Spin correlations broken in decays.

[Juergen Reuter]

Spin correlations do not work. In all cases with factorized decays, I get the pure phase space (i.e. w\ spin correlations). Damned!

[Juergen Reuter]

Casc_dec output

[Juergen Reuter]

This should be definitely fixed before the MC school!

[kilian]

After the fixes in r5348, the output looks ok. The problem was a rest_frame flag for decay processes, which should not be set at all when generating a cascade.

There are still two potential issues, with a minor impact:

• there is a high number of excess events in the full-process simulation. Maybe this is bogus (since the results look reasonable), but should be investigated.
• there are small differences between the 2.1 and 2.2 results which may or may not be significant. What really should be done (and hasn't been done for 2.1 either) is a detailed and high-statistics investigation of the spin correlations. This is, however, beyond the scope of the current example, and should probably be postponed.

[Juergen Reuter]

WK, what exactly do you have in mind? What should be done here? For point 1 (the excess events) this is not clear to me. For point 2, this is something we have never investigated. Maybe suited for a summer student!? Is this still for 2.2.0, or for a later release?

[kilian]

1. I'll do that, but only after some bigger issues have been resolved. Therefore P4.
2. Yes, it's actually a good topic for a bachelor thesis.

[Juergen Reuter]

Point 1 should be looked at before the release, point 2 will be addressed (partially) by JG (summy).

[kilian]

Looking into the excess-event issue.

[kilian]

Printing intermediate values of weights doesn't show anything suspicious. The formulas appear to be correct. Apparently, for this process with O(100) channels and rather singular kinematics, the sampling during integration easily misses some actual peaks, although they are not relevant for the integration.

Regarding the example, I just inserted a safety_factor=5 which reduces the efficiency accordingly, but eliminates most of the excess.

On the long run, we should look into the problem from a broader perspective. Created ticket #639 for this.

Regarding the current issue, there remains the major task of checking spin correlations in general.

[Juergen Reuter]

There has been substantial work on the checks of the spin correlations by Jan Gerken, see his summer student work. However, there are still open issues, e.g. the questions why the decay process does not give the correct invariant mass distribution.

[Juergen Reuter]

Some previous modification broke this, found by Carsten Bittrich.

I've fixed this already, but not yet checked in. Thinking about a possible unit test.

The fix will also add a Theta_star observable, which was missing. The existing Theta_RF observable will be removed, the way it was defined it's useless.

[Juergen Reuter]

WK, please also remember to adapt the SINDARIN reference in the manual once the Theta_RF vs. Theta_star changes are checked in!

[kilian]

overlooked this one. OK, nevertheless...

[kilian]

Tentative fix in r6494. The distribution is now checked in test spincor_1, which evaluates the Theta_star observable (which is also new) for correlated and isotropic decay. [I can't promise that this test is numerically safe, but something along this lines was necessary.]

The changes raised issues with recreating event transforms when reading events from file, see #698. I had to temporarily disable (XFAIL) tests lhef_7 and lhef_8; those need to be reinstated before release. Ranking this ticket down, but not closing yet.

[Juergen Reuter]

Jan Gerken confirmed that the issues he found in SUSY cascades are solved by r6494. So it is the left-over remnant from #698 that prevents closing this ticket (and the two XFAIL tests).

[kilian]

The XFAILs of lhef_7 and lhef_8 were unrelated to this ticket. Since we currently don't know any more problems with spin correlations, I close.