EPIC-ANALYSIS

General purpose analysis software for (SI)DIS at the EIC

This repository provides a set of common tools for the analysis of both full and fast simulations, including the following features:

• General event loops for reading upstream data structures; for example, src/AnalysisDelphes.cxx for reading Delphes trees
• Kinematics reconstruction methods (e.g., leptonic, hadronic, Jacquet-Blondel, etc.)
  • see Kinematics Documentation for more information
  • see Jet Kinematics Documentation for jet kinematics
• Calculations of SIDIS variables, such as PhiH and qT, for single particles, as well as jet variables
• Automation for downloading or streaming simulation data from S3, along with the capability to combine data from varying Q2 ranges using weights
• Ability to specify arbitrary multi-dimensional binning schemes and cuts using Adage
• Output data structures include multi-dimensionally binned histogram sets, tables, and TTrees
• An analysis is primarily driven by macros, used to set up the binning and other settings

If you prefer to use your own analysis code, but would still like to make use of the common tools provided in this repository (e.g., kinematics reconstruction), this is also possible; you only need to stream the data structure you need, most likely within the event loop. In this situation, it is recommended you fork the repository (pull requests are also welcome).

Here is a flowchart showing the main classes (underlined) and the connections to upstream simulation output:

---

Setup and Dependencies

Initial setup

First, clone this epic-analysis Github repository:

git clone [email protected]:eic/epic-analysis.git      # if you have SSH permission
git clone https://github.com/eic/epic-analysis.git  # if you do not have SSH permission

This will create the directory epic-analysis, which you can then cd into.

Upstream Dependencies

These are common dependencies used for the upstream simulation, some of which are needed for epic-analysis as well.

Follow the EIC Software Environment Setup Guide to obtain and install the EIC software image.

• The eic-shell script is used to start a container shell
• This image contains all the upstream dependencies needed for EIC simulations
• All documentation below assumes you are running in eic-shell

If you upgrade your image (eic-shell --upgrade), you may need to clean build everything: make all-clean && make

Local Dependencies

These are additional dependencies needed by epic-analysis; they will be built locally and stored in the deps/ directory (see deps/README.md for more details). This section documents how to obtain and build local dependencies:

Delphes is the only local dependency that is not mirrored in deps/, so you must download and build it first:

deps/install_delphes.sh

• Alternatively, if you already have a delphes build elsewhere, symlink deps/delphes to it
• All other dependencies in deps/ are mirrors, and are already included with epic-analysis; they will be built automatically later

While you are waiting for Delphes to build, you may want to:

• Prepare to analyze some data from S3, following s3tools documentation
• Read through the Kinematics class header and source, along with documentation, to see what physics reconstruction methods are available
• Tutorial macros in the tutorial/ directory, to learn how to run epic-analysis

Building

First, set environment variables:

source environ.sh

Then compile analysis-epic (and some other local dependencies):

make

• We have not yet upgraded to cmake in this repository, and still use Makefiles
• Build target locations are not yet configurable, and all will stay within epic-analysis (e.g., libaries will be installd in lib/)
• Additional make targets are available (see Makefile), for more control during development:

make                     # builds dependencies, then `epic-analysis` (equivalent to `make all`)
make release             # build with optimization enabled
make debug               # build with debugging symbols
make clean               # clean `epic-analysis` (but not dependencies)

make deps                # builds only dependencies
make deps-clean          # clean dependencies
make all-clean           # clean `epic-analysis` and dependencies

make <dependency>        # build a particular `<dependency>`
make <dependency>-clean  # clean a particular `<dependency>`

Additional build options are available:

INCLUDE_CENTAURO=1 make  # build with fastjet plugin Centauro (not included in Delphes by default!)
EXCLUDE_DELPHES=1 make   # build without Delphes support; primarily used to expedite CI workflows
INCLUDE_PODIO=1 make     # build with support for reading data with PODIO

Quick Start: Tutorial Macros

If you're ready to try the software hands-on, follow the tutorials in the tutorial/ directory. Otherwise continue reading below.

---

Simulation

Delphes Fast Simulation

Delphes Wrapper

• for convenience, the wrapper script deps/run_delphes.sh is provided, which runs delphes on a given hepmc or hepmc.gz file, and sets the output file names and the appropriate configuration card
  • configuration cards are stored in the deps/delphes_EIC/ directory, a mirror of eic/delphes_EIC
  • environment must be set first (source environ.sh)
  • run deps/run_delphes.sh with no arguments for usage guide
  • in the script, you may need to change exeDelphes to the proper executable, e.g., DelphesHepMC2 or DelphesHepMC3, depending on the format of your generator input
  • if reading a gunzipped file (*.hepmc.gz), this script will automatically stream it through gunzip, so there is no need to decompress beforehand
  • there are some hepmc files on S3; follow s3tools documentation for scripts and guidance
• the output will be a TTree stored in a root file
  • output files will be placed in datarec/
  • input hepmc(.gz) files can be kept in datagen/

AnalysisDelphes

• The class AnalysisDelphes contains the event loop for reading Delphes trees
  • There are several classes which derive from the base Analysis class; Analysis handles common setup and final output, whereas the derived classes are tuned to read the upstream data formats
• See the event loop in src/AnalysisDelphes.cxx for details of how the full simulation data are read

ePIC Full Simulation

• Full simulation files are stored on S3; follow s3tools documentation for scripts and guidance
• In general, everything that can be done in fast simulation can also be done in full simulation; just replace your usage of AnalysisDelphes with AnalysisEpic
  • In practice, implementations may sometimes be a bit out of sync, where some features exist in fast simulation do not exist in full simulation, or vice versa
• See the event loop in src/AnalysisEpic.cxx for details of how the full simulation data are read

ATHENA and ECCE Full Simulations

• Similar implementation as ePIC full simulation, but use AnalysisEcce or AnalysisAthena

---

Analysis Procedure

After simulation, this repository separates the analysis procedure into two stages: (1) the Analysis stage includes the event loop, which processes either fast or full simulation output, kinematics reconstruction, and your specified binning scheme, while (2) the Post-processing stage includes histogram drawing, comparisons, table printouts, and any feature you would like to add.

The two stages are driven by macros. See examples in the tutorial directory, and follow the README.

• Note: most macros stored in this repository must be executed from the epic-analysis top directory, not from within their subdirectory, e.g., run root -b -q tutorial/analysis_template.C; this is because certain library and data directory paths are given as relative paths

In general, these macros will run single-threaded. See HPC documentation for guidance how to run multi-threaded or on a High Performance Computing (HPC) cluster.

Analysis Stage

Analysis Macro and Class

• the Analysis class is the main class that performs the analysis; it is controlled at the macro level
  • a typical analysis macro must do the following:
    • instantiate an Analysis derived class (e.g., AnalysisDelphes)
    • set up bin schemes and bins (arbitrary specification, see below)
    • set any other settings (e.g., a maximum number of events to process, useful for quick tests)
    • execute the analysis
  • the input is a config file, which contains a list of files to analyze together with settings such as beam energy and Q2 ranges; see doc/example.config for an example config file and more details
  • the output will be a root file, filled with TObjArrays of histograms
    • each TObjArray can be for a different subset of events (bin), e.g., different minimum y cuts, so that their histograms can be compared and divided; you can open the root file in a TBrowser to browse the histograms
    • the Histos class is a container for the histograms, and instances of Histos will also be streamed to root files, along with the binning scheme (handled by the Adage BinSet class); downstream post processing code makes use of these streamed objects, rather than the TObjArrays
  • derived classes are specific to upstream data structures:
    • AnalysisDelphes for Delphes trees (fast simulations)
    • AnalysisAthena for trees from the DD4hep+Juggler stack (ATHENA full simulations)
    • AnalysisEcce for trees from the Fun4all+EventEvaluator stack (ECCE full simulations)
  • the Kinematics class is used to calculate all kinematics
    • Analysis-derived classes have one instance of Kinematics for generated variables, and another for reconstructed variables, to allow quick comparison (e.g., for resolutions)
    • calculations are called by Analysis-derived classes, event-by-event or particle-by-particle or jet-by-jet
    • see Kinematics Documentation for details of Kinematics

Bin Specification

• The bins may be specified arbitrarily, using the Adage BinSet and CutDef classes
  • see example analysis_*C macros in tutorial/
  • CutDef can store and apply an arbitrary cut for a single variable, such as:
    • ranges: a<x<b or |x-a|<b
    • minimum or maximum: x>a or x<a
    • no cut (useful for "full" bins)
  • The set of bins for a variable is defined by BinSet, a set of bins
    • These bins can be defined arbitrarily, with the help of the CutDef class; you can either:
      • Automatically define a set of bins, e.g., N bins between a and b
        • Equal width in linear scale
        • Equal width in log scale (useful for x and Q2)
        • Any custom TAxis
      • Manually define each bin
        • example: specific bins in z and pT:
          • |z-0.3|<0.1 and |pT-0.2|<0.05
          • |z-0.7|<0.1 and |pT-0.5|<0.05
        • example: 3 different y minima:
          • y>0.05
          • y>0.03
          • y>0 (no cut)
          • note that the arbitrary specification permits bins to overlap, e.g., an event with y=0.1 will appear in all three bins
• Multi-dimensional binning
  • Binning in multi-dimensions is allowed, e.g., 3D binning in x,Q2,z
  • See Adage documentation for more information on how multi-dimensional binning is handled, as well as the Adage syntax reference
  • Be careful of the curse of dimensionality

Simple Tree

• The Analysis class is also capable of producing a simple TTree, handled by the SidisTree class, which can also be useful for analysis
  • As the name suggests, it is a flat tree with a minimal set of variables, specifically needed for SIDIS spin asymmetry analysis
  • The tree branches are configured to be compatible with asymmetry analysis code built on the BruFit framework
  • There is a switch in Analysis to enable/disable whether this tree is written

Post-Processing Stage

Post-Processing Macro and Class

• results processing is handled by the PostProcessor class, which does tasks such as printing tables of average values, and drawing ratios of histograms
  • this class is steered by postprocess_*.C macros, which includes the following:
    • instantiate PostProcessor, with the specified root file that contains output from the analysis macro
    • loops over bins and perform actions, using Adage
• see src/PostProcessor.h and src/PostProcessor.cxx for available post-processing routines; you are welcome to add your own

---

Contributions

• Add your own analysis scripts (macros, etc.) in macro/, either in the main directory or in a subdirectory of macro/.

  • The macro/ci directory is for scripts used by the CI (see .github/workflows/ci.yml); you are welcome to add new analysis scripts to the CI
  • Make changes in classes such as PostProcessor as needed

• Git workflow:

  • Contributions are welcome via pull requests and issues reporting; it is recommended to fork this repository or ask to be a contributor
  • Continuous Integration (CI) will trigger on pull requests, which will build and test your contribution
    • see Actions tab for workflows for details
    • many CI jobs will not work properly from forks (for security), but you may ask to be a contributor
  • It is recommended to keep up-to-date with developments by browsing the pull requests, issues, and viewing the latest commits by going to the Insights tab, and clicking Network to show the commit graph