Overview of the LHCb trigger system.

Sketch of the different types of tracks within LHCb .

Schematic view of the real-time alignment and calibration procedure starting at the beginning of each fill, as used for 2018 data taking.

Sketch of the HLT1 track and vertex reconstruction.

The PV $x$ (left) and $z$ (right) resolution as a function of the number of tracks in the PV for the Run 1 off\-line and Run 2 (used both off\-line and online) PV reconstruction algorithms.

HLT1 dimuon efficiency as a function of the minimum $p_{\rm T}$ of the two muons. A large gain, especially at low $p_{\rm T}$ , can be seen from the comparison of the Run 1 and Run 2 algorithms.

HLT1 muon identification efficiency for (left) muons from $ { J \mskip -3mu/\mskip -2mu\psi \mskip 2mu} \rightarrow \mu ^+\mu ^- $ decays and (right) pions from $ D^0 \rightarrow K ^- \pi ^+ $ decays. Green circles show only the identification efficiency (HLT1 Muon ID) while red squares show the efficiency of the additional trigger line (named HLT1TrackMuon) requirements (see text).

Sketch of the HLT2 track and vertex reconstruction sequence.

Performance of the fake-track classifier on (left) $ D \rightarrow K ^- \pi ^+ $ and (right) $ K ^0_{\rm\scriptscriptstyle S} \rightarrow \pi ^- \pi ^+ $ decays. For these plots, the clones have been removed.

Comparison of the track reconstruction efficiency in 2015 and 2012 data as a function of the momentum (left) and pseudorapidity (right).

Comparison of the invariant mass distributions for a subset of the 2012 (left) and 2016 (right) data set, using $ { J \mskip -3mu/\mskip -2mu\psi \mskip 2mu} \rightarrow \mu ^+\mu ^- $ decays, with the $ { J \mskip -3mu/\mskip -2mu\psi \mskip 2mu}$ originating from a $ b $ -hadron.

Resolution of the $x$ (left) and $y$ (right) components of the impact parameter comparing the 2012 (blue), 2015 (orange), 2016 (red) and 2017 (green) data-taking periods. The resolution as a function of $p_{\rm T}$ is given in the bottom right corner.

Decay time resolution for $ B ^0_ s \rightarrow { J \mskip -3mu/\mskip -2mu\psi \mskip 2mu} \phi$ decays (in their rest frame) as a function of momentum. The filled histogram shows the distribution of $ B ^0_ s $ meson momenta, to give an idea of the relative importance of the different resolution bins for the analysis sensitivity.

Efficiency and fake rate of the RICH identification for the 2012 (left) and the 2016 (right) data.

Invariant mass of $B^0 \rightarrow (K^+ \pi^-) \gamma$ candidates in Run 1 (left) and Run 2 (right). The fit model includes the (red) signal component, (dashed green) combinatorial background, (dot-dashed turqoise) misidentified physics backgrounds (e.g. $ B ^0_ s \rightarrow \phi\gamma$ where a kaon is misidentified as a pion) and (dotted magenta) partially reconstructed physics backgrounds.

Efficiencies per signal mode for (top) 2016 and (bottom) 2017 data-taking periods measured in simulation. Red (left-slanted) hatched plots are when the entire \texttt{L0} bandwidth is granted to this signal mode, whereas blue (right-slanted) hatched plots are following the bandwidth division. Signals which appear only in blue are used for performance validation and are not part of the optimization itself. Channels followed by "(S)" are selected in a kinematic and geometric volume which is particularly important for spectroscopy studies.

Disk buffer usage projections during (left) and at the end of (right) the 2017 data-taking period. During data taking, simulations (red, left) are used every two weeks to determine the probability of exceeding the 80% usage threshold. In 2017, the loose HLT1 configuration was used for the entire year leading to a maximum buffer capacity of 48% (black, right). LHC Technical Stops and Machine Development (MD) periods are shown in dark and light grey, respectively. The schedule changed between when this simulation was run in week 32 and the end of the year. An MD period was removed and the duration of the data taking was reduced.

Charm candidates used for the evaluation of the trigger performance.

Beauty candidates used for the evaluation of the trigger performance.

Efficiencies of the \texttt{L0} trigger lines in Run 2 data for $ c$ -hadron decays. The left plot shows the efficiency as a function of the hadron $p_{\rm T}$ , while the right plot shows the evolution of the efficiency as a function of the different trigger configurations used during data taking. The three blocks visible in the plot, separated by vertical gaps, correspond to the three years of data taking (2015--2017). The \texttt{L0} hadron efficiency is shown.

Efficiencies of the \texttt{L0} trigger lines in Run 2 data for $ b$ -hadron decays. The left plot shows the efficiency as a function of the hadron $p_{\rm T}$ , while the right plot shows the evolution of the efficiency as a function of the different trigger configurations used during data taking. The three blocks visible in the plot, separated by vertical gaps, correspond to the three years of data taking (2015--2017). The plotted \texttt{L0} efficiency for each $ b$ -hadron is described in the legend above the plots.

The SPD hit multiplicity of events containing $ B ^+ \rightarrow \overline{ D }{} {}^0 \pi ^+ $ candidates in Run 2 data.

Two-dimensional efficiencies of the \texttt{L0} trigger lines in Run 2 data: (top left) \texttt{L0} hadron; (top right) \texttt{L0} electron; (bottom left) \texttt{L0} muon; and (bottom right) \texttt{L0} dimuon. The \texttt{L0} hadron efficiency is evaluated using $ D ^0 \rightarrow K ^- \pi ^+ $ decays, whereas the others are evaluated using the relevant signals listed in Fig. 20 and Fig. 21.

Efficiency of the HLT1 inclusive trigger lines as a function of (left) $ c$ -hadron $p_{\rm T}$ and (right) decay time. The decay time plots are drawn such that the x-axis is binned in units of the lifetime for each hadron in its rest frame. The plots in each column show, from top to bottom, the single-track, two-track, and combined HLT1 inclusive performance.

Efficiency of the HLT1 inclusive trigger lines as a function of (left) $ b$ -hadron $p_{\rm T}$ and (right) decay time. The decay time plots are drawn such that the x-axis is binned in units of the lifetime for each hadron in its rest frame. The plots in each column show, from top to bottom, the single-track, two-track, and combined HLT1 inclusive performance.

The HLT1 efficiency as a function of the different trigger configurations used during data taking for (left) $ c$ -hadrons and (right) $ b$ -hadrons. The three blocks visible in the plot, separated by vertical gaps, correspond to the three years of data taking (2015--2017).

The efficiency of the HLT1 muon trigger lines as a function of the (left) $ b$ -hadron $p_{\rm T}$ and (right) units of the average $ B ^+ $ decay time. The decay time plot is drawn such that the x-axis is binned in units of the $ B ^+ $ lifetime in its rest frame. The efficiency of the inclusive single-track HLT1 trigger is plotted for reference.

The $D^0$ (left) and $J/\psi$ (right) candidates selected by the HLT1 calibration lines. Both plots show candidates reconstructed online.

Rates of the main groups of HLT1 trigger lines and the total HLT1 rate as a function of the year of data taking, shown for the trigger configuration used to take most of the luminosity in each year.

Efficiency of the HLT2 topological trigger lines as a function of the (left) $ b$ -hadron $p_{\rm T}$ and (right) in units of the average $ b$ -hadron decay time. The decay time plots are drawn such that the x-axis is binned in units of the lifetime for each hadron in its rest frame. The plots show the combined efficiency of the topological trigger lines for each $ b$ -hadron decay mode.

Efficiency of the HLT2 topological trigger lines as a function of the (left) $ b$ -hadron $p_{\rm T}$ and (right) in units of the average $ b$ -hadron decay time. The decay time plots are drawn such that the x-axis is binned in units of the lifetime for each hadron in its rest frame. The plots show the different contributions of the 2-, 3-, and 4-body topological trigger lines to a specific decay.

Evolution of the HLT2 efficiency as a function of the different trigger configurations used during data taking.

The TOS efficiency of the HLT2 muon trigger lines as a function of the (left) $ B ^+ $ $p_{\rm T}$ and (right) units of the average $ B ^+ $ decay time. The decay time plot is drawn such that the x-axis is binned in units of the $ B ^+ $ lifetime in its rest frame. The efficiency of the inclusive topological ("any topological") trigger lines and topological trigger lines requiring one track identified as a muon ("any muon topological") are plotted for reference.

HLT2 trigger efficiencies of the dedicated selections for low-multiplicity events: (left) for dimuon candidates as a function of dimuon mass, and (right) for $\phi(1020)$ candidates as a function of candidate $p_{\rm T}$ .

Rates of the main categories of HLT2 trigger lines and the total HLT2 rate for each year of data taking, shown for the trigger configuration used to take most of the luminosity in the given year. TURBO, CALIBRATION, and FULL refer to different output data streams as discussed in Ref. [2].

Animated gif made out of all figures.

The \texttt{L0} thresholds for the different trigger lines used to take the majority of the data for each indicated year. Technical trigger lines and those used for special areas of the physics programme are excluded for brevity. The Hadron, Photon, and Electron trigger lines select events based on the $E_{\rm T}$ of reconstructed ECAL and HCAL clusters. The Muon, Muon High, and Dimon trigger lines select events based on the $p_{\rm T}$ reconstruced MUON stubs, where the Dimuon selection is based on the product of the largest and second largest $p_{\rm T}$ stubs found in the event. As some of the subdetectors also read out hits associated to other bunch crossings, the use of bandwidth is further optimised in most of the L0 lines by rejecting events with a large $ E_{\rm T} $ ($>24\mathrm{ Ge V} $) for the previous bunch crossing [19].