Incorrect summation of synaptic currents for IF_curr_exp model

[apdavison]

Hajime Yamauchi reported:

I’m currently implementing a simulation model which uses leaky integrated and fire neurons with a type of PSCs whose profile would be fitted with a bi-exponential function. An equation of this type should have two time constants, namely “tau_syn_rise” and “tau_syn_decay.” <e.g., exp(-(t-t0)/tau_syn_decay) - exp((t-t0)/tau_syn_rise)>. Some of you may notice I’ve asked a related question to the NEST community because there is no standard model which accepts two PSC time constants in its synaptic specification. One solution I was given was to connect two identical synapses to a neuron while using different tau_syn values and reversed weights. Regarding the weights, they are the same size but one is positive, and the other is negative. Originally, I was told I should use the iaf_psc_exp_multisynapse model, but I’ve confirmed that the same result will be obtained if I I use the standard iaf_psc_exp model with tau_syn_ex for tau_syn_decay, and tau_syn_in for tau_syn_rise.

However, if I reconstruct this in PyNN (with NEST as its base), no activity is observed. The membrane potential is completely flat. If I change the spike time of one of the synapses, NEST and PyNN give the identical result. Am I coding wrongly?

Here I give two sample codes (related part only) :
NEST 2.6.0
PyNN 0.8dev2
Python 3.4

—————NEST—————

neuron = nest.Create('iaf_psc_exp',1,{'tau_syn_ex':12., 'tau_syn_in':4., 'E_L':-65.,'V_reset':-65.,'V_m':-65.,'V_th':-50.,'t_ref':0.,'C_m':1000.,'tau_m':20.})

nest.SetStatus(spike_sourceD,'spike_times',[[40.]])
nest.SetStatus(spike_sourceR, 'spike_times',[[40.]])

nest.CopyModel('static_synapse', 'decay_syn',{'weight':200.})
nest.CopyModel('static_synapse','rise_syn',{'weight':-200.})

nest.Connect(spike_sourceD, neuron,'all_to_all', syn_spec = 'decay_syn')
nest.Connect(spike_sourceR, neuron,'all_to_all', syn_spec = 'rise_syn')

—————PyNN—————

neuron = sim.Population(1, sim.IF_curr_exp(tau_syn_E=12., tau_syn_I=4.))

spike_sourceD = sim.Population(1, sim.SpikeSourceArray(spike_times=[40.]))
spike_sourceR = sim.Population(1, sim.SpikeSourceArray(spike_times=[40.]))

sim.Projection(spike_sourceD, neuron, connector=sim.AllToAllConnector(), synapse_type = sim.StaticSynapse(weight=0.2), receptor_type='excitatory')
sim.Projection(spike_sourceR, neuron, connector=sim.AllToAllConnector(), synapse_type = sim.StaticSynapse(weight=-0.2), receptor_type='inhibitory')

[apdavison]

First, my apologies for the very long delay in investigating this. It turns out to be a bug in the iaf_psc_exp_ps model in NEST, which is the model PyNN uses by default for IF_curr_exp. I have created a ticket for this in the NEST issue tracker: nest/nest-simulator#368

As a workaround, in PyNN you can use the spike_precision argument to setup(), i.e.:

setup(spike_precision="on_grid")

This will cause PyNN to select iaf_psc_exp (which works correctly) instead of iaf_psc_exp_ps.

[apdavison]

Now fixed in the NEST master branch (nest/nest-simulator#810)