Consider the diffusion of some unknown $R$ on a physically discrete, 1-D ring:

n = 0         1         2         3

We let grid point $n=0$ be the same as grid point $n=3$, giving us our ring. We are given a diffusion coefficient $D$, which is still multiplied by some to some notion of the "Laplacian" (which in 1-D, is just the second order difference) of $R$ in order to tell us how $R$ moves diffuses between neighbouring nodes

We can use techniques such as 2nd order central differences to calculate some notion of second order difference of $R$ on these physically discrete points, if we consider the numerical discretization of the grid to be the same as the physical discretization.

What if the physical discretization is not uniform?

n = 0                1  2       3

What notion of "second order difference" in $R$ might I now consider given that:

  1. I can't simply consider the physical discretization to be the same as the numerical discretization, without complicating things (non-uniform numerical grid spacing);
  2. but more importantly, I don't have any continuous physical situation to go back in order to derive a different second order differencing scheme using Taylor series, so I can't use some finer uniform numerical grid, which also happens to coincide with the physical points -- it would make no sense to have a numerical grid point that doesn't coincide with a physical grid point.

Is my only solution to make some sort of assumption regarding the "continuous" nature of $R$ between the grid points? Maybe I assume that $R$ varies linearly between the physically discrete points? Then the problem I outline in point 2. isn't there, and I can define a finer numerical grid? This is the solution advocated in a similar question here: How can I numerically differentiate an unevenly sampled function?

What other ideas might I consider?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.