A core challenge for both physics and artificial intelligence (AI) is symbolic regression: finding a symbolic expression that matches data from an unknown function. Although this problem is likely to be NP-hard in principle, functions of practical interest often exhibit symmetries, separability, compositionality, and other simplifying properties. In this spirit, we develop a recursive multidimensional symbolic regression algorithm that combines neural network fitting with a suite of physics-inspired techniques. We apply it to 100 equations from the Feynman Lectures on Physics, and it discovers all of them, while previous publicly available software cracks only 71; for a more difficult physics-based test set, we improve the state-of-the-art success rate from 15 to 90%.

A major challenge of AI plus science lies in its inherent incompatibility: Today’s AI is primarily based on connectionism, while science depends on symbolism. To bridge the two worlds, we propose a framework to seamlessly synergize Kolmogorov-Arnold networks (KANs) and science. The framework highlights KANs’ usage for three aspects of scientific discovery: identifying relevant features, revealing modular structures, and discovering symbolic formulas. The synergy is bidirectional: science to KAN (incorporating scientific knowledge into KANs), and KAN to science (extracting scientific insights from KANs). We highlight major new functionalities in : (1) MultKAN, KANs with multiplication nodes, (2) kanpiler, a KAN compiler that compiles symbolic formulas into KANs; (3) tree converter, convert KANs (or any neural networks) into tree graphs. Based on these tools, we demonstrate KANs’ capability to discover various types of physical laws, including conserved quantities, Lagrangians, symmetries, and constitutive laws.

There is growing interest in using 3-dimensional neutral hydrogen mapping with the redshifted 21 cm line as a cosmological probe. However, its utility depends on many assumptions. To aid experimental planning and design, we quantify how the precision with which cosmological parameters can be measured depends on a broad range of assumptions, focusing on the 21 cm signal from $6<z<20$. We cover assumptions related to modeling of the ionization power spectrum, to the experimental specifications like array layout and detector noise, to uncertainties in the reionization history, and to the level of contamination from astrophysical foregrounds. We derive simple analytic estimates for how various assumptions affect an experiment’s sensitivity, and we find that the modeling of reionization is the most important, followed by the array layout. We present an accurate yet robust method for measuring cosmological parameters that exploits the fact that the ionization power spectra are rather smooth functions that can be accurately fit by 7 phenomenological parameters. We find that for future experiments, marginalizing over these nuisance parameters may provide constraints almost as tight on the cosmology as if 21 cm tomography measured the matter power spectrum directly. A future square kilometer array optimized for 21 cm tomography could improve the sensitivity to spatial curvature and neutrino masses by up to 2 orders of magnitude, to $\Delta {\Omega }_k\approx 0.0002$ and $\Delta m_{\nu }\approx 0.007\mathrm{eV}$, and give a $4\sigma$ detection of the spectral index running predicted by the simplest inflation models.

The Murchison Widefield Array (MWA) has collected hundreds of hours of Epoch of Reionization (EoR) data and now faces the challenge of overcoming foreground and systematic contamination to reduce the data to a cosmological measurement. We introduce several novel analysis techniques, such as cable reflection calibration, hyper-resolution gridding kernels, diffuse foreground model subtraction, and quality control methods. Each change to the analysis pipeline is tested against a two-dimensional power spectrum figure of merit to demonstrate improvement. We incorporate the new techniques into a deep integration of 32 hours of MWA data. This data set is used to place a systematic-limited upper limit on the cosmological power spectrum of mK² at k = 0.27 h Mpc⁻¹ and z = 7.1, consistent with other published limits, and a modest improvement (factor of 1.4) over previous MWA results. From this deep analysis, we have identified a list of improvements to be made to our EoR data analysis strategies. These improvements will be implemented in the future and detailed in upcoming publications.