Scribd
Upload
11 views2 pages

HW 2025 3

The document outlines a homework assignment for a Numerical Linear Algebra course, focusing on complex rotation, Householder reflectors, and QR factorization. It includes problems related to Cholesky decomposition, orthogonal matrices, complex rotators, and reflectors, as well as algorithms for QR factorization. The assignment requires students to demonstrate understanding through proofs and algorithm development.

Uploaded by

bansallitesh
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
11 views2 pages

HW 2025 3

The document outlines a homework assignment for a Numerical Linear Algebra course, focusing on complex rotation, Householder reflectors, and QR factorization. It includes problems related to Cholesky decomposition, orthogonal matrices, complex rotators, and reflectors, as well as algorithms for QR factorization. The assignment requires students to demonstrate understanding through proofs and algorithm development.

Uploaded by

bansallitesh
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/ 2

Homework - 3

MA571 : Numerical Linear Algebra 2025 R. Alam

Complex rotation, complex Householder reflector, and QR factorization

1. Let A ∈ Rn×n be SPD with Cholesky decomposition > n


 A>=  v ∈ R . Explain why
 GG . Let
G R
A+vv > has a unique Cholesky decomposition. Let > = Q be a QR factorization,
v 0
where R ∈ Rn×n is upper triangular. Explain why R is nonsingular. Show that R can be
chosen to have positive diagonal entries. Assuming that R has positive diagonal entries,
deduce the Cholseky decomposition of A + vv > .
2. If Q ∈ R2×2 is orthogonal then show that there exists θ ∈ [0, 2π) such that either
   
cos θ − sin θ cos θ sin θ
Q= or Q = .
sin θ cos θ sin θ − cos θ
Obviously, Q is a rotation in the first case. Show that, in the second case, Q is a
Householder reflector in R2 which reflects a vector through the line y = tan(θ/2)x.
Determine a unit Householder vector u and show that Q = I − 2uu> . Conclude that a
2 × 2 orthogonal matrix is either a rotation or a reflection.
3. 
Complex
 rotator: This problem is about howto construct
 a complex rotator.
  Given
 
f 2 c −s̄ ∗ f r
∈ C , determine c, s and r such that Q = is unitary and Q = .
g s c̄ g 0
Such a matrix Q is called a complex rotator. Show that Q is not uniquely determined.
Your task is to determine c, s and r so that your rotator has the following properties:
The definition for real and complex data should be consistent, so that real data
passed to the complex algorithm (for rotator) should result in the same answer
(modulo roundoff) as from the real algorithm.
[Hint: See stable generation of rotation in the slides on Givens rotation.]
Define a complex plane rotator in Cn that rotates a vector in the xi xj plane.
4. Complex reflector: Given u ∈ Cn with kuk2 = 1, define Q := I − 2uu∗ . Show that
Qu = −u and Qv = v if hu, vi = 0. Further show that Q = Q∗ = Q−1 . The matrix Q is
called a complex reflector.
Let x, y ∈ Cn be such that kxk2 = kyk2 . Suppose that x and y are linearly independent
and hx, yi is real, that is, hx, yi ∈ R. Show that there exists a complex reflector Q such
that Qx = y. If x ∈ Cn is nonzero then show that there is a complex reflector H such
that  iθ 
−e kxk2
 0 
Hx =  ,
 
..
 . 
0
where θ is such that x1 = |x1 |eiθ and θ = 0 if x1 = 0. Write an algorithm that generates
the reflector H.

1
5. Let A ∈ Rn×n . Consider the QR algorithm

For k =1:n-1
Compute reflector Q = I- aplha * u * u’ such that
(QA)(k,k) = - sigma and (QA)(k+1:n, k) = 0
Compute w = A’ * u and A = A - alpha * u * w’
end

(a) Show that at the k-th step, the rank one update A ← A - alpha * u * w’ does
not affect the first k − 1 rows and columns of A.
(b) Modify above algorithm so that the strict upper triangular part of A contains the
strict upper triangular part of R (where A = QR), the diagonal entries of R are
stored in an extra array and the lower triangular part of A contains the householder
vectors (that is, vectors u’s that determine the reflectors).

6. This problem is about how to accumulate reflectors to produce Q in the QR factorization


of A ∈ Rn×n . Let Q1 , . . . , Qn−1 be reflectors such that Qn−1 Qn−2 · · · Q2 Q1 A = R, where
R is upper triangular and Qk = I − αk uk u∗k for k = 1 : n − 1. Set Q = Q1 Q2 · · · Qn−1 .
Then A = QR. Suppose that the matrix Q is explicitly required. Describe an efficient
algorithm to compute Q and give the total flop count.

********** End ***********

You might also like