Fellows

Background

An integral part of a learning process is applying the acquired skills to concrete problems. Within MCgen mentoring program, a group of selected graduate students and junior postdoctoral researchers work, under guidance of senior members, on carefully crafted projects that will have an immediate impact on the particle/nuclear physics community. The selection process and sample projects are listed below (students are also allowed to propose their own projects in consultation with one of the senior members). Students selected from a national competition remain enrolled at their home institutions, while MCgen provides financial support to students so that they can focus on their project at the institution of the mentor (UC Berkeley or U. Cincinnati, in the pilot phase of MCgen).

Call for Fellows

The current call is open: apply here. The deadline is Nov 30th 2024.

The pilot phase of MCgen consists of eight 6-month long traineeships to be selected from a national pool of graduate students and junior postdocs at institutions in the US. We will have two calls for the positions: four positions will be offered each for 2025-26 (current call) and then for 2026-27 academic year, equally split among the two host campuses.

Eligibility

Be a graduate student or junior postdoc at a US institution. The participants are not required to be US citizens.

Benefits

Fellows will receive a stipend that will cover the relocation costs and the local living expenses (the stipends are cost-of-living adjusted, $23.5k/6 months for U. Cincinnati and $27.5k/6 months for UC Berkeley campus). Partial/additional support by their home institution is allowed, but not required. The host institute will provide office space and a working environment to collaborate on the selected project, with the fellows fully integrated in the activities of the respective groups.

Selection

Apply here. Applicants are requested to submit a brief (1 page) description of their proposed research project/research interests, including the preferred host institution, their CV (including software experience), and two letters of recommendation, out of which one should be from the advisor. We will conduct video interviews with short-listed candidates. The selected candidates will be placed equally between the two institutions, based on their preferences, and will each be assigned a primary mentor.

Potential Projects

The examples below should be taken only as guidance: it is ok to highlight in your research statement one or even two areas you are most intersted in working on, without a very concrete project in mind at the time of application.

1. AI/ML and HPC Integration with MC Simulations
   • AI/ML Models for Hadronization. Unlike the reactions involving high-energy quarks and gluons, the process of hadron formation from these fundamental particles is not understood from first principles. Therefore, existing hadronization models are physics-inspired, but ultimately limited to simple parameterized fits to data. The flexibility of ML-based models offer an attractive alternative to existing parametric models. The goal is to continue the development and training of an ML-based hadronization model and interface it with the Pythia or Herwig MC event generators using inference libraries like ONNX.
   • AI/ML Models for Material Interactions. Material interactions are usually the slowest part of a full event simulation. This is particularly the case for dense materials, as is the case for calorimeter detectors, since many secondary particles are produced and must be tracked. ML-based models can be surrogates for the complex microphysics of material interactions, but much faster to sample. The goal is to train and integrate a model into Geant.
   • Offloading all or parts of MC calculations to hardware accelerators. Machine learning models are naturally compatible with hardware accelerators like graphical processing units (GPUs), which can be much faster than CPU-based computing. However, when only one component of an MC simulation is a neural network (or written in a GPU-compatible differential programming language like JAX), additional CI is needed to make use of the accelerator. The goal is to build inference-as-a-service into MC simulations to efficiently offload expensive calculations to GPUs.
2. Generic Tools and Interfaces
   • Automated Validation/Continuous Integration. Given the large number of users for MC simulations, it is critical that even small changes are tracked and documented so that end users can know which updates are expected to introduce changes in physics and which ones should only introduce changes in efficiency. This is true also for developers. Tools that can automatically run a series of tests for any new commit are extremely valuable. The goal is to build the continuous integration pipeline for Pythia using probing tests (e.g. comparing histograms of observables for important reactions) and to track pseudo-random number branchings.
   • Universal Interfaces. For a particular step in the generation of a single event, there are typically multiple simulators capable of transforming the same inputs into the same type of outputs. This is critical, so that different physics model assumptions can be tested against each other. However, each MC program has its own interface, with its own schemes and naming conventions. The goal of this project is to create a universal interface to run MC simulators so that the exact syntax of a particular generator is not required for the end user. Such interfaces already exist in the software suites of most particle/nuclear physics experiments, but they are not easy to decouple from the framework for general use.
   • Automated Documentation. Despite the best efforts by MC simulation authors, documentation is always a challenge to maintain. However, well-documented software is essential for the end-users to deploy methods accurately. The goal of this project is to build CI around large language models for automating the process of creating documentation and providing real-time, automated user support.
   • Coordinated Particle Data. The Particle Data Group (PDG) provides a centralized repository of experimentally measured paricle/nuclear physics data, including particle masses, lifetimes, spin configurations, and other particle properties. The major MC event generators use these particle properties as reported by the PDG, but each generator uses an independently maintained database. Further complicating the issue, the measured decays for many particles are incomplete, and so each generator group performs non-trivial theory extrapolations of the data to create bespoke decay tables. The goal of this project is to provide a unified set of particle data, including decay tables, that can be easily updated from the PDG, but is also accessible to all the major MC event generators.
3. Physics Improvements to Specific Simulators
   • Simulations of Dark Sector Models. Discerning the nature of dark matter, i.e. new matter that is not optically visible but is known to exist from cosmological and astrophysical observations, is one of the major open problems in physics. Not surprisingly, a significant effort is therefore devoted to searching for dark matter in a number of different ways. Here, one of the main challenges is to understand the possible ways in which dark matter or related particles (usually termed a dark sector) could give rise to detectable signals in particle/nuclear physics detectors. MC generators are an important part of both devising the experimental searches and interpreting results. Reliable simulations of dark sectors are essential, but state-of-the-art MC tools are lacking in several respects. There are a number of possible projects appropriate for six month traineeships, that would improve the precision of these MC tools: (1) including helicity correlations in generic dark sector models, (2) including a data-driven description of heavy neutral leptons in Pythia, (3) improving dark matter direct detection description in GAMBIT, (4) implement dark sector/dark showers inside the parton shower modules in Pythia and/or Herwig.
   • Improvements to the Pythia Parton Shower. The parton shower (PS) algorithm, a variant of Markov-chain MC, is at the core of MC event generators. Physics improvements there are most likely to have the largest impact on expected high-precision measurements in PNP. The PS algorithm is extremely flexible, but also prone to inefficiencies. One possible project is to use the new plug-in library in Pythia to enable corrections to more PS splittings. Another is to expand on the recent progress in PS-induced quarkonium production to include double heavy-baryon production. The Pythia simulation of heavy-ion collisions could also be expanded to include strong medium effects in the PS. Finally, rare decay searches require a simulation of photon radiation from hadronic final states, best included with a multi-pole PS.
   • Interfaces between Pythia and Geant4. Pythia includes extensive models of hadron-hadron interactions at low and medium energies. This is the same physics as simulated by Geant4, which has ported some of the Pythia code. We envision standardizing an interface of our hadronic interaction, hadronic rescattering, and string hadronization models so that users of our models are synchronized with our developments. In general, we aim to explore how recent developments in the use of the MC veto algorithm can be applied to improve hadronic interaction algorithms.
   • Hybrid Hadronization and Hadronic Interactions. The observation of correlated hadrons in heavy-ion collisions has spurred interest in exploring the quark-gluon plasma. A challenge is to harmonize the hydrodynamic phase of a collision with the standard particle-physics description. One possible project is to combine hydrodynamic and jet shower simulations in a coupled approach in the \jetscape simulation. In this way, soft partons hadronize using a hydrodynamic approach, while hard partons are hadronized through string fragmentation. Furthermore, given the high density of hadrons in heavy-ion collisions, a project could incorporate hadronic interaction in the MC simulations after hadronization.