Fall 2018 CIS 515

Fundamentals of Linear Algebra and Optimization

Jean Gallier
Homework 1
September 5, 2018; Due September 19, 2018

Problem B1 (30 pts). (i) Prove that the axioms of vector spaces imply that

α·0=0
0·v =0
α · (−v) = −(α · v)
(−α) · v = −(α · v),

for all v ∈ E and all α ∈ K, where E is a vector space over K.

(ii) For every λ ∈ R and every x = (x1 , . . . , xn ) ∈ Rn , define λx by

λx = λ(x1 , . . . , xn ) = (λx1 , . . . , λxn ).

Recall that every vector x = (x1 , . . . , xn ) ∈ Rn can be written uniquely as

x = x1 e1 + · · · + xn en ,

where ei = (0, . . . , 0, 1, 0, . . . , 0), with a single 1 in position i. For any operation · : R × Rn →

Rn , if · satisfies the axiom (V1) of a vector space, then prove that for any α ∈ R, we have

α · x = α · (x1 e1 + · · · + xn en ) = α · (x1 e1 ) + · · · + α · (xn en ).

Conclude that · is completely determined by its action on the one-dimensional subspaces of

Rn spanned by e1 , . . . , en .
(iii) Use (ii) to define operations · : R × Rn → Rn that satisfy the axioms (V1–V3), but
for which axiom V4 fails.
(iv) Extra credit (20 pts). For any operation · : R × Rn → Rn , prove that if · satisfies
the axioms (V2–V3), then for every rational number r ∈ Q and every vector x ∈ Rn , we
have
r · x = r(1 · x).
In the above equation, 1 · x is some vector (y1 , . . . , yn ) ∈ Rn not necessarily equal to x =
(x1 , . . . , xn ), and
r(1 · x) = (ry1 , . . . , ryn ),

1
as in part (ii).
Use (iv) to conclude that any operation · : Q×Rn → Rn that satisfies the axioms (V1–V3)
is completely determined by the action of 1 on the one-dimensional subspaces of Rn spanned
by e1 , . . . , en .
Problem B2 (45 pts). (In solving this problem, do not use determinants). (1) Let
(u1 , . . . , um ) and (v1 , . . . , vm ) be two families of vectors in some vector space E. Assume that
each vi is a linear combination of the uj s, so that

vi = ai 1 u1 + · · · + ai m um , 1 ≤ i ≤ m,

and that the matrix A = (ai j ) is an upper-triangular matrix, which means that if 1 ≤ j <
i ≤ m, then ai j = 0. Prove that if (u1 , . . . , um ) are linearly independent and if all the
diagonal entries of A are nonzero, then (v1 , . . . , vm ) are also linearly independent.
Hint. Use induction on m.
(2) Let A = (ai j ) be an upper-triangular matrix. Prove that if all the diagonal entries of
A are nonzero, then A is invertible and the inverse A−1 of A is also upper-triangular.
Hint. Use induction on m.
Prove that if A is invertible, then all the diagonal entries of A are nonzero (do not use
determinants or eigenvalues!).
(3) Prove that if the families (u1 , . . . , um ) and (v1 , . . . , vm ) are related as in (1), then
(u1 , . . . , um ) are linearly independent iff (v1 , . . . , vm ) are.
Problem B3 (40 pts). (In solving this problem, do not use determinants). Consider
the n × n matrix  
1 2 0 0 ... 0 0
0 1 2 0 . . . 0 0
 
0 0 1 2 . . . 0 0
 
A =  ... .. . . . . . . .. ..  .

. . . . . .
 
0 0 . . . 0 1 2 0
 
0 0 . . . 0 0 1 2
0 0 ... 0 0 0 1

(1) Find the solution x = (x1 , . . . , xn ) of the linear system

Ax = b,

for  
b1
 b2 
b =  ..  .
 
.
bn

2
 (2) Prove that the matrix A is invertible and find its inverse A−1 . Given that the number
of atoms in the universe is estimated to be ≤ 1082 , compare the size of the coefficients the
inverse of A to 1082 , if n ≥ 300.
(3) Prove that (A − I)n = 0.
Problem B4 (80 pts). Consider the polynomials

B02 (t) = (1 − t)2 B12 (t) = 2(1 − t)t B22 (t) = t2

B03 (t) = (1 − t)3 B13 (t) = 3(1 − t)2 t B23 (t) = 3(1 − t)t2 B33 (t) = t3 ,

known as the Bernstein polynomials of degree 2 and 3.

(1) Show that the Bernstein polynomials B02 (t), B12 (t), B22 (t) are expressed as linear com-
binations of the basis (1, t, t2 ) of the vector space of polynomials of degree at most 2 as
follows:  2    
B0 (t) 1 −2 1 1
B12 (t) = 0 2 −2  t  .
B22 (t) 0 0 1 t2
Prove that
B02 (t) + B12 (t) + B22 (t) = 1.

(2) Show that the Bernstein polynomials B03 (t), B13 (t), B23 (t), B33 (t) are expressed as linear
combinations of the basis (1, t, t2 , t3 ) of the vector space of polynomials of degree at most 3
as follows:  3    
B0 (t) 1 −3 3 −1 1
B13 (t) 0 3 −6 3   t 
 3 =  .
B2 (t) 0 0 3 −3 t2 
B33 (t) 0 0 0 1 t3
Prove that
B03 (t) + B13 (t) + B23 (t) + B33 (t) = 1.

(3) Prove that the Bernstein polynomials of degree 2 are linearly independent, and that
the Bernstein polynomials of degree 3 are linearly independent.
(4) Recall that the binomial coefficient m


k
is given by
 
m m!
= ,
k k!(m − k)!

with 0 ≤ k ≤ m.
For any m ≥ 1, we have the m + 1 Bernstein polynomials of degree m given by
 
m m
Bk (t) = (1 − t)m−k tk , 0 ≤ k ≤ m.
k

3
Prove that m   
X m j j
Bkm (t) = (−1) j−k
t. (∗)
j=k
j k
m
Use the above to prove that B0m (t), . . . , Bm (t) are linearly independent.
(5) Prove that
m
B0m (t) + · · · + Bm (t) = 1.

Extra credit (20 pts). What can you say about the symmetries of the (m + 1) × (m + 1)
m
matrix expressing B0m , . . . , Bm in terms of the basis 1, t, . . . , tm ?
Prove your claim (beware that in equation (∗) the coefficient of tj in Bkm is the entry on
the (k+1)th row of the (j +1)th column, since 0 ≤ k, j ≤ m. Make appropriate modifications
to the indices).
What can you say about the sum of the entries on each row of the above matrix? What
about the sum of the entries on each column?
(6) (This is no longer for extra credit!) The purpose of this question is to express
m
the ti in terms of the Bernstein polynomials B0m (t), . . . , Bm (t), with 0 ≤ i ≤ m.
First, prove that
m−i
X
ti = ti Bjm−i (t), 0 ≤ i ≤ m.
j=0

Then prove that      

m m−i m i+j
= .
i j i+j i
Use the above facts to prove that
m−i
X i+j


i i
m
t = m Bi+j (t).
j=0 i

Conclude that the Bernstein polynomials B0m (t), . . . , Bm

m
(t) form a basis of the vector
space of polynomials of degree ≤ m.
Compute the matrix expressing 1, t, t2 in terms of B02 (t), B12 (t), B22 (t), and the matrix
expressing 1, t, t2 , t3 in terms of B03 (t), B13 (t), B23 (t), B33 (t).
You should find  
1 1 1
0 1/2 1
0 0 1

4
and  
1 1 1 1
0 1/3 2/3 1
0 0 1/3 1 .
 

0 0 0 1

(7) A
 polynomial
 curve C(t) of degree m in the plane is the set of points
x(t)
C(t) = given by two polynomials of degree ≤ m,
y(t)

x(t) = α0 tm1 + α1 tm1 −1 + · · · + αm1

y(t) = β0 tm2 + β1 tm2 −1 + · · · + βm2 ,

with 1 ≤ m1 , m2 ≤ m and α0 , β0 6= 0.
Prove that there exist m + 1 points b0 , . . . , bm ∈ R2 so that
 
x(t)
C(t) = = B0m (t)b0 + B1m (t)b1 + · · · + Bm
m
(t)bm
y(t)

for all t ∈ R, with C(0) = b0 and C(1) = bm . Are the points b1 , . . . , bm−1 generally on the
curve?
We say that the curve C is a Bézier curve and (b0 , . . . , bm ) is the list of control points of
the curve (control points need not be distinct).

Remark: Because B0m (t) + · · · + Bm m

(t) = 1 and Bim (t) ≥ 0 when t ∈ [0, 1], the curve
segment C[0, 1] corresponding to t ∈ [0, 1] belongs to the convex hull of the control points.
This is an important property of Bézier curves which is used in geometric modeling to
find the intersection of curve segments. Bézier curves play an important role in computer
graphics and geometric modeling, but also in robotics because they can be used to model
the trajectories of moving objects.
Problem B5 (40 pts). (a) Let A be an n × n matrix. If A is invertible, prove that for any
x ∈ Rn , if Ax = 0, then x = 0.
The converse is true: If for all x ∈ Rn , Ax = 0 implies that x = 0, then A is invertible.
We will prove this fact later, and you may use it without proof in part (b) of this problem.
(b) Let A be an m × n matrix and let B be an n × m matrix. Prove that Im − AB is
invertible iff In − BA is invertible.

5
Problem B6 (40 pts). Consider the following n × n matrix, for n ≥ 3:
 
1 −1 −1 −1 · · · −1 −1
1 −1 1 1 ··· 1 1
 
1 1 −1 1 · · · 1 1
 
B=
1 1 1 −1 · · · 1 1 

 .. .. .. .. .. .. .. 
. . . . . . . 
 
1 1 1 1 · · · −1 1 
1 1 1 1 · · · 1 −1
(1) If we denote the columns of B by b1 , . . . , bn , prove that
(n − 3)b1 − (b2 + · · · + bn ) = 2(n − 2)e1
b1 − b2 = 2(e1 + e2 )
b1 − b3 = 2(e1 + e3 )
.. ..
. .
b1 − bn = 2(e1 + en ),
where e1 , . . . , en are the canonical basis vectors of Rn .
(2) Prove that B is invertible and that its inverse A = (aij ) is given by
(n − 3) 1
a11 = , ai1 = − 2≤i≤n
2(n − 2) 2(n − 2)
and
(n − 3)
aii = − , 2≤i≤n
2(n − 2)
1
aji = , 2 ≤ i ≤ n, j =
6 i.
2(n − 2)

(3) Show that the n diagonal n × n matrices Di defined such that the diagonal entries of
Di are equal the entries (from top down) of the ith column of B form a basis of the space of
n × n diagonal matrices (matrices with zeros everywhere except possibly on the diagonal).
For example, when n = 4, we have
   
1 0 0 0 −1 0 0 0
0 1 0 0  0 −1 0 0
D1 = 0 0 1 0
 D 2 = 
0
,
0 1 0
0 0 0 1 0 0 0 1
   
−1 0 0 0 −1 0 0 0
 0 1 0 0 0 1 0 0
D3 =  0 0 −1 0 ,
 D4 =  0 0 1 0 .


0 0 0 1 0 0 0 −1

6
Problem B7 (30 pts). Let H be the set of 3 × 3 upper triangular matrices given by
  
 1 a b 
H = 0 1 c | a, b, c ∈ R .

0 0 1
 

(1) Prove that H with the binary operation of matrix multiplication is a group; find
explicitly the inverse of every matrix in H. Is H abelian (commutative)?
(2) Given two groups G1 and G2 , recall that a homomorphism if a function ϕ : G1 → G2
such that
ϕ(ab) = ϕ(a)ϕ(b), a, b ∈ G1 .
Prove that ϕ(e1 ) = e2 (where ei is the identity element of Gi ) and that
ϕ(a−1 ) = (ϕ(a))−1 , a ∈ G1 .
(3) Let S 1 be the unit circle, that is
S 1 = {eiθ = cos θ + i sin θ | 0 ≤ θ < 2π},
and let ϕ be the function given by
 
1 a b
ϕ 0 1 c  = (a, c, eib ).
0 0 1
Prove that ϕ is a surjective function onto G = R × R × S 1 , and that if we define
multiplication on this set by
(x1 , y1 , u1 ) · (x2 , y2 , u2 ) = (x1 + x2 , y1 + y2 , eix1 y2 u1 u2 ),
then G is a group and ϕ is a group homomorphism from H onto G.
(4) Extra credit (10 pts). The kernel of a homomorphism ϕ : G1 → G2 is defined as
Ker (ϕ) = {a ∈ G1 | ϕ(a) = e2 }.
Find explicitly the kernel of ϕ and show that it is a subgroup of H.
Problem B8 (10 pts). For any m ∈ Z with m > 0, the subset mZ = {mk | k ∈ Z} is an
abelian subgroup of Z. Check this.
(1) Give a group isomorphism (an invertible homomorphism) from mZ to Z.
(2) Check that the inclusion map i : mZ → Z given by i(mk) = mk is a group homomor-
phism. Prove that if m ≥ 2 then there is no group homomorphism p : Z → mZ such that
p ◦ i = id.

Remark: The above shows that abelian groups fail to have some of the properties of vector
spaces. We will show later that a linear map satisfying the condition p ◦ i = id always exists.

TOTAL: 315 + 50 points.