67639 research outputs found
Sort by
Beyond the Ideal: Analyzing the Inexact Muon Update
No full textThe Muon optimizer has rapidly emerged as a powerful, geometry-aware alternative to AdamW, demonstrating strong performance in large-scale training of neural networks. However, a critical theory-practice disconnect exists: Muon's efficiency relies on fast, approximate orthogonalization, yet all prior theoretical work analyzes an idealized, computationally intractable version assuming exact SVD-based updates. This work moves beyond the ideal by providing the first analysis of the inexact orthogonalized update at Muon's core. We develop our analysis within the general framework of Linear Minimization Oracle (LMO)-based optimization, introducing a realistic additive error model to capture the inexactness of practical approximation schemes. Our analysis yields explicit bounds that quantify performance degradation as a function of the LMO inexactness/error. We reveal a fundamental coupling between this inexactness and the optimal step size and momentum: lower oracle precision requires a smaller step size but larger momentum parameter. These findings elevate the approximation procedure (e.g., the number of Newton-Schulz steps) from an implementation detail to a critical parameter that must be co-tuned with the learning schedule. NanoGPT experiments directly confirm the predicted coupling, with optimal learning rates clearly shifting as approximation precision changes.The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST): i) KAUST Baseline Research Scheme, ii) Center of Excellence for Generative AI, under award number 5940, iii) SDAIA-KAUST Center of Excellence in Artificial Intelligence and Data Science
Degenerate free boundary problems with oscillatory singularities
No full textWe extend the findings of Araújo et al.(On free boundary problems shaped by oscillatory singularities, arXiv:2401.08071) concerning free boundary problems involving varying singularities to the degenerate scenario, establishing fine geometric properties for minimizers under minimal assumptions on the oscillatory exponent.Open access publishing provided by King Abdullah University of Science and Technology (KAUST).JMU is partially supported by the King Abdullah University of Science and Technology (KAUST) and the Centre for Mathematics of the University of Coimbra (funded by the Portuguese Government through FCT/MCTES, DOI 10.54499/UIDB/00324/2020). AS is supported by the King Abdullah University of Science and Technology (KAUST)
Integration of RIS into CloudRT Simulator for Railway Tunnel Scenarios
No full textRadio communication in tunnels presents significant challenges due to the high likelihood of signal obstruction by elements like tunnel walls, ceilings, train carriages, or blocked trains. Reconfigurable Intelligent Surfaces (RIS) offer a promising and cost-effective solution to mitigate signal-blocking issues by transforming the unpredictable radio environment into a controllable one through managed reflections. This paper extends the CloudRT ray-tracing simulator to accommodate RIS technology and assesses its feasibility for railway tunnel scenarios. RIS is integrated into the CloudRT simulator as a phased antenna array acting as a virtual receiver (Rx)/ transmitter (Tx), The study then explores how various key parameters influence received power and compares two scenarios: one where a masking train obstructs the direct Tx-Rx path, and another where there is no obstruction. The results underscore the potential benefits of deploying RIS in obstructed tunnel environments.This work was funded by the council of the Region Bretagne, under the grant MILLIRIS and the French ANR project
mmW4Rail
A Fractional Graph La plus ψ Approach to Image Reconstruction
No full textWe investigate a variational method for ill-posed problems, which embeds the fractional power of the standard graph Laplacian operator in the regularization term. We explore the dependence of the regularizer on a preliminary approximation of the solution, which is obtained using various existing reconstruction methods
from the literature. As a result, the regularization term is both dependent on and adaptive to the observed data, noise, and the choice of the fractional exponent.
We present a selected numerical example problem on 2D computerized tomography, for which we consider various reconstruction techniques
, including Filtered Back Projection, Total Variation, and a trained deep neural network. Incorporating the fractional power of the graph Laplacian operator into the regularization term significantly enhances the quality of the approximated solutions for each method
. Additionally, we show that our proposal behaves as a regularization method and is also stable with respect to variations in the noise level.The first, second, and fourth authors are partially supported by the PRIN project 2022ANC8HL funded by the Italian Ministry of University and Research.
The first and second authors are partially supported by the “INdAM - GNCS Project”, code CUP_E53C24001950001. This research was also supported by the Swiss National Science Foundation SNSF via the projects Stress-Based Methods for Variational Inequalities in Solid Mechanics n. 186407 and ExaSolvers n. 162199
Indian Institute of Tropical Meteorology (IITM) High-Resolution Global Forecast Model version 1: an attempt to resolve monsoon prediction deadlock
No full textAbstract. The prediction of Indian monsoon rainfall variability, affecting a country with a population of billions, remained one of the major challenges of the numerical weather prediction community. While in recent years, there has been a significant improvement in the prediction of the synoptic-scale transients associated with the monsoon circulation, the intricacies of rainfall variability remained a challenge. Here, an attempt is made to develop a global model using a dynamic core of a cubic octahedral grid that provides a higher resolution of 6.5 km over the global tropics. This high-resolution model has been developed to resolve the monsoon convection. Reforecasts with the Indian Institute of Tropical Meteorology (IITM) High-Resolution Global Forecast Model (HGFM) have been run daily from June through September 2022. HGFM has a wavenumber truncation of 1534 in the cubic octahedral grid. The monsoon events have been predicted with a 10 d lead time. HGFM is compared to the operational Global Forecast System (GFS) T1534. While HGFM provides skills comparable to GFS, it shows better skills for higher precipitation thresholds. This model is currently being run in experimental mode and will be made operational.This research has been supported by IITM, Ministry of Earth Sciences, government of India. IITM is fully funded by the Ministry of Earth Sciences, government of India. We would like to thank ECMWF for their support during the model development and for providing the ERA5 dataset. We thank NCMRWF for providing the GFS initial conditions used for conducting simulations. We acknowledge the Pratyush high-performance computing system at IITM, Pune, for providing the computing facility to carry out the simulations. We thank Subhash B. Vaisakh for helping to archive the data on the ARDC server. The authors thank the secretary of the Ministry of Earth Sciences, government of India, and the director of IITM, Pune, for their support and for the facilities provided for this study. We thank IMD for providing the IMD-GPM rainfall and cyclone best-track data
Physics-Informed Waveform Inversion Using Pretrained Wavefield Neural Operators
No full textFull waveform inversion (FWI) is crucial for reconstructing high-resolution subsurface models, but it is often hindered, considering the limited data, by its null space resulting in low-resolution models, and more importantly, by its computational cost, especially if needed for real-time applications. Recent attempts to accelerate FWI using learned wavefield neural operators have shown promise in efficiency and differentiability, but typically suffer from noisy and unstable inversion performance. To address these limitations, we introduce a novel physics-informed FWI framework to enhance the inversion in accuracy while maintaining the efficiency of neural operator-based FWI. Instead of relying only on the L2 norm objective function via automatic differentiation, resulting in noisy model reconstruction, we integrate a physics constraint term in the loss function of FWI, improving the quality of the inverted velocity models. Specifically, starting with an initial model to simulate wavefields and then evaluating the loss over how much the resulting wavefield obeys the physical laws (wave equation) and matches the recorded data, we achieve a reduction in noise and artifacts. Numerical experiments using the OpenFWI and Overthrust models demonstrate our method’s superior performance, offering cleaner and more accurate subsurface velocity than vanilla approaches. Considering the efficiency of the approach compared to FWI, this advancement represents a significant step forward in the practical application of FWI for real-time subsurface monitoring.The authors thank KAUST and the DeepWave Consortium sponsors for their support and the SWAG Group for the collaborative environment. This work utilized the resources of the Supercomputing Laboratory, KAUST, and they are grateful
for that
CCDC 2389642: Experimental Crystal Structure Determination : catena-((mu-Iodo)-(3-bromopyridine)-copper(i))
No full textAn entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures
Efficient atmospheric water generation using mechanical vapor compression: An improved system for sustainable freshwater production
No full textThis study addresses the global freshwater scarcity challenge by introducing an energy-efficient atmospheric water generation system that employs a desiccant-based moisture extraction process coupled with a mechanical vapor compression cycle. A thermodynamic model is developed and evaluated across a range of operating conditions, accounting for key parameters such as desiccant and air mass flow rates, ambient environmental factors, and the thermophysical properties of the desiccant. The proposed approach offers a scalable and environmentally sustainable solution, contributing to the advancement of modern water resource management technologies. The proposed system achieves up to 60 % lower specific energy consumption than conventional humidification dehumidification-based atmospheric water generator systems. Optimal performance occurs at a desiccant-to-air mass flow ratio of 4, with diminishing returns beyond this point. The proposed system operates at 8.78 kWh/m3 with compact heat transfer areas: 2.73 m2 (evaporator), 0.63 m2 (brine preheater), and 0.14 m2 (distillate preheater).The authors acknowledge the support provided by KFUPM through the project DUP241-04
CCDC 2353443: Experimental Crystal Structure Determination : bis(N,N,N-trimethylanilinium) tetrachloro-manganese(ii)
No full textAn entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures
Fast lithium ion diffusion in brownmillerite Li<i>x</i>Sr2Co2O5
No full textTransition metal oxides not only exhibits novel magnetic properties but also provides outstanding ionic transports. Ionic conductors have great potential for interesting tunable physical properties via ionic liquid gating and novel energy storage applications such as all-solid-state lithium batteries. In particular, low migration barriers and high hopping attempt frequency are the keys to achieve fast ion diffusion in solids. Taking advantage of the oxygen-vacancy channel in LixSr2Co2O5, we show that migration barriers of lithium ion are as small as 0.28–0.17 eV depending on the lithium concentration rates. Our first-principles calculation also investigated hopping attempt frequency and concluded the room temperature ionic diffusivity and ion conductivity are high as 10−7–10−6cm2s−1 and 10−3–10−2Scm−1, respectively, which outperform most of perovskite-type, garnet-type, and sulfide Li-ion solid-state electrolytes. This work proves LixSr2Co2O5 as a promising super-ionic conductorWe are grateful for fruitful discussions with Professor Pu Yu. This work was financially supported by the National Natural Science Foundation of China (NNSFC) (No. 12274309). Analysis of magnetic states done at UNH was supported by the U.S. Department of Energy, Director, Office of Science, Office of Basic Energy Sciences, Division of Materials Sciences and Engineering under Contract No. DE-SC0020221. The cl-NEB calculations were performed at Shaheen II in King Abdullah University of Science and Technology (KAUST)