I am doing reaction diffusion computing and I get artifacts related to the Cartesian coordinate system in combination with the design of the Laplacian operator. The initial circularly symmetric seed develops into
where the upper left structure uses a traditional vertical horizontal Laplacian and the upper right structure a Laplacian with diagonals added. Lower left is a Laplacian calculated from the 12 closest points and lower right 20 points. Even the 20 point version decays after a while. The code for this is at https://www.shadertoy.com/view/3sGXWG
Is there a better or a "best" Laplacian operator to use in situations like this? Someone suggested hexagonal and Fourier coordinates.
I even accept answers involving other coordinate systems, provided they are transformable back to the Cartesian coordinates for viewing on computer screens.