4.10.1 Proof of Proposition 4.1 in Section 4.4We follow the technique used in [203] by defining the function Equation 4.184
Notice that Equation 4.185
Equation 4.186
Equation 4.187
Equation 4.188
where (4.188) follows from the assumption that y '( x ) m . In (4.188), denotes that the matrix ( A - B ) is positive semidefinite. It then follows from (4.185), (4.187), and (4.188) that d ( t ) 0, for any t . Now on setting Equation 4.189
we obtain Equation 4.190
Assume that the penalty function r ( x ) is convex and bounded from below; then the cost function C ( q ) is convex and has a unique minimum C ( q *). Therefore, q * is the unique solution to (4.15) such that z ( q *) = . Since the sequence C ( q l ) is decreasing and bounded from below, it converges. Therefore, from (4.190) we have Equation 4.191
Since for any realization of r , the probability that z ( q l ) falls in the null space of the matrix ( SR -1 S T ) is zero, then (4.191) implies that z ( q l ) with probability 1. Since z ( q ) is a continuous function of q and has a unique minimum point q *, we then have q l q * with probability 1, as l . 4.10.2 Proof of Proposition 4.2 in Section 4.5Denote . Then (4.75) can be written in matrix form as Equation 4.192
Denote . Then from (4.73) and (4.74) we obtain Equation 4.193
Using (4.192) and (4.193), we obtain Equation 4.194
Equation 4.195
Equation 4.196
where in (4.194) denotes the Moore “Penrose generalized matrix inverse [189]; in (4.195) we have used the fact that , which can easily be verified by using the definition of the Moore “Penrose generalized matrix inverse [189]; in (4.196) we have used the facts that ; and (4.196) is the matrix form of (4.76). Finally, we notice that Equation 4.197
It follows from (4.197) that the k th diagonal element of the diagonal matrix A -2 satisfies Equation 4.198
|