Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “CONVOLUTION THEORY”

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.

At least 163 records · Page 9
All sourcesNASA NTRSDOE OSTI
DOE OSTI2026

An ultra-fast method for generating synthetic down-scattered neutron data for inertial confinement fusion implosions

In inertial confinement fusion experiments at the National Ignition Facility, asymmetries are probed by a variety of neutron diagnostics, including neutron imaging systems, real-time neutron activation diagnostics (RTNADs), and neutron spectrometers. It is often useful to generate synthetic data based on these diagnostics to validate and tune models. However, current methods of doing so using Monte Carlo particle tracing are time-consuming. In this paper, an ultra-fast method is presented for generating synthetic neutron images, RTNAD data, and spectrometry data using line integrals and 3D convolutions. While it does not contain as much physics as particle tracing codes, it is thousands of times faster and produces nearly identical data. This enables analysis techniques that depend on generating large amounts of synthetic data, which will prove very useful for the study of asymmetries going forward.

Deuterium↗
NASA NTRS1994

Spin glasses and error-correcting codes

In this article, we study a model for error-correcting codes that comes from spin glass theory and leads to both new codes and a new decoding technique. Using the theory of spin glasses, it has been proven that a simple construction yields a family of binary codes whose performance asymptotically approaches the Shannon bound for the Gaussian channel. The limit is approached as the number of information bits per codeword approaches infinity while the rate of the code approaches zero. Thus, the codes rapidly become impractical. We present simulation results that show the performance of a few manageable examples of these codes. In the correspondence that exists between spin glasses and error-correcting codes, the concept of a thermal average leads to a method of decoding that differs from the standard method of finding the most likely information sequence for a given received codeword. Whereas the standard method corresponds to calculating the thermal average at temperature zero, calculating the thermal average at a certain optimum temperature results instead in the sequence of most likely information bits. Since linear block codes and convolutional codes can be viewed as examples of spin glasses, this new decoding method can be used to decode these codes in a way that minimizes the bit error rate instead of the codeword error rate. We present simulation results that show a small improvement in bit error rate by using the thermal average technique.

Belongie, M. L.↗
DOE OSTI2024

Unifying Combinatorial and Graphical Methods in Artificial Intelligence

Recently, a new graph Laplacian, called the inner product Laplacian, was introduced which generalizes many existing Laplacians, including the normalized and combinatorial Laplacian and their weighted variants. The key observation behind the inner product Laplacian is that by defining appropriate inner product spaces on the vertices and edges, the standard Laplacians can be recovered as Hodge Laplacians over the simplicial complex formed by the edges and vertices. These inner product spaces form a natural way to incorporate non-combinatorial information into the definition of a domain-specific Laplacian. In particular, in contrast to current domain-specific weighting schemes which rely solely on edge weights, information regarding the similarity of non-adjacent vertices and arbitrary pairs of edges can be effectively incorporated into the Laplacian. In order to illustrate this approach we consider the problem of calculating the potential energy of an atomistic configuration using Graph Neural Networks. In comparison with start-of-the-art approaches, such as SchNet, our approach replaces a learned (via auto-encoder) representation of the atom types with an inner product space on atoms based on scientific knowledge (e.g., electronegativity). We will illustrate how this approach captures key chemical properties of the molecules and compare the energy calculations with state-of-the-art neural network approaches. However, to compute the resulting Laplacian involves a mixture of sparse and dense matrix computation and yields a dense matrix as the basis for the graph convolution. This dense convolutional kernel necessitates moving away from the standard message passing framework for graph neural networks and increases the computational cost of applying the kernel. In order to mitigate these costs we investigate means of leveraging the mixed sparse and dense computations to reduce the overall computational cost and how these approaches can be automatically transferred to energy efficient hardware (e.g., field programmable gate arrays (FPGAs)).

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Equation-of-Motion Coupled-Cluster Cumulant Green’s Function for Excited States and X-ray Spectra

Green’s function methods within many-body perturbation theory provide a general framework for treating electronic correlations in excited states and spectra. Conventional methods using the Dyson equation or the cumulant expansion are typically based on the GW self-energy approximation. In order to extend this approximation in molecular systems, a non-perturbative real-time coupled-cluster cumulant Green’s function approach has been introduced, where the cumulant is obtained as the solution to a set of coupled first order, non-linear differential equations. This approach naturally includes non-linear corrections to conventional cumulant Green’s function techniques where the cumulant is linear in the GW self-energy. The method yields the spectral function associated with the core Green’s function, which is directly related to the x-ray photoemission spectra (XPS) of molecular systems. The approach also yields very good results for binding energies and satellite excitations. The x-ray absorption spectrum (XAS) is then obtained as a convolution of the core spectral function and an effective one-body absorption spectrum. Here this approach is extended to include the full coupled-cluster-singles (CCS) core Green’s function by including the complete form of the non-linear contributions to the cumulant as well as all single, double, and triple cluster excitations in the CC amplitude equations. This approach naturally builds in orthogonality and shake-up effects analogous to those in the Mahan-Noizeres-de Dominicis edge singularity corrections that enhance the XAS near the edge. Themethod is illustrated for the XPS and XAS of NH 3 .

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

Twisted bilayer graphene. V. Exact analytic many-body excitations in Coulomb Hamiltonians: Charge gap, Goldstone modes, and absence of Cooper pairing

We find exact analytic expressions for the energies and wave functions of the charged and neutral excitations above the exact ground states (at rational filling per unit cell) of projected Coulomb Hamiltonians in twisted bilayer graphene. Our exact expressions are valid for any form of the Coulomb interaction and any form of A A and A B / B A tunneling. The single charge excitation energy is a convolution of the Coulomb potential with a quantum geometric tensor of the TBG bands. The neutral excitations are (high-symmetry group) magnons, and their dispersion is analytically calculated in terms of the form factors of the active bands in TBG. The two-charge excitation energy and wave functions are also obtained, and a sufficient condition on the graphene eigenstates for obtaining a Cooper pair from Coulomb interactions is obtained. For the actual TBG bands at the first magic angle, we can analytically show that the Cooper pair binding energy is zero in all such projected Coulomb models, implying that either phonons and/or nonzero kinetic energy are needed for superconductivity. Since Vafek and Kang [Phys. Rev. Lett. 125, 257602 (2020)] showed that the kinetic energy bounds on the superexchange energy are less 10 -3 in Coulomb units, the phonon mechanism becomes then very likely. If nonetheless the superconductivity is due to kinetic terms which render the bands nonflat, one prediction of our theory is that the highest T c would not occur at the highest DOS.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2020

A Multiple Filter Based Neural Network Approach to the Extrapolation of Adsorption Energies on Metal Surfaces for Catalysis Applications

Computational catalyst discovery involves the development of microkinetic reactor models based on estimated parameters determined from density functional theory (DFT). For complex surface chemistries, the number of reaction intermediates can be very large, and the cost of calculating the adsorption energies by DFT for all surface intermediates even for one active site model can become prohibitive. In this paper, we have identified appropriate descriptors and machine learning models that can be used to predict a significant part of these adsorption energies given data on the rest of them. Furthermore, our investigations also included the case when the species data used to train the predictive model are of different size relative to the species the model tries to predict—this is an extrapolation in the data space which is typically difficult with regular machine learning models. Due to the relative size of the available data sets, we have attempted to extrapolate from the larger species to the smaller ones in the current work. Here, we have developed a neural network based predictive model that combines an established additive atomic contribution based model with the concepts of a convolutional neural network that, when extrapolating, achieves a statistically significant improvement over the previous models.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2024

Non-Electricity Based Renewable Fuels: Theory and Computation for Solar Thermochemical Hydrogen

Dominated by photovoltaics and wind, current renewable energy sources generate mostly electricity, but 80% of the global final energy consumption occurs in form of fuels. Therefore, direct solar fuel generation would be a major breakthrough for the energy transition. Solar thermochemical hydrogen (STCH) is one of the very few potential routes towards scalable renewable fuels, but currently suffers from lack of an oxide working material that could optimally perform energy conversion within the thermodynamic boundary conditions. Theory and computation can contribute in two distinct ways, through materials search and discovery, but also by providing detailed mechanistic models for specific systems so to advance our understanding of possible design strategies. To enable high-throughput materials screening, we developed a defect graph neural network (dGNN) machine learning approach,[1] which accelerates the prediction of defect formation energies by replacing the tedious density functional theory (DFT) supercell calculations for all possible defect sites. This approach enables high-throughput database screening of oxides, which was integrated with thermodynamic modeling to extract the reduction entropies as additional selection criterion for STCH. Once potential candidate materials are identified, detailed models can guide materials design by predicting performance characteristics. One challenge is to quantitatively predict thermochemical equilibria at high concentrations when the redox active defects start to interact with each other, thereby impeding the formation of additional defects. Introducing a model for the free energy of defect interaction, parametrized on the basis of DFT data, we simulated the complete STCH redox cycle for (Sr,Ce)MnO3 alloys, achieving near-quantitative agreement with experimental data.[2] The analysis of these simulations reveals how defect interactions diminish the reduction entropy and H2 yield, suggesting to include these interactions in design considerations. Finally, we revisit the popular van't Hoff method for analyzing reduction enthalpies and entropies. This method is not ideal, as it involves a temperature-dependent convolution of gas-phase and solid-state entropies, causing uncertainties in the same order of magnitude as the physical quantities of interest. To avoid this problem, we suggest a simple alternative approach which can be applied to experimental and simulated data alike.

first-principles calculations↗
DOE OSTI2024

Tunneling magnetoresistance in MgO tunnel junctions with Fe-based leads in empirically corrected density functional theory

The minority-spin Fe/MgO interface states are at the Fermi level in density functional theory (DFT), but experimental evidence and GW calculations place them slightly higher in energy. This small shift can strongly influence tunneling magnetoresistance (TMR) in junctions with a thin MgO barrier and its dependence on the concentration of Co in the electrodes. Here, in this study, an empirical potential correction to DFT is introduced to shift the interface states up to match the tunnel spectroscopy data. With this shift, TMR in Fe/MgO/Fe junctions exceeds 800% and 3000% at 3 and 4 monolayers (ML) of MgO, respectively. We further consider the effect of alloying of the Fe electrodes with up to 30% Co or 10% V, treating them in the coherent potential approximation (CPA). Alloying with Co broadens the interface states and brings a large incoherent minority-spin spectral weight to the Fermi level. Alloying with V brings the minority-spin resonant states close to the Fermi level. However, in both cases the minority-spin spectral weight at the Fermi level resides primarily at the periphery of the Brillouin zone, which is favorable for spin filtering. Using convolutions of k II -resolved barrier densities of states calculated in CPA, it is found that TMR is strongly reduced by alloying with Co or V but still remains above 500% at 4 ML of MgO up to 30% of Co or 5% V. At 5 ML, the TMR increases above 1000% in all systems considered. However, while TMR declines sharply with increasing bias up to 0.2 eV in the tunnel junctions with pure Fe leads, it remains almost constant up to 0.5 eV if leads are alloyed with Co.

74 ATOMIC AND MOLECULAR PHYSICS↗
DOE OSTI2020

Projecting Antarctica's contribution to future sea level rise from basal ice shelf melt using linear response functions of 16 ice sheet models (LARMIP-2)

Abstract. The sea level contribution of the Antarctic ice sheet constitutes a large uncertainty in future sea level projections. Here we apply a linear response theory approach to 16 state-of-the-art ice sheet models to estimate the Antarctic ice sheet contribution from basal ice shelf melting within the 21st century. The purpose of this computation is to estimate the uncertainty of Antarctica's future contribution to global sea level rise that arises from large uncertainty in the oceanic forcing and the associated ice shelf melting. Ice shelf melting is considered to be a major if not the largest perturbation of the ice sheet's flow into the ocean. However, by computing only the sea level contribution in response to ice shelf melting, our study is neglecting a number of processes such as surface-mass-balance-related contributions. In assuming linear response theory, we are able to capture complex temporal responses of the ice sheets, but we neglect any self-dampening or self-amplifying processes. This is particularly relevant in situations in which an instability is dominating the ice loss. The results obtained here are thus relevant, in particular wherever the ice loss is dominated by the forcing as opposed to an internal instability, for example in strong ocean warming scenarios. In order to allow for comparison the methodology was chosen to be exactly the same as in an earlier study (Levermann et al., 2014) but with 16 instead of 5 ice sheet models. We include uncertainty in the atmospheric warming response to carbon emissions (full range of CMIP5 climate model sensitivities), uncertainty in the oceanic transport to the Southern Ocean (obtained from the time-delayed and scaled oceanic subsurface warming in CMIP5 models in relation to the global mean surface warming), and the observed range of responses of basal ice shelf melting to oceanic warming outside the ice shelf cavity. This uncertainty in basal ice shelf melting is then convoluted with the linear response functions of each of the 16 ice sheet models to obtain the ice flow response to the individual global warming path. The model median for the observational period from 1992 to 2017 of the ice loss due to basal ice shelf melting is 10.2 mm, with a likely range between 5.2 and 21.3 mm. For the same period the Antarctic ice sheet lost mass equivalent to 7.4 mm of global sea level rise, with a standard deviation of 3.7 mm (Shepherd et al., 2018) including all processes, especially surface-mass-balance changes. For the unabated warming path, Representative Concentration Pathway 8.5 (RCP8.5), we obtain a median contribution of the Antarctic ice sheet to global mean sea level rise from basal ice shelf melting within the 21st century of 17 cm, with a likely range (66th percentile around the mean) between 9 and 36 cm and a very likely range (90th percentile around the mean) between 6 and 58 cm. For the RCP2.6 warming path, which will keep the global mean temperature below 2 ∘C of global warming and is thus consistent with the Paris Climate Agreement, the procedure yields a median of 13 cm of global mean sea level contribution. The likely range for the RCP2.6 scenario is between 7 and 24 cm, and the very likely range is between 4 and 37 cm. The structural uncertainties in the method do not allow for an interpretation of any higher uncertainty percentiles. We provide projections for the five Antarctic regions and for each model and each scenario separately. The rate of sea level contribution is highest under the RCP8.5 scenario. The maximum within the 21st century of the median value is 4 cm per decade, with a likely range between 2 and 9 cm per decade and a very likely range between 1 and 14 cm per decade.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2021

The ABACUS cosmological N -body code

We present abacus, a fast and accurate cosmological N-body code based on a new method for calculating the gravitational potential from a static multipole mesh. The method analytically separates the near- and far-field forces, reducing the former to direct 1/r 2 summation and the latter to a discrete convolution over multipoles. The method achieves 70 million particle updates per second per node of the Summit supercomputer, while maintaining a median fractional force error of 10 -5 . We express the simulation time-step as an event-driven ‘pipeline’, incorporating asynchronous events such as completion of co-processor work, input/output, and network communication. abacus has been used to produce the largest suite of N-body simulations to date, the abacussummit suite of 60 trillion particles, incorporating on-the-fly halo finding. abacus enables the production of mock catalogues of the volume and resolution required by the coming generation of cosmological surveys.

79 ASTRONOMY AND ASTROPHYSICS↗
NASA NTRS1998

Determining the Molecular Growth Mechanisms of Tetragonal Lysozyme Crystals

Studies of the growth of tetragonal lysozyme crystals employing atomic force microscopy (AFM) have shown the advantages of this technique in investigating the growth mechanisms of protein crystals [1]. The resolution of these studies was in the micron range, which revealed surface features such as the occurrence of dislocations and 2D nucleation islands, similar to those found in inorganic systems. They clearly showed that the crystals grew by these surface growth mechanisms. However, the studies also revealed some surprising features, such as bimolecular growth step heights and pronounced growth anisotropies on the (110) face, which could not be explained. In previous studies we employed Periodic Bond Chain (PBC) theory to tetragonal lysozyme crystal growth and found that the crystals were constructed by strongly bonded molecular chains forming helices about the 43 axes [2,3]. The helices were connected to each other with weaker bonds. The growth process was shown to proceed by the formation of these 43 helices, resulting in bimolecular growth steps on the (110) face. It was also shown to explain many other observations on tetragonal lysozyme crystal growth. Although PBC analysis is not a new technique [4], it has not been widely used as the mechanisms predicted from it could not be experimentally verified. In this study the growth process of these crystals was investigated, particularly for the (110) face, employing some newly developed high resolution AFM techniques. These techniques allowed individual lysozyme molecules on the crystal faces to be resolved and predictions from PBC analyses to be tested. The analyses had shown that of the two possible packing arrangements on (110) faces, only one would actually occur. Employing the first of the newly developed techniques, these faces were scanned by high resolution AFM. The resulting images were then compared with the theoretically constructed images for the two possible packing arrangements on the (110) face. The theoretical images were constructed by convolution of the crystal surface shape obtained from crystallographic data with the AFM tip shape. The comparison confirmed the prediction that the molecular packing arrangement of these faces corresponded to that for complete 43 helices. The second AFM technique that was developed was used to follow the growth process by measuring the dimensions of individual growth units on the (110) face. Linescans across a growth step, performed near the saturation limit of the crystals, allowed the growth unit dimensions to be measured. These measurements showed that growth on the (110) face proceeded by the formation of new 43 helices from the addition of at least tetramer units in the [110] direction. In the [001] direction growth proceeded by the addition of various aggregate units corresponding to the 4(sub 3) helices.

Li, Huayu↗
DOE OSTI2025

Revisiting single inclusive jet production: timelike factorization and reciprocity

Factorization theorems for single inclusive jet production play a crucial role in the study of jets and their substructure. In the case of small radius jets, the dynamics of the jet clustering can be factorized from both the hard production dynamics, and the dynamics of the low scale jet substructure measurement, and is described by a matching coefficient that can be computed in perturbative Quantum Chromodynamics (QCD). A proposed factorization formula describing this process has been previously presented in the literature, and is referred to as the semi-inclusive, or fragmenting jets formalism. By performing an explicit two-loop calculation, we show the inconsistency of this factorization formula, in agreement with another recent result in the literature. Building on recent progress in the factorization of single logarithmic observables, and the understanding of reciprocity, we then derive a new all-order factorization theorem for inclusive jet production. The use of a jet algorithm, being only a modification of the infrared structure of the measurement, modifies the structure of convolutions in the factorization theorem, as compared to inclusive fragmentation, but maintains the universality of the inclusive hard function and its associated Dokshitzer-Gribov-Lipatov-Altarelli-Parisi (DGLAP) evolution, which are ultraviolet properties. However, the non-trivial structure of convolutions in the factorization theorem implies that the jet functions exhibit a modified evolution. We perform an explicit two-loop calculation of the jet function in both N = 4 super Yang-Mills (SYM), and for all color channels in QCD, finding exact agreement with the structure derived from our renormalization group equations. In addition, we derive several new results, including an extension of our factorization formula to jet substructure observables, a jet algorithm definition of a generating function for the energy correlators, and new results for exclusive jet functions. Our results are a key ingredient for achieving precision jet substructure at colliders.

Effective Field Theories↗
DOE OSTI2021

Catalytic site ensembles: A context to reexamine the Langmuir-Hinshelwood kinetic description

The Langmuir-Hinshelwood formalism describes catalytic reactions of Langmuirian surface species under the assumption that all adsorbates are randomly-distributed – enabling adjacency of surface-bound intermediates to be determined solely by coverages of single-site occupants. We demonstrate herein that this approximation is inappropriate even for simple catalytic reactions (e.g. A + A → A 2 ) and manifestly neglects islanding of slowly-consumed species and partitioning of highly-reactive species inherently engendered by ≥ two-site elementary steps (e.g. A*–A* → A 2(g) + *–*). Rigorous description of kinetically-consequential islanding/partitioning phenomena requires explicit description of the coverage and chemical dynamics of all multi-site ensembles. Higher-order, ensemble-specific rate terms identify the particular microscopic events relevant to each ensemble, and, in doing so, reveal that each elementary step (e.g. A (g) adsorption) describes not one event (e.g. A (g) + * → A*) , but a sum over all ensemble-specific paths (e.g. A (g) + *–* → A*–* and A (g) + A*–* → A*–A*). De-convoluting each elementary step into its constituent multi-site paths proffers kinetic detail otherwise inaccessible – enabling (i) identification of rate- and selectivity-determining site ensembles, (ii) calculation of rates and degrees of rate control of ensemble-specific elementary steps, (iii) incorporation of adsorbate surface diffusion, (iv) incorporation of lateral adsorbate interactions, and (v) quantitative description of catalysis of multi-site-occupying intermediates (e.g. *C n H m * species in hydrocarbon (de-)hydrogenation and C-C bond coupling/cleavage reactions) which we demonstrate here is inaccessible to the Langmuir-Hinshelwood formalism even if adsorbate surface diffusion is infinitely-fast.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2020

A Bayesian framework for adsorption energy prediction on bimetallic alloy catalysts

Abstract For high-throughput screening of materials for heterogeneous catalysis, scaling relations provides an efficient scheme to estimate the chemisorption energies of hydrogenated species. However, conditioning on a single descriptor ignores the model uncertainty and leads to suboptimal prediction of the chemisorption energy. In this article, we extend the single descriptor linear scaling relation to a multi-descriptor linear regression models to leverage the correlation between adsorption energy of any two pair of adsorbates. With a large dataset, we use Bayesian Information Criteria (BIC) as the model evidence to select the best linear regression model. Furthermore, Gaussian Process Regression (GPR) based on the meaningful convolution of physical properties of the metal-adsorbate complex can be used to predict the baseline residual of the selected model. This integrated Bayesian model selection and Gaussian process regression, dubbed as residual learning, can achieve performance comparable to standard DFT error (0.1 eV) for most adsorbate system. For sparse and small datasets, we propose an ad hoc Bayesian Model Averaging (BMA) approach to make a robust prediction. With this Bayesian framework, we significantly reduce the model uncertainty and improve the prediction accuracy. The possibilities of the framework for high-throughput catalytic materials exploration in a realistic setting is illustrated using large and small sets of both dense and sparse simulated dataset generated from a public database of bimetallic alloys available in Catalysis-Hub.org.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2021

An adaptive approach to machine learning for compact particle accelerators

Abstract Machine learning (ML) tools are able to learn relationships between the inputs and outputs of large complex systems directly from data. However, for time-varying systems, the predictive capabilities of ML tools degrade if the systems are no longer accurately represented by the data with which the ML models were trained. For complex systems, re-training is only possible if the changes are slow relative to the rate at which large numbers of new input-output training data can be non-invasively recorded. In this work, we present an approach to deep learning for time-varying systems that does not require re-training, but uses instead an adaptive feedback in the architecture of deep convolutional neural networks (CNN). The feedback is based only on available system output measurements and is applied in the encoded low-dimensional dense layers of the encoder-decoder CNNs. First, we develop an inverse model of a complex accelerator system to map output beam measurements to input beam distributions, while both the accelerator components and the unknown input beam distribution vary rapidly with time. We then demonstrate our method on experimental measurements of the input and output beam distributions of the HiRES ultra-fast electron diffraction (UED) beam line at Lawrence Berkeley National Laboratory, and showcase its ability for automatic tracking of the time varying photocathode quantum efficiency map. Our method can be successfully used to aid both physics and ML-based surrogate online models to provide non-invasive beam diagnostics.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2023

Extracting the gamma-ray source-count distribution below the Fermi-LAT detection limit with deep learning

We reconstruct the extra-galactic gamma-ray source-count distribution, or dN/dS, of resolved and unresolved sources by adopting machine learning techniques. Specifically, we train a convolutional neural network on synthetic 2-dimensional sky-maps, which are built by varying parameters of underlying source-counts models and incorporate the Fermi-LAT instrumental response functions. The trained neural network is then applied to the Fermi-LAT data, from which we estimate the source count distribution down to flux levels a factor of 50 below the Fermi-LAT threshold. We perform our analysis using 14 years of data collected in the (1,10) GeV energy range. The results we obtain show a source count distribution which, in the resolved regime, is in excellent agreement with the one derived from cataloged sources, and then extends as dN/dS ~ S -2 in the unresolved regime, down to fluxes of 5 · 10 -12 cm -2 s -1 . The neural network architecture and the devised methodology have the flexibility to enable future analyses to study the energy dependence of the source-count distribution.

79 ASTRONOMY AND ASTROPHYSICS↗
DOE OSTI2022

Accurate Prediction of Voltage of Battery Electrode Materials Using Attention-Based Graph Neural Networks

Performing first-principles calculations to discover electrodes’ properties in the large chemical space is a challenging task. While machine learning (ML) has been applied to effectively accelerate those discoveries, most of the applied methods ignore the materials’ spatial information and only use predefined features: based only on chemical compositions. Here, we propose two attention-based graph convolutional neural network techniques to learn the average voltage of electrodes. Our proposed methods, which combine both atomic composition and atomic coordinates in 3D-space, improve the accuracy in voltage prediction significantly when compared to composition-based ML models. The first model directly learns the chemical reaction of electrodes and metal ions to predict their average voltage, whereas the second model combines electrodes’ ML predicted formation energy (E form ) to compute their average voltage. Our E form -based model demonstrates improved accuracy in transferability from our subset of learned Li ions to Na ions. Moreover, we predicted the theoretical voltage of 10 Na x MPO 4 F (M = Ti, Cr, Fe, Cu, Mn, Co, and Ni) fluorophosphate battery frameworks, which are unavailable in the Material Project database. It could be shown that we can expect average voltages higher than 3.1 V from those Na battery frameworks except from the NaTiPO 4 F and TiPO 4 F pair of electrodes, which offer an average voltage of 1.32 V.

25 ENERGY STORAGE↗
DOE OSTI2024

Enhancing Lattice Kinetic Schemes for Fluid Dynamics with Lattice-Equivariant Neural Networks

A new class of equivariant neural networks is presented, hereby dubbed lattice-equivariant neural networks (LENNs), designed to satisfy local symmetries of a lattice structure. The approach develops within a recently introduced framework aimed at learning neural network-based surrogate models’ lattice Boltzmann collision operators. Whenever neural networks are employed to model physical systems, respecting symmetries and equivariance properties has been shown to be key for accuracy, numerical stability, and performance. Here, hinging on ideas from group representation theory, trainable layers are defined whose algebraic structure is equivariant with respect to the symmetries of the lattice cell. In this work, the presented method naturally allows for efficient implementations, in terms of both memory usage and computational costs, supporting scalable training/testing for lattices in two spatial dimensions and higher (in which the size of symmetry group grows). The approach is validated and tested considering 2D and 3D flowing dynamics, both in laminar and turbulent regimes. It is compared with group-averaged-based symmetric networks and with plain, nonsymmetric, networks, showing how the presented approach unlocks the (a posteriori) accuracy and training stability of the former models and the train/inference speed of the latter networks. (LENNs are about one order of magnitude faster than group-averaged networks in 3D.) The work in this paper opens toward practical use of machine learning-augmented lattice Boltzmann CFD in real-world simulations.

97 MATHEMATICS AND COMPUTING↗