Fatemi, Renee H

Introduction:

The PI, Renee Fatemi, has been continuously funded by the National Science Foundation since 2009. During that time, she has demonstrated leadership in a variety of experimental efforts, ranging from the study of how quarks and gluons interact to form the properties of nuclear matter in the universe, to using precision measurements of the muon anomalous magnetic moment to look for signs of new forces and particles. These experiments have been carried out at existing facilities, namely at Fermi National Accelerator Laboratory (FNAL) and the Relativistic Heavy Ion Collider (RHIC) at Brookhaven National Lab (BNL), using detectors that were built by other institutions and laboratories. This project, which involves the design, testing and construction of the backward and central electromagnetic calorimeters for the ePIC detector, provides a unique opportunity for the PI and her group to play a substantial role in the decade-long process of designing, building, and commissioning a new piece of hardware at a completely new facility, the Electron-Ion-Collider (EIC).

The EIC is a powerful and versatile new accelerator facility, capable of colliding high-energy beams ranging from heavy ions to polarized light ions and protons with high-energy polarized electron beams. The EIC, accompanied by a general-purpose large-acceptance detector, ePIC, will be a discovery machine that addresses fundamental questions such as the origin of mass and spin of the proton as well as probing dense gluon systems in nuclei. It will allow for the exploration of new landscapes in QCD, permitting the “tomography”, or high-resolution multidimensional mapping of the quark and gluon components inside of nucleons and nuclei.

Both the PI and postdoc Dmitry Kalinkin have experience with building and commissioning calorimeters at RHIC, but these detectors were of smaller scale and they were not part of the design and optimization process that occurs when planning for an entirely new facility. Fatemi and Kalinkin plan to utilize newly developed artificial intelligence (AI) and machine learning guided design algorithms to assist in the design optimization, diving into the rapidly evolving field of AI assisted design as well as the standard suite of detector design simulation and design tools.

AI guided design for the Endcap and Barrel Electromagnetic Calorimeters for the EPIC detector at the future Electron-Ion Collider

The EIC science case is very broad, but the analysis of nearly all reaction channels will require the precise reconstruction of the energy and position of the scattered electron. The best way to reconstruct the energy of the scattered electron is via an electromagnetic calorimeter, a large homogenous cylindrical device composed of half a meter of lead-tungstate and silicon glass that will serve to both stop the electron and collect all the energy deposited in the calorimeter. The calorimeters are immersed in a 1.5T magnetic field and must be designed to fit within the coils of the magnet. While the materials have been chosen to provide the highest precision on the electron energy measurement, many other aspects of the design still need to be explored and optimized. Examples include the spacing between and angle of tilt for individual tower components. The backward calorimeter serves as a “cap” to the cylindrical central calorimeter, so the overlap region must be carefully studied to maintain the excellent resolution and electron detection capability. The goal of this project will be to produce a final, integrated design of both the backward and central detectors. The completion of the final design is required for the CD2/3a review by the Department of Energy which is currently scheduled for January of 2024. This project should result in a published paper on the outcome of the AI guided design studies. The final design will also be included in a Technical Design Report, which is an unpublished but public document submitted to the Department of Energy for the CD 2/3a review. 

Personnel:

Renee Fatemi (faculty)

Dmitry Kalinkin (postdoc), Added on LCC 09/211/2022

Hannah A Smith, Graduate, Added on LCC 08/18/2023 

Computational Methods:

The project will ultimately require computationally-intensive simulations for passage of particles through the alternative detector geometries. The Geant4 simulation toolkit will be used to represent the geometries in-memory, as well as track particles through those while sampling possible physical interaction processes. The DD4hep framework is used to provide reusable parametrized geometry descriptions and physics process settings to be loaded into Geant4. It also interfaces with Geant4 to provide energy depositions to individual segments of the detector (i.e., calorimeter towers) from particles and their associated cascade showers. Each geometry will be evaluated with respect to detector response to single particles (electrons, photons and charged pions). A custom benchmark code will evaluate energy depositions for each simulated event to numerically quantify detector performance in the two key aspects: energy resolution EM showers (in GeV) and pion rejection fraction. The alternative detector geometries are to be sampled from a many-dimensional design space, however, parameter sets will have to be checked to not produce invalid geometries, such as self-overlapping geometries, geometries extruding past the space allocated for the detector, or geometries going outside of limitations imposed by engineering consideration. To control for this, facilities implemented in DD4hep using the Geometry package in the ROOT framework will be used. Finally, the parameters of the geometry are to be optimized for multiple objectives using Machine Learning methods such as Non-dominated Sorting Genetic Algorithm (NSGA), as implemented in the pymoo framework, to produce a pareto front in the multi-dimensional space performance numbers and, possibly, additional detector cost. As the use of the Genetic Algorithms imposes a requirement of parameter evaluation to be done in batches, the implementation of the pymoo interface to the problem will utilize a parallel job scheduling using the Dask Distributed framework.

Software: 

Geant4, EPIC software stack (including but not limited to: DD4hep, EDM4hep, JANA2), Nix, Python (dask, distributed, numpy, pymoo, tensorflow), Pythia, CERN ROOT

Collaborators:

None

Grants:

Publications: