[Ipopt] The scaling of problem.

[Lewis I]

Dear all, 

I found that when I try to scale up the initial point (multiple with 100); the
optimization is much faster (100 times faster) and provide a more precise
optimal point than other scaling on initial point (multiple with 10000 or
multiple with 1). 
But all the constraints and objective
function are the same throughout these testing, and I have turned on
the build-in scaling of ipopt. 
What is the possible reason for
that situation?
 
Thank you so much.
 
Regards, 
Lewis