Thank you very much! I used gradient descent and it's much faster.
However, I have a problem that: when I used the linear fitting method, the resulted FA looks okay, but when I used the non-linear method (no matter Powell, hybrid, or gradient descent), the resulted FA looks very weird. Please see the two attached images. The very bright one is by nonlinear method. They are both scaled between [0 1]. Actually when I used the 3dDWItoDT old version of Aug 2007, the FA map fitted by default nonlinear method and linear method both look fine and similar, but I could not find the older version any more.
Thank you!