Chronological feed of everything captured from Elad Gil.
Prior constraints on axion couplings from SN 1987A used only the coherent (large-scale) component of the Milky Way magnetic field for axion-to-photon conversion. This paper shows that incorporating small-scale turbulent magnetic field components — from both the Milky Way and the Large Magellanic Cloud — dramatically boosts conversion probability and extends sensitivity to higher axion masses. The result is an improvement of up to two orders of magnitude in constraints on the combined axion-proton and axion-photon coupling product g_{ap} × g_{aγ}. The authors argue this turbulence effect should broadly impact axion searches in other high-conversion environments such as starburst galaxies.
A spectroscopic follow-up of 255 Gaia-selected white dwarfs within 500 pc using GTC/OSIRIS (R~1000) reveals that the automated "massive helium-rich DB" classification is severely contaminated: only 4.46% of such objects are genuine DBs, with the majority being magnetic white dwarfs or warm DQ (carbon-dominated) stars. Random Forest ML classifiers applied to Gaia's low-resolution spectra achieve >90% accuracy for well-represented spectral types, but fail specifically for rare or physically distinct subtypes not well-sampled in training sets. Warm DQs cluster along the Gaia Q branch with anomalously high tangential velocities, consistent with a merger remnant origin.
Gil and Kaminski establish an explicit bijection between rooted bicubic planar maps on 2n vertices and Dyck paths of semilength 3n whose ascents have lengths divisible by 3, with each 3j-ascent colored by one of g_j colors indexing rooted 3-connected bicubic maps on 2j vertices. This bijection provides a combinatorial proof of Tutte's classical decomposition into 3-connected components and offers a constructive assembly method from primitive building blocks. They further show that every 3-connected bicubic planar map on 2n vertices (n ≥ 4) can be generated from a smaller primitive map via two insertion operations adding 4 or 6 vertices, and characterize extremal rootings using prism graphs.
Current next-best-view (NBV) planners optimize surface coverage but assume static objects, while motion-aware active perception methods prioritize tracking over reconstruction completeness — leaving a critical gap for dynamic scene reconstruction. This paper closes that gap with a framework that evaluates candidate viewpoints by their expected observation quality over a distribution of plausible future object states, rather than a single predicted pose. A fixed-lag Gaussian Process smoother estimates and propagates object state uncertainty from noisy planar position measurements, enabling a mobile robot to select views that are both reachable and coverage-maximizing under motion uncertainty. Simulation and real-world results show the approach outperforms both non-predictive NBV and prediction-only tracking baselines in reconstruction completeness.