In an earlier work, we proposed a frame-based kernel analysis approach to the problem of recovering erasures from unknown locations. The new approach led to the stability question on recovering a signal from noisy partial frame coefficients with erasures occurring at unknown locations. In this continuing work, we settle this problem by obtaining a complete characterization of frames that provide stable reconstructions. We show that an encoding frame provides a stable signal recovery from noisy partial frame coefficients at unknown locations if and only if it is totally robust with respect to erasures. We present several characterizations for either totally robust frames or almost robust frames. Based on these characterizations several explicit construction algorithms for totally robust and almost robust frames are proposed. As a consequence of the construction methods, we obtain that the probability for a randomly generated frame to be totally robust with respect to a fixed number of erasures is one.
The DGMRES method for solving Drazin-inverse solution of singular linear systems is generally used with restarting.But the restarting often slows down the convergence and DGMRES often stagnates.We show that adding some eigenvectors to the subspace can improve the convergence just like the method proposed by R.Morgan in[R.Morgan,A restarted GMRES method augmented with eigenvectors,SIAM J.Matrix Anal.Appl.,16:1154-1171,1995.We derive the implementation of this method and present some numerical examples to show the advantages of this method.
