Speaker
Description
Numerical methods are essential in many scientific applications for their ability to make discrete data
sets continuous. Interpolation is one commonly used method, which helps one to predict the value
of any point between two adjacent data points of a data set. Cubic spline interpolation is highly
regarded as the most useful method for its ability to interpolate between data points in a piece-wise
fashion. However, this advantageous method comes with the price of high computation time, especially
in cases such as molecular dynamics (MD) simulations where it must be applied repetitively. An
alternative representation of a cubic spline was proposed in previous research, where each spline could
be described as the linear combination of basis set functions, thus naming the method B-spline
interpolation 1 . It was proposed that implementing this new formulation can accelerate many scientific
computing operations which involve interpolation.
In this study, the performance of B-spline method was evaluated and compared to cubic spline
interpolation using a series of analytic functions. The effects were also evaluated in a Free Energies
from Adaptive Reaction Coordinates (FEARCF) 2 reaction sampling of protonation of ammonia into
ammonium 2 . Significant CPU speed up was observed at four-dimensional B-spline up to 10 5 . It
provided identical results regarding the accuracy. Furthermore, the speed up of the splining method led
to a 4-fold overall increase of the CPU speed of reaction sampling.
