Beam read_from_file also reads spin

[MaxThevenet]

As reported by @pots007, when reading a beam from file the spin data is currently ignored. This PR proposes to fix this. The behavior is determined by the user: if do_spin_tracking is specified, the corresponding runtime components are allocated and read from file. Otherwise, they are neither allocated nor read.

[MaxThevenet]

Thanks for the review, some tests are on the way, I'll merge once we have confirmation that it works as expected.

[AlexanderSinn]

Restart test

[MaxThevenet]

Looks great! was this on GPU? If not, do you know why there is a tiny difference, don't we expect them to be equal to machine precision?

[AlexanderSinn]

Unfortunately, there are tiny differences even on CPU. I added some oscillations in x to the witness so the spin has a signal. Without that, the spin is very constant, and both cases are within 1e-6, but with the autoscale plot the restart looks different. One possible reason for this could be that the momentum is multiplied with the speed of light and its inverse.

[pots007]

Hey!
I tested this WIP fix and there are still some differences. The scripts I used are attached: I restart after 2.5m propagation and run for another 5. The plot below is from the restart point: beam parameters start from exactly values, however, for spin there is a small offset.

Scripts:
halhf_spin_test.zip

I also ran a total of 6 stages (disclaimer: I ran stage 1 with randomised spins, but this should not make any difference). The differences end up getting bigger and bigger:

Could you check if there's still any reason why the spins would be initialised different from what was read in?
Cheers,
Kris

[AlexanderSinn]

@pots007 indeed, there was an indexing issue of the spin when reading the beam from a file, related to the per slice sorting that is done.

[pots007]

Same crappy plot as above from the handover. Seems this has really done the trick, the spin tracks are now identically the same. I will run the longer test case I had.

[MaxThevenet]

Cool, thanks @AlexanderSinn! Hm it seems from these plots (also the previous ones) that the rest of the handshake may have a hiccup? A comparison between w/ and w/o restart would be helpful indeed.