Beruflich Dokumente
Kultur Dokumente
T
Definition: A linear transformation Rn −→ Rn is orthogonal if it preserves dot products—
that is, if ~x · ~y = T (~x) · T (~y ) for all vectors ~x and ~y in Rn .
T
Theorem: Let Rn −→ Rn be a linear transformation. Then T is orthogonal if and only if
it preserves the length of every vector—that is, ||T (~x)|| = ||~x|| for all ~x ∈ Rn .
*********************************************************************************************
A. Discuss the definition and theorem with your table: What, intuitively and geometrically,
really is an orthogonal transformation? What does an orthogonal transformation of
R3 do to the unit cube? How does this differ from a general (non-orthonormal) linear
transformation?
Solution note: The unit cube is taken to another unit cube, also with one
vertex at the origin. In general, a linear transformation can stretch the cube to a
parallelopiped (if rank T = 3) or smash the cube to 2 dimensional parallelogram
(if rank T is 2), or even a linear segment (if rank T = 1.).
B. Which of the following maps are orthogonal transformations? Short, geometric justifica-
tions are preferred.
5. Dilation R3 → R3 by a factor of 3.
cosθ −sinθ
6. Multiplication by .
sinθ cosθ
3 1
7. Multiplication by .
−2 5
Solution note:
1. Yes, dot product is obviously the same before or after doing nothing.
2. Yes, rotation obviously preserves the LENGTH of every vector, so by the theorem, this
means rotation is an orthogonal transformation.
3. Yes, reflection obviously preserves the LENGTH of every vector, so by the theorem,
this means reflection is an orthogonal transformation.
5. NO, dilation by three obviously takes each vector to a vector of length 3 times as long.
T
C. Prove that orthogonal transformations preserve lengths: if Rn −→ Rn is an orthogonal
transformation, then ||~x|| = ||T (~x)|| for all ~x ∈ Rn . Is the converse true? What is the book’s
definition of an orthogonal transformation and why is it equivalent to ours?
Solution note: Say that T is orthogonal. Take arbitrary ~x ∈ Rn . We want
to show ||~x|| = ||T (~x)||. By definition ||~x|| = (~x · ~x)1/2 and ||T (~x)|| = (T (~x) ·
T (~x))1/2 . Since T is orthogonal, we know that ~x · ~x = T (~x) · T (~x), so the result
follows. The converse is also true. This is the Theorem at the top! The book
DEFINES an orthogonal transformation as one which preserves lengths. This
is equivalent to ours by the Theorem above.
T
D. Let Rn −→ Rn be an orthogonal transformation.
3. Prove that the matrix of T (in standard coordinates) has columns that are orthonormal.
1. Let ~v ∈ ker T . We want to show that ~v = 0. We know that T (~v ) = 0, so ||T (~v )|| = 0.
Since T is orthogonal, it preserves lengths (by the theorem), so also ||~v || = 0. This
means ~v = 0 since no other vector has length 0. So the kernel of T is trivial and T is
injective.
3. Let A be the matrix of T . The columns of A are T (~e1 ), . . . , T (~en ) by our old friend
the ”Crucial Theorem.” We need to show these are orthonormal. Compute for each i:
T (~ei ) · T (~ei ) = ~ei · ~ei = 1 (where we have used the fact that T preserves dot product).
Also compute for each pair i 6= j: T (~ei ) · T (~ej ) = ~ei · ~ej = 0. This means that
T (~e1 ), . . . , T (~en ) are orthonormal.
4. Yes! Assume T and S are orthonormal transformations with source and target Rn .
To show that S ◦ T is orthogonal, we can show it preserves LENGTHS. Take any
vector ~x ∈ Rn . Then ||~x|| = ||T (~x)|| since T is orthogonal (by the Theorem, or by the
book definition). So also ||T (~x)|| = ||S(T (~x))|| since S is orthogonal. Putting these
together we have ||~x|| = ||S(T (~x))||. Since ~x was arbitrary, we see that S ◦ T preserves
every length. QED. Note: This is a different proof that what I did in class, using the
Theorem instead of the definition.
E. Recall that if A is an d × n matrix with columns ~v1 , . . . , ~vn , then the transpose of A is
the n × d matrix AT whose rows are ~v1T , . . . , ~vnT .
3
2. Suppose that A = [~v1 ~v2 ~v3 ] is a 3 ×3 matrix with columns ~v1 , ~v2 , ~v3 ∈ R . Use block
~v1 · ~v1 ~v1 · ~v2 ~v1 · ~v3
multiplication to show that AT A = ~v2 · ~v1 ~v2 · ~v2 ~v2 · ~v3
~v3 · ~v1 ~v3 · ~v2 ~v3 · ~v3
3. Prove that a square matrix A is orthogonal if and only if its columns are orthonormal.1
" #
3/5 4/5 0
4. Is B = −4/5 3/5 0 orthogonal? Find B −1 . [Be clever! There is an easy way!]
0 0 1
1. Yes. It is easy to check that AT A is the identity matrix in both cases (we use
sin2 + cos2 = 1 for the first).
T T
~v1 ~v1 ~v1 ~v1T ~v2 ~v1T ~v3
2. Multiply AT A = ~v2T ~v1 ~v2 ~v3 = ~v2T ~v1 ~v2T ~v2 ~v2T ~v3 , which produces the de-
~v3T ~v3T ~v1 ~v3T ~v2 ~v3T ~v3
sired matrix since for any n × 1 matrices (column vectors) w ~ and ~v , the matrix product
T
w
~ ~v is the same as the dot product w ~ · ~v .
3. This follows immediately from the n-dimensional version of (2). Let A have columns
~v1 , . . . , ~vn . Then the ij-th entry of AT A is ~viT ~vj = ~vi ·~vj . So the columns are orthonormal
if and only if AT A is the identity matrix.
" #
3/5 −4/5 0
4. Yes. The inverse is the transpose, which is 4/5 3/5 0
0 0 1
with the first equality coming from the definition of the length of the vector and the
second coming from the definition of T being orthogonal. For the second, T (~ei )·T (~ej ) =
~ei · ~ej by the Theorem above. This is zero since the standard basis is orthonormal.
2). Assume A has orthonormal columns. We need to show that for any ~x ∈ Rn ,
||T (~x)|| = ||~x||. Since the length is always a non-negative number, it suffices to show
||T (~x)||2 = ||~x||2 . That is, it suffices to show T (~x) · T (~x) = ~x · ~x. For, this we take an
arbitrary ~x and write it in the basis {~e1 , . . . , ~en }. Note that
X
~x · ~x = (x1~e1 + · · · + xn~en ) · (x1~e1 + · · · + xn~en ) = (xi~ei ) · (xj ~ej ) = x21 + · · · + x2n .
ij
here the third equality is using some basic properties of dot product (like ”foil”), and
the third equality is using the fact that the ~ei are orthonormal so that (xi~ei )·(xj ~ej ) = 0
if i 6= j. On the other hand, we also have
with the last equality coming from the fact that the T (~ei )’s are the columns of A and
hence orthonormal. QED.
F. Prove the very important Theorem on the top of the first page. [Hint: consider x + y.]
Solution note: We need to show two things: 1) If T is orthogonal, then T
preserves lengths; and 2) If T preserves lengths, then T is orthogonal. We
already did (1) in C. We need to do 2. Assume that for all ~x, ||T (~x)|| = ||~x||.
We need to show that for all ~x and ~y , ~x · ~y = T (~x) · T (~y ). Consider T (~x + ~y ).
We have ||~x + ~y || = ||T (~x + ~y )|| so also (~x + ~y ) · (~x + ~y ) = (T ~x + T ~y ) · (T ~x + T ~y )
This expands as ~x · ~x + 2~x · ~y + ~y · ~y = T ~x · T ~x + 2T ~x · T ~y + T ~y · T ~y . This is
the same as ||x||2 + 2~x · ~y + ||y||2 = ||T (~x)||2 + 2T (~x) · T (~y ) + ||T (~y )||2 . But we
know that ||x|| = ||T (~x)|| and ||y|| = ||T (~y )||, so we can cancel from both sides.
We get 2~x · ~y = 2T ~x · T ~y , so dividing by 2 we have ~x · ~y = T ~x · T ~y .
G. A square matrix with orthonormal columns is invertible, and its inverse is easy to find:
the inverse will be the transpose. Why? Prove in general that if the columns of A are
orthonormal, then AT A = Id for some d. Can AAT also be an identity matrix if A is not
square?
Solution note: We checked if the columns are orthonormal, then AT A = Id
where d is the number of rows of A, the argument really did not use square
matrix for this computation. However, if A is not square, then AAT will not be
an identity matrix. Any non-square matrix with orthogonal columns, such as
A = [~e1 ~e2 ], gives a counter example.
2. [(AB)C]T = C T (AB)T = C T B T AT .
b1
..
3. Say A = a1 . . . an and B = . . Then AB = ni=1 ai bi , which is a 1 × 1 matrix
P
bn
T T T
a1
so equal to its transpose (AB) . Also B A = b1 . . . bn . . ., which gives the
an
same product.
R1 R1 C1 R1 C2 · · · R1 Cn
R2 R2 C1 R2 C2 · · · R2 Cn
4. AB = .. C1 C2 · · · Cn = .. .. . So transposing, we
. . .
Rn Rn C1 Rn C2 · · · Rn Cn
R1 C1 R2 C1 · · · Rn C1
R1 C2 R2 C2 · · · Rn C2
T T T T
get (AB) = .. .. . Each Ri Cj = (Ri Cj ) = Cj Ri , since it is
. .
R1 Cn R2 Cn · · · Rn Cn
a scalar. So this product is also the same as
C1T
C T
2
B T AT = .. R1T R2T . . . RnT .
.
CnT