I.4.3.2: A Family of Quadratics

This example shows some of the properties of coordinate relaxation. We want to minimize For each this is a quadratic in and . If we fix and , the resulting function is a convex quadratic in . And if we fix and , the resulting function is a convex quadratic in . Thus coordinate relaxation can always be carried out, with a unique minimum in each substep.

The partials are given by and the Hessian is the matrix Thus the eigenvalues of the Hessian are and .

If then the function has a unique isolated minimum equal to zero at . If it has a minimum equal to zero on the line and if it has a minimum equal to zero on the line . If the unique stationary point at is a saddle point, and there are no minima or maxima.

Coordinate relaxation gives the algorithm or Thus we have convergence to if and only if In that case convergence is linear, with rate Convergence is immediate, to or , if

For the function values we have If then the function values decrease, and diverge to . Also, diverges to infinity. By defining we easily change the problem into an equivalent one with the same iterates, for which function values converge to zero, but since is the same sequence as before it still diverges to infinity.

Note that while Thus function values converge twice as fast as the coordinates of the solution vector.