Evaluating a function by approximating it in a rather large domain may require polynomial or rational functions of large degrees. This may lead to long delays of computation, and this may also make the numerical error control difficult. A natural way to deal with this problem is to split the interval where the function is to be approximated into several smaller subintervals. It suffices to store in a table, for each subinterval, the coefficients of a low-degree approximation that is valid in that interval. Such a method is not new (a PDP-9 implementation is reported in ), but it becomes very attractive nowadays, since memory is less and less expensive.
KeywordsDouble Precision Minimax Approximation Machine Number Target Precision Representable Number
Unable to display preview. Download preview PDF.