I think stop_d, stop_c and stop_p have nothing to do in NLPModel. Similarly, the methods that compute KKT residuals are not properties of a model, and so should be moved out of there. I propose to expand the KKTresiduals class to be a general "KKT residual computer". Instead of assigning values to the residuals, we would create an instance based on a model, e.g.,
update model.x, model.y, etc., and then update the residuals using something like kkt.update().
Currently, no solver uses the methods that compute KKT residuals to check stopping conditions.
I think
stop_d,stop_candstop_phave nothing to do inNLPModel. Similarly, the methods that compute KKT residuals are not properties of a model, and so should be moved out of there. I propose to expand theKKTresidualsclass to be a general "KKT residual computer". Instead of assigning values to the residuals, we would create an instance based on a model, e.g.,update
model.x,model.y, etc., and then update the residuals using something likekkt.update().Currently, no solver uses the methods that compute KKT residuals to check stopping conditions.