Three high performance simplex solvers

Event Sponsor: 
Mathematics and Computer Science Division LANS Seminar
Start Date: 
May 16 2017 - 10:00am
Building/Room: 
Building 240/Room 1405
Location: 
Argonne National Laboratory
Speaker(s): 
Julian Hall
Speaker(s) Title: 
The University of Edinburgh
Host: 
Sven Leyffer

This talk gives an overview of the computational features of three high performance implementations of the revised simplex method for solving large scale sparse linear programming (LP) problems. The first (EMSOL) was developed during relatively early work on parallelising the revised simplex method, but also yielded an important serial technique. The second (PIPS-S) was developed in collaboration with researchers at Argonne, and solves stochastic MIP relaxations in parallel. The third (hsol) was also written to study parallelism, and its underlying techniques will be introduced. Each solver has an associated article which won a best paper award in Computational Optimization and Applications. A particular crashing technique which leads to significantly improved solution time for certain classes of problem will also be presented. Final observations point the way to the use of hsol as a major open source linear optimization resource.
 

Miscellaneous Information: 

Goodies and coffee will be served.

Please click below to add this event to your calendar.

[schedule.ics]

 

Upcoming Seminars:
 
May 17, 2017, "Stratified Markov chain Monte Carlo" Brian Van Koten, Department of Statistics, University of Chicago, [more info]
May 23, 2017, "Y-Tune: Performance Portability Using Compiler-Directed Autotuning" Mary Hall, Professor, School of Computing at the University of Utah, [more info]
May 31, 2017, "TBD" Kibaek Kim, Mathematics/Statistics 2, MCS/ANL, [more info]