Skip to main content
. Author manuscript; available in PMC: 2011 Jul 1.
Published in final edited form as: Neuroimage. 2010 Mar 4;51(3):1071–1081. doi: 10.1016/j.neuroimage.2010.02.060

Table 1.

Projected gradient descent algorithm.

  1. Form an initial guess x0 via truncated SVD, projected onto the non-negative space.

  2. Calculate the gradient vector according to Eq. (3).

  3. Perform a line search along the direction of the negative gradient to determine the optimal step length δ.

  4. Update the FOD estimate x according to Eqs. (4) and (5).

  5. Repeat steps 2–4 until the J-divergence of successive FOD estimates falls below the termination threshold.