The Extrapolated Taylor Model

The Taylor model [8] is one of the inclusion functions available to compute the range enclosures. It has the property of (m+ 1) convergence order, where, m is the order of the Taylor model used. It computes a high order polynomial approximation to a multivariate Taylor expansion, with a remainder term that rigorously bound the approximation error. The sharper bounds on the enclosures computed using the Taylor model can be obtained either by successively partitioning the domain x using suitable subdivision factors, or by increasing the convergence rate of the Taylor model using higher order Taylor models. However, higher order Taylor forms require higher degrees of the polynomial part, which in turn require more computational effort and more memory. This is the major drawback of increasing the order m of Taylor models for obtaining range enclosures with higher order convergence rates. In this paper, we attempt to overcome these drawbacks by using a lower order Taylor model, and then using extrapolation to accelerate the convergence process of the sequences generated with the lower order Taylor model. The effectiveness of all the proposed algorithms is tested on various multivariate examples and compared with the conventional methods. The test results show that the proposed extrapolation-based methods offer considerable speed improvements over the conventional methods.