No label defined (Q4938247)

The paper is devoted to the construction of simple approximation formulas with the rate \({\mathcal O}(h^N)\) for arbitrary \(N\in\mathbb{N}\) and certain range of stepsize \(h\) relevant in numerical computations. The concept of approximate approximations reflects the fact that for computers ``\(h\) tending to 0'' is impossible, since floating point arithmetic and memory resources always give lower bounds for stepsizes. On the other hand, the numerical results for any relevant practical application are needed only within some tolerance, prescribed for example by the accuracy of input data. Therefore, numerical computations do not need approximations for arbitrarily small \(h\), but rather high-order approximations in some range of \(h\) accessible to the computing system. For the entire collection see [Zbl 0924.00038].