Improve conditioning of linear system

[JonasBreuling]

I stumbeled across a different method to make the linear system less i'll conditioned. The approach in eq. (30) [1] seems to be very simple and easy to implement. It is based in the ideas discussed in [2]. Do you think there will be time on your side to investigate this? Removing the iterative refinement seems to increase the performance since only a single linear system has to be solve in each iteration.

NB: Thanks for this great library.

[govindchari]

Thanks for sending this over! Yes I will look into it! (when I have time; the end of PhD is busy 😄).

This seems like a simple change and it would be if QOCO was in a higher level language, but due to the fact that the linear system is in CSC form and everything is written in C will make this a bit more involved to implement (~1-2 weeks I estimate).

I am interested in anything that will help the robustness/speed of QOCO, so this is a great find! I will get around to this when my schedule clears up a bit.

[JonasBreuling]

Nice to here. Indeed I ment "simple" in lerms of a higher level language. But the reasoning in the papers seem plausible to me, so I Just wanted you to know, since the library is one of the best I used Up to now. Good luck with your PhD! The end is hard but worth it!