How does Acts handle 1D, 2D, 3D measurements?

[chrishanw]

Hi,

for Belle II, I am working on improving genfit by replacing TVector, TMatrix, TVector3 etc and everything else that is inherited from TObject to speed up operations. To avoid introducing another dependency, for now I decided to replace TVector and TMatrix with SVector and SMatrix from ROOT::Math instead of using e.g. Eigen (and TVector3 anyway "needs" to be replaced with XYZVector in most cases for the API to the outside world).

However, now I am at a point where I am stuck. After adding template arguments for the matrix and vector sizes to half of the classes in genfit, including many of the abstract base classes, the compilation starts requiring concrete values for the template arguments. Usually this happens in the places where genfit deals with measurements of different dimensions for now, i.e. 1D measurements from e.g. wire measurements or silicon strip detectors, or 2D measurements from pixel detectors. Of course in principle these values are all known at compile time, but I did not (yet) find a way of providing them in a constexpr manner, and thus mostly functions / function calls that require template arguments.

It might just be a lack of knowledge of C++ on my side. But I anyway thought it would be a good idea to learn how
Acts deals with these different kinds of measurements. Or is this not an issue after all for Acts because of the way the hits and surfaces are defined?

By now, I am second-guessing whether sticking with ROOT and SVector and SMatrix was a good choice, I guess my life would be much easier if I could use the Dynamic "dimension" of Eigen, a functionality which the ROOT matrices don't provide as far as I can tell. But I also hoped, that because everything should essentially be known at compile time, I could take full advantage of the compile time optimisation of SMatrix algebra.

Thanks a lot in advance for your help and insights into Acts and how to potentially solve the issues that I face. I am currently working alone on this project, but it's quite likely that I'll soon involve a few more people from Belle II. So far I was confident I could do this by myself.

Best,
Christian

[andiwand]

In Acts we deal with up to 6D measurements and "project" them down to the dimension of the actual size. This mechanism also knows which dimensions are the ones that are measured.

Acts abstracts measurements away and we basically let the user deal with it but we have containers in the Examples and interfaces for the EDM in Core

• https://github.com/acts-project/acts/blob/c4761042be0b30407c6170c64bf6ee6a1911645e/Examples/Framework/include/ActsExamples/EventData/Measurement.hpp

• acts/Core/include/Acts/EventData/TrackStateProxy.hpp

  Lines 694 to 702 in c476104

  	/// Full calibrated measurement vector. Might contain additional zeroed
  	/// dimensions.
  	/// @return The measurement vector
  	/// @note Const version
  	template <std::size_t measdim>
  	ConstCalibrated<measdim> calibrated() const {
  	assert(has<hashString("calibrated")>());
  	return m_traj->self().template calibrated<measdim>(m_istate);
  	}

We usually also deal with templates there and "unroll" them during runtime with the dimension information.

Generally we store the measurements and their covariance in a dense flat double vector and address to the start of the measurement and know its size, i.e. jagged vector.