Constraining cubic and f(P) gravity models with linear regression using RANSAC and polynomial regression

The universe is expanding, a fact established nearly a century ago when astronomers noticed that distant galaxies are moving away from us. For decades, scientists believed this expansion would eventually slow down, pulled back by the gravity of all the matter in the cosmos. However, observations in the late 1990s revealed a startling truth: the universe is not slowing down; it is speeding up. To explain this acceleration, physicists proposed the existence of "dark energy," a mysterious force pushing space apart. While the standard theory of gravity, developed by Albert Einstein, works exceptionally well for most of the universe, it struggles to explain this cosmic acceleration without invoking dark energy. This has led researchers to explore "modified gravity," theories that tweak the fundamental laws of how gravity works on the largest scales, potentially removing the need for dark energy entirely. Two such theories, known as cubic gravity and f(P) gravity, suggest that the equations governing space and time are more complex than previously thought, involving higher-order mathematical terms that become significant only in the vast emptiness of the cosmos.

A team of researchers from GLA University in India has recently put these two modified gravity theories to a rigorous test using modern computational tools. Their goal was to see if these complex mathematical models could accurately predict the rate at which the universe is expanding at different points in its history. To do this, they turned to a dataset containing thirty specific measurements of the Hubble parameter, a value that tells us how fast the universe is expanding at a given distance. Instead of relying solely on traditional statistical methods, the team employed a suite of machine learning techniques—computer algorithms designed to find patterns in data. They trained these algorithms on a portion of the data and then asked them to predict the expansion rates for the remaining points. The researchers compared the predictions made by the machine learning models against the actual observational data and the predictions made by the theoretical gravity models themselves.

The study focused on four specific machine learning approaches: standard linear regression, which looks for straight-line relationships; artificial neural networks, which mimic the way the human brain processes information; linear regression with RANSAC, a method designed to ignore outliers or "bad" data points; and polynomial regression, which looks for curved, non-linear relationships. The researchers found that the relationship between the distance of galaxies and their expansion speed is not a simple straight line, but a curve. This made the polynomial regression technique particularly effective. When the team measured how closely the machine learning predictions matched the real-world observations, they used a metric called "alpha deviation," where a lower number indicates a better fit. The results showed that the polynomial regression technique performed exceptionally well, producing values that were nearly identical to the theoretical predictions of the cubic and f(P) gravity models. In fact, the polynomial regression outperformed the other machine learning methods, including the neural networks and the linear regression with RANSAC, in capturing the subtle, non-linear behavior of the universe's expansion.

One of the most significant findings was that the machine learning models did not just guess; they confirmed the validity of the theoretical frameworks. The polynomial regression model produced an alpha deviation of 0.1362 for both the cubic and f(P) gravity models, a figure that is remarkably close to the theoretical value of 0.1360 for cubic gravity and 0.1362 for f(P) gravity. This suggests that the complex equations proposed by these modified gravity theories are consistent with the actual data we observe in the sky. While the standard linear regression method struggled because it tried to force a straight line onto a curved reality, the polynomial regression successfully navigated the curvature of the data. The study also noted that the linear regression with RANSAC, which is designed to handle messy data, performed better than standard linear regression but still did not match the precision of the polynomial approach.

The researchers conclude that these advanced machine learning techniques are powerful tools for testing the limits of our understanding of gravity. By using polynomial regression, they were able to demonstrate that the cubic and f(P) gravity models are robust and capable of describing the universe's acceleration without necessarily requiring the introduction of dark energy. The study does not claim to have solved the mystery of dark energy or to have proven that these modified theories are the final answer, but it does provide strong evidence that these specific mathematical models align well with observational data. The authors suggest that future work should involve larger datasets and the exploration of other cosmological models with even more advanced machine learning techniques. For now, this research offers a clear demonstration that when the universe behaves in a curved, complex way, the right kind of mathematical curve-fitting can reveal the underlying laws of nature with surprising clarity.