I’ll scan the key sources for the latest science and tech developments from the past day.AI Solves Multiple Erdős Problems
OpenAI’s unreleased Astra model recently produced solutions or advances on several long-standing Erdős problems, including three additional ones announced August 1, following a May counterexample to the unit-distance problem. These mark some of the first historically significant mathematical proofs generated by AI, with techniques quickly applied to related open questions. The results stand out because they demonstrate AI moving beyond verification into original discovery in pure mathematics.
Paul Erdős posed thousands of problems in number theory, combinatorics and graph theory, many with cash prizes. Mathematician Thomas Bloom’s erdosproblems.com catalog, launched in 2023 and expanded with community comments, created a public testbed. Human-AI collaborations and specialized prompting have accelerated progress, particularly in discrete math domains where large language models show relative strength.
Key uncertainties remain around the depth and novelty of the AI contributions: some early claims were later found to rediscover known results, and human mathematicians rapidly refined the initial AI outputs. It is unclear how far these methods generalize beyond Erdős-style problems to other areas of mathematics or whether they produce fully rigorous, publishable proofs without extensive human oversight.
Sources: Quanta Magazine.
Hidden Acceleration in Quark-Gluon Plasma
New simulations published August 3 reveal that extreme acceleration, previously understudied, builds up along the edges of quark-gluon plasma created in heavy-ion collisions and helps drive its explosive expansion. Peak proper accelerations reach hundreds of MeV, concentrated at the fireball boundary due to sharp pressure drops. This force may act as a thermodynamic control parameter, potentially reshaping the QCD phase diagram.
Quark-gluon plasma is the hottest fluid known, briefly formed when atomic nuclei collide near light speed and recreating conditions microseconds after the Big Bang. Researchers have long examined its vorticity and electromagnetic fields using facilities like RHIC and the LHC. The Fudan University-led team combined transport models (AMPT and UrQMD) with Gaussian smearing to map acceleration across energies from 3.5 GeV to 2.76 TeV.
Uncertainties include whether the acceleration produces measurable experimental signatures, such as altered hyperon spin polarization, and how strongly it couples to chiral or confinement transitions via effects like the Unruh temperature. More realistic hydrodynamic modeling is still needed, and the relative importance of acceleration versus other forces at different collision energies remains to be quantified.
Sources: ScienceDaily, Nuclear Science and Techniques.
Moonquakes as Probes for Buried Lunar Ice
Seismic waves from moonquakes can detect and roughly quantify subsurface water ice because ice stiffens lunar soil, speeding wave travel by two to three times and creating reflective echoes. The method, detailed in a July 31 Science Advances paper and highlighted August 3, offers a way to map ice deposits beyond the reach of orbital instruments ahead of Artemis and Chang’e-7 missions.
Permanently shadowed polar craters may hold ancient ice useful for drinking water, oxygen and rocket fuel, reducing Earth-launched mass for sustained lunar presence. The University of Maryland-led team combined lab experiments on frozen volcanic rock analogs, thermal modeling of cold traps, and simulations of seismic propagation to identify distinct ice signatures.
Open questions include the actual depth, purity and total volume of ice deposits, the practical density of seismometers needed for useful maps, and whether natural moonquake rates or artificial sources will provide sufficient signals. Confirmation awaits real data from upcoming landers carrying seismometers near candidate sites such as Shackleton Crater.
Sources: ScienceDaily, University of Maryland, Science Advances.