Track Seeding

To reduce the time needed to reconstruct particle tracks, a track seed (henceforth: seed) is created which serves as the initial direction for the track reconstruction algorithm (henceforth: the tracking). The tracking then tries to find all measurements belonging to a single particle in this direction in order to reconstruct the track. This means, if no seed exists for a particle, this particle will not be reconstructed. On the other hand, finding too many seeds which either correspond to particles with already existing seeds or which do not correspond to particles at all increases the time needed for tracking.

A good seeding algorithm, therefore, has the following properties:

  • It finds at least one seed for each particle that should be found

  • It doesn’t find many seeds which do NOT correspond to particles

  • It doesn’t find many seeds per particle

The most typical way to create seeds is to combine measurements. In a homogeneous magnetic field, 3 measurements perfectly describe the helical path of a charged particle. One such triplet of measurements would then constitute a seed and defines, in close bounds, where the tracking needs to look for additional measurements to create a track spanning the whole detector. The difficulty is in choosing the correct measurements, as a helix can be fitted through any 3 measurements in a collision event with potentially tens of thousands of measurements. Therefore, many constraints or “cuts” are defined to reduce the number of candidates. Cuts may define where particles originate or the range of energy of particles to be found or otherwise restrict the combination of measurements for seed creation.

Acts Implementation

The seeding implementation in Core/include/Acts/Seeding/ was written with a focus on parallelism and maintainability and as detector agnostic as possible, only assuming a (near) homogeneous magnetic field with particles originating from the central detector region. Cuts are configurable and can be plugged in as an algorithm which is called by the seeding. The seeding works on measurements or “SpacePoints” (SP), which need to provide \((x,y,z)\) coordinates with the \(z\) axis being along the magnetic field, and \(x\) and \(y\). For the seeding algorithm to function the particle point of origin has to have a radius smaller than the radius of the detector layer closest to the interaction region. In other words, this seeding algorithm is not suitable for secondary particles originating far from the interaction region.


Fig. 21 Sketch of the detector with 3 layers. The interaction region is supposed to be located along the \(z\)-axis and has a size significantly smaller than the radius of the innermost detector layer.


Note that the seeding algorithm breaks down for particles with a particle track whose helix diameter is smaller than the detector radius until which seeds are to be created. This is due to ordering assumptions of SP locations as well as due to approximations which become inaccurate for lower energy particles.

SP Grid and Groups Formation

The SPs in each detector layer are projected on a rectangular grid of configurable granularity. The search for seed starts by selecting SP in the middle detector layer. Then matching SPs are searched in the inner and outer layers. Grouping of the SPs in the aforementioned grid allows to limit the search to neighbouring grid cells thus improving significantly algorithm performance (see Fig. 22). The number of neighboring bins used in the SP search can be defined separately for the bottom and top layer SPs in the \(z\) and \(\phi\) directions.


Fig. 22 Representation of the search for triplet combinations in the \((r, z)\) plane. The bins used in the search are represented in different colours.

The Seed Finder

The SeedFilter::createSeedsForGroup() function receives three iterators over SPs constructed from detector layers of increasing radii. The seedfinder will then attempt to create seeds, with each seed containing exactly one SP returned by each of the three iterators. It starts by iterating over SPs in the middle layer (2nd iterator), and within this loop separately iterates once over the bottom SP and once over the top SP. Within each of the nested loops, SP pairs are tested for compatibility by applying a set of configurable cuts that can be tested with two SP only (pseudorapidity, origin along \(z\)-axis, distance in \(r\) between SP, compatibility with interaction point).

For all pairs passing the selection the triplets of bottom-middle-top SPs are formed. Each triplet is then confronted with the helix hypothesis. In order to perform calculations only once, the circle calculation is spread out over the three loops.


Fig. 23 The x-y projection of the detector with the charged particle helical track originating from the centre of the detector. Signals left by the passage of the track through the detector layers are marked with green crosses.

From the helix circle (see Fig. 23), particle energy and impact parameters can be estimated. To calculate the helix circle in the \(x/y\) plane, the \(x,y\) coordinates are transformed into a \(u/v\) plane to calculate the circle with a linear equation instead of a quadratic equation for speed. The conformal transformation is given by:

\[\begin{equation*} u = \frac{x}{x^2+y^2}, \quad \quad v = \frac{y}{x^2+y^2} , \end{equation*}\]

where the circle containing the three SPs are transformed into a line with equation \(v = Au + B\). The angular coefficient \(A\) can be evaluated by the slope of the linear function between the top and bottom layer SPs, after transforming the coordinates of these SPs from \(x/y\) to \(u/v\) using the previous equations:

\[\begin{equation*} A = \frac{v_t-v_b}{u_t-u_b} , \end{equation*}\]

Then, \(B\) can be obtained by inserting \(A\) into the linear equation for the bottom layer SP:

\[\begin{equation*} v_b = Au_b + B \rightarrow B = v_b - Au_B . \end{equation*}\]

Inserting the coefficients in the circle equation and assuming that the circle goes through the origin we obtain:

\[\begin{equation*} (2R)^2 = \frac{A^2+1}{B^2} . \end{equation*}\]

Now we can apply a cut on the estimate of the minimum helix diameter minHelixDiameter2 without the extra overhead of conversions or computationally complex calculations. The seed is accepted if

\[\begin{equation*} \frac{A^2+1}{B^2} > (2 R^{min})^2 = \left ( \frac{2 \cdot p_T^{min}}{300 \cdot B_z} \right)^2 \equiv \textnormal{minHelixDiameter2} , \end{equation*}\]

where \(B_z\) is the magnetic field.


Fig. 24 The r-z projection of the detector with the same charged particle track. The track is depicted with the same colours as in the previous figure.

The track is not an ideal helix. At each detector layer (or any other material) scattering may occur making the helix approximate. The algorithm will check if the triplet forms a nearly straight line in the \(r/z\) plane (see Fig. 24) as the particle path in the \(r/z\) plane is unaffected by the magnetic field. This is split into two parts; the first test occurs before the calculation of the helix circle. Therefore, the deviation from a straight line is compared to the maximum allowed scattering at minimum \(p_T\) scaled by the forward angle:

\[\begin{equation*} \left (\cot \theta_b - \cot \theta_t \right )^2 < \sigma^2_{p_T^{min}} + \sigma_f^2, \end{equation*}\]

This check takes into account the squared uncertainty in the difference between slopes (\(\sigma_f^2\)) and the scattering term scatteringInRegion2 (\(\sigma^2_{p_T^{min}}\)), which is calculated from sigmaScattering, the configurable number of sigmas of scattering angle to be considered, and maxScatteringAngle2, which is evaluated from the Lynch & Dahl correction1\(^,\)2 of the Highland equation assuming the lowest allowed \(p_T\):

\[\begin{equation*} \Theta_0^{min} = \frac{13.6 \text{MeV}}{p_T^{min}} \cdot \sqrt{\frac{z_q^2 L}{\beta^2 L_0}} \left(1+0.038 \ln \left(\frac{z_q^2 L}{\beta^2 L_0}\right) \right), \end{equation*}\]

The second part check against the multiple scattering term p2scatterSigma (\(\sigma^2_{p_T^{estimated}}\)) assuming the seed \(p_T\) estimation, instead of the minimum allowed \(p_T\). This term is inversely proportional to the momentum and accounts for the curvature of the seed; smaller scattering angle is permitted for higher momentum.

\[\begin{equation*} \left (\cot \theta_b - \cot \theta_t \right ) ^2 < \sigma^2_{p_T^{estimated}} + \sigma_f^2, \end{equation*}\]

The last cut applied in this function is on the transverse impact parameter (or DCA - distance of closest approach), which is the distance of the perigee of a track from the interaction region in \(mm\) of detector radius. It is calculated and cut on before storing all top SP compatible with both the current middle SP and current bottom SP.


Fig. 25 Helix representation in \(x/y\) reference frame with central space-point (SP\(_m\)) in the origin.

Assuming the middle layer SP is in the origin of the \(x/y\) frame, as in Fig. 25. The distance between the centre of the helix and the interaction point (IP) is given by

\[\begin{equation*} (x_0 + r_m)^2 + y_0^2 = (R + d_0)^2 \quad \xrightarrow{R^2 = x_0^2 + y_0^2} \quad \frac{d_0^2}{R^2} + 2 \frac{d_0}{R} = \frac{2 x_0 r_m + r_m^2}{R^2} . \end{equation*}\]

Considering that \(d_0 << R\) (we can neglect the term proportional to \(d_0^2\)) and using the \(u/v\) line equation calculated previously, the cut can now be estimated using a linear function in the \(u/v\) plane instead of a quartic function:

\[\begin{equation*} d_0 \leq \left| \left( A - B \cdot r_M \right) \cdot r_M \right| \end{equation*}\]

The Seed Filter

After creating the potential seeds we apply a seed filter procedure that compares the seeds with other SPs compatible with the seed curvature. This process ranks the potential seeds based on certain quality criteria and selects the ones that are more likely to produce high-quality tracks The filter is divided into two functions SeedFilter::filterSeeds_2SpFixed() and SeedFilter::filterSeeds_1SpFixed().

The first function compares the middle and bottom layer SPs of the seeds to other top layer SPs; seeds only differing in top SP are compatible if they have similar helix radius with the same sign (i.e. the same charge). The SPs must have a minimum distance in detector radius, such that SPs from the same layer cannot be considered compatible. The second function iterates over the seeds with only a common middle layer SP and selects the higher quality combinations.

virtual void Acts::SeedFilter::filterSeeds_2SpFixed(InternalSpacePoint<external_spacepoint_t> &bottomSP, InternalSpacePoint<external_spacepoint_t> &middleSP, std::vector<InternalSpacePoint<external_spacepoint_t>*> &topSpVec, std::vector<float> &invHelixDiameterVec, std::vector<float> &impactParametersVec, SeedFilterState &seedFilterState, std::vector<std::pair<float, std::unique_ptr<const InternalSeed<external_spacepoint_t>>>> &outCont) const

This function assigns a weight (which should correspond to the likelihood that a seed is good) to all seeds and applies detector-specific selection of seeds based on weights. The weight is a “soft cut”, which means that it is only used to discard tracks if many seeds are created for the same middle SP. This process is important to improving computational performance and the quality of the final track collections by rejecting lower-quality seeds.

The weight is calculated by:

\[\begin{equation*} w = (c_1 \cdot N_{t} - c_2 \cdot d_0 - c_3 |z_0| ) + \textnormal{detector specific cuts}. \end{equation*}\]

The transverse (\(d_0\)) and longitudinal (\(z_0\)) impact parameters are multiplied by a configured factor and subtracted from the weight, as seeds with higher impact parameters are assumed to be less likely to stem from a particle than another seed using the same middle SP with smaller impact parameters. The number of compatible seeds (\(N_t\)) is used to increase the weight, as a higher number of measurements will lead to higher quality tracks. Finally, the weight can also be affected by optional detector-specific cuts.

The SeedFilter::filterSeeds_2SpFixed() function also includes a configurable Acts::SeedConfirmationRangeConfig seed confirmation step that, when enabled, classifies higher quality seeds as “quality confined” seeds if they fall within a predefined range of parameters (\(d_0\), \(z_0\) and \(N_t\)) that also depends on the region of the detector (i.e., forward or central region). If the seed is not classified as “quality confined” seed, it will only be accepted if its weight is greater then a certain threshold and no other high-quality seed has been found.

The seed confirmation also sets a limit on the number of seeds produced for each middle SP, which retains only the higher quality seeds. If this limit is exceeded, the algorithm checks if there is any low-quality seed in the seed container of this middle SP that can be removed.

virtual void Acts::SeedFilter::filterSeeds_1SpFixed(std::vector<std::pair<float, std::unique_ptr<const InternalSeed<external_spacepoint_t>>>> &seedsPerSpM, int &numQualitySeeds, std::back_insert_iterator<std::vector<Seed<external_spacepoint_t>>> outIt) const

This function allows the detector-specific cuts to filter based on all seeds with a common middle SP and limits the number of seeds per middle SP to the configured limit. It sorts the seeds by weight and, to achieve a well-defined ordering in the rare case weights are equal, sorts them by location. The ordering by location is only done to make sure reimplementations (such as the GPU code) are comparable and return the bitwise exactly the same result.


M. Tanabashi et al. (Particle Data Group), Passage of Particles Through Matter, Phys. Rev. D 98 0300001 (2018) 2.


G.R. Lynch and O.I Dahl, Nucl. Instrum. Methods B58, Phys. Rev. D 98 6 (1991) 2.