strange behavior of Opamp / transistor circuit

For another project the students had to set up a LDO with and Op Amp and a transistor (i know, there is already a volt. regulator available).
It works quite well in 0.4.13-SR5 / 0.4.14-SR3 (WIN10). I again tried to reduce the simulation - here it is still a kind of complex.
The simu- and hex-file is attached. Changing the duty cycle changes the input voltage U_ref of the Op Amp and this changes U_out to the 2 times U_ref (works up to the limit of the transistor of about 4.0 ... 4.3V). The duty cycle can be changed by the buttons.

However there were 2 strange things:

1. Changing the ReactStep behaves counterintuitive: there seem to be a sweet spot for ReactStep = 5. For 5 I get about 30...50% real speed, when the duty cycle is about 0..40%. Lower or higher values result in slower real speed down to 1%.
   
2. The simulation not really works in 0.5.15-RC2/RC3. It only starts, when the switch nearby the RC element is open or when the 5V source is off. When the 5V source is of it even shows valid values - but it's not a real world situation. When the switch is closed the 328 pauses working. On  0.5.15-RC1 it works until the scope is connected.
   

The second effect could be related to changing / improving the cores and the scope.  For the first effect it could be the implementation of the Op Amp.

Changing the ReactStep behaves counterintuitive: there seem to be a sweet spot for ReactStep = 5. For 5 I get about 30...50% real speed, when the duty cycle is about 0..40%. Lower or higher values result in slower real speed down to 1%.

Non-linear component don't have a fixed time to converge to the correct solutiuon.
It can happen that it converges faster in certain conditions, just by chance.

The simulation not really works in 0.5.15-RC2/RC3. It only starts, when the switch nearby the RC element is open or when the 5V source is off. When the 5V source is of it even shows valid values - but it's not a real world situation.

When the 5V source is of, then it is working throught the BE diode.
But then is not working as a voltage regulator, just as analog 2x output for the PWM signal.

When the switch is closed the 328 pauses working.

I think it get stucked trying to solve the non-linear feedbacks forever.
I will try to solve this for the next RC.

I'm working in this right now, just created a discussion to talk about it:
https://simulide.forumotion.com/t66-non-linear-components#214