We study the problem of the appropriate choice of the interpolating kernel to be used in the evaluation of gradients of functions. Such interpolation technique is often used in applications, e.g. it is typical for Smoothed Particle Hydrodynamics (SPH). We propose a minimization procedure for selecting kernels in n-dimensions, in the class of regular, normalizable, symmetric functions having finite moments up to a sufficiently high order; the method is valid when the kernel width is position-dependent and allows to recover conservation laws at the same order of approximation that SPH, as interpolation technique, has when the kernel size is constant.

A modern approach to the evolution of a mixed (stars and gas) self-gravitating system is the fully Lagrangian particle approach. The gaseous (particle) phase differs from the compact because the mutual force is given by the sum of gravity and pressure gradient. In this note we report of some characteristics, advantages and limitations of this approach for what regards the evaluation of forces in the system. In particular, a comparison between classic tree-code and fast multipole algorithm to evaluate gravitational forces is discussed.

We discuss capability of Smooth Particle Hydrodynamics to represent adequately the dynamics of self-gravitating systems, in particular for what regards the quality of approximation of force fields in the motion equations. When cubic spline kernels are used, we find that a good estimate of the pressure field cannot be obtained in non uniform situations using the commonly adopted scheme of adapting the kernel sizes to include a fixed number of neighbours. We find that a fixed number of neighbours gives the best approximation of just the intensity of the force field, while the determination of the direction of the force requires a number of neighbours which strongly depends on the particle position.