The convex model about an iterate: the caller's own subproblem, solved by whatever inner solve it chooses.
The model.
OptionalcorrectThe second-order correction of a rejected step, against the Maratos
effect (Nocedal and Wright 2006, section 15.6, pp. 443-444): the step
p plus a correction p^ that meets the constraints' linearisation at
the trial, A(x) p^ + c(x + p) = 0 on the rows the step meets, its
minimum-norm solution being their (15.36). Asked only for a step whose
ratio did not exceed ISchedule.eta and whose predicted decrease is
positive, once per step and before the region shrinks; the corrected
point is judged on the step's own predicted decrease, taken where its
ratio exceeds eta and dropped otherwise, never searched along
(their Algorithm 15.2, p. 443, in the trust region of their Algorithm
18.4, p. 549). Absent, no correction is tried.
The iterate.
The rejected step.
The reading at the step's trial.
The corrected point x + p + p^, or null where the caller has none.
Whether a reading meets the constraints, to the caller's tolerance.
The reading.
Whether it is feasible.
The problem, as the caller reads it.