In article <1993Sep8.200728.16169 at news.arc.nasa.gov>, brp at neuron.arc.nasa.gov (Bruce Parnas) writes:
>> As the subject says, I'm working on a distributed HH simulation. Multi-compartment,
> each one HH in nature (I'm not fond of the HH model, mind you, but I have to do this).
>> Here's the problem:
>> I'm coupling compartments passively, i.e. I(k) = {v(k-1) - v(k+1)}/axial_res.
>> If I compute the HH model per unit area (the way the parameters are given in the
> original paper), the input between compartments is too small to cause propagation.
>> If I normalize to the area, then Cm is ver small, 1/Cm is large and the input from
> neighboring compartments is too large and the thing goes unstable, even for very
> small timesteps.
I believe the problem is here. Cm is in units of uF/cm^2.
The various conductances (Gna, Gk, Gl) are in mS/cm^2. When
you divide the currents by Cm, the cm^2 is cancelled (You will
have to include your coupling current -- I(k) above -- in this
consideration as well.)
The thing to remember (when looking at single membrane site) is
that although everything seems to describe activity at a cm^2 level,
it is really independent of the size of membrane simulated.
>> In the first case, it works ok if i reduce axial_res by about 10,000
> In the second case, I have to increase it by about 3000.
>> Neither of these scale factors is justifiable.
See the arguments above. Check all your units.
>> Anybody have any suggestions for what might be going on here? Or a different
> way of coupling the compartments?
>> btw, i've run this code for a point HH model, and it works just fine. It's the
> compartmental coupling that's causing the problems here.
>> thanx for any help.
>>> --
> brp
> bruce parnas
>brp at psychomo.arc.nasa.gov> /usr/local/Std.Disclaimer: The opinions expressed here are mine and
> not those of NASA, but you probably could have guessed that.
Shameless plug for my site:
The Army High Performance Computing Research Center at the
University of Minnesota
There are two of us here looking at similar problems.
First, Rogene Eichler has built a network of HH compartments
and simulated them on our Connection Machines CM-2 and CM-5.
She has also taken realistic morphology for Hippocampal neurons
and used more accurate (than HH) models to start looking at
activity in groups of neurons. She has done quite a bit of work
on looking at good numerical methods as well.
Second, I have been looking at large networks (sheets and cubes)
of cardiac cells where instead of the 4 nonlinear differential
equations in HH we have 17. Both of us have preprints that we
would be happy to send out.
Tony Varghese