
.) The universal mapping property of the tensor product
now gives us a linear map g : K
R
(M/M
tor
) K
R
M where g(x m) = x m.
The composites g f and f g are both linear and x elementary tensors, so they x all
tensors and thus f and g are inverse isomorphisms.
(2) It is immediate from (1) that K
R
M = 0 if M is a torsion module, since K
R
M

=
K
R
(M/M
tor
) = K
R
0 = 0. We could also prove this in a direct way, by showing all
elementary tensors in K
R
M are 0: for x K and m M, let rm = 0 with r = 0, so
x m = r(x/r) m = x/r rm = x/r 0 = 0.
To show K
R
M = 0 when M is not a torsion module, from the isomorphism K
R
M

=
K
R
(M/M
tor
), we may replace M with M/M
tor
and are reduced to the case when M is
torsion-free. For torsion-free M we will create a nonzero R-module and a bilinear map onto
it from K M. This will require a fair bit of work (as it usually does to prove a tensor
product doesnt vanish).
We want to consider formal products xm with x K and m M. To make this precise,
we will use equivalence classes of ordered pairs in the same way that a domain is enlarged
to its fraction eld. On the product set K M, dene an equivalence relation by
(a/b, m) (c/d, n) adm = bcn in M.
Here a, b, c, and d are in R and b and d are not 0. The proof that this relation is well-dened
(independent of the choice of numerators and denominators) and transitive requires M be
torsion-free (check!). As an example, (0, m) (0, 0) for all m M.
TENSOR PRODUCTS 19
Dene KM = (K M)/ and write the equivalence class of (x, m) as x m. Give KM
the addition and K-scaling formulas
a
b
m+
c
d
n =
1
bd
(adm+bcn), x(y m) = (xy) m.
It is left to the reader to check these operations on KM are well-dened and make KM into
a K-vector space (so in particular an R-module). The zero element of KM is 0 0 = 0 m.
The function M KM given by m 1 m is injective, since if 1 m = 1 m

then
(1, m) (1, m

), so m = m

in M. So KM = 0 since M = 0.
The function K M KM given by (x, m) x m is R-bilinear and onto, so there is
a linear map K
R
M
f
KM such that f(x m) = x m, which is onto. Since KM = 0
we have K
R
M = 0, and in fact K
R
M

= KM by f (exercise).
(3) Since N M, there is an obvious bilinear map K N K
R
M, namely (x, n)
xn. So we get automatically a linear map f : K
R
N K
R
M where f(xn) = xn.
(This is not the identity function: on the left x n is in K
R
N and on the right x n is
in K
R
M.) To get an inverse map to f, let KM K
R
N by (x, m) (1/r)x rm,
where r R 0 is chosen so rm N. There is such r since M/N is a torsion module.
(We cant simplify (1/r)x rm by moving r through since rm is in N but usually m is
not.) To check this function is well-dened, if also r

m N with r

R 0, then
1
r

x r

m =
r
rr

x r

m =
1
rr

x rr

m =
1
rr

x r

(rm) =
r

rr

x rm =
1
r
x rm.
So our function KM K
R
N is well-dened, and the reader can check it is bilinear. It
leads to a linear map g : K
R
M K
R
N where g(xm) = (1/r)xrm when rm N,
r = 0. As an exercise, check f(g(x m)) = x m and g(f(x n)) = x n, so f g and
g f are both the identity by additivity.
Corollary 4.22. Let R be a domain with fraction eld K. In K
R
M, x m = 0 if and
only if x = 0 or m M
tor
. In particular, M
tor
= ker(M K
R
M) where m 1 m.
Proof. If x = 0 then xm = 0m = 0. If m M
tor
, with rm = 0 for some nonzero r R,
then x m = (x/r)r m = (x/r) rm = (x/r) 0 = 0.
Conversely, suppose x m = 0. We want to show x = 0 or m M
tor
. Write x = a/b,
so (1/b) am = 0. If a = 0 then x = 0, so we suppose a = 0 and will show m M
tor
.
Multiply by b to get 1 am = 0. From the isomorphism K
R
M

= K
R
(M/M
tor
),
1 am = 0 in K
R
(M/M
tor
). Since M/M
tor
is torsion-free, applying the R-linear map
K
R
(M/M
tor
) K(M/M
tor
) from the proof of Theorem 4.21 tells us that 1 am = 0 in
K(M/M
tor
). The function m 1 m from M/M
tor
to K(M/M
tor
) is injective, so am = 0,
so am M
tor
. Therefore there is nonzero r R such that 0 = r(am) = (ra)m. Since
ra = 0, m M
tor
.
Example 4.23. The tensor product Q
Z
A is 0 when A is a torsion abelian group, so we
recover Example 3.6.
5. General Properties of Tensor Products
There are canonical isomorphisms M N

= N M and (M N) P

= M (N P).
We want to show similar isomorphisms for tensor products: M
R
N

= N
R
M and
(M
R
N)
R
P

= M
R
(N
R
P). Furthermore, there is a distributive property over
direct sums: M
R
(N P)

= (M
R
N) (M
R
P). How these modules are isomorphic
is much more important than just that they are isomorphic.
20 KEITH CONRAD
Theorem 5.1. There is a unique R-module isomorphism M
R
N

= N
R
M where
mn n m.
Proof. We want to create a linear map M
R
N N
R
M sending mn to n m. To
do this, we back up and start o with a map out of M N to the desired target module
N
R
M. Dene M N N
R
M by (m, n) n m. This is a bilinear map since
n m is bilinear in m and n. Therefore by the universal mapping property of the tensor
product, there is a unique linear map f : M
R
N N
R
M such that f(mn) = n m
on elementary tensors.
M
R
N
f

M N

q
q
q
q
q
q
q
q
q
q
(m,n)nm

M
M
M
M
M
M
M
M
M
M
N
R
M
Running through the above arguments with the roles of M and N interchanged, there
is a unique linear map g : N
R
M M
R
N where g(n m) = m n on elementary
tensors. We will show f and g are inverses of each other.
To show f(g(t)) = t for all t N
R
M, it suces to check this when t is an elementary
tensor, since both sides are R-linear (or even just additive) in t and N
R
M is spanned
by its elementary tensors: f(g(n m)) = f(mn) = n m. Therefore f(g(t)) = t for all
t N
R
M. The proof that g(f(t)) = t for all t M
R
N is similar. We have shown f
and g are inverses of each other, so f is an R-module isomorphism.
Theorem 5.2. There is a unique R-module isomorphism (M
R
N)
R
P

= M
R
(N
R
P)
where (mn) p m(n p).
Proof. By Lemma 3.3, (M
R
N)
R
P is linearly spanned by all (m n) p and M
R
(N
R
P) is linearly spanned by all m (n p). Therefore linear maps out of these two
modules are determined by their values on these
8
elementary tensors. So there is at most
one linear map (M
R
N)
R
P M
R
(N
R
P) with the eect (mn)p m(np),
and likewise in the other direction.
To create such a linear map (M
R
N)
R
P M
R
(N
R
P), consider the function
M N P M
R
(N
R
P) given by (m, n, p) m (n p). Since m (n p) is
trilinear in m, n, and p, for each p we get a linear map f
p
: M
R
N M
R
(N
R
P)
where f
p
(mn) = m(n p) on elementary tensors in M
R
N.
Now we consider the function (M
R
N) P M
R
(N
R
P) given by (t, p) f
p
(t).
This is bilinear! First, it is linear in t with p xed, since each f
p
is a linear function. Next
we show it is linear in p with t xed:
f
p+p
(t) = f
p
(t) +f
p
(t) and f
rp
(t) = rf
p
(t)
8
A general elementary tensor in (M R N) R P is not (m n) p, but t p where t M R N and
t might not be elementary itself. Similarly, elementary tensors in M R (N R P) are more general than
m(n p).
TENSOR PRODUCTS 21
for any p, p

, and r. Both sides of these identities are additive in t, so to check them it

suces to check the case when t = mn:
f
p+p
(mn) = (mn) (p +p

)
= (mn) p + (mn) p

= f
p
(mn) +f
p
(mn)
= (f
p
+f
p
)(mn).
That f
rp
(m n) = rf
p
(m n) is left to the reader. Since f
p
(t) is bilinear in p and t,
the universal mapping property of the tensor product tells us there is a unique linear map
f : (M
R
N)
R
P M
R
(N
R
P) such that f(t p) = f
p
(t). Then f((mn) p) =
f
p
(m n) = m (n p), so we have found a linear map with the desired values on the
tensors (mn) p.
Similarly, there is a linear map g : M
R
(N
R
P) (M
R
N)
R
P where g(m(np)) =
(mn)p. Easily f(g(m(np))) = m(np) and g(f((mn)p)) = (mn)p. Since
these particular tensors linearly span the two modules, these identities extend by linearity
(f and g are linear) to show f and g are inverse functions.
Theorem 5.3. There is a unique R-module isomorphism
M
R
(N P)

= (M
R
N) (M
R
P)
where m(n, p) (mn, mp).
Proof. Instead of directly writing down an isomorphism, we will put to work the essential
uniqueness of solutions to a universal mapping problem by showing (M
R
N) (M
R
P)
has the universal mapping property of the tensor product M
R
(N P). Therefore by
abstract nonsense these modules must be isomorphic. That there is an isomorphism whose
eect on elementary tensors in M
R
(NP) is as indicated in the statement of the theorem
will fall out of our work.
For (M
R
N) (M
R
P) to be a tensor product of M and N P, it needs a bilinear
map to it from M(NP). Let b: M(NP) (M
R
N)(M
R
P) by b(m, (n, p)) =
(mn, mp). This function is bilinear. We verify one part:
b(m, (n, p) + (n

, p

)) = b(m, (n +n

, p +p

))
= (m(n +n

), m(p +p

))
= (mn +mn

, mp +mp

)
= (mn, mp) + (mn

, mp

)
= b(m, (n, p)) +b(m, (n

, p

)).
To show (M
R
N) (M
R
P) equipped with the map b has the universal mapping
property of M
R
(N P) and its canonical bilinear map , let B: M (N P) Q be
any bilinear map. We seek an R-linear map L making the diagram
(5.1) (M
R
N) (M
R
P)
L

M (N P)
b

j
j
j
j
j
j
j
j
j
j
j
j
j
j
j
B

T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
Q
22 KEITH CONRAD
commute. Being linear, L would be determined by its values on the direct summands, and
these values would be determined by its values on all pairs (m n, 0) and (0, m p) by
additivity. These values are forced by commutativity of (5.1) to be
L(mn, 0) = L(b(m,(n, 0))) = B(m,(n, 0)) and L(0, mp) = L(b(m,(0, p))) = B(m,(0, p)).
To construct L, we are inspired by these formulas to contemplate the maps MN Q
and M P Q given by (m, n) B(m, (n, 0)) and (m, p) B(m, (0, p)). Both are
bilinear, so there are R-linear maps M
R
N
L
1
Q and M
R
P
L
2
Q where
L
1
(mn) = B(m, (n, 0)) and L
2
(mp) = B(m, (0, p)).
Dene L on (M
R
N) (M
R
P) by L(t
1
, t
2
) = L
1
(t
1
) +L
2
(t
2
). (Notice we are dening
L not just on ordered pairs of elementary tensors, but on all pairs of tensors. We need L
1
and L
2
to be dened on the whole tensor product modules M
R
N and M
R
P.) The
map L is linear since L
1
and L
2
are linear, and (5.1) commutes:
L(b(m, (n, p))) = L(b(m, (n, 0) + (0, p)))
= L(b(m, (n, 0)) +b(m, (0, p)))
= L((mn, 0) + (0, mp)) by the denition of b
= L(mn, mp)
= L
1
(mn) +L
2
(mp) by the denition of L
= B(m, (n, 0)) +B(m, (0, p))
= B(m, (n, 0) + (0, p))
= B(m, (n, p)).
Now that weve shown (M
R
N) (M
R
P) and the bilinear map b have the universal
mapping property of M
R
(N P) and the canonical bilinear map , there is a unique
linear map f making the diagram
(M
R
N) (M
R
P)
f

M (N P)
b

j
j
j
j
j
j
j
j
j
j
j
j
j
j
j

T
T
T
T
T
T
T
T
T
T
T
T
T
T
T
M
R
(N P)
commute, and f is an isomorphism of R-modules because it transforms one solution of a
universal mapping problem into another. Taking (m, (n, p)) around the diagram both ways,
f(b(m, (n, p))) = f(mn, mp) = m(n, p).
Therefore the inverse of f is an isomorphism M
R
(N P) (M
R
N) (M
R
P) with
the eect m(n, p) (mn, mp).
Theorem 5.4. There is a unique R-module isomorphism
M
R

iI
N
i

=

iI
(M
R
N
i
)
where m(n
i
)
iI
(mn
i
)
iI
.
TENSOR PRODUCTS 23
Proof. We extrapolate from the case #I = 2 in Theorem 5.3. The map b: M(

iI
N
i
)

iI
(M
R
N
i
) by b((m, (n
i
)
iI
)) = (mn
i
)
iI
is bilinear. We will show

iI
(M
R
N
i
)
and b have the universal mapping property of M
R

iI
N
i
and .
Let B: M (

iI
N
i
) Q be bilinear. For each i I the function M N
i
Q where
(m, n
i
) B(m, (. . . , 0, n
i
, 0, . . . )) is bilinear, so there is a linear map L
i
: M
R
N
i
Q
where L
i
(mn
i
) = B(m, (. . . , 0, n
i
, 0, . . . )). Dene L:

iI
(M
R
N
i
) Q by L((t
i
)
iI
) =

iI
L
i
(t
i
). All but nitely many t
i
equal 0, so the sum here makes sense, and L is linear.
It is left to the reader to check the diagram

iI
(M
R
N
i
)
L

iI
N
i
b

l
l
l
l
l
l
l
l
l
l
l
l
l
B

R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
Q
commutes. Any linear map L making this diagram commute has its value on (. . . , 0, m
n
i
, 0, . . . ) = b(m, (. . . , 0, n
i
, 0, . . . )) determined by B, so it is unique. Thus

iI
(M
R
N
i
)
and the bilinear map b to it have the universal mapping property of M
R

iI
N
i
and the
canonical map , so there is an R-module isomorphism f making the diagram

iI
(M
R
N
i
)
f

iI
N
i
b

l
l
l
l
l
l
l
l
l
l
l
l
l

R
R
R
R
R
R
R
R
R
R
R
R
R
M
R

iI
N
i
commute. Sending (m, (n
i
)
iI
) around the diagram both ways shows f((mn
i
)
iI
) = m
(n
i
)
iI
, so the inverse of f is an isomorphism with the eect m(n
i
)
iI
(mn
i
)
iI
.
Remark 5.5. The analogue of Theorem 5.4 for direct products is false. While there is a
natural R-linear map
(5.2) M
R

iI
N
i

iI
(M
R
N
i
),
namely m (n
i
)
iI
(m n
i
)
iI
on elementary tensors, it is not an isomorphism in
general. Taking R = Z, M = Q, and N
i
= Z/p
i
Z (i 1), the right side of (5.2) is 0 since
Q
Z
(Z/p
i
Z) = 0 for all i 1 (Example 3.6). The left side of (5.2) is Q
Z

i1
Z/p
i
Z,
which is not 0 by Theorem 4.21 since

i1
Z/p
i
Z is not a torsion abelian group.
In our proof of associativity of the tensor product, we started with a function on a direct
product MNP and collapsed this domain to an iterated tensor product (M
R
N)
R
P
using bilinearity twice. It is useful to record a rather general result in that direction, as a
technical lemma for future convenience.
Theorem 5.6. Let M
1
, . . . , M
k
, N be R-modules, with k > 2, and suppose
M
1
M
k2
M
k1
M
k

N
24 KEITH CONRAD
is a function which is bilinear in M
k1
and M
k
when other coordinates are xed. There is
a unique function
M
1
M
k2
(M
k1

R
M
k
)

N
that is linear in M
k1

R
M
k
when the other coordinates are xed and satises
(5.3) (m
1
, . . . , m
k2
, m
k1
m
k
) = (m
1
, . . . , m
k2
, m
k1
, m
k
).
If is multilinear in M
1
, . . . , M
k
, then is multilinear in M
1
, . . . , M
k2
, M
k1

R
M
k
.
Proof. Assuming a function exists satisfying (5.3) and is linear in the last coordinate,
its value everywhere is determined by additivity in the last coordinate: write any tensor
t M
k1

R
M
k
in the form t =

p
i=1
x
i
y
i
, and then
(m
1
, . . . , m
k2
, t) =

m
1
, . . . , m
k2
,
p

i=1
x
i
y
i

=
p

i=1
(m
1
, . . . , m
k2
, x
i
y
i
)
=
p

i=1
(m
1
, . . . , m
k2
, x
i
, y
i
).
It remains to show exists with the desired properties.
Fix m
i
M
i
for i = 1, . . . , k 2. Dene
m
1
,...,m
k2
: M
k1
M
k
N by

m
1
,...,m
k2
(x, y) = (m
1
, . . . , m
k2
, x, y).
By hypothesis
m
1
,...,m
k2
is bilinear in x and y, so from the universal mapping property of
the tensor product there is a linear map
m
1
,...,m
k2
: M
k1

R
M
k
N such that

m
1
,...,m
k2
(x y) =
m
1
,...,m
k2
(x, y) = (m
1
, . . . , m
k2
, x, y).
Dene : M
1
M
k2
(M
k1

R
M
k
) N by
(m
1
, . . . , m
k2
, t) =
m
1
,...,m
k2
(t).
Since
m
1
,...,m
k2
is a linear function on M
k1

R
M
k
, (m
1
, . . . , m
k2
, t) is linear in t when
m
1
, . . . , m
k2
are xed.
If is multilinear in M
1
, . . . , M
k
we want to show is multilinear in M
1
, . . . , M
k2
,
M
k1

R
M
k
. We already know is linear in M
k1

R
M
k
when the other coordinates are
xed. To show is linear in each of the other coordinates (xing the rest), we carry out
the computation for M
1
(the argument is similar for other M
i
s): is
(x +x

, m
2
, . . . , m
k2
, t)
?
= (x, m
2
, . . . , m
k2
, t) + (x

, m
2
, . . . , m
k2
, t)
(rx, m
2
, . . . , m
k2
, t)
?
= r(x, m
2
, . . . , m
k2
, t)
when m
2
, . . . , m
k2
, t are xed in M
2
, . . . , M
k2
, M
k1

R
M
k
? In these two equations, both
sides are additive in t so it suces to verify these two equations when t is an elementary
tensor, say t = m
k1
m
k
. Then from (5.3), these two equations are true since were
assuming is linear in M
1
(xing the other coordinates).
TENSOR PRODUCTS 25
Theorem 5.6 is not specic to functions which are bilinear in the last two coordinates:
any two coordinates can be used when the function is bilinear in those two coordinates. For
instance, lets revisit the proof of associativity of the tensor product in Theorem 5.3. Dene
: M N P M
R
(N
R
P)
by (m, n, p) = m(n p). This function is trilinear, so Theorem 5.6 says we can replace
M N with its tensor product: there is a bilinear function
: (M
R
N) P M
R
(N
R
P)
such that (mn, p) = m(n p). Since is bilinear, there is a linear function
f : (M
R
N)
R
P M
R
(N
R
P)
such that f(tp) = (t, p), so f((mn)p) = (mn, p) = m(np). The construction
of the functions f
p
in the proof of Theorem 5.3 are now seen to be special cases of Theorem
5.6.
The remaining module properties we treat with the tensor product in this section in-
volve its interaction with the Hom-module construction, so in particular the dual module
construction (M

= Hom
R
(M, R)).
Theorem 5.7. There are R-module isomorphisms
Hom
R
(M, Hom
R
(N, P))

= Bil
R
(M, N; P)

= Hom
R
(M
R
N, P).
Proof. A member of Hom
R
(M, Hom
R
(N, P)) is a linear map f : M Hom
R
(N, P), and
from f we can create a bilinear map B
f
: MN P by B
f
(m, n) = f(m)(n). In the other
direction, from any bilinear map B: MN P we get a linear map f
B
: M Hom
R
(N, P)
by f
B
(m): n B(m, n). The correspondences f B
f
and B f
B
are obviously in-
verses, and the reader can check they are linear as well, so Hom
R
(M, Hom
R
(N, P))

=
Bil
R
(M, N; P). The R-module isomorphism between Bil
R
(M, N; P) and Hom
R
(M
R
N, P)
comes from the universal mapping property of the tensor product.
Heres a high-level way of interpreting the isomorphism between the two Hom-modules
in Theorem 5.7. Write F
N
(M) = Hom
R
(N, M) and G
N
(M) = M
R
N, so F
N
and G
N
turn R-modules into new R-modules. Theorem 5.7 says
Hom
R
(M, F
N
(P))

= Hom
R
(G
N
(M), P).
This is analogous to the relation between a matrix and its transpose inside the dot product:
v Aw = A

v w.
So F
N
and G
N
are transposes of each other. Actually, F
N
and G
N
are really called
adjoints of each other (F
N
is left adjoint to G
N
, and G
N
is right adjoint to F
N
) because
pairs of operators L and L

in linear algebra that satisfy the relation L(v) w = v L

(w)
are called adjoints and the relation between F
N
and G
N
is formally similar.
Corollary 5.8. For R-modules M and N, there are R-module isomorphisms
Hom
R
(M, N

= Hom
R
(N, M

= Bil
R
(M, N; R)

= (M
R
N)

.
Proof. Let P = R in Theorem 5.7: Hom
R
(M, N

)

= Bil
R
(M, N; R)

= (M
R
N)

. By
Theorem 5.1, M
R
N

= N
R
M, so (M
R
N)

= (N
R
M)

, and the latter dual module

is isomorphic to Hom
R
(N, M

) by Theorem 5.7 with the roles of M and N there reversed

and P = R. Thus we have obtained isomorphisms between the desired modules.
26 KEITH CONRAD
Explicitly, the isomorphism between Hom
R
(M, N

) and Hom
R
(N, M

) amounts to real-
izing linear maps in both Hom-modules are the same as bilinear maps B: MN R.
The construction of M
R
N is symmetric in M and N in the sense that M
R
N

=
N
R
M in a natural way, but Corollary 5.8 is not saying Hom
R
(M, N)

= Hom
R
(N, M)
since those are not the Hom-modules in the corollary. For instance, if R = M = Z and
N = Z/2Z then Hom
R
(M, N)

= Z/2Z and Hom

R
(N, M) = 0.
Theorem 5.9. For R-modules M and N, there is a unique linear map
M

R
N Hom
R
(M, N)
sending the elementary tensor n to [m (m)n]. This is an isomorphism if M and
N are nite free. In particular, if F is nite free then F

R
F

= End
R
(F) as R-modules.
Concretely, the last part is saying (R
n
)

R
R
n
= M
n
(R) as R-modules, in a natural way.
Proof. We need to make an element of M

and element of N act together as a linear map

M N. The function M

N M N given by (, n, m) (m)n is trilinear.

Here the parameter acts on m to give a scalar, which is then multiplied by n. By
Theorem 5.6, this trilinear map induces a bilinear map B: (M

R
N) M N where
B( n, m) = (m)n. For t M

R
N, B(t, ) is in Hom
R
(M, N), so we have a linear
map f : M

R
N Hom
R
(M, N) by f(t) = B(t, ). (Explicitly, the elementary tensor
n acts on M by the rule ( n)(m) = (m)n.)
Now let M and N be nite free. To show f is an isomorphism, we may suppose M and
N are nonzero. Pick bases e
i
of M and e

j
of N. Then f lets e

i
e

j
act on M by
sending e
k
to e

i
(e
k
)e

j
=
ik
e

j
. So f(e

i
e

j
) Hom
R
(M, N) sends e
i
to e

j
and sends every
other member of the basis of M to 0. Such linear maps are a basis of Hom
R
(M, N), so f is
onto.
To show f is one-to-one, suppose f(

i,j
c
ij
e

i
e

j
) = O in Hom
R
(M, N). Applying both
sides to any e
k
, we get

i,j
c
ij

ik
e

j
= 0, which says

j
c
kj
e

j
= 0, so c
kj
= 0 for all j and
all k. Thus every c
ij
is 0. This concludes the proof that f is an isomorphism.
Lets work out the inverse map explicitly. For L Hom
R
(M, N), write L(e
i
) =

j
a
ji
e

j
,
so L has matrix representation (a
ji
). (The matrix indices here look reversed from usual
practice because we use i as the index for basis vectors in M and j as the index for basis
vectors in N; review how linear maps become matrices when bases are chosen. If we had
indexed bases of M and N with i and j in each others places, then L(e
j
) =

a
ij
e

i
.) From
the isomorphism f, let L correspond to

i,j
c
ij
e

i
e

j
in M

R
N, with the coecients
c
ij
to be determined. Then
L(e
k
) =

i,j
c
ij
(e

i
e

j
)(e
k
) =

j
c
kj
e

j
,
so a
jk
= c
kj
. Therefore c
ij
= a
ji
, so L Hom
R
(M, N) with matrix representation (a
ji
)
corresponds to

i,j
a
ji
e

i
e

j
. Thats nice. It just says e

i
e
j
corresponds to the matrix
unit E
ji
.
9

Example 5.10. For nite-dimensional vector spaces V and W over the eld K, V

K
W

=
Hom
K
(V, W) by ( w)(v) = (v)w. This is one of the most basic ways tensor products
occur in linear algebra. Understand this isomorphism! When V = W = K
2
, with standard
9
Notice the index switch: e

i
ej goes to Eji and not Eij.
TENSOR PRODUCTS 27
basis e
1
and e
2
, the matrix (
a b
c d
) M
2
(K) = Hom
K
(V, W) corresponds to the tensor
e

a
c

+e

b
d

since this tensor sends e

1
to e

1
(e
1
)

a
c

+e

2
(e
1
)

b
d

a
c

, and similarly
this tensor sends e
2
to

b
d

.
Remark 5.11. The linear map M

R
N Hom
R
(M, N) in Theorem 5.9 is not an
isomorphism, or even injective or surjective, in general. For example, let p be prime,
R = Z/p
2
Z, and M = N = Z/pZ as an R-module. Check as an exercise that M

= M,
M
R
M

= M, Hom
R
(M, M)

= M, and the natural map M

R
M Hom
R
(M, M) is
identically 0 (it suces to show every elementary tensor in M

R
M acts on M as 0).
Remark 5.12. We could have bypassed the injectivity part of the proof, for the following
reason. When M and N are nite free, M

R
N and Hom
R
(M, N) are both nite free R-
modules of the same rank, so they are abstractly isomorphic. Therefore a linear surjection
between them has to be an isomorphism (consequence of Cayley-Hamilton theorem), so f
is an isomorphism.
When M and N are nite free R-modules, the isomorphisms in Corollary 5.8 and Theorem
5.9 lead to a description of M
R
N that makes no mention of universal mapping properties.
Identify M with M

by double duality, so Theorem 5.9 assumes the form

M
R
N

= Hom
R
(M

, N),
where m n goes over to the linear map (m)n. Since N

= N

by double duality,
Hom
R
(M

, N)

= Hom
R
(M

, (N

= Bil
R
(M

, N

; R) by Corollary 5.8. Therefore

(5.4) M
R
N

= Bil
R
(M

, N

; R),
where m n acts as the bilinear map M

R sending (, ) to (m)(n). The

denition of the tensor product of nite-dimensional vector spaces in [1, p. 65] and [4, p. 35]
is essentially (5.4).
10
It is a good exercise to check these interpretations of m n as a
member of Hom
R
(M

, N) and Bil
R
(M

, N

; R) are identied with each other by Corollary

5.8 and double duality.
But watch out! The formula (5.4) for the tensor product does not work for general
modules M and N (where double duality doesnt hold). While there is always a linear map
M
R
N Bil
R
(M

, N

; R) given on elementary tensors by mn [(, ) (m)(n)],

it is generally not an isomorphism.
Example 5.13. Let p be prime, R = Z/p
2
Z, and M = Z/pZ as an R-module. The natural
map M
R
M Bil
R
(M

, M

; R) is identically 0 while both sides have size p.

6. Base Extension
In algebra, there are many times a module over one ring is replaced by a related module
over another ring. For instance, in linear algebra it is useful to enlarge R
n
to C
n
, creating
in this way a complex vector space by letting the real coordinates be extended to complex
coordinates. In ring theory, irreducibility tests in Z[X] involve viewing a polynomial in
Z[X] as a polynomial in Q[X] or reducing the coecients mod p to view it in (Z/pZ)[X].
We will see that all these passages to modules with new coecients (R
n
C
n
, Z[X]
Q[X], Z[X] (Z/pZ)[X]) can be described in a uniform way using tensor products.
10
Using the last isomorphism in Corollary 5.8 and double duality, M R N

= BilR(M, N; R)

for nite
free M and N, where mn goes to the (evaluation) function B B(m, n). This is how tensor products of
nite-dimensional vector spaces are dened in [3, p. 40], namely V K W is the dual space to BilK(V, W; K).
28 KEITH CONRAD
Let f : R S be a homomorphism of commutative rings. We use f to consider any
S-module N as an R-module by rn := f(r)n. In particular, S itself is an R-module by
rs := f(r)s. Passing from N as an S-module to N as an R-module in this way is called
restriction of scalars.
Example 6.1. If R S, f can be the inclusion map (e.g., R C or Q C). This is
how a C-vector space is thought of as an R-vector space or Q-vector space.
Example 6.2. If S = R/I, f can be reduction modulo I (e.g., R = K[T] and S =
K[T]/(m(T))). This is how a K-vector space V on which a linear operator A acts is turned
into a K[T]-module by letting T act as A. First V becomes a module over the ring of
operators K[A] acting on V . Since K[A]

= K[T]/(m(T)), where m(T) is the minimal
polynomial of A, V is a K[T]/(m(T))-module. Then V becomes a K[T]-module by pulling
back the scalars from K[T]/(m(T)) to K[T] using reduction mod m(T).
Here is a simple illustration of restriction of scalars.
Theorem 6.3. Let N and N

be S-modules. Any S-linear map N N

is also an R-linear
map when we treat N and N

as R-modules.
Proof. Let : N N

be S-linear, so (sn) = s(n) for any s S and n N. For r R,

(rn) = (f(r)n) = f(r)(n) = r(n),
so is R-linear.
As a notational convention, since we will be going back and forth between R-modules
and S-modules a lot, we will write M (or M

, and so on) for R-modules and N (or N

, and
so on) for S-modules. Since N is also an R-module by restriction of scalars, we can form
the usual R-module M
R
N, where
r(mn) = (rm) n = mrn,
with the third expression really being mf(r)n since rn := f(r)n.
We want to reverse the process of restriction of scalars. For any R-module M we want
to create an S-module of products sm which matches the old meaning of rm if s = f(r).
This new module is called an extension of scalars or base extension. It will be the R-module
S
R
M equipped with a new structure of S-module.
Since S is a ring and not just an R-module, we can make S
R
M into an S-module by
(6.1) s

(s m) = s

s m.
But does this S-scaling extend unambiguously to all tensors?
Theorem 6.4. The additive group S
R
M has a unique S-module structure satisfying
(6.1), and this is compatible with the R-module structure in the sense that rt = f(r)t for all
r R and t S
R
M.
Proof. Suppose the additive group S
R
M has an S-module structure satisfying (6.1). We
will show the S-scaling on all tensors in S
R
M is determined by this. Any t S
R
M is
a nite sum of elementary tensors, say
t = s
1
m
1
+ +s
k
m
k
.
TENSOR PRODUCTS 29
For s S,
st = s(s
1
m
1
+ +s
k
m
k
)
= s(s
1
m
1
) + +s(s
k
m
k
)
= ss
1
m
1
+ +ss
k
m
k
by (6.1),
so st is determined, although this formula for it is not obviously well-dened. (Does a
dierent expression for t as a sum of elementary tensors change st?)
Now we show there really is an S-module structure on S
R
M satisfying (6.1). Describing
the S-scaling on S
R
M means creating a function S (S
R
M) S
R
M satisfying
the relevant scaling axioms:
(6.2) 1 t = t, s(t
1
+t
2
) = st
1
+st
2
, (s
1
+s
2
)t = s
1
t +s
2
t, s
1
(s
2
t) = (s
1
s
2
)t.
For each s

S we consider the function S M S

R
M given by (s, m) (s

s) m.
This is R-bilinear, so by the universal mapping property of tensor products there is an
R-linear map
s
: S
R
M S
R
M where
s
(s m) = (s

s) m on elementary tensors.
Dene a multiplication S (S
R
M) S
R
M by s

t :=
s
(t). This will be the scaling
of S on S
R
M. We check the conditions in (6.2):
(1) To show 1t = t means showing
1
(t) = t. On elementary tensors,
1
(s m) =
(1 s) m = s m, so
1
xes elementary tensors. Therefore
1
xes all tensors by
additivity.
(2) s(t
1
+t
2
) = st
1
+st
2
since
s
is additive.
(3) To show (s
1
+ s
2
)t = s
1
t + s
2
t means showing
s
1
+s
2
=
s
1
+
s
2
as functions on
S
R
M. Both sides are additive functions on S
R
M, so it suces to check they
agree on tensors s m, where both sides have common value (s
1
+s
2
)s m.
(4) To show s
1
(s
2
t) = (s
1
s
2
)t means
s
1

s
2
=
s
1
s
2
as functions on S
R
M. Both
sides are additive functions of t, so it suces to check they agree on tensors s m,
where both sides have common value (s
1
s
2
s) m.
Lets check the S-module structure on S
R
M is compatible with its original R-module
structure. For r R, if we treat r as f(r) S then scaling by f(r) on an elementary tensor
sm has the eect f(r)(sm) = f(r)sm. Since f(r)s is the denition of rs (this is how
we make S into an R-module), f(r)sm = rsm = r(sm). Thus f(r)(sm) = r(sm),
so f(r)t = rt for all t in S
R
M.
By exactly the same kind of argument, M
R
S has a unique S-module structure where
s

(ms) = ms

s. So whenever we meet M
R
S or S
R
M, we know they are S-modules
in a specic way. Moreover, these two S-modules are naturally isomorphic: by Theorem
5.1, there is an isomorphism : S
R
M M
R
S of R-modules where (s m) = ms.
To show is in fact an isomorphism of S-modules, all we need to do is check S-linearity
since is known to be additive and a bijection. To show (s

t) = s

(t) for all s

and t,
additivity of both sides in t means we may focus on the case t = s m:
(s

(s m)) = ((s

s) m) = ms

s = s

(ms) = s

(s m).
This idea of creating an S-module isomorphism by using a known R-module isomorphism
that is also S-linear will be used many more times, so watch for it.
Now we must be careful to refer to R-linear and S-linear maps, rather than just linear
maps, so it is clear what our scalar ring is each time.
30 KEITH CONRAD
Example 6.5. In Example 4.5 we saw (R/I)
R
M

= M/IM as R-modules by rm rm.
Since M/IM is naturally an R/I-module, and now we know (R/I)
R
M is an R/I-module,
the R-module isomorphism turns out to be an R/I-module isomorphism too since it is R/I-
linear (check!).
Example 6.6. In Theorem 4.21, we looked at K
R
M when R is a domain with fraction
eld K. It was treated there as an R-module, but now we see it is also a K-vector space
with the K-scaling rule x(y m) = xy m on elementary tensors.
Theorem 6.7. If F is a free R-module with basis e
i

iI
then S
R
F is a free S-module
with basis 1 e
i

iI
.
Proof. Since S is an R-module, we know from Theorem 4.13 that every element of S
R
F
has a unique representation in the form

iI
s
i
e
i
, where all but nitely many s
i
equal 0.
Since s
i
e
i
= s
i
(1e
i
) in the S-module structure on S
R
F, every element of S
R
F is a
unique S-linear combination

s
i
(1 e
i
), which says 1 e
i
is an S-basis of S
R
F.
Example 6.8. As an S-module, S
R
R
n
has S-basis 1e
1
, . . . , 1e
n
where e
1
, . . . , e
n

is the standard basis of R

n
, so S
n
= S
R
R
n
as S-modules by
(s
1
, . . . , s
n
)
n

i=1
s
i
(1 e
i
) =
n

i=1
s
i
e
i
because this map is S-linear (check!) and sends an S-basis to an S-basis. In particular,
S
R
R

= S as S-modules by s r sr.
For instance,
C
R
R
n

= C
n
, C
R
M
n
(R)

= M
n
(C).
These are isomorphisms as C-vector spaces. For any ideal I in R, (R/I)
R
R
n
= (R/I)
n
as R/I-modules.
Example 6.9. As an S-module, S
R
R[X] has S-basis 1 X
i

i0
, so S
R
R[X]

= S[X]
as S-modules
11
by

i0
s
i
X
i

i0
s
i
X
i
.
As particular examples, C
R
R[X]

= C[X] as C-vector spaces, Q

Z
Z[X]

= Q[X] as
Q-vector spaces and (Z/pZ)
Z
Z[X]

= (Z/pZ)[X] as Z/pZ-vector spaces.

Example 6.10. If we treat C
n
as a real vector space, then its base extension to C is the
complex vector space C
R
C
n
where c(z v) = cz v for c in C. Since C
n
= R
2n
as real
vector spaces, we have a C-vector space isomorphism
C
R
C
n

= C
R
R
2n

= C
2n
.
Thats interesting: restricting scalars on C
n
to make it a real vector space and then ex-
tending scalars back up to C does not give us C
n
back, but instead two copies of C
n
. The
point is that when we restrict scalars, the real vector space C
n
forgets it is a complex vector
space. So the base extension to C doesnt remember that it used to be a complex vector
space.
Quite generally, if V is a nite-dimensional complex vector space and we view it as a real
vector space, its base extension C
R
V as a complex vector space is not V but a direct
sum of two copies of V . Lets do a dimension check: as real vector spaces, dim
R
(C
R
V ) =
dim
R
(C) dim
R
(V ) = 2(2 dim
C
V ) = 4 dim
C
V and dim
R
(V V ) = 2 dim
C
(V V ) =
11
We saw S R R[X] and S[X] are isomorphic as R-modules in Example 4.14 when S R, and it holds
now for any R
f
S.
TENSOR PRODUCTS 31
4 dim
C
V , so the two dimensions match. This match is of course not a proof that there
is a natural isomorphism C
R
V V V of complex vector spaces. Work out such an
isomorphism as an exercise.
To get our bearing on this example, lets compare an S-module N with the S-module
S
R
N (where s

(s n) = s

s n). Since N is already an S-module, should S

R
N

= N?
If you think so, reread Example 6.10 (R = R, S = C, N = C
n
). Scalar multiplication
SN N is R-bilinear, so there is an R-linear map : S
R
N N where (sn) = sn.
This map is also S-linear: (st) = s(t). To check this, since both sides are additive in t it
suces to check the case of elementary tensors, and
(s(s

n)) = ((ss

) n) = ss

n = s(s

n) = s(s

n).
In the other direction, the function : N S
R
N where (n) = 1 n is R-linear but is
generally not S-linear since (sn) = 1sn has no reason to be s(n) = sn because were
using
R
, not
S
. We have created natural maps : S
R
N N and : N S
R
N;
are they inverses? Its unlikely, since is S-linear and is generally not. But lets work
out the composites and see what happens. In one direction,
((n)) = (1 n) = 1 n = n.
In the other direction,
((s n)) = (sn) = 1 sn = s n
in general. So is the identity but is usually not the identity. Since = id
N
,
is a section to , so N is a direct summand S
R
N. Explicitly, S
R
N

= ker N by
s n (s n 1 sn, sn) and its inverse map is (k, n) k + 1 n. The phenomenon
that S
R
N is typically larger than N when N is an S-module can be remembered by the
example C
R
C
n
= C
2n
.
Theorem 6.11. For R-modules M
i

iI
, there is an S-module isomorphism
S
R

iI
M
i

=

iI
(S
R
M
i
).
Proof. Since S is an R-module, by Theorem 5.4 there is an R-module isomorphism
: S
R

iI
M
i

iI
(S
R
M
i
)
where (s (m
i
)
iI
) = (s m
i
)
iI
. To show is an S-module isomorphism, we just have
to check is S-linear, since we already know is additive and a bijection. It is obvious that
(st) = s(t) when t is an elementary tensor, and since both (st) and s(t) are additive
in t the case of general tensors follows.
The analogue of Theorem 6.11 for direct products is false. There is a natural S-linear
map S
R

iI
M
i

iI
(S
R
M
i
), but it need not be an isomorphism: Q
Z

i1
Z/p
i
Z
is nonzero (Remark 5.5) but

i1
(Q
Z
Z/p
i
Z) is 0.
We now put base extensions to work. Let M be a nitely generated R-module, say with
n generators. That is the same as saying there is a linear surjection R
n
M. To say M
contains a subset of d linearly independent elements is the same as saying there is a linear
injection R
d
M. If both R
n
M and R
d
M, it is natural to suspect d n, i.e., the
size of a spanning set should always be an upper bound on the size of a linearly independent
subset. Is it really true? If R is a eld, so modules are vector spaces, we can use dimension
32 KEITH CONRAD
inequalities on R
d
, M, and R
n
to see d n. But if R is not a eld, then what? We will
settle the issue in the armative when R is a domain, by tensoring M with the fraction eld
of R to reduce to the case of vector spaces. We rst tensored R-modules with the fraction
eld of R in Theorem 4.21, but not much use was made of the vector space structure of the
tensor product with a eld. Now we exploit it.
Theorem 6.12. Let R be a domain with fraction eld K. For a nitely generated R-module
M, K
R
M is nite-dimensional as a K-vector space and dim
K
(K
R
M) is the maximal
number of R-linearly independent elements in M and is a lower bound on the size of a
spanning set for M. In particular, the size of any linearly independent subset of M is less
than or equal to the size of any spanning set of M.
Proof. If x
1
, . . . , x
n
is any spanning set for M as an R-module then 1 x
1
, . . . , 1 x
n
span
K
R
M as a K-vector space, so dim
K
(K
R
M) n.
Let y
1
, . . . , y
d
be R-linearly independent in M. We will show 1 y
i
is K-linearly
independent in K
R
M, so d dim
K
(K
R
M). Suppose

d
i=1
c
i
(1 y
i
) = 0 with c
i
K.
Write c
i
= a
i
/b using a common denominator b in R. Then 0 = 1/b

d
i=1
a
i
y
i
in K
R
M.
By Corollary 4.22, this implies

d
i=1
a
i
y
i
M
tor
, so

d
i=1
ra
i
y
i
= 0 in M for some nonzero
r R. By linear independence of the y
i
s over R, every ra
i
is 0, so every a
i
is 0 (R is a
domain). Thus every c
i
= a
i
/b is 0.
It remains to prove M has a linearly independent subset of size dim
K
(K
R
M). Let
e
1
, . . . , e
d
be a linearly independent subset of M, where d is maximal. (Since d
dim
K
(K
R
M), there is a maximal d.) For every m M, e
1
, . . . , e
d
, m has to be
linearly dependent, so there is a nontrivial R-linear relation
a
1
e
1
+ +a
d
e
d
+am = 0.
Necessarily a = 0, as otherwise all the a
i
s are 0 by linear independence of the e
i
s. In
K
R
M,
d

i=1
a
i
(1 e
i
) +a(1 m) = 0
and from the K-vector space structure on K
R
M we can solve for 1 m as a K-linear
combination of the 1e
i
s. Therefore 1e
i
spans K
R
M as a K-vector space. This set
is also linearly independent over K by the previous paragraph, so it is a basis and therefore
d = dim
K
(K
R
M).
While M has at most dim
K
(K
R
M) linearly independent elements and this upper bound
is achieved, any spanning set has at least dim
K
(K
R
M) elements but this lower bound is
not necessarily reached. For example, if R is not a eld and M is a torsion module (e.g., R/I
for I a nonzero proper ideal) then K
R
M = 0 and M certainly doesnt have a spanning
set of size 0 if M = 0. It is also not true that niteness of dim
K
(K
R
M) implies M is
nitely generated as an R-module. Take R = Z and M = Q, so Q
Z
M = Q
Z
Q

= Q
(Theorem 4.17), which is nite-dimensional over Q but M is not nitely generated over Z.
The maximal number of linearly independent elements in an R-module M, for R a do-
main, is called the rank of M.
12
This use of the word rank is consistent with its usage
for nite free modules: if M is free with an R-basis of size n then K
R
M has a K-basis
of size n by Theorem 6.7.
12
When R is not a domain, this concept of rank for R-modules is not quite the right one.
TENSOR PRODUCTS 33
Example 6.13. A nonzero ideal I in a domain R has rank 1. We can see this in two ways.
First, any two nonzero elements in I are linearly dependent over R, so the maximal number
of R-linearly independent elements in I is 1. Second, K
R
I

= K as K-vector spaces (in
Theorem 4.19 we showed they are isomorphic as R-modules, but that isomorphism is also
K-linear; check!), so dim
K
(K
R
I) = 1.
Example 6.14. A nitely generated R-module M has rank 0 if and only if it is a torsion
module, since K
R
M = 0 if and only if M is a torsion module.
Since K
R
M

= K
R
(M/M
tor
) as K-vector spaces (the isomorphism between them
as R-modules in Theorem 4.21 is easily checked to be K-linear check!), M and M/M
tor
have the same rank.
We return to general R, no longer a domain, and see how to make the tensor product of
an R-module and S-module into an S-module.
Theorem 6.15. Let M be an R-module and N be an S-module.
(1) The additive group M
R
N has a unique structure of S-module such that s(mn) =
msn for s S. This is compatible with the R-module structure in the sense that
rt = f(r)t for r R and t M
R
N.
(2) The S-module M
R
N is isomorphic to (S
R
M)
S
N by sending m n to
(1 m) n.
The point of part 2 is that it shows how the S-module structure on M
R
N can be
described as an ordinary S-module tensor product once we know the idea of base extending
M to the S-module S
R
M.
Part 2 has both R-module and S-module tensor products. This is the rst time that we
must decorate the tensor product sign explicitly. Up to now it was actually unnecessary, as
all the tensor products were over R.
Proof. (1) This is similar to the proof of Theorem 6.4 (which is the special case N = S).
We just sketch the idea.
Since every tensor is a sum of elementary tensors, declaring how s S scales elementary
tensors in M
R
N determines its scaling on all tensors. To show the rule s(mn) = msn
really corresponds to an S-module structure, for each s S we consider the function
M N M
R
N given by (m, n) m sn. This is R-bilinear in m and n, so there is
an R-linear map
s
: M
R
N M
R
N such that
s
(m n) = m sn on elementary
tensors. Dene a multiplication S (M
R
N) M
R
N by st :=
s
(t). Checking the
four scaling axioms to make M
R
N an S-module is left to the reader as an exercise.
To check rt = f(r)t for r R and t M
R
N, both sides are additive in t so it suces
to check equality when t = mn is an elementary tensor. In that case r(mn) = mrn =
mf(r)n = f(r)(mn).
(2) Let MN (S
R
M)
S
N by (m, n) (1m) n. Lets check this is R-bilinear:
Biadditivity is clear. For R-scaling, we want to check
(1 rm) n
?
= r((1 m) n), (1 m) rn
?
= r((1 m) n).
Decorating elementary tensors with
R
or
S
to emphasize in what kind of tensor product
they live, we have
(1
R
rm)
S
n = (r(1
R
m))
S
n = (f(r)(1
R
m))
S
n = f(r)((1
R
m)
S
n)
34 KEITH CONRAD
and
(1
R
m)
S
rn = (1
R
m)
S
f(r)n = f(r)((1
R
m)
S
n).
Now the universal mapping property of tensor products gives an R-linear map : M
R
N
(S
R
M)
S
N where (m
R
n) = (1
R
m)
S
n. This is exactly the map we were looking
for, but we only know it is R-linear so far. It is also S-linear: (st) = s(t). To check this,
it suces by additivity of to focus on the case of an elementary tensor:
(s(m
R
n)) = (m
R
sn) = (1
R
m)
S
sn = s((1
R
m)
S
n) = s(m
R
n).
To show is an isomorphism, we create an inverse map (S
R
M)
S
N M
R
N. The
function S MN M
R
N given by (s, m, n) msn is R-trilinear, so by Theorem
5.6 there is an R-bilinear map B: S
R
M N M
R
N where B(s m, n) = msn.
This function is in fact S-bilinear:
B(st, n) = sB(t, n), B(t, sn) = sB(t, n).
To check these equations, the additivity of both sides of the equations in t reduces us to
case when t is an elementary tensor. Writing t = s

m,
B(s(s

m), n) = B(ss

m, n) = mss

n = ms(s

n) = s(ms

n) = sB(s

m, n)
and
B(s

m, sn) = ms

(sn)ms(s

n) = s(ms

n) = sB(s

m, n).
Now the universal mapping property of the tensor product for S-modules tells us there is
an S-linear map : (S
R
M)
S
N M
R
N such that (t n) =
n
(t).
It is left to the reader to check and are identity functions, so is an S-module
isomorphism.
In addition to M
R
N being an S-module because N is, the tensor product N
R
M
in the other order has a unique S-module structure where s(n m) = sn m, and this is
proved in a similar way.
Example 6.16. For an S-module N, lets show R
k

R
N

= N
k
as S-modules. By Theorem
5.4, R
k

R
N

= (R
R
N)
k
= N
k
as R-modules, an explicit isomorphism : R
k

R
N N
k
being ((r
1
, . . . , r
k
) n) = (r
1
n, . . . , r
k
n). Lets check is S-linear: (st) = s(t). Both
sides are additive in t, so we only need to check when t is an elementary tensor:
(s((r
1
, . . . , r
k
) n)) = ((r
1
, . . . , r
k
) sn) = (r
1
sn, . . . , r
k
sn) = s((r
1
, . . . , r
k
) n).
To reinforce the S-module isomorphism
(6.3) M
R
N

= (S
R
M)
S
N
from Theorem 6.15(2), we will write out the isomorphism in both directions on appropriate
tensors:
mn (1 m) n, (s m) n msn.
Corollary 6.17. If M and M

are isomorphic R-modules then M

R
N and M

R
N are
isomorphic S-modules, as are N
R
M and N
R
M

.
Proof. We will show M
R
N

= M

R
N as S-modules. The other one is similar.
Let : M M

be an R-module isomorphism. To write down an S-module isomorphism

M
R
N M

R
N, we will write down an R-module isomorphism which is also S-linear.
Let M N M

R
N by (m, n) (m) n. This is R-bilinear (check!), so we get
TENSOR PRODUCTS 35
an R-linear map : M
R
N M

R
N such that (m n) = (m) n. This is also
S-linear: (st) = s(t). Since is additive, it suces to check this when t = mn:
(s(mn)) = (msn) = (m) sn = s((m) n) = s(mn).
Using the inverse map to we get an R-linear map : M

R
N M
R
N which is
also S-linear, and a computation on elementary tensors shows and are inverses of each
other.
Example 6.18. We can use tensor products to prove the well-denedness of ranks of nite
free R-modules when R is not the zero ring. Suppose R
m
= R
n
as R-modules. Pick a
maximal ideal m in R (Zorns lemma) and R/m
R
R
m
= R/m
R
R
n
as R/m-vector spaces
by Corollary 6.17. Therefore (R/m)
m
= (R/m)
n
as R/m-vector spaces (Example 6.8), so
taking dimensions of both sides over R/m tells us m = n.
Heres a conundrum. If N and N

are both S-modules, then we can make N

R
N

into
an S-module in two ways: s(nn

) = snn

and s(nn

) = nsn

. In the rst S-module

structure, N

only matters as an R-module. In the second S-module structure, N only

matters as an R-module. These two S-module structures on N
R
N

are not generally the

same because the tensor product is
R
, not
S
, so snn

need not equal nsn

. But are
the two S-module structures on N
R
N

at least isomorphic to each other? In general, no.

Example 6.19. Let R = Z, S = Z[

14], and p = (3, 1 +

14) be an ideal in S. We
will show the two S-module structures on S
Z
p, coming from scaling by S on the left and
the right in elementary tensors, are not isomorphic to each other.
First we show p is nonprincipal. Let p = x : x p = (3, 1

14). Check pp = 3S. If

p = (a +b

14), then p = (a b

14) and pp = (a +b

14)(a b

14) = (a
2
+14b
2
).
For this to be (3) forces a
2
+ 14b
2
= 3, but this equation has no integral solution.
As Z-modules, S and p are both free of rank 2, with respective Z-bases 1,

14 and
3, 1 +

14. When S
Z
p is an S-module by scaling on the left, p only matters as a Z-
module, so S
Z
p

= S
Z
Z
2
as S-modules by Corollary 6.17. By Example 6.8, S
Z
Z
2
= S
2
as S-modules. So S
Z
p is a free S-module of rank 2 (with S-basis 13 and 1(1+

14)).
Making S
Z
p into an S-module by scaling on the right, S
Z
p

= Z
2

Z
p

= p
2
as S-modules.
It remains to show p
2
= p p is not isomorphic to S
2
, i.e., p
2
is not free of rank 2
as an S-module. The only proof I know for showing p p is not free of rank 2 as an S-
module uses exterior powers, which lie beyond the scope of this note, so we cant discuss the
proof here. However, its important to address a natural question: since p is not principal,
isnt it obvious that p p cant be free because p is not free? No. A direct sum of two
nonfree modules can be free. For instance, in Z[

5] the ideals I = (3, 1 +

5) and
J = (3, 1

5) are both nonprincipal but it can be shown that I J

= Z[

5] Z[

5]
as Z[

5]-modules. The reason a direct sum of non-free modules could be free is that there
is more room to move around in the direct sum than just within the direct summands, and
this extra room might contain a basis. Therefore showing p
2
is not a free S-module of rank
2 requires some real work.
The upshot is that if you want to make N
R
N

into an S-module where N and N

are
both naturally S-modules, you generally have to specify whether S scales on the left or the
right. It would be a theorem to prove that in some particular example the two S-modules
structures on N
R
N

are in fact identical.

36 KEITH CONRAD
The next theorem collects a number of earlier tensor product isomorphisms for R-modules
and shows the same maps are S-module isomorphisms when one of the R-modules in the
tensor product is an S-module.
Theorem 6.20. Let M and M

be R-modules and N and N

be S-modules.
(1) There is a unique S-module isomorphism
M
R
N N
R
M
where mn nm. In particular, S
R
M and M
R
S are isomorphic S-modules.
(2) There are unique S-module isomorphisms
(M
R
N)
R
M

N
R
(M
R
M

)
where (mn) m

n (mm

) and
(M
R
N)
R
M

= M
R
(N
R
M

)
where (mn) m

m(n m

).
(3) There is a unique S-module isomorphism
(M
R
N)
S
N

M
R
(N
S
N

)
where (mn) n

m(n n

).
(4) There is a unique S-module isomorphism
N
R
(M M

) (N
R
M) (N
R
M

)
where n (m, m

) (n, m) (n, m

).
In the rst, second, and fourth parts, we are using R-module tensor products only and
then endowing them with S-module structure from one of the factors being an S-module
(Theorem 6.15). In the third part we have both
R
and
S
.
Proof. There is a canonical R-module isomorphism M
R
N N
R
M where m n
n m. This map is S-linear using the S-module structure on both sides (check!), so it is
an S-module isomorphism. This settles part 1.
Part 2, like part 1, only involves R-module tensor products, so from earlier work we
know there is an R-module isomorphism : (M
R
N)
R
M

N
R
(M
R
M

) where
((mn)m

) = n(mm

). Using the S-module structure on M

R
N, (M
R
N)M

,
and N
R
(M
R
M

), is S-linear (check!), so it is an S-module isomorphism. To derive

(M
R
N)
R
M

= M
R
(N
R
M

) from (M
R
N)
R
M

= N
R
(M
R
M

), use a
few commutativity isomorphisms.
Part 3 resembles associativity of tensor products. We will in fact derive part 3 from such
associativity for
S
:
(M
R
N)
S
N

= ((S
R
M)
S
N)
S
N

by (6.3)

= (S
R
M)
S
(N
S
N

) by associativity of
S

= M
R
(N
S
N

) by (6.3).
These successive S-module isomorphisms have the eect
(mn) n

((1 m) n) n

(1 m) (n n

)
m(n n

),
which is what we wanted.
TENSOR PRODUCTS 37
For part 4, there is an R-module isomorphism N
R
(MM

) (N
R
M) (N
R
M

)
by Theorem 5.4. Now its just a matter of checking this map is S-linear using the S-module
structure on both sides coming from N being an S-module, and this is left to the reader.
As an alternate proof, we have a chain of S-module isomorphisms
N
R
(M M

)

= N
S

S
R
(M M

by part 1 and (6.3)

= N
S
((S
R
M) (S
R
M

)) by Theorem 6.11

= (N
S
(S
R
M)) (N
S
(S
R
M

)) by Theorem 5.4

= (N
R
M) (N
R
M

) by part 1 and (6.3).

Of course one needs to trace through these isomorphisms to check the overall result has the
eect intended on elementary tensors, and it does (exercise).
The last part of Theorem 6.20 extends to arbitrary direct sums: the natural R-module
isomorphism N
R

iI
M
i

=

iI
(N
R
M
i
) is also an S-module isomorphism.
As an application of Theorem 6.20, we can show the base extension of an R-module
tensor product is the S-module tensor product of the base extensions:
Corollary 6.21. For R-modules M and M

, there is a unique S-module isomorphism

S
R
(M
R
M

) (S
R
M)
S
(S
R
M

)
where s
R
(m
R
m

) s((1
R
m)
S
(1
R
m

)).
Proof. Since M
R
M

is additively spanned by all mm

, S
R
(M
R
M

) is additively
spanned by all s
R
(m
R
m

). Therefore an S-linear (or even additive) map out of

S
R
(M
R
M

) is determined by its values on the tensors s

R
(m
R
m

).
We have the S-module isomorphisms
S
R
(M
R
M

)

= M
R
(S
R
M

) by Theorem 6.20(2)

= (S
R
M)
S
(S
R
M

) by (6.3).
The eect of these isomorphisms on s
R
(m
R
m

) is
s
R
(m
R
m

) m
R
(s
R
m

)
(1
R
m)
S
(s
R
m

)
= (1
R
m)
S
s(1
R
m

)
= s((1
R
m)
S
(1
R
m

)),
as desired. The eect of the inverse isomorphism on (s
1

R
m)
S
(s
2

R
m

) is
(s
1

R
m)
S
(s
2

R
m

) m
R
s
1
(s
2

R
m

)
= m
R
((s
1
s
2
)
R
m

)
s
1
s
2

R
(m
R
m

).

Theorem 6.21 could also be proved by showing the S-module S

R
(M
R
M

) has the
universal mapping property of (S
R
M)
S
(S
R
M

) as a tensor product of S-modules.

That is left as an exercise.
38 KEITH CONRAD
Corollary 6.22. For any R-modules M
1
, . . . , M
k
,
S
R
(M
1

R

R
M
k
)

= (S
R
M
1
)
S

S
(S M
k
)
as S-modules, where s
S
(m
1

R

R
m
k
) s((1
R
m
1
)
S

S
(1
R
m
k
)). In
particular, S
R
(M

R
k
)

= (S
R
M)

S
k
as S-modules.
Proof. Induct on k.
Example 6.23. For any real vector space V , C
R
(V
R
V )

= (C
R
V )
C
(C
R
V ).
The middle tensor product sign on the right is over C, not R. Note that C
R
(V
R
V )

=
(C
R
V )
R
(C
R
V ) when V = 0, as the two sides have dierent dimensions over R
(what are they?).
The base extension M S
R
M turns R-modules into S-modules in a systematic way.
So does M M
R
S, and this is essentially the same construction. This suggests there
should be a universal mapping problem about R-modules and S-modules which is solved
by base extension, and there is: it is the universal device for turning R-linear maps from
M to S-modules into S-linear maps.
Theorem 6.24. Let M be an R-module. For every S-module N and R-linear map : M
N, there is a unique S-linear map
S
: S
R
M N such that the diagram
M
m1m

A
A
A
A
A
A
A
A
S
R
M

S
.
N
commutes.
This says the single R-linear map M S
R
M from M to an S-module explains all
other R-linear maps from M to S-modules using composition of it with S-linear maps from
S
R
M to S-modules.
Proof. Assume there is such an S-linear map
S
. We will derive a formula for it on elemen-
tary tensors:

S
(s m) =
S
(s(1 m)) = s
S
(1 m) = s(m).
This shows
S
is unique if it exists.
To prove existence, consider the function S M N by (s, m) s(m). This is R-
bilinear (check!), so there is an R-linear map
S
: S
R
M N such that
S
(sm) = s(m).
Using the S-module structure on S
R
M,
S
is S-linear.
For in Hom
R
(M, N),
S
is in Hom
S
(S
R
M, N). Because
S
(1 m) = (m), we can
recover from
S
. But even more is true.
Theorem 6.25. Let M be an R-module and N be an S-module. The function
S
is
an S-module isomorphism Hom
R
(M, N) Hom
S
(S
R
M, N).
How is Hom
R
(M, N) an S-module? Values of these functions are in N, an S-module, so
S scales any function M N to a new function M N by just scaling the values.
Proof. For and

in Hom
R
(M, N), ( +

)
S
=
S
+
S
and (s)
S
= s
S
by checking
both sides are equal on all elementary tensors in S
R
M. Therefore
S
is S-linear.
Its injectivity is discussed above (
S
determines ).
TENSOR PRODUCTS 39
For surjectivity, let h: S
R
M N be S-linear. Set : M N by (m) = h(1 m).
Then is R-linear and
S
(s m) = s(m) = sh(1 m) = h(s(1 m)) = h(s m), so
h =
S
since both are additive and are equal at all elementary tensors.
The S-module isomorphism
(6.4) Hom
R
(M, N)

= Hom
S
(S
R
M, N)
should be thought of as analogous to the R-module isomorphism
(6.5) Hom
R
(M, Hom
R
(N, P))

= Hom
R
(M
R
N, P)
from Theorem 5.7, where
R
N is left adjoint to Hom
R
(N, ). (In (6.5), N and P are
R-modules, not S-modules! Were using the same notation as in Theorem 5.7.) If we look
at (6.4), we see S
R
is applied to M on the right but nothing special is applied to N
on the left. Yet there is something dierent about N on the two sides of (6.4). It is an
S-module on the right side of (6.4), but on the left side it is being treated as an R-module
(restriction of scalars). That changes N, but we have introduced no notation to reect this.
We still just write it as N. Lets now write Res
S/R
(N) to denote N as an R-module. It is
the same underlying additive group as N, but the scalars are now taken from R with the
rule rn = f(r)n. The appearance of (6.4) now looks like this:
(6.6) Hom
R
(M, Res
S/R
(N))

= Hom
S
(S
R
M, N).
So extension of scalars (from R-modules to S-modules) is left adjoint to restriction of scalars
(from S-modules to R-modules) in a similar way that
R
M is left adjoint to Hom
R
(M, ).
Using this new notation for restriction of scalars, the important S-module isomorphism
(6.3) can be written more explicitly as
M
R
Res
S/R
(N)

= (S
R
M)
S
N,
Theorem 6.26. Let M be an R-module and N and P be S-modules. There is an S-module
isomorphism
Hom
S
(M
S
N, P)

= Hom
R
(M, Res
S/R
(Hom
S
(N, P))).
Example 6.27. Taking N = S,
Hom
S
(S
R
M, P)

= Hom
R
(M, Res
S/R
(P))
since Hom
S
(S, P)

= P. We have recovered S
R
being left adjoint to Res
S/R
.
Example 6.28. Taking S = R, so N and P are now R-modules,
Hom
R
(M
R
N, P)

= Hom
R
(M, Hom
R
(N, P)).
We have recovered
R
N being left adjoint to Hom
R
(N, ) for R-modules N.
These two consequences of Theorem 6.26 are results we have already seen, and in fact we
are going to use them in the proof, so they are together equivalent to Theorem 6.26.
Proof. Since M
R
N

= (S
R
M)
S
N as S-modules,
Hom
S
(M
R
N, P)

= Hom
S
((S
R
M)
S
N, P).
Since
S
N is left adjoint to Hom
S
(N, ),
Hom
S
((S
R
M)
S
N, P)

= Hom
S
(S
R
M, Hom
S
(N, P)).
Since S
R
is left adjoint to Res
S/R
,
Hom
S
(S
R
M, Hom
S
(N, P))

= Hom
R
(M, Res
S/R
(Hom
S
(N, P))).
40 KEITH CONRAD
Combining these three isomorphisms,
Hom
S
(M
R
N, P)

= Hom
R
(M, Res
S/R
(Hom
S
(N, P))).
Here is an explicit (overall) isomorphism. If : M
R
N P is S-linear there is an R-linear
map L

: M Hom
S
(N, P) by L

(m) = (m()). If : M Hom

S
(N, P) is R-linear
then M N P by (m, n) (m)(n) is R-bilinear and (m)(sn) = s(m)(n), so the
corresponding R-linear map

L

: M
R
N P where

L

(m n) = (m)(n) is S-linear.
The functions L

and

L

are S-linear and are inverses.

References
[1] D. V. Alekseevskij, V. V. Lychagin, A. M. Vinogradov, Geometry I, Springer-Verlag, Berlin, 1991.
[2] R. Grone, Decomposable Tensors as a Quadratic Variety, Proc. Amer. Math. Soc. 64 (1977), 227230.
[3] P. Halmos, Finite-Dimensional Vector Spaces, Springer-Verlag, New York, 1974.
[4] B. ONeill, Semi-Riemannian Geometry, Academic Press, New York, 1983.
TENSOR PRODUCTS, II
KEITH CONRAD
1. Introduction
Continuing our study of tensor products, we will see how to combine two linear maps
M M

and N N

into a linear map M

R
N M

R
N

. This leads to at modules

and linear maps between base extensions. Then we will look at special features of tensor
products of vector spaces and discuss tensor products of R-algebras.
2. Tensor Products of Linear Maps
If M

M

and N

N

are linear, then we get a linear map between the direct

sums, MN

M

, dened by ()(m, n) = ((m), (n)). We want to dene

a linear map M
R
N M

R
N

such that mn (m) (n).

Start with the map M N M

R
N

where (m, n) (m) (n). This is R-

bilinear, so the universal mapping property of the tensor product gives us an R-linear map
M
R
N

M

R
N

where ( )(mn) = (m) (n), and more generally

( )(m
1
n
1
+ +m
k
n
k
) = (m
1
) (n
1
) + +(m
k
) (n
k
).
We call the tensor product of and , but be careful to appreciate that is not
denoting an elementary tensor. This is just notation for a new linear map on M
R
N.
When M

M

is linear, the linear maps N

R
M
1
N
R
M

or M
R
N
1

R
N are called tensoring with N. The map on N is the identity, so (1 )(n m) =
n(m) and (1)(mn) = (m) n. This construction will be particularly important
for base extensions in Section 4.
Example 2.1. Tensoring inclusion aZ
i
Z with Z/bZ is aZ
Z
Z/bZ
i1
Z
Z
Z/bZ,
where (i 1)(ax y mod b) = ax y mod b. Since Z
Z
Z/bZ

= Z/bZ by multiplication,
we can regard i 1 as a function aZ
Z
Z/bZ Z/bZ where ax y mod b axy mod b.
Its image is {az mod b : z Z/bZ}, which is dZ/bZ where d = (a, b); this is 0 if b|a and is
Z/bZ if (a, b) = 1.
Example 2.2. Let A = (
a b
c d
) and A

= (
a

) in M
2
(R). Then A and A

are both linear

maps R
2
R
2
, so AA

is a linear map from (R

2
)
2
= R
2

R
R
2
back to itself. Writing e
1
and e
2
for the standard basis vectors of R
2
, lets compute the matrix for AA

on (R
2
)
2
1
2 KEITH CONRAD
with respect to the basis {e
1
e
1
, e
1
e
2
, e
2
e
1
, e
2
e
2
}. By denition,
(AA

)(e
1
e
1
) = Ae
1
A

e
1
= (ae
1
+ce
2
) (a

e
1
+c

e
2
)
= aa

e
1
e
1
+ac

e
1
e
2
+ca

e
2
e
1
+cc

e
2
e
2
,
(AA

)(e
1
e
2
) = Ae
1
A

e
2
= (ae
1
+ce
2
) (b

e
1
+d

e
2
)
= cb

e
1
e
1
+ad

e
1
e
2
+cb

e
2
e
2
+cd

e
2
e
2
,
and similarly
(AA

)(e
2
e
1
) = ba

e
1
e
1
+bc

e
1
e
2
+da

e
2
e
1
+dc

e
2
e
2
,
(AA

)(e
2
e
2
) = bb

e
1
e
1
+bd

e
1
e
2
+db

e
2
e
1
+dd

e
2
e
2
.
Therefore the matrix for AA

is

aa

ab

ba

bb

ac

ad

bc

bd

ca

cb

da

db

cc

cd

dc

dd

aA

bA

cA

dA

.
So Tr(AA

) = a(a

+d

) +d(a

+d

) = (a +d)(a

+d

) = (Tr A)(Tr A

), and det(AA

)
looks painful to compute from the matrix. Well do this later, in Example 2.7, in an almost
painless way.
If, more generally, A M
n
(R) and A

M
n
(R) then the matrix for AA

with respect
to the standard basis for R
n

R
R
n

is the block matrix (a

ij
A

) where A = (a
ij
). This
nn

nn

matrix is called the Kronecker product of A and A

, and is not symmetric in the

roles of A and A

in general (just as AA

= A

A in general). In particular, I
n
A

has
block matrix representation (
ij
A

), whose determinant is (det A

)
n
.
The construction of tensor products (Kronecker products) of matrices has the following
application to nding polynomials with particular roots.
Theorem 2.3. Let K be a eld and suppose A M
m
(K) and B M
n
(K) have eigenvalues
and in K. Then AI
n
+I
m
B has eigenvalue + and AB has eigenvalue .
Proof. We have Av = v and Bw = w for some v K
m
and w K
n
. Then
(AI
n
+I
m
B)(v w) = Av w +v Bw
= v w +v w
= ( +)(v w)
and
(AB)(v w) = Av Bw = v w = (v w),

TENSOR PRODUCTS, II 3
Example 2.4. The numbers

2 and

3 are eigenvalues of A = (
0 2
1 0
) and B = (
0 3
1 0
). A
matrix with eigenvalue

2 +

3 is
AI
2
+I
2
B =

0 0 2 0
0 0 0 2
1 0 0 0
0 1 0 0

0 3 0 0
1 0 0 0
0 0 0 3
0 0 1 0

0 3 2 0
1 0 0 2
1 0 0 3
0 1 1 0

,
whose characteristic polynomial is T
4
10T
2
+1. So this is a polynomial with

2 +

3 as
a root.
Although we stressed that is not an elementary tensor, but rather is the notation
for a linear map, and belong to the R-modules Hom
R
(M, M

) and Hom
R
(N, N

), so
one could ask if the actual elementary tensor in Hom
R
(M, M

)
R
Hom
R
(N, N

) is
related to the linear map : M
R
N M

R
N

.
Theorem 2.5. There is a linear map
Hom
R
(M, M

)
R
Hom
R
(N, N

) Hom
R
(M
R
N, M

R
N

)
which sends the elementary tensor to the linear map . When M, M

, N, and
N

are nite free, this is an isomorphism.

Proof. We adopt the temporary notation T(, ) for the linear map we have previously
written as , so we can use to mean an elementary tensor in the tensor product
of Hom-modules. So T(, ): M
R
N M

R
N

is the linear map sending every mn

to (m) (n).
Dene Hom
R
(M, M

)Hom
R
(N, N

) Hom
R
(M
R
N, M

R
N

) by (, ) T(, ).
This is R-bilinear. For example, to show T(r, ) = rT(, ), both sides are linear maps
so to prove they are equal it suces to check they are equal at the elementary tensors in
M
R
N:
T(r, )(mn) = (r)(m) (n) = r(m) (n) = r((m) (n)) = rT(, )(mn).
The other bilinearity conditions are left to the reader.
From the universal mapping property of tensor products, there is a unique R-linear map
Hom
R
(M, M

)
R
Hom
R
(N, N

) Hom
R
(M
R
N, M

R
N

) where T(, ).
Suppose M, M

, N, and N

are all nite free R-modules. Let them have respective bases
{e
i
}, {e

}, {f
j
}, and {f

}. Then Hom
R
(M, M

) and Hom
R
(N, N

) are both free with bases

{E
i

i
} and {

E
j

j
}, where E
i

i
: M M

is the linear map sending e

i
to e

and is 0 at other
basis vectors of M, and

E
j

j
: N N

is dened similarly. (The matrix representation

of E
i

i
with respect to the chosen bases of M and M

has a 1 in the (i

, i) position and
0 elsewhere, thus justifying the notation.) A basis of Hom
R
(M, M

)
R
Hom
R
(N, N

) is
4 KEITH CONRAD
{E
i

E
j

j
} and T(E
i

E
j

j
): M
R
N M

R
N

has the eect

T(E
i

E
j

j
)(e

) = E
i

i
(e

)

E
j

j
(f

)
=
i
e

i

j
f

, if = i and = j,
0, otherwise,
so T(E
i

i
E
j

j
) sends e
i
f
j
to e

and sends other members of the basis of M

R
N to 0.
That means the linear map Hom
R
(M, M

)
R
Hom
R
(N, N

) Hom
R
(M
R
N, M

R
N

)
sends a basis to a basis, so it is an isomorphism when the modules are nite free.
The upshot of Theorem 2.5 is that Hom
R
(M, M

)
R
Hom
R
(N, N

) naturally acts as
linear maps M
R
N M

R
N

and it turns the elementary tensor into the linear

map weve been writing as . This justies our use of the notation for the linear
map, but it should be kept in mind that we will continue to write for the linear map
itself (on M
R
N) and not for an elementary tensor in a tensor product of Hom-modules.
Properties of tensor products of modules carry over to properties of tensor products of
linear maps, by checking equality on all tensors. For example, if
1
: M
1
N
1
,
2
: M
2

N
2
, and
3
: M
3
N
3
are linear maps, we have
1
(
2

3
) = (
1

2
) (
1

3
)
and (
1

2
)
3
=
1
(
2

3
), in the sense that the diagrams
M
1

R
(M
2
M
3
)

1
(
2

3
)

N
1

R
(N
2
N
3
)

(M
1

R
M
2
) (M
1

R
M
3
)
(
1

2
)(
1

3
)

(N
1

R
N
2
) (N
1

R
N
3
)
and
M
1

R
(M
2

R
M
3
)

1
(
2

3
)

N
1

R
(N
2

R
N
3
)

(M
1

R
M
2
)
R
M
3
(
1

2
)
3

(N
1

R
N
2
)
R
N
3
commute, with the vertical maps being the canonical isomorphisms.
The properties of the next theorem are called the functoriality of the tensor product of
linear maps.
Theorem 2.6. For R-modules M and N, id
M
id
N
= id
M
R
N
. For linear maps M

, M

, N

N

, and N

,
(

) ( ) = (

) (

)
as linear maps from M
R
N to M

R
N

.
Proof. The function id
M
id
N
is a linear map from M
R
N to itself which xes every
elementary tensor, so it xes all tensors.
Since (

) () and (

) (

) are linear maps, to prove their equality it

suces to check they have the same value at any elementary tensor m n, at which they
both have the value

((m))

((n)).
TENSOR PRODUCTS, II 5
Example 2.7. The composition rule for tensor products of linear maps helps us compute
determinants of tensor products of linear operators. Let M and N be nite free R-modules
of respective ranks k and . For linear operators M

M and N

N, we will compute
det( ) by breaking up into a composite of two maps M
R
N M
R
N:
= ( id
N
) (id
M
),
so the multiplicativity of the determinant implies det( ) = det( id
N
) det(id
M
)
and we are reduced to the case when one of the factors is an identity map. Moreover, the
isomorphism M
R
N N
R
M where mn n m converts id
N
into id
N
, so
det( id
N
) = det(id
N
), so
det( ) = det(id
N
) det(id
M
).
What are the determinants on the right side? Pick bases e
1
, . . . , e
k
of M and e

1
, . . . , e

of
N. We will use the k elementary tensors e
i
e

j
as a bases of M
R
N. Let [] be the
matrix of in the ordered basis e
1
, . . . , e
k
. Since (id
N
)(e
i
e

j
) = (e
i
) e

j
, lets order
the basis of M
R
N as
e
1
e

1
, . . . , e
k
e

1
, . . . , e
1
e

, . . . , e
k
e

.
The k k matrix for id
N
in this ordered basis is the block diagonal matrix

[] O O
O [] O
.
.
.
.
.
.
.
.
.
.
.
.
O O []

,
whose determinant is (det )

.
Thus
det( ) = (det )

(det )
k
.
Note is the rank of the module on which is dened and k is the rank of the module on
which is dened. In particular, in Example 2.2 we have det(AA

) = (det A)
2
(det A

)
2
.
Lets review the idea in this proof. Since N

= R

, M
R
N

= M
R
R

= M

. Under
such an isomorphism, id
N
becomes the -fold direct sum , which has a block
diagonal matrix representation in a suitable basis. So its determinant is (det )

.
Example 2.8. Taking M = N and = , the tensor square
2
has determinant (det )
2k
.
Corollary 2.9. Let M be a free module of rank k 1 and : M M be a linear map.
For every i 1, det(
i
) = (det )
ik
i1
.
Proof. Use induction and associativity of the tensor product of linear maps.
Lets see how the tensor product of linear maps behaves for isomorphisms, surjections,
and injections.
Theorem 2.10. If : M M

and : N N

are isomorphisms then is an

isomorphism.
Proof. The composite of with
1

1
in both orders is the identity.
Theorem 2.11. If : M M

and : N N

are surjective then is surjective.

6 KEITH CONRAD
Proof. Since is linear, to show it is onto it suces to show every elementary tensor in
M

R
N

is in the image. For such an elementary tensor m

, we can write m

= (m) and
n

= (n) since and are onto. Therefore m

= (m) (n) = ()(mn).

It is a fundamental feature of tensor products that if and are both injective then
might not be injective. This can occur even if one of or is the identity function.
Example 2.12. Taking R = Z, let : Z/pZ Z/p
2
Z be multiplication by p: (x) = px.
This is injective, and if we tensor with Z/pZ we get the linear map 1: Z/pZ
Z
Z/pZ
Z/pZ
Z
Z/p
2
Z with the eect a x a px = pa x = 0, so 1 is identically 0 and
its domain is Z/pZ
Z
Z/pZ

= Z/pZ = 0, so 1 is not injective.

This provides an example where the natural linear map
Hom
R
(M, M

)
R
Hom
R
(N, N

) Hom
R
(M
R
N, M

R
N

)
in Theorem 2.5 is not an isomorphism; R = Z, M = M

= N = Z/pZ, and N

= Z/p
2
Z.
Because the tensor product of linear maps does not generally preserve injectivity, a tensor
has to be understood in context: it is a tensor in a specic tensor product M
R
N. If
M M

and N N

, it is generally false that M

R
N can be thought of as a submodule
of M

R
N

since the natural map M

R
N M

R
N

might not be injective. We might

say it this way: a tensor product of submodules need not be a submodule.
Example 2.13. Again with R = Z, let i : pZ Z be inclusion. Tensoring i with Z/pZ gives
the linear map 1i : Z/pZ
Z
pZ Z/pZ
Z
Z with the eect apx apx = pax = 0,
so 1 i is identically 0. Since pZ

= Z, Z/pZ
Z
pZ

= Z/pZ
Z
Z

= Z/pZ by Theorem
2.10, so 1 i is not injective.
This example also shows the image of M
R
N

M

R
N

need not be isomorphic

to (M)
R
(N), since 1 i has image 0 and Z/pZ
Z
i(pZ)

= Z/pZ
Z
Z

= Z/pZ.
Example 2.14. Again with R = Z, let i : Z Q be the inclusion. Tensor this with Z/pZ
to get the Z-linear map 1 i : Z/pZ
Z
Z Z/pZ
Z
Q with domain isomorphic to Z/pZ
(not 0) and target 0, so 1 i is not injective.
Example 2.15. Here is an example of a linear map i : M N which is injective and its
tensor square i
2
: M
2
N
2
is not injective.
Let R = A[X, Y ] with A a nonzero commutative ring and I = (X, Y ). Let i : I R
be the inclusion map. We will show i
2
: I
2
R
2
is not injective, which means the
natural way to think of I
R
I inside R
R
R is not an embedding. In other words, for
polynomials f and g in I, you have to distinguish between the tensor f g in I
R
I and
the tensor f g in R
R
R.
The tensor XY in I
2
gets sent by i
2
to the tensor XY in R
2
, and in R
2
we can
write X Y = XY (1 1). (We cant do that in I
2
, as 1 is not in I.) Similarly, i
2
sends
Y X in I
2
to Y X = Y X(1 1) in R
2
. Since XY = Y X, i
2
(X Y ) = i
2
(Y X).
We will now show that in I
2
the dierence X Y Y X is not zero, so the kernel of
i
2
is not 0.
Letting the partial derivatives of a polynomial f(X, Y ) with respect to X and Y be
denoted f
X
and f
Y
, the function I I A given by (f, g) f
X
(0, 0)g
Y
(0, 0) is R-bilinear
(where A is an R-module through multiplication by the constant term of polynomials in R,
or just view A as R/I), so there is an R-linear map I
2
A sending fg to f
X
(0, 0)g
Y
(0, 0).
In particular, X Y 1 and Y X 0, so X Y = Y X in I
2
.
TENSOR PRODUCTS, II 7
The tensor product I
2
in Example 2.15 exhibits another interesting feature when A is
a domain: I is torsion-free but I
2
is not: XY (X Y ) = XY XY = XY (Y X), so
XY (XY Y X) = 0 and XY Y X = 0. Therefore a tensor product of torsion-free
modules (even over a domain) need not be torsion-free!
Generalizing Example 2.15, let R = A[X
1
, . . . , X
n
] where n 2 and I = (X
1
, . . . , X
n
).
The inclusion i : I R is injective but the nth tensor power i
n
: I
n
R
n
is not
injective because the tensor

Sn
(sign )X
(1)
X
(n)
I
n
gets sent to

Sn
(sign )X
1
X
n
(1 1) in R
n
, which is 0, but the original sum is
not 0 because we can construct a linear map I
n
A sending it to 1 (using a product of
partial derivatives at (0, 0, . . . , 0), as in the n = 2 case above).
While we have just seen a tensor power of an injective linear map need not be injective,
here is a condition where injectivity holds.
Theorem 2.16. Let : M N be injective and (M) be a direct summand of N. For
k 0,
k
: M
k
N
k
is injective and the image is a direct summand of N
k
.
Proof. Write N = (M) P. Let : N M by ((m) + p) = m, so is linear and
= id
M
. Then
k
: M
k
N
k
and
k
: N
k
M
k
are linear maps and

k

k
= ( )
k
= id
k
M
= id
M
k,
so
k
has a left inverse. That implies
k
is injective and M
k
is isomorphic to a direct
summand of N
k
by criteria for when a short exact sequence of modules splits.
We can apply this to vector spaces: if V is a vector space and W is a subspace, there
is a direct sum decomposition V = W U (U is non-canonical), so tensor powers of the
inclusion W V are injective linear maps W
k
V
k
.
Other criteria for a tensor power of an injective linear map to be injective will be met in
Corollary 3.9 and Theorem 4.8.
We will now compute the kernel of M
R
N

M

R
N

in terms of the kernels of

and , assuming and are onto.
Theorem 2.17. Let M

M

and N

N

be R-linear and surjective. The kernel

of M
R
N

M

R
N

is the submodule of M
R
N spanned by all m n where
(m) = 0 or (n) = 0. In terms of the inclusion maps ker
i
M and ker
j
N,
ker( ) = (i 1)((ker )
R
N) + (1 j)(M
R
(ker )).
The reason we dont write the kernel of as (ker )
R
N + M
R
(ker ), even
though that is the intuitive idea, is that strictly speaking these terms are their own tensor
product modules and only the application of i 1 and 1 j which might not be injective
puts them inside M
R
N.
Proof. Both (i 1)((ker )
R
N) and (1j)(M(ker )) are killed by : if m ker
and n N then ( )((i 1)(m n)) = ( )(m n)
1
= (m) (n) = 0 since
(m) = 0. Similarly (1 j)(mn) is killed by if m M and n ker . Set
U = (i 1)((ker )
R
N) + (1 j)(M (ker )),
1
This is mn in M R N, whereas the previous mn is in (ker ) R N.
8 KEITH CONRAD
so U ker( ), which means induces a linear map
: (M
R
N)/U M

R
N

where (m n mod U) = ( )(m n) = (m) (n). We will now write down an

inverse map, which proves is injective, so the kernel of is U.
Because and are assumed to be onto, every elementary tensor in M

R
N

has the
form (m) (n). Knowing (m) and (n) only determines m and n up to addition by
elements of ker and ker . For m

ker and n

ker ,
(m+m

) (n +n

) = mn +m

n +mn

+m

mn +U,
so the function M

(M
R
N)/U dened by ((m), (n)) mn mod U is well-
dened. It is R-bilinear, so we have an R-linear map : M

R
N

(M
R
N)/U where
((m) (n)) = mn mod U on elementary tensors.
Easily the linear maps and x spanning sets, so they are both the identity.
Remark 2.18. If we remove the assumption that and are onto, Theorem 2.17 does not
correctly compute the kernel. For example, if and are both injective then the formula
for the kernel in Theorem 2.17 is 0, and we know need not be injective.
Unlike the kernel computation in Theorem 2.17, it is not easy to describe the torsion
submodule of a tensor product in terms of the torsion submodules of the original modules.
While (M
R
N)
tor
contains (i1)(M
tor

R
N)+(1j)(M
R
N
tor
), with i : M
tor
M and
j : N
tor
N being the inclusions, it is not true that this is all of (M
R
N)
tor
, since M
R
N
can have nonzero torsion when M and N are torsion-free (so M
tor
= 0 and N
tor
= 0). We
saw this at the end of Example 2.15.
Corollary 2.19. Let f : R S be a homomorphism of commutative rings and M N as
R-modules, with M
i
N the inclusion map. The following are equivalent:
(1) S
R
M
1i
S
R
N is onto.
(2) S
R
(N/M) = 0.
Proof. Let N

N/M be the reduction map, so we have the sequence S
R
M
1i

S
R
N
1
S
R
(N/M). The map 1 is onto, and ker = M, so ker(1 ) =
(1 i)(S
R
M). Therefore 1 i is onto if and only if ker(1 ) = S
R
N if and only if
1 = 0, and since 1 is onto we have 1 = 0 if and only if S
R
(N/M) = 0.
Example 2.20. If M N and N is nitely generated, we show M = N if and only if the
natural map R/m
R
M
1i
R/m
R
N is onto for all maximal ideals m in R, where
M
i
N is the inclusion map. The only if direction is clear. In the other direction, if
R/m
R
M
1i
R/m
R
N is onto then R/m
R
(N/M) = 0 by Corollary 2.19. Since N
is nitely generated, so is N/M, and we are reduced to showing R/m
R
(N/M) = 0 for all
maximal ideals m if and only if N/M = 0. When P is a nitely generated module, P = 0
if and only if P/mP = 0 for all maximal ideals m in R, so we can apply this to P = N/M
since P/mP

= R/m
R
P.
Corollary 2.21. Let f : R S be a homomorphism of commutative rings and I be an
ideal in R[X
1
, . . . , X
n
]. Write I S[X
1
, . . . , X
n
] for the ideal generated by the image of I in
S[X
1
, . . . , X
n
]. Then
S
R
R[X
1
, . . . , X
n
]/I

= S[X
1
, . . . , X
n
]/(I S[X
1
, . . . , X
n
]).
TENSOR PRODUCTS, II 9
as S-modules by s h mod I sh mod I S[X
1
, . . . , X
n
].
Proof. The identity S S and the natural reduction R[X
1
, . . . , X
n
] R[X
1
, . . . , X
n
]/I
are both onto, so the tensor product of these R-linear maps is an R-linear surjection
(2.1) S
R
R[X
1
, . . . , X
n
] S
R
(R[X
1
, . . . , X
n
]/I)
and the kernel is (1 j)(S
R
I) by Theorem 2.17, where j : I R[X
1
, . . . , X
n
] is the
inclusion. Under the natural R-module isomorphism
(2.2) S
R
R[X
1
, . . . , X
n
]

= S[X
1
, . . . , X
n
],
(1 j)(S
R
I) on the left side corresponds to I S[X
1
, . . . , X
n
] on the right side, so (2.1)
and (2.2) say
S[X
1
, . . . , X
n
]/(I S[X
1
, . . . , X
n
])

= S
R
(R[X
1
, . . . , X
n
]/I).
as R-modules. The left side is naturally an S-module and the right side is too using extension
of scalars. It is left to the reader to check the isomorphism is S-linear.
Example 2.22. For h(X) Z[X], Q
Z
Z[X]/(h(X))

= Q[X]/(h(X)) as Q-vector spaces

and Z/mZ
Z
Z[X]/(h(X)) = (Z/mZ)[X]/(h(X)) as Z/mZ-modules where m > 1.
3. Flat Modules
Because a tensor product of injective linear maps might not be injective, it is important
to give a name to those R-modules N which always preserve injectivity, in the sense that
M

M

being injective implies N

R
M
1
N
R
M

is injective. (Notice the map

on N is the identity.)
Denition 3.1. An R-module N is called at if for all injective linear maps M

M

the linear map N

R
M
1
N
R
M

is injective.
Example 3.2. For a nonzero torsion abelian group A, the natural map Z Q is injective
but if we tensor with A we get the map A 0, which is not injective, so A is not a at
Z-module.
The concept of a at module is pointless unless one has some good examples. The next
two theorems provide some.
Theorem 3.3. Any free R-module F is at: if the linear map : M M

is injective,
then 1 : F
R
M

F
R
M

is injective.
Proof. When F = 0 it is clear, so take F = 0 with basis {e
i
}
iI
. From our previous
development of the tensor product, every element of F
R
M can be written as

i
e
i
m
i
for a unique choice of m
i
M, and similarly for F
R
M

.
For t ker(1 ), we can write t =

i
e
i
m
i
with m
i
M. Then
0 = (1 )(t) =

i
e
i
(m
i
),
in F
R
M

, which forces each (m

i
) to be 0. So every m
i
is 0, since is injective, and we
get t =

i
e
i
0 = 0.
Note that in Theorem 3.3 we did not need to assume F has a nite basis.
Theorem 3.4. Let R be a domain and K be its fraction eld. As an R-module, K is at.
10 KEITH CONRAD
Proof. Let M

M

be an injective linear map of R-modules. Every tensor in K

R
M is
elementary (use common denominators in K) and an elementary tensor in K
R
M is 0 if
and only if its rst factor is 0 or its second factor is torsion. (Here we are using properties
of K
R
M proved in part I.)
Supposing (1 )(t) = 0, we may write t = x m, so 0 = (1 )(t) = x (m).
Therefore x = 0 in K or (m) M

tor
. If (m) M

tor
then r(m) = 0 for some nonzero
r R, so (rm) = 0, so rm = 0 in M ( is injective), which means m M
tor
. Thus x = 0
or m M
tor
, so t = x m = 0.
If M is a submodule of the R-module M

then Theorem 3.4 says we can consider K

R
M
as a subspace of K
R
M

since the natural map K

R
M K
R
M

is injective. (See
diagram below.) Notice this works even if M or M

has torsion; although the natural maps

M K
R
M and M

R
K
R
M

might not be injective, the map K

R
M K
R
M

is injective.
M

K
R
M


1

K
R
M

Example 3.5. The natural inclusion Z Z/3Z Z is Z-linear and injective. Tensoring
with Q and using properties of tensor products turns this into the identity map Q Q,
which is also injective.
Remark 3.6. Theorem 3.4 generalizes: for any commutative ring R and multiplicative set
D in R, the localization R
D
is a at R-module.
Theorem 3.7. A tensor product of two at modules is at.
Proof. Let N and N

be at. For any injective linear map M

M

, we want to show the

induced linear map (N
R
N

)
R
M
1
(N
R
N

) M

is injective.
Since N

is at, N

R
M
1
N

R
M

is injective. Tensoring now with N, N

R
(N

R
M)
1(1)
N
R
(N

R
M

) is injective since N is at. The diagram

N
R
(N

R
M)

 1(1)

N
R
(N

(N
R
N

)
R
M
1

(N
R
N

) M

commutes, where the vertical maps are the natural isomorphisms, so the bottom map is
injective. Thus N
R
N

is at.
Theorem 3.8. Let M

M

and N

N

be injective linear maps. If the four modules

are all at then M
R
N

M

R
N

is injective. More precisely, if M and N

are
at or M

and N are at then is injective.

TENSOR PRODUCTS, II 11
Proof. The trick is to break up the linear map into a composite of linear maps 1
and 1 in the following commutative diagram.
M
R
N

N
N
N
N
N
N
N
N
N
N
N
M
R
N
1

q
q
q
q
q
q
q
q
q
q
q

R
N

Both 1 and 1 are injective since N

and M are at, so their composite

is injective. Alternatively, we can write as a composite tting in the commutative
diagram
M

R
N
1

N
N
N
N
N
N
N
N
N
N
N
M
R
N
1

q
q
q
q
q
q
q
q
q
q
q

R
N

and the two diagonal maps are injective from atness of N and M

, so is injective.
Corollary 3.9. Let M
1
, . . . , M
k
, N
1
, . . . , N
k
be at R-modules and
i
: M
i
N
i
be injective
linear maps. Then the linear map

1

k
: M
1

R

R
M
k
N
1
N
k
is injective. In particular, if : M N is an injective linear map of at modules then the
tensor powers
k
: M
k
N
k
are injective for all k 1.
Proof. We argue by induction on k. For k = 1 there is nothing to show. Suppose k 2 and

1

k1
is injective. Then break up
1

k
into the composite
(N
1

R

R
N
k1
)
R
M
k
1
k

V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
V
M
1

R

R
M
k1

R
M
k
(
1

k1
)1

g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g
g

N
1

R

R
N
k
.
The rst diagonal map is injective because M
k
is at, and the second diagonal map is
injective because N
1

R

R
N
k1
is at (Theorem 3.7 and induction).
Corollary 3.10. If M and N are free R-modules and : M N is an injective linear
map, any tensor power
k
: M
k
N
k
is injective.
Proof. Free modules are at by Theorem 3.3.
Note the free modules in Corollary 3.10 are completely arbitrary. We make no assump-
tions about nite bases.
Corollary 3.10 is not a special case of Theorem 2.16 because a free submodule of a free
module need not be a direct summand (consider the ring Z and any proper subgroup of
rank n in Z
n
as a Z-module inside of Z
n
).
Corollary 3.11. If M is a free module and {m
1
, . . . , m
s
} is a nite linearly independent
subset then for any k s the s
k
elementary tensors
(3.1) m
i
1
m
i
k
where i
1
, . . . , i
k
{1, 2, . . . , s}
are linearly independent in M
k
.
12 KEITH CONRAD
Proof. There is an embedding R
s
M by

s
i=1
r
i
e
i


s
i=1
r
i
m
i
. Since R
s
and M are
free, the kth tensor power (R
s
)
k
M
k
is injective. This map sends the basis
e
i
1
e
i
k
of (R
s
)
k
, where i
1
, . . . , i
k
{1, 2, . . . , s}, to the elementary tensors in (3.1), so they are
linearly independent in M
k

Corollary 3.11 is not saying the elementary tensors in (3.1) can be extended to a basis of

k
(M), any more than m
1
, . . . , m
s
can be extended to a basis of M.
4. Tensor Products of Linear Maps and Base Extension
Fix a ring homomorphism R
f
S. Every S-module becomes an R-module by restriction
of scalars, and every R-module M has a base extension S
R
M, which is an S-module. In
part I we saw S
R
M has a universal mapping property among all S-modules: an R-linear
map from M to any S-module extends uniquely to an S-linear map from S
R
M to the
S-module. We discuss in this section an arguably more important role for base extension:
it turns an R-linear map M

M

between two R-modules into an S-linear map between

S-modules. Tensoring M

M

with S gives us an R-linear map S

R
M
1
S
R
M

which is in fact S-linear: (1 )(st) = s(1 )(t) for all s S and t S

R
M. Since
both sides are additive in t, to prove this it suces to consider the case when t = s

m is
an elementary tensor. Then
(1 )(s(s

m)) = (1 )(ss

m)) = ss

(m) = s(s

(m)) = s(1 )(s

m).
We will write the base extended linear map 1 as
S
to make the S-dependence clearer,
so

S
: S
R
M S
R
M

by
S
(s m) = s (m).
Since 1 id
M
= id
S
R
M
and (1 ) (1

) = 1 (

), we have (id
M
)
S
= id
S
R
M
and (

)
S
=
S

S
. That means the process of creating S-modules and S-linear maps
out of R-modules and R-linear maps is functorial.
If an R-linear map M

M

is an isomorphism or is surjective then so is S

R
M

S

S
R
M

(Theorems 2.10 and 2.11). But if is injective then

S
need not be injective.
(Examples 2.12, 2.13, and 2.14, which all have S as a eld).
Theorem 4.1. Let R be a nonzero commutative ring. If R
m
= R
n
as R-modules then
m = n. If there is a linear surjection R
m
R
n
then m n.
Proof. Pick a maximal ideal m in R. Tensoring R-linear maps R
m
= R
n
or R
m
R
n
with R/m produces R/m-linear maps (R/m)
m
= (R/m)
n
or (R/m)
m
(R/m)
n
. Taking
dimensions over the eld R/m implies m = n or m n, respectively.
We cant extend this method of proof to show a linear injection R
m
R
n
forces m n
because injectivity is not generally preserved under base extension. We will return to this
later when we meet exterior powers.
Theorem 4.2. Let R be a PID and M be a nitely generated R-module. Writing
M

= R
d
R/(a
1
) R/(a
k
),
where a
1
| |a
k
, the integer d equals dim
K
(K
R
M), where K is the fraction eld of R.
Therefore d is uniquely determined by M.
TENSOR PRODUCTS, II 13
Proof. Tensoring the displayed R-module isomorphism by K gives a K-vector space iso-
morphism K
R
M

= K
d
since K
R
(R/(a
i
)) = 0. Thus d = dim
K
(K
R
M).
Example 4.3. Let A = (
a b
c d
) M
2
(R). Regarding A as a linear map R
2
R
2
, its base
extension A
S
: S
R
R
2
S
R
R
2
is S-linear and S
R
R
2
= S
2
as S-modules.
Let {e
1
, e
2
} be the standard basis of R
2
. An S-basis for S
R
R
2
is {1e
1
, 1e
2
}. Using
this basis, we can compute a matrix for A
S
:
A
S
(1 e
1
) = 1 A(e
1
) = 1 (ae
1
+ce
2
) = a(1 e
1
) +c(1 e
2
)
and
A
S
(1 e
2
) = 1 A(e
2
) = 1 (be
1
+de
2
) = b(1 e
1
) +d(1 e
2
).
Therefore the matrix for A
S
is (
a b
c d
) M
2
(S). (Strictly speaking, we should have entries
f(a), f(b), and so on.)
The next theorem says base extension doesnt change matrix representations, as in the
previous example.
Theorem 4.4. Let M and M

be nonzero nite-free R-modules and M

M

be an
R-linear map. For any bases {e
j
} and {e

i
} of M and M

, the matrix for the S-linear map

S
R
M

S
S
R
M

with respect to the bases {1 e

j
} and {1 e

i
} equals the matrix for
with respect to {e
j
} and {e

i
}.
Proof. Say (e
j
) =

i
a
ij
e

i
, so the matrix of is (a
ij
). Then

S
(1 e
j
) = 1 (e
j
) = 1

i
a
ij
e
i
=

i
a
ij
(1 e
i
),
so the matrix of
S
is also (a
ij
).
Example 4.5. Any n n real matrix acts on R
n
, and its base extension to C acts on
C
R
R
n
= C
n
as the same matrix. An n n integral matrix acts on Z
n
and its base
extension to Z/mZ acts on Z/mZ
Z
Z
n
= (Z/mZ)
n
as the same matrix reduced mod m.
Theorem 4.6. Let M and M

be R-modules. There is a unique S-linear map

S
R
Hom
R
(M, M

) Hom
S
(S
R
M, S
R
M

)
sending s to the function s
S
: t s
S
(t) and it is an isomorphism if M and M

are
nite free. In particular, there is a unique S-linear map
S
R
(M

R
) (S
R
M)

S
where s s
S
, and it is an isomorphism if M is nite-free.
The point of this theorem in the nite-free case is that it says base extension on linear
maps accounts (through S-linear combinations) for all S-linear maps between base extended
R-modules. This doesnt mean every S-linear map is a base extension, which would be like
saying every tensor is an elementary tensor rather than just a sum of them.
Proof. The function S Hom
R
(M, M

) Hom
S
(S
R
M, S
R
M

) where (s, ) s
S
is
R-bilinear (check!), so there is a unique R-linear map
S
R
Hom
R
(M, M

)
L
Hom
S
(S
R
M, S
R
M

)
14 KEITH CONRAD
such that L(s) = s
S
. The map L is S-linear (check!). If M

= R and we identify S
R
R
with S as S-modules by multiplication, then L becomes an S-linear map S
R
(M

R
)
(S
R
M)

S
.
Now suppose M and M

are both nite free. We want to show L is an isomorphism. If

M or M

is 0 it is clear, so we may take them both to be nonzero with respective R-bases

{e
i
} and {e

j
}, say. Then S-bases of S
R
M and S
R
M

are {1 e
i
} and {1 e

j
}. An
R-basis of Hom
R
(M, M

) is the functions
ij
sending e
i
to e

j
and other basis vectors e
k
of
M to 0. An S-basis of S
R
Hom
R
(M, M

) is the tensors {1
ij
}, and
L(1
ij
)(1 e
i
) = (
ij
)
S
(1 e
i
) = (1
ij
)(1 e
i
) = 1
ij
(e
i
) = 1 e

j
while L(1
ij
)(1 e
k
) = 0 for k = i. That means L sends a basis to a basis, so L is an
isomorphism.
Example 4.7. Let R = Z/p
2
Z and S = Z/pZ. Take M = Z/pZ as an R-module. The
linear map
S
R
M

R
(S
R
M)

S
has image 0, but the right side is isomorphic to Hom
Z/pZ
(Z/pZ, Z/pZ)

= Z/pZ, so this
map is not an isomorphism.
In Corollary 3.10 we saw any tensor power of an injective linear map between free modules
is injective. Using base extension, we can drop the requirement that the target module be
free provided we are working over a domain.
Theorem 4.8. Let R be a domain and : M N be an injective linear map where M is
free. Then
k
: M
k
N
k
is injective for any k 1.
Proof. We have a commutative diagram
M
k

N
k

K
R
M
k

1(
k
)

K
R
N
k

(K
R
M)
k
(1)
k

(K
R
N)
k
where the top vertical maps are the natural ones (t 1 t) and the bottom vertical maps
are the base extension isomorphisms. (Tensor powers along the bottom are over K while
those on the rst and second rows are over R.) From commutativity, to show
k
along
the top is injective it suces to show the composite map along the left side and the bottom
is injective. The K-linear map map K
R
M
1
K
R
N is injective since K is a
at R-module, and therefore the map along the bottom is injective (tensor products of
injective linear maps of vector spaces are injective). The bottom vertical map on the left
is an isomorphism. The top vertical map on the left is injective since M
k
is free and thus
torsion-free (R is a domain).
TENSOR PRODUCTS, II 15
5. Vector Spaces
Because all (nonzero) vector spaces have bases, the results we have discussed for modules
assume a simpler form when we are working with vector spaces. We will review what we
have done in the setting of vector spaces and then discuss some further special properties
of this case.
Let K be a eld. Tensor products of K-vector spaces involve no unexpected collapsing: if
V and W are nonzero K-vector spaces then V
K
W is nonzero and in fact dim
K
(V
K
W) =
dim
K
(V ) dim
K
(W) in the sense of cardinal numbers.
For any K-linear maps V

V

and W

W

, we have the tensor product linear

map V
K
W

V

K
W

which sends v w to (v) (w). When V

V

and
W

W

are isomorphisms or surjective, so is V

K
W

V

K
W

(Theorems
2.10 and 2.11). Moreover, because all K-vector spaces are free a tensor product of injective
K-linear maps is injective (Theorem 3.3).
Example 5.1. If V

W is an injective K-linear map and U is any K-vector space, the
K-linear map U
K
V
1
U
K
W is injective.
Example 5.2. A tensor product of subspaces is a subspace: if V V

and W W

the
natural linear map V
K
W V

K
W

is injective.
Because of this last example, we can treat a tensor product of subspaces as a subspace
of the tensor product. For example, if V

V

and W

W

are linear then (V ) V

and (W) W

, so we can regard (V )
K
(W) as a subspace of V

K
W

, which we
couldnt do with modules in general. The following result gives us some practice with this
viewpoint.
Theorem 5.3. Let V V

and W W

where V

and W

are nonzero. Then V

K
W =
V

K
W

if and only if V = V

and W = W

.
Proof. Since V
K
W is inside both V
K
W

and V

K
W, which are inside V

K
W

,
by reasons of symmetry it suces to assume V V

and show V
K
W

K
W

.
Since V is a proper subspace of V

, there is a linear functional : V

K which vanishes
on V and is not identically 0 on V

, so (v

0
) = 1 for some v

0
V

. Pick nonzero W

,
and say (w

0
) = 1. Then the linear function V

K
W

K where v

(v

)(w

)
vanishes on all of V
K
W

by checking on elementary tensors but its value on v

0
w

0
is
1. Therefore v

0
w

0
V
K
W

, so V
K
W

K
W

.
When V and W are nite-dimensional, the K-linear map
(5.1) Hom
K
(V, V

)
K
Hom
K
(W, W

) Hom
K
(V
K
W, V

K
W

)
sending the elementary tensor to the linear map denoted is an isomorphism
(Theorem 2.5). So the two possible meanings of (elementary tensor in a tensor product
of Hom-spaces or linear map on a tensor product of vector spaces) really match up. Taking
V

= K and W

= K in (5.1) and identifying K

K
K with K by multiplication, (5.1) says
V

K
W

= (V
K
W)

using the obvious way of making a tensor in V

K
W

act on V
K
W, namely through multiplication of the values: ( )(v w) = (v)(w).
By induction on the numbers of terms,
V

1

K

K
V

k

= (V
1

K

K
V
k
)

16 KEITH CONRAD
when the V
i
s are nite-dimensional. Here an elementary tensor
1

k

k
i=1
V

i
acts on an elementary tensor v
1
v
k

k
i=1
V
i
with value
1
(v
1
)
k
(v
k
) K. In
particular,
(V

)
k

= (V
k
)

when V is nite-dimensional.
Lets turn now to base extensions to larger elds. When L/K is any eld extension,
2
base
extension turns K-vector spaces into L-vector spaces (V L
K
V ) and K-linear maps
into L-linear maps (
L
:= 1 ). Provided V and W are nite-dimensional over K,
base extension of linear maps V W accounts for all the linear maps between L
K
V
and L
K
W using L-linear combinations, in the sense that the natural L-linear map
(5.2) L
K
Hom
K
(V, W)

= Hom
L
(L
K
V, L
K
W)
is an isomorphism (Theorem 4.6). When we choose K-bases for V and W and use the
corresponding L-bases for L
K
V and L
K
W, the matrix representations of a K-linear
map V W and its base extension by L are the same (Theorem 4.4). Taking W = K, the
natural L-linear map
(5.3) L
K
V

= (L
K
V )

is an isomorphism for nite-dimensional V , using K-duals on the left and L-duals on the
right.
3
Remark 5.4. We dont really need L to be a eld; K-vector spaces are free and therefore
their base extensions to modules over any commutative ring containing K will be free as
modules over the larger ring. For example, we could dene the characteristic polynomial
of a linear operator V

V in a coordinate-free way using base extension of V from
K to K[T]: the characteristic polynomial of is the determinant of the linear operator
T id
V

K[T]
: K[T]
K
V K[T]
K
V since det(T id
V

K[T]
) = det(TI
n
A),
where A is a matrix representation of .
We will make no nite-dimensionality assumptions in the rest of this section.
The next theorem tells us the image and kernel of a tensor product of linear maps of
vector spaces, with no surjectivity hypotheses as in Theorem 2.17.
Theorem 5.5. Let V
1

1
W
1
and V
2

2
W
2
be linear. Then
ker(
1

2
) = ker
1

K
V
2
+V
1

K
ker
2
, Im(
1

2
) =
1
(V
1
)
K

2
(V
2
).
In particular, if V
1
and V
2
are nonzero then
1

2
is injective if and only if
1
and
2
are injective, and if W
1
and W
2
are nonzero then
1

2
is surjective if and only if
1
and

2
are surjective.
Here we are taking advantage of the fact that in vector spaces a tensor product of sub-
spaces is naturally a subspace of the tensor product: ker
1

K
V
2
can be identied with
its image in V
1

K
V
2
and
1
(V
1
)
K

2
(V
2
) can be identied with its image in W
1

K
W
2
under the natural maps. Theorem 2.17 for modules has weaker conclusions (e.g., injectivity
of
1

2
doesnt imply injectivity of
1
and
2
).
2
We allow innite or even non-algebraic extensions, such as R/Q.
3
If we drop nite-dimensionality assumptions, (5.1), (5.2), and (5.3) are all still injective but generally
not surjective.
TENSOR PRODUCTS, II 17
Proof. First we handle the image of
1

2
. The diagrams

1
(V
1
)
i
1

G
G
G
G
G
G
G
G

2
(V
2
)
i
2

G
G
G
G
G
G
G
G
V
1
v
1

1
(v
1
)

x
x
x
x
x
x
x
x
x

W
1
V
2
v
2

2
(v
2
)

x
x
x
x
x
x
x
x
x

W
2
commute, with i
1
and i
2
being injections, so the composite diagram

1
(V
1
)
K

2
(V
2
)
i
1
i
2

R
R
R
R
R
R
R
R
R
R
R
R
R
V
1

K
V
2
v
1
v
2

1
(v
1
)
2
(v
2
)

m
m
m
m
m
m
m
m
m
m
m
m
m

W
1

K
W
2
commutes. As i
1
i
2
is injective, both maps out of V
1

K
V
2
have the same kernel. The
kernel of the map V
1

K
V
2

1
(V
1
)
K

2
(V
2
) can be computed by Theorem 2.17 to be
ker
1

K
V
2
+V
1

K
ker
2
, where we identify tensor products of subspaces with a subspace
of the tensor product.
If
1

2
is injective then its kernel is 0, so 0 = ker
1

K
V
2
+ V
1

K
ker
2
from the
kernel formula. Therefore the subspaces ker
1

K
V
2
and V
1

K
ker
2
both vanish, so
ker
1
and ker
2
must vanish (because V
2
and V
1
are nonzero, respectively). Conversely,
if
1
and
2
are injective then we already knew
1

2
is injective, but the formula for
ker(
1

2
) also shows us this kernel is 0.
If
1

2
is surjective then the formula for its image shows
1
(V
1
)
K

2
(V
2
) = W
1

K
W
2
,
so
1
(V
1
) = W
1
and
2
(V
2
) = W
2
by Theorem 5.3 (here we need W
1
and W
2
nonzero).
Conversely, if
1
and
2
are surjective then so is
1

2
because thats true for all modules.

Corollary 5.6. Let V V

and W W

. Then
(V

K
W

)/(V
K
W

+V

K
W)

= (V

/V )
K
(W

/W).
Proof. Tensor the natural projections V

1
V

/V and W

2
W

/W to get a linear
map V

K
W

2
(V

/V )
K
(W

/W) which is onto with ker(

1

2
) = V
K
W

+
V

K
W by Theorem 5.5.
Remark 5.7. It is false that (V

K
W

)/(V
K
W)

= (V

/V )
K
(W

/W). The subspace

V
K
W is generally too small
4
to be the kernel. This is a distinction between tensor
products and direct sums (where (V

)/(V W)

= (V

/V ) (W

/W)).
Corollary 5.8. Let V

W be a linear map and U be a K-vector space. The linear map
U
K
V
1
U
K
W has kernel and image
(5.4) ker(1 ) = U
K
ker Im(1 ) = U
K
(V ).
In particular, for nonzero U the map is injective or surjective if and only if 1 has
that property.
Proof. This is immediate from Theorem 5.5 since were using the identity map on U.
4
Exception: V

or W

is 0, or V = V

and W = W

.
18 KEITH CONRAD
Example 5.9. Let V

W be a linear map and L/K be a eld extension. The base
extension L
K
V

L
L
K
W has kernel and image
ker(
L
) = L
K
ker , Im(
L
) = L
K
Im().
The map is injective if and only if
L
is injective and is surjective if and only if
L
is
surjective.
Lets formulate this in the language of matrices. If V and W are nite-dimensional then
can be written as a matrix with entries in K once we pick bases of V and W. Then

L
has the same matrix representation relative to the corresponding bases of L
K
V and
L
K
W. Since the base extension of a free module to another ring doesnt change the size
of a basis, dim
L
(L
K
Im()) = dim
K
Im() and dim
L
(L
K
ker()) = dim
K
ker(). That
means and
L
have the same rank and the same nullity: the rank and nullity of a matrix
in M
mn
(K) do not change when it is viewed in M
mn
(L) for any eld extension L/K.
In the rest of this section we will look at tensor products of many vector spaces at once.
Lemma 5.10. For v V with v = 0, there is V

such that (v) = 1.

Proof. The set {v} is linearly independent, so it extends to a basis {v
i
}
iI
of V . Let v = v
i
0
in this indexing. Dene : V K by

i
c
i
v
i

= c
i
0
.
Then V

and (v) = (v
i
0
) = 1.
Theorem 5.11. Let V
1
, . . . , V
k
be K-vector spaces and v
i
V
i
. Then v
1
v
k
= 0 in
V
1

K

K
V
k
if and only if some v
i
is 0.
Proof. The direction () is clear. To prove (), we show the contrapositive: if every v
i
is nonzero then v
1
v
k
= 0. By Lemma 5.10, for i = 1, . . . , k there is
i
V

i
with

i
(v
i
) = 1. Then
1

k
is a linear map V
1

K

K
V
k
K having the eect
v
1
v
k

1
(v
1
)
k
(v
k
) = 1 = 0,
so v
1
v
k
= 0.
Corollary 5.12. Let
i
: V
i
W
i
be linear maps between K-vector spaces for 1 i k.
Then the linear map
1

k
: V
1

K

K
V
k
W
1

K

K
W
k
is O if and only
if some
i
is O.
Proof. For (), if some
i
is O then (
1

k
)(v
1
v
k
) =
1
(v
1
)
k
(v
k
) = 0
since
i
(v
i
) = 0. Therefore
1

k
vanishes on all elementary tensors, so it vanishes
on V
1

K

K
V
k
, so
1

k
= O.
To prove (), we show the contrapositive: if every
i
is nonzero then
1

k
= O.
Since
i
= O, we can nd some v
i
in V
i
with
i
(v
i
) = 0 in W
i
. Then
1

k
sends
v
1
v
k
to
1
(v
1
)
k
(v
k
). Since each
i
(v
i
) is nonzero in W
i
, the elementary tensor

1
(v
1
)
k
(v
k
) is nonzero in W
1

K

K
W
k
by Theorem 5.11. Thus
1

k
takes a nonzero value, so it is not the zero map.
Corollary 5.13. If R is a domain and M and N are R-modules, for non-torsion x in M
and y in N, x y is non-torsion in M
R
N.
TENSOR PRODUCTS, II 19
Proof. Let K be the fraction eld of R. The torsion elements of M
R
N are precisely the
elements that go to 0 under the map M
R
N K
R
(M
R
N) sending t to 1 t. We
want to show 1 (x y) = 0.
The natural K-vector space isomorphism K
R
(M
R
N)

= (K
R
M)
K
(K
R
N)
identies 1 (x y) with (1 x) (1 y). Since x and y are non-torsion in M and
N, 1 x = 0 in K
R
M and 1 y = 0 in K
R
N. An elementary tensor of nonzero
vectors in two K-vector spaces is nonzero (Theorem 5.11), so (1 x) (1 y) = 0 in
(K
R
M)
K
(K
R
N). Therefore 1 (x y) = 0 in K
R
(M
R
N), which is what we
wanted to show.
Remark 5.14. If M and N are torsion-free, Corollary 5.13 is not saying M
R
N is torsion-
free. It only says all (nonzero) elementary tensors have no torsion. There could be tensors
with torsion that are not elementary, as we saw at the end of Example 2.15.
In Theorem 5.11 we saw an elementary tensor in a tensor product of vector spaces is 0 only
under the obvious condition that one of the vectors appearing is 0. We now show two nonzero
elementary tensors in vector spaces are equal only under the obvious circumstances.
Theorem 5.15. Let V
1
, . . . , V
k
be K-vector spaces. Pick pairs of nonzero vectors v
i
, v

i
in
V
i
for i = 1, . . . , k. If v
1
v
k
= v

1
v

k
in V
1

K

K
V
k
then there are constants
c
1
, . . . , c
k
in K such that v
i
= c
i
v

i
and c
1
c
k
= 1.
Proof. If v
i
= c
i
v

i
for all i and c
1
c
k
= 1 then v
1
v
k
= c
1
v

1
c
k
v

k
=
(c
1
c
k
)v

1
v

k
= v

1
v

k
.
Now we want to go the other way. It is clear for k = 1, so we may take k 2.
By Theorem 5.11, v
1
v
k
is not 0. Pick
i
V

i
for 1 i k1 such that
i
(v
i
) = 1.
For any V

k
, let h

=
1

k1
, so h

(v
1
v
k1
v
k
) = (v
k
). Also
h

(v
1
v
k1
v
k
) = h

(v

1
v

k1
v

k
) =
1
(v

1
)
k1
(v

k1
)(v

k
) = (c
k
v

k
),
where c
k
=
1
(v

1
)
k1
(v

k1
) K. So we have
(v
k
) = (c
k
v

k
)
for all V

k
. Therefore (v
k
c
k
v

k
) = 0 for all V

k
, so v
k
c
k
v

k
= 0, which says
v
k
= c
k
v

k
.
In the same way, for every i = 1, 2, . . . , k there is c
i
in K such that v
i
= c
i
v

i
. Then
v
1
v
k
= c
1
v

1
c
k
v

k
= (c
1
c
k
)(v

1
v

k
). Since v
1
v
k
= v

1
v

k
= 0,
we get c
1
c
k
= 1.
Here is the analogue of Theorem 5.15 for linear maps (compare to Corollary 5.12).
Theorem 5.16. Let
i
: V
i
W
i
and

i
: V
i
W
i
be nonzero linear maps between K-
vector spaces for 1 i k. Then
1

k
=

k
as linear maps V
1

K

K
V
k
W
1

K

K
W
k
if and only if there are c
1
, . . . , c
k
in K such that
i
= c
i

i
and c
1
c
2
c
k
= 1.
Proof. Since each
i
: V
i
W
i
is not identically 0, for i = 1, . . . , k 1 there is v
i
V
i
such
that
i
(v
i
) = 0 in W
i
. Then there is f
i
W

i
such that f
i
(
i
(v
i
)) = 1.
Pick any v V
k
and f W

k
. Set h
f
= f
1
f
k1
f (W
1

K

K
W
k
)

where
h
f
(w
1
w
k1
w
k
) = f
1
(w
1
) f
k1
(w
k1
)f(w
k
).
Then
h
f
((
1

k1

k
)(v
1
v
k1
v)) = f
1
(
1
(v
1
)) f
k1
(
k1
(v
k1
))f(
k
(v)),
20 KEITH CONRAD
and since f
i
(
i
(v
i
)) = 1 for i = k, the value is f(
k
(v)). Also
h
f
((

k1

k
)(v
1
v
k1
v)) = f
1
(

1
(v
1
)) f
k1
(

k1
(v
k1
))f(

k
(v)).
Set c
k
= f
1
(

1
(v
1
)) f
k1
(

k1
(v
k1
)), so the value is c
k
f(

k
(v)) = f(c
k

k
(v)). Since

1

k1

k
=

k1

k
,
f(
k
(v)) = f(c
k

k
(v)).
This holds for all f W

k
, so
k
(v) = c
k

k
(v). This holds for all v V
k
, so
k
= c
k

k
as
linear maps V
k
W
k
.
In a similar way, there is c
i
K such that
i
= c
i

i
for all i, so

1

k
= (c
1

1
) (c
k

k
)
= (c
1
c
k
)

k
= (c
1
c
k
)
1

k
,
so c
1
c
k
= 1 since
1

k
= O.
Remark 5.17. When the V
i
s and W
i
s are nite-dimensional, the tensor product of linear
maps between them can be identied with elementary tensors in the tensor product of
the vector spaces of linear maps (Theorem 2.5), so in this special case Theorem 5.16 is a
special case of Theorem 5.15. Theorem 5.16 does not assume the vector spaces are nite-
dimensional.
When we have k copies of a vector space V , any permutation S
k
acts on the direct
sum V
k
by permuting the coordinates:
(v
1
, , v
k
) (v
(1)
, , v
(k)
).
There is a similar action of S
k
on the kth tensor power.
Corollary 5.18. For S
k
, there is a linear map P

: V
k
V
k
such that
v
1
v
k
v
(1)
v
(k)
on elementary tensors. Then P

= P

for and in S
k
.
When dim
K
(V ) > 1, P

= P

if and only if = . In particular, P

is the identity map

if and only if is the identity permutation.
Proof. The function V V V
k
given by
(v
1
, . . . , v
k
) v
(1)
v
(k)
is multilinear, so the universal mapping property of tensor products gives us a linear map
P

with the indicated eect on elementary tensors. It is clear that P

1
is the identity map.
For any and in S
k
, P

= P

by checking equality of both sides at all elementary

tensors in V
k
. Therefore the injectivity of P

is reduced to showing if P

is the
identity map on V
k
then is the identity permutation.
We prove the contrapositive. Suppose is not the identity permutation, so (i) = j = i
for some i and j. Choose v
1
, . . . , v
k
V all nonzero such that v
i
and v
j
are not on the
same line. (Here we use dim
K
V > 1.) If P

(v
1
v
k
) = v
1
v
k
then v
j
Kv
i
by
Theorem 5.15, which is not so.
TENSOR PRODUCTS, II 21
The linear maps P

provide an action of S
k
on V
k
by linear transformations. We usually
write (t) for P

(t). Not only does S

k
acts on V
k
but also the group GL(V ) acts on V
k
via tensor powers of linear maps:
g(v
1
v
k
) := g
k
(v
1
v
k
) = gv
1
gv
k
on elementary tensors. These actions of the groups S
k
and GL(V ) on V
k
commute with
each other, as a computation on elementary tensors shows.
Tensors t V
k
satisfying (t) = t for all S
k
are called symmetric and tensors
t V
k
satisfying (t) = (sign )t for all S
k
are called skew-symmetric or anti-
symmetric. Both kinds of tensors occur in physics. Well see some examples in Section
6. The symmetric tensors and skew-symmetric tensors each form subspaces of V
k
. If K
does not have characteristic 2, every tensor in V
2
is a unique sum of a symmetric and
skew-symmetric tensor:
t =
t +(t)
2
+
t (t)
2
.
where is the non-identity permutation in S
2
. (It is the ip automorphism of V
2
sending
v w to w v.) For k > 2, the symmetric and anti-symmetric tensors in V
k
do not span
the whole space. There are additional subspaces of tensors in V
k
which are needed to ll
out the whole space, and these subspaces are connected to the representation theory of the
group GL(V ). The appearance of representations of GL(V ) inside tensor powers of V is an
important role for tensor powers in algebra.
6. Tensors in Physics
The name tensor was introduced by the mathematical physicist Woldemar Voigt in the
late 19th century in the context of measuring stress and strain. In the table below are
examples of tensors in dierent areas of physics.
Area Name of Tensor Symmetry
Mechanics Stress Symmetric
Strain Symmetric
Elasticity Symmetric
Moment of Inertia Symmetric
Electromagnetism Electromagnetic Anti-symmetric
Polarization Symmetric
Relativity Metric Symmetric
Stress-Energy Symmetric
What is it about tensors that makes them show up in physics? Lets rst understand why
vectors show up in physics. The physical meaning of a vector is not just displacement, but
linear displacement. For instance, forces at a point combine in the same way that vectors
add (this is an experimental observation), so force is treated as a vector. While the physical
meaning of vector is a linear displacement, the physical meaning of a tensor is multilinear
displacement.
Here is a denition of a tensor that can be found (more or less) in physics textbooks.
For k 1, a rank k tensor on an n-dimensional vector space
5
V is a choice, for each basis
5
The physicist may write R
3
or R
n
in place of V .
22 KEITH CONRAD
B = {e
1
, . . . , e
n
}, of a collection of n
k
numbers {T
i
1
,...,i
k
B
}
1i
1
,...,i
k
n
such that for two bases
B = {e
1
, . . . , e
n
} and B

= {e

1
, . . . , e

n
} of V with change-of-basis matrix A = (a
ij
) from B
to B

(so e
j
=

n
i=1
a
ij
e

i
) the two systems of n
k
numbers are related by the formula
(6.1) T
i
1
,...,i
k
B

1j
1
,...,j
k
n
T
j
1
,...,j
k
B
a
i
1
j
1
a
i
k
j
k
.
Do you see what a rank k tensor on V means in mathematical language? What is
being dened here is a member of the kth tensor power V
k
. Place the array of numbers
{T
i
1
,...,i
k
B
}
1i
1
,...,i
k
n
associated to a particular basis B = {e
1
, . . . , e
n
} of V as the coecients
for the associated basis {e
i
1
e
i
k
}
1i
1
,...,i
k
n
of V
k
:
t :=

1i
1
,...,i
k
n
T
i
1
,...,i
k
B
e
i
1
e
i
k
.
Lets write t in terms of the basis B

= {e

1
, . . . , e

n
} for V : when e
j
=

n
i=1
a
ij
e

i
,
t =

1j
1
,...,j
k
n
T
j
1
,...,j
k
B
e
j
1
e
j
k
=

1j
1
,...,j
k
n
T
j
1
,...,j
k
B

i
1
=1
a
i
1
j
1
e

i
1

i
k
=1
a
i
1
kj
k
e

i
k

1i
1
,...,i
k
n

1j
1
,...,j
k
n
T
j
1
,...,j
k
B
a
i
1
j
1
a
i
k
j
k

i
1
e

i
k
.
If we label the coecients of t with respect to the basis {e

i
1
e

i
k
}
1i
1
,...,i
k
n
as T
i
1
,...,i
k
B

then we recover the formula (6.1). So the physicists rank k tensor is just all the dierent
coordinate representations of a single element of V
k
.
Physics textbooks say tensors are indexed quantities with components that transform
in a denite way
6
, but this is like dening a group to be a set equipped with a denite law
of composition and then forgetting to add that the group law is supposed to satisfy some
actual properties (not every set with a law of composition is a group). The key structure in
the rule (6.1) is its multilinearity via the products a
i
1
j
1
a
i
k
j
k
. I have never seen a physics
textbook whose discussion of tensors uses the word multilinear. The transformation rule
(6.1) under a change of coordinates is not simply a denite rule but rather is a rule which
depends multilinearly on the change of coordinate data. Tensors are multilinear functions
of several directions.
7
Lets compare how the mathematician and physicist think about a tensor:
(Mathematician) A tensor belongs to a tensor space, which is a vector space dened
by a multilinear universal mapping property. This is a coordinate-free point of view.
(Physicist) A tensor is something dened by a multi-indexed system of components
in all coordinate systems on V , and its components in two dierent coordinate
systems are related by the transformation formula (6.1).
Mathematicians and physicists can check two rank k tensors t and t

are equal in the

same way: check t and t

have the same components in one coordinate system. But the

reason they consider this to be a verication of equality is not the same. The mathematician
6
G. B. Arfken and H. J. Weber, Mathematical Methods for Physicists, 6th ed., p. 133
7
G. F. J. Temple, Cartesian Tensors, 1960, p. 9
TENSOR PRODUCTS, II 23
thinks about the condition t = t

in a coordinate-free way but knows that to check t = t

it
suces to check t and t

have the same coordinates in a basis. The physicist, who thinks

about tensors in component form, considers the condition t = t

to mean (by denition!)

the components of t and t

match in all coordinate systems, and the rule (6.1) implies that
if the components of t and t

are equal in one coordinate system then they are equal in any
other coordinate system. Thats why the physicist is content to look only in one coordinate
system.
An operation on tensors (like the ip vw wv in V
2
) is checked to be well-dened by
the mathematician and physicist in dierent ways. The mathematician checks the operation
respects the universal mapping property that denes tensor products of vector spaces, while
the physicist checks the explicit formula for the operation on individual tensors changes in
dierent coordinate systems by the rule (6.1). The physicist would say an operation on
tensors makes sense because it transforms tensorially.
While the mathematicians may shake their heads and wonder how physicists can think
about tensors in component-dependent ways, that more concrete viewpoint is crucial to
understanding how tensors show up in physics. A physical quantity whose descriptions
in two dierent coordinate systems are related to each other in the same way that the
coordinates of a tensor in two dierent coordinate systems are related is asking to be
mathematically described as a tensor.
A nice example of this is in electromagnetism. Electric elds and magnetic elds are
rst learned about separately as vector elds in R
3
. Experiments show that a changing
electric eld produces a magnetic eld and a changing magnetic eld produces an electric
eld, so electric and magnetic elds are not really well-dened objects independent of the
coordinate system: an electric eld that vanishes in one coordinate system is nonzero in a
second coordinate system moving with respect to the rst one. But the formulas describing
how the components of the electric and magnetic elds interact (6 components total, 3 from
each eld) show that under a change of coordinates they transform together in the same way
as an antisymmetric rank 2 tensor on a 4-dimensional space. This is why physicists describe
an electromagnetic eld as a rank 2 tensor (in the 4-dimensional Minkowski spacetime).
Physicists adopt the habit of identifying R
n
with its dual space (R
n
)

via the dot product:

v is associated to the linear map w v w. For example, a bilinear form B: R
n
R
n
R
is the same thing as a linear function R
n

R
R
n
R, and all such functions form the dual
space (R
n

R
R
n
)

= (R
n
)

R
(R
n
)

, which can be thought of as R

n

R
R
n
, so bilinear
forms on R
n
are also rank 2 tensors on R
n
. Since M
n
(R) = Hom
R
(R
n
, R
n
)

= (R
n
)

R
R
n
can be thought of as R
n

R
R
n
, linear maps from R
n
to R
n
are rank 2 tensors on R
n
. This
accounts for the wide use of rank 2 tensors in physics, as they include both bilinear forms
and linear transformations. General relativity is full of rank 2 tensors for this reason.
The most basic example of a tensor in mechanics is the stress tensor, which has rank 2.
When a force is applied to a body the stress it imparts at a point may not be in the direction
of the force but in some other direction (compressing a piece of clay, say, can push it out
orthogonally to the direction of the force), so stress is described by a linear transformation,
and thus is a rank 2 tensor. The link between stress and strain tensors is the elasticity
tensor. The stress and strain tensors are rank 2, so they are elements of V
2
(V = R
3
),
and the elasticity tensor is a linear map in Hom
R
(V
2
, V
2
)

= (V
2
)

R
V
2
= V
4
, so
the elasticity tensor has rank 4. Since the stress from an applied force can act in dierent
directions at dierent points, the stress tensor is not really a single tensor but a varying
family of tensors at dierent points: stress is a tensor eld, which is a generalization of a
24 KEITH CONRAD
vector eld. Tensors in physics usually occur as part of a tensor eld, so tensor really
means tensor eld in physics. A change of variables between two coordinate systems {x
i
}
and {x

j
} in a region of R
n
involves partial derivatives
x
i
x

j
, and the tensor transformation
formula (6.1) shows up in physics with the products
x
i
1
x

j
1

x
i
k
x

j
k
, which vary from point to
point, in the role of a
i
1
j
1
a
i
k
j
k
.
Quantum mechanics is another part of physics where tensors play a role, but for rather
dierent (non-computational) reasons than weve seen already. In classical mechanics, the
states of a system are modeled by the points on a nite-dimensional manifold, and when
we combine two systems the corresponding manifold is the direct product of the manifolds
for the original two systems. The states of a quantum system, on the other hand, are
represented by the nonzero vectors (really, the 1-dimensional subspaces) in an innite-
dimensional Hilbert space, such as L
2
(R
6
). (A point in R
6
has three position and three
momentum coordinates, which is the classical description of a particle.) When we combine
two quantum systems the corresponding Hilbert space is the tensor product of the original
two Hilbert spaces, essentially because L
2
(R
6
R
6
) = L
2
(R
6
)
C
L
2
(R
6
), which is the
analytic
8
analogue of R[X, Y ]

= R[X]
R
R[Y ]. While in electromagnetism and relativity
the physicist uses specic tensors (e.g., the electromagnetic or metric tensor), in quantum
mechanics it is a whole tensor product space H
1

C
H
2
that gets used.
The distinction between direct products MN of manifolds and tensor products H
1

C
H
2
of vector spaces reects mathematically some of the non-intuitive features of quantum
mechanics. Every point in MN is a pair (x, y) where x M and y N, so we get a direct
link to something in M and something in N. On the other hand, most tensors in H
1

C
H
2
are not elementary. The non-elementary tensors in H
1

C
H
2
have no simple-minded
description in terms of H
1
and H
2
separately, and this reects the diculty of trying to
describe quantum phenomena for a combined system (e.g., the two-slit experiment) purely
in a classical language about the two original systems. Ive been told that physics students
who get used to computing with tensors in mechanics and relativity by learning to work
with the transform by a denite rule description of tensors can nd the role of tensors in
quantum mechanics to be dicult to learn, because the conceptual role of the tensors is so
dierent.
Well end this discussion of tensors in physics with a story. I was the mathematics
consultant for the 4th edition of the American Heritage Dictionary of the English Language
(2000). The editors sent me all the words in the 3rd edition with mathematical denitions,
and I had to correct any errors. Early on I came across a word I had never heard of
before: dyad. It was dened in the 3rd edition as an operator represented as a pair of
vectors juxtaposed without multiplication. Thats a ridiculous denition, as it conveys no
meaning at all. I obviously had to x this denition, but rst I had to know what the word
meant! In a physics book
9
a dyad is dened as a pair of vectors, written in a denite
order AB. This is just as useless, but the physics book also does something with dyads,
which gives a clue about what they really are. The product of a dyad AB with a vector
C is A(B C), where B C is the usual dot product (A, B, and C are all vectors in R
n
).
This reveals what a dyad is. Do you see it? Dotting with B is an element of the dual
space (R
n
)

, so the eect of AB on C is reminiscient of the way V V

acts on V by
(v )(w) = (w)v. A dyad is really an elementary tensor v in R
n
(R
n
)

. In the 4th
8
This tensor product should be a completed tensor product, including innite sums of products f(x)g(y).
9
H. Goldstein, Classical Mechanics, 2nd ed., p. 194
TENSOR PRODUCTS, II 25
edition of the dictionary, I included two denitions for a dyad. For the general reader, a
dyad is a function that draws a correspondence
10
from any vector u to the vector (v u)w
and is denoted vw, where v and w are a xed pair of vectors and v u is the scalar product
of v and u. For example, if v = (2, 3, 1), w = (0, 1, 4), and u = (a, b, c), then the dyad
vw draws a correspondence from u to (2a +3b +c)w. The more concise second denition
was: a dyad is a tensor formed from a vector in a vector space and a linear functional on
that vector space.
More general than a dyad is a dyadic. The physicists dene them
11
as a sum of dyads:
AB+CD+. . . . So a dyadic is a general element of R
n

R
(R
n
)

= Hom
R
(R
n
, R
n
). The
terminology of dyads and dyadics goes back to Gibbs (1884), who was aiming to develop a
coordinate-free way of describing linear transformations. Elementary tensors in the tensor
product of 3, 4, or more vector spaces are called triads, tetrads, and more generally polyads.
7. Tensor Product of R-Algebras
Many times our tensor product isomorphisms have involved rings, e.g., R[X]
R
R[Y ]

=
R[X, Y ], but we only said the isomorphism holds at the level of R-modules since we didnt
have a multiplication of tensors. Now we will show how to turn the tensor product of two
rings into a ring. Then we will revisit a number of previous module isomorphisms where
the modules are also rings and nd that the isomorphism holds at the level of rings too.
Because we want to be able to say C
R
M
n
(R)

= M
n
(C) as rings, not just as vector
spaces (over R or C), and matrix rings are noncommutative, we are going to allow our
R-modules to be possibly noncommutative rings. But R itself remains commutative!
Our rings will all be R-algebras. An R-algebra is an R-module A equipped with an R-
bilinear map AA A, called multiplication or product. The bilinearity of multiplication
includes the distributive laws for multiplication over addition as well as the rule r(ab) =
(ra)b = a(rb) for r R and a and b in B, which says R-scaling commutes with multiplication
in the R-algebra. We also want 1 a = a for a A, where 1 is the identity element of R.
Examples of R-algebras include the matrix ring M
n
(R), a quotient ring R/I, and the
polynomial ring R[X
1
, . . . , X
n
]. We will assume, as in all these examples, that our algebras
have associative multiplication and a multiplicative identity, so they are genuinely rings
(perhaps not commutative) and being an R-algebra just means they have a little extra
structure related to scaling by R.
12
The dierence between an R-algebra and a ring is exactly like that between an R-module
and an abelian group. An R-algebra is a ring on which we have a scaling operation by R
that behaves nicely with respect to the addition and multiplication in the R-algebra, in the
same way that an R-module is an abelian group on which we have a scaling operation by
R that behaves nicely with respect to the addition in the R-module. While Z-modules are
nothing other than abelian groups, Z-algebras in our lexicon are nothing other than rings
(possibly noncommutative).
Because of the universal mapping property of the tensor product, to give an R-bilinear
multiplication A A A in an R-algebra A is the same thing as giving an R-linear map
A
R
A A. So we could dene an R-algebra as an R-module A equipped with an R-
linear map A
R
A
m
A, and declare the product of a and b in A to be ab := m(a b).
10
Yes, this terminology sucks. Blame the unknown editor at the dictionary for that one.
11
H. Goldstein, p. 194
12
Lie algebras are an important class of nonassociative algebras; they are not rings.
26 KEITH CONRAD
Associativity of multiplication can be formulated in tensor language: the diagram
A
R
A
R
A
1m

m1

A
R
A
m

A
R
A
m

A
commutes.
Theorem 7.1. Let A and B be R-algebras. There is a unique R-algebra structure on A
R
B
such that
(7.1) (a b)(a

) = aa

bb

for all elementary tensors. The multiplicative identity is 1 1.

Proof. If there is an R-algebra structure on A
R
B such that (7.1) holds then multiplication
between any two tensors is determined:
k

i=1
a
i
b
i

j=1
a

j
b

j
=

i,j
(a
i
b
i
)(a

j
b

j
) =

i,j
a
i
a

j
b
i
b

j
.
So the R-algebra structure on A
R
B satisfying (7.1) is unique if it exists at all. Our task
now is to write down an R-algebra structure on A
R
B satisfying (7.1).
One way to do this is to start from scratch and try to dene what left multiplication by
an elementary tensor a b should be by introducing a suitable bilinear map and making
it into a linear map. But rather than proceed by this route, well take advantange of
various maps we already know between tensor products. Writing down a multiplication
function on A
R
B which commutes with R-scaling means writing down an R-linear map
(A
R
B)
R
(A
R
B) A
R
B, and thats what were going to do. Using the commutativity
and associativity isomorphisms on tensor products, there are natural isomorphisms
(A
R
B)
R
(A
R
B)

= ((A
R
B)
R
A)
R
B

= (A
R
(B
R
A))
R
B

= (A
R
(A
R
B))
R
B

= ((A
R
A)
R
B)
R
B

= (A
R
A)
R
(B
R
B).
Let A
R
A
m
A
A and B
R
B
m
B
B be the linear maps corresponding to multiplication
in A and B. Their tensor product is a linear map (A
R
A)
R
(B
R
B)
m
A
m
B
A
R
B,
so composing m
A
m
B
with the above isomorphism (A
R
B)
R
(A
R
B)

= (A
R
A)
R
(B
R
B) creates a linear map (A
R
B)
R
(A
R
B) A
R
B. Tracking the eect of
these maps on (a b) (a

),
(a b) (a

) ((a b) a

) b

(a (b a

)) b

(a (a

b)) b

((a a

) b) b

(a a

) (b b

)
aa

bb

,
TENSOR PRODUCTS, II 27
where m
A
m
B
is used in the last step. This R-linear map (A
R
B)
R
(A
R
B) A
R
B
can be pulled back to an R-bilinear map (A
R
B) (A
R
B) A
R
B with the eect
(a b, a

) aa

bb

on pairs of elementary tensors, which is what we wanted for our

multiplication on A
R
B.
To prove 11 is an identity and that multiplication in A
R
B is associative, the equations
(1 1)t

= t

, t(1 1) = t, (t
1
t
2
)t
3
= t
1
(t
2
t
3
)
involving general tensors t

, t
1
, t
2
, and t
3
are additive in each tensor appearing on both
sides, so verifying these equations reduces to computations with elementary tensors, and
those can be checked directly.
Corollary 7.2. If A and B are commutative R-algebras then A
R
B is a commutative
R-algebra.
Proof. We want to check tt

= t

t for all t and t

in A
R
B. Both sides are additive in t, so
it suces to check the equation when t = ab is an elementary tensor: (ab)t

?
= t

(ab).
Both sides of this are additive in t

, so we are reduced further to the special case when

t

= a

is also an elementary tensor: (a b)(a

)
?
= (a

)(a b). The validity of

this is immediate from (7.1) since A and B are commutative.
A homomorphism of R-algebras is a function between R-algebras that is both R-linear
and a ring homomorphism. An isomorphism of R-algebras is a bijective R-algebra homo-
morphism. That is, an R-algebra isomorphism is simultaneously an R-module isomorphism
and a ring isomorphism. For example, the reduction map R[X] R[X]/(X
2
+X+1) is an
R-algebra homomorphism (it is R-linear and a ring homomorphism) and R[X]/(X
2
+1)

= C
as R-algebras by a + bX a + bi: this function is not just a ring isomorphism, but also
R-linear.
For any R-algebras A and B, there is an R-algebra homomorphism A A
R
B by
a a 1 (check!). The image of A in A
R
B might not be isomorphic to A. For instance,
in Z
Z
(Z/5Z) (which is isomorphic to Z/5Z by a (b mod 5) = ab mod 5), the image
of Z by a a 1 is isomorphic to Z/5Z. There is also an R-algebra homomorphism
B A
R
B by b 1 b. Even when A and B are noncommutative, the images of A and
B in A
R
B commute: (a 1)(1 b) = a b = (1 b)(a 1). This is like groups G and
H commuting in GH even if G and H are nonabelian.
It is worth contrasting the tensor product A
R
B and the direct product A B (com-
ponentwise addition and multiplication, with r(a, b) = (ra, rb)). Both are R-algebras. The
most basic dierence is that there are natural R-algebra homomorphisms AB

1
A and
AB

2
B by projection, while there are natural R-algebra homomorphisms A A
R
B
and B A
R
B in the other direction (out of A and B to the tensor product rather than
to A and B from the direct product). The projections out of the direct product A B
are surjective, but the maps to the tensor product A
R
B need not be injective, e.g.,
Z Z
Z
Z/5Z. The maps A A
R
B and B A
R
B are ring homomorphisms
and the images are subrings, but although there are natural functions A A B and
B AB given by a (a, 0) and b (0, b), these are not ring homomorphisms and the
images are ideals rather than subrings.
The R-algebras A B and A
R
B have dual universal mapping properties. For any
R-algebra C and R-algebra homomorphisms C

A and C

B, there is a unique
28 KEITH CONRAD
R-algebra homomorphism C AB making the diagram
C

AB

1
.x
x
x
x
x
x
x
x
x

F
F
F
F
F
F
F
F
F
A B
commute. For any R-algebra C and R-algebra homomorphisms A

C and B

C
such that the images of A and B in C commute ((a)(b) = (b)(a)), there is a unique
R-algebra homomorphism A
R
B C making the diagram
A

aa1

H
H
H
H
H
H
H
H
H
B

.
b1b
.v
v
v
v
v
v
v
v
v
A
R
B

C
commute.
A practical criterion for showing an R-linear map of R-algebras is an R-algebra homo-
morphism is as follows. If : A B is an R-linear map of R-algebras and {a
i
} is a spanning
set for A as an R-module (that is, A =

i
Ra
i
), then is multiplicative as long as it is so
on these module generators: (a
i
a
j
) = (a
i
)(a
j
) for all i and j. Indeed, if this equation
holds then

i
r
i
a
i

j
r
j
a
j

i,j
r
i
r
j
a
i
a
j

i,j
r
i
r
j
(a
i
a
j
)
=

i,j
r
i
r
j
(a
i
)(a
j
)
=

i,j
r
i
(a
i
)

j
r
j
(a
j
)
=

i
r
i
a
i

j
r
j
a
j

.
This will let us bootstrap a lot of known R-module isomorphisms between tensor products
to R-algebra isomorphisms by checking the behavior only on products of elementary tensors
(and checking the multiplicative identity is preserved, which is always easy). We give some
concrete examples before stating some general theorems.
Example 7.3. For ideals I and J in R, there is an isomorphism : R/I
R
R/J
R/(I +J) of R-modules where (x y) = xy. Then (1 1) = 1 and
((x y)(x

)) = (xx

yy

) = xx

yy

= xy x

= (x y)(x

).
TENSOR PRODUCTS, II 29
So R/I
R
R/J

= R/(I + J) as R-algebras, not just as R-modules. In particular, the
additive isomorphism Z/aZ
Z
Z/bZ

= Z/(a, b)Z is in fact an isomorphism of rings.

Example 7.4. There is an R-module isomorphism : R[X]
R
R[Y ] R[X, Y ] where
(f(X) g(Y )) = f(X)g(Y ). Then (1 1) = 1 and
((f
1
(X) g
1
(Y ))(f
2
(X) g
2
(Y ))) = (f
1
(X)f
2
(X) g
1
(Y )g
2
(Y ))
= f
1
(X)f
2
(X)g
1
(Y )g
2
(Y )
= f
1
(X)g
1
(Y )f
2
(X)g
2
(Y )
= (f
1
(X) g
1
(Y ))(f
2
(X) g
2
(Y )),
so R[X]
R
R[Y ]

= R[X, Y ] as R-algebras, not just as R-modules. (It would have suced

to check is multiplicative using monomial tensors X
i
Y
j
.)
In a similar way, R[X]
k
= R[X
1
, . . . , X
n
] as R-algebras. The indeterminate X
i
on the
right corresponds on the left to the tensor 1 X 1 with X in the ith position.
Example 7.5. When R is a domain with fraction eld K, K
R
K

= K as R-modules by
x y xy. This sends 1 1 to 1 and preserves multiplication on elementary tensors, so it
is an R-algebra isomorphism.
Theorem 7.6. Let A, B, and C be R-algebras. The standard R-module isomorphisms
A
R
B

= B
R
A,
A
R
(B C)

= (A
R
B) (A
R
C)
(A
R
B)
R
C

= A
R
(B
R
C).
are all R-algebra isomorphisms.
Proof. Exercise. Note the direct product of two R-algebras is the direct sum as R-modules
with componentwise multiplication, so rst just treat the direct product as a direct sum.
Corollary 7.7. For R-algebras A and B, A
R
B
n
= (A
R
B)
n
as R-algebras.
Proof. Induct on n.
We turn now to base extensions. Fix a homomorphism f : R S of commutative rings.
We can restrict scalars from S-modules to R-modules and extend scalars from R-modules to
S-modules. What about between R-algebras and S-algebras? An example is the formation
of C
R
M
n
(R), which ought to look like M
n
(C) as rings (really, as C-algebras) and not
just as complex vector spaces.
If A is an S-algebra, then we make A into an R-module in the usual way by ra = f(r)a,
and this makes A into an R-algebra (restriction of scalars). More interesting is extension
of scalars. For this we need a lemma.
Lemma 7.8. If A, A

, B, and B

are all R-algebras and A

A

and B

B

are
R-algebra homomorphisms then the R-linear map A
R
B

A

R
B

is an R-algebra
homomorphism.
Proof. Exercise.
Theorem 7.9. Let A be an R-algebra.
(1) The base extension S
R
A, which is both an R-algebra and an S-module, is an
S-algebra by its S-scaling.
30 KEITH CONRAD
(2) If A

B is an R-algebra homomorphism then S
R
A
1
S
R
B is an
S-algebra homomorphism.
Proof. 1) We just need to check multiplication in S
R
A commutes with S-scaling (not
just R-scaling): s(tt

) = (st)t

= t(st

). Since all three expressions are additive in t and t

,
it suces to check this when t and t

are elementary tensors:

s((s
1
a
1
)(s
2
a
2
))
?
= (s(s
1
a
1
))(s
2
a
2
)
?
= (s
1
a
1
)(s(s
2
a
2
)).
From the way S-scaling on S
R
A is dened, all these products equal ss
1
s
2
a
1
a
2
.
2) For an R-algebra homomorphism A

B, the base extension S
R
A
1
S
R
B is
S-linear and it is an R-algebra homomorphism by Lemma 7.8. Therefore it is an S-algebra
homomorphism.
We can also give A
R
S an S-algebra structure by S-scaling and the natural S-module
isomorphism S
R
A

= A
R
S is an S-algebra isomorphism.
Example 7.10. Let I be an ideal in R[X
1
, . . . , X
n
]. Check the S-module isomorphism
S
R
R[X
1
, . . . , X
n
]/I

= S[X
1
, . . . , X
n
]/(I S[X
1
, . . . , X
n
]) is an S-algebra isomorphism.
In one-variable, with I = (h(X)) a principal ideal in R[X],
13
Example 7.10 gives us an
S-algebra isomorphism
S
R
R[X]/(h(X))

= S[X]/(h
f
(X)),
where h
f
(X) is the result of applying f : R S to the coecients of h(X). (If f : Z
Z/pZ is reduction mod p, for instance, then h
f
(X) = h(X) mod p.) This isomorphism is
particularly convenient, as it lets us compute a lot of tensor products of elds. For instance,
R
Q
Q(

2)

= R
Q
Q[X]/(X
2
2)

= R[X]/(X
2
2)

= RR
as R-algebras since X
2
2 factors into distinct linear polynomials in R[X], and
R
Q
Q(
3

2)

= R
Q
Q[X]/(X
3
2)

= R[X]/(X
3
2)

= RC
as R-algebras since X
3
2 is a linear times a quadratic in R[X]. And
C
R
C

= C
R
R[X]/(X
2
+ 1)

= C[X]/(X
2
+ 1) = C[X]/(X i)(X +i)

= CC
as R-algebras. (Lets make the R-algebra isomorphism C
R
C

= C C explicit, as
it is not z w (z, w); thats not additive or even a bijection, since C
R
C contains
non-elementary tensors. Tracing the eect of the isomorphisms on elementary tensors,
z (a +bi) z (a +bX) za +zbX (za +zbi, za +ab(i)) = (z(a +bi), z(a bi)),
so z w (zw, zw). Thus 1 1 (1, 1), z 1 (z, z), and 1 w (w, w).)
In these examples, a tensor product of elds is not a eld. But sometimes it can be. For
instance,
Q(

2)
Q
Q(

3)

= Q(

2)
Q
Q[X]/(X
2
3)

= Q(

2)[X]/(X
2
3),
which is a eld because X
2
3 is irreducible in Q(

2)[X]. As an example of a tensor

product involving a nite eld and a ring,
Z/5Z
Z
Z[i]

= Z/5Z
Z
Z[X]/(X
2
+ 1)

= (Z/5Z)[X]/(X
2
+ 1)

= Z/5Z Z/5Z
since X
2
+ 1 = (X 2)(X 3) in (Z/5Z)[X].
13
Not all ideals in R[X] have to be principal, but this is just an example.
TENSOR PRODUCTS, II 31
Example 7.11. For any R-module M, there is an S-linear map
S
R
End
R
(M) End
S
(S
R
M)
where s s
S
= s(1 ). Both sides are S-algebras. Check this S-linear map is
an S-algebra map. When M is nite free this map is a bijection (chase bases), so it is an
S-algebra isomorphism. For other M it might not be an isomorphism.
As a concrete instance of this, when M = R
n
we get S
R
M
n
(R)

= M
n
(S) as S-algebras.
Example 7.12. If I is an ideal in R and A is an R-algebra, R/I
R
A

= A/IA rst as R-
modules, then as R-algebras (the R-linear isomorphism is also multiplicative and preserves
identities), and nally as R/I-algebras since it is R/I-linear too.
Theorem 7.13. If A is R-algebra and B is an S-algebra, then the S-module structure on
the R-algebra A
R
B makes it an S-algebra, and
A
R
B

= (A
R
S)
S
B
as S-algebras sending a b to (a 1) b.
Proof. It is left as an exercise to check the S-module and R-algebra structure on A
R
B
make it an S-algebra. As for the isomorphism, from part I we know there is an S-module
isomorphism with the indicated eect on elementary tensors. This function sends 1 1 to
(1 1) 1, which are the multiplicative identities. It is left to the reader to check this
function is multiplicative on products of elementary tensors too.
Theorem 7.13 is particularly useful in eld theory. Consider two eld extensions L/K
and F/K with an intermediate eld K E L, as in the following diagram.
L
F E







K
}
}
}
}
}
}
}
@
@
@
@
@
@
@
@
Then there is a ring isomorphism
F
K
L

= (F
K
E)
E
L
which is also an isomorphism as E-algebras, F-algebras (from the left factor) and L-algebras
(from the right factor).
Theorem 7.14. Let A and B be R-algebras. There is an S-algebra isomorphism
S
R
(A
R
B) (S
R
A)
S
(S
R
B)
by s (a b) s((1 a) (1 b)).
Proof. By part I, there is an S-module isomorphism with the indicated eect on tensors of
the form s (a b). This function preserves multiplicative identities and is multiplicative
on such tensors (which span S
R
(A
R
B)), so it is an S-algebra isomorphism.