MATH 223. Orthogonal Vector Spaces. U = {v ∈ R

advertisement
MATH 223. Orthogonal Vector Spaces.
Let U, V be vector spaces with U ⊆ V . We consider
U ⊥ = {v ∈ Rn : for all u ∈ U, < u, v >= 0}
Theorem 0.1 U ⊥ is a vector space.
Proof: We have that U T is a vector space. Here we must verify that 0 ∈ U ⊥ since this will not
follow from the two closure ideas. That is because < u, 0 >= 0 always. Also if x, y ∈ U ⊥ , then
< x + y, u >=< x, u > + < y, u > and < cx, u >= c < x, u > by our inner product axioms. Thus
if for all u ∈ U , < x, u >= 0 and < y, u >= 0, then we conclude that < x + y, u >=< x, u > + <
y, u >= 0 + 0 = 0 and also < cx, u >= c < x, u >= c · 0 = 0. Thus we have x + y and cx in U ⊥ ,
verifying closure. So U ⊥ is a vector space.
Consider a vector space U ⊆ Rn . Thus we are thinking of V = Rn with the standard basis
e1 , e2 , . . . , en . Let {u1 , u2 , . . . , uk } be a basis for U . Then if we write each ui with respect to the
standard basis we can form a matrix A = (aij ) with the ith row A being uTi . Thus row space(A) = U
and dim(U ) = rank(A). Then
null space(A) = {x : Ax = 0} = {x : < x, ui >= 0 for i = 1, 2, . . . , k}
= {x : < x, u >= 0 for all u ∈ U } = U ⊥
Thus dim(U ) + dim(U T ) = n using our result that dim(nullsp(A)) + rank(A) = n where n is the
number of columns in A These ideas will happily generalize to two vector spaces U, V with U ⊆ V .
We do not need V = Rn but we can benefit from an orthonormal basis for V in order to use the
null space idea. If we apply Gram Schmidt or otherwise, we can obtain a basis {v1 , v2 , . . . , vn }
with the orthonormal properties:
(
< vi , vj >=
if i 6= j
if i = j
0
1
Now proceed much as before writing
ui =
n
X
(∗)
aij vj
j=1
since {v1 , v2 , . . . , vn } is a basis for V and ui ∈ V . Let A be the associated k × n matrix. Now
P
consider any vector w ∈ V which we can write as w = nj=1 wj vj . Let w denote the vector in the
coordinates of the orthonormal basis so w = (w1 , w2 , . . . , wn )T Then
< ui , w >=<
n
X
aij vj ,
j=1
=
n
X
=
n
X
j=1
aij
w ` v` >
`=1
aij < vj ,
j=1
n
X
n
X
!
w ` v` >
`=1
n
X
!
w` (< vj , v` >)
`=1
=
n
X
aij wj
j=1
using properties of (*). Now nj=1 aij wj is the ith entry of Aw. Thus we have a way of expressing
U ⊥ as above and we have the desired result
P
Theorem 0.2 Let U, V be vector spaces with U a subspace of V and V is finite dimensional. Then
dim(U ) + dim(U T ) = dim(V ).
Another approach that doesn’t use an orthonormal basis of V (with respect to the given inner
product) but just any basis v1 , v2 , . . . , vn , we use the observation that for a given ui , the function
< ui , x > is a linear transformation V → R and so has an associated 1 × n matrix. Now we verify
that the k linear transformations < ui , x > are linearly independent. Assume
k
X
ci < ui , x >≡ 0
i=1
where we use the notation ≡ 0 to mean the identically 0 function, namely the 0 vector in the space
of functions. But now
k
X
i=1
ci < ui , x >=<
k
X
ci ui , x > for all x
i=1
but when we evaluate the righthand side at x = ki=1 ci ui , we obtain < x, x >= 0 and so by the
P
axioms of an inner product we have x = 0 i.e. ki=1 ci ui = 0 which forces c1 = c2 = · · · = ck = 0
since the vectors u1 , u2 , . . . , uk are linearly independent.
P
Theorem 0.3 Let U, V be vector spaces with U a subspace of V and V is finite dimensional. Then
⊥
U ⊥ = U.
Download