These are chat archives for elemental/chat

Mar 2017
Jack Poulson
Mar 05 2017 06:12
there is now a pretty easy way to test the LP IPM against the entire netlib/LP_data test set:
bnl2 is giving some problems right now if the primal and dual step sizes are not forced to be equal
Jack Poulson
Mar 05 2017 22:16
for what it's worth, the fix seems to be to prevent the primal (dual) step size from being larger than the dual (primal) step size when its norm is significantly smaller, and to prevent different step sizes before achieving approximate feasibility
80bau3b and bnl2 seem to be the two most challenging problems
maybe maros is third