Figure 1

Timings for the M2L kernel. The time taken by the M2L kernel (using double precision without vectorisation) as function of expansion order p. The rotation accelerated version of Appendix A.6.2 is faster than the direct implementation of equation (3b) for $p\ge 5$.