Introduction
The Numerics module includes support for:
- FFTs
- Limited memory BFGS optimization
- b-Splines
- Erf and Erfc
- Tensor recursions
Fast Fourier Transformation
The FFT classes implement:
- 1D Real to Complex
- 1D Complex to Complex
- 3D Real to Complex
- 3D Complex to Complex
- 3D Real to Complex Convolutions
- 3D Complex to Complex Convolutions
The 1D methods compute the FFT of real or complex, double precision data of arbitrary length n using a mixed radix method that has special methods to handle factors of [2, 3, 4, 5, 6, 7] and a general method for larger prime factors.
The 3D methods are serial or SMP parallel. The convolutions are designed for the reciprocal space portion of Particle Mesh Ewald summation.
L-BFGS Optimization
The limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) algorithm is appropriate for large-scale multidimensional unconstrained optimization problems. It is derived from Robert Dodier's Java translation of orignal FORTRAN code by Jorge Nocedal.