This pages describes changes to how we decide which associations are most important

 Sorting track-hit associations

Currently all the associations are stored on a single collection of AcdTkrHitPoca objects.  These are sorted by active distance.   This means that AcdTkrHitPoca's from later tracks can be placed before ones from the best track.  

In the new algorithm the  AcdTkrHitPoca are sorted by Veto esitmator.  This is constructed by combining terms for the geometry and terms for the signal size
   VetoSigmaProj = active3D < 0 ? 0 :  active3D / activeDist3DErrProj;
   VetoSigmaHit = 2. / pocaData.m_cosTheta < totalMips ? 0 :  (( 2. / pocaData.m_cosTheta) - totalMips)/0.45  [ tiles ]
   VetoSigmaHit =1.5 < totalMips ? 0 :  (1.5 - totalMips) [ ribbons ]

Both terms are constructed so that:

  1. They are 0. for anything that looks like a good MIP (ie, track point at hit tile, signal is at MIP peak)
  2. They become positive as the event looks less MIP like (ie, track missed hit tile, signal is below MIP peak) 

Then we can combine them in quadrature
  VetoSigma = sqrt( VetoSigmaProj*VetoSigmaProj + VetoSigmaHit*VetoSigmaHit)

This means that for tiles there is some contribution for expected signals less that a mip.  From ribbons this contribution only gets up to 1.5 sigma since very smal signals are common.

 Sorting track-gap associations

Currently all the associations are stored on a single collection of AcdTkrGapPoca objects.  These are sorted by active distance.   This means that AcdTkrGapPoca's from later tracks can be placed before ones from the best track.  

In the new algorithm the  AcdTkrGapPoca are sorted by VetoProbabliity.  This is constructed by adding terms for the geometry and terms for the signal size
   VetoSigmaProj = Intergal of  (distance / activeDist3DErrProj ) across the gap, then converted to sigma.
We take the integral instead of just using the sigma to the near edge because care about the chance that the track went into the gap, not the chance that it missed the particular tile in question.  This reduces the esimator if the gap is very small, or the error bars are large.
   VetoSigmaHit =1.5  [ gaps covered by ribbons ]
   VetoSigmaHit = 0 [ gaps not covered by ribbons ]

Both terms are constructed so that:

  1. They are 0. for anything that looks it really went into a gap  (ie, high energy track points at gap)
  2. They become positive as the event looks less MIP like (ie, track missed gap, gap is covered by ribbon)

Then we can combine them in quadrature
  VetoSigma = sqrt( VetoSigmaProj*VetoSigmaProj + VetoSigmaHit*VetoSigmaHit)

 Associations between tracks and hit tiles

In AcdValsTool there is a test if a track comes close to a tile,  this is defined as -300./sqrt(CAL_EnergyRaw/100). 

When looping over all tracks, the algorithm then selects the AcdTkrHitPoca with the largest energy that passed the distance cut.  On the other hand, for the best track the algorithm simply takes the largest active distance.

For each track the new algorithm sorts all AcdTkrHitPoca by increasing VetoSigma.  In AcdValsTool we select the AcdTkrHitPoca with the smallest veto sigma, making sure to only use those AcdTkrHitPoca that actually have a hit in the tile. 

So, if a track comes close to two tiles with signal, we may expect to see a different tile selected.
Note however that with the new algorithm for the loop over all tracks energy will always be equal or lower, while for the best track the active distance will always be equal or lower.

  Best Track

This plot shows old v. new for AcdTkr1ActiveDist (ie, the best active distance for the best track) for all events.  
 

 
The huge majority of events lie on the diagonal, ie, they are unchanged.   Most of the remainder lie above the diagonal, ie, we have selected an association with a larger miss distances, (probably b/c it has a larger signal).  However, we do see some events below the diagonal.  Looking more closely we can distinguish events where the POCA occurs at the head of the track from those events where it occurs upstream.
  

AtTrackHead

Upstream


 
So, if the POCA is upstream from the track head (right plot) things look sensible so far (almost all the entries are the same, in a few cases a different tile is picked b/c it has larger signal, or the error projection is larger, either of which makes it more likely to veto the event).

 
On the other hand, if the POCA occurs at the track head things get confused by two effects. 

  1. Some fraction of the time new calculation gives a sigma > 10000 and is ignored.   Only happens if the orignal active distance was < -400.  No great loss.  This explains the stripe at -2000 in the left plot.
  2.  Sometimes the POCA moves slightly because of minor changes to how the edge case is handled, this explains the off-diagonal entries in the left plot.

This plot shows old v. new for  log10( AcdTkr1ActDistTileEnergy) (ie, the Energy of best active distance for the best track) for all events. 


 
Again, the large majority of events lie on the diagonal.  The stripe at 0 is an artifact from effect 1) above.  Aside from that we see some events different energies.  This is b/c of the changes in select the "best" association.

 All Tracks

As mentioned above, In the case for all tracks AcdValTool first applies an energy scaled cut on active distance, then picks the association with the largest signal.  This means that we expect the Energy in the new case to almost always be less than or equal to the previous value. 


 
 
 

 
 
 There is a special case when two tracks both point at a tile with a MIP-like signal.  With the new logic both associations have a VetoSigma value of 0.  Therefore the one from the earlier track is selected.  With the old logic whichever track had a larger active distance was selected.  This can be seen in the large excess in the 0 column in plot below, which compares the track number of the best association.   In either case we will reject the event, and it is probably wise to use the better tracking information. 

 

 

 Associations between tracks and hit ribbons

In AcdValsTool we simply select the ribbon with the largest active distance. 

With the new algorithm this changes a little bit since we are now using the active distance / the error projection. 

Best Track

Most of the differences come from cases where the POCA occurs at the head of the track. 


 

 All Tracks


 
 

 Assocations between tracks and gaps

In AcdValsTool we simply select the gap with the largest active distance.

 With the new algorithm this changes a little bit since we are now using the active distance / the error projection. 

Best Track

For the best track, if we used the tile intersection to calculate the distance to the gap we expect to see almost exactly the same results since the gap esimator is dominated by the distance and all the associations are using the same tracking errors.
One the other hand, if we used the ribbon to calculate the distance then we expect small changes, (depending on the geometry).  The latter case explains the horizontal stripe at positive values in this plot: 

 

 All Tracks

Since different tracks can have slightly different error projections selecting by the gap estimator can sometimes pick a value with slightly higher distance (and correspondingly higher error bars).  This explains the entries above the diagonal in this plot.

The entries below the diagonal come in because we are consider all the gaps, not just the ribbons, in the new version. 

  • No labels