Inconsistency in current injection initiation in NEST implementation

[appukuttan-shailesh]

Take a look at the output for this code in NEST:

It appears that the DCSource here is activated some time after the intended t_start. But after exploring further it is found that the issue is not with DCSource, but rather with the first instantiation, irrespective of current type. For example, this code shows the same issue for two ACSource's:

The delay here is 0.9 ms. That, I believe, doesn't correspond to the simulator's dt. So not sure about its origin.

[appukuttan-shailesh]

Interestingly, on rearranging the lines in this file to:

acsource1 = ACSource(start=start, stop=stop, amplitude=mean, offset=0.0, frequency=100.0, phase=0.0)
acsource2 = ACSource(start=start, stop=stop, amplitude=mean, offset=0.0, frequency=100.0, phase=0.0)
cells[0].inject(acsource1)
cells[1].inject(acsource2)

we get:

i.e. both the signals now have a delayed activation of 0.9 ms.

[apdavison]

The bug is related to the default "auto" setting for the min_delay parameter. If you explicitly set min_delay,
using setup(min_delay=1.5) (for example) then the current sources are correctly activated at 50.0 ms

[appukuttan-shailesh]

@apdavison : Any thoughts on how to fix this? How should we handle the default "auto" setting?

[apdavison]

I think the only way to handle it properly is to defer connection of the current sources to the neurons until run() is called.