Bayesian Active Search on Parameter Space: a 95 GeV Spin-0 Resonance in the (B−L)SSM

Code repository for https://arxiv.org/abs/2404.18653.

Overview

In the attempt to explain possible data anomalies from collider experiments in terms of New Physics (NP) models, computationally expensive scans over their parameter spaces are typically required in order to match theoretical predictions to experimental observations. Assuming the anomalies observed at around 95 GeV by the Large Electron-Positron (LEP) and Large Hadron Collider (LHC) experiments signify an NP signal, we interpret it as a spin-0 resonance within the $(B-L)$ Supersymmetric Standard Model ($(B-L)$SSM). We introduce a novel Machine Learning (ML) approach, named b-CASTOR, to efficiently scan the parameter space. This method leverages Gaussian Processes surrogate to approximate multiple objectives constrained by experimental measurements and employs a volume-based acquisition function to ensure a comprehensive characterisation of the satisfactory region in the parameter space. Our method outperforms traditional competing algorithms, such as those based on Markov-Chain Monte Carlo (MCMC) methods, offering a more effective strategy for exploring parameter spaces in computationally expensive BSM scenarios.

Requirements

The hep-aid library is required. To install it, clone the hep-aid repository and run pip install -e ..

To run the experiments from the paper, the following HEP tools need to be installed: SPheno, MadGraph, HiggsBounds, and HiggsSignals.

If the HEP tools are not already installed, run the Python script install_hepstack.py, located in experiments_paper/blssm/hepstack. The code will install all necesary HEP Software using the SPheno and UFO files can be found in the BLSSM_SPheno and BLSSM_UFO folders, also located in experiments_paper/blssm/hepstack.

Lastly modify the configuration file located experiments_paper/blssm/configs/hep_stack_config.yaml with this format:

model: 'BLSSM'
reference_slha: 'configs/hep_files/diphoton_paper_v2'        
directory: '/path/to/spheno_directory/'

higgsbounds:
    neutral_higgs: 6
    charged_higgs: 1
    directory: '/path/to/higgsbounds_directory/build'

higgssignals:
    neutral_higgs: 6
    charged_higgs: 1
    directory: '/path/to/higgssignals_directory/build'

madgraph:
    directory: '/path/to/madgraph_directory/'
    scripts: 
        gghaa: "configs/hep_files/mg5/blssm_pphaa_LHC13.txt"
hep_stack: 
    name: 'SPhenoHBHSMG5'
    scan_dir: '/path/to/scan_directory/'
    final_dataset: 'datasets'
    delete_on_exit: True

In this case we are using some relative paths to the experiments_paper/blssm directory. A manual for the hep-aid library is currently under development.

Manual installation of HEP Software

The SPheno and UFO filese files can also be generated using SARAH by running it in a Mathematica notebook.

Needs["SARAH`","/SARAH-4.14.5/SARAH.m"]

Start["B-L-SSM"]
ModelOutput[EWSB]

MakeSPheno[]
MakeUFO[]

Quit[];

SPheno is set up following section 6.1.2 in [1]. After downloading SPheno, copy the ouput of SARAH as follows,

mkdir $SPHENO_PATH/BLSSM/
cp $SARAH_PATH/SARAH/Output/B-L-SSM/EWSB/SPheno/* BLSSM/

and run

make Model=BLSSM

an executable SPhenoBLSSM will be created which will be used by the script in this package. HiggsBoundsand HiggsSignals, which are part of HiggsTools now, should installed. This code uses only HB and HS. Finally MadGraph with the UFO model file copied in the models directory.

Results

The main Python script for both case studies is located in run_search.py located in experiments_paper/test_function/ and experiments_paper/blssm/. It defines a hyper-parameter sweep function that updates the configuration for the b-CASTOR algorithm based on a specified combination of the priority parameter $\alpha$, TPE trials $N_{TPE}$, and resolution $r$ as defined in the paper. The function can perform either a sweep for constant resolution or sweep_decay for decaying resolution. Each combination is accessed using the --n argument in the script for easier replication.

2D Double Objective Test Function

The working directory for the test funcion study is located in experiments_paper/test_function. Configuration files are in experiments_paper/test_function/configs.

For the b-CASTOR algorithm run,

python run_search.py --method bcastor --n 8 --mode sweep_decay --config configs/bcastor_hyper_params.yaml --channels himboo

For the MCMC search,

python run_search.py --method mcmc --config configs/mcmc.yaml --channels himboo

$(B - L)SSM$ for a 95 GeV Higgs

The working directory for the B-L SSM model study case is located in experiments_paper/blssm. Configuration files are in experiments_paper/blssm/configs/hep_files, with the SPheno reference LHE file at configs/hep_files/diphoton_paper_v2 and the Madgraph script files for calculating cross-sections of each channel located in experiments_paper/blssm/configs/hep_files/mg5.

Modify the corresponding hyper-parameter configuration for bcastor in experiments_paper/blssm/configs/bcastor*, specifically for local runs depending on the number of cores of your machine. These scripts were run on a node of the Iridis5 cluster, which has 40 cores per node.

b-CASTOR

For the $\gamma\gamma$ case run,

python run_search.py --method bcastor --n 19 --mode sweep_decay --config configs/aa_obj_hep.yml --channels aa

For the search in $\gamma\gamma$ and $bb$,

python run_search.py --method bcastor --n 19  --mode sweep_decay --config configs/aa_bb_obj_hep.yml --channels aabb

Finally, for the search in the three channels $\gamma\gamma$,$bb$ and $\tau\tau$,

python run_search.py --method bcastor --n 19  --mode sweep_decay --config configs/aa_bb_tautau_obj_hep.yml --channels all

MCMC Metropolis Hastings

For the $\gamma\gamma$ case run,

python run_search.py --method mcmc  --config configs/aa_obj_hep.yml --channels aa

For the search in $\gamma\gamma$ and $bb$,

python run_search.py --method mcmc --config configs/aa_bb_obj_hep.yml --channels aabb

Finally, for the search in the three channels $\gamma\gamma$,$bb$ and $\tau\tau$,

python run_search.py --method mcmc  --config configs/aa_bb_tautau_obj_hep.yml --channels all