You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Mathematically, my understanding is that this is used for computing the relative errors, and should be
$$
\text{primal\_feasibility\_in\_rhs\_0} = \max\{ \lVert C x \rVert_{\infty}, \lVert x \rVert_{\infty} \}
$$
However, the problem right now is that qpresults.si.tail(qpmodel.dim) does not contain $x$. It seems like this is not assigned in this function, and either contains zero or some other expression, i.e., from here
Dear @oswinso,
Thanks a lot for reporting this typo. qpresults.si.tail(qpmodel.dim) was indeed not assigned at this stage.
There are several ways for computing relative errors. One usual way (adapted to the box constraint case) is to indeed use $||x||_\infty$ and also the norm of its projection onto the box constraints. I have added these two in PR #258.
I think maybe there is a typo when computing
primal_feasibility_in_rhs_0
in theglobal_primal_residual
function when there are box constraints?proxsuite/include/proxsuite/proxqp/dense/utils.hpp
Lines 220 to 221 in c9586f8
Mathematically, my understanding is that this is used for computing the relative errors, and should be
However, the problem right now is that$x$ . It seems like this is not assigned in this function, and either contains zero or some other expression, i.e., from here
qpresults.si.tail(qpmodel.dim)
does not containproxsuite/include/proxsuite/proxqp/dense/solver.hpp
Lines 980 to 981 in c9586f8
This seems like a typo?
Perhaps it should instead be
which does contain$\lVert x \rVert_{\infty}$ , as assigned from earlier in the function?
proxsuite/include/proxsuite/proxqp/dense/utils.hpp
Lines 196 to 200 in c9586f8
The text was updated successfully, but these errors were encountered: