[Ipopt] Warm Starting Options Suggestions

[Gabriel Hackebeil]

Breannan,

If you are launching Ipopt via AMPL (or with NL files in general), I believe there is an option called warm_start_init_point that indicates whether or not initial starting points for variable bound multipliers should be respected. By default they are ignored, but they also do not appear in the NL file by default unless indicated using an AMPL suffix. More details here (page 57): http://web.mit.edu/ipopt_v3.11.6/doc/documentation.pdf

Regards,
Gabe Hackebeil

> On Nov 5, 2014, at 6:46 PM, Breannan Smith <smith at cs.columbia.edu> wrote:
> 
> Dear Ipopt Community,
> 
> I am solving a sequence of QPs, some of which are quite similar, and thus wanted to experiment with warm starting. Despite the caveats against warm starting an interior point method, I was able to obtain a reasonable speedup by turning down the values of warm_start_bound_push and warm_start_mult_bound_push, as well as by setting mu_strategy to adaptive.
> 
> Has anyone had luck warm starting Ipopt? Are there options beyond the ones above that I should experiment with?
> 
> Thank your for your advice.
> 
> Best regards,
> Breannan Smith
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt