Add documentation page listing and naming available solvers?

[pierre-haessig]

Hello,

I wonder whether it may be useful to add a documentation page listing the available solvers in HiGHS. In particular, this may clarify some aspects:

• provide fully developed names to the short names (simplex, ipm and pdlp) used in the solver setting option https://ergo-code.github.io/HiGHS/dev/options/definitions/#solver
• is the "Dual simplex" mentioned on the Parallelism page (https://ergo-code.github.io/HiGHS/dev/parallel/#Dual-simplex) the same as pdlp? (seems yes from previous discussion)
  • if yes, add the discussion comment that it is meant for GPU usage?
• finally (but this is a separate issue), a parallel interior point method is mentioned as future plans https://ergo-code.github.io/HiGHS/dev/parallel/#Future-plans. Is it indeed a work in progress?

[jajhall]

I wonder whether it may be useful to add a documentation page listing the available solvers in HiGHS. In particular, this may clarify some aspects:

There is already some information on Solution-algorithms

• provide fully developed names to the short names (simplex, ipm and pdlp) used in the solver setting option https://ergo-code.github.io/HiGHS/dev/options/definitions/#solver

This isn't necessary, as these are three wholly distinct solution techniques for LP. Variants within these techniques are

• is the "Dual simplex" mentioned on the Parallelism page (https://ergo-code.github.io/HiGHS/dev/parallel/#Dual-simplex) the same as pdlp? (seems yes from previous discussion)

  • if yes, add the discussion comment that it is meant for GPU usage?

No, "Dual simplex" is a variant of the simplex algorithm, but "PDLP" is a first order method.

I've realised that Solution-algorithms doesn't say anything about what PDLP is, so I'll add that, stressing that PDLP is currently only available for CPUs, and may be totally uncompetitive.

• finally (but this is a separate issue), a parallel interior point method is mentioned as future plans https://ergo-code.github.io/HiGHS/dev/parallel/#Future-plans. Is it indeed a work in progress?

Yes, and much progress has been made on the most challenging part of the development. We could make a prototype serial version available soon - that may be significantly faster than the current IPM solver on some problem classes. However, a full parallel implementation is unlikely to be available within the next 12 months.

[jajhall]

Closed by #1916

[pierre-haessig]

Thanks a lot for the feedback and PR #1916. I've made a minor Markdown syntax correction in #1939.

Overall PR #1916 is fine to me, except that now that there are links from the solvers explanation page to the options solver and simplex_strategy, I would also add the matching back-links from these options to the solvers explanation page.

Also perhaps add "see also simplex_strategy" in the "solver" option