Exploring Beyond Standard Models (BSM) Using Machine Learning

Project Description

This project aims to identify the Wilson coefficients within the Standard Model Effective Field Theory (SMEFT). The Lagrangian can be expressed as:

$$ \mathcal{L}{BSM}=\mathcal{L}{{SM}}+\sum_o \frac{f_o}{\Lambda^2} {O}_o\ . $$

Workflow

Data Acquisition:

Identify New Interactions: Establish the Lagrangian form.
Sample Parameter Points: Define the approximation around the SM. Utilize MadGraph to produce particle-level metadata (four-momentum) for various Wilson coefficient combinations. (Note: MadGraph only runs in the SM; subsequent data from different parameter points are acquired through reweighting).
Data Generation: Use Pythia8 and Delphes for detector-level data creation.

Machine Learning:

Construct Features and Labels: Detector data builds features, and particle-level data forms the target label.
Neural Network Training: Train the neural network.

Testing:

Predict and Output Likelihood: The ML model provides the Likelihood Score from the test set features. (Done)
Identify Wilson Coefficients: Employ Maximum Likelihood to determine the new physics' Wilson coefficients. (To do)
Assess Confidence Intervals: Measure the confidence intervals.

Dependencies

MadGraph: Check MadGraph’s website for setup details. Requires Fortran compiler and Python 3.7+.
Pythia8:
```
$ ./bin/mg5_aMC
> install pythia8
```
Delphes: Requires Root.
```
$ ./bin/mg5_aMC
> install Delphes
```
Madminer:
```
pip install madminer
```
Alternatively, use the library in this repository and run:
```
pip install -r environment.yml
```

Usage

Three sample Jupyter notebooks in the examples directory showcase the full workflow, focusing on the electron-positron collision process, leading to the production and decay of a top quark pair, emphasizing operators involving $Z$.

$$ O_{\phi Q}^{(3)}=(\phi^{\dagger} \tau^{I} i \overset{\leftrightarrow}{D}{\mu} \phi) \bar{Q}{L} \gamma^{\mu} \tau^{I} Q_{L} $$

$$ O_{\phi u}=(\phi^{\dagger} i \overset{\leftrightarrow}{D}{\mu} \phi) \bar{Q}{L} \gamma^{\mu} u_{R} $$

With the corresponding Lagrangian:

$$ \mathcal{L}{BSM}=\mathcal{L}{{SM}}+ \frac{C_{\phi Q}^{(3)}}{\Lambda^2} O_{\phi Q}^{(3)}+ \frac{C_{\phi u}}{\Lambda^2} O_{\phi u} $$

More To Do:

Investigate various processes and interactions.
Introduce new features (observables).
Add background noise.
Explore deep learning models (rethink probability modeling, alternate loss functions, different objective functions, and deep learning model tuning).

License

This project is licensed under the MIT License.

KeriYuu/Machine-Learning-for-Particle-Physics