I noticed in one of your Hot Tips posts (viewtopic.php?f=28&t=562) you wrote:
Note that chemical signals tend to have a much shorter effective length constant than
electircal signals do. Therefore nseg will probably have to be much larger than if you were
interested only in electrical signals. As always, choose odd values for nseg, and test
for adequate spatial accuracy by following this protocol:
It make sense to me why one would need a finer spatial scale to study diffusion since it's length constant will most likely be on a fine spatial scale, however why does the number of segments need to be odd?
I have had some issues with modeling diffusion and I have convinced myself (hopefully incorrectly) that the problem is due to adding too many segments (or shells in radial diffusion). Without getting in to specifics (yet :) ), I was thinking that the model might be constrained by the fact that there can only be flux between two adjacent compartments. If the segments get too thin then you'd actually want to have molecules diffusing through more than one segment in less than a single time step. I guess I was thinking that perhaps the model wasn't appropriately shrinking the time-steps to deal with the thin segments. Now that I write that, I bet it does do that appropriately, so maybe I'm just not interpreting my results correctly? Still, what about the case where a diffusion coefficient is very large (simulating instantaneous equilibrium), will the time-step shrink enough so that an ion could equilibrate fast enough in all the compartments? For my own sanity I would love to know that NEURON will perform correctly in these cases (motivation below).
Let's consider the radial diffusion of calcium with influx and eflux happening only in the outer shell:
1) [Ca] transients should be the same in the case where nshells = 1 and nshells = 100 with diffusion coefficient = very large
2) As nshells approaches a large number, the results should approach a single solution
Are these tests reasonable? Does the calcium diffusion model in the NEURON book pass these tests? I can't put my finger on the code I used to test these right now, but I seem to remember that these tests do not pass. I'll run them again if you're interested though...
Now I'm running into discrepancies in similar controls for longitudinal diffusion and I'm wondering if I'm just confused, or if there really could be problems with these "control experiments" that I'm trying. Anyone have thoughts about these questions? Any insight is welcomed. Thanks!