88 Applications of vector algebra to analytic geometry

13.17.5 n. 5 (p. 496)

(a) A normal vector for the plane π through the points P ≡ (1, 1, −1), Q ≡ (3, 3, 2),
R ≡ (3, −1, −2) is
−→ −→
n ≡ P Q × QR = (2, 2, 3) × (0, −4, −4) = (4, 8, −8)

(b) A cartesian equation for π is

x + 2y − 2z = 5

(c) The distance of π from the origin is 53 .

13.17.6 n. 6 (p. 496)
Proceeding as in points (a) and (b) of the previous exercise, a normal vector for the
plane through the points P ≡ (1, 2, 3), Q ≡ (2, 3, 4), R ≡ (−1, 7, −2) is
−→ −→
n ≡ P Q × RQ = (1, 1, 1) × (3, −4, 6) = (10, −3, −7)

and a cartesian equation for it is

10x − 3y − 7z + 17 = 0

13.17.7 n. 8 (p. 496)

A normal vector to the plane is given by any direction vector for the given line

n ≡ (2, 4, 12) − (1, 2, 3) = (1, 2, 9)

A cartesian equation for the plane is

(x − 2) + 2 (y − 3) + 9 (z + 7) = 0

or

x + 2y + 9z = 55

13.17.8 n. 9 (p. 496)

A direction vector for the line is just the normal vector of the plane. Thus the
parametric equations are

x = 2 + 4h y = 1 − 3h z = −3 + h
Exercises 89

13.17.9 n. 10 (p. 496)

(a) The position of the point at time t can be written as follows:

x= 1−t y = 2 − 3t z = −1 + 2t

which are just the parametric equations of a line.

(b) The direction vector of the line L is

d = (−1, −3, 2)

(c) The hitting instant is determined by substituting the coordinates of the moving
point into the equation of the given plane π

2 (1 − t) + 3 (2 − 3t) + 2 (−1 + 2t) + 1 = 0

−7t + 7 = 0

yielding t = 1. Hence the hitting point is (0, −1, 1).

(d) At time t = 3, the moving point has coordinates (−2, −7, 5). Substituting,

2 · (−2) + 3 · (−7) + 2 · 5 + d = 0
d = 15

A cartesian equation for the plane π 0 which is parallel to π and contains (−2, −7, 5)
is

2x + 3y + 2z + 15 = 0

(e) At time t = 2, the moving point has coordinates (−1, −4, 3). A normal vector for
the plane π 00 which is orthogonal to L and contains (−1, −4, 3) is −d. Thus

−1 · −1 − 3 · −4 + 2 · 3 + d = 0
d = −19

A cartesian equation for π 00 is

(x + 1) + 3 (y + 4) − 2 (z − 3) = 0

or

x + 3y − 2z + 19 = 0
90 Applications of vector algebra to analytic geometry

13.17.10 n. 11 (p. 496)

From
ci = π
n cj = π
n nck =
π
3 4 3
we get
n 1³ √ ´
= 1, 2, 1
knk 2
A cartesian equation for the plane in consideration is
√
(x − 1) + 2 (y − 1) + (z − 1) = 0
or
√ √
x+ 2y + z = 2 + 2
13.17.11 n. 13 (p. 496)
First I find a vector of arbitrary norm which satisfies the given conditions:
l + 2m − 3n = 0
l − m + 5n = 0
¡ ¢
Assigning value 1 to n,³ the other values ´are easily obtained: (l, m) = − 73 , 83 . Then
7 8 3
the required vector is − √122 , √122 , √122
13.17.12 n. 14 (p. 496)
A normal vector for the plane π which is parallel to both vectors i + j and j + k is
n ≡ (i + j) × (j + k) = i − j + k
Since the intercept of π with the X-axis is (2, 0, 0), a cartesian equation for π is
x−y+z = 2
13.17.13 n. 15 (p. 496)
I work directly on the coefficient matrix for the equation system:
3 1 1 | 5
3 1 5 | 7
1 −1 3 | 3
With obvious manipulations
0 4 −8 | −4
0 0 4 | 2
1 −1 3 | 3
¡ ¢
I obtain a unique solution (x, y, z) = 32 , 0, 12 .
The conic sections 91

13.17.14 n. 17 (p. 497)

If the line ` under consideration is parallel to the two given planes, which have as
normal vectors n1 ≡ (1, 2, 3) and n2 ≡ (2, 3, 4), a direction vector for ` is

d ≡ (2, 3, 4) × (1, 2, 3) = (1, −2, 1)

Since ` goes through the point P ≡ (1, 2, 3), parametric eqautions for ` are the
following:

x=1+t y = 2 − 2t z =3+t

13.17.15 n. 20 (p. 497)

A cartesian equation for the plane under consideration is

2x − y + 2z + d = 0

The condition of equal distance from the point P ≡ (3, 2, −1) yields

|6 − 2 − 2 + d| |6 − 2 − 2 + 4|
=
3 3
that is,

|2 + d| = 6

The two solutions of the above equation are d1 = 4 (corresponding to the plane
already given) and d2 = −8. Thus the required equation is

2x − y + 2z = 8

13.18 The conic sections

13.19 Eccentricity of conic sections

13.20 Polar equations for conic sections

13.21 Exercises
92 Applications of vector algebra to analytic geometry

13.22 Conic sections symmetric about the origin

13.23 Cartesian equations for the conic sections

13.24 Exercises

13.25 Miscellaneous exercises on conic sections

 Chapter 14
CALCULUS OF VECTOR-VALUED FUNCTIONS
94 Calculus of vector-valued functions
 Chapter 15
LINEAR SPACES

15.1 Introduction

15.2 The definition of a linear space

15.3 Examples of linear spaces

15.4 Elementary consequences of the axioms

15.5 Exercises
15.5.1 n. 1 (p. 555)
The set of all real rational functions is a real linear space. Indeed, let P , Q, R, and
S be any four real polynomials, and let
P (x) R (x)
f : x 7→ g : x 7→
Q (x) S (x)
Then for every two real numbers α and β
αP (x) S (x) + βQ (x) R (x)
αf + βg : x 7→
Q (x) S (x)
is a well defined real rational function. This shows that the set in question is closed
with respect to the two linear space operations of function sum and function multipli-
cation by a scalar. From this the other two existence axioms (of the zero element and
of negatives) also follow as particular cases, for (α, β) = (0, 0) and for (α, β) = (0, −1)
respectively. Of course it can also be seen directly that the identically null function
x 7→ 0 is a rational function, as the quotient of the constant polynomials P : x 7→ 0
96 Linear spaces

and Q : x 7→ 1. The remaining linear space axioms are immediately seen to hold, tak-
ing into account the general properties of the operations of function sum and function
multiplication by a scalar
15.5.2 n. 2 (p. 555)
The set of all real rational functions having numerator of degree not exceeding the
degree of the denominator is a real linear space. Indeed, taking into account exercise
1, and using the same notation, it only needs to be proved that if deg P ≤ deg Q and
deg R ≤ deg S, then

deg [αP S + βQR] ≤ deg QS

This is clear, since for every two real numbers α and β

deg [αP S + βQR] ≤ max {deg P S, deg QR}

deg P S = deg P deg S ≤ deg Q deg S
deg QR = deg Q deg R ≤ deg Q deg S

so that the closure axioms hold. It may be also noticed that the degree of both
polynomials P and Q occurring in the representation of the identically null function
as a rational function is zero.
15.5.3 n. 3 (p. 555)
The set of all real valued functions which are defined on a fixed domain containing 0
and 1, and which have the same value at 0 and 1, is a real linear space. Indeed, for
every two real numbers α and β,

(αf + βg) (0) = αf (0) + βg (0) = αf (1) + βg (1) = (αf + βg) (1)

so that the closure axioms hold. Again, the two existence axioms follow as particular
cases; and it is anyway clear that the identically null function x 7→ 0 achieves the same
value at 0 and 1. Similarly, that the other linear space axioms hold is a straightforward
consequence of the general properties of the operations of function sum and function
multiplication by a scalar.
15.5.4 n. 4 (p. 555)
The set of all real valued functions which are defined on a fixed domain containing
0 and 1, and which achieve at 0 the double value they achieve at 1 is a real linear
space. Indeed, for every two real numbers α and β,

(αf + βg) (0) = αf (0) + βg (0) = α2f (1) + β2g (1) = 2 (αf + βg) (1)

so that the closure axioms hold. A final remark concerning the other axioms, of a
type which has become usual at this point, allows to conclude.
Exercises 97

15.5.5 n. 5 (p. 555)

The set of all real valued functions which are defined on a fixed domain containing
0 and 1, and which have value at 1 which exceeds the value at 0 by 1, is not a real
linear space. Indeed, let α and β be any two real numbers such that α + β 6= 1. Then

(αf + βg) (1) = αf (1) + βg (1) = α [1 + f (0)] + β [1 + g (0)]

= α + β + αf (0) + βg (0) = α + β + (αf + βg) (0)
6= 1 + (αf + βg) (0)

and the two closure axioms fail to hold (the above shows, however, that the set in
question is an affine subspace of the real linear space). The other failing axioms are:
existence of the zero element (the identically null function has the same value at 0
and at 1), and existence of negatives

f (1) = 1 + f (0) ⇒ (−f) (1) = −1 + (−f) (0) 6= 1 + (−f ) (0)

15.5.6 n. 6 (p. 555)

The set of all real valued step functions which are defined on [0, 1] is a real linear
space. Indeed, let f and g be any two such functions, so that for some nonnegative
integers m and n, some increasing (m + 1)-tuple (σ r )r∈{0}∪m of elements of [0, 1] with
σ 0 = 0 and σ m = 1, some increasing (n + 1)-tuple (τ s )s∈{0}∪n of elements of [0, 1]
with τ 0 = 0 and τ m = 1, some (m + 1)-tuple (γ r )r∈{0}∪m of real numbers, and some
n-tuple (δ s )s∈{0}∪n of real numbers∗ ,
X X
f ≡ γ 0 χ{0} + γ r χIr g ≡ δ 0 χ{0} + δ s χJs
r∈m s∈n

where for each subset C of [0, 1], χC is the characteristic function of C

¿
1 if x ∈ C
χC ≡ x 7→
0 if x ∈
/C

and (Ir )r∈m, (Js )s∈n are the partitions of (0, 1] associated to (σ r )r∈{0}∪m , (τ s )s∈{0}∪n

Ir ≡ (σ r−1 , σ r ] (r ∈ m)
Js ≡ (τ s−1 , τ s ] (s ∈ n)

Then, for any two real numbers α and β,

X
αf + βg = (αγ 0 + βδ 0 ) χ{0} + (αγ r + βδ s ) χKrs
(r,s)∈m×n

∗
I am using here the following slight abuse of notation for degenerate intervals: (σ, σ] ≡ {σ};
This is necessary in order to allow for “point steps”.
98 Linear spaces

where (Krs )(r,s)∈m×n is the “meet” partition of (Ir )r∈m and (Js )s∈n
Krs ≡ Ir ∩ Js ( (r, s) ∈ m × n)
which shows that αf + βg is a step function too, with no more† than m + n steps on
[0, 1].
Thus the closure axioms hold, and a final, usual remark concenrning the other
linear space axioms applies. It may also be noticed independently that the identically
null function [0, 1] 7→ R, x 7→ 0 is indeed a step function, with just one step (m = 1,
γ 0 = γ 1 = 0), and that the opposite of a step function is a step function too, with
the same number of steps.
15.5.7 n. 7 (p. 555)
The set of all real valued functions which are defined on R and convergent to 0 at
+∞ is a real linear space. Indeed, by classical theorems in the theory of limits, for
every two such functions f and g, and for any two real numbers α and β,
lim (αf + βg) (x) = α lim f (x) + β lim g (x)
x→+∞ x→+∞ x→+∞
= 0
so that the closure axioms hold. Final remark concerning the other linear space
axioms. It may be noticed independently that the identically null function [0, 1] 7→
R, x 7→ 0 indeed converges to 0 at +∞ (and, for that matter, at any x0 ∈ R∪{−∞}).
15.5.8 n. 11 (p. 555)
The set of all real valued and increasing functions of a real variable is not a real linear
space. The first closure axiom holds, because the sum of two increasing functions is an
increasing function too. The second closure axiom does not hold, however, because
the function αf is decreasing if f is increasing and α is a negative real number.
The axiom of existence of the zero element holds or fails, depending on whether
monotonicity is meant in the weak or in the strict sense, since the identically null
function is weakly increasing (and, for that matter, weakly decreasing too, as every
constant function) but not strictly increasing. Thus, in the former case, the set of all
real valued and (weakly) increasing functions of a real variable is a convex cone. The
axiom of existence of negatives fails, too. All the other linear space axioms hold, as
in the previous examples.
15.5.9 n. 13 (p. 555)
The set of all real valued functions which are defined and integrable on [0, 1], with
the integral over [0, 1] equal to zero, is a real linear space. Indeed, for any two such
functions f and g, and any two real numbers α and β,
Z 1 Z 1 Z 1
(αf + βg) (x) dx = α f (x) dx + β g (x) dx = 0
0 0 0
†
Some potential steps may “collapse” if, for some (r, s) ∈ m × n, σr−1 = τ s−1 and ασr + βτ s =
ασr−1 + βτ s−1 .
Exercises 99

15.5.10 n. 14 (p. 555)

The set of all real valued functions which are defined and integrable on [0, 1], with
nonnegative integral over [0, 1], is a convex cone, but not a linear space. For any two
such functions f and g, and any two nonnegative real numbers α and β,
Z 1 Z 1 Z 1
(αf + βg) (x) dx = α f (x) dx + β g (x) dx ≥ 0
0 0 0

R1 R1
It is clear that if α < 0 and 0
(αf ) (x) dx > 0, then 0
(αf ) (x) dx < 0, so that
axioms 2 and 6 fail to hold.
15.5.11 n. 16 (p. 555)
First solution. The set of all real Taylor polynomials of degree less than or equal to
n (including the zero polynomial) is a real linear space, since it coincides with the set
of all real polynomials of degree less than or equal to n, which is already known to be
a real linear space. The discussion of the above statement is a bit complicated by the
fact that nothing is said concerning the point where our Taylor polynomials are to be
centered. If the center is taken to be 0, the issue is really easy: every polynomial is
the Taylor polynomial centered at 0 of itself (considered, as it is, as a real function of
a real variable). This is immediate, if one thinks at the motivation for the definition
of Taylor polynomials: best n-degree polynomial approximation of a given function.
At any rate, if one takes the standard formula as a definition
n
X f (j) (0)
Taylorn (f ) at 0 ≡ x 7→ xj
j=0
j!

here are the computations. Let

n
X
P : x 7→ pi xn−i
i=0

Then
P
P0 : x 7→ n−1 (n − i) pi xn−1−i P 0 (0) = pn−1
Pi=0
n−2
P 00 : x→ 7 i=0 (n − i) (n − (i + 1)) pi x
n−2−i
P 00 (0) = 2pn−2
.. .. ..
. . .
P
P (j) : x 7→ n−j (n−i)!
i=0 (n−i−j)! pi x
n−j−i
P (j) (0) = j!pn−j
.. .. ..
. . .
P (n−1) : x 7→ n!p0 x + (n − 1)!p1 P (n−1) (0) = (n − 1)!p1
P (n) : x→ 7 n!p0 P (n) (0) = n!p0
100 Linear spaces

and hence
n
X P (j) (0)
n
X j!pn−j
n
X
xj = xj = pi xn−i = P (x)
j=0
j! j=0
j! i=0

On the other hand, if the Taylor polynomials are meant to be centered at some x0 6= 0
n
X f (j) (x0 )
Taylorn (f ) at x0 ≡ x 7→ (x − x0 )j
j=0
j!

it must be shown by more lengthy arguments that for each polynomial P the following
holds:
Xn
P (j) (x0 ) Xn
(x − x0 )j = pi xn−i
j=0
j! i=0

Second solution. The set of all real Taylor polynomials (centered at x0 ∈ R)

of degree less than or equal to n (including the zero polynomial) is a real linear space.
Indeed, let P and Q be any two such polynomials; that is, let f and g be two real
functions of a real variable which are m and n times differentiable in x0 , and let
P ≡ Taylorm (f ) at x0 , Q ≡ Taylorn (g) at x0 , that is,
m
X f (i) (x0 ) i
n
X g (j) (x0 )
P : x 7→ (x − x0 ) Q : x 7→ (x − x0 )j
i=0
i! j=0
j!

Suppose first that m = n. Then for any two real numbers α and β
Xm ½ · (k) ¸ · (k) ¸¾
f (x0 ) g (x0 )
αP + βQ = α +β (x − x0 )k
k! k!
k=0
Xm
(αf + βg)(k) (x0 )
= (x − x0 )k
k=0
k!
= Taylorm (αf + βg) at x0
Second, suppose (without loss of generality) that m > n. In this case, however,
n
X αf (k) (x0 ) + βg (k) (x0 ) k
m
X αf (k) (x0 )
αP + βQ = (x − x0 ) + (x − x0 )k
k! k!
k=0 k=n+1

a polynomial which can be legitimately considered the Taylor polynomial of degree

n at x0 of the function αf + βg only if all the derivatives of g of order from n + 1
to m are null at x0 . This is certainly true if g itself a polynomial of degree n. In
fact, this is true only in such a case, as it can be seen by repeated integration. It is
hence necessary, in addition, to state and prove the result asserting that each Taylor
polynomial of any degree and centered at any x0 ∈ R can be seen as the Taylor
polynomial of itself.
Exercises 101

15.5.12 n. 17 (p. 555)

The set S of all solutions of a linear second-order homogeneous differential equation
∀x ∈ (a, b) , y 00 (x) + P (x) y 0 (x) + Q (x) y (x) = 0
where P and Q are given everywhere continuous real functions of a real variable,
and (a, b) is some open interval to be determined together with the solution, is a real
linear space. First of all, it must be noticed that S is nonempty, and that its elements
are indeed real functions which are everywhere defined (that is, with (a, b) = R), due
to the main existence and solution continuation theorems in the theory of differential
equations. Second, the operator
L : D2 → RR , y 7→ y 00 + P y 0 + Qy
where RR is the set of all the real functions of a real variable, and D2 is the subset
of RR of all the functions having a second derivative which is everywhere defined, is
linear:
∀y ∈ D2 , ∀z ∈ D2 , ∀α ∈ R, ∀β ∈ R,

L (αy + βz) = (αy + βz)00 + P (αy + βz)0 + Q (αy + βz)

= α (y 00 + P y 0 + Qy) + β (z 00 + P z 0 + Qz)
= αL (y) + βL (z)
Third, D2 is a real linear space, since
∀y ∈ D2 , ∀z ∈ D2 , ∀α ∈ R, ∀β ∈ R,

αy + βz ∈ D2
by standard theorems on the derivative of a linear combination of differentiable
functions, and the usual remark concerning the other linear space axioms. Finally,
S = ker L is a linear subspace of D2 , by the following standard argument:
L (y) = 0 ∧ L (z) = 0 ⇒ L (αy + βz) = αL (y) + βL (z) = 0
and the usual remark.
15.5.13 n. 18 (p. 555)
The set of all bounded real sequences is a real linear space. Indeed, for every two real
sequences x ≡ (xn )n∈N and y ≡ (yn )n∈N and every two real numbers α and β, if x
and y are bounded, so that for some positive numbers ε and η and for every n ∈ N
the following holds:
|xn | < ε |yn | < η
then the sequence αx + βy is also bounded, since for every n ∈ N
|αxn + βyn | ≤ |α| |xn | + |β| |yn | < |α| ε + |β| η
Thus the closure axioms hold, and the usual remark concerning the other linear space
axioms applies.
102 Linear spaces

15.5.14 n. 19 (p. 555)

The set of all convergent real sequences is a real linear space. Indeed, for every two
real sequences x ≡ (xn)n∈N and y ≡ (yn )n∈N and every two real numbers α and β, if
x and y are convergent to x and y respectively, then the sequence αx + βy converges
to αx + βy. Thus the closure axioms hold. Usual remark concerning the other linear
space axioms.
15.5.15 n. 22 (p. 555)
The set U of all elements of R3 with their third component equal to 0 is a real linear
space. Indeed, by linear combination the third component remains equal to 0; U is
the kernel of the linear function R3 → R, (x, y, z) 7→ z.
15.5.16 n. 23 (p. 555)
The set W of all elements of R3 with their second or third component equal to 0 is
not a linear subspace of R3 . For example, (0, 1, 0) and (0, 0, 1) are in the set, but
their sum (0, 1, 1) is not. The second closure axiom and the existence of negatives
axiom fail too. The other axioms hold. W is not even an affine subspace, nor it is
convex; however, it is a cone.
15.5.17 n. 24 (p. 555)
The set π of all elements of R3 with their second component which is equal to the
third multiplied by 5 is a real linear space, being the kernel of the linear function
R3 → R, (x, y, z) 7→ 5x − y.
15.5.18 n. 25 (p. 555)
The set ` of all elements (x, y, z) of R3 such that 3x + 4y = 1 and z = 0 (the line
through the point P ≡ (−1, 1, 0) with direction vector v ≡ (4, −3, 0)) is an affine
subspace of R3 , hence a convex set, but not a linear subspace of R3 , hence not a
linear space itself. Indeed, for any two triples (x, y, z) and (u, v, w) of R3 , and any
two real numbers α and β

3x + 4y = 1  ½

z=0 3 (αx + βu) + 4 (αy + βv) = α + β
⇒
3u + 4v = 1  z+w =0

w=0

Thus both closure axioms fail for `. Both existence axioms also fail, since neither
the null triple, nor the opposite of any triple in `, belong to `. The associative and
distributive laws have a defective status too, since they make sense in ` only under
quite restrictive assumptions on the elements of R3 or the real numbers appearing in
them.
15.5.19 n. 26 (p. 555)
The set r of all elements (x, y, z) of R3 which are scalar multiples of (1, 2, 3) (the line
through the origin having (1, 2, 3) as direction vector) is a real linear space. For any
Subspaces of a linear space 103

two elements (h, 2h, 3h) and (k, 2k, 3k) of r, and for any two real numbers α and β,
the linear combination

α (h, 2h, 3h) + β (k, 2k, 3k) = (αh + βk) (1, 2, 3)

belongs to r. r is the kernel of the linear function R3 → R2 , (x, y, z) 7→ (2x − y, 3x − z).

15.5.20 n. 27 (p. 555)
The set of solutions of the linear homogenous system of equations

A (x, y, z)0 = 0

is the kernel of the linear function R3 → R, (x, y, z) 7→ A (x, y, z)0 and hence a real
linear space.
15.5.21 n. 28 (p. 555)
The subset of Rn of all the linear combinations of two given vectors a and b is a
vector subspace of Rn , namely span {a, b}. It is immediate to check that every linear
combination of linear combinations of a and b is again a linear combination of a and
b.

15.6 Subspaces of a linear space

15.7 Dependent and independent sets in a linear space

15.8 Bases and dimension

15.9 Exercises
15.9.1 n. 1 (p. 560)
The set S1 of all elements of R3 with their first coordinate equal to 0 is a linear
subspace of R3 (see exercise 5.22 above). S1 is the coordinate Y Z-plane, its dimension
is 2. A standard basis for it is {(0, 1, 0) , (0, 0, 1)}. It is clear that the two vectors are
linearly independent, and that for every real numbers y and z

(0, y, z) = y (0, 1, 0) + z (0, 0, 1)

104 Linear spaces

15.9.2 n. 2 (p. 560)

The set S2 of all elements of R3 with the first and second coordinate summing up to
0 is a linear subspace of R3 (S2 is the plane containing the Z-axis and the bisectrix of
the even-numbered quadrants of the XY -plane). A basis for it is {(1, −1, 0) , (0, 0, 1)};
its dimension is 2. It is clear that the two vectors are linearly independent, and that
for every three real numbers x, y and z such that x + y = 0

(x, y, z) = x (1, −1, 0) + z (0, 0, 1)

15.9.3 n. 3 (p. 560)

The set S3 of all elements of R3 with the coordinates summing up to 0 is a linear
subspace of R3 (S3 is the plane through the origin and normal vector n ≡ (1, 1, 1)).
A basis for it is {(1, −1, 0) , (0, −1, 1)}; its dimension is 2. It is clear that the two
vectors are linearly independent, and that for every three real numbers x, y and z
such that x + y + z = 0

(x, y, z) = x (1, −1, 0) + z (0, −1, 1)

15.9.4 n. 4 (p. 560)

The set S4 of all elements of R3 with the first two coordinates equal is a linear subspace
of R3 (S4 is the plane containing the Z-axis and the bisectrix of the odd-numbered
quadrants of the XY -plane). A basis for it is {(1, 1, 0) , (0, 0, 1)}; its dimension is 2.
It is clear that the two vectors are linearly independent, and that for every three real
numbers x, y and z such that x = y

(x, y, z) = x (1, 1, 0) + z (0, 0, 1)

15.9.5 n. 5 (p. 560)

The set S5 of all elements of R3 with all the coordinates equal is a linear subspace of
R3 (S5 is the line through the origin and direction vector d ≡ (1, 1, 1)). A basis for it
is {d}; its dimension is 1.
15.9.6 n. 6 (p. 560)
The set S6 of all elements of R3 with the first coordinate equal either to the second
or to the third is not a linear subspace of R3 (S6 is the union of the plane S4 of
exercise 4 and the plane containing the Y -axis and the bisectrix of the odd-numbered
quadrants of the XZ-plane). For example, (1, 1, 0) and (1, 0, 1) both belong to S6 ,
but their sum (2, 1, 1) does not.
15.9.7 n. 7 (p. 560)
The set S7 of all elements of R3 with the first and second coordinates having identical
square is not a linear subspace of R3 (S7 is the union of the two planes containing
the Z-axis and one of the bisectrices of the odd and even-numbered quadrants of the
XY -plane). For example, (1, −1, 0) and (1, 1, 0) both belong to S7 , but their sum
Exercises 105

(2, 0, 0) or their semisum (1, 0, 0) do not. S7 is not an affine subspace of R3 , and it

is not even convex. However, S7 is a cone, and it is even closed with respect to the
external product (multiplication by arbitrary real numbers).
15.9.8 n. 8 (p. 560)
The set S8 of all elements of R3 with the first and second coordinates summing up to
1 is not a linear subspace of R3 (S8 is the vertical plane containing the line through
the points P ≡ (1, 0, 0) and Q ≡ (0, 1, 0)) For example, for every α 6= 1, α (1, 0, 0) ,
and α (0, 1, 0) do not belong to S8 . S8 is an affine subspace of R3 , since

∀ (x, y, z) ∈ R3 , ∀ (u, v, w) ∈ R3 , ∀α ∈ R

(x + y = 1) ∧ (u + v = 1) ⇒ [(1 − α) x + αu] + [(1 − α) y + αv] = 1

15.9.9 n. 9 (p. 560)

See exercise 26, p.555.
15.9.10 n. 10 (p. 560)
The set S10 of all elements (x, y, z) of R3 such that

x+y+z = 0
x−y−z = 0

is a line containing the origin, a subspace of R3 of dimension 1. A base for it is

{(0, 1, −1)}. If (x, y, z) is any element of S10 , then

(x, y, z) = y (0, 1, −1)

15.9.11 n. 11 (p. 560)

The set S11 of all polynomials of degree not exceeding n, and taking value 0 at 0 is a
linear subspace of the set of all polynomials of degree not exceeding n, and hence a
linear space. If P is any polynomial of degree not exceeding n,
X
P : t 7→ p0 + ph th (15.1)
h∈n

then P belongs to S11 if and only if p0 = 0. It is clear that any linear combination of
polynomials in S11 takes value 0 at 0. A basis for S11 is
© ª
B ≡ t 7→ th h∈n

Indeed, by the general principle of polynomial identity (which is more or less explicitly
proved in example 6, p.558), a linear combination of B is the null polynomial, if and
only if all its coefficients are null. Moreover, every polynomial belonging to S11 is a
linear combination of B. It follows that dim S11 = n.
106 Linear spaces

15.9.12 n. 12 (p. 560)

The set S12 of all polynomials of degree not exceeding n, with first derivative taking
value 0 at 0, is a linear subspace of the set of all polynomials of degree not exceeding
n. Indeed, for any two suach polynomials P and Q, and any two real numbers α and
β,

(αP + βQ)0 (0) = αP 0 (0) + βQ0 (0) = α · 0 + β · 0 = 0

A polynomial P as in (15.1) belongs to S12 if and only if p1 = 0. A basis for S is

n¡ ¢ o
h+1
B ≡ t 7→ t h∈n−1
, (t 7→ 1)

That B is a basis can be seen as in the previous exercise. Thus dim S12 = n.
15.9.13 n. 13 (p. 560)
The set S13 of all polynomials of degree not exceeding n, with second derivative
taking value 0 at 0, is a linear subspace of the set of all polynomials of degree not
exceeding n. This is proved exactly as in the previous exercise (just replace 0 with 00 ).
A polynomial P as in (15.1) belongs to S13 if and only if p2 = 0. A basis for S is
n¡ ¢ o
B ≡ t 7→ th+2 h∈n−2 , (t 7→ 1) , (t 7→ t)

That B is a basis can be seen as in the exercise 11. Thus dim S13 = n.
15.9.14 n. 14 (p. 560)
The set S14 of all polynomials of degree not exceeding n, with first derivative taking
value at 0 which is the opposite of the polynomial value at 0, is a linear subspace of
the set of all polynomials of degree not exceeding n. A polynomial P as in (15.1)
belongs to S14 if and only if p0 + p1 = 0. If P and Q are any two polynomials in S,
and α and β are any two real numbers,

(αP + βQ) (0) + (αP + βQ)0 (1) = αP (0) + βQ (0) + αP 0 (0) + βQ0 (0)
= α [P (0) + P 0 (0)] + β [Q (0) + Q0 (0)]
= α·0+β·0

A basis for S14 is

n ¡ ¢ o
h+1
B ≡ t 7→ 1 − t, t 7→ t h∈n−1

Indeed, if α is the n-tuple of coefficients of a linear combination R of B, then

X
R : t 7→ α1 − α1 t + αh+1 th+1
h∈n−1
Exercises 107

By the general principle of polynomial identity, the n-tuple α of coefficients of any

linear combination of B spanning the null vector must satisfy the conditions
α1 = 0
αh+1 = 0 (h ∈ n − 1)
that is, the combination must be the trivial one. If P is any polynomial such that
p0 + p1 = 0, P belongs to span B, since by the position
α1 = p0 = −p1
αh+1 = ph+1 (h ∈ n − 1)
the linear combination of B with α as n-tuple of coefficients is equal to P . It follows
that B is a basis of S14 , and hence that dim S = n.
15.9.15 n. 15 (p. 560)
The set S15 of all polynomials of degree not exceeding n, and taking the same value
at 0 and at 1 is a linear subspace of the set of all polynomials of degree not exceeding
n, and hence a linear space. This can be seen exactly as in exercise 3, p.555. A
polynomial P belongs to S15 if and only if
X
p0 = p0 + ph
h∈n

that is, if and only if

X
ph = 0
h∈n

A basis for S15 is

n ¡ ¢ o
B ≡ (t 7→ 1) , t 7→ (1 − t) th h∈n−1

and the dimension of S15 is n.

15.9.16 n. 16 (p. 560)
The set S16 of all polynomials of degree not exceeding n, and taking the same value
at 0 and at 2 is a linear subspace of the set of all polynomials of degree not exceeding
n, and hence a linear space. This can be seen exactly as in exercise 3, p.555. A
polynomial P belongs to S16 if and only if
X
p0 = p0 + 2h ph
h∈n

that is, if and only if

X
2h ph = 0
h∈n
108 Linear spaces

A basis for S16 is

n ¡ ¢ o
B ≡ (t 7→ 1) , t 7→ (2 − t) th h∈n−1

and the dimension of S16 is n.

15.9.17 n. 22 (p. 560)
(b) It has alredy been shown in the notes that
\
lin S ≡ W
W subspace of V
S⊆W
( )
X
= v : ∃n ∈ N, ∃u ≡ (ui )i∈n ∈ V n , ∃a ≡ (αi )i∈n ∈ F n , v = αi ui
i∈n

hence if T is a subspace, and it contains S, then T contains lin S.

(c) If S = lin S, then of course S is a subspace of V because lin S is so. Conversely, if
S is a subspace of V , then S is one among the subspaces appearing in the definition
of lin S, and hence lin S ⊆ S; since S ⊆ lin S always, it follows that S = lin S.
(e) If S and T are subspaces of V , then by point c above lin S = S and lin T = T ,
and hence

S ∩ T = lin S ∩ lin T

Since the intersection of any family of subspaces is a subspace, S ∩ T is a subspace

of V .
(g) Let V ≡ R2 , S ≡ {(1, 0) , (0, 1)}, T ≡ {(1, 0) , (1, 1)}. Then

S ∩ T = {(1, 0)}
L (S) = L (T ) = R2
L (S ∩ T ) = {(x, y) : y = 0}

15.9.18 n. 23 (p. 560)

(a) Let

f : R → R, x 7→ 1
g : R → R, x 7→ eax
h : R → R, x 7→ ebx
0 : R → R, x 7→ 0

and let (u, v, w) ∈ R3 be such that

uf + vg + wh = 0
Exercises 109

In particular, for x = 0, x = 1, and x = 2, we have

u+v+w = 0
u + ea v + eb w = 0 (15.2)
u + e2a v + e2b w = 0

The determinant of the coefficient matrix of the above system is

¯ ¯ ¯ ¯
¯ 1 1 1 ¯¯ ¯ 1 1 1 ¯
¯ ¯ ¯
¯ 1 ea eb ¯¯ = ¯¯ 0 ea − 1 eb − 1 ¯¯
¯
¯ 1 e2a e2b ¯ ¯ 0 e2a − 1 e2b − 1 ¯
¡ ¢£ ¤
= (ea − 1) eb − 1 eb + 1 − (ea + 1)
¡ ¢¡ ¢
= (ea − 1) eb − 1 eb − ea

Since by assumption a 6= b, at most one of the two numbers can be equal to zero. If
none of them is null, system (15.2) has only the trivial solution, the triple (f, g, h)
is linearly independent, and dim lin {f, g, h} = 3. If a = 0, then g = f ; similarly, if
b = 0, then h = f . Thus if either a or b is null then (f, g, h) is linearly dependent (it
suffices to take (u, v, w) ≡ (1, −1, 0) or (u, v, w) ≡ (1, 0, −1), respectively).
It is convenient to state and prove the following (very) simple

Lemma 4 Let X be a set containing at least two distinct elements, and let f : X → R
be a nonnull constant function. For any function g : X → R, the couple (f, g) is
linearly dependent if and only if g is constant, too.

Proof. If g is constant, let Im f ≡ {yf } and Im g ≡ {yg }. Then the function

yg f − yf g is the null function. Conversely, if (f, g) is linearly dependent, let (u, v) ∈
R2 ∼ (0, 0) be such that uf + vg is the null function. Thus

∀x ∈ X, uyf + vg (x) = 0

Since f is nonnull, yf 6= 0. This yields

v=0⇒u=0

and hence v 6= 0. Then

u
∀x ∈ X, g (x) = − yf
v
and g is constant, too.
It immediately follows from the above lemma (and from the assumption a 6= b)
that if either a = 0 or b = 0, then dim lin {f, g, h} = 2.
(b) The two functions

f : x 7→ eax g : x 7→ xeax
110 Linear spaces

are linearly independent, since

[∀x ∈ R, αex + βxeax = 0] ⇔ [∀x ∈ R, (α + βx) eax = 0]

⇔ [∀x ∈ R, (α + βx) = 0]
⇔ α=β=0

Notice that the argument holds even in the case a = 0, Thus dim lin {f, g} = 2.
(c) Arguing as in point a, let (u, v, w) ∈ R3 be such that

uf + vg + wh = 0

where

f : x 7→ 1 g : x 7→ eax h : x 7→ xeax

In particular, for x = 0, x = 1, and x = −1, we have

u+v+w = 0
u + ea v + ea w = 0 (15.3)
u + e−a v − e−a w = 0

a homogeneous system of equations whose coefficient matrix has determinant

¯ ¯ ¯ ¯
¯ 1 1 1 ¯ ¯ 1 1 1 ¯
¯ ¯ ¯ ¯
¯ 1 e a
e ¯¯ = ¯¯ 0 e − 1
a a
e − 1 ¯¯
a
¯
¯ 1 e−a −e−a ¯ ¯ 0 e−a − 1 −e−a − 1 ¯
¡ ¢
= − (ea − 1) e−a + 1 + e−a − 1
= 2e−a (1 − ea )

Thus if a 6= 0 the above determinant is different from zero, yielding (u, v, w) =

(0, 0, 0). The triple (f, g, h) is linearly independent, and dim lin {f, g, h} = 3. On the
other hand, if a = 0 then f = g and h = IdR , so that dim lin {f, g, h} = 2.
(f) The two functions

f : x 7→ sin x g : x 7→ cos x

are linearly independent. Indeed, for every two real numbers α and β which are not
both null,

∀x ∈ R, α sin x + β cos x = 0
m
∀x ∈ R, √ α2 2 sin x + √ β2 2 cos x = 0
α +β α +β
m
∀x ∈ R, sin (γ + x) = 0
Inner products. Euclidean spaces. Norms 111

where

α
γ ≡ sign β arccos p
α2 + β 2

since the sin function is not identically null, the last condition cannot hold. Thus
dim lin {x 7→ sin x, x 7→ cos x} = 2.
(h) From the trigonometric addition formulas

∀x ∈ R cos 2x = cos2 x − sin2 x = 1 − 2 sin2 x

Let then

f : x 7→ 1 g : x 7→ cos x g : x 7→ sin2 x

The triple (f, g, h) is linearly dependent, since

f − g + 2h = 0

By the lemma discussed at point a, dim lin {f, g, h} = 2.

15.10 Inner products. Euclidean spaces. Norms

15.11 Orthogonality in a euclidean space

112 Linear spaces

15.12 Exercises

15.12.1 n. 9 (p. 567)

Z ¯1
1
t2 ¯¯ 1 1
hu1 , u2 i = t dt = ¯ = − = 0
−1 2 −1 2 2
Z 1 ¯1 µ ¶
2 t3 ¯¯ 1 1 2
hu2 , u3 i = t + t dt = 0 + ¯ = − − =
−1 3 −1 3 3 3
Z 1
hu3 , u1 i = 1 + t dt = t|1−1 + 0 = 1 − (−1) = 2
−1
Z 1 √
ku1 k2 = 1 dt = 2 ku1 k = 2
−1
Z 1 r
2 2
ku2 k2 = t2 dt = ku2 k =
−1 3 3
Z 1 r
2 8
ku3 k2 = 1 + 2t + t2 dt = 2 + ku3 k =
−1 3 3
2
√
1 3 2 3
2 u3 = q q =
cos u[ 1 u3 = √ q =
cos u[
2 8 2 2 83 2
3 3
π π
2 u3 =
u[ 1 u3 =
u[
3 6

15.12.2 n. 11 (p. 567)

(a) Let for each n ∈ N

Z +∞
In ≡ e−t tn dt
0

Then

Z +∞ Z x
−t
I0 = e dt = lim e−t dt
0 x→+∞ 0
¯
−t ¯x
= lim −e 0 = lim − e−x + 1
x→+∞ x→+∞
= 1
Exercises 113

and, for each n ∈ N,

Z +∞ Z x
−t n+1
In+1 = e t dt = lim e−t tn+1 dt
0 x→+∞ 0
½ Z x ¾
¯
−t n+1 ¯x −t n
= lim −e t 0
+ (n + 1) e t dt
x→+∞ 0
½Z x ¾
© −x n+1 ª −t n
= lim −e x − 0 + (n + 1) lim e t dt
x→+∞ x→+∞ 0
Z +∞
= 0 + (n + 1) e−t tn dt
0
= (n + 1) In

The integral involved in the definition of In is always convergent, since for each n ∈ N

lim e−x xn = 0
x→+∞

It follows that

∀n ∈ N, In = n!

Let now

X X
f : t 7→ α0 + αi ti g : t 7→ β 0 + β j tj
i∈m i∈n

be two real polynomials, of degree m and n respectively. Then the product f g is a

real polynomial of degree m + n, containing for each k ∈ m + n a monomial of degree
k of the form αi ti β j tj = αi β j ti+j whenever i + j = k. Thus

 
X  X  k
f g = α0 β 0 +  α β 
j t
 i
k∈m+n i∈m, j∈n
i+j=k
114 Linear spaces

and the scalar product of f and g is results in the sum of m + n + 1 converging

integrals
   
Z +∞ X  X
  k
hf, gi = e−t 
α 0 β 0 + 
 αi β j  
 t  dt
0 k∈m+n i∈m, j∈n
i+j=k
 
Z X  X Z
+∞  +∞
= α0 β 0 −t
e dt +  αi β j  e−t tk dt
 
0 k∈m+n i∈m, j∈n 0
i+j=k
 
X  X 
= α0 β 0 I0 +  α β  Ik
 i j 
k∈m+n i∈m, j∈n
i+j=k
 
X  X 
= α0 β 0 +  αi β j 
  k!
k∈m+n i∈m, j∈n
i+j=k

(b) If for each n ∈ N

xn : t 7→ tn

then for each m ∈ N and each n ∈ N

Z +∞
hxm , xn i = e−t tm tn dt
0
= Im+n = (m + n)!

(c) If

f : t 7→ (t + 1)2 g : t 7→ t2 + 1

then

fg t 7→ t4 + 2t3 + 2t2 + 2t + 1
:
Z +∞
¡ ¢
hf, gi = e−t t4 + 2t3 + 2t2 + 2t + 1 dt
0
= I4 + 2I3 + 2I2 + 2I1 + I0
= 4! + 2 · 3! + 2 · 2! + 2 · 1! + 1
= 43
Construction of orthogonal sets. The Gram-Schmidt process 115

(d) If

f : t 7→ t + 1 g : t 7→ αt + β

then

f g : t 7→ αt2 + (α + β) t + β
hf, gi = αI2 + (α + β) I1 + βI0
= 3α + 2β

and

hf, gi = 0 ⇔ (α, β) = (2γ, −3γ) (γ ∈ R)

that is, the set of all polynomials of degree less than or equal to 1 which are orthogonal
to f is

P1 ∩ perp {f } = {gγ : t 7→ 2γt − 3γ}γ∈R

15.13 Construction of orthogonal sets. The Gram-Schmidt process

15.14 Orthogonal complements. projections

15.15 Best approximation of elements in a euclidean space by elements

in a finite-dimensional subspace

15.16 Exercises
15.16.1 n. 1 (p. 576)
(a) By direct inspection of the coordinates of the three given vectors. it is seen that
they all belong to the plane π of equation x−z = 0. Since no two of them are parallel,
lin {x1 , x2 , x3 } = π and dim lin {x1 , x2 , x3 } = 2. A unit vector in π is given by

1
v≡ (2, 1, 2)
3
Every vector belonging to the line π ∩ perp {v} must have the form

(x, −4x, x)
116 Linear spaces
√
with norm 3 2 |x|. Thus a second unit vector which together with v spans π, and
which is orthogonal to v, is
√
2
w≡ (1, −4, 1)
6
The required orthonormal basis for lin {x1 , x2 , x3 } is {v, w}.
(b) The answer is identical to the one just given for case a.
15.16.2 n. 2 (p. 576)
(a) First solution. It is easily checked that

x2 ∈
/ lin {x1 }
x3 ∈
/ lin {x1 , x2 }
0 = x1 − x2 + x3 − x4

Thus dim W ≡ dim lin {x1 , x2 , x3 , x4 } = 3, and three vectors are required for any
basis of W . The following vectors form an orthonormal one:
√
x1 2
y1 ≡ = (1, 1, 0, 0)
kx1 k 2
√ √
2 2
x2 − hx2 , y1 i y1 (0, 1, 1, 0) − (1, 1, 0, 0)
y2 ≡ =°°
2 2
√ √ °
°
kx2 − hx2 , y1 i y1 k °(0, 1, 1, 0) − 2 2
(1, 1, 0, 0)°
2 2
¡ 1 1 ¢ √
− , , 1, 0 6
= q2 2 = (−1, 1, 2, 0)
1
+ 1
+ 1 6
4 4
√ √
x3 − hx3 , y1 i y1 − − hx3 , y2 i y2 (0, 0, 1, 1) − 0 − 2 66 66 (−1, 1, 2, 0)
y3 ≡ = ° √ √ °
kx3 − hx3 , y1 i y1 − − hx3 , y2 i y2 k ° °
°(0, 0, 1, 1) − 0 − 2 66 66 (−1, 1, 2, 0)°
¡1 1 1 ¢ √
, − , , 1 3
= q3 3 3 = (1, −1, 1, 3)
1
+ 1
+ 1
+ 1 6
9 9 9

Second solution. More easily, it suffices to realize that

© ª
lin {x1 , x2 , x3 , x4 } = H(1,−1,1,−1),0 ≡ (x, y, z, t) ∈ R4 : x − y + z − t = 0

is the hyperplane of R4 through the origin, with (1, −1, 1, −1) as normal vector, to
directly exhibit two mutually orthogonal unit vectors in lin {x1 , x2 , x3 , x4 }
√ √
2 2
u≡ (1, 1, 0, 0) v≡ (0, 0, 1, 1)
2 2
Exercises 117

It remains to find a unit vector in H(1,−1,1,−1),0 ∩ perp {u, v}. The following equations
characterize H(1,−1,1,−1),0 ∩ perp {u, v}:
x−y+z−t = 0
x+y = 0
z+t = 0
yielding (by addition) 2 (x + z) = 0, and hence
1 1
w≡ (1, −1, −1, 1) or w ≡ − (1, −1, −1, 1)
2 2
Thus an orthonormal basis for lin {x1 , x2 , x3 , x4 } is {u, v, w}.
(b) It is easily checked that
x2 ∈
/ lin {x1 }
0 = 2x1 − x2 − x3
Thus dim W ≡ dim lin {x1 , x2 , x3 } = 2, and two vectors are required for any basis of
W . The following vectors form an orthonormal one:
√
x1 3
y1 ≡ = (1, 1, 0, 1)
kx1 k 3
√ √
x2 − hx2 , y1 i y1 (1, 0, 2, 1) − 2 33 33 (1, 1, 0, 1)
y2 ≡ =°° √ √ °
°
kx2 − hx2 , y1 i y1 k °(0, 1, 1, 0) − 2 33 33 (1, 1, 0, 1)°
¡1 2 1
¢ √
, − , 2, 42
= q3 3 3
= (1, −2, 6, 1)
1
+ 4
+ 4 + 1 42
9 9 9

15.16.3 n. 3 (p. 576)

Z π
1 1 1
hy0 , y0 i = √ √ dt = π = 1
0 π π π
Z πr r Z
2 2 2 π
hyn , yn i = cos nt cos nt dt = cos2 nt dt
0 π π π 0
Let
Z π Z π
2
In ≡ cos ntdt = cos nt cos nt dt
0 0
integrating by parts
¯π Z π
sin nt ¯¯ sin nt
In = cos nt ¯ − −n sin nt dt
n 0 0 n
Z π Z π
2
= sin nt dt = 1 − cos2 nt dt
0 0
= π − In
118 Linear spaces

Thus
π
In =
2
hyn , yn i = 1
and every function yn has norm equal to one.
To check mutual orthogonality, let us compute
Z π ¯π
1 1 sin nt ¯¯
hy0 , yn i = √ cos nt dt = √ =0
0 π π n ¯0
Z π
hym , yn i = cos mt cos nt dt
0
¯ Z π
sin nt ¯¯π sin nt
= cos mt ¯ − −m sin mt dt
n 0 0 n
Z
m π
= sin mt sin nt dt
n 0
µ ¶¯π Z µ ¶
m cos nt ¯¯ m π cos nt
= sin mt − ¯ − n m cos mt − dt
n n 0 0 n
³ m ´2
= hym , yn i
n
¡ ¢2
Since m and n are distinct positive integers, m n
is different from 1, and the equation
³ m ´2
hym , yn i = hym , yn i
n
can hold only if hym , yn i = 0.
That the set {yn }+∞ +∞
n=0 generates the same space generated by the set {xn }n=0
is trivial, since, for each n, yn is a multiple of xn .
15.16.4 n. 4 (p. 576)
We have
Z 1
hy0 , y0 i = 1dt = 1
0
Z 1 Ã ¯1 !
¡ 2 ¢ 4 3 ¯
hy1 , y1 i = 3 4t − 4t + 1 dt = 3 t − 2t2 + t¯¯
0 3 0
= 1Z
1 ¡ ¢
hy2 , y2 i = 5 36t4 − 72t3 + 48t2 − 12t + 1 dt
0
à ¯1 !
36 5 ¯
= 5 t − 18t4 + 16t3 − 6t2 + t¯¯
5 0
= 1
Exercises 119

which proves that the three given functions are unit vectors with respect to the given
inner product.
Moreover,
Z 1√ √ ³ ¯1 ´
hy0 , y1 i = 3 (2t − 1) dt = 3 t2 − t¯0 = 0
0
Z 1√ √ ³
¡ ¢ ¯1 ´
hy0 , y2 i = 5 6t2 − 6t + 1 dt = 5 2t3 − 3t2 + t¯0 = 0
Z0 1 √
¡ ¢
hy1 , y2 i = 15 12t3 − 18t2 + 8t − 1 dt
0
√ ³ 4 ¯1 ´
= 15 3t − 6t + 4t − t¯0 = 0
3 2

which proves that the three given functions are mutually orthogonal. Thus {y1 , y2 , y3 }
is an orthonormal set, and hence linearly independent.
Finally,
√ Ã √ ! √ √
1 3 5 3 5
x0 = y0 x1 = y0 + y1 x2 = 1 − y1 + y1 + y2
2 2 30 3 30

which proves that

lin {y1 , y2 , y3 } = lin {x1 , x2 , x3 }

120 Linear spaces
 Chapter 16
LINEAR TRANSFORMATIONS AND MATRICES

16.1 Linear transformations

16.2 Null space and range

16.3 Nullity and rank

16.4 Exercises
16.4.1 n. 1 (p. 582)
T is linear, since for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R, for
every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αy + βv, αx + βu)
= α (y, x) + β (v, u)
= αT [(x, y)] + βT [(u, v)]

The null space of T is the trivial subspace {(0, 0)}, the range of T is R2 , and hence

rank T = 2 nullity T = 0

T is the orthogonal symmetry with respect to the bisectrix r of the first and third
quadrant, since, for every (x, y) ∈ R2 , T (x, y) − (x, y) is orthogonal to r, and the
midpoint of [(x, y) , T (x, y)], which is (x + y, x + y), lies on r.
122 Linear transformations and matrices

16.4.2 n. 2 (p. 582)

T is linear, since for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R, and for
every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αy + βv, αx + βu)
= α (y, x) + β (v, u)
= αT [(x, y)] + βT [(u, v)]

The null space of T is the trivial subspace {(0, 0)}, the range of T is R2 , and hence

rank T = 2 nullity T = 0

T is the orthogonal symmetry with respect to the X-axis.

16.4.3 n. 3 (p. 582)
T is linear , since for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R, and
for every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αx + βu, 0)
= α (x, 0) + β (u, 0)
= αT [(x, y)] + βT [(u, v)]

The null space of T is the Y -axis, the range of T is the X-axis, and hence

rank T = 1 nullity T = 1

T is the orthogonal projection on the X-axis.

16.4.4 n. 4 (p. 582)
T is linear, since for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R, and for
every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αx + βu, αx + βu)
= α (x, x) + β (u, u)
= αT [(x, y)] + βT [(u, v)]

The null space of T is the Y -axis, the range of T is the bisectrix of the I and III
quadrant, and hence

rank T = nullity T = 1
Exercises 123

16.4.5 n. 5 (p. 582)

T is not a linear function, since, e.g., for x and u both different from 0,
¡ ¢
T (x + u, 0) = x2 + 2xy + u2 , 0
¡ ¢
T (x, 0) + T (u, 0) = x2 + u2 , 0

16.4.6 n. 6 (p. 582)

T is not linear; indeed, for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R,
for every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

¡ ¢
= eαx+βu , eαy+βv
h i
= (ex )α · (eu )β , (ey )α · (ev )β
αT [(x, y)] + βT [(u, v)] = α (ex , ey ) + β (eu , ev )
= (αex + βeu , αey + βev )

so that, e.g., when x = y = 0 and u = v = α = β = 1,

T [(0, 0) + (1, 1)] = (e, e)

T [(0, 0)] + T [(1, 1)] = (1 + e, 1 + e)

16.4.7 n. 7 (p. 582)

T is not an affine function, but it is not linear. Indeed, for every (x, y) ∈ R2 , for
every (u, v) ∈ R2 , for every α ∈ R, and for every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αx + βu, 1)
αT [(x, y)] + βT [(u, v)] = α (x, 1) + β (u, 1)
= (αx + βu, α + β)

16.4.8 n. 8 (p. 582)

T is affine, but not linear; indeed, for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for
every α ∈ R, and for every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αx + βu + 1, αy + βv + 1)
αT [(x, y)] + βT [(u, v)] = α (x + 1, y + 1) + β (u + 1, v + 1)
= (αx + βu + α + β, αy + βv + α + β)
124 Linear transformations and matrices

16.4.9 n. 9 (p. 582)

T is linear, since for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R, and for
every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (αx + βu − αy − βv, αx + βu + αy + βv)
= α (x − y, x + y) + β (u − v, u + v)
= αT [(x, y)] + βT [(u, v)]

The null space of T is the trivial subspace {(0, 0)}, the range of T is R2 , and hence

rank T = 2 nullity T = 0

The matrix representing T is

· ¸ " √ √ #
1 −1 √ 2
− 2
A ≡ = 2 √2 √2
1 1 2 2
2 2
· √ ¸· π
¸
2 √0 cos 4
− sin π4
= π
0 2 sin 4
cos π4

Thus T is the composition of a counterclockwise rotation by an angle of π4 with a

√
homothety of modulus 2.
16.4.10 n. 10 (p. 582)
T is linear, since for every (x, y) ∈ R2 , for every (u, v) ∈ R2 , for every α ∈ R, and for
every β ∈ R,

T [α (x, y) + β (u, v)] = T [(αx + βu, αy + βv)]

= (2 (αx + βu) − (αy + βv) , (αx + βu) + (αy + βv))
= α (2x − y, x + y) + β (2u − v, u + v)
= αT [(x, y)] + βT [(u, v)]

The null space of T is the trivial subspace {(0, 0)}, the range of T is R2 , and hence

rank T = 2 nullity T = 0

The matrix representing T is

· ¸
2 −1
A≡
1 1

the characteristic polynomial of A is

λ2 − 3λ + 3
Exercises 125

and the eigenvalues of A are

√ √
3 + 3i 3 − 3i
λ1 ≡ λ2 ≡
2 2
An eigenvector associated to λ1 is
µ ¶
2√
z≡
1 − 3i
The matrix representing T with respect to the basis
½µ ¶ µ ¶¾
0
√ 2
{Im z, Re z} = ,
− 3 1
is
" √ # " √ #
3
− 3 √ 3
− 1
B ≡ √2 2 = 3 2 √2
3 3 1 3
2 2 2 2
· √ ¸· ¸
3 √0 cos π6 − sin π
6
=
0 3 sin π6 cos π6
Thus T is the composition of a counterclockwise rotation by an angle of π6 with a
√
homothety of modulus 3.
16.4.11 n. 16 (p. 582)
T is linear, since for every (x, y, z) ∈ R3 , for every (u, v, w) ∈ R3 , for every α ∈ R,
for every β ∈ R,
T [α (x, y, z) + β (u, v, w)] = T [(αx + βu, αy + βv, αz + βw)]
= (αz + βw, αy + βv, αx + βu)
= α (z, y, x) + β (w, v, u)
= αT [(x, y, z)] + βT [(u, v, w)]
The null space of T is the trivial subspace {(0, 0, 0)}, the range of T is R3 , and hence
rank T = 3 nark T = 0
16.4.12 n. 17 (p. 582)
T is linear (as every projection on a coordinate hyperplane), since for every (x, y, z) ∈
R3 , for every (u, v, w) ∈ R3 , for every α ∈ R, for every β ∈ R,
T [α (x, y, z) + β (u, v, w)] = T [(αx + βu, αy + βv, αz + βw)]
= (αx + βu, αy + βv, 0)
= α (x, y, 0) + β (u, v, 0)
= αT [(x, y, z)] + βT [(u, v, w)]
The null space of T is the Z-axis, the range of T is the XY -plane, and hence
rank T = 2 nark T = 1
126 Linear transformations and matrices

16.4.13 n. 23 (p. 582)

T is linear, since for every (x, y, z) ∈ R3 , for every (u, v, w) ∈ R3 , for every α ∈ R,
for every β ∈ R,

T [α (x, y, z) + β (u, v, w)] = T [(αx + βu, αy + βv, αz + βw)]

= (αx + βu + αz + βw, 0, αx + βu + αy + βv)
= [α (x + z) , 0, α (x + y)] + [β (u + w) , 0, β (u + v)]
= α (x, y, z) + β (u, v, w)
= αT [(x, y, z)] + βT [(u, v, w)]

The null space of T is the axis of central symmetry of the (+, −, −)-orthant and of
the (−, +, +)-orthant, of parametric equations

x=t y = −t z = −t

the range of T is the XZ-plane, and hence

rank T = 2 nark T = 1

16.4.14 n. 25 (p. 582)

Let D1 (−1, 1) or, more shortly, D1 be the linear space of all real functions of a real
variable which are defined and everywhere differentiable on (−1, 1). If

T : D1 → R(−1,1) , f 7→ (x 7→ xf 0 (x))

then T is a linear operator. Indeed,

¡ ¢
T (f + g) = x 7→ x [f + g]0 (x)
= (x 7→ x [f 0 (x) + g 0 (x)])
= (x 7→ xf 0 (x) + xg 0 (x))
= (x 7→ xf 0 (x)) + (x 7→ +xg 0 (x))
= T (f ) + T (g)

Moreover,

ker T = {f ∈ D1 : ∀x ∈ (−1, 1) , xf 0 (x) = 0}

= {f ∈ D1 : ∀x ∈ (−1, 0) ∪ (0, 1) , f 0 (x) = 0}

By an important though sometimes overlooked theorem in calculus, every function f

which is differentiable in (−1, 0) ∪ (0, 1) and continuous in 0, is differentiable in 0 as
well, provided lim f 0 (x) exists and is finite, in which case
x→0

f 0 (0) = lim f 0 (x)

x→0
Exercises 127

Thus
ker T = {f ∈ D1 : f 0 = 0}
(here 0 is the identically null function defined on (−1, 1)). If f belongs to ker T , by
the classical theorem due to Lagrange,
∀x ∈ (−1, 1) , ∃ϑx ∈ (0, x)

f (x) = f (0) + xf 0 (ϑx )

and hence f is constant on (−1, 1). It follows that a basis of ker T is given by the
constant function (x 7→ 1), and nullity T = 1. Since T (D1 ) contains, e.g., all the
restrictions to (−1, 1) of the polynomials with vanishing zero-degree monomial, and
it is already known that the linear space of all polynomials has an infinite basis, the
dimension of T (D1 ) is infinite.
16.4.15 n. 27 (p. 582)
Let D2 be the linear space of all real functions of a real variable which are defined
and everywhere differentiable on R. If
L : D2 → RR , y 7→ y 00 + P y 0 + Q
where P and Q are real functions of a real variable which are continuous on R, it has
been shown in the solution to exercise 17, p.555, that L is a linear operator. Thus
ker L is the set of all solutions to the linear differential equation of the second order
∀x ∈ R, y 00 (x) + P (x) y 0 (x) + Q (x) y (x) = 0 (16.1)
By the uniqueness theorem for Cauchy’s problems in the theory of differential equa-
tions, for each (y0 , y00 ) ∈ R2 there exists a unique solution to equation (16.1) such
that y (0) = y0 and y 0 (0) = y00 . Hence the function
µ ¶
2 y (0)
ϕ : ker L → R , y 7→
y 0 (0)
is injective and surjective. Moreover, let u be the solution to equation (16.1) such
that (u (0) , u0 (0)) = (1, 0), and let v be the solution to equation (16.1) such that
(v (0) , v 0 (0)) = (0, 1). Then for each (y0 , y00 ) ∈ R2 , since ker L is a linear subspace of
D2 , the function
y : x 7→ y0 u (x) + y00 v (x)
is a solution to equation (16.1), and by direct inspection it is seen that y (0) = y0 and
y 0 (0) = y00 . In other words, ϕ−1 ((y0 , y00 )), and
ker L = span {u, v}
Finally, u and v are linearly independent; indeed, since u (0) = 1 and v (0) = 0,
αu (x) + βv (x) = 0 for each x ∈ R can only hold (by evaluating at x = 0) if α = 0;
in such a case, from βv (x) = 0 for each x ∈ R and v 0 (0) = 1 it is easily deduced that
β = 0 as well. Thus nullity L = 2.
128 Linear transformations and matrices

16.5 Algebraic operations on linear transformations

16.6 Inverses

16.7 One-to-one linear transformations

16.8 Exercises
16.8.1 n. 15 (p. 589)
T is injective (or one to one), since

 x = x0
0 0 0
T (x, y, z) = T (x , y , z ) ⇔ 2y = 2y 0 ⇔ (x, y, z) = (x0 , y 0 , z 0 )

3z = 3z 0

³ v w´
T −1 (u, v, w) = u, ,
2 3
16.8.2 n. 16 (p. 589)
T is injective, since

 x = x0
0 0 0
T (x, y, z) = T (x , y , z ) ⇔ y = y0 ⇔ (x, y, z) = (x0 , y 0 , z 0 )

x + y + z = x0 + y 0 + z 0

T −1 (u, v, w) = (u, v, w − u − v)

16.8.3 n. 17 (p. 589)

T is injective, since

 x + 1 = x0 + 1
0 0 0
T (x, y, z) = T (x , y , z ) ⇔ y + 1 = y 0 + 1 ⇔ (x, y, z) = (x0 , y 0 , z 0 )

z − 1 = z0 − 1

T −1 (u, v, w) = (u − 1, v − 1, w + 1)
Linear transformations with prescribed values 129

16.8.4 n. 27 (p. 590)

Let
p = x 7→ p0 + p1 x + p2 x2 + · · · + pn−1 xn−1 + pn xn

· Z x ¸
DT (p) = D [T (p)] = D x 7→ p (t) dt
0
= x 7→ p (x)
the last equality being a consequence of the fundamental theorem of integral calculus.
£ ¤
T D (p) = T [D (p)] = T x 7→ p1 + 2p2 x + · · · (n − 1) pn−1 xn−2 + npn xn−1
Z x
= x 7→ p1 + 2p2 t + · · · (n − 1) pn−1 tn−2 + npn tn−1 dt
0
¯x
= x 7→ p1 t + p2 t2 + · · · pn−1 tn−1 + pn tn ¯0
= x 7→ p1 x + p2 x2 + · · · pn−1 xn−1 + pn xn
= p − p0
Thus T D acts as the identity map only on the subspace W of V containing all
polynomials having the zero degree monomial (p0 ) equal to zero. Im T D is equal to
W , and ker T D is the set of all constant polynomials, i.e., zero-degree polynomials.

16.9 Linear transformations with prescribed values

16.10 Matrix representations of linear transformations

16.11 Construction of a matrix representation in diagonal form

16.12 Exercises
16.12.1 n. 3 (p. 596)
(a) Since T (i) = i + j and T (j) = 2i − j,
T (3i − 4j) = 3T (i) − 4T (j) = 3 (i + j) − 4 (2i − j)
= −5i + 7j
2
T (3i − 4j) = T (−5i + 7j) = −5T (i) + 7T (j)
= −5 (i + j) + 7 (2i − j) = 9i − 12j
130 Linear transformations and matrices

(b) The matrix of T with respect to the basis {i, j} is

· ¸
£ ¤ 1 2
A ≡ T (i) T (j) =
1 −1
and the matrix of T 2 with respect to the same basis is
· ¸
2 3 0
A ≡
0 3
(c) First solution (matrix for T ). If e1 = i − j and e2 = 3i + j, the matrix
· ¸
£ ¤ 1 3
P ≡ e1 e2 =
−1 1
transforms the (canonical) coordinates (1, 0) and (0, 1) of e1 and e2 with respect to
the basis {e1 , e2 } into their coordinates (1, −1) and (3, 1) with respect to the basis
{i, j}; hence P is the matrix of coordinate change from basis {e1 , e2 } to basis {i, j},
and P −1 is the matrix of coordinate change from basis {i, j} to basis {e1 , e2 }. The
operation of the matrix B representing T with respect to the basis {e1 , e2 } can be
described as the combined effect of the following three actions: 1) coordinate change
from coordinates w.r. to basis {e1 , e2 } into coordinates w.r. to the basis {i, j} (that
is, multiplication by matrix P ); 2) transformation according to T as expressed by
the matrix representing T w.r. to the basis {i, j} (that is, multiplication by A); 3)
coordinate change from coordinates w.r. to the basis {i, j} into coordinates w.r. to
basis {e1 , e2 } (that is, multiplication by P −1 ). Thus
· ¸· ¸· ¸
−1 1 1 −3 1 2 1 3
B = P AP =
4 1 1 1 −1 −1 1
· ¸· ¸
1 −2 5 1 3
=
4 2 1 −1 1
· ¸
1 −7 −1
=
4 1 7
Second solution (matrix for T ). The matrix B representing T with respect to the
basis {e1 , e2 } is
£ ¤
B ≡ T (e1 ) T (e2 )
provided T (e1 ) and T (e2 ) are meant as coordinate vectors (α, β) and (γ, δ) with
respect to the basis {e1 , e2 }. Since, on the other hand, with respect to the basis {i, j}
we have
T (e1 ) = T (i − j) = T (i) − T (j) = (i + j) − (2i − j)
= −i + 2j
T (e2 ) = T (3i + j) = 3T (i) + T (j) = 3 (i + j) + (2i − j)
= 5i + 2j
Exercises 131

it suffices to find the {e1 , e2 }-coordinates (α, β) and (γ, δ) which correspond to the
{i, j}-coordinates (−1, 2) and (5, 2). Thus we want to solve the two equation systems
(in the unknowns (α, β) and (γ, δ), respectively)

αe1 + βe2 = −i + 2j
γe1 + δe2 = 5i + 2j

that is,
½ ½
α + 3β = −1 γ + 3δ = 5
−α + β = 2 −γ + δ = 2

The (unique) solutions are (α, β) = 14 (−7, 1) and (γ, δ) = 14 (−1, 7), so that
· ¸
1 −7 −1
B=
7 1 7

Continuation (matrix for T 2 ). Since T 2 is represented by a scalar diagonal matrix

with respect to the initially given basis, it is represented by the same matrix with
respect to every basis (indeed, since scalar diagonal matrices commute with every
matrix of the same order, P −1 DP = D for every scalar diagonal matrix D).
16.12.2 n. 4 (p. 596)

T : (x, y) (reflection w.r. to the Y -axis) 7→ (−x, y)

(length doubling) 7 → (−2x, 2y)

Thus T may be represented by the matrix

µ ¶
−2 0
AT ≡
0 2

and hence T 2 by the matrix

µ ¶
4 0
A2T ≡
0 4

16.12.3 n. 5 (p. 596)

a)

T (i + 2j + 3k) = T (k) + T (j + k) + T (i + j + k)
= (2i + 3j + 5k) + i + (j − k)
= 3i + 4j + 4k
132 Linear transformations and matrices

The three image vectors T (k), T (j + k), T (i + j + k) form a linearly independent

triple. Indeed, the linear combination

xT (k) + yT (j + k) + zT (i + j + k) = x (2i + 3j + 5k) + yi + z (j − k)

= (2x + y) i + (3x + z) j + (5x − z) k

spans the null vector if and only if

2x + y = 0
3x + z = 0
5x − z = 0

which yields (II + III) x = 0, and hence (by substitution in I and II) y = z = 0.
Thus the range space of T is R3 , and rank T is 3. It follows that the null space of T
is the trivial subspace {(0, 0, 0)}, and nark T is 0.
b) The matrix of T with respect to the basis {i, j, k} is obtained by aligning as
columns the coordinates w.r. to {i, j, k} of the image vectors T (i), T (j), T (k). The
last one is given in the problem statement, but the first two need to be computed.

T (j) = T (j + k − k) = T (j + k) − T (k)
= −i − 3j − 5k
T (i) = T (i + j + k − j − k) = T (i + j + k) − T (j + k)
= −i + j − k

 
−1 −1 2
AT ≡  1 −3 3 
−1 −5 3

16.12.4 n. 7 (p. 597)

(a)

T (4i − j + k) = 4T (i) − T (j) + T (k) = (0, −2)

Since {T (j) , T (k)} is a linearly independent set,

ker T = {0} rank T = 2

(b)

µ ¶
0 1 1
AT =
0 1 −1
Exercises 133

(c)
Let C = (w1 , w2 )
T (i) = 0w1 + 0w2
T (j) = 1w1 + 0w2
1 3
T (k) = − w1 + w2
2 2
µ ¶
0 1 − 12
(AT )C =
0 0 32
(d)
Let B ≡ {j, k, i} and C = {w1 , w2 } = {(1, 1) , (1, −1)}. Then
µ ¶
1 0 0
(AT )B,C =
0 1 0
16.12.5 n. 8 (p. 597)
2 3
(a) I shall distinguish the canonical unit vectors
¡ ¢ of R and R by marking the former
with an underbar. Since T (i) = i + k and T j = −i + k,
¡ ¢ ¡¢
T 2i − 3j = 2T (i) − 3T j = 2 (i + k) − 3 (i − k)
= −i + 5k
For any two real numbers x and y,
¡ ¢ ¡¢
T xi + yj = xT (i) + yT j = x (i + k) + y (i − k)
= (x + y) i + (x − y) k
It follows that
Im T = lin {i, k}
rank T = 2
nullity T = 2 − 2 = 0
© ª
(b) The matrix of T with respect to the bases i, j and {i, j, k} is
 
£ ¡¢ ¤ 1 1
A ≡ T (i) T j = 0 0 
1 −1
(c) First solution. If w1 = i + j and w2 = i + 2j, the matrix
· ¸
£ ¤ 1 1
P ≡ w1 w2 =
1 2
134 Linear transformations and matrices

transforms the (canonical) coordinates (1, 0) and (0, 1) of w1 and w2 with respect to
the ª
© basis {w1 , w2 } into their coordinates (1, 1) and (1, 2) with respect to the©basis
ª
i, j ; hence P is the matrix of coordinate change from basis {w1 , w2 } to basis i, j ,
© ª
and P −1 is the matrix of coordinate change from basis i, j to basis {w1 , w2 }. The
operation of the matrix B of T with respect to the bases {w1 , w2 } and {i, j, k} can be
described as the combined effect of the following two actions: 1) coordinate © ªchange
from coordinates w.r. to basis {w1 , w2 } into coordinates w.r. to the basis i, j (that
is, multiplication by matrix P ); 2) transformation
© ª according to T as expressed by the
matrix representing T w.r. to the bases i, j and {i, j, k} (that is, multiplication by
A). Thus
 
1 1 · ¸
1 1
B = AP =  0 0 
1 2
1 −1
 
2 3
=  0 0 
0 −1
Second solution (matrix for T ). The matrix B representing T with respect to the
bases {w1 , w2 } and {i, j, k} is
£ ¤
B ≡ T (w1 ) T (w2 )
where
¡ ¢ ¡¢
T (w1 ) = T i + j = T (i) + T j = (i + k) + (i − k)
= i
¡ ¢ ¡¢
T (w2 ) = T i + 2j = T (i) + 2T j = (i + k) + 2 (i − k)
= 3i − k
Thus
 
1 3
B= 0 0 
0 −1
(c) The matrix C representing T w.r. to bases {u1 , u2 } and {v1 , v2 , v3 } is diagonal,
that is,
 
γ 11 0
C =  0 γ 22 
0 0
if and only if the following holds
T (u1 ) = γ 11 v1 T (u2 ) = γ 22 v2
Exercises 135

There are indeed very many ways to achieve that. In the present situation the simplest
way seems to me to be given by defining

u1 ≡ i v1 ≡ T (u1 ) = i + k
u2 ≡ j v2 ≡ T (u2 ) = i − k
v3 ≡ any vector such that {v1 , v2 , v3 } is lin. independent

thereby obtaining
 
1 0
C = 0 1 
0 0

16.12.6 n. 16 (p. 597)

We have
D (sin) = cos D2 (sin) = D (cos) = − sin
D (cos) = − sin D2 (cos) = D (− sin) = − cos
D (id · sin) = sin + id · cos D2 (id · sin) = D (sin + id · cos) = 2 cos − id · sin
D (id · cos) = cos − id · sin D2 (id · cos) = D (cos − id · sin) = −2 sin − id · cos

and hence the matrix representing the differentiation operator D and its square D2
acting on lin {sin, cos, id · sin, id · cos} with respect to the basis {sin, cos, id · sin, id · cos}
is
   
0 −1 1 0 −1 0 0 −2
 1 0 0 1   0 −1 2 0 
A=  0 0 0 −1 
 A2 =  0

0 −1 0 
0 0 1 0 0 0 0 −1