Archives ouvertes de Paris-Saclay
Not a member yet
272730 research outputs found
Sort by
Modélisation haute résolution des panaches de méthane : validation et expériences de sensibilité pour explorer les approches de quantification des émissions
No full textInternational audienceMethane (C H 4 ) emissions from natural gas, waste, and industrial sources are routinely detected by satellite and aerial platforms; however, quantifying these plumes remains challenging due to their complex structure and rapidly changing fine-scale atmospheric dynamics. This study directly addresses the resulting uncertainties in UAV flight measurements by employing the Fire Dynamics Simulation (FDS) in Large Eddy Simulation (LES) mode, leveraging mean wind data derived from LiDAR datasets. The FDS model was validated with in-situ CH 4 concentration data collected by Uncrewed Aerial Vehicles (UAVs) during controlled release experiments. Our analysis of ten extensively sampled plumes shows that FDS accurately reproduces the magnitude and spatiotemporal variations of the observed plumes for sufficiently large pipes (diameter > 0.6 cm). We found that factors such as gas exit velocity, obstacles, and terrain topography significantly affect the near-field dynamics of the plumes. To a lesser extent, the temperature of the gas influences plume behaviour at higher mass-flow rates. We highlight the added value of high-resolution LES modelling to understand CH 4 plume dynamics captured by various sensors, aiming to improve current emissions quantification methodsLes émissions de méthane (CH4) provenant du gaz naturel, des déchets et des sources industrielles sont couramment détectées par des plateformes satellitaires et aériennes ; cependant, la quantification de ces panaches reste complexe en raison de leur structure multidimensionnelle et de la dynamique atmosphérique fine, qui évolue rapidement. Cette étude traite directement les incertitudes résultant des mesures par vol de drones (UAV) en employant la simulation de dynamique d'incendie (FDS) en mode de simulation des grandes échelles (LES), en exploitant les données de vent moyen dérivées de jeux de données LiDAR. Le modèle FDS a été validé avec des données de concentration de CH4 recueillies in situ par des drones lors d'expériences de rejets contrôlés. Notre analyse de dix panaches largement échantillonnés montre que le modèle FDS reproduit avec précision l'ampleur et les variations spatiotemporelles des panaches observés pour des conduites de diamètre suffisant (diamètre > 0,6 cm). Nous avons constaté que des facteurs tels que la vitesse de sortie du gaz, les obstacles et la topographie du terrain affectent de manière significative la dynamique en champ proche des panaches. Dans une moindre mesure, la température du gaz influence le comportement du panache à des débits massiques plus élevés. Nous soulignons la valeur ajoutée de la modélisation LES à haute résolution pour comprendre la dynamique des panaches de CH4 captée par divers capteurs, dans le but d'améliorer les méthodes actuelles de quantification des émissions
Polymerized tungstate-molybdenum sulfide electrocatalysts for the hydrogen evolution reaction under acidic conditions
No full textInternational audiencePolyoxometalate-based electrocatalysts represent a promising class of earth-abundant electrocatalysts for hydrogen production; however, their large-scale synthesis remains challenging due to the multi-step procedures
“ Theorem” for a Wide Range of
No full textInternational audienceIn this chapter, we will study and prove the so-called “ Theorem” of Patarin (2005). More precisely, we will study the case for a wide range of values that is sufficient for most cryptographic applications. Then, in Chap. 17, we will use this result to prove some very strong security bound on generic Feistel ciphers. We also illustrate the usefulness of the result with the case , which has its own interest from a cryptographic point of view. Indeed, as we will see, it is closely related to the problem of distinguishing where f is a random permutation on n bits from a random function. The proof presented in this chapter follows the recent work by Cogliati et al. (2023), in which a complete and compact proof of the result is provided. We extend this paper by providing additional explanations and clarifications
Multiqubit monogamy relations beyond shadow inequalities
No full textInternational audienceMultipartite quantum systems are subject to monogamy relations that impose fundamental constraints on the distribution of quantum correlations between subsystems. These constraints can be studied quantitatively through sector lengths, defined as the average value of -body correlations, which have applications in quantum information theory and coding theory. In this work, we derive a set of monogamy inequalities that complement the shadow inequalities, enabling a complete characterization of the numerical range of sector lengths for systems with qubits in a pure state. This range forms a convex polytope, facilitating the efficient extremization of key physical quantities, such as the linear entropy of entanglement and the quantum shadow enumerators, by a simple evaluation at the polytope vertices. For larger systems (), we highlight a significant increase in complexity that neither our inequalities nor the shadow inequalities can fully capture
China’s land carbon sinks: from improved estimates to opportunities for better management
No full textInternational audienc
Multi-satellite derived data reveals spatiotemporal dynamics of carbon-water coupling and its drivers in tropical ecosystems
No full textInternational audienc
Low latency global carbon budget indicates reduced land carbon sink in the year 2024
No full textInternational audienceIn 2024, the atmospheric CO2 growth rate based on the globally averaged marine boundary layer (MBL) observations from the National Oceanic and Atmospheric Administration (NOAA) network reached 3.73 ± 0.08 ppm yr−1, marking a record high since continuous measurements began in 1959 (Fig. 1a) [1]. The whole-atmosphere growth rate derived from independent OCO-2 satellite observations for 2024 was 3.20 ± 0.1 ppm yr−1 using the Growth Rates from Satellite Observations data-driven approach (GRESO) from Ref. [2], the highest value of the OCO-2 record since 2015. The whole-atmosphere growth rate derived from our flux inversion models assimilating OCO-2 observations mainly over land for 2024 was 3.23 ± 0.12 ppm yr−1, thus unsurprisingly being almost equal to GRESO and less than the MBL stations but still a record high in the OCO-2 inversions record since 2015. This 38.15% increase in the CO2 growth rate between 2023 and 2024 occurred despite fossil fuel CO2 emissions increasing by only 0.85% [3]. This highlights an unprecedented weakening of the net carbon uptake on land and ocean. Net land uptake is defined here as the sum of non-fossil land CO2 fluxes including photosynthesis, respirations, fire, rivers and land-use change emissions. Here, we present a low latency global and regional carbon budget for 2024, using top-down inversions and bottom-up models, revealing a strong weakening of the global net land carbon sink and widespread transitions of terrestrial regions from carbon sinks to sources
COVID-19 containment and control reduced lake turbidity around the world
No full textInternational audienc
Statistical Advantage of Softmax Attention: Insights from Single-Location Regression
No full textInternational audienceLarge language models rely on attention mechanisms with a softmax activation. Yet the dominance of softmax over alternatives (e.g., component-wise or linear) remains poorly understood, and many theoretical works have focused on the easier-to-analyze linearized attention. In this work, we address this gap through a principled study of the single-location regression task, where the output depends on a linear transformation of a single input token at a random location. Building on ideas from statistical physics, we develop an analysis of attention-based predictors in the high-dimensional limit, where generalization performance is captured by a small set of order parameters. At the population level, we show that softmax achieves the Bayes risk, whereas linear attention fundamentally falls short. We then examine other activation functions to identify which properties are necessary for optimal performance. Finally, we analyze the finite-sample regime: we provide an asymptotic characterization of the test error and show that, while softmax is no longer Bayes-optimal, it consistently outperforms linear attention. We discuss the connection with optimization by gradient-based algorithms