KinKal implements a kinematic Kalman filter track fit, as described in https://www.epj-conferences.org/articles/epjconf/abs/2024/05/epjconf_chep2024_06001/epjconf_chep2024_06001.html. The primary class of KinKal is Track, which shares the state describing the fit inputs (hits, material interactions, BField corrections, etc), and owns the result of the fit, and the methods for computing it. The fit result is expressed as a piecewise kinematic covariant particle trajectory, providing 4-vector position and momentum information about the particle with covariance as a function of physical time.
The Track is templated on a simple kinematic trajectory class representing the 1-dimensional path and momentum of a particle traveling through empty space in a constant magnetic field, as a function of physical time. The piecewise kinematic trajectory fit result is expressed as a time sequence of these simple trajectory objects. Material effects and spatial variations of magnetic fields are modeled through changes between adjacent simple trajectories. The simple kinematic trajectory class must satisfy the interface defined in the Track header, including the 4-vector position and 4-vector momentum of the particle as a function of time. The simple kinematic trajectory class must provide access to its parameterization, and the derivatives of those parameters WRT physical effects of material interactions and magnetic field inhomogeneity. KinKal can be instantiated with any simple kinematic trajectory which satisfies the interface. KinKal provides 3 fully-implemented and tested examples
- LoopHelix = low-momentum looping helix parameterized in terms of curvature radius and longitudinal wavelength
- CentralHelix = high-momentum helix parameterized in terms of inverse curvature radius and initial direction
- KLine = linear path with no geometric momentum interpretation
Measurements provided to Track constructor must provide a calculation of the Residual (difference between measurement and kinematic trajectory prediction) and the derivatives of that residual WRT the simple kinematic trajectory parameters, for the given choice of simple kinematic trajectory class. KinKal provides several fully-implemented measurement classes, based on Residuals computed from the calculation of the TPOCA (Time and Position of Closest Approach) between the measurement and the simple kinematic trajectory. The fully-functional examples include:
- WireHit = wire chamber hit, with incomplete material an drift field properties
- StrawHit = WireHit subclass with specific material and drift field properties
- ScintHit = rectilinear sensor with a prompt (light-based) signal, such as a scintillating crystal or plastic extrusion
The underlying processing model used in KinKal is a progressive BLUE fit first used in the geometric track fit implementation used by the BaBar collaboration, described in "D.N. Brown, E.A. Charles, D.A. Roberts, The BABAR track fitting algorithm, Proceedings of CHEP 2000, Padova, Italy, 2000"
Track is constructed from a configuration object which can be shared between many instances, and a unique set of hit and material interactions. The configuration object includes a convergence schedule, which holds the parameters used in successive meta-iteration steps. Each meta-iteration is described by the temperature of the simulated annealing applied to the measurement uncertainties, convergence critera, and flags to specify which physical effects (like material interactions) should be updated based on the current complete kinematic trajectory estimate. KinKal iterates each meta-iteration to algebraic convergence, by re-evaluating the extended Kalman filter derivatives, holding the physical paramters of the fit fixed.
The fit is performed on construction. After a fit, the track may be extrapolated by dead-reckoning through a magnetic field map and additional materials to make predictions.
KinKal uses the root SVector and SMatrix classes for algebraic manipulation, and GenVector classes to represent geometric and kinematic vectors, both part of the root Math package. These are described on the root website
The KinKal package includes a number of unit test programs to verify the individual components of the fit (simple kinematic trajectory class, hits, etc), as well as test program with an embedded toy Monte Carlo simulation which exercises the fit, and verifies its performance.
The KinKal package is licensed under Apache v2, and is hosted at GitHub
David N. Brown, Lawrence Berkeley National Lab
To build, you must have ROOT (see https://root.cern.ch/) installed, and the bin/ directory should be on your PATH.
There are 2 build configurations: Debug or Release. To build
- First, clone this repo
git clone https://github.com/KFTrack/KinKal.git- Set up a new build directory; lets say
kinkal_profilefor a profile build (orkinkal_debugfor a debug build)
mkdir kinkal_profile
cd kinkal_profile
- Run
cmake, and build withmake
cmake ../KinKal -DCMAKE_BUILD_TYPE=Release (or Debug)
make -j <jobs to run>- Optionally, run unit tests
make testTest programs will be built in the bin/ directory. Run them with --help in the build directory to get a list of run parameters.
Make sure clang-tidy is installed, and add an extra argument to the cmake command.
mkdir kinkal_clangtidy
cd kinkal_clangtidy
cmake ../KinKal -DCMAKE_BUILD_TYPE=[Release/Debug] -DENABLE_CLANG_TIDY=ON
make -j <jobs to run>The code will appear to compile as normal, although in this case clang-tidy is actually checking the code for problems. No warnings or errors mean that no problems were found.
The build tries to find ROOT using cmake. You should ensure before building that brew added the ROOT bin/ directory correctly to the $PATH environment variable. Sometimes re-installing the package can fix the issue.
If not installed via a package manager, you should make sure to source the <root_location>/bin/thisroot.sh shell script before building.
source <ROOT>/bin/thisroot.sh
The ROOT binaries need to be compiled with C++17 (-std=c++17) support.