Skip to content

A translation of crucial parts of GRTL in torch for accelerated learning

License

Notifications You must be signed in to change notification settings

ThomasHelfer/TorchGRTL

Repository files navigation

Unittest License: MIT

TorchGRTL - Numerical General Relativity in PyTorch

no alignment

Overview

TorchGRTL is a Python library for the application of deep learning to Numerical Relativity. It is based on the GRTL codebase (formerly known as GRChombo) codebase.

Installation

Prerequisites

Before installing TorchGRTL, ensure you have the following prerequisites:

  • Python 3.8 or higher
  • pip package manager

Steps

  1. Clone the TorchGRTL repository:

    git clone https://github.com/ThomasHelfer/TorchGRTL.git
    cd TorchGRTL
  2. Install the package:

    pip install .
  3. (Optional) Set up pre-commit hooks for code formatting and linting:

    pre-commit install

Usage

Training models

To first learn a model please download the training dataset from https://huggingface.co/datasets/thelfer/BinaryBlackHole and adapt the path in yaml <configs/factor_2.yaml>.

factor: 2
# Data Source
filenamesX: "<ADAPT TO YOUR LOCAL PATH>/outputXdata_level{res_level}_step*.dat"
filenames_check: "/home/thelfer1/scr4_tedwar42/thelfer1/high_end_data_4/outputXdata_level{res_level}_step*.dat"
# Restarting
restart: False

To run the code, simply run

python learn_error.py configs/factor_2.yaml

Evaluate models metrics

Copy your model fold in the /models folder and run

python evaluate_models.py 

the corresponding metrics in the outputfile 'metrics_results.csv' can be visualised in notebook.ipynb

Computing Constraints

You can compute the Christoffel symbols, which are crucial in the context of general relativity for defining the Levi-Civita connection and geodesic equations:

# Compute the Christoffel symbols using the standard method
chris = compute_christoffel(d1['h'], h_UU)

In these examples, d1['h'] refers to the first derivatives of the metric tensor, and h_UU is the inverse metric tensor.

Calculating Hamiltonian and Momentum Constraints The library can also compute more complex quantities like the Hamiltonian and Momentum constraints, which are fundamental in ensuring the consistency of solutions in numerical relativity:

# Compute the Hamiltonian and Momentum constraints
out = constraint_equations(vars, d1, d2, h_UU, chris)

Here, vars contains various tensor fields, d1 and d2 are the first and second derivatives of these tensor fields, and chris is the computed Christoffel symbols.

Self-Contained Example

For a full, self-contained example that demonstrates the library's capabilities, refer to example.py in the repository. This example will guide you through a typical use case, showing how to leverage TorchGRTL for numerical relativity simulations and calculations.

License

TorchGRTL is released under the MIT License. See LICENSE for more details.

Contact

For questions or support, please contact Thomas Helfer at thomashelfer@live.de.

About

A translation of crucial parts of GRTL in torch for accelerated learning

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published