Stars are the fundamental component of our universe. Like our Sun, most stars host planets that host our solar system. More broadly, groups of stars make up huge structures such as clusters and galaxies. Therefore, before astrophysicists can try to understand these large-scale structures, they must first understand the fundamental properties of stars, such as stars, radius, and temperature.
However, these basic properties have proven to be extremely difficult to measure. This is because the stars are literally at astronomical distances. If our sun was basketball on the East Coast of the US, then the nearest star, Proxima, will be Hawaiian orange. Even the world's largest telescopes can't solve orange in Hawaii. Measuring radius and star masses appears to be out of reach of scientists.
Enter the binary star. A binary is a system of two stars that rotate around the mutual centre of the masses. Their movement is governed by Kepler's harmonic method, which connects three important quantities. It is the size of each orbit, the time taken to a trajectory called the orbital duration, and the total mass of the system.
I am an astronomer and my research team has worked on advances in theoretical understanding and modeling approaches to binary stars and multiple star systems. For the past 20 years, we have pioneered the use of artificial intelligence in interpreting observations of these cornerstone celestial bodies.
Measurement of the mass of a star
Astronomers can measure the orbital size and duration of binary systems easily from observations. So, these two pieces allow you to calculate the total mass of the system. Kepler's harmonic method serves as a measure for weighing celestial bodies.
Think of a seesaw for a playground. If two children weigh almost the same, they should sit at approximately the same distance from the midpoint. However, if one child is large, he or she must sit nearby, and the small child is far from the midpoint.
It's the same as a star. The larger the binary pair star, the closer the center, the slower the more it rotates. When astronomers measure the speed at which a star travels, they can see how large the star orbit is and, as a result, they also need weight.
Measurement of the stellar radius
Unfortunately, Kepler's harmony method tells astronomers nothing about the radius of the star. For them, astronomers rely on another accidental feature of Mother Nature.
The binary star orbit is randomly oriented. Sometimes the telescope's gaze is flat and the binary star system is on track. This accidental alignment means that stars scream at each other as they unfold. These solar eclipse shapes allow astronomers to find the radius of the star using simple geometry. These systems are called ecliping binary stars.
More than half of all sun-like stars are found in binary, with Eclipsing Binaries accounting for around 1% to 2% of all stars. That may sound low, but the universe is vast, so there are many appetizing systems for hundreds of millions of people in the galaxy alone.
By observing the eclipse binary, astronomers can measure not only the masses and radius of the stars, but how hot and how bright they are.
Complex problems require complex computing
Even using an Equilap Binary, measuring the characteristics of a star is not easy. Stars deform when rotated and pulled together in a binary system. They can interact, illuminate each other, have spots and magnetic fields, and tilt them like this.
To study them, astronomers use complex models with many knobs and switches. As input, the model uses parameters – for example the shape and size of a star, its orbital properties, or the amount of light it emits – predicts how observers will see such appetizing binary systems.
Computer models take time. Predicting a computing model typically takes several minutes. To make sure we can trust them, we need to try many parameter combinations – usually tens of millions.
Many of these combinations require hundreds of millions of minutes of computation time just to determine the basic characteristics of the star. That amounts to over 200 years of computer time.
Computers linked to clusters can calculate faster, but even with computer clusters it takes more than three weeks to “solve” a single binary, determine all parameters. This task explains why astronomers only have around 300 stars to accurately measure basic parameters.
The models used to solve these systems have already been significantly optimized and cannot proceed much faster than they already do. Therefore, researchers need a whole new approach to reduce computing time.
Use deep learning
One solution my research team explored involves deep learning neural networks. The basic idea is simple. I wanted to replace the expensive computational physics models with a much faster, AI-based model.
First, we calculated a huge database of predictions about hypothetical binary stars – using features that astronomers can easily observe – altered the properties of virtual binary stars. We're talking about hundreds of millions of parameter combinations. We then compared these results with actual observations to see which ones best match. AI and neural networks are perfect for this task.
In short, neural networks are mapping. Maps a specific known input to a specific output. In our case, they map the characteristics of binary covered by the expected predictions. Neural networks emulate binary models, but do not need to explain all the complexity of a physical model.
A neural network displays each prediction from a database and trains a set of properties that are used to generate it. Once fully trained, neural networks can accurately predict what astronomers should observe from the specific characteristics of a binary system.
Compared to the runtime of a physical model for a few minutes, neural networks use artificial intelligence to get the same results in one second.
Enjoy the benefits
A small portion of the second means a runtime reduction of approximately 1 million times. This reduces the time from weeks on a supercomputer to minutes on a single laptop. It also means that computer clusters can analyze hundreds of thousands of binary systems in a few weeks.
This reduction means that you can obtain the basic properties (star mass, radius, temperature, luminosity) of all eclipse binary stars observed within a month or two. The big remaining challenge is to show that AI results actually give the same results as physical models.
This task is at the heart of my team's new paper. In it, we showed that in fact, AI-driven models yield the same results as physical models at over 99% of parameter combinations. This result means that AI performance is robust. Our next step? Deploy AI to all observed eclipse binaries.
Above all? We applied this methodology to binary, but the basic principles apply to the complex physical models there. Similar AI models have already accelerated many real-world applications, from weather forecasting to stock market analysis.
Andrej Prša receives funding from the National Aeronautics and Space Administration.
/Commentary of the conversation. This material of the Organization of Origin/Author is a point-in-time nature and may be edited for clarity, style and length. Mirage.news does not take any institutional position or aspect, and all views, positions and conclusions expressed here are the authors alone.
