[Ipopt] Warm starting IPOPT with the solution of the optimization problem

[Stefan Vigerske]

Hi,

Ipopt is an interior-point solver, but solutions are usually not on the 
interior. So if you start it from a solution point, Ipopt will first 
walk into the interior and then back to a solution. Changing the initial 
value for the Barrier parameter mu may make this less severe.

Also Ipopt moves a given starting point away from the bounds at the 
beginning. There are some options to change that, too.

Stefan

On 4/26/20 6:14 PM, Gianpaolo Gulletta wrote:
> Dear all,
> 
> my question is pretty trivial and leads to the implementation of the IPOPT.
> 
> I first solve my non-linear constrained optimization problem with IPOPT 
> (cold-started) and I save the full solution (variables and multipliers) 
> into a file with a precision of 15 numbers after the decimal point.
> 
> Then, I solve the same problem with IPOPT warm-started with the previous 
> solution loaded from the file. It takes much less iterations to converge 
> successfully, but it does not take 0 iterations. Why?
> 
> I have an idea of the reasons, but I would like to fully understand why 
> this happens.
> 
> Thank you in advance.
> 
> Best Regards,
> 
> Gianpaolo
> 
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> https://list.coin-or.org/mailman/listinfo/ipopt