Iterative refinement explained

Iterative refinement is an iterative method proposed by James H. Wilkinson to improve the accuracy of numerical solutions to systems of linear equations.^{[1] [2]}

When solving a linear system

A x = b , due to the compounded accumulation of rounding errors, the computed solution \hat{ x } may sometimes deviate from the exact solution x _{\star.} Starting with x _{1} = \hat{ x }\,, iterative refinement computes a sequence \{ x _{1,} x _{2,} x _{3,...\}} which converges to x _{\star,} when certain assumptions are met.

Description

For

m = 1 , 2 , 3 , ... , the th iteration of iterative refinement consists of three steps:

The crucial reasoning for the refinement algorithm is that although the solution for in step (ii) may indeed be troubled by similar errors as the first solution,

\hat x , the calculation of the residual in step (i), in comparison, is numerically nearly exact: You may not know the right answer very well, but you know quite accurately just how far the solution you have in hand is from producing the correct outcome . If the residual is small in some sense, then the correction must also be small, and should at the very least steer the current estimate of the answer,, closer to the desired one, x _{\star.}

The iterations will stop on their own when the residual is zero, or close enough to zero that the corresponding correction is too small to change the solution which produced it; alternatively, the algorithm stops when is too small to convince the linear algebraist monitoring the progress that it is worth continuing with any further refinements.

Note that the matrix equation solved in step (ii) uses the same matrix

A for each iteration. If the matrix equation is solved using a direct method, such as Cholesky or LU decomposition, the numerically expensive factorization of A is done once and is reused for the relatively inexpensive forward and back substitution to solve for at each iteration.^[2]

Error analysis

As a rule of thumb, iterative refinement for Gaussian elimination produces a solution correct to working precision if double the working precision is used in the computation of, e.g. by using quad or double extended precision IEEE 754 floating point, and if is not too ill-conditioned (and the iteration and the rate of convergence are determined by the condition number of).^[3]

More formally, assuming that each step (ii) can be solved reasonably accurately, i.e., in mathematical terms, for every, we have$$A\; \backslash left(I\; +\; F\_m\; \backslash right)\; \backslash mathbf\_m\; =\; \backslash mathbf\_m$$

where, the relative error in the -th iterate of iterative refinement satisfies$$\backslash frac\; \backslash leq\; \backslash bigl(\backslash sigma\backslash ,\backslash kappa(A)\backslash ,\backslash varepsilon\_1\backslash bigr)^m\; +\; \backslash mu\_1\backslash ,\; \backslash varepsilon\_1\; +\; n\backslash ,\; \backslash kappa(A)\backslash ,\; \backslash mu\_2\backslash ,\; \backslash varepsilon\_2$$

where

• denotes the -norm of a vector,
• is the -condition number of,
• is the order of,
• and are unit round-offs of floating-point arithmetic operations,
• , and are constants that depend on, and

if is "not too badly conditioned", which in this context means

and implies that and are of order unity.

The distinction of and is intended to allow mixed-precision evaluation of where intermediate results are computed with unit round-off before the final result is rounded (or truncated) with unit round-off . All other computations are assumed to be carried out with unit round-off .

Notes and References

1. Book: Wilkinson, James H. . 1963 . Rounding Errors in Algebraic Processes . . Englewood Cliffs, NJ . registration.
2. Cleve B. . Moler . April 1967 . Iterative refinement in floating point . . 14 . 2 . 316–321 . . New York, NY. 10.1145/321386.321394. free .
3. Book: Higham, Nicholas . Accuracy and Stability of Numerical Algorithms . 2 . SIAM . 2002 . 232.