Templated class to fit the Point of Closest Approach (PCA) of secondary vertex with N prongs. Allows minimization of either absolute or weighted Distances of Closest Approach (DCA) of N tracks to their common PCA.

For every N (prongs) a separate specialization must be instantiated, e.g.

using Track = o2::track::TrackParCov;
o2::vertexing::DCAFitterN<2,Track> ft2; // 2-prongs fitter
// or, to set at once some parameters
float bz = 5.;           // field in kGauss
bool useAbsDCA = true;   // use abs. DCA minimizaition instead of default weighted
bool propToDCA = true;   // after fit, create a copy of tracks at the found PCA
o2::vertexing::DCAFitterN<3,Track> ft3(bz, useAbsDCA, propToDCA); // 3-prongs fitter

One can also use predefined aliases o2::vertexing::DCAFitter2 and o2::vertexing::DCAFitter3; The main processing method is

o2::vertexing::DCAFitterN<N,Track>::process(const Track& trc1,..., cons Track& trcN);

The typical use case is (for e.g. 3-prong fitter):

using Vec3D    =  ROOT::Math::SVector<double,3>; // this is a type of the fitted vertex
o2::vertexing::DCAFitter3 ft;
ft.setBz(5.0);
ft.setPropagateToPCA(true); // After finding the vertex, propagate tracks to the DCA. This is default anyway
ft.setMaxR(200);            // do not consider V0 seeds with 2D circles crossing above this R. This is default anyway
ft.setMaxDZIni(4);          // do not consider V0 seeds with tracks Z-distance exceeding this. This is default anyway
ft.setMaxDXYIni(4);         // do not consider V0 seeds with tracks XY-distance exceeding this. This is default anyway
ft.setMinParamChange(1e-3); // stop iterations if max correction is below this value. This is default anyway
ft.setMinRelChi2Change(0.9);// stop iterations if chi2 improves by less that this factor
ft.setMaxChi2(10);          // discard vertices with chi2/Nprongs (or sum{DCAi^2}/Nprongs for abs. distance minimization) 

Track tr0,tr1,tr2;                  // decide candidate tracks
int nc = ft.process(tr0,tr1,tr2);   // one can have up to 2 candidates, though the 2nd (if any) will have worse quality
if (nc) {
   Vec3D vtx = ft.getPCACandidate(); // same as ft.getPCACandidate(0);
   LOG(INFO) << "found vertex " << vtx[0] << ' ' << vtx[1] << ' ' << vtx[2];
   // access the track's X parameters at PCA
   for (int i=0;i<3;i++) {
    LOG(INFO) << "Track " << i << " at PCA for X = " << ft.getTrackX(i);
   }
   // access directly the tracks propagated to the DCA
   for (int i=0;i<3;i++) {
     const auto& track = ft.getTrack(i);
     track.print();
   }
}

See O2/Detectors/Base/test/testDCAFitterN.cxx for more extended example. Currently only 2 and 3 prongs permitted, thought this can be changed by modifying DCAFitterN::NMax constant.

The workflow is o2-primary-vertexing-workflow, the vertexing parameters are provided via configurable param pvertexer... of PVertexerParams class. The vertexing first runs an improvized version of DBSCan to group tracks losely converging to MeanVertex into time-Z clusters, then finds for each such a cluster vertices using as a seed the peaks from histogrammed tracks time-Z values.

Lot of numerical values in the params must be fine-tuned when the tracking performance will be close to final, particularly the fake ITS-TPC matches (which are prone to create fake vertices). By default the vertices will be fitted using the MeanVertex as an extra measured point.

Two groups of parameters need particular attention:

1. Debris (split vertices) reduction: after finding the vertices it tries to suppress low-multiplicity vertices in a close proximity (in Z and in time) of high-multiplicity ones. See PVertexerParams.*Debris parameters comments and PVertexer::reduceDebris method.
2. Tracks re-attachment: after finding the vertices and optionally reducing debris, it finds for each track closest vertex (in time and in Z) and refits these groups of tracks using corresponding vertices as seeds. The reason to apply this procedure is the bad time resolution of ITS standalone tracks (=ROF duration): high multiplicity vertex has high chances to steal such tracks belonging to nearby low-multiplicity ones (since it is found first). The reattachment allows to eliminate this effect of particular order of vertex finding. It should be applied only in case the debris reduction was performed, otherwise the low-multiplicity split vertices will steal tracks from high-multiplicity ones. The tracks are tested for belonging to given vertex only if they are in certain time-range from the fitted vertex time. This time-range is defined as PVertexerParams.timeMarginReattach + half of DBSCan time difference cut value PVertexerParams.dbscanDeltaT or half ITS strobe length (ROF), whichever larger.

In order to tune the parameters, a special debug output file is written when the code is compiled with _PV_DEBUG_TREE_ uncommented in PVertexer.h. It contains the (i) tree of time-Z clusters found by DBSCan (pvtxDBScan), the seeding histograms for every time-Z cluster after every vertexing iteration; (ii) the pvtxComp tree containing the pairs of vertices which were considered as close by the reduceDebris routine, their mutual chi2 in Z and time, as well as the decision to reject the vertex with lower multiplicity (2nd one); (iii) the pvtx tree with final vertices and their belonging tracks.

To see the effect of running with and w/o re-attachment, one can compare the outputs of 2 tests, e.g.

o2-primary-vertexing-workflow --run --configKeyValues "pvertexer.useMeanVertexConstraint=true;pvertexer.applyDebrisReduction=true;pvertexer.applyReattachment=false" 

and

o2-primary-vertexing-workflow --run --configKeyValues "pvertexer.useMeanVertexConstraint=true;pvertexer.applyDebrisReduction=true;pvertexer.applyReattachment=true" 

The workflow is o2-secondary-vertexing-workflow