AI Network Tests General Relativity by Reconstructing Dark Energy Evolution
A physics-informed neural network analysis of massive cosmic datasets points to evolving dark energy dynamics while remaining consistent with standard gravity.
Cosmologists mapping the expansion of the universe are testing the limits of standard gravitational physics by training an advanced artificial intelligence network on massive astronomical datasets. The approach uses a specialized deep learning tool called a physics-informed neural network to reconstruct how dark energy and matter have evolved over billions of years without relying on preset mathematical formulas.
The study, available on the arXiv preprint server, was conducted by researcher Andronikos Paliathanasis from the Durban University of Technology and several international institutions. By embedding the fundamental laws of physics directly into the training loss function, the network ensures that its mathematical outputs never violate the known laws of gravity across different cosmic epochs.
Probing the Dark Side of the Universe with Neural Networks
Traditional cosmological analyses often assume rigid parametric models to describe how the universe behaves, which can bias the final interpretations. To bypass this restriction, the author expanded a framework known as Cosmo-PINN, dividing the artificial intelligence architecture into three distinct functional modules: a background module, a matter growth module, and a growth index module.
The background module ingests cosmic observations to learn the dark energy density profile, allowing researchers to reconstruct the dark energy equation of state parameter, denoted as $w_{\mathrm{DE}}(z)$, through the mathematical residual of the energy conservation equation. This reconstructed background is subsequently fed into the growth modules to evaluate how matter clumps together under gravity.
What the Training Data Reveals About Cosmic Expansion
The neural network was trained using a robust combination of observational data, including Cosmic Chronometers measuring ancient galaxy expansion rates, Baryon Acoustic Oscillations from the Dark Energy Spectroscopic Instrument Data Release 2, multiple Type Ia supernova compilations, and Redshift-Space Distortion measurements that track the growth of large-scale cosmic structures.
To quantify uncertainties thoroughly, the team applied Bayesian inference via Hamiltonian Monte Carlo sampling starting from the optimal training solution. The resulting posterior median for the present-day dark energy equation of state parameter fell within the range of $w_{\mathrm{DE},0} \in [-0.930, -0.867]$, placing it squarely in the quintessence regime where dark energy dynamics change over time rather than behaving as a static cosmological constant.
Testing General Relativity Through Independent Growth Reconstructions
One of the most notable features of the extended network is its ability to perform a dual reconstruction of the growth index parameter, designated as $\gamma(z)$, which quantifies how matter perturbations evolve and helps scientists test General Relativity on cosmological scales. The first approach derives the growth index directly from the perturbation equations of General Relativity using the background history, while the second approach reconstructs it independently from Redshift-Space Distortion data.
Comparing these two pathways serves as a direct consistency test for standard gravity. The results indicate that the growth index derived directly from General Relativity yields a tightly constrained present-day value of $\gamma_{\mathrm{GR},0} \simeq 0.555$, while direct reconstruction from the observational data gives a slightly higher range of $\gamma_0 \simeq 0.60$ to $0.62$, suggesting a minor suppression in the growth of cosmic structure.
Uncertainties and Future Directions in Cosmological Reconstruction
Although the data shows a slight preference for values deviating from the standard cosmological model limit of $6/11$, the statistical significance remains low, meaning the current observations still align comfortably with General Relativity within acceptable confidence intervals. Furthermore, the growth index becomes difficult to constrain tightly at high redshifts because matter density approaches unity, rendering the growth rate largely independent of $\gamma(z)$.
The author concludes that future astronomical surveys offering higher-precision measurements will significantly tighten these constraints, opening the door to even more rigorous tests of gravity. Subsequent research efforts aim to apply this physics-informed neural network architecture to explore modified theories of gravity and complex interacting dark sector cosmological models.
The study was posted to the arXiv preprint server on November 14, 2026.
This article has been fact checked for accuracy, with information verified against reputable sources. Learn more about us and our editorial process.
Last reviewed on .
Article history
- Latest version
Reference(s)
- Paliathanasis, Andronikos. “Testing General Relativity by Reconstructing Dark Energy and the Growth Index with Physics-Informed Neural Networks.” arXiv.org <http://arxiv.org/abs/2611.19117v1>.
Cite this page:
- Posted by Aisha Ahmed