Fix a drift integration mistake in the coupling of hydro and gravity

Store a_grav back in the xpart to avoid an incorrect drift operation where the gpart->a_grav was sometimes accessed after the next step had already started accumulating accelerations. That was a bad mistake introduced by my change to the gravity mesh integration, which primarily affected the Planetary calculations due to their choice of system of units.

In the cosmology system of units, the access to an incorrect value was mitigated by the fact that (in this system) G>1, making the incorrect accelerations (a/G) much smaller than the correct values but not leading to a crash, just to a slightly too small drift (which is a small effect as the hydro forces are usually larger for all interesting regimes).

However, in "planetary units", G<<1, making the incorrect accelerations much too large and leading to a very wrong drift. Then to planet evaporation.

added SPH bug labels

@jborrow could you check that the Evrard collapse is still fine? (or rather fine again)

@jkeger could you check whether this indeed solves your problem?

Yep looks good! Just running some duplicate tests to be sure

Confirmed, all back to normal

added 1 commit

• 90939bcf - Fix comment typos

Compare with previous version

@jborrow sorry to add to your plate but one of your low-res EAGLE-25 would also be great.

running now

mentioned in merge request !1190 (merged)

Evrard has completed and shows significant differences:

Old

New

The only problem is that in the new case it looks worse...

Well, the shock looks worse. The central region is much more smooth (see r=5e-2)

Is the new one back to how it was before the original change was made? b74398ba

old == current master? or old == version from september? Current master is wrong.

Old I guess is a version from September.

I guess it is unexpected that those two are different. They use exactly the same ICs, so should be very reproducible.

grad_h has changed in SPHENIX in between the two plots.

That's a much older version than September though. The gravity solver is very different beyond just the time integration. Setting Gravity:MAC:geometric would restore the old behaviour.

Do the plots here match the ones from your draft?

merged

mentioned in commit d0e9c6f5