embotech/ecos

Question about the feastol option (absolute vs. relative).

Opened this issue · 0 comments

Viech commented

Hello,

I wonder if the feastol option refers to an absolute measure of feasibility violation, or if the violation is measured in relation to the magnitude of other numbers that appear in the problem. For instance, CVXOPT divides the 2-norm of the vector of violations by the 2-norm of the right hand side (constant) part of all constraints (see here).

I looked at the code, but I did not understand the meaning of the factors that make up the "primal residual" that is compared to feastol:

info->pres = MAX(nry, nrz) / w->tau;