Thesis

AutoBSM - Scalable Automation of Beyond the Standard Model (BSM) Physics Validation

Details

  • Call:

    PT-CERN Call 2021/1

  • Academic Year:

    2021/2022

  • Domain:

    Astrophysics

  • Supervisor:

    Miguel Crispim Romão

  • Co-Supervisor:

    Nuno Castro

  • Institution:

    Universidade do Minho

  • Host Institution:

    Laboratório de Instrumentação e Física Experimental de Partículas

  • Abstract:

    As of 2021, the Large Hadron Collider (LHC) has yet to find compelling evidence of New Physics (NP), leaving the High-Energy Physics (HEP) community with a tantalising collection of possible BSM candidate theories to be tested against experimental data. This validation task comprises the search for regions of the parameter space where predictions of the BSM model are in agreement with current experimental data. This involves exploring multidimensional parameter spaces and heavy numerical computations of predictions that are then compared to experimental data. To mitigate the overhead of this step, BSM researchers often simplify their models in order to find viable regions of the parameter space within a reasonable time. This is often done by either simplifying the model, which reduces the dimensionality of the parameter space and the numerical computations involved, or by reducing the constraints on which the BSM model is tested against, effectively increasing the volume of the valid parameter space. In either case, one ends up validating a simplified variation of the BSM model solely due to computational constraints. The student will help develop AutoBSM, a scalable tool for automated BSM physics validation. For this, AutoBSM needs to tackle two challenges in the BSM researchers' workflow: (1) how to efficiently validate a BSM candidate model against experimental data, i.e. how to perform a thorough scan of the parameter space; and (2) how to allow for a platform-agnostic deployment of the code, e.g. on a High Performance Computing (HPC) cluster or in the Worldwide LHC Computing Grid (WLCG). For (1) Machine Learning (ML) algorithms will be used to improve the efficiency of parameter space scanning, reducing the time needed, by orders of magnitude, to find valid regions of the parameter space. For (2) containerised solutions, such as Docker, will be used to deploy such scans in a platform-agnostic way, which will include all the required software dependencies.