Specifically, we selected the TNC (Truncated Newton Conjugate-Gradient) method because of the implementation's ability to 
bounds
different scales - relative error metric.

Results