[Coin-ipopt] Re: Re: IPOPT + warm-start

[Andreas Waechter]

Hi Victor,

Good to hear you are interested in looking at this!

Just one clarification:

> As for
> the bound multipliers, I've looked at the AMPL documentation and there seems
> not to be a clean way to do this. Our only option would be specify the
> multipliers using the suffixes utility or to generate a method in the AMPL
> interface to read those values (from a file or an external routine).
> Additionally, we would need a special method similar to the
> "GetWarmStartIterate" in the IPOPTNLP Class just to make sure that the warm
> start point is in the interior of the feasible region (push the primal
> variables if not).

I like the suffix alternative most for the bound multiplier 
initialization.  However, you will not need to (and probably can't) use 
the GetWarmStartIterate method for the NLP (since that really needs the 
entire Ipopt internal iterate vector, which is more information than you 
have in the AMPL interface - that thing was mainly implemented to try 
something with Bonmin).  Instead, you would simply provide the bound 
multipliers information in the get_starting_point method of the AmplTNLP, 
whenever init_z is true (currently, all z_L and z_U values are set to 
zero).

I hope this makes sense.  When you want to discuss the suffix 
implementation, please let me know.

Cheers

Andreas