testsuite::test_iaf_dc_aligned - li script for test of iaf_psc_alpha resolution independence
(test_iaf_dc_aligned) run -> compare response with reference data
test_iaf_dc_aligned.sli is a modified version of test_iaf_dc. Here,
we discuss and remove one of the problems of test_iaf_dc leading to
simulation results not independent of the computation step size
(resolution). The remaining flaw is corrected in the follow up script
test_iaf_dc_delay.sli .
The expected output is documented and briefly commented at the end of
the script.
The intention of this script is that the current starts to affect the
neuron (is present from) at time t=3ms. By default, the delay of the
connection between DC generator and neuron assumes the minimal value
of 1 computation step h. Therefore, the time stamp of the emitted
current event needs to be 3ms - h. according to the definitions of
the timing scheme, the DC generator consequently needs to emit the
appropriate event in the step 3ms - 2h -> 3ms - h.
The property /start of a device specifies in ms the earliest time at
which the device emits an event. This enables resolution independent
specification of the emission time because at any computation step
size h the device can be activated in the step start - h ->
start. The only requirement being that start - h is a time larger or
equal to the current time. Otherwise the device cannot fulfill the
request. The additional property /origin enables a shift of the
device's time axis, see the documentation of devices for details. If
resolution independence is desired origin should be specified in ms,
often 0. is a useful value. The above considerations result in the
following settings for the DC generator device:
/origin 0. % in ms
/start arrival h sub % in ms
It should be pointed out that not specifying the delay of a
connection in the .sli script, like it is done here
dc_gen neuron Connect ;
results in a simulation script that is not independent of the
computation step size. This is because at the time of creation of the
connection the minimal delay (1 computation step) is
assumed. Therefore we have to compensate for this dependency in the
start time of the DC generator as described above. The alternative
approach would be to specify a delay between the DC generator and the
Neuron larger or equal to the largest computation time step to be
used. This approach is investigated in
test_iaf_dc_aligned.sli
In earlier versions /start was defined to specify the time step t at
which the device becomes activated, which is the transition t -> t+h.
On the basis of this definition, the correct code was:
/start arrival 2 h mul sub % in ms
During the development of this script a conversion problem of device
start time expressed in ms was observed:
dc_gen << /start 1.8 >> SetStatus
dc_gen GetStatus /start get == --> 1.800000e+00
dc_gen << /start 2.8 >> SetStatus
dc_gen GetStatus /start get == --> 2.799000e+00
For the purpose of this script, the problem was temporarily fixed by
adding an epsilon 0.0001
/start arrival 2 h mul sub 0.0001 add % in ms
The problem was fixed by Gewaltig 3.3.2004 and the code of this test
script changed accordingly, Diesmann .
February 2004, Diesmann, Gewaltig, Plesser
/var/www/debian/nest/nest-simulator-2.20.0/testsuite/unittests/test_iaf_dc_aligned.sli