Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
X(a) = {x Rn | a x = 0}.
(i) Show that X(a) is a linear subspace.
(ii) Show the dimension of X(a).
Solution:
(i) If x, x0 X(a) then a x = a x0 = 0. Hence a x + a x0 = a (x + x0 ) = 0 and this
implies that x + x0 X(a). Again let x X(a) then a x = 0 and a x = 0 as well.
Moving the constant we see that a (x) = 0 which means that x X(a).
(ii) Clearly the dimension of Rn is n. The vector a represents a linear mapping with
one row. The linear span of a is contained in a subspace with dimension 1. The
set X(a) contains all the vectors which are orthogonal to a. By the rank nullity
theorem we know that rank( f ) + Null( f ) = dim(Rn ). Hence the dimension of the
null space X(a) is dim(Rn ) rank( f ) = n 1.
Another way to look at this is by using what you found in Problem 1 and recognizing that X(a) is a hyperplane in Rn passing through the origin. In Problem 1,
you found that a plane in R3 passing through the origin had dimension 3 1 = 2.
Well a hyperplane is just a plane, so if X(a) is a hyperplane through the origin in
Rn it must have dimension n 1.
1
Some Definitions The following definitions describe the greater than or equal to
type of ordering on Rn that will be important for answering Problem 3.
Definition 1. Let x and y be vectors in Rn . We define the following relations:
x = y iff xi = yi for all i = 1, 2, . . . , n.
x y iff xi yi for all i = 1, 2, . . . , n.
x > y iff xi yi and x , y meaning there is at least one element j such that
x j > y j.
x y iff xi > yi for all i = 1, 2, . . . , n.
Problem 3. Suppose a, x, y Rn and a x > a y. Does it follow that x > y? [Hint: do
not divide both sides by a]
Solution:
Let a, x, y Rn (note these are vectors) and let ax > ay. Note that both ax and ay are
real numbers so the > here refers to the ordering on the real line, while the comparison
we want to investigate x > y is an ordering on vectors (as defined before the problem).
Lets consider the value r y = ay. Consider the hyperplane X(a, r y ) = {z Rn : az = r y }
and the upper half-space X+ (a, r y ) = {z Rn : az r y }. Now we know that ax > r y
so we can be sure x X+ (a, r y ). Let Y = {z Rn : zi yi i = 1, . . . , n} be the set of all
vectors z such that z y. When Y ( X+ (a, r y ) then its possible for a x > a y and yet
(x > y). Consider the two-dimensional graph below. Each of the points x, x0 , x00 are
contained in X+ (a, r y ) yet are not in Y. So its possible for a x > a y and yet x y.
Figure 1: Hyperplanes, Half-spaces and Vector Ordering
x
x0
{z R2 : z > y}
a
y
x1
X+ (a, r y )
x2
X(a, r y )
x00
10. 1u = u
All the axioms should be tested.
(iv) For part (iv) the idea is to use what we just learned in part (ii), that every linear
functional f X can be uniquely determined by a vector a = ( f (e1 ), . . . , f (en )).
We can think about finding a set of such a vectors. For example suppose we
have a set of n vectors, say the coordinate vectors again, (e1 , . . . , en ), do each of
these represent their own linear functional? Yes. Now that we are calling them
functionals, lets represent the set now as { f1 , . . . , fn }. At this point we establish
that any linear functional is just a linear combination of these n linear functionals
and that the set is linearly independent. Once youve done that you will have
shown that a set of n vectors is a basis for X and therefore dim (X ) = n.
Problem 5. Prove that the set Z is a subspace of R3 .
Z = {[x1 , x2 , x3 ] | 4x1 x2 + 5x3 = 0}
Solution:
Determine the null space and range of each of the following linear
operators on R3 .
(i) L(x) = (x3 , x2 , x1 )T
(ii) L(x) = (x1 , x1 , x1 )T
Solution:
(i) Null(L) = {0} and L(R3 ) = R3 .
(ii) Null(L) = Span(e2 , e3 ) and L(R3 ) = Span(e1 , e2 ).
Definition 2. Let X be a vector space and let a X and b R. A hyperplane in X is
a set of the form Ha (b) = {x X : a x = b} and associated with hyperplane X are two
half-spaces Ha (b) = {x X : a x b} and Ha (b) = {x X : a x b}.
Problem 7. What are the range of angles between vectors in x Ha (0) and the vector
a? What are the ranges of the angles between vectors x Ha (0) and the vector a?
Solution: For Ha (0) the range on the angles between vectors x and the vector a are
= [0, 90] [270, 360] and the range on the angles between vectors x in Ha (0) and a
are = [90, 270].
Problem 8.
(iii) the set of 2 2 real singular matrices (remember M is singular if det(M) = 0).
(iv) the zero matrix.
(v) the set of all 2 2 nonsingular matrices
Solution:
(i)
#
"
# "
#
a11 a12
b11 b12
a11 + b11 a12 + b12
+
=
a12 a22
b12 b22
a12 + b12 a22 + b22
"
(ii)
#
"
# "
#
a11 0
b11 0
a11 + b11
0
+
=
0 a22
0 b22
0
a22 + b22
"
{x Rn : f (x) = y} is a singleton.
(i) Is the linear function f invertible?
(ii) If f can be represented by matrix A, then show A is invertible if and only if
rank(A) = m = n.
(iii) Suppose that an n n matrix A is invertible. Does it follow that [Ax = 0] =
[x = 0]?
(iv) Suppose that A is an n n matrix and that [Ax = 0] = [x = 0]. Does it follow
that A is invertible?
Solution:
(i) Yes, the function is a bijection and so the inverse correspondence will be a
function.
(ii) () Suppose that rank(A) = m = n and assume to the contrary that A is not
invertible. Then there must exist y, x, x0 Rn with x , x0 such that y = Ax = Ax0 .
But this means that A(x x0 ) = 0 with (x x0 ) , 0 and thus the columns of A are
linearly dependent. This means there can be, at most, n 1 linearly independent
columns making the rank of A less than n - which is a contradiction.
In x = 0
x=0
which establishes the result.
(iv) Let A be an n n matrix such that Null(A) = {0}. Let u , v be two nonzero
vectors. Suppose Au = Av then Au Av = 0 and finally A(u v) = 0. But
(u v) , 0 which is impossible since only 0 is in the null space of A. This
contradiction implies that A is injective. By the rank nullity theorem we have
dim Rn = dim Null(A) + dim img(A) and since dim Null(A) = 0 we have n =
0 + dim img(A) which implies dim img(A) = n. This means the image of A ( or
of f ) fills all of Rn and A is surjective. Because it is injective and surjective A is
invertible.
Problem 10. Let y Rn be a netput vector where each element yi for i = 1, . . . , n is a
commodity. If yi < 0 then yi is an input into a production process. If yi > 0 then yi is
an output of the production process. Let F : Rn R be a transformation function and
we define the set Y = {y Rn : F(y) 0} and well call Y a technology. The technology
Y describes all the feasible production plans a firm can choose (i.e., combinations of
feasible inputs and outputs). We will assume that Y is a convex set. Now let p Rn
where p 0 be vector prices for the n commodities in any bundle y Y. Note that
for fixed p, we can define a function Tp (y) = p y for all y Y. The function Tp (y)
represents the profit (revenue minus costs) of choosing production plan y when input
and output prices are given by p. Firms want to choose some feasible production plan
y that satisfies Tp (y ) = max{p y : y Y} = (p).
(i) If (p) is the maximum achievable profit for firms given technology Y and prices
p the profit-maximizing production plans are {y Y : p y = (p)}. Now consider
the set {y Rn : p y = (p)}. Is this a hyperplane? If so, show that Y is contained
in one of the half-spaces of p y = (p) and specify which one (i.e., upper or
lower).
(ii) Does the set {y Rn : p y = (p)} touch the technology Y? i.e, is it the case
that
min |p y (p)| = 0
yY
If so, what does this say about the value of F(y ) for any y {y Y : p y = (p)}?
(iii) (Duality) Consider the sets of the form A(p) = {y Rn : p y (p)} and we
create a collection of sets {A(p) Rn : p Rn }. Show that the following equality
is true.
\
Y=
A(p)
pRn