Original authors:

Michael Feindt (Email: M.Feindt@cern.ch), Christof Kreuter, Oliver Podobrin

Now maintained by:

Zoltan Albrecht (Email: Thomas.Albrecht@cern.ch)

Gary Barker (Email: Gary.Barker@cern.ch)

Michael Feindt (Email: Michael.Feindt@cern.ch)

Markus Moch (Email: Markus.Moch@cern.ch)

Last updated: 14-1-2000

BSAURUS News

BSAURUS Performance Plots

What BSAURUS does

• the rapidity algorithm used e.g. in the DELPHI B* analysis , for the observation of the B* Dalitz decay and a measurement of the b fragmentation function , providing an estimate of the B energy and direction,
• inclusive secondary vertex search and fit , an improved version of those used for the DELPHI B** analysis and the preliminary analyses on B** and B_S ** , Sigma_b and Sigma_b* , and the possible evidence for B pi pi resonances , providing extimates of the secondary vertex coordinates with covariance matrix and a calculation of the decay length,
• a calculation of the probability for each track that it stems from B decay rather from fragmentation, based on a neural net essentially exploiting vertex fit probabilities, rapidity and energy,
• an estimate of the B hadron charge with error estimator, using the track probabilities and binomial distribution,
• a neural network to distinguish between b and b-bar hemispheres combining jet charge, B vertex charge and its error, lepton p_t, charge of the leading fragmentation particle, fast kaons and protons from the secondary vertex. This may be used for asymmetry measurements and as tagging tools for B mixing analyses. Cuts on the continuous output variable allow to select a working point in the efficiency - purity plane. Work is going on to optimize the performance.
• NEW 9/Apr/97 NEW: a B+ tag exploiting information from both hemispheres, which may be used for b+ lifetime and B+ branching ratio measurements.
• some useful generator information is put into COMMON blocks to ease access necessary to a large number of B-analyses.

Which analyses can benefit from BSAURUS ?

• B+ lifetime
• B+ --> lepton branching ratio. Interesting because there is no need to cut hard on the p_T of the lepton, since the B sign fixes the b->l sign.
• B+ --> K+, K- branching ratios
• B+ fragmentation correlations -- leading particle correlation
• B+, B0, B_s, Lambda_b (Xi_b) production fractions
• B* isospin splitting
• B** --> B(*) pi, B_s** --> B(*) K, Sigma_b(*) --> Lambda_b pi, B**, B_radial (?) --> B pi pi reanalyses
• B_d mixing
• B_s mixing
• b asymmetry

What do you have to do ?

CALL BSAURUS

once per event fills 3 COMMON arrays, one for the whole event, one for the two hemispheres, one for each track. The track index of the latter corresponds to the VECSUB/SKELANA pointer, if these arrays are filled before the call to BSAURUS.

The COMMONS are described below.
 

Where do you find the code ?

What information do you find in the COMMONS ?

*     Array                Nval  description
*     -----------------------------------------------------------------

*     Quantities related to single particles BSPAR
*     =============================================
*     Array                 Nval  description
*     -----------------------------------------------------------------
*     BSPAR(IBP_HEM,IPART)   1   hemisphere (or jet) number
*     BSPAR(IBP_IYB,IPART)   1   index of the matching VECSUB track
*     BSPAR(IBP_QUAL,IPART)  1   track quality: 20(if isrt=0) + 10(if from
*                                interaction) + 1(if has ambiguous hits)
*     BSPAR(IBP_Y,IPART)     1   rapidity with respect to jet
*     BSPAR(IBP_PRP,IPART)   1   probability to fit primary vertex
*     BSPAR(IBP_PRS,IPART)   1   probability to fit secondary vertex
*     BSPAR(IBP_BRP,IPART)   1   Borisov probability in rphi
*     BSPAR(IBP_BZ,IPART)    1   Borisov probability in z
*     BSPAR(IBP_B3D,IPART)   1   Borisov probability in 3D
*     BSPAR(IBP_BSRP,IPART)  1   Borisov probability in rphi w/r secondary vtx
*     BSPAR(IBP_BSZ,IPART)   1   Borisov probability in z  w/r secodary vtx
*     BSPAR(IBP_BS3D,IPART)  1   Borisov probability in 3D w/r secodary vtx
*     BSPAR(IBP_NET,IPART)   1   net output (-->0 for good fragmentation,
*                                            -->1 for good B decay product)
*     BSPAR(IBP_DNET,IPART)  1   cascade D track net output
*     BSPAR(IBP_NRP,IPART)   1   number of rphi VD hits
*     BSPAR(IBP_NZ,IPART)    1   number of z VD hits
*     BSPAR(IBP_KST,IPART)   1   momentum of track in B CMS
*     BSPAR(IBP_THETST,IPART)1   helicity angle of the track
*     BSPAR(IBP_BSAU,IPART)  1   pointer to BSAURUS internal particle array
*     BSPAR(IBP_SV,IPART)    1   1--> trk in sec. vertex 0--> trk not in SV
*     BSPAR(IBP_TRKE,IPART)  1  Track energy
*     BSPAR(IBP_TRKM,IPART)  1   Track mass
*     BSPAR(IBP_TRKP,IPART)  1   Track momentum
*     BSPAR(IBP_TRKQ,IPART)  1   Track charge
*     BSPAR(IBP_TRKL,IPART)  1   Track length
*     BSPAR(IBP_ERRE,IPART)  1   Error on track energy
*     BSPAR(IBP_DELPP,IPART) 1   Delta(P)/P
*     BSPAR(IBP_DFLBS,IPART) 1   charge/decay flavour correlation net for B_s
*     BSPAR(IBP_DFLB0,IPART) 1   charge/decay flavour correlation net for B0
*     BSPAR(IBP_DFLBP,IPART) 1   charge/decay flavour correlation net for B+
*     BSPAR(IBP_DFLLB,IPART) 1   charge/decay flavour correlation net for LamB
*     BSPAR(IBP_FFLBS,IPART) 1   charge/fragm. flavour correlation net for B_s
*     BSPAR(IBP_FFLB0,IPART) 1   charge/fragm. flavour correlation net for B0
*     BSPAR(IBP_FFLBP,IPART) 1   charge/fragm. flavour correlation net for B+
*     BSPAR(IBP_FFLLB,IPART) 1   charge/fragm. flavour correlation net for LamB
*     BSPAR(IBPMC_KF,IPART)  1   MC Lund KF code
*     BSPAR(IBPMC_KFP,IPART) 1   MC Lund KF parent code
*     BSPAR(IBPMC_TYP,IPART) 1   MC code
*                                0 : unidentified
*                                1 : ordinary fragmentation
*                                2 : fragmentation partner of B-hadron
*                                3 : decay of primary (not weak) C
*                                4 : decay of D*
*                                5 : decay of weak B
*                                6 : decay of excited B
*                                7 : decay of weak C
*                                1x : same as for x but decay of
*                                    long living particles (e.g. K_l)
*                                2x : same as for x but from
*                                     hadronic interaction or photo conv.
*                                -/+ : b/bbar quark
*                                      6       : content of primary B
*                                      3,4,5,7 : content of weak B
*
*     BSPAR(IBPMC_KST,IPART)     1 MC momentum of track in MC B CMS
*     BSPAR(IBPMC_THETST,IPART)  1 MC helicity angle of the track
*     BSPAR(IBPMC_TRKP,IPART)    1 truth track momentum
*     BSPAR(IBPMC_TRKE,IPART)    1 truth track energy
*    ---------------------------------------------------------------------

Additional wishes or ideas, problems or questions?

The success of BSAURUS will depend on the feedback of users. We will be happy to include other physics ideas and features into the program or optimize the algorithms where appropriate.

BSAURUS News

16 Mar 1997 Bsaurus is interfaced to KAL by Anders Borgland -- in just one day at the weekend. Thanks, Anders! The convention for BSAURUS variables in KAL can be found in here

9 Apr 1997 BSAURUS is in the official libraries. You may use the command dellib bsaurus or the f77 option -lbsaurusxx to link the library. You don't have to do anything special

9 Apr 1997 A first version of the B+ tag is introduced in the (new) variable BSHEM(IBH_BPLUS,IHEM).

10 Apr 1997 interference between electron id. neural net and BSAURUS neural nets removed, improved b-bbar net, some small bug fixes

24 Apr 1997 A wrong array dimension in YBREC had the effect to set NYCUT to zero after the first event with more than two jets was processed. After that the algorithm was effectively switched off and therefore the status word ISAURUS was always equal to zero. This error occured only on HP-UX due to different memory alignment on OSF1. This does not imply that the results obtained on OSF1 are correct.

29 Apr 1997 Some more bug fixes due to usage of B confidence method. All net outputs should improve...

1 May 1997 Development going on on new variables to distinguish direct B and D decay products: lifetime signed Borisov track probabilities with respect to the fitted secondary vertex. Direct b decay particles should have negative lifetimes (i.e. in our convention values near 2), whereas D decay particles should have positive lifetimes (i.e. in our convention values near 0). Based on those quantities Borisov probabilities for the fragmentation part of the jet (i.e. fragmentation/b neural net output <0.5 ) and the B decay part of the jet (i.e. fragmentation/b neural net output >0.5 ) are calculated with respect to the primary and the secondary vertex. In particular decays with a long D lifetime should have a low secondary vertex Borisov probability of the B part of the jet. For those particles also a "dipole moment" is calculated which should somehow be proportional to the D lifetime. The code is still under development and will probably become standard next week.

27 May 1997 Many more new variables, especially for Monte Carlo truth. Thanks to Jong Yi for the implementation! See the updated variable list above.

27 Nov 1998 New variables: new Borisov b-tagging, new dipole moment, output of the kaon net and proton net (soon), new net for separation of tracks from fragmentation and B decay, new flavour net (soon), charge correlation coefficients of best lepton and kaon candidate in hemisphere, for Monte Carlo truth: the quark flavour at production time and a new oscillation flag. Thanks to Christian Weiser for some useful routines! Look also at the updated variable list above.

9 Feb 1999 Improved kaon and proton net variables stored in PINFO(8,PARTICLE) for kaons and in PINFO(9,PARTICLE) for protons. The net information is even available by direct call of the MACRIB package. For further details see MACRIB (ps).

27 Sep 1999  version of BSAURUS released on ~/bcteam/export  directory . New ,  improved, versions of the track net, flavour net and  B^+  net.  Inclusion of new networks  to :   identify tracks  from  the  cascade D vertex  in B-decays,  B^0 identification, and an
additional flavour net that includes opposite hemipshere information.
 

20 Oct 1999  version of BSAURUS including B-D track separation, decay and production flavour (b bbar separators) for given B species hypotheses for single tracks and complete hemispheres.

14 Jan 2000  Change default version of BSAURUS on the afs export directory which will, in due course, be made into the library. The new version contains new track level output word BSPAR(IBP_IYB,IPART). This gives the matching VECSUB track index for BSAURUS (ie VECTOP) track IPART. Use of this word is essential for all users running BSAURUS in the SKELANA framework. The new version also contains hemisphere output word, BSHEM(IBH_QUAL2,IH), giving the hemisphere quality in a continuous variable. This is then useful for eg) the weighting of BSAURUS hemisphere net output distributions. Finally, the new version contains some additional network outputs at the track and hemisphere level that have been recently developed for various analyses. For a short definition go to What information do you find in the COMMONS? A more detailed definition will eventually appear in a BSAURUS description DELPHI note. 

 1)  For the track net, 'signal' is defined as all tracks originating from a displaced secondary vertex in B-decays and 'background'
      constitutes tracks from the primary vertex. The perfomance  distribution is based on  tagging  the signal tracks in bbbar Monte Carlo.
Track net output distribution
Track net performance

2) For the flavour net, 'signal' and 'background' are defined as hemipsheres  originating from a b or bbar quark respectively. The
     performance  distribution is based on tagging signal hemispheres in bbar Monte Carlo.
Flavour net output distribution
Flavour net performance

3) For the B+ net the 'signal'  hemispheres are those containing a B+ decay and the 'background'   is everything else.  The
    perfomance plot is based on tagging signal hemispheres in bbar Monte Carlo.
B+ net output distribution
B+ net performance

4)  The f performance of the lavour net  including  information from the opposite hemipshere in training  is shown compared to the
      standard BSAURUS flavour net performance.
Performance of the flavour net including opposite hemisphere information

5)  Now also existing: A neural net that tries to distinguish tracks from the cascade B->D (or Dbar) decay and tracks from the B decay directly. A neural net that estimates the correlation between the electric charge of a charged particle and the decay quark flavour of the hemisphere for a given B species hypothesis (i.e. B+, Bs etc). And another one for fragmentation charged particles to estimate the production quark flavour (b or bbar). For all tracks in a hemisphere these are combined in corresponding HEMISPHERE. Very useful for the construction of an optimal flavour net and for MIXING ANALYSES.
Performance of B/D track separation, track decay and fragmentation flavour nets and hemisphere decay and fragmentation flavour variables for different B species

6)  Using all this a new optimised hemisphere flavour net is created (FLAV45). The performance is even better than that of BH_FLAV, and especially the variable BH_FLAV45 (this hemisphere) - BH_FLAV45 (opposite hemisphere) is really nice. This should be used for a FORWARD BACKWARD ASYMMETRY measurement.
Performance of optimised flavour nets BH_FLAV45 in one and both hemispheres, compared to original (ok, just one month old...) BH_FLAV, and jet charge.

7)  A production flavour for Bs analyses from the likelihood combination of opposite hemisphere FLAV45 and this hemisphere's Bs fragmentation flavour net. This should be used for a Bs MIXING measurement.
Performance of the Bs production flavour variable for B_s hemispheres

8)  A production flavour for B0 analyses from the likelihood combination of opposite hemisphere FLAV45 and this hemisphere's B0 fragmentation flavour net. This should be used for a B0 MIXING measurement.
Performance of the B0 production flavour variable for B_0 hemispheres

9)  B hadron identification nets using only one hemisphere information.
Performance of the B0, B+ , Lambda_b and B_s enrichment algorithms. The normalisation is as in the 94/95 Monte Carlo, i.e. with only about 7% B_s (probably an underestimate!)

10)  B hadron identification nets also using opposite hemisphere information (exploiting charge correlations).
Performance of the B0, B+ , Lambda_b and B_s enrichment algorithms. The normalisation is as in the 94/95 Monte Carlo, i.e. with only about 7% B_s (probably an underestimate!)

11)  Mixed B_s selection net, all the rest background.
Performance of the mixed B_s net. The normalisation is as in the 94/95 Monte Carlo, i.e. with only about 7% B_s (probably an underestimate!)

For details of the BSAURUS  outputs a  note is in preparation. Please refer any  interrim questions to Gary.Barker@cern.ch

Return to DELPHI b/c physics superteam working pages

Michael.Feindt@cern.ch, Christof.Kreuter@cern.ch, Oliver.Podobrin@cern.ch) 

Created : 14 Mar 1997 mif

Last update: 27 Sep 1999 mm and gb