Sei sulla pagina 1di 3

Applied Math Practicum

c 2006 QuIC Financial Technologies Inc. All rights reserved.


Copyright

Question 1

q
Define the norm of a real n × n matrix A by kAk = max kxk=1 kAxk. Here kxk = x21 + · · · + x2n is
the usual n–dimensional Euclidean norm and the maximum is taken over the unit sphere in R n .

1. Suggest algorithms for computing kAk in two cases; where A is a dense 100 × 100 matrix and
where A is a tridiagonal 3333 × 3333 matrix. Discuss the rationale behind and the complexity
of each approach.

2. Prove that kAk = kA> k. Among the real invertible matrices A of norm kAk ≤ 1, show that
the orthogonal matrices are the only ones for which kAk = kA −1 k.

3. Let A be a real n × n orthogonal matrix. If B is any real n × n matrix, show that the matrices
m
1 X
Bm = A−k BAk , m = 0, 1, 2, · · ·
m + 1 k=0

converge entrywise as m −→ ∞. [Hint. By using a suitable matrix decomposition, it suffices


to prove this when A is a diagonal unitary complex matrix].

Question 2

Take the following fact as given:

Given any complex n × n matrix A, there exists a unitary matrix Q and an upper
triangular matrix T , both n × n, such that Q ∗ AQ = T . The eigenvalues of A are the
diagonal entries of T .

There are efficient methods for computing Q and T from A. In what follows, assume that A is an
n × n complex matrix while Q and T are as described above.

1. Define exp(M ) = ∞ k 1 Show that exp(A) =


P
k=0 M /k! for any complex square matrix M .
Q exp(T )Q∗ . Thus, to compute exp(A) it suffices to compute exp(T ).

2. Show that (exp(T ))ii = eTii .


1
Implicit in this definition is a convergence concept similar to that involved in Question 1. Feel free to comment
on this should you be so inclined.

1
3. If A has n distinct eigenvalues, show that the i th superdiagonal of exp(T ) can be computed
from the (i−1)th and lower superdiagonals. Thus, starting from the diagonal of exp(T ), which
is easily computed from Part 2 above, we can recursively compute the first superdiagonal,
second superdiagonal, and so on. [Hint. T and exp(T ) commute].

4. Discuss how you would implement a general matrix exponentiation algorithm based on the
following fact.
Qr
If A has distinct eigenvalues λ1 , · · · , λr , minimal polynomial M (x) = j=1 (x −
λj )mj , and A is in Jordan normal form, then
m−1
X
exp(A) = fk Ak ,
k=0
Pr
where m = j=1 mj and the fk s satisfy the following system of m equations:
m−1
!
k e λj
λjk−i fk =
X
, i = 0, 1, · · · , mj − 1, j = 1, · · · , r.
i i!
k=i

What complexity issues and practical pitfalls are there to watch out for? If A has n distinct
eigenvalues, are there circumstances where you might expect the algorithm outlined in Part
3 to do better or worse than this more general algorithm?

Question 3

Consider the initial value problem

∂f ∂f ∂2f
− (t, x) + ax (t, x) + b 2 (t, x) = 0, t ≥ 0, −∞ < x < ∞,
∂t ∂x ∂x

where a < 0 < b are constants. We wish to calculate f (1, 0) given the boundary condition f (0, x) =
f0 (x), −∞ < x < ∞.

1. Express f (1, 0) as an expectation of the form E[f 0 (Y )], where Y is a Gaussian distributed
random variable. Based on this representation, suggest a numerical solution algorithm.

2. Without referring to the probabilistic representation of f (1, 0) in Part 1, suggest finite differ-
ence solution methods under each of the following conditions. In each case, you may assume
that Dirichlet–type boundary conditions are specified for |x| −→ ∞ and for all t ∈ (0, 1].

• all derivatives of f0 (x) exist and are bounded. a  −1 and b = 10 −2 .


• f0 (x) as above, a = −10−2 , and b  1.
• f0 (x) = max{x, 0}. Consider both of the previous conditions on the coefficients a, b.

Is there a single finite difference scheme that will work reasonably well under each of the
previous conditions? Is there a change of variable x 7→ y that transforms the original problem
into one more amenable to finite differencing?

2
Question 4

Your favorite baseball team is playing against your uncle’s favorite team in the World Series. At
the beginning of each game, you and your uncle bet on the game’s outcome. Your uncle, being
wealthy and carefree, always lets you choose the amount of the bet. If you bet b dollars and your
team wins the game, your uncle gives you an IOU for b dollars. But if they lose the game, you give
him an IOU for b dollars. When the series is over, all outstanding IOUs are settled in cash. You
would like to walk away with $100 in cash if your team wins the series, and lose $100 if your team
loses the series. How much should you bet on the opening game? (For non–baseball fans, the first
team to win a total of four games wins the series).

Potrebbero piacerti anche