Design, implementation, and optimization of a highly efficient multilevel fast multipole algorithm

In this work we present a highly optimized sequential multilevel fast multipole algorithm. Departing from Fortran coding tradition, we have implemented the algorithm in C++, making extended use of the object-oriented features of the language, which leads to a great data locality. Great effort has also been put into optimizing the interpolation routines, which take up a significant fraction of the computation time. Finally, we also show that, by ordering carefully the interpolations and alpha translations, it is possible to reduce the octtree memory to almost half of what is normally needed.