Laboratoire Jean Kuntzmann

Glacio-seismology, which investigates the dynamics and processes of the cryosphere using seismic observations and methods, has strongly grown in interest over the past two decades in the context of global warming. To study the Argentière Glacier (French Alps), a dense array of 98 3-component seismic sensors was deployed in spring 2018 for 35 days. This period coincided with a temperature increase, which enhances the glacier’s seismic activity. The recordings bear the imprint of several thousand icequakes associated with ice-fracturing phenomena such as crevassing. We build a catalog of icequakes and their location using Matched Field Processing (MFP), which is a beamforming based approach.Then, we jointly reconstruct icequake source mechanisms and the 3D glacier structure by exploiting the full waveform of the recorded 3-component data. The reconstruction relies on elastic wave modelling through numerical solution of the 3D elastodynamic equations using the Spectral Element Method (SEM). Accounting for the glacier surface topography is essential in order to correctly model surface waves, which mainly dominate the icequake data. We apply an alternating optimisation strategy that iterates between two sub-problems: estimating source mechanisms and updating glacier model parameters. The estimation of an icequake mechanism is formulated as the solution of a bi-quadratic minimisation problem depending on the moment tensor and the source wavelet. The model-parameter update is based on the application of a classical elastic Full Waveform Inversion (FWI).This joint inversion strategy is applied to a set of selected icequakes with a high signal-to-noise ratio. We are able to reconstruct average model heterogeneities that align with the orientation of surface crevasses in several areas of the glacier. Some heterogeneities are reconstructed down to 100 m below the surface, enabling us to estimate the depth of crevasse fields surrounding the sensor network. Finally, we note a clear improvement in the reconstruction of the SH-wave in the updated model compared to what is obtained in a homogeneous medium. In the homogeneous approximation, the Rayleigh wave is reconstructed accurately, whereas the SH-wave is less well recovered. This improvement suggests that the SH-wave is strongly impacted by surface heterogeneities, more than the Rayleigh wave, and mainly drives the reconstruction of the structure. The estimated icequake source mechanisms do not appear to change significantly between the homogeneous model and the updated model obtained during the alternating strategy. This suggests a relative decoupling between source parameters and model parameters in the joint reconstruction problem, mediated by the Rayleigh and SH- waves. Such a decoupling is generally not observed in classical seismology, and therefore seems to be rather specific to the glacial context.

Lipid transfer proteins (LTPs) maintain the specialized lipid compositions of organellar membranes1,2. In humans, many LTPs are implicated in diseases3, but the cargo and auxiliary lipids that facilitate the transfer of the majority of LTPs remain unknown. Here we combined biochemical, lipidomic and computational methods to systematically characterize LTP-lipid complexes4 and measure how LTP gains of function affect cellular lipidomes. We identified bound lipids for around half of the hundreds of LTPs that we analysed, confirming known ligands and identifying new ones across most LTP families. Gains in LTP function affected the cellular abundance of both their known and newly identified lipid ligands, indicating comparable functional relevance of the two ligand sets. Using structural bioinformatics, we characterized mechanisms that contribute to lipid selectivity and identified preferences based on headgroup or acyl chain. We demonstrate some basic principles of how LTPs mobilize their ligands. They commonly interact with several classes of lipids and exhibit broad but selective preference for particular headgroups and for lipid species with shorter acyl chains that contain one or two unsaturated carbons, suggesting that only subsets of lipid species are efficiently mobilized. The datasets represent a resource for further analysis in different cell types and states, such as those associated with pathologies.

Abstract During an earthquake, slip occurs in a localized shear zone that features a heavily granulated fault core that can be characterized as a shear band. We study the formation of this fault core in a granular rock such as sandstone by developing a model of crushable granular media within the framework of Breakage Mechanics. This model accounts for the evolution of the grain size distribution, while also accounting for the co‐evolution of the solid fraction. An enrichment with the Cosserat continuum allows for the model to predict finite‐width shear bands. The model is then calibrated against experimental data taken from tests on Bentheim sandstone, and a parametric study of the mechanical parameters is conducted using linear stability analysis. We find that for deeply‐buried rocks the shear bands have a compactive component, and the initial value of the solid fraction does not play a strong role in the initial band thickness, but can influence the rate of delocalization of the band. Post‐localization behavior is studied with the finite element method, which shows the formation of zones of dilation outside the band in addition to the compaction within the band. Using a modified Kozeny–Carman permeability law, it is shown that within the band the permeability reduces by several orders of magnitude, but can increase outside the band. Our results highlight the importance of modeling grain size and solid fraction evolution as they exert a controlling influence on hydromechanical properties that play an important role in fault formation and seismic slip.

ABSTRACT This paper is focused on the statistical analysis of data consisting of a collection of multiple series of probability measures that are indexed by distinct time instants and supported over a bounded interval of the real line. By modeling these time‐dependent probability measures as random objects in the Wasserstein space, we propose a new auto‐regressive model for the statistical analysis of multivariate distributional time series. Using the theory of iterated random function systems, results on the second‐order stationarity of the solution of such a model are provided. We also propose a consistent estimator for the autoregressive coefficients of this model. Due to the simplex constraints that we impose on the model coefficients, the proposed estimator that is learned under these constraints naturally has a sparse structure. The sparsity allows the application of the proposed model in learning a graph of temporal dependency from multivariate distributional time series. We explore the numerical performances of our estimation procedure using simulated data. To shed some light on the benefits of our approach for real data analysis, we also apply this methodology to two data sets, respectively made of observations from age distribution in different countries and those from the bike sharing network in Paris.

Abstract Diapycnal mixing in the ocean interior has diverse numerical representations in current global ocean models. These representations affect the simulated transport and storage of oceanic tracers in ways that remain little studied. Here we present the impacts of three different tidal mixing representations in thousand‐year‐long simulations with the NEMO global ocean model at one‐degree resolution. The first model experiment includes local bottom‐intensified mixing at internal tide generation sites and a constant background diffusivity. The second explicitly includes both local and remote tidal mixing, with no background diffusivity. The third experiment is identical to the second but has the added contribution of bottom‐trapped (subinertial) internal tides, known to be important in polar regions. The three simulations show broadly similar circulation and stratification but important regional differences. Explicit representation of remote tidal mixing strengthens the Atlantic Meridional Overturning Circulation by up to 1.5 × 10 6 m 3 s −1 . Inclusion of bottom‐trapped internal tides reduces heat reaching Antarctica by eroding Circumpolar Deep Water at southern high latitudes, and reduces the mean age of the global deep (>2 km) ocean by 10%. The results call for more observational constraints on polar ocean mixing, and point to multi‐faceted climatic repercussions of tidal mixing representations.

Abstract We consider $U$-statistics on row-column exchangeable matrices, which are arrays invariant to separate permutations of rows and columns and are common in bipartite data. Under the standard dissociation assumption, we develop a graph-indexed analogue of the Hoeffding decomposition tailored to row-column exchangeable dependence. We present a new decomposition based on orthogonal projections onto probability spaces generated by sets of Aldous-Hoover-Kallenberg variables. These sets are indexed by bipartite graphs, enabling the application of graph-theoretic concepts to describe the decomposition. This framework provides new insights into the characterization of $U$-statistics on row-column exchangeable matrices, particularly regarding their asymptotic behaviour, including in degenerate cases. Notably, the limit distribution depends only on specific terms in the decomposition, corresponding to non-zero components indexed by the smallest graphs, namely the principal support graphs. We show that the asymptotic behaviour of a $U$-statistic is characterized by the properties of its principal support graphs. The number of nodes in these graphs (the principal degree) dictates the convergence rate to the limit distribution, with degeneracy occurring if and only if this number is strictly greater than 1. Furthermore, when the principal support graphs are connected, the limit distribution is Gaussian, even in degenerate cases. Applications to network analysis illustrate these findings.

A few years after AlphaFold revolutionised the field of protein structure prediction, the new frontiers and limitations in structural biology have become clearer. Predicting protein-nucleic acid interactions currently stands as one of the major unresolved challenges in the field. This knowledge gap stems from the scarcity and limited diversity of experimental data, as well as the unique geometric, physicochemical, and evolutionary properties of nucleic acids. Despite these challenges, innovative ideas and promising methodological developments have emerged for both predicting protein-nucleic acid complex structures and designing nucleic acids capable of binding to specific protein conformations. This review presents these recent advances and discusses promising avenues, including the integration of high-throughput profiling data, the development of more rigourous and richer evaluation benchmarks, and the discovery of biologically meaningful regulatory and structural signals using self-supervised learning.

Dealing with extreme values is a key challenge in probabilistic modeling, relevant to economics, engineering, and life sciences. Standard GANs, built on light-tailed noise, fail to capture heavy-tail behavior and dependence in extreme regions. We propose HTGAN, a modified GAN with heavy-tailed input noise. Using the stable tail dependence function (stdf) from extreme-value theory, we derive a worst-case error bound for approximating the stdf of the target, scaling as N − 1 / ( d − 1 ) , where N is the latent noise dimension and the data dimension. Thus, higher latent dimension improves dependence estimation. Experiments on synthetic and real heavy-tailed datasets confirm that HTGAN better reproduces extreme dependence than classical light-tailed GANs, with accuracy improving as latent dimension increases.