Decoding the Imprints of Energy-Momentum Squared Gravity in Neutron Stars with Machine Learning Analysis
This study demonstrates that supervised machine learning classifiers, particularly Random Forest, can achieve over 99% accuracy in distinguishing between General Relativity and Energy-Momentum Squared Gravity models using neutron star observables (mass, radius, tidal deformability, and oscillation frequency), even after applying observational constraints.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Gravity is the force that holds the universe together, shaping everything from falling apples to the orbits of planets. For over a century, Albert Einstein's theory of General Relativity has been the gold standard for describing how gravity works. It has passed every test we have thrown at it, from the bending of starlight to the ripples in space-time created by colliding black holes. Yet, many physicists suspect that this theory is not the final word. In the most extreme environments in the cosmos, where matter is crushed to densities higher than an atomic nucleus, the rules might change. To find out, scientists look to neutron stars. These are the collapsed cores of dead stars, so dense that a single teaspoon of their material would weigh billions of tons on Earth. Because they pack so much mass into such a small space, they create gravitational fields far stronger than anything we can create in a laboratory, making them the perfect natural laboratories for testing if gravity behaves differently than Einstein predicted.
One such alternative theory is called Energy-Momentum Squared Gravity. In this framework, the equations that describe gravity include an extra term that depends on how much energy and pressure are packed into a specific region. While this theory behaves exactly like Einstein's in empty space, it predicts that inside dense objects like neutron stars, gravity should act slightly differently. The strength of this extra effect is controlled by a single number, a coupling parameter. If this number is positive, the theory suggests the star's interior is pushed outward, allowing it to be larger and heavier. If the number is negative, the star is squeezed tighter, becoming more compact and lighter. The big question is whether we can actually tell the difference between these scenarios using the data we can observe from Earth.
A team of researchers set out to answer this by simulating thousands of neutron stars under different conditions. They did not just look at one or two models; they generated a massive dataset of about 10,000 different neutron stars, each built using a different set of rules for how nuclear matter behaves. For each of these stars, they calculated what the star would look like if the extra gravity term were positive, negative, or zero (which would mean it behaves exactly like Einstein's theory). They focused on four key properties that astronomers can actually measure or infer: the star's mass, its radius, how easily it can be stretched by a companion star's gravity, and the frequency at which it vibrates like a bell after being disturbed.
The researchers then applied a filter to their massive dataset, keeping only the stars that matched what we currently know about the real universe. For instance, they discarded any star that was too small or too light, because we have already observed neutron stars that are heavier and larger than those limits. This left them with a collection of "observationally viable" stars—models that could actually exist in our universe. The next step was to see if a computer could look at the four properties of these surviving stars and correctly guess which version of gravity they belonged to. To do this, they used machine learning, a type of artificial intelligence that learns patterns from data. They trained several different computer algorithms to act as classifiers, teaching them to recognize the subtle fingerprints left by the different gravity theories.
The results were strikingly clear. Even after filtering out the impossible stars, the remaining ones still carried distinct signatures of the gravity theory they were built under. The machine learning models were able to distinguish between the positive, negative, and zero cases with incredible accuracy. One specific algorithm, known as a Random Forest, correctly identified the gravity type for nearly every single star in the test set, achieving an accuracy rate of about 99.85%. Other methods also performed very well, with accuracy rates exceeding 99%. The computer was essentially looking at a star's size, weight, stretchiness, and vibration, and from those numbers alone, it could tell with near-perfect certainty whether the star was governed by Einstein's gravity or one of the modified versions.
This finding suggests that the imprints of modified gravity are not washed out by the natural variations in how stars are made. Even when we restrict our view to only the stars that fit our current observations, the differences in their structure remain sharp enough to be detected. The study implies that as our telescopes and gravitational wave detectors become more precise, we may soon be able to use these same machine learning tools to analyze real data from neutron stars. By doing so, we could potentially confirm or rule out these alternative theories of gravity, moving us closer to understanding how the universe works in its most extreme corners. The work does not prove that Einstein's theory is wrong, but it demonstrates that if a different theory is true, our instruments and computers are already capable of finding the evidence.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.