siren-lartpc

Implementation of sinusoidal representation network (siren) for modeling the transport of optical photon signals in Liquid Argon Time Projection Chambers (LArTPCs).

Installation

This repository requires photonlib package that you can download from here. Install the photonlib following the instructions on its webpage.

Then git clone this repository and:

pip install .

Siren for modeling optical photon transport

In the simplest term, the optical photon transport can be modeled by a function that calculates the probability for the detector $d$ to observe a photon produced at a position $\vec{r}$. A naive form of implementation is to only estimate the mean probability in a deterministic manner (i.e. ignore stochasticity over a distribution), namely $f:R^N\mapsto R$.

Traditionally, LArTPC experiments have employed Photon Library (here) for this modeling. We have shown that siren, a neural network designed to model a continuous field in space (and learns accurate gradients), brings significant advantages to replace Photon Library. Read the original paper to learn about siren here. The study is shown in this paper where known issues (in particular, scalability) for Photon Library are also discussed. This repository implements siren for LArTPC optical photon transport including scripts to run optimization of siren.

Training siren

You need two items:

• a data file for Photon Library
  • You can train siren given a Photon Library data file. Follow the instructions here and download the photon library file.
• a configuration file to run the training script
  • You can prepare by yourself or use an example provided in this repository at slar/configs directory.

Example configuration files can be found from the terminal:

python3 -c "from slar.utils import get_config;print(get_config('icarus_train'))"