As others have mentioned in the forum, OSQP is a great product - many congratulations and thank you. My questions is generic, please forgive me if the answer is already out there in a textbook.
Can the primal and dual residual be connected to the distance (in units of objective function) from the optimal point?
I have a problem which is rather large and time-sensitive, therefore I’d like to solve it in as few iterations as possible, as long as I have some guarantee that I won’t be more than a given amount (this itself can be a function of the residuals / eps_abs / eps_rel) far from the optima. If such an ansatz exists, that would help me specify eps_abs and eps_rel with confidence. Currently I am trying the naive approach of incrementally increasing precision and observing the effects on the optimal objective, but this does not generalize.
I see in Section 3.3.1 in https://web.stanford.edu/~boyd/papers/pdf/admm_distr_stats.pdf something that may provide some direction, but I’m not sure if this applies to the osqp algo equally?
Thanks a lot