4 1. For N log(N) strategies requiring CGAL, in CGAL versions 3.7, 3.8,
5 and 3.9-beta1, in rare case of near degeneracies, the program may
6 abort with the following error message:
8 terminate called after throwing an instance of
9 'CGAL::Assertion_exception'
10 what(): CGAL ERROR: assertion violation!
12 File: .../include/CGAL/Triangulation_2.h
14 This is due to an excessively conservative check in versions 3.7,
15 3.8, and 3.9-beta1 of CGAL that has been fixed in CGAL-3.9. Note
16 also that versions<=3.6.1 do not show that behaviour. If you
17 encounter this problem, please switch to an appropriate version of
18 CGAL. [Thanks to the CGAL development team and to Olivier Devillers
19 in particular for the quick fix.]
21 2. With JetMedianBackgroundEstimator, the computation of sigma()
22 implicitly assumes that all jets have the same area. The fact that
23 they don't introduces a bias at the level of a couple of tens of
24 percent. GridMedianBackgroundEstimator should not suffer from this
25 problem. Thanks to Christopher Young for bringing this to our
28 ----------------------------------------------------------------------
29 Other issues to be aware of
30 ---------------------------
32 1. Some algorithms can have ambiguities in the clustering steps and we
33 do not guarantee that different internal clustering strategies (or
34 different fastjet versions) will always resolve the ambiguities
35 identically. This issue doesn't arise with normal particle inputs.
39 - Some (older) versions of Pythia 6 sometimes produce massive
40 particles with pt=0; phi is then ill-defined, while the rapidity
41 is finite and we make no guarantees about how we treat the
44 - inputs that lie on a perfect grid in rapidity-phi (e.g. one
45 interpretation of a calorimeter) cause many interparticle
46 distances to be identical. In the kt, Cambridge/Aachen and anti-kt
47 algorithms, many dij can therefore also be identical. The choice
48 of which dij to take is then ambiguous and what fastjet actually
49 does may depend on: the compiler, the machine architecture
50 (because of rounding errors), the fastjet version and the
53 Physically this issue should not change the jets
54 much. Nevertheless, one might choose to break any degeneracy
55 explicity, e.g. using information from the location of energy
56 deposits in each calorimeter tower, so that the inputs are not on
59 2. For some of the plugins (listed below), the result of the clustering
60 depends on how "sort" treats equal entities. Since this is
61 supposedly undefined, FastJet has no control over the different
62 results that one may obtain using those plugins.
64 - The ATLAS-Cone plugin: At the beginning of the stable-cone search,
65 the input particles are sorted in Et. In that sort, particles with
66 an Et difference below 0.001 are considered as having the same Et
67 (see line 80 of Jet.hh), resulting in the undefined behaviour
70 A consequence of this is that, if 2 input particles have too
71 similar an Et, the stable cone search will consider them in an
72 undefined order. If the 2 resulting stable cones are too close
73 (deta<0.05, dphi<0.05) one will be accepted and the other
74 rejected. Which one depends on the ordering and is thus
75 undefined. If the 2 stable cones do not have the same constituents
76 this could affect the result of the clustering.
78 - In the TrackJet plugin, input particles are sorted in Et. If two
79 particles have the same Et (within machine precision), the order
80 in which they will be considered is undefined. Note however that to
81 have exactly the same pt, the particles are very likely to be
82 back-to-back. In that case, only the order of the clustering will
83 be affected but not the final jets.
85 Relative to the original code, we have replaced the use of 'sort'
86 with 'stable_sort' so that in case of such a degeneracy the order
87 will be the same as that of the original particles and no
88 random behaviour is to be expected. Note however that the issue
89 could still arise if one changes the ordering of the input
92 3. Copy of ClusterSequenceArea (and the other CS area-related classes)
93 is not fully implemented.
95 ----------------------------------------------------------------------
99 1. Compilation issues have been reported under windows with the VC7.1
100 compiler. We have incorporated fixes proposed by Ivan Belyaev for
101 this, but do not have access to an appropriate machine to test it
102 ourselves. (NB: it in any case works with cygwin).
1.8.8
