Scribd
Upload
57 views

EC2R Answer Key

The document contains an answer key for a test with multiple questions. Question 1 asks students to provide explanations and derivations for various concepts related to eigenvalues and eigenvectors. Question 2 involves calculating derivatives and determining properties of matrices. Question 3 covers topics like quadratic forms, singular value decompositions, and gradient calculations.

Uploaded by

srinivasa.reddy1
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
57 views

EC2R Answer Key

The document contains an answer key for a test with multiple questions. Question 1 asks students to provide explanations and derivations for various concepts related to eigenvalues and eigenvectors. Question 2 involves calculating derivatives and determining properties of matrices. Question 3 covers topics like quadratic forms, singular value decompositions, and gradient calculations.

Uploaded by

srinivasa.reddy1
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Birla Institute of Technology and Science, Pilani

Work Integrated Learning Programmes Division


Answer Key

Q1 Expected solutions
a) Since the given property is satisfied, it means that x = [1 1 .... 1]T is a
vector satisfying Ax = 0. Hence zero is one of the eigenvalues. (1 mark)
Since zero is one of the eigenvalues, it can be concluded that determi-
nant=0 (1 mark)
Finally, its clear from Ax = 0 that an eigenpair is (0, x) where x =
[1 1 .... 1]T ( 1 mark).

b) Derivation of first row (1 mark)


Derivation of second row (1 mark)
Derivation of third row (2 marks)
Hence the final matrix L is
 
6.4807 0 0
L = 4.9377 3.1015 0 
5.7092 0.9059 1.2586
Students are expected to provide derivation of each entry by constructing
6 equations.

c) If matrix has only nonzero eigenvalues


√ , then determinant is not zero
2
This means 2 − β ̸= 0. So β ̸= 2 (1.5 marks)
Similarly, one necessary
√ condition
√ for all positive eigenvalues is 2−β 2 > 0
In summary − 2 < β < 2 (1.5 Marks)
Q2 Expected solutions (1 + 1 + 0.5)
a)
 1

∂g u
= 2u exp(u2 )
∂u cos(u)
∂f
= 2xT
∂x
∂f ◦ g 1
= 2( log(u) + 2u exp(2u2 ) + sin(u) cos(u))
∂u u

b) Given: A ∈ Rn×n is real symmetric matrix.


Now x ∈ N (A) ⇒ Ax = 0 = 0x. If x ̸= 0 ⇒ x is an eigenvector
corresponding to 0 eigenvalue and hence eigenspace EA (0) = N (A). So
dimEA (0) = dimN (A) = n − r where r =rank(A).
Since, geometric multiplicity is same as algebraic multiplicity for sym-
metric matrices, number of zero eigenvalues is n − r and hence number
of nonzero eigenvalues is equal to r. (1 mark)
T
Clearly A = bb is a rank 1 matrix by looking into the REF(A). Hence
from i), we can say that the two eigenvalues are 0 and nonzero eigenvalue
1
is equal to trace of A = x2 + y 2 + z 2 where b = [x, y, z]T . Putting x = 1
we get the required answer. (2 marks)

c) Now  
k 2 2k −1 0
(A ) = A =
0 −1 − δ
where δ > 0. Let Ak = B ⇒ A2k = B2 So, we have
 
2 −1 0
(B ) =
0 −1 − δ
where δ > 
0. 
a b
If B = then
c d
 
2 a2 + bc (a + d)b
(B ) =
(a + d)c bc + d2
Equating terms we get (a + d)b = (a + d)c = 0. If (a + d) = 0 ⇒
a = −d and hence δ = 0 which is a contradiction. So then we have
b = c = 0 which implies a2 , d2 are negative and hence not real which is
a contradiction.
Thus no such B and hence no such real A exists
(2.5 marks)

d) (1 + 1)  
2x1 + 2 cos(x2 ) α − 2x1 sin(x2 )
i) ∇x f =
β − 2x2 sin(x1 ) 2x2 + 2 cos(x1 )
 
2 α
∇x f (0, 0) =
β 2
 
2 α
ii) ConsiderA =
β 2
Then [x, y]A[x, y]T > 0 ⇒ 2x2 + 2y 2 + (α + β)xy > 0, ∀[x, y] ̸= [0, 0]
If x or y is equal to 0, clearly the other must be non zero and the
above term will be positive. If x and y both are non zero then
y = rx for some real r and the above expression becomes
2x2 (r2 + α+β
2
r + 1) > 0 ∀r ∈ R
2 α+β
⇒ (r + 2 r + 1) > 0 ∀r ∈ R
This is possible only when (r2 + α+β 2
r + 1) ̸= 0 ∀r ∈ R
q
(α+β)2 2
That means (1/2)(− α+β 2
± 4
− 4) is not real ⇒ (α+β)
4
−4 < 0
2
⇒ 0 ≤ (α + β) < 16. Further if A is symmetry then α = β and
this will imply −2 < α < 2.

Q3 Expected solutions
 
T 2 −1
a (i) The given quadratic form may be written as x Ax where A = .
−1 3
(1 mark)
This is a symmetric matrix, so it has real eigenvalues and a set of eigen-
vectors which form an orthonormal basis. We can express any vector
x ∈ R2 as x = c1 v1 + c2 v2 where v1 and v2 are orthonormal eigenvectors
of A. Then xT Ax = c21 λ1 + c22 λ2 , subject to c21 + c22 = 1 where λ1 ≥ λ2 .
The maximum
√
value
√
of xT Ax equals λ1 . The eigenvalues√of the matrix
A are 5+2 5 and 5−2 5 , so the answer to the problem is 5+2 5 . (2 marks)

a (ii) Since A is a symmetric matrix, the eigendecomposition of A is SΛS T .


(0.5 mark)
The SVD is U ΣV T where V is the eigenvector matrix of AT A = A2 .
The eigenvector matrix of A2 is also S. Thus V = S. Since S is
a matrix of orthonormal eigenvectors we know that SS T = I. The
eigenvectors ui can be obtained as Av σi
i
which means that we can write
−1 T −1 −1
U = AV Σ = SΛS V Σ = SΛΣ = S. The last step in the chain
of equalities is justified since the singular values in Σ are the square-roots
of the eigenvalues of A2 and the eigenvalues of A2 are the squares of the
eigenvalues of A. Therefore the singular values are just the eigenvalues
of A. Thus ΛΣ−1 = I, and U = S. Thus the SVD of A is SΛS T . (1.5
marks)
b (i) Since D is the Euclidean distance from the origin of the point [x1 , x2 , x3 ],
it is a function of x1 , x2 and x3 we have the gradient
dD x1 x2 x3
=p 2 ,p 2 ,p 2
dx 2
x1 + x2 + x3 2 2 2
x1 + x2 + x3 x1 + x22 + x23
. (1 mark)
Then we can calculate
dD ∂D dx1 ∂D dx2 ∂D dx3
= + +
dt ∂x1 dt ∂x2 dt ∂x3 dt
dx1
, where dt
= ωcos(ωt), dx
dt
2
= −ωrsin(ωt), dx
dt
3
= k. (1 mark)

b (ii) We can write g(x, y) = f (x0 , y0 ) + ∂f |


∂x x0 ,y0
(x − x0 ) + ∂f |
∂y x0 ,y0
(y − y0 ) using
the Taylor’s series expansion. Substituting for the partial derivatives we
have g(x, y) = αx20 + βy02 + 2αx0 (x − x0 ) + 2β(y − y0 ). (1 mark)
We shall now study the condition under which f (x, y) ≥ g(x, y). We
have αx2 + βy 2 − [αx20 + βy02 + 2αx0 (x − x0 ) + 2β(y − y0 )] ≥ 0. This can
be rearranged to
α(x2 − 2x0 (x − x0 ) − x20 ] + β(y 2 − 2y0 (y − y0 ) − y02 ) ≥ 0
α(x2 − 2xx0 + x20 ) + β(y 2 − 2y0 y + y02 ) ≥ 0
α(x − x0 )2 + β(y − y0 )2 ≥ 0
From the last inequality we see that α ≥ 0, β ≥ 0 will ensure that
f (x, y) ≥ g(x, y) for all x, y. (2 marks)

You might also like