[Question] cvode active and setting h.tstop

[xanderladd]

Hello! I have a brief question about using adaptive integration and using h.continuerun(ms) or h.tstop = ms where ms is some stopping time.

Is the expected recorded vector supposed to be a certain length? I've been using h.continuerun with and without an adaptive timestep and when I used a fixed timestep the length of the simulated voltage trace is ms / dt but when I turn on h.cvode.active(1) I get a shorter voltage trace. Is this the expected behavior? Is there a setting I am missing to ensure interpolation over a fixed set of timesteps? If there is some way to ensure the simulation is recorded at certain timesteps using cvode I will package my code into a working example but before I do that I wanted to ask this preliminary question.

Thanks in advance for all your support.

[ramcdougal]

This is expected behavior.

If you're using CVode, by definition the time steps are not evenly spaced and you could have more or less of them than in your fixed-step case. Always record time as well as voltage; e.g.

t = h.Vector().record(h._ref_t)
v = h.Vector().record(soma(0.5)._ref_v)

If you really want values interpolated to a regular interval (but you may not need this), you can always record as above and then use e.g. numpy.interp to do the interpolation after-the-fact.

One other suggestion: True and False are probably more intuitive than 1 and 0 here, so consider writing:

h.cvode.active(True)

[xanderladd]

Thanks so much for responding so quickly. I see that I will have to do the interpolation which makes sense given the design of adaptive dt. This answers my question so I'll close this.

[nrnhines]

numpy.interp to do the interpolation after-the-fact

Just a small caution about piece-wise linear interpolation. I don't know any situation where it matters, but there will be a small loss of accuracy compared to the use of https://nrn.readthedocs.io/en/latest/python/simctrl/cvode.html#CVode.solve at specific tout request times, as the latter uses the current cvode integration order interpolation formulae to calculate the state values at the tout value of CVode.solve(tout) . Unfortunately you are more or less on your own with regard to recording variables after each tout-tenter step. But a Vector.record will definitely have entries at each at each tout (except the last, unless you use the last tout value twice). Vector.record does have

vdest = vdest.record(var_reference, Dt)
vdest = vdest.record(var_reference, tvec)

variants which will accomplish recording with cvode's current order at the moment of recording, but comes at the cost of a full cvode interpolation along with an extra calculation of the system f(y, t) at each of those time points.