Science
Physics and the wider sciences — research, discoveries and the ideas behind them.
It's the end of the world as we know it, per a physicist's 1960 prediction
A 1960 paper by physicist Heinz von Foerster predicted that humanity would reach a breaking point due to overpopulation by November 13, 2026. While his model accurately measured past population growth, it failed to account for changing trends and the eventual decline in birth rates. Despite its flaws, the paper's projections were surprisingly accurate for several decades before diverging from reality.
Physics2Weight: Direct Neural Network Architecture and Weight Generation from Physical Laws Using Large Language Models
Researchers have developed Physics2Weight (PtW), a framework that uses large language models to directly generate neural network weights and edit architectures for solving partial differential equations. PtW employs physics-constrained prompting and architecture editing, along with iterative weight refinement, to improve solution accuracy and physical consistency without gradient-based optimization. Experiments show PtW achieves lower solution errors than conventional physics-informed neural networks for several PDEs, though it has limitations with nonlinear equations.
What happens to mathematics when AI becomes an oracle?
A new book, "Big Math: The Hidden Codes That Run Our World," by applied mathematicians Steven Strogatz and Alex Townsend, explores the evolving relationship between mathematics and advanced computing. The authors discuss how AI, large language models, and quantum computing are pushing the boundaries of "big math" beyond human capacity. They argue that externalizing mathematical processes into tools is a long-standing human practice, now reaching new heights with modern technology.
Modulation Induced Memory Resonances in Dynamically Driven Optical Cavities
This research explores periodically driven optical cavities using a time-dependent cavity-field model, demonstrating modulation-induced tuning of cavity resonances and analyzing intracavity field evolution. The study reveals hysteretic behavior and memory effects due to the interplay between periodic modulation and cavity dissipation, identifying distinct memory resonances. The findings offer a framework for engineering memory effects in optical resonators and provide opportunities for tunable photonic environments and dynamic control of light-matter interactions.
Just CaloCHILL: fast hadronic shower generation in a learned latent space
CaloCHILL, a new generative model, significantly accelerates the simulation of hadronic calorimeter showers by generating a compact summary in a learned latent space rather than individual energy deposits. This three-stage model, which includes flow-matching and a variational autoencoder, reproduces main shower observables and their time structure. CaloCHILL is 600 times faster than its predecessor while also improving accuracy.
How evolution can make cells smaller without slowing down their growth
A new study by Marco Fumasoni and colleagues demonstrates that through experimental evolution in yeast over 1,500 generations, cells can become significantly smaller without impairing their growth rate, revealing genetic changes in pathways related to growth and the cell cycle. This work challenges previous assumptions about the constraints on reducing cell size due to biological costs.
Dark matter cannot be ruled out as cause of gamma ray glow at the Milky Way's center, machine learning shows
Researchers from the University of Vienna and Lawrence Berkeley National Laboratory used machine learning to reassess the Galactic Center Excess (GCE), a gamma-ray glow at the Milky Way's center, which has been debated for over a decade. Their analysis, published in Physical Review Letters, suggests that dark matter could still be a viable explanation due to the faintness of potential millisecond pulsar sources, indicating more research is needed before ruling out dark matter as a cause.