www.neuron.yale.edu

Hi everyone,

I have a very general question which has been in my mind lately (brought about by a very specific issue).

I have a multicompartmental model, and I noticed that when I run the exact same simulation (no pseudorandom variables) it on my Mac OS X machine I get different results than if I run it on my unix machine.

I was wondering what the possible causes of this issue is? I know that there are indeed differences in processor architcture but I currently don't understand what those differences are or could be, and how they would affect the results.

Does this necessarily imply that the model is insensitive to changes in parameters or inputs (ie not robust?)

Cheers and thanks.

adamimos wrote:I get different results than if I run it on my unix machine.

First question:
suppose you run a simulation, then without exiting NEURON run another simulation. The two runs should generate identical results. If they don't, there is a problem with initialization. That could account for different results on different machines (uninitialized memory locations might contain different garbage values on different machines or under different OSes).

A comment: model setup code that implements pseudorandom variation (variability of parameters) or specifies fluctuation of variables at run time (e.g. "noisy" signal sources, stochastic variation of channel gating or transmitter release) can be very difficult or impossible to debug unless it is written in such a way that the randomness itself is reproducible.

Does this necessarily imply that the model is insensitive to changes in parameters or inputs (ie not robust?)

It suggests the existence of a bug somewhere. Just to make sure it isn't a problem in NEURON itself, you might want to update to the latest standard version, which was just released--http://www.neuron.yale.edu/neuron/download