Discussion Forum

Convergence is sometimes tricky.

What have you tried already?

Hi Alexis,

I've tried adjusting the relative tolerance by orders of magnitude, adjusting min and max time stepping in the solver to be relatively small, changed the jacobian to update on every iteration and increased max number of iterations to 25. Ive gotten the solution to converge when I begin with a non-zero current source, but the solution at time t=0 is given for every specified time from 0-600e-6 seconds for any time step. I suspect that the solver is having trouble adapting to rapid changes in the solution between time steps due to the pulses, but I'm not sure if it is indeed the problem or how to fix it. Any advice would be greatly appreciated.

Thanks, Brandon

Brandon,

it is quite obvious that you will need to apply some kind of smoothing to the waveform.

Cheers Edgar

Hi Brandon

Try a smaller, simpler model first, to troubleshoot your implementation.

For example:

• definitions: shifting your rect(t) functions by 50e-6 (lower and upper limits) so that your Biphasic_pulse starts at 0, and using a transition zone of 50e-6. this is important. you must have some finite derivative, or you will have problems. you can decrease this further as you continute to improve your model. note that probably you will have to modify the solver as you mention (change the jacobian to update on every iteration and increased max number of iterations to 25) in order for it to converge.
• geometry: disable all your geometry after TCRE location, so you only get the Gold and Paste materials
• physics: instead of a floating potential, use a terminal boundary condition ( tried just because, I have no experience using electric currents)
• mesh: coarser mesh for all, physics controlled
• time range: range(0,0.1e-5,6.0e-4) (this does not affect the solutions, they are just the times for which the solver "saves" the solution)
• tolerance: physics controlled
• solver configurations > solver 4 > Time-Dependent Solver 1 > enable direct instead of iterative (this is just from my experience modelling magnetic fields, does not mean it applies to all)

It solved in 3 minutes on my computer, in Comsol 5.4, though.

Good luck :)

Gustavo,

Thank you for all the helpful suggestions! I have implemented all of them, but my model still has trouble converging to V=0 when the current source is set to 0 A. I do see the soution getting very close to zero before the model does not converge. Is there any way to allow the solver to round the solution to zero when it is very close? Or instead of having the current source turned off should I just set it to an arbitrarily low number?

Thanks again

my model still has trouble converging to V=0 when the current source is set to 0 A

I'm not sure what you're trying to do here.

If your excitation is 0, and your initial values are 0, your solution will be 0.

What is this V? Do you set it up explicitly anywhere? I've tried setting the current source to I=0, and it runs fine.

Hi everyone, Thanks for your contributuons Brandon, when I enabled the direct solution, I was able o get result.

I am actaually working on something similar with Brandon's study. I am trying to apply charge balanced bihasic current on boundary of the electrode array. I have managed to built charge balanced current function (pw(t)), but can anyone suggest me how to implement this current pulse each electrode contacts (anode and cathode) seperately? Lets say I have 20 contacts (10 source and 10 sinks), should I generate pulse for each contact seperately or just generate one pulse for anode and one for cathode and apply them to ?

Thank you in advance.

Enver