SCR followed by resistor fails to converge; converges with just wire

[nabijaczleweli]

2023-03-16.17-56-40.mp4

Minimal circuit (resistor value doesn't matter, nonexistence of the gate resistor also doesn't):

$ 65 0.0000049999999999999996 10.20027730826997 50 5 50 5e-11
177 416 48 480 48 1 0.6138908161995182 1.7000000000000002 0.000049999999999999996 0.002 50
R 416 48 416 16 0 0 40 5 0 0 0.5
R 480 -32 448 -32 0 0 40 3.3 0 0 0.5
g 528 48 544 48 0 0
r 480 48 528 48 0 2200
s 480 16 480 -32 0 1 false
o 3 64 0 4099 0.0000762939453125 409.6 0 2 3 3

When you toggle the switch, the simulator crashes into "Convergence failed!". If you replace the resistor with just a wire, it behaves as-expected and ~infinite current flows.

I played around with it, and if you increase the top voltage to 4.5V it starts to converge again (but the time step becomes 76ps).

[esaruoho]

I've submitted a fix for this in pfalstad#297 (targeting the v3-dev branch).

Root cause: The SCR's doStep() method was evaluating the on/off state (switching the anode resistance between 0.0105 and 100K ohms) on every Newton-Raphson sub-iteration. When a series resistor is present, the current feedback through that resistor causes the state to flip-flop between on and off on alternating iterations, preventing convergence.

Fix: Move the state evaluation to startIteration(), which is called once per timestep before the Newton-Raphson loop. This ensures the resistance stays constant during the iterative solve. The state is re-evaluated at the next timestep based on converged currents. This matches the pattern already used by TriacElm and DiacElm, which both evaluate their state in startIteration().