Research Article

Smoothing and Regularization with Modified Sparse Approximate Inverses

Table 3

Optimal reconstruction error 𝑥 1 ̃ 𝑥 1 2 for the 1D operator 𝐻 8 1 and original data 𝑥 1 of size 𝑛 = 1 0 3 affected by random noise of order 0.01%.

Regularization method Optimal valueReached at iteration

CG ( 𝐻 8 1 ) 2 . 1 8 7 0 1 0 4 5
CG ( 𝐻 8 1 𝑇 𝐻 8 1 ) 1 . 8 9 9 8 1 0 4 62

PCG ( 𝐻 8 1 ) with 𝑀 3 𝜌 𝜌 = 0 . 4 2 . 3 6 9 9 1 0 4 4
𝜌 = 1 1 . 9 7 4 6 1 0 4 12
𝜌 = 5 1 . 9 2 3 6 1 0 4 34
𝜌 = 1 0 2 1 . 9 2 2 6 1 0 4 48