The problem; its numeric storage remains unchanged during this synchronous solve.
Optionalsession: ISession
The caller's session of solves of one optimisation (HumanQuadraticInteriorSession.ISession): its first solve's scales serve the later ones, a solve of the same rows starts from the one before, and the solver's scratch is kept between them; absent, the solve is alone and cold.
The unscaled iterates with status solved, to the interior method's own accuracy; undefined when it did not converge, and the caller solves by solve or a prepared model.
The minimiser by the interior-point method (
HumanQuadraticInterior), for a caller whose curvature is badly conditioned for the operator splitting ofsolve: a second-order method's iteration count does not grow with that conditioning.