Fall 2024 Advanced Linear Algebra (MATH 6351) November 13

Solutions to Assignment 7

1. (a) For the matrix of f` relative to (C, B ∗ ), the (i, j)-entry is the i-th coordinate of
f` (cj ) relative to B ∗ , which equals f` (cj )(bi ) = f (cj , bi ) = aji . Therefore, the
matrix of f` is AT .
Similarly, for the matrix of fr relative to (B, C ∗ ), the (i, j)-entry is fr (bj )(ci ) =
f (ci , bj ) = aij . Therefore, the matrix of fr is A.
(b) Let A and A0 be the matrices of a bilinear function f : U ×V → F relative to (C, B)
and to (C 0 , B 0 ), respectively. By part (a), these are also the matrices of the linear
operator fr : V → U ∗ relative to (B, C ∗ ) and (B 0 , (C 0 )∗ ), respectively. Suppose
P is the transition matrix from B to B 0 and Q is the transition matrix from C
to C 0 . Then the transition matrix from C ∗ to (C 0 )∗ is (QT )−1 (see problem 2 on
Assignment 4). Therefore, A0 = ((QT )−1 )−1 AP = QT AP .

2. Define a function g̃ : V /N × V /N → F by g̃(u + N, v + N ) = g(u, v). This g̃ is well

defined, because if u0 is another representative of the coset u + N and v 0 of v + N , then
u0 = u + x and v 0 = v + y for some x, y ∈ N , and hence g(u0 , v 0 ) = g(u + x, v + y) =
g(u, v) + g(u, y) + g(x, v) + g(x, y) = g(u, v) since g(x, V ) = 0 and g(V, y) = 0.
The function g̃ is linear in the second variable: for all u, v, w ∈ V and c ∈ F , we have
g̃(u + N, c(v + N )) = g̃(u + N, cv + N ) = g(u, cv) = cg(u, v) = cg̃(u + N, v + N ) and
g̃(u + N, (v + N ) + (w + N )) = g̃(u + N, (v + w) + N ) = g(u, v + w) = g(u, v) + g(u, w) =
g̃(u + N, v + N ) + g̃(u + N, w + N ). If g is symmetric (respectively, skew-symmetric or
Hermitian), then so is g̃. It follows that g̃ is linear in the first variable in the symmetric
and skew-symmetric cases, and antilinear in the first variable in the Hermitian case.
To show that g̃ is nondegenerate, suppose u+N ∈ V /N is such that g̃(u+N, v +N ) = 0
for all v ∈ V . By definition of g̃, this means g(u, v) = 0 for all v ∈ V , so u ∈ V ⊥ = N ,
i.e., u + N = N = 0V /N .

3. (a) First we check that this is a symmetric inner product. For all ϕ, ψ1 , ψ2 ∈ C[a, b]
Rb Rb
and c1 , c2 ∈ R, hϕ, c1 ψ1 + c2 ψ2 i = a ϕ(t)(c1 ψ1 (t) + c2 ψ2 (t))dt = c1 a ϕ(t)ψ1 (t) +
Rb
c2 a ϕ(t)ψ2 (t))dt = c1 hϕ, ψ1 i + c2 hϕ, ψ2 i, proving linearity in the second variable.
Also, hϕ, ψi = hψ, ϕi, because ϕ(t)ψ(t) = ψ(t)ϕ(t) for all t ∈ [a, b]. Linearity in
the first variable follows.
Rb
For positivity, consider hϕ, ϕi = a ϕ(t)2 dt for some 0 6= ϕ ∈ C[a, b]. Then
ϕ(t)2 ≥ 0 for all t ∈ [a, b], and, by continuity, ϕ(t)2 > 0 for t in some open
interval around any point t0 where ϕ(t0 ) 6= 0. It follows that hϕ, ϕi > 0.
(b) First we check that this is a Hermitian inner product. For all X, Y ∈ Mn (C) and
T T T
c1 , c2 ∈ C, hX, c1 Y1 + c2 Y2 i = tr(X (c1 Y1 + c2 Y2 )) = c1 tr(X Y1 ) + c2 tr(X Y2 ) =
c1 hX, Y1 i + c2 hX, Y2 i, where we have used the linearity of trace. Also, hY, Xi =
 T
 T 
T
tr(Y X) = tr(X Y ) = tr X Y = hX, Y i, where we have used the obvious
properties tr(AT ) = tr(A) and tr(Ā) = tr(A).
For positivity, observe that, for any 0 6= X = (xij ) ∈ Mn (C), the (i, j)-entry of
T
X X equals nk=1 x̄ki xkj , so hX, Xi = ni,k=1 x̄ki xki = ni,k=1 |xki |2 > 0.
P P P

(c) Linearity in the second variable is checked as in part (b). The symmetry is a
property of trace: tr(XY ) = tr(Y X) (prove it!).
For nondegeneracy, observe that hEij , Ek` i = tr(EijP Ek` ) = δjk tr(Ei` ) = δjk δi` .
Hence, for any X ∈ Mn (F ), we have hEij , Xi = hEij , nk,`=1 xk` Ek` i = xji . There-
fore, hMn (F ), Xi = 0 implies X = 0.
(d) First we check that this is a skew-symmetric (in fact, alternating) inner product.
For all (ϕ, v), (ψ1 , w1 ), (ψ2 , w2 ) ∈ V ∗ ⊕ V and c1 , c2 ∈ F , we have
h(ϕ, v), c1 (ψ1 , w1 ) + c2 (ψ2 , w2 )i = h(ϕ, v), (c1 ψ1 + c2 ψ2 , c1 w1 + c2 w2 )i
= ϕ(c1 w1 + c2 w2 ) − (c1 ψ1 + c2 ψ2 )(v)
= c1 ϕ(w1 ) − c1 ψ1 (v) + c2 ϕ(w2 ) − c2 ψ2 (v)
= c1 h(ϕ, v), (ψ1 , w1 )i + c2 h(ϕ, v), (ψ2 , w2 )i.
Also, h(ϕ, v), (ϕ, v)i = ϕ(v) − ϕ(v) = 0, so this inner product is alternating and
hence skew-symmetric.
For nondegeneracy, observe that h(ϕ, 0), (ψ, w)i = ϕ(w) and h(0, v), (ψ, w)i =
−ψ(v). Therefore, hV ∗ ⊕ V, (ψ, w)i = 0 implies ϕ(w) = 0 for all ϕ ∈ V ∗ and
ψ(v) = 0 for all v ∈ V , hence w = 0 and ψ = 0.
4. Let B1 = {b1 , . . . , bn }, so B2 = {f (b1 ), . . . , f (bn )}. By definition, the (i, j)-entry of the
matrix [g1 ]B1 is g1 (bi , bj ) and the (i, j)-entry of the matrix [g2 ]B2 is g2 (f (bi ), f (bj )). If f
is an isometry, then g1 (bi , bj ) = g2 (f (bi ), f (bj )), so the matrices are equal. Conversely,
suppose [g1 ]B1 = [g2 ]B2 and denote P this matrix by A. For any u ∈ VP 1 , let x = [u]B1 ∈
F . By definition, this means u = j=1 xj bj , which implies f (u) = nj=1 xj f (bj ) and
n n

hence x = [f (u)]B2 . Similarly, for v ∈ V , let y = [v]B1 = [f (v)]B2 . Then g1 (u, v) =

xT Ay = g2 (f (u), f (v)) in the symmetric and skew-symmetric cases, and g1 (u, v) =
x̄T Ay = g2 (f (u), f (v)) in the Hermitian case. Therefore, f is an isometry.
5. (a) By the definition of W , we have W = {x ∈ R4 | u1 · x = u2 · x = 0} where
u1 = (2, 2, 1, 1) and u2 = (1, 2, 1, 2), which means that W = U ⊥ where U is
the subspace generated by u1 and u2 . Hence, W ⊥ = U has a basis {u1 , u2 }.
·u2
To make an orthonormal basis, we compute u02 = u2 − uu11 ·u 1
u1 = u2 − 10 9
u1 =
1 1 1
10
(−8, 2, 1, 11) (which is orthogonal to u1 ) and take b1 = ku1 k u1 = √10 (2, 2, 1, 1)
and b2 = ku10 k u02 = √190
1
(−8, 2, 1, 11).
2

(b) The distance from v := (2, 4, 0, −1) to W is the length of the orthogonal projection
q
2 √
+ (−19)
2
p
of v onto W ⊥ , which is (b1 · v)2 + (b2 · v)2 = 11 10 190
= 14.

6. Observe that the condition g|U ×U = 0 is equivalent to U ⊂ U ⊥ . Indeed, U ⊂ U ⊥ means

that, for any x ∈ U , we have g(x, U ) = 0, and this is the same as saying g(x, y) = 0
for all x, y ∈ U .
(⇒) Suppose g|U ×U = 0 and U is maximal with this property. As just observed,
we have U ⊂ U ⊥ . For the inclusion U ⊥ ⊂ U , let v ∈ U ⊥ and assume, by way of
contradiction, that v ∈ / U . Then Ũ := U ⊕ F v contains U properly, and we claim
that g|Ũ ×Ũ = 0, which will contradict the maximality of U . Indeed, for any x, y ∈ U
and a, b ∈ F , we have g(x + av, y + bv) = g(x, y) + ag(v, y) + bg(x, v) + abg(v, v) = 0,
since each of the terms is equal to 0: the first because U is null, the second and third
because v ∈ U ⊥ , and the last because g is alternating.
(⇐) Suppose U ⊥ = U . Then g|U ×U = 0, since U ⊂ U ⊥ . For maximality, suppose
U ⊂ Ũ and g|Ũ ×Ũ = 0. Then, on the one hand, Ũ ⊥ ⊂ U ⊥ and, on the other hand,
Ũ ⊂ Ũ ⊥ . Therefore, Ũ ⊂ U ⊥ = U and we conclude Ũ = U , as required for maximality.
Since g : W × W → F is nondegenerate, we have dim U + dim U ⊥ = dim W , hence
U = U ⊥ implies dim U = 21 dim W .
Remark. In 3(d), U = V and U = V ∗ are examples of Lagrangian subspaces of
W =V∗⊕V.