Assignments Problem

Download as pdf or txt
Download as pdf or txt
You are on page 1of 3

Indian Institute of Science Banglore

Department of Computational and Data Sciences (CDS)


DS284: Numerical Linear Algebra
Assignment 1 [Posted Aug 20, 2022]
Faculty Instructor: Dr. Phani Motamarri
TAs: Ashish Rout, Dibya Nayak, Gourab Panigrahi, Nikhil Kodali

Submission Deadline: Sept. 4, 2022 23:59 hrs Max Points: 100

Notations: Vectors and matrices are denoted below by bold-faced lower case and upper
case alphabets, respectively.

Problem 1 [30 marks]


Let A ∈ Rm×n and B ∈ Rn×p , R(.) denotes the range of the matrix, N (.) denotes the
null space of a given matrix, dim(.) denotes the dimension of a vector space, then prove
the following:
(a) dim[R(AB)] ≤ dim[R(A)] [3 marks]
(b) If the matrix B is non-singular then dim[R(AB)] = dim[R(A)] [3 marks]
(c) dim[N (AB)] ≤ dim[N (A)] + dim[N (B)] [4 marks]
(d) dim[R(A)] + dim[N (A)] = n [4 marks]
(e) rank(A) + rank(B) − n ≤ rank(AB) ≤ min(rank(A), rank(B)) [5 marks]
Hint: Use the result in (a)
(f) Given a vector u ∈ Rn , rank (uuT ) is 1. [5 marks]
Hint: Use the result in (a)
(g) Row rank always equals column rank. [6 marks]

Problem 2 [10 marks]


Suppose there always exists a set of real coefficients c1 , c2 , c3 , ...c10 for any set of real
numbers d1 , d2 , d3 , ...d10
10
X
cj fj (i) = di for i ∈ {1, 2, ...10}
j=1

where f1 , f2 , f3 , ...f10 are a set of functions defined on the interval [1,10]


(a) Use the concepts discussed in class to show that d1 , d2 , d3 , ...d10 determine c1 , c2 , c3 , ...c10
uniquely.
[6 marks]

(a) Let A be a 10×10 matrix representing the linear mapping from data d1 , d2 , d3 , ...d10
to coefficients c1 , c2 , c3 , ...c10 . What is the i,j th entry of A−1 ? [4 marks]

Problem 3 [15 marks]


A matrix S is said to be symmetric if ST = S and skew-symmetric if ST = −S. Now
verify the following:

(a) The matrix Q = (I − S)−1 (I + S) is an orthogonal matrix for any skew-symmetric


matrix S. [3 marks]

(b) Note that a symmetric matrix A ∈ Rm×m can be decomposed as QDQT where Q
is an orthogonal matrix and D is a diagonal matrix. Using this result, show that
uT Au = 0 ∀ u ∈ Rm , if and only if A = 0. [5 marks]

(c) Show that “uT Su = 0 ∀ u ∈ Rm , if and only if S is a skew-symmetric matrix.”


[7 marks]

Problem 4 [35 marks]


If x ∈ Rm and A ∈ Rm×n , then show the following.

(a) ∥x∥∞ ≤ ∥x∥2 [5 marks]



(b) ∥x∥2 ≤ m∥x∥∞ [5 marks]

(c) ∥A∥∞ ≤ n∥A∥2 [5 marks]

(d) ∥A∥2 ≤ m∥A∥∞ [5 marks]
q
(e) ∥A∥F = tr(AT A) [5 marks]

(f) √1 ∥A∥1 ≤ ∥A∥2 ≤ n∥A∥1 [5 marks]
m
p
(g) ∥A∥2 ≤ ∥A∥1 ∥A∥∞ [5 marks]

Give an example of a non-zero vector or matrix for which equality is achieved in the
above inequalities.
Problem 5 [10 marks]
Induced matrix norm is defined as ∥A∥(m,n) = max ∥Ax∥(m) , where x ∈ Rn and is a unit
x
vector. ∥.∥ corresponds to p-norm (1 ≤ p < ∞). For this exercise, let us consider p to
be a natural number.
Using MATLAB/Octave/Python programming environment, create a matrix using
“A = randn(100, 2)”. Subsequently, create random unit vectors x using “temp =
randn(2, 1)” and normalize x using “x = temp/norm(temp)”. Check for multiple
random vectors x (use a loop, and check for about 1000 random vectors x) using
“norm of Ax = norm(Ax, p)” for p = 1, 2, 3, 4, 5, 6, ∞. What is the maximum value of
p-norm for the vector Ax? Now calculate p-norm of A using “norm of A = norm(A, p)”
for p = 1, 2, ∞ within the same programming environment you used before. Verify the
equality ∥A∥(m,n) = max ∥Ax∥(m) for p = 1, 2, ∞. Note that this equality is true for
x
other values of p as well but you are restricting to p = 1, 2, ∞ in this exercise.

You might also like