Linear Algebra and Its Applications · Problem 7.4.6
Linear Programming and Game Theory — Gilbert Strang
Mathematics · iterative-refinement
Problem
The true solution to Ax=b is slightly different from the elimination solution to LUx0=b; A−LU misses zero because of roundoff. One strategy is to do everything in double precision, but a better and faster way is iterative refinement: Compute only one vector r=b−Ax0 in double precision, solve LUy=r, and add the correction y to x0. Problem: Multiply x1=x0+y by LU, write the result as a splitting Sx1=Tx0+b, and explain why T is extremely small. This single step brings us almost exactly to x.