Fusion

Physics is all about approximation. No one formula can account for every factor and predict any event with full accuracy. Over the years, scientists have used different methods to make these predictions, such as mathematical formulas and statistical models. This has been particularly difficult in fields such as astrophysics, where the subjects being studied are mind-blowing distances away from Earth. In recent years, the advancements in machine learning, specifically in neural networks, allow us to computationally model far more complex relationships than before. The Fusion Stellar Model is designed to model, predict, and classify stellar parameters more accurately than traditional methods. It takes inputs of a star’s Effective Temperature, Luminosity, and Radius, and runs its Physics-Informed Neural Network (PINN) algorithm to find twenty-three other parameters of a star such as Mass and surface Gravity. It was trained on the Gaia mission dataset with over 403 million stellar parameters. The performance of this model varies depending on the star type and parameter it predicts but has an average accuracy of 90.018% and an average speed of 0.008 seconds per round of predictions. While this model is currently limited to stars on the Hertzsprung-Russell diagram (i.e., the model cannot make accurate predictions on other objects in the universe, such as a neutron star), its main use is to better model the distant lights in our universe.

The Project

The Fusion model is a deep neural network that can tackle regression and classification problems, depending on the parameter of a star it is predicting. Here are the required inputs and output parameters the model is compatible with:

Inputs (at least two of the following)

The target stars:

• Effective Temperature (in Kelvin)
• Radius (in solar radii)
• Luminosity (in solar luminosity)

Outputs

Outputs calculated outside of the model based on input values

• Diameter (D/Do) (based on the radius input)
• Volume (V/Vo) (based on the radius input)
• Surface Area (SA/SAo) (based on the radius input)
• Great Circle Circumference (GCC/GCCo) (based on the radius input)
• Great Circle Area (GCA/GCAo) (based on the radius input)

Outputs predicted by the model (and whether they are a regression or classification prediction)

• Absolute Bolometric Magnitude (Mbol) (regression)
• Absolute Magnitude (M)(Mv) (regression)
• Absolute Bolometric Luminosity (Lbol) (log(W)) (regression)
• Mass (M/Mo) (regression)
• Average Density (D/Do) (regression)
• Central Pressure (log(N/m^2)) (regression)
• Central Temperature (log(K)) (regression)
• Lifespan (SL/SLo) (regression)
• SurfaceGravity (log(g)...log(N/kg)) (regression)
• Gravitational Binding Energy (log(J)) (regression)
• Bolometric Flux (log(W/m^2)) (regression)
• Metallicity (log(MH/MHo)) (regression)
• Spectral Class (classification) (class M, K, G, F, A, B, or O)
• Luminosity Class (classification) (class D, V, IV, III, II, Ib, or Ia)
• Star Peak Wavelength (nm) (regression)
• Star Type (classification) (class Brown Dwarf, Red Dwarf, White Dwarf, Main Sequence, Supergiant, or Hypergiant)

Outputs calculated outside the model but based on model predictions

• Bolometric Correction (based on the predicted bolometric and absolute magnitude values)
• Stellar Classification (based on the predicted Spectral and Luminosity classes. For example, the Sun is a G2V star—the '2' is a subclass (from 0–9, where 0 is the hottest subclass and 9 is the coolest) of the 'G' spectral class, and the 'V' is the star's luminosity class)

The model uses a TensorFlow/Keras backend with custom functions tailored to the specific regression and classification problems described above as well as overall model improvement.

Usage

Please take a look at the license found in this directory under the filename LICENSE

Install the application

The User Interface of the project is complete. Once it is packaged, I will upload the application files for download here. The User Interface can currently be used by cloning the repository and running Fusion/FusionUI/widget.py as shown below. To model a single star, simply input at least two of the following three input parameters of the star:

• Effective Temperature (Kelvin)
• Luminosity (Solar Luminosities)
• Radius (Solar Radii)

and click the Model Star(s) button.

You can export the simulation data as a Comma-Separated Values file (.csv) by clicking the Export button.

Furthermore, you can import csv files containing input data for multiple starts by clicking the middle import button on the top of the front page. The prediction data for each of these stars can also be exported as a csv file after clicking the button. On the left of this button links to this GitHub repository for help, and the button to the right of that middle button gives application information.

Clone the repository

Make sure Python is installed.

With git, run:

git clone https://github.com/anshmenghani/Fusion.git

Be sure to create a virtual environment in the resultant Fusion directory and activate it.

Assuming Python is already installed, run (from the resultant Fusion directory)

pip install -r requirements.txt

Fusion is now set up in your directory. Each file has a description in the comments. Run Fusion/FusionUI/widget.py to launch the User Interface from the directory.

Contact Information for Inquiries:

ansh.menghani@gmail.com