Costas,
The nonlinear solver requires that the functions are smooth (and
convex). The problem must also be primal and dual feasible.
Generalized linear-fractional problems can be solved using bisection
(ie, by implementing the algorithm on page 146). For the problem in
the exercise, the feasibility problem in step 2 of Algorithm 4.1 would
be:
Sk(p) >= t * ( Ik(p) + sigma_k ), k=1,...,n
plus the other linear inequalities in p.
This is a set of linear inequalities, so the feasibility problem can
be solved using the LP solver.
You may find the cvxmod package (
cvxmod.net) very helpful. It
automates part of the problem specification for cvxopt.
Lieven