[Ipopt] MU_TARGET question

Vivek Periaraj vivek.periaraj at gmail.com
Sun Jan 20 02:50:20 EST 2013

>> I still get a 'close-to-zero' solution after setting mu_target to a value like 10. Is there something else I have to set to get all x_i to be positive? 
>> I guess I have to set the mu to be a really large value? As mu_target could be 'sum(x_i*nu_i)' - would that be correct? 

> If my understanding is correct then x_i*nu_i=mu_target holds for each i (not the sum). If you still get close-to-zero solutions, maybe your 
> multipliers are very large (scaling of the objective might help)? 

> If you really want to make sure that you get positive values, try setting the bound_relax_factor 
> http://www.coin-or.org/Ipopt/documentation/node87.html to zero.

Thanks! It seems the IP algorithms generally have sum(x_i * nu_i)/n = mu_target. Where n is the number of primal variables. I am not sure that's the reason why I am not getting a strictly feasible point. I will try your suggestions.


More information about the Ipopt mailing list