抄録
We propose a multiple-precision arithmetic environment, which reduces influence of rounding errors, as a clue of qualitative improvement of scientific computations. The usual double precision system is not satisfactory for computations of numerically unstable problems although it has about 15 digits accuracy in representation and computation of real numbers, but our proposed environment gives a new method to deal with them and enable us to carry out high accurate computations. We show the effective use of multiple-precision arithmetic for numerical treatments of analytic functions and high-accurate numerical integration formula. Such ultimate accuracy is necessary in numerical computations of inverse and ill-posed problems. Proposed environment is designed for new 64-bit personal computer architectures.