next up previous
Next: Format of data Up: PM Code Previous: Equations and dimensionless variables

Scheme of integration

Equations (8 - 9) are solved using finite differences with a constant step in space $\Delta$x = $\Delta$y = $\Delta$z = 1 and a constant step in the expansion parameter $\Delta$a . We use the ``leap-frog'' scheme to advance coordinates and velocities from one moment to another. (In the following we drop tildas for all dimensionless variables.) At any moment an = a0 + n$\Delta$a , we have the coordinates x n and the potential $\phi_{n}^{}$ . Velocities p n - 1/2 are defined at an - 1/2 = an - $\Delta$a/2 . The coordinates and the velocities for the next moment are found using:


p n + 1/2 = p n - 1/2 - F(an)$\displaystyle\nabla$$\displaystyle\phi_{n}^{}$$\displaystyle\Delta$a,   
x n + 1 = x n + $\displaystyle{F(a_{n+1/2})\over a_{n+1/2}^2}$p n + 1/2$\displaystyle\Delta$a (14)

In order to solve eq.(9) we approximate the Laplacian operator using the 7-point ``crest'' template:

$\displaystyle\nabla^{2}_{}$$\displaystyle\phi$ $\displaystyle\approx$ $\displaystyle\phi_{i\pm 1,j,k}^{}$ + $\displaystyle\phi_{i,j\pm 1,k}^{}$ + $\displaystyle\phi_{i,j,k\pm 1}^{}$ - 6$\displaystyle\phi_{i,j,k}^{}$, (15)

where (i,j,k) = 1,...,Ngrid . This leads to a large system of linear equations relating unknown variables $\phi_{i,j,k}^{}$ with known right-hand side of the discrete form of the Poisson equation 3$\Omega_{0}^{}$($\rho_{i,j,k}^{}$ - 1)/2a . The system of equations is solved exactly by the FFT technique.

The density on the mesh $\rho_{i,j,k}^{}$ is obtained from particle positions using the Cloud-In-Cell method. In order to find the ``acceleration'' g = - $\nabla$$\phi$ for each particle, the gravitational potential is differentiated on the mesh:

gx = - ($\displaystyle\phi_{i+1,j,k}^{}$ - $\displaystyle\phi_{i-1,j,k}^{}$)/2, gy = ..., gz = ... (16)

Then the acceleration is interpolated to the position of the particle using a three-linear interpolation. This scheme for the force interpolation (the same interpolation as in the density assignment) is very important because it does not produce a force acting on the particle itself. (Thus, an isolated point does not produce a force at the position of the particle). While this might sound like a natural condition for any realistic method, only two methods - PM and TREE - do not have this self-force. In P 3M the effect is minimized. In the case of multigrid methods the self-force cannot be excluded - only minimized. Typically this is achieved by placing extended buffer zones around regions with high resolution (e.g. Kravtsov 1997). No precautions were made in AP 3M method, which might result in spurious effects in regions were multi-level grids are introduced.

Thus, the main scheme of the PM method consists of the following four blocks repeated every time step:

next up previous
Next: Format of data Up: PM Code Previous: Equations and dimensionless variables
Jon Holtzman