Truth-Based 1l1p Selection

A. Mastbaum [email protected], 2017/08/13

Loop through MCTracks and MCShowers, doing PID based on dE/dx and other observable parameters, to build a list of tracks and showers in the event, then select based on that.

Tools

TSSelection: Truth-Based Selection

This module is the main truth-based selection code. It scans through Monte Carlo events to extract MCTracks and MCShowers, assigns them particle IDs based on dE/dx information (using PDFs from TSPDFGen), applies selection cuts (e.g. kinetic energy), and classifies events as 1e1p and 1mu1p. The selected events are written into an output analysis tree.

Usage:

$ python scripts/selection.py OUTPUT.root DATASET "INPUT*.root"

The output file OUTPUT.root will contain the output analysis trees. The DATASET is an integer flag which is written into the output trees, as a convenience for bookkeeping when working with multiple MC samples. The final argument is a set of input art ROOT files.

TSCovariance: Covariance Matrix Generator

This module takes analysis trees with selected event samples and Monte Carlo event weights, and computes a covariance matrix for a user-specified subset of weights. The output is a weighted event histogram, a graph with systematic uncertainty bands, and the covariance and correlation matrices.

Usage:

$ python scripts/cov.py input.root

Here, the input is an analysis tree ROOT file, i.e. the output of TSSelection.

The user should modify cov.py to select an appropriate combinations of weights.

TSPDFGen: PDF Generator

The PDF Generator produces probability distributions used by TSSelection for track and particle identification. It constructs a 2D histogram in dE/dx vs. residual range for tracks, and a 1D histogram of dE/dx for showers, based on the Monte Carlo truth information in the events in the input dataset.

Usage:

$ python scripts/make_pdfs.py "input*.root"

The input is a set of art ROOT files.

This will create a file ana_out.root containing histograms named like htrackdedx_<PDG> for track-like PDG codes, and hshowerdedx_<PDG> for shower-like PDGs.

A set of PDFs (which is used by TSSelection is provided in data/dedx_pdfs.root. To update the PDFs for the selection, run TSPDFGen and replace that file.

TSUtil: Utilities

This file contains common utilities used by the truth-based selection modules, which are in namespace tsutil.

Scripts

The scripts directory contains:

• Python scripts used to run each module (see the Tools section)
• Utilities in scripts/util
  • File handling and MC POT counting
• Analysis scripts in scripts/analysis

See header blocks at the top of each script for more information, including usage.

Workflow

The workflow begins with raw art ROOT files, produced by LArSoft (uboonecode). In practice we use two Monte Carlo samples: one fully inclusive BNB neutrino sample with MC cosmic ray overlays, and one with only nue events (to improve the statistics for nues, since they constitute only a small fraction of the BNB inclusive sample). Additionally, there are MC samples produced for various signal models.

In general, MC weighting functions that assign weights based on variations in the systematics parameters are not run on the production MC files, so we must run them manually using grid-based LArSoft jobs that run the EventWeight module (and can also drop most data products, since this analysis only looks at MC truth). In this way we obtain copies of the MC datasets with only MC information, imbued with a set of systematics weights attached to each event. The POT counting script in scripts/util/pot.py can be used to extract the equivalent POT for these datasets, which is necessary for scaling to various exposures.

The selection script (see TSSelection above) operates on a list of these weighted art ROOT files, applying the truth-based selection. Typically this would be run three times -- for the electron, inclusive, and signal samples -- and produce three output files. The electron and inclusive samples can now be added together, using scripts/util/add.C. This (in contrast to hadd) takes CC nue and nuebar events from the electron sample and everything else from the inclusive sample. Splitting in this way simplifies the reweighting, as we combine samples representing different POT exposures. After the addition, we have two ROOT files -- a combined 1e1p + 1mu1p background file, and a signal file -- containing simple analysis ROOT trees.

We can now run scripts/cov.py on the background ROOT file. This script will iterate through the chosen weights (the set of which comprise a systematics "universe") and produce a number of weighted spectra with and without systematic error bands, plus the covariance and correlation matrices. The script allows the user to toggle on and off different combinations of parameters, so that one can produce a covariance for e.g. the flux uncertainties alone, one particular GENIE uncertainty, or everything together. The plotcov.py script is provided for convenience, to draw these plots into reasonable-looking PDF files.

The scripts/analysis directory provides some basic analysis code which operates on either the analysis tree files, or the histograms output from cov.py The main script, chi2.py reads in the latter and can make a variety of plots, and compute a chi^2 significance for observing the signal prediction under (background-only) null hypothesis using an analytic approximation or a Monte Carlo approach, which represents the final product of this analysis chain.