<nop>SignalGridInterpolation

This page provides a sharepoint for techniques, such as signal morphing and reweighting, that are being used by analyses in ATLAS to extend or interpolate within a signal grid. The goal is to provide a pool of ideas for analysis groups who are interested in extending their existing signal grid but cannot (or do not want to) request additional MC samples.

All groups using these techniques are invited to share their approaches!

Your contribution

If you would like to add a description of the approach used in your analysis, please provide the following information:

1. A short description of the analysis grid or link to internal or public analysis documents.
2. A short description of the approach (e.g, background-to-signal vs signal-to-signal reweighting, histogram vs event-by-event reweighting).
3. Caveats of the approach (e.g. what could go wrong?)
4. Link to slides/twiki with technical details/step-by-step documentation
5. Contact person/expert who can help with further questions

Reweighting

Search for heavy Higgs bosons A/H→ttbar

The goal of this search is to probe the parameter region of 2HDMs (and extensions thereof, such as the 2HDM+pseudoscalar mediator used as a new dark matter benchmark model). In particular, a dense grid of signal samples is needed to populate the 2D parameter space defined by the mass of the heavy pseudoscalar, mA, and the parameter tanβ.

The search is complicated by the strong interference between the gg→A/H→>ttbar signal process and the background from (mostly gg-induced) SM ttbar production, which distorts the signal shape from a Breit-Wigner peak to a peak-dip structure. To constrain the signal hypothesis for a given set of parameters, the shape of both the pure signal S and the signal-plus-interference shape S+I have to be generated.

Further details can be found in the latest public result.

Signal-to-signal reweighting

This approach relies on or several simulated (AF2 or FullSim) samples for the pure signal process S for one or several points in the 2HDM parameter space. For each event, a new weight is assigned. The weight is the ratio of the matrix elements calculated for the particle configuration in the event assuming the target (numerator) and the input (denominator) signal parameters. The details of this reweighting approach are given on slide 3 of this presentation.

Caveat: Particular care must be taken regarding the choice of the input sample. It is imperative that the input sample covers a sufficiently large kinematic range to provide enough statistics within all regions of interest in the target sample. In particular, the input sample should cover a larger range in mttbar than the target sample, as illustrated in the following figure.

Contact: Katharina Behr

Background-to-signal reweighting

Like the signal-to-signal reweighting described above, this approach relies on event-by-event weights calculated as ratios of matrix elements of the target and input samples. However, in this case, an inclusive SM ttbar sample (LO in QCD) is chosen as input. The advantage of this approach is that no extra signal samples need to be generated and that the existing background sample provides abundant statistics and therefore good coverage of a large kinematic range.

Caveat: This approach is not fully functional yet. The S and S+I shapes are reproduced correctly but the normalisation is off by a factor √2.

Further details can be found on slide 32 here.

Contact: Noam Tal Hod

Search for mono-Higgs(gamgam)

Event-by-event weights are derived by dividing histograms for the target and input/base samples of a number of generator-level kinematic distributions. The weights are applied to events in the input/base sample to derive the kinematic distributions for the target sample at reco-level.

Caveats:

• Choice of base sample: The base sample must have a broader spectrum of the variables used in the reweighting than all the target samples.
• Choice of reweighting variables: For the simple mono-X (jet+X, y+X, Z(ll)+X, lepton+X, H(yy)+X), the final states are simple, and the main acceptance difference comes from the MET spectrum, therefore MET variable is the best variable to choose. If the closure test (i.e. comparing reco-level variables between reweighed and fast-sim samples) is not good enough, adding a second variable to build two-dimensional reweighting could be useful. For some complicated final states, also analyses using jet substructures, this method may not work easily, hence more extensive tests may be required.

The reweighting approach is described in Appendix E here.

Contact: Ren-Jie Wang

Morphing

Search for mono-Higgs(bb)

To match the fine mono-Higgs(gamgam) grid in the mono-Higgs combination, use interpolation+morphing for mono-Higgs(bb). For each mono-Higgs(bb) signal grid point to be made, take the normalization from TGraph2D interpolation, and take the shape from RooMomentMorphND morphing.

Caveats: cannot interpolate from on-shell region to off-shell region.

The interpolation+morphing approach is described in here.

Contact: Chen Zhou

Trigger-object Level Analysis (TLA)

To supplement the observable mass distributions constructed from Z’ MC samples for the TLA, "moment morphing" [1] as implemented in RooMomentMorph ("Linear" setting) was used to interpolate additional mass shapes for model points in between the masses of those for which MC samples were available. Cross-section and acceptances are fitted with empirical functional forms encoding the expected dependence with mass. More information can be found on page 59 of https://cds.cern.ch/record/2226514/files/ATL-COM-PHYS-2016-1498.pdf. As we are currently concentrating on the paper circulation, documentation is being prepared as this can be used for the V/AV summary plots.

[1] M. Baak et al., Interpolation between multi-dimensional histograms using a new non-linear moment morphing method, Nucl. Instrum. Meth. A771 (2015) p. 39, arXiv: 1410.7388 [physics.data-an].

Contact: Antonio Boveia, Bryan Reynolds

Note: Other studies of cross-section interpolation for these samples can be found in Conrad Schwanitz's thesis: https://lup.lub.lu.se/student-papers/search/publication/8922141

Major updates:
-- KatharinaBehr - 2017-12-19