feat: Implement Hans Drevermanns z-finder and V-plot track finder

[asalzburger]

Alternatively to (re-)implementing the seed finder in traccc I suggest to add an implementation of Hans Drevermann's z-finder and v-plot. I have both the original Fortran code and a transcribed version in C++, but it would need serious adaption to fit into the traccc repository.

However, it would offer a different pattern recognition algorithm with different container access and would thus build a great alternative reconstruction path and test of vecmem.

[stephenswat]

Out of curiosity, do you have any written sources I can use to learn more about this? 😄

[asalzburger]

Yes, I do - I also have the original fortran code, let me add the links here for the notes

• Vertex Finder: https://cernbox.cern.ch/index.php/s/86tzRtMElNuCmIa
• Cleaning: https://cernbox.cern.ch/index.php/s/UzyQL8BeGK059oC

I am happy to discuss these ones, I am also sure Hans would be delighted to explain the algorithm

[stephenswat]

I see, so if I understand correctly the idea of the z-finder is to estimate a number of possible primary vertices along the z-axis, and then you can perform a much cheaper seed finding, because seed finding is much easier if you know the z-position of the vertex. Did I grok that correctly?

For the average event what is the ratio (very roughly) of primary vertices compared to the number of spacepoints?

[asalzburger]

Yes, although it has it's shortcomings, if you don't find the primary vertex (e.g. in Higgs-> gamma gamma) you miss the entire thing. A possible extension would be e.g. a scan of the vertex region, if the filter algorithm is fast enough and you can parallelise, then you can omit the z-finder and run e.g. on 1000 z-vertex bins in parallel.

For Run-3: ~60 primary vertices, ~40-60k space points
For Run-4: ~140-200 primary vertices, ~100k space points