One more comment, thanks to running this by Aidan.
We're already doing something unnatural/incorrect with
granular potentials by using them in a velocity-Verlet
scheme. Which is evaluating forces with a velocity
that is 1/2 step out-of-date. On every step!
That error is of the same order as what happens when
you restart or continue a run with the glitch in
the velocities that I described in the last emails.
So we shouldn't worry about exact restarts or
continuations (run 100, run 100 = run 200).
All we should do is make sure the shear history values are correct
across continued or restarted runs. There might still be a bug with
that?
But I still think we don't need Peter's post-force
method to accomplish that, unless I'm missing something.
Steve