| Abstract: |
| In this talk, we introduce an efficient implementation of affine
arithmetic. To compute tight enclosures of trajectories of dynamical
systems, it is essential to suppress the wrapping effect. Although
affine arithmetic can effectively reduce the wrapping effect, it tends
to require significantly longer computation time compared to other
methods. To address this issue, we accelerated affine arithmetic by
utilizing error-free transformation, SIMD vectorization, and
mixed-precision arithmetic. We will demonstrate its effectiveness by
applying the accelerated implementation to several dynamical systems. |
|