Generative AI for Validating Physics Laws

TL;DR

The paper addresses validating fundamental physics laws, focusing on the Stefan-Boltzmann law that links stellar temperature, radius, and luminosity. It introduces a generative AI approach within a causal counterfactual framework to simulate luminosity under alternative temperatures and learns personalized temperature–luminosity dependencies via a Fourier-series neural network. Using Gaia DR3 main-sequence stars, it shows that the temperature effect on luminosity is positive on average, increases with radius, and is stronger for intrinsically brighter stars, aligning with $L \propto R^2 T^4$. The approach provides a data-driven, counterfactual method to refine physical theories and supports evidence-based practice, with potential to extend to other laws.

Abstract

We present generative artificial intelligence (AI) to empirically validate fundamental laws of physics, focusing on the Stefan-Boltzmann law linking stellar temperature and luminosity. Our approach simulates counterfactual luminosities under hypothetical temperature regimes for each individual star and iteratively refines the temperature-luminosity relationship in a deep learning architecture. We use Gaia DR3 data and find that, on average, temperature's effect on luminosity increases with stellar radius and decreases with absolute magnitude, consistent with theoretical predictions. By framing physics laws as causal problems, our method offers a novel, data-driven approach to refine theoretical understanding and inform evidence-based policy and practice.

Figures (4)

• Figure 1: Architecture of the generative AI in a causal framework. The model takes star features ($X_s$), temperature state ($D_s$), and quantile level ($q$) as inputs. Through a deep neural network, it learns Fourier coefficients ($\beta_k$) that combine with cosine basis functions to estimate potential luminosities under both temperature states ($Y_s(0)$, $Y_s(1)$). The difference between these potential outcomes yields the estimated temperature effect ($\hat{\theta}_s$) for each star. Each hidden layer $h_i$ for $i \in [1, 2]$ uses a ReLU activation function. We use Adam optimization algorithm for updating parameters.
• Figure 2: The distribution of stellar temperature effect on luminosity.
• Figure 3: The distribution of temperature effects on stellar luminosity. Panel (a) shows how temperature effects vary with stellar radius (R/Ro), while panel (b) illustrates the relationship between temperature effects and absolute magnitude (Mv).
• Figure 4: Distribution of the estimated effect of stellar temperature on luminosity using the generalized random forest method.