Determinant and Volume

1. Notation. If S is a set in R
n
and A is a matrix, then A(S) denotes the set of all vectors Ax such that
x is in S. So, A(S) is the set of outputs you get by applying A to inputs from S.
vol
n
denotes n-dimensional volume. For example, for n = 1, 2, 3 respectively this would be length, area
and the usual volume.
Given k vectors v
1
, . . . , v
k
in R
n
, the parallelepiped generated by v
1
, . . . , v
k
is the set of all vectors of the form
c
1
v
1
+ + c
k
v
k
such that 0 c
i
1, for each i. (If k = 2 and the vectors v
1
, v
2
are linearly independent,
then this is a parallelogram, with corners 0, v
1
, v
2
and v
1
+v
2
.)
If A is a matrix, then P
A
denotes the parallelepiped generated by its column vectors.
2. The facts. This is what you need to know.
1. Suppose A is an n n matrix and S is a subset of R
n
. Then
vol
n
(A(S)) = |det(A)| vol
n
(S) .
2. If A is n n, then
vol
n
(P
A
) = | det(A| .
3. If A is n k with k n, then
vol
k
(P
A
) =
_
| det(A
tr
A)| .
3. Remarks.
Actually, in this course you will never get a test question on n dimensional volume with n > 3. And
for the last item, I would only test a case with k n 3, probably k = 2 and n = 3 (area of a
parallelogram in R
3
). On a test, you should be able to do a computation similar to one of the examples
of the next section.
Above, Fact 2 can be considered a special case of Fact 3: if A is square, then det A is dened and
det(A
tr
A) = det(A) det(A
tr
) = (det(A))
2
.
Above, Fact 2 follows from Fact 1, because P
A
equals A(S) if S is the unit cube. (The unit cube in R
n
is the parallelepiped spanned by the standard basis vectors e
1
, ..., e
n
. Its n dimensional volume is 1.)
4. Examples.
1. If A =
_
5 2
3 0
_
, then P
A
is the parallelogram whose four corners are the column vectors (0, 0), (2, 0), (5, 3)
and (7, 3). The area of P
A
is | det(A)| = |(5)(0) (2)(3)| = | 6| = 6 .
(Draw the paralleleogram you can easily check the area is indeed 6.)
2. Let B denote the unit ball in R
3
, (the set of points (x
1
, x
2
, x
3
) such that (x
1
)
2
+ (x
2
)
2
+ (x
3
)
2
1.
Now dene an ellipsoid S and a matrix A as follows:
S = {y :
(y
1
)
2
4
+
(y
2
)
2
9
+ (y
3
)
2
1} and A =
_
_
2 0 0
0 3 0
0 0 1
_
_
.
Then y = Ax is written out as (y
1
, y
2
, y
3
) = (2x
1
, 3x
2
, x
3
). It follows that x is in B if and only if Ax
is a point y in S. That is, A(B) = S. Consequently,
vol
3
(S) = | det(A)|vol
3
(B) = 6(4/3)(1)
3
= 8 .
3. The area of the parallelogram P in R
3
generated by the column vectors (1, 2, 3) and (2, 0, 4) is computed
from A =
_
_
1 2
2 0
3 4
_
_
by
area(P) =
_
| det(A
tr
A)|
=

_| det(
_
1 2 3
2 0 4
_
_
_
1 2
2 0
3 4
_
_
)|
=

| det(
_
14 14
14 20
_
| =

84 .
5. A remark on Fact 1. Consider the one dimensional case. Now A =
_
a
_
for some number a, and Ax
is just ax, the product of two numbers. If S is any interval [c, d], then A(S) is an interval whose length is
|a|(d c), the length of S multiplied by |a|.
For n > 1, multiplication of length-of-interval everywhere by the same number |a| becomes multiplication of
every vol
n
(S) by the same number, | det(A)|.
For n = 2, there was some discussion of why this holds in class, and there is discussion in Lay. For n > 2,
the arguments are essentially the same. I wont repeat this here.
6. Sketch of proof of Fact 3. This is purely optional reading, for the curious (such as math majors).
To be concrete, let us consider k = 2 and n = 3: computing the area of a parallelogram P in R
3
, where P
is generated by linearly independent vectors v and w. (If v and w were linearly dependent, then det(A
tr
A)
and the area of P
A
would both be zero.) The basic idea of the proof works in the general case stated as Fact
3.
We can imagine P as a sti piece of cardboard, with one corner at the origin. Let u be a unit vector
perpendicular to P. The parallelepiped Q generated by u, v, w is a boxof height 1 built over a base P.
The three dimensional volume of Q as a number is the same as the area of P. Let M be the matrix whose
columns in order are v, w, u. Then | det(M)| = vol(Q) = area(P).
How to relate | det(M)| to A? Let u
1
, u
2
be an orthonormal basis for the subspace (plane) spanned by v
and w. Let u
3
= u. Now U = {u
1
, u
2
, u
3
} is an orthonormal basis for R
3
. Let U be the matrix whose
columns in order are u
1
, u
2
, u
3
. The matrix U is an orthogonal matrix: U
1
= U
tr
. Also, if e
1
, e
2
, e
3
are the
standard basis, then for i = 1, 2, 3 we have Ue
i
= u
i
, and therefore U
1
u
i
= e
i
. Since every vector in W is a
linear combination of u
1
and u
2
, it follows that U
1
maps W to the horizontal plane H through the origin,
H = {(a, b, 0) : a, b are in R}. Let us use V to denote U
1
. Since the rst two columns of M (v and w) are
in W, the rst two columns of V M are in H. The third column of V M is e
3
. Therefore V M has the form
V M =
_
_
b
11
b
12
0
b
21
b
22
0
0 0 1
_
_
and then (V M)
tr
(V M) has the form
(V M)
tr
V M =
_
_
b
11
b
21
0
b
12
b
22
0
0 0 1
_
_
_
_
b
11
b
12
0
b
21
b
22
0
0 0 1
_
_
=
_
_
c
11
c
12
0
c
21
c
22
0
0 0 1
_
_
.
Note, by the nature of matrix multiplication and transpose, here c
ij
is the dot product of column i of
V M with column j of V M. Because V is an orthognal matrix, this equals the dot product of column i of
M with column j of M. The rst two columns of M are the columns of A, so for i, j in {1, 2} we have
((V M)
tr
V M)(i, j) = (A
tr
A)(i, j). Since (A
tr
A) is 2 2, this means
_
c
11
c
12
c
21
c
22
_
= A
tr
A .
Because V is an orthogonal matrix, we have V
1
= V
tr
, and we can compute
A
tr
A = A
tr
IA = A
tr
V
tr
V A
= (V A)
tr
V A
and remembering det V = 1 implies det(V M) = det M, we have (at last)
(area(P))
2
= (volQ)
2
= (det M)
2
= (det V M)
2
= det(V M)
tr
det(V M)
= det(V M)
tr
V M = det
_
c
11
c
12
c
21
c
22
_
= det(A
tr
A) .
Taking square root, we nish with
area(P) =
_
det(A
tr
A) .
We do not have to write
_
| det(A
tr
A)|, because det(A
tr
A) is always nonnegative, because above it is the
square of a number.
7. One more example. Let us compute the area of the parallelgram P in R
3
generated by the column
vectors (2, 5, 0) and (3, 1, 0). Dene matrices
A =
_
_
2 3
5 1
0 0
_
_
and B =
_
2 3
5 1
_
.
Our formula for the area of P should agree with area(P) = | det(B)|. Lets check.
By the nature of matrix multiplication,
A
tr
A =
_
2 5 0
3 1 0
_
_
_
2 3
5 1
0 0
_
_
=
_
2 3
5 1
__
1 5
3 1
_
= B
tr
B .
Therefore
area(P) =
_
det(A
tr
A) =
_
det(B
tr
B) = | det(B)|
as desired.