Search NASA⌕ Search

SEARCH · Search NASA

Results for “Function approximation”

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 289 records · Page 16
All sourcesNASA NTRSDOE OSTI
DOE OSTI2023

Probabilistic partition of unity networks for high–dimensional regression problems

We explore the probabilistic partition of unity network (PPOU-Net) model in the context of high-dimensional regression problems and propose a general framework focusing on adaptive dimensionality reduction. With the proposed framework, the target function is approximated by a mixture of experts model on a low-dimensional manifold, where each cluster is associated with a fixed-degree polynomial. We present a training strategy that leverages the expectation maximization (EM) algorithm. During the training, we alternate between (i) applying gradient descent to update the DNN coefficients; and (ii) using closed-form formulae derived from the EM algorithm to update the mixture of experts model parameters. Under the probabilistic formulation, step (ii) admits the form of embarrassingly paralleliazable weighted least-squares solves. The PPOU-Nets consistently outperform the baseline fully-connected neural networks of comparable sizes in numerical experiments of various data dimensions. Here, we also explore the proposed model in applications of quantum computing, where the PPOU-Nets act as surrogate models for cost landscapes associated with variational quantum circuits.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2022

Improved Architectures and Training Algorithms for Deep Operator Networks

Operator learning techniques have recently emerged as a powerful tool for learning maps between infinite-dimensional Banach spaces. Trained under appropriate constraints, they can also be effective in learning the solution operator of partial differential equations (PDEs) in an entirely self-supervised manner. In this work we analyze the training dynamics of deep operator networks (DeepONets) through the lens of Neural Tangent Kernel theory, and reveal a bias that favors the approximation of functions with larger magnitudes. To correct this bias we propose to adaptively re-weight the importance of each training example, and demonstrate how this procedure can effectively balance the magnitude of back-propagated gradients during training via gradient descent. We also propose a novel network architecture that is more resilient to vanishing gradient pathologies. Taken together, our developments provide new insights into the training of DeepONets and consistently improve their predictive accuracy by a factor of 10-50x, demonstrated in the challenging setting of learning PDE solution operators in the absence of paired input-output observations.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2021

Initial Fermi orbital descriptors for FLOSIC calculations: The quick-FOD method

Fermi orbital descriptors (FODs) play a key role in Fermi-Löwdin orbital self-interaction correction (FLOSIC) calculations used to remove self-interaction from approximate density functionals. Optimal FODs are obtained by minimizing the self-interaction-corrected total energy, and, in this process, identifying initial sets of FODs becomes crucial. Here we propose, implement and test a novel method for automatically initializing FODs, quick-FOD, based on the minimization of an empirical energy expression that involves a Coulomb-like FOD-electron density attraction, an FOD-FOD short-range repulsion, and an exchange-like FOD-density repulsion. Quick-FOD successfully reproduces FOD arrangements in qualitatively good agreement with Lewis theory and with full-fledged FLOSIC calculations.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2018

Gradient-based optimization for regression in the functional tensor-train format

Predictive analysis of complex computational models, such as uncertainty quantification (UQ), must often rely on using an existing database of simulation runs. In this paper we consider the task of performing low-multilinear-rank regression on such a database. Specifically we develop and analyze an efficient gradient computation that enables gradient-based optimization procedures, including stochastic gradient descent and quasi-Newton methods, for learning the parameters of a functional tensor-train (FT). We compare our algorithms with 22 other nonparametric and parametric regression methods on 10 real-world data sets and show that for many physical systems, exploiting low-rank structure facilitates efficient construction of surrogate models. Here, we use a number of synthetic functions to build insight into behavior of our algorithms, including the rank adaptation and group-sparsity regularization procedures that we developed to reduce overfitting. Finally we conclude the paper by building a surrogate of a physical model of a propulsion plant on a naval vessel.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2021

Norm-Conserving Pseudopotentials and Basis Sets to Explore Actinide Chemistry in Complex Environments

We constructed and tested a new set of norm-conserving pseudopotentials, as well as companion Gaussian basis sets, for the actinide series (Ac–Lr) using the Goedecker, Teter and Hutter (GTH) formalism within the generalized gradient approximation (GGA) functional for exchange-correlation of Perdew, Burke and Ernzerhof (PBE). This fills a critical void in the literature for studying the chemistry of 5f–block elements in condensed phase. The accuracy and reliability of the newly parameterized An–GTH pseudopotentials and basis sets, a variety of tests on actinide-containing molecules were carried out and compared to all-electron and available experimental results. The new pseudopotentials include both medium ( [Xe]4f 14 ) and large core ( [Xe]4f 14 5d 10 ) options that successfully reproduced structure and energetics, particularly redox processes. The medium core size set, in particular, reproduced all electron calculations over multiple oxidation states from 0 to VII, whereas the large core set was suitable only for the early series elements and low oxidation states.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Pareto Optimization of Oligomer Polarizability and Dipole Moment Using a Genetic Algorithm

High-performance electronic components are highly sought after in order to produce increasingly smaller and cheaper electronic devices. Drawing inspiration from inorganic dielectric materials, in which both polarizability and polarization contribute, organic materials can also maximize both. For a large set of small molecules drawn from PubChem, a Pareto-like front appears between the polarizability and dipole moment, indicating the presence of an apparent trade-off between these two properties. We tested this balance in π-conjugated materials by searching for novel conjugated hexamers with simultaneously large polar- izabilities and dipole moments with potential use for dielectric materials. Using a genetic algorithm (GA) screening technique in conjunction with an approximate density functional tight-binding method for property calculations, we were able to efficiently search chemical space for optimal hexamers. Given the scope of chemical space, using the GA technique saves considerable time and resources by speeding up molecular searches compared to a systematic search. Here, we also explored the underlying structure–function relationships, including sequence and monomer properties, that characterize large polarizability and dipole moment regimes.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Libration of hydroxyl groups in layered aluminum (oxy)hydroxides and other material analogs: insights from inelastic neutron scattering and theory

We analyzed the hydroxyl librational signatures of five structurally related aluminum (oxy)hydroxides, using inelastic neutron scattering (INS) and plane-wave lattice dynamics simulations. A clear trend across these aluminum-containing phases illustrates the relationship between hydrogen bonding, local atomic structure, and the spectral location and profile of the librational bands. The INS spectra have been compared to previous optical spectroscopy and computational studies, highlighting the complementary nature of the INS technique. Taking into account other structurally or chemically related material analogs, we have identified a correlation between a blueshift (to higher energy) of the upper librational band edge and the geometry of the hydrogen bond interactions, mirroring (with opposite correlation) the well-known redshift in the intramolecular O–H stretching energy with increasing hydrogen bond strength. For hydroxyl groups that do not participate in hydrogen bonding effectively, the bending librations occur at lower energies and hybridize with metal–oxygen lattice modes. Standard density functional theory approximations, including dispersion corrections, struggle to correctly predict vibrational frequencies of motions dominated by H but perform well for metal–oxygen modes, allowing us to make detailed mode assignments in several cases, including a demonstration of how layer-to-layer disorder in boehmite hydrogen bond orientations is reflected in the sharp but minor low energy peaks (at ∼70–80 meV) of the INS spectrum.

Wang, Hsiu-Wen [Oak Ridge National Laboratory (ORN↗
DOE OSTI2024

A near X-point charge exchange neutral spectroscopy (CENS) system for DIII-D

A 16 channel spectroscopy system has been installed on DIII-D to provide information about the energy distribution of the atomic neutrals using the Doppler shift and broadening of passive Balmer-α emission. Here, the lines of sight are dominantly in the toroidal direction, with tangencies near the lowest point on closed magnetic flux surfaces moving from the lower divertor upward into the confined plasma. This allows the system to provide details of the neutrals as they undergo various atomic physics processes while traveling upward from the lower divertor. The spectrally resolved measurements provide several advantages that complement typical optical filter based measurements of hydrogenic spectral lines. These include more direct measurements of the neutrals that fuel deeper in the plasma and capturing neutral velocity distribution information via the Doppler broadened and shifted line. In addition, the spectral separation of higher energy “thermal” neutrals and bright cold emission from the scrape-off layer allows for a more straightforward calculation of the underlying neutral densities based on the emission because of the dependence of the thermal emission on confined plasma properties that are approximately flux functions and well measured.

Atomic and molecular physics↗
DOE OSTI2026

Mott vs Kondo: Influence of various density functional based methods on the Ce isostructural phase transition mechanism

The cerium iso-structural phase transition (γ–α) is dominated by f-electron localization changes that results in a magnetic ordering change and a volume collapse. Generally, these physics are difficult to capture with ab initio and first principles methods. However, previous works have shown various methods to be successful in predicting at least some of the physics of the γ to α phase transition. Therefore, here, we perform a broad survey of density functional based methods across three levels of theory and types of functions [GGA (generalized gradient approximation), MetaGGA, and hybrid functionals] and compare the results, focusing on hydrostatic compression across the phase boundary at zero kelvin. For the methods that best reproduce experimental results, we directly probe the predicted mechanisms and frame the results in the Mott/Kondo debate, assessing how the underlying methods and assumptions of different functionals can assess the physical drivers in the phase transition, providing insight into the governing processes of this unique phase transition.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Spatiotemporal forecasting of the edge localized modes in tokamak plasmas using neural networks

Artificial intelligence techniques have been increasingly adopted by the plasma and fusion science to address problems like plasma reconstruction, surrogate modeling, and tokamak/stellarator optimization. A key focus in sustained fusion research is the prediction and mitigation of edge-localized-modes (ELMs), instabilities that occur in short, periodic bursts and can cause erosion to the tokamak vessel wall. Recent research has demonstrated the power of neural networks in approximating continuous functions. In this work, we build spatiotemporal forecasting models that can predict the onset of ELMs and their evolution at early stages. We leverage recent advances in generative modeling, sequence-to-sequence modeling, and Fourier neural operators to propose architectures and training strategies that can learn to forecast short to long term dynamics of the noisy signals due to ELMs. We benchmark the developed model against a state-of-the-art foundation model using the beam emission spectroscopy (BES) data that captures the plasma fluctuations due to ELMs over a 8 x 8 spatial grid. Our models demonstrate high accuracy, outperforming the baselines, in predicting the evolution of BES signals during ELM events. Furthermore, the developed models exhibit high accuracy in predicting the rapid rise and relaxation of the signals due to ELMs within 30–80 µs.

edge localized modes↗
DOE OSTI2020

Two-particle correlation functions in cluster perturbation theory: Hubbard spin susceptibilities

Cluster perturbation theory (CPT) is a computationally economic method commonly used to estimate the momentum- and energy-resolved single-particle Green's function. It has been used extensively in direct comparisons with experiments that effectively measure the single-particle Green's function, e.g., angle-resolved photoemission spectroscopy. However, many experimental observables are given by two-particle correlation functions. CPT can be extended to compute two-particle correlation functions by approximately solving the Bethe-Salpeter equation. We implement this method and focus on the transverse spin susceptibility, measurable via inelastic neutron scattering or with optical probes of atomic gases in optical lattices. Finally, we benchmark the method with the one-dimensional Fermi-Hubbard model by comparing with known results.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2022

Effect of radiation-induced defects on the superfluid density and optical conductivity of overdoped La 2 - x Sr x CuO 4

Using a combination of time-domain THz spectroscopy (TDTS) and mutual inductance measurements, we have investigated the low-energy electrodynamic response of overdoped La$_{2-x}$Sr$_x$CuO$_4$ films that have been exposed to ion irradiation. Films went through three rounds of irradiation (2, 4, and 6 $\times 10^{13}$ ions/cm$^2$) and mututal inductance and TDTS experiments were performed between each step. Together with the as-grown film, this gives four different levels of disorder. The transport scattering rate that is measured directly in the THz experiments is an approximately linear function of the radiation dose at all temperatures. This is consistent with a proportionate increase in elastic scattering from point-like 2D disorder. Here, in the superconducting state we find that the relation between T$_c$, the superfluid density, and the scattering rates are quantitatively at odds with the predictions based on the extant theory of Abrikosov-Gorkov-like pair breaking in a dirty d-wave superconductor. Increasing disorder causes only a small change in T$_c$ for the overdoped films, but the changes to the $\omega \sim 0$ superfluid density are much larger.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2024

High-throughput determination of Hubbard $U$ and Hund $J$ values for transition metal oxides via the linear response formalism

DFT+U provides a convenient, cost-effective correction for the self-interaction error (SIE) that arises when describing correlated electronic states using conventional approximate density functional theory (DFT). The success of a DFT+U(+J) calculation hinges on the accurate determination of its Hubbard U and Hund J parameters, and the linear response (LR) methodology has proven to be computationally effective and accurate for calculating these parameters. This study provides a high-throughput computational analysis of the U and J values for transition metal d-electron states in a representative set of over 1000 magnetic transition metal oxides (TMOs), providing a frame of reference for researchers who use DFT+U to study transition metal oxides. In order to perform this high-throughput study, an ATOMATE workflow is developed for calculating U and J values automatically on massively parallel supercomputing architectures. Here, to demonstrate an application of this workflow, the spin-canting magnetic structure and unit cell parameters of the multiferroic olivine LiNiPO4 are calculated using the computed Hubbard U and Hund J values for Ni-d and O-p states, and are compared with experiment. Both the Ni-d U and J corrections have a strong effect on the Ni-moment canting angle. Additionally, including a O-pU value results in a significantly improved agreement between the computed lattice parameters and experiment

36 MATERIALS SCIENCE↗
DOE OSTI2018

Field Characteristics of Spin-Torque Diode Sensitivity in the Presence of a Bias Current

The use of the spin-torque diode effect, which is generated upon the current-induced transfer of spin angular momentum in magnetic tunnel junctions, opens the prospect of considerably improving microwave sensitivity in the gigahertz frequency band over that of semiconductor Schottky diodes. In this work, the spin-diode effect of microwave signal rectification in a magnetic tunnel junction upon the resonant excitation of spin waves in the free magnetic layer as a result of the current-induced spin-transfer torque effect is analyzed theoretically. Frequency characteristics of the resonant response of a spin-torque diode to a microwave signal are calculated in the linear macrospin approximation as functions of the direction and value of the applied magnetic field and bias current. It is shown that with zero bias current, the maximum rectified voltage across a junction is obtained in the geometry of mutually perpendicular magnetizations of magnetic layers; as the resonant frequency of oscillations in a magnetic field increases, this voltage drops at the retained equilibrium orientation of spins in the layers. After the bias current is switched on, the resonance amplitude of forced oscillations of the free layer spins upon microwave excitation grows sharply at the critical point of the spin-torque diode’s loss of equilibrium state stability. The linewidth at this point is limited only by nonlinear effects. Improving spin-torque diode sensitivity is important for its use in microwave holographic vision systems.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2019

Improved unconstrained energy functional method for eigensolvers in electronic structure calculations

This paper reports on the performance of a preconditioned conjugate gradient based iterative eigensolver using an unconstrained energy functional minimization scheme. In contrast to standard implementations, this scheme avoids an explicit reorthogonalization of the trial eigenvectors and becomes an attractive alternative for the solution of very large problems. The unconstrained formulation is implemented in the first-principles materials and chemistry CP2K code, which performs electronic structure calculations based on a density functional theory approximation to the solution of the many-body Schrödinger equation. We study the convergence of the unconstrained formulation, as well as its parallel scaling, on a Cray XC40 at the National Energy Research Scientific Computing Center (NERSC). The systems we use in our studies are bulk liquid water, a supramolecular catalyst gold(III)-complex, a bilayer of MoS2-WSe2 and a divacancy point defect in silicon, with the number of atoms ranging from 2,247 to 12,288. We show that the unconstrained formulation with an appropriate preconditioner has good convergence properties and scales well to 230k cores, roughly 38% of the full machine.

Del Ben, M↗
DOE OSTI2019

Multi-fidelity Gaussian Process Bandit Optimisation

In many scientific and engineering applications, we are tasked with the maximisation of an expensive to evaluate black box function f. Traditional settings for this problem assume just the availability of this single function. However, in many cases, cheap approximations to f may be obtainable. For example, the expensive real world behaviour of a robot can be approximated by a cheap computer simulation. We can use these approximations to eliminate low function value regions cheaply and use the expensive evaluations of f in a small but promising region and speedily identify the optimum. We formalise this task as a multi-fidelity bandit problem where the target function and its approximations are sampled from a Gaussian process. We develop MF-GP-UCB, a novel method based on upper confidence bound techniques. In our theoretical analysis we demonstrate that it exhibits precisely the above behaviour and achieves better bounds on the regret than strategies which ignore multi-fidelity information. Empirically, MF-GP-UCB outperforms such naive strategies and other multi-fidelity methods on several synthetic and real experiments.

97 MATHEMATICS AND COMPUTING↗