Chronological feed of everything captured from Justine Moore.
The MicroBooNE collaboration has performed the first-ever measurement of energy resolution in a liquid argon time projection chamber (LArTPC) at the MeV scale, achieving (7.52 ± 0.78 stat ± 0.92 syst)% resolution at ~1.5 MeV. The calibration leverages monoenergetic 2.614 MeV gamma-rays from ²⁰⁸Tl decays undergoing pair production within the detector — a novel in-situ approach that avoids external radioactive sources. The measured resolution is consistent with simulation at the 1.6σ level and establishes a viable monoenergetic calibration pathway for next-generation LArTPC experiments targeting MeV-scale neutrino physics.
SN 2023xgo, a Type Ibn/Icn supernova observed out to +377 days post-explosion, provides one of the most detailed IR characterizations of dust formation in this rare SN subclass. JWST (NIRSpec + MIRI) reveals a large dust mass (≥3×10⁻² M☉ in cool silicate dust) co-located near the shock radius, alongside evidence of ongoing circumstellar interaction at late times. Rapid dust formation — detectable as early as ~70–100 days — and progressive optical/NIR line profile evolution point to dust condensing interior to the unshocked CSM, confirming that the He/C-rich, H-poor circumstellar environments of SNe Ibn/Icn are unusually efficient dust factories both before and after explosion.
AdaWeather is an adaptive ensemble framework that combines multiple probabilistic weather forecasts using a hybrid of machine learning and mixture-of-experts methods. Its key theoretical contribution is extending the classical "best single expert in hindsight" regret bound to logarithmic regret against the best static mixture of experts — a strictly stronger guarantee. No single forecasting model dominates across all spatial and temporal contexts, making adaptive combination both practically motivated and theoretically well-founded. Empirical validation on temperature forecasting shows improvements over existing combination methods.
Microsoft's topological qubit team demonstrates that substituting aluminum with lead (a higher-gap superconductor) in InAs-based tetron devices dramatically improves topological phase robustness, yielding parity lifetimes of ~20 seconds — and in some cases minute-scale — compared to typical qubit gate operation times of ~microseconds. The result is a direct experimental validation of the theoretical prediction that increasing the excitation gap suppresses quasiparticle poisoning. Interferometric single-shot parity measurements, enabled by a new RF technique with μeV energy resolution, confirm h/2e-periodic bimodal quantum capacitance shifts consistent with topological parity switching. These lifetimes have direct implications for Pauli measurement fidelity in topological quantum computing architectures.
Dynamic Modulated Arc Therapy (DMAT) reframes VMAT planning by treating delivery time as a co-optimized parameter rather than an emergent byproduct, coupling machine-aware emulation (axis synchronization, finite acceleration) with dosimetric and modulation objectives. A user-controlled modulation level (−3 to +3) steers leaf travel, MU behavior, and control point (CP) density, with adaptive CP redistribution concentrating resolution in high-complexity arc sectors. Evaluation on H&N, lung SBRT, and prostate SBRT cases demonstrates a transparent quality-vs-efficiency trade-off, with H&N cases showing the largest gains from increased modulation and SBRT cases showing diminishing returns. The framework is positioned as directly applicable to time-constrained workflows such as adaptive and motion-sensitive radiotherapy.
PoisonCap introduces a new "poison" capability format for CHERI architectures that upgrades temporal memory safety from use-after-reallocation to strict use-after-free protection, while also adding initialisation safety — two gaps left by the current state-of-the-art, Cornucopia Reloaded. The design replaces Cornucopia's shadow bitmap, communicates memory quarantine state to the microarchitecture for cache management, and delegates poisoning privilege via capability bounds to support nested allocators. Critically, PoisonCap imposes no fundamental overhead over a Cornucopia baseline that already zeros memory before reallocation, making it a strict improvement in both security guarantees and architectural elegance.
MERSEM is a multi-objective scheduling framework that combines evolutionary search with reinforcement learning to optimize graph analytics workloads across heterogeneous edge-cloud environments. The evolutionary component handles broad global exploration of scheduling solutions, while the RL agent performs adaptive local refinement — a division of labor that addresses both exploration breadth and exploitation precision. The framework jointly targets SLA compliance and carbon footprint reduction by incorporating dynamic carbon intensity signals and resource heterogeneity into scheduling decisions. Benchmarked against state-of-the-art baselines, MERSEM achieves up to 45% fewer SLA violations and up to 12% lower carbon emissions.
LLM inference — not training — dominates the energy lifecycle of large models, accounting for up to 90% of total energy use, making inference optimization a critical sustainability lever. MARLIN is a multi-agent game-theoretic reinforcement learning framework that jointly optimizes four competing objectives: latency (TTFT), carbon emissions, water consumption, and energy cost in cloud datacenter deployments. Against state-of-the-art inference management baselines, MARLIN achieves simultaneous reductions across all four metrics, demonstrating that Pareto-style multi-objective optimization via game theory is viable at inference serving scale.
Researchers have demonstrated the first Z₃ discrete time crystal (DTC) using a chain of 15 superconducting qutrits implementing a Floquet chiral clock model. Unlike Z₂ (qubit-based) DTCs, stabilizing higher-symmetry time-crystalline order requires managing complex domain-wall physics, a challenge solved here by introducing a tunable chiral angle that controls domain-wall dynamics and spectral degeneracies. The system exhibits robust period-tripling subharmonic response across a wide range of drive parameters and independent of initial state — a property that critically depends on the presence of chirality. This work positions native qudit hardware as a viable platform for exploring the broader landscape of non-equilibrium quantum phases beyond the standard Ising paradigm.