PROBLEMS ON TENSOR PRODUCTS

(SOLUTIONS TO HOMEWORK ASSIGNMENT #9)

DR. ROSENBERG

1. Let V , W , and U be finite-dimensional vector spaces over a field

F . (All Homs and tensor and exterior products in this problem are
over F .) Prove that there are natural isomorphisms
Hom (V ⊗ W, U ) ∼ = Hom (V, W ∗ ⊗ U )
and ^n ∗ ^n
V ∼= V∗
(for any n).
Solution. For the first part, first observe that we have a natural iso-
morphism
Hom (V ⊗ W, U ) ∼ = Bilin (V × W, U )
to the bilinear maps V × W → U (by the universal property of the
tensor product). Then we get a natural isomorphism
Bilin (V × W, U ) ∼= Hom (V, Hom(W, U ))
via restriction to the first factor. (I.e., a bilinear map V × W → U is
the same thing as a family of linear maps W → U varying linearly in
V .)
So to finish the first part, we have to show there is a natural isomor-
phism Hom(W, U ) ∼ = W ∗ ⊗ U and plug this in. Now Hom(W, F ) = W ∗
(by definition) and we can tensor both sides by U . But there is a nat-
ural isomorphism Hom(W, F ) ⊗F U ∼ = Hom(W, U ) via f ⊗ u 7→ (w 7→
f (w)u). (To check this is well-defined, start with the bilinear map
(f, u) 7→ (w 7→ f (w)u) and observe that it has to factor through the
tensor product.)
For the second part, let v1 , · · · , vm , m = dim V , be a basis for V
and let v1∗ , · · · , vm
∗ ∗ ∗
Vnbe the dual basis for V (so vj (vk ) = δjk ). Recall
that a basis for V , n ≤ m, consists of vI = vi1 ∧ · · · ∧ vin with
I = (i1 , · · · , in ), where 1 ≤ i1 V< · · · < in ≤ m. Similarly, the vI∗ =
vi1 ∧ · · · ∧ vinVwill be a basis for n V ∗ . We simply identify vI∗ with the
∗ ∗
∗
element of ( n V ) that Vn sends vIVto 1 and vJ to 0 for J 6= I; this gives
∗
our isomorphism Φ : V → ( nV) .
∗

Date: Nov. 19, 2013.

1
2 DR. ROSENBERG

We have to show this isomorphism is independent of the P choice of

basis. Let w1 , · · · , wm be another basis for V , and let wj = k ajk vk .
Here A = (ajk ) is an invertible “change of basis” matrix. The P change
of basis matrix B = (bjk ) for the dual bases satisfies wj∗ = k bjk vk∗ ,
and since
X
X X
δj` = wj∗ (w` ) = bjk vk∗ a`r vr = bjk a`r δkr bjk a`k ,
k,r k,r k
t t −1
we see that BA = I, or B = (A ) .
∗
The new basis defines an isomorphism Φ0 : n V ∗ → ( n V ) sending
V V
∗
wI to the linear functional with wI 7→ 1, wJ 7→ 0 for J 6= I. Let’s
compare it with Φ. First suppose n = m (perhaps the most interesting
case). Then the is only one applicable multi-index I = (1, · · · , n), and
wI = (det A)vI , while wI∗ = (det B)vI∗ . So
Φ(wI∗ )(wI ) = Φ (det B)vI∗ (det A)vI

= (det B)(det A)Φ(vI∗ )(vI ) = (det B)(det A)

= 1 = Φ0 (wI∗ )(wI ),
since det B = (det At )−1 = (det A)−1 , and so Φ0 = Φ.
In the general case the calculation is a bit messier:
X X 
δIJ = Φ(wI∗ )(wJ ) = Φ bi1 k1 vk∗1 ∧ · · · ∧ bin kn vk∗n
k1 kn
X X 
aj1 `1 v`1 ∧ · · · ∧ ajn `n v`n .
`1 `n

Only terms where k1 , · · · , kn are distinct, `1 , · · · , `n are distinct, and

k1 , · · · , kn are a permutation of `1 , · · · , `n survive and give ±1 (multi-
plied of course by the appropriate product of a’s and b’s), depending
on the sign of the permutation. So the result can be written as
X
δKL det(BIK ) det(AJL )
K,L

for appropriate n × n minors of A and B, and this gives δIJ via the
relationship between the matrices A and B. 

2. Show that (as rings) there is an isomorphism H⊗R H ∼= M4 (R), the

4×4 matrices over the reals. Here H is the division ring of quaternions,
which is a 4-dimensional algebra over R. Also show that there is an
isomorphism of rings H ⊗R C ∼ = M2 (C).
Solution. Let’s start with the first isomorphism. Note that H is an H-
H-bimodule (though the left and right actions do not coincide). Since
 SOLUTIONS TO HOMEWORK ASSIGNMENT #9 3

both the left and right actions of H on itself are R-linear, we get a
homomorphism of rings
Φ : H ⊗R Hop → EndR (H) ∼= M4 (R),
where Hop is H with the multiplication reversed. (The op is needed
because a right H-action is the same as a left action of Hop .) Note that
Φ is natural; no choice of basis is involved. Furthermore, Hop ∼
= H, since
quaternionic conjugation (1 7→ 1, i 7→ −i, j 7→ −j, k 7→ −k) is an anti-
isomorphism from H to itself (it reverses the order of multiplication).
Thus Φ can be viewed as a homomorphism
H ⊗R H → M4 (R).
It is injective since the domain and codomain both have dimension
42 = 16 and there cannot be any kernel, since the Φ(a ⊗ b) for a, b
running through 1, i, j, k (an R-basis of H) can be seen to be linearly
independent.
The second isomorphism is similar. Note that H is an H-C bimodule
via the left action of H on itself and the right action of C = R + Ri.
So we get a homomorphism Ψ : H ⊗R Cop → EndR (H). Since C is
commutative, this time we can drop the op . We can identify H with
C2 via the C-basis {1, j}, and the image of Ψ consists of C-linear maps
(for the right action of C) since left multiplication commutes with right
multiplication and C commutes with itself. Again, Ψ is an isomorphism
by dimension counting.