Is this a numerical error or just how math works? ...
# xscheml
Luke Harold Miles
03/22/2024, 9:19 PMIs this a numerical error or just how math works? (integral(rollingaverage(x)) very different from integral(x))
Luke Harold Miles
03/22/2024, 9:22 PMHmm ravg(integ(x)) also very different so I'm pretty sure it's floating point stuff not how math actually works. Not a bug I think...
Luke Harold Miles
03/22/2024, 9:22 PMraw file if of use
Luke Harold Miles
03/22/2024, 9:30 PMoh it's probably me misunderstanding how ravg() or integ() work cuz if I scroll the x axis right slightly then the plots all look the same
s
Stefan Schippers
03/22/2024, 9:50 PM@Luke Harold Miles For low current values there are numerical oscillations in simulations. Add this:
.options method=gear🙌 1
Stefan Schippers
03/22/2024, 9:52 PM@Luke Harold Miles integration and ravg() are done on the interval shown in the graph.
Stefan Schippers
03/22/2024, 10:02 PMAlso do not expect numerical simulations to show reliable data in the Femto-Coulomb range, unless -may be- you tune the precision parameters of the simulator, and set the above method=gear parameter (and cross fingers)
👍 1
Stefan Schippers
03/22/2024, 10:12 PMIf you integrate a current of a nA over 100ns you get 1e-16C of charge , ~600 electrons. Too low for a concentrated parameters circuit solver.
👍 1
Stefan Schippers
03/22/2024, 10:15 PMAnd this is the numerical oscillation of the i(vmt2) voltage source: zoom in into the waveform you will see it.
👍 1
l
Luke Harold Miles
03/25/2024, 8:36 PMThank you very much. This knowledge is very valuable to me. I am still newbing along 😅
Luke Harold Miles
05/04/2024, 5:42 PM@Stefan Schippers how much charge for the numbers to be somewhat trustworthy? 10^-15C? 10^-14C? (Assuming a very reasonable and simple circuit.)
If you integrate a current of a nA over 100ns you get 1e-16C of charge , ~600 electrons. Too low for a concentrated parameters circuit solver.
s
Stefan Schippers
05/04/2024, 11:56 PMDifficult to give a number, Charge conservation also depends on length of simulation. A very long transient simulation with millions of integration timesteps will accumulate considerable charge errors due to numerical approximations. I can't give you a number, but I often have seen charge conservation violations in simulations.
4 Views