Numerical Methods Qualification Exam Problems and Solutions (University of Maryland)/January 2004

From Wikibooks, open books for an open world
Jump to navigation Jump to search

Problem 1[edit | edit source]

Let be an arbitrary fixed partition of the interval . A function is a quadratic spline function if


(i)


(ii) On each subinterval , is a quadratic polynomial


The problem is to construct a quadratic spline interpolating data points . The construction is similar to the construction of the cubic spline but much easier.

Problem 1a[edit | edit source]

Show that if we know , we can construct .

Solution 1a[edit | edit source]

Consider interval . Since is linear on this interval, using the point slope form we have



Integrating, we have



or, in a more convenient form,


Problem 1b[edit | edit source]

Find equations which enable us to determine the . You should find that one of the can be prescribed arbitrarily, for instance

Solution 1[edit | edit source]

Since is continuous on ,



i.e.



Simplifying and rearranging terms yields the reoccurrence formula


Problem 2a[edit | edit source]

Give the definition of the algorithm for finding the eigenvalues of a matrix . Define both the unshifted version and the version with shifts

Solution 2a[edit | edit source]

Unshifted Version[edit | edit source]

Shifted Version[edit | edit source]

Problem 2b[edit | edit source]

Show that in each case the matrices generated by the algorithm are unitarily equivalent to (i.e. unitary).

Solution 2b[edit | edit source]

Suppose for some unitary . Since we have

which is as desired as is unitary.

For the shifted case, the same argument holds using the fact that .

Problem 2c[edit | edit source]

Let



Use plane rotations (Givens rotations) to carry out one step of the algorithm on , first without shifting and then using the shift . Which seems to do better? The eigenvalues of are . (Recall that a plane rotation is a matrix of the form



with

Solution 2c[edit | edit source]

The shifted iteration appears to work better because its diagonal entries are closer to the actual eigenvalues than the diagonal entries of the unshifted iteration.

Unshifted Iteration[edit | edit source]

Shifted Iteration[edit | edit source]

Problem 3[edit | edit source]

Let be an symmetric, positive definite matrix. Then we know that solving is equivalent to minimizing the functional where denotes the standard inner product in . To solve the problem by minimization of we consider the general iterative method

Problem 3a[edit | edit source]

When and are given, show that the value of which minimizes as a function of is given in terms of the residual



Solution 3a[edit | edit source]

Useful Relationship[edit | edit source]

Since is symmetric



This relationship will used throughout the solutions.


Substitute into Functional[edit | edit source]

Take Derivative With Respect to Alpha[edit | edit source]

Set Derivative Equal To Zero[edit | edit source]


which implies


Problem 3b[edit | edit source]

Let be an -orthogonal basis of , . Consider the expansion of the solution in this basis:



Use that orthogonality of the to compute the in terms of the solution and the 's

Solution 3b[edit | edit source]


which implies


Problem 3c[edit | edit source]

Let denote the partial sum



so that where the 's are the coefficients found in (b). Use the fact that and the -orthogonality of the 's to conclude that the coefficient is given by the optimal i.e.


Solution 3c[edit | edit source]


which implies