Data assimilation method for fractured reservoirs using mimetic finite differences and ensemble Kalman filter
Not provided.
SEARCH · Search NASA
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Not provided.
Abstract not provided
Explore the source record for details and available documents.
Explore the source record for details and available documents.
Not provided.
Model ensembles are effective tools for estimating prediction uncertainty in deep learning atomistic force fields. However, their widespread adoption is hindered by high computational costs and overconfident error estimates. In this work, we address these challenges by leveraging distributions of per-sample errors obtained during training and employing a distance-based similarity search in the model latent space. Our method, which we call LTAU (Loss Trajectory Analysis for Uncertainty), efficiently estimates the full probability distribution function of errors for any test point using the logged training errors, achieving speeds that are 2–3 orders of magnitudes faster than typical ensemble methods and allowing it to be used for tasks where training or evaluating multiple models would be infeasible. We apply LTAU towards estimating parametric uncertainty in atomistic force fields (LTAU-FF), demonstrating that it produces well-calibrated confidence intervals and predicts errors that correlate strongly with the true errors for data near the training domain. Furthermore, we show that the errors predicted by LTAU-FF can be used in practical applications for detecting out-of-domain data, tuning model performance, and predicting failure during simulations. We believe that LTAU will be a valuable tool for uncertainty quantification in atomistic force fields and is a promising method that should be further explored in other domains of machine learning.
When parameter estimation is solved in a high-dimensional space, the dimensionality reduction strategy becomes the primary consideration for alleviating the tremendous computational cost. Here, the discrete empirical interpolation method (DEIM) is explored to retrieve the initial condition (IC) by combining the polynomial chaos (PC) based ensemble Kalman filter (i.e. PC-EnKF), where a non-intrusive PC expansion is considered as a surrogate model in place of the forward model in the prediction step of the ensemble Kalman filter, resulting in fewer forward model integrations but with a comparable accuracy as Monte Carlo-based approaches. The DEIM acts as a hyper-reduction tool to provide the low-dimensional input for the high-dimensional initial field, which can be reconstructed using the information on the sparse interpolation grid points that is adaptively obtained through PC-EnKF data assimilation method. Thus an innovative framework to reconstruct the IC is developed. The detailed procedure at each assimilation iteration includes: the determination of the spatial interpolation points, the estimation of the initial values on the interpolation locations using the optimal observations, and the reconstruction of IC in the full space. The current study uses the reconstruction field of initial conditions of the Navier-Stokes equations as an example to illustrate the efficacy of our method. The experimental results demonstrate the proposed algorithm achieves a satisfactory reconstruction for the initial field. The proposed method helps to extend the applicable area of DEIM in solving inverse problems.
We assess the heterogeneous electrochemistry and mechanics in a composite electrode of commercial batteries using synchrotron X-ray tomography analysis and microstructure-resolved computational modeling. We visualize the morphological defects at multi-scales ranging from the macroscopic composite, particle ensembles, to individual single particles. Particle fracture and interfacial debonding are identified in a large set of tomographic data of active particles. Mechanical failure in the regime near the separator is more severe than toward the current collector. The active particles close to the separator experience deeper charge and discharge over cycles and thus are more mechanically loaded. The difference in the Li activity originates from the polarization of the electrolyte potential and the non-uniform distribution of the activation energy for the charge transfer reaction. We model the kinetics of intergranular fracture and interfacial degradation to confirm that the various Li activities are the major cause of the heterogeneous damage. The interfacial failure may reconstruct the conductive network and redistribute the electrochemical activities that render a dynamic nature of electrochemistry and mechanics evolving over time in the composite electrode. We further quantify the influence of the mechanical damage on the metrics of battery performance. We simulate the electrochemical impedance profile to build a relationship between the interfacial debonding and the impedance of electron transport and surface charge transfer. Lastly, the mechanical failure disrupts the conduction path of electrons and results in significant polarization and capacity loss in batteries.
Phonon dispersion and phonon density of states of graphene are calculated using classical molecular dynamics simulations. In this method, the dynamical matrix is constructed based on linear response theory by computing the displacement of atoms during the simulations. The computed phonon dispersions show excellent agreement with experiments. The simulations are done in both NVT and NPT ensembles at 300 K and found that the LO/TO modes are getting hardened at the Γ point. The NPT ensemble simulations capture the anharmonicity of the crystal accurately and the hardening of LO/TO modes is more pronounced. We also found that at 300 K the C-C bond length reduces below the equilibrium value and the ZA bending mode frequency becomes imaginary close to Γ along K-Γ direction, which indicates instability of the flat 2D graphene sheets.
With the coming data deluge from synoptic surveys, there is a need for frameworks that can quickly and automatically produce calibrated classification probabilities for newly observed variables based on small numbers of time-series measurements. In this paper, we introduce a methodology for variable-star classification, drawing from modern machine-learning techniques. We describe how to homogenize the information gleaned from light curves by selection and computation of real-numbered metrics (features), detail methods to robustly estimate periodic features, introduce tree-ensemble methods for accurate variable-star classification, and show how to rigorously evaluate a classifier using cross validation. On a 25-class data set of 1542 well-studied variable stars, we achieve a 22.8% error rate using the random forest (RF) classifier; this represents a 24% improvement over the best previous classifier on these data. This methodology is effective for identifying samples of specific science classes: for pulsational variables used in Milky Way tomography we obtain a discovery efficiency of 98.2% and for eclipsing systems we find an efficiency of 99.1%, both at 95% purity. The RF classifier is superior to other methods in terms of accuracy, speed, and relative immunity to irrelevant features; the RF can also be used to estimate the importance of each feature in classification. Additionally, we present the first astronomical use of hierarchical classification methods to incorporate a known class taxonomy in the classifier, which reduces the catastrophic error rate from 8% to 7.8%. Excluding low-amplitude sources, the overall error rate improves to 14%, with a catastrophic error rate of 3.5%.
Non-coding ribonucleic acids (ncRNA) are functional RNA molecules that are not translated into protein. They are extremely dynamic, adopting diverse conformational substates, which enables them to modulate their interaction with a large number of other molecules. The flexibility of ncRNA provides a challenge for probing their complex 3D conformational landscape, both experimentally and computationally. As a result, despite their conformational diversity, ncRNAs mostly preserve their secondary structure throughout the dynamic ensemble. Here we present a kinematics-based procedure to morph an RNA molecule between conformational substates, while avoiding inter-atomic clashes. We represent an RNA as a kinematic linkage, with fixed groups of atoms as rigid bodies and rotatable bonds as degrees of freedom. Our procedure maintains RNA secondary structure by treating hydrogen bonds between base pairs as constraints. The constraints define a lower-dimensional, secondary-structure constraint manifold in conformation space, where motions are largely governed by molecular junctions of unpaired nucleotides. On a large benchmark set, we show that our morphing procedure compares favorably to peer algorithms, and can approach goal conformations to within a low all-atom RMSD by directing fewer than 1% of its atoms. Furthermore, our results suggest that molecular junctions can modulate 3D structural rearrangements, while secondary structure elements guide large parts of the molecule along the transition to the correct final conformation.
We explore, from a theoretical perspective, photoelectron angular distributions (PADs) of the Na clusters Na{sub 8}, Na{sub 10}, Na{sub 12}, Na{sub 18}, Na{sub 3}{sup +}, Na{sub 11}{sup +}, Na{sub 13}{sup +}, and Na{sub 19}{sup +}. The basis of the description is the time-dependent local-density approximation (TDLDA), augmented by a self-interaction correction (SIC) to describe ionization properties correctly. The scheme is solved on a numerical grid in coordinate space with absorbing bounds. We assume for each cluster system an isotropic ensemble of free clusters and develop for the case of one-photon emission analytical formulas for computing the orientation-averaged PAD on the basis of a few TDLDA-SIC calculations for properly chosen reference orientations. It turns out that all the information in the averaged PAD is contained in one anisotropy parameter. We find that this parameter varies very little with system size, but as a whole is crucially influenced by the detailed ionic structure. We also make comparisons with direct orientation averaging and consider one example reaching outside the perturbative regime.
Recently, we made significant advancements in improving the computational efficiency of lattice QCD calculations for generalized parton distributions (GPDs). This progress was achieved by adopting calculations of matrix elements in asymmetric frames, deviating from the computationally-expensive symmetric frame typically used, and allowing freedom in the choice for the distribution of the momentum transfer between the initial and final states. A crucial aspect of this approach involves the adoption of a Lorentz covariant parametrization for the matrix elements, introducing Lorentz-invariant amplitudes. This approach also allows us to propose an alternative definition of quasi-GPDs, ensuring frame independence and potentially reduce power corrections in matching to light cone GPDs. In our previous work, we presented lattice QCD results for twist-2 unpolarized GPDs ( H and E ) of quarks obtained from calculations performed in asymmetric frames at zero skewness. Building upon this work, we now introduce a novel Lorentz covariant parametrization for the axial-vector matrix elements. We employ this parametrization to compute the axial-vector GPD H ˜ at zero skewness, using an N f = 2 + 1 + 1 ensemble of twisted mass fermions with clover improvement. The light-quark masses employed in our calculations correspond to a pion mass of approximately 260 MeV. Published by the American Physical Society 2024
In extreme astrophysical environments such as core-collapse supernovae and binary neutron star mergers, neutrinos play a major role in driving various dynamical and microphysical phenomena, such as baryonic matter outflows, the synthesis of heavy elements, and the supernova explosion mechanism itself. The interactions of neutrinos with matter in these environments are flavor-specific, which makes it of paramount importance to understand the flavor evolution of neutrinos. Flavor evolution in these environments can be a highly nontrivial problem thanks to a multitude of collective effects in flavor space, arising due to neutrino-neutrino ($v$-$v$) interactions in regions with high neutrino densities. A neutrino ensemble undergoing flavor oscillations under the influence of significant $v$-$v$ interactions is somewhat analogous to a system of coupled spins with long-range interactions among themselves and with an external field (‘long-range’ in momentum-space in the case of neutrinos). As a result, it becomes pertinent to consider whether these interactions can give rise to significant quantum correlations among the interacting neutrinos, and whether these correlations have any consequences for the flavor evolution of the ensemble. In particular, one may seek to utilize concepts and tools from quantum information science and quantum computing to deepen our understanding of these phenomena. In this article, we attempt to summarize recent work in this field. In conclusion, we also present some new results in a three-flavor setting, considering complex initial states.
Abstract Climate change research uses an ensemble of general circulation model runs (GCMs-runs) to predict future climate under uncertainties. To reduce computational costs, this study selects representative GCM-runs (RGCM-runs) for Western North America (WNA) based on their performance in replicating historical climate conditions from 1981 to 2005 and projecting future changes from 1981–2010 to 2071–2100. This evaluation is conducted under two representative concentration pathways (RCPs) scenarios, RCP4.5 and RCP8.5, from the Coupled Model Intercomparison Project 5. By using an envelope-based selection technique and a multi-objective distance-based approach, we identify four RGCM-runs per RCP representing diverse climatic conditions, including wet-warm, wet-cold, dry-warm, and dry-cold. Compared to the full-set, these selected runs show a decreased mean absolute error (MAE) between the reference and RGCM-runs concerning the monthly average mean air temperature (T̄) and precipitation (P̄). For RCP4.5, T̄ MAE is 0.45 (vs. 0.58 in the full-set) and P̄ MAE is 0.31 (vs. 0.42). For RCP8.5, T̄ MAE is 0.51 (vs. 0.75) and P̄ MAE is 0.25 (vs. 0.36). The lower MAE values in the RGCM-run set indicate closer alignment between predicted and reference values, making the RGCM-run suitable for climate impact assessments in the region.
A challenge common to many constituent data assimilation applications is the fact that one observes a much smaller fraction of the phase space that one wishes to estimate. For example, remotely sensed estimates of the column average concentrations are available, while one is faced with the problem of estimating 3D concentrations for initializing a prognostic model. This problem is exacerbated in the case of aerosols because the observable Aerosol Optical Depth (AOD) is not only a column integrated quantity, but it also sums over a large number of species (dust, sea-salt, carbonaceous and sulfate aerosols. An aerosol transport model when driven by high-resolution, state-of-the-art analysis of meteorological fields and realistic emissions can produce skillful forecasts even when no aerosol data is assimilated. The main task of aerosol data assimilation is to address the bias arising from inaccurate emissions, and Lagrangian misplacement of plumes induced by errors in the driving meteorological fields. As long as one decouples the meteorological and aerosol assimilation as we do here, the classic baroclinic growth of error is no longer the main order of business. We will describe an aerosol data assimilation scheme in which the analysis update step is conducted in observation space, using an adaptive maximum-likelihood scheme for estimating background errors in AOD space. This scheme includes e explicit sequential bias estimation as in Dee and da Silva. Unlikely existing aerosol data assimilation schemes we do not obtain analysis increments of the 3D concentrations by scaling the background profiles. Instead we explore the Lagrangian characteristics of the problem for generating local displacement ensembles. These high-resolution state-dependent ensembles are then used to parameterize the background errors and generate 3D aerosol increments. The algorithm has computational complexity running at a resolution of 1/4 degree, globally. We will present the result of assimilating AOD retrievals from MODIS (on both Aqua and TERRA satellites) from AERONET for validation. The impact on the GEOS-5 Aerosol Forecasting will be fully documented.
As one of the most adopted sequential data assimilation methods in many areas, especially those involving complex nonlinear dynamics, the ensemble Kalman filter (EnKF) has been under extensive investigation regarding its properties and efficiency. Compared to other variants of the Kalman filter (KF), EnKF is straightforward to implement, as it employs random ensembles to represent solution states. This, however, introduces sampling errors that affect the accuracy of EnKF in a negative manner. Though sampling errors can be easily reduced by using a large number of samples, in practice this is undesirable as each ensemble member is a solution of the system of state equations and can be time consuming to compute for large-scale problems. In this paper we present an efficient EnKF implementation via generalized polynomial chaos (gPC) expansion. The key ingredients of the proposed approach involve (1) solving the system of stochastic state equations via the gPC methodology to gain efficiency; and (2) sampling the gPC approximation of the stochastic solution with an arbitrarily large number of samples, at virtually no additional computational cost, to drastically reduce the sampling errors. The resulting algorithm thus achieves a high accuracy at reduced computational cost, compared to the classical implementations of EnKF. Numerical examples are provided to verify the convergence property and accuracy improvement of the new algorithm. We also prove that for linear systems with Gaussian noise, the first-order gPC Kalman filter method is equivalent to the exact Kalman filter.
In this paper, we present a method combining multilevel Monte Carlo (MLMC) and a graph-based primary subnetwork identification algorithm to provide estimates of the mean and variance of the distribution of first passage times in fracture media at significantly lower computational cost than standard Monte Carlo (MC) methods. Simulations of solute transport are performed using a discrete fracture network (DFN), and instead of using various grid resolutions for levels in the MLMC, which is standard practice in MLMC, we identify a hierarchy of subnetworks in the DFN based on the shortest topological paths through the network using a graph-based method. While the mean of these ensembles is of critical importance, the variance is also essential in fractured media where uncertainty is an overarching theme, and understanding variability across an ensemble is a requirement for safety assessments. The method provides good estimates of the mean and variance at two orders of magnitude lower computational cost than MC.