Search NASA⌕ Search

SEARCH · Search NASA

Results for “High efficiency acceleration”

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 1,531 records · Page 85

Novel Low Cost Organic Vapor Jet Printing of Striped High Efficiency Phosphorescent OLEDs for White Lighting

In this program, Universal Display Corporation and University of Michigan proposed to integrate three innovative concepts to meet the DOE's Solid State Lighting (SSL) goals: (1) high-efficiency phosphorescent organic light emitting device (PHOLED{trademark}) technology, (2) a white lighting design that is based on a series of red, green and blue OLED stripes, and (3) the use of a novel cost-effective, high rate, mask-less deposition process called organic vapor jet printing (OVJP). Our PHOLED technology offers up to four-times higher power efficiency than other OLED approaches for general lighting. We believe that one of the most promising approaches to maximizing the efficiency of OLED lighting sources is to produce stripes of the three primary colors at such a pitch (200-500 {mu}m) that they appear as a uniform white light to an observer greater than 1 meter (m) away from the illumination source. Earlier work from a SBIR Phase 1 entitled 'White Illumination Sources Using Striped Phosphorescent OLEDs' suggests that stripe widths of less than 500 {mu}m appear uniform from a distance of 1m without the need for an external diffuser. In this program, we intend to combine continued advances in this PHOLED technology with the striped RGB lighting design to demonstrate a high-efficiency, white lighting source. Using this background technology, the team has focused on developing and demonstrating the novel cost-effective OVJP process to fabricate these high-efficiency white PHOLED light sources. Because this groundbreaking OVJP process is a direct printing approach that enables the OLED stripes to be printed without a shadow mask, OVJP offers very high material utilization and high throughput without the costs and wastage associated with a shadow mask (i.e. the waste of material that deposits on the shadow mask itself). As a direct printing technique, OVJP also has the potential to offer ultra-high deposition rates (> 1,000 Angstroms/second) for any size or shaped features. As a result, we believe that this work will lead to the development of a cost-effective manufacturing solution to produce very-high efficiency OLEDs. By comparison to more common ink-jet printing (IJP), OVJP can also produce well-defined patterns without the need to pattern the substrate with ink wells or to dry/anneal the ink. In addition, the material set is not limited by viscosity and solvent solubility. During the program we successfully demonstrated a 6-inch x 6-inch PHOLED lighting panel consisting of fine-featured red, green and blue (R-G-B) stripes (1mm width) using an OVJP deposition system that was designed, procured and installed into UDC's cleanroom as part of this program. This project will significantly accelerate the DOE's ability to meet its 2015 DOE SSL targets of 70-150 lumens/Watt and less than $10 per 1,000 lumens for high CRI lighting index (76-90). Coupled with a low cost manufacturing path through OVJP, we expect that this achievement will enable the DOE to achieve its 2015 performance goals by the year 2013, two years ahead of schedule. As shown by the technical work performed under this program, we believe that OVJP is a very promising technology to produce low cost, high efficacy, color tunable light sources. While we have made significant progress to develop OVJP technology and build a pilot line tool to study basic aspects of the technology and demonstrate a lighting panel prototype, further work needs to be performed before its full potential and commercial viability can be fully assessed.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

Acceleration of GPU-based Krylov solvers via data transfer reduction

Krylov subspace iterative solvers are often the method of choice when solving large sparse linear systems. At the same time, hardware accelerators such as graphics processing units continue to offer significant floating point performance gains for matrix and vector computations through easy-to-use libraries of computational kernels. However, as these libraries are usually composed of a well optimized but limited set of linear algebra operations, applications that use them often fail to reduce certain data communications, and hence fail to leverage the full potential of the accelerator. In this study, we target the acceleration of Krylov subspace iterative methods for graphics processing units, and in particular the Biconjugate Gradient Stabilized solver that significant improvement can be achieved by reformulating the method to reduce data-communications through application-specific kernels instead of using the generic BLAS kernels, e.g. as provided by NVIDIA’s cuBLAS library, and by designing a graphics processing unit specific sparse matrix-vector product kernel that is able to more efficiently use the graphics processing unit’s computing power. Furthermore, we derive a model estimating the performance improvement, and use experimental data to validate the expected runtime savings. Finally, considering that the derived implementation achieves significantly higher performance, we assert that similar optimizations addressing algorithm structure, as well as sparse matrix-vector, are crucial for the subsequent development of high-performance graphics processing units accelerated Krylov subspace iterative methods.

97 MATHEMATICS AND COMPUTING↗

Design and analysis of graphene–MoS{sub 2} hybrid layer based SPR biosensor with TiO{sub 2}–SiO{sub 2} nano film for formalin detection: numerical approach

In this paper, a Surface Plasmon Resonance (SPR) biosensor has been numerically developed that used Graphene–MoS{sub 2}–Au–TiO{sub 2}–SiO{sub 2} hybrid structure for the detection of formalin. This developed sensor has been sensed the presence the formalin based on attenuated total reflection (ATR) method by observing the change of “surface plasmon resonance angle-the change of minimum reflectance attributor” and “the resonance frequency characteristics (RFC)-maximum transmittance attributor”. Chitosan has been used as probe sensing medium to accelerate particular reaction with the formalin. Here, graphene is used as biomolecular recognition element because of its high adsorption ability and optical characteristics which helps to improve sensor sensitivity, MoS{sub 2} used for it has larger band gap, high fluroscence quenching ability, higher optical absorption efficiency, TiO{sub 2}–SiO{sub 2} bilayer as the improvement of sensitivity and Gold (Au) as the sharp SPR curve. Numerical results give the impression that the variation of RFC and SPR angle for improper sensing of formalin is quiet negligible that confirms no formalin is detected whereas for proper sensing these change are considerably countable that confirms the detection of formalin. From the sensor sensitivity analysis, owing to add TiO{sub 2}–SiO{sub 2} bilayer with Graphene–MoS{sub 2} Hybrid layer, maximum sensitivity of 85.375% has been numerically resulted. This high sensor performance is for taking advantages of Graphene surface high selectively which detect bio molecular compounds through pi-stacking force, larger work function (5.1 eV) of MoS{sub 2} which allows the high sensitive detection of bio targets and Rich plasmon happens at the TiO{sub 2}–SiO{sub 2} interface.

77 NANOSCIENCE AND NANOTECHNOLOGY↗

Modeling in High Temperature Corrosion: A Review and Outlook

Realizing higher operating temperatures to increase efficiency of future applications for energy conversion and storage while minimizing cost is a challenge for development of high-temperature materials. Simultaneous optimization of mechanical properties and corrosion resistance continues to be a difficult task but is essential due to the need to significantly accelerate the transition between technology readiness levels in the future. Furthermore, oxidation-induced degradation will be a critical life-limiting mechanism at increased operating temperatures. Suitable high-temperature materials cannot be solely identified by time-consuming experiments and reliable computational methods incorporating the relevant physics of processes must be considered to complement the experimental efforts. In the present work, a review of the methods employed to model oxidation-induced material degradation described in literature will be discussed. Furthermore, their capability to predict lifetime and aid in material selection will be evaluated.

36 MATERIALS SCIENCE↗

Multi-variance replica exchange SGMCMC for inverse and forward problems via Bayesian PINN

Physics-informed neural network (PINN) has been successfully applied in solving a variety of nonlinear non-convex forward and inverse problems. However, the training is challenging because of the non-convex loss functions and the multiple optima in the Bayesian inverse problem. In this work, we propose a multi-variance replica exchange stochastic gradient Langevin dynamics method to tackle the challenge of the multiple local optima in the optimization and the challenge of the multiple modal posterior distribution in the inverse problem. Replica exchange methods are capable of escaping from the local traps and accelerating the convergence; two chains with different temperatures are designed where the low temperature chain aims for the local convergence, and the target of the high temperature chain is to travel globally and explore the whole loss function entropy landscape. However, it may not be efficient to solve mathematical inversion problems by using the vanilla replica method directly since the method doubles the computational cost in evaluating the forward solvers (likelihood functions) in the two chains. To address this issue, we propose to make different assumptions on the energy function estimation and this facilities one to use solvers of different fidelities in the likelihood function evaluation. More precisely, one can use a solver with low fidelity in the high temperature chain while using a solver with high fidelity in the low temperature chain. Our proposed method significantly lowers the computational cost in the high temperature chain, meanwhile preserving the accuracy and converging very fast. Here we give an unbiased estimate of the swapping rate and give an estimation of the discretization error of the scheme. To verify our idea, we design and solve four inverse problems which have multiple modes. The proposed method is also employed to train the Bayesian PINN to solve the forward and inverse problems; faster and more accurate convergence has been observed when compared to the stochastic gradient Langevin dynamics (SGLD) method and vanilla replica exchange methods.

97 MATHEMATICS AND COMPUTING↗

Emerging materials intelligence ecosystems propelled by machine learning

We report that the age of cognitive computing and artificial intelligence (AI) is just dawning. Inspired by its successes and promises, several AI ecosystems are blossoming, many of them within the domain of materials science and engineering. These materials intelligence ecosystems are being shaped by several independent developments. Machine learning (ML) algorithms and extant materials data are utilized to create surrogate models of materials properties and performance predictions. Materials data repositories, which fuel such surrogate model development, are mushrooming. Automated data and knowledge capture from the literature (to populate data repositories) using natural language processing approaches is being explored. The design of materials that meet target property requirements and of synthesis steps to create target materials appear to be within reach, either by closed-loop active-learning strategies or by inverting the prediction pipeline using advanced generative algorithms. AI and ML concepts are also transforming the computational and physical laboratory infrastructural landscapes used to create materials data in the first place. Surrogate models that can outstrip physics-based simulations (on which they are trained) by several orders of magnitude in speed while preserving accuracy are being actively developed. Automation, autonomy and guided high-throughput techniques are imparting enormous efficiencies and eliminating redundancies in materials synthesis and characterization. The integration of the various parts of the burgeoning ML landscape may lead to materials-savvy digital assistants and to a human-machine partnership that could enable dramatic efficiencies, accelerated discoveries and increased productivity. Here, we review these emergent materials intelligence ecosystems and discuss the imminent challenges and opportunities. The materials research landscape is being transformed by the infusion of approaches based on machine learning. This Review discusses the emerging materials intelligence ecosystems and the potential of human-machine partnerships for fast and efficient virtual materials screening, development and discovery.

36 MATERIALS SCIENCE↗

Uncovering fast solid-acid proton conductors based on dynamics of polyanion groups and proton bonding strength

Achieving high proton conductivity in inorganic solids is key for advancing many electrochemical technologies, including low-energy nano-electronics and energy-efficient fuel cells and electrolyzers. A quantitative understanding of the physical traits of a material that regulate proton diffusion is necessary for accelerating the discovery of fast proton conductors. In this work, we have mapped the structural, chemical and dynamic properties of solid acids to the elementary steps of the Grotthuss mechanism of proton diffusion. Our approach combines ab initio molecular dynamics simulations, analysis of phonon spectra and atomic structure calculations. We have identified the donor–hydrogen bond lengths and the acidity of polyanion groups as key descriptors of local proton transfer and the vibrational frequencies of the cation framework as the key descriptor of lattice flexibility. The latter facilitates rotations of polyanion groups and long-range proton migration in solid acid proton conductors. The calculated lattice flexibility also correlates with the experimentally reported superprotonic transition temperatures. Using these descriptors, we have screened the Materials Project database and identified potential solid acid proton conductors with monovalent, divalent and trivalent cations, including Ag + , Sr 2+ , Ba 2+ and Er 3+ cations, which go beyond the traditionally considered monovalent alkali cations (Cs + , Rb + , K + , and NH 4 + ) in solid acids.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Combined Photoneutron And X Ray Interrogation Of Containers For Nuclear Materials

Effective cargo inspection systems for nuclear material detection require good penetration by the interrogating radiation, generation of a sufficient number of fissions, and strong and penetrating detection signatures. Inspection systems need also to be sensitive over a wide range of cargo types and densities encountered in daily commerce. Thus they need to be effective with highly hydrogenous cargo, where neutron attenuation is a major limitation, as well as with dense metallic cargo, where x-ray penetration is low. A system that interrogates cargo with both neutrons and x-rays can, in principle, achieve high performance over the widest range of cargos. Moreover, utilizing strong prompt-neutron ({approx}3 per fission) and delayed-gamma ray ({approx}7 per fission) signatures further strengthens the detection sensitivity across all cargo types. The complementary nature of x-rays and neutrons, used as both probing radiation and detection signatures, alleviates the need to employ exceedingly strong sources, which would otherwise be required to achieve adequate performance across all cargo types, if only one type of radiation probe were employed. A system based on the above principles, employing a commercially-available 9 MV linac was developed and designed. Neutrons are produced simultaneously with x-rays by the photonuclear interaction of the x-ray beam with a suitable converter. A total neutron yield on the order of 10{sup 11} n/s is achieved with an average electron beam current of 100 {mu}A. If fissionable material is present, fissions are produced both by the high-energy x-ray beam and by the photoneutrons. Photofission and neutron fission dominate in hydrogenous and metallic cargos, respectively. Neutron-capture gamma rays provide information on the cargo composition. The prompt neutrons resulting from fission are detected by two independent detector systems: by very efficient Differential Die Away Analysis (DDAA) detectors, and by direct detection of neutrons with energies higher than 3 MeV using a recently developed fluorine-based threshold activation detector (TAD). The delayed gamma-ray signals are measured with high efficiency with the same TAD and with additional lower-cost plastic scintillators.

07 ISOTOPE AND RADIATION SOURCES↗

Terahertz free electron superradiation from mimicking surface plasmons-two electron beams interaction within a 3-mirror quasi-optical cavity

Interaction between mimicking surface plasmons and electron beams provides a good opportunity to develop terahertz (THz) radiation sources. In this paper, such an interaction in a special 3-mirror quasi-optical cavity is presented and explored. The 3-mirror quasi-optical cavity acts as a resonant tunable system, the resonance frequency of which can be tuned by adjusting the distance between mirrors. The study demonstrated THz free electron superradiation from the interaction of mimicking surface plasmons and two electron beams that formed a resonance within the structure, with the 3-mirror cavity enhancing the intensity of superradiation. Moreover, this system can work in the high-harmonic superradiation region with relatively high efficiency and low current density. This concept thus shows the application potential for electron-beam-driven terahertz sources.

43 PARTICLE ACCELERATORS↗

New generation bunch shape monitor for ion accelerators

Measuring longitudinal beam parameters is important for operation and development of high intensity linear accelerators, but it is notoriously difficult for proton and ion beams at non-relativistic energies. The Bunch Shape Monitor (BSM) is a device used for measuring the longitudinal bunch distribution in ion linacs. The existing BSM models have poor electron collection efficiency from the wire and are limited to one-dimensional measurements of the phase coordinate. In response to this problem, we have developed a new generation BSM with improved performance. The proposed design incorporates three major innovations: First, the collection efficiency was improved by adding a focusing field between the wire and the entrance slit, which will also allow measurements over a much wider dynamic range. Second, an improvement in the measurement speed was achieved by sampling longitudinal profiles of multiple energy slices simultaneously, where the BSM wire is placed at the exit of an ion spectrometer so that ions with different energies hit the wire at different horizontal coordinates along the wire. Finally, the design incorporates a motion system that can shift the wire and deflecting cavity together, enabling transverse profile measurements like a wire scanner. Here, in this paper, we will provide the design of the new BSM and report on its beam test results at the Spallation Neutron Source facility in Oak Ridge National Laboratory.

43 PARTICLE ACCELERATORS↗

Machine learning-accelerated discovery of iron cobalt phosphides as rare-earth-free magnets

Here, the discovery of rare-earth-free permanent magnets has been a goal of scientists for decades. The absence of rare-earth elements will alleviate a pressing concern about the availability of rare-earth elements used in permanent magnets. These magnets are crucial for applications such as wind turbines, electric cars, and memory devices. Rare-earth magnets are special owing to a large magnetic anisotropy energy (K 1 ). In contrast, iron cobalt phosphides hold promise since doping P into cubic FeCo can induce anisotropy, leading to a large coercivity, without introducing rare-earth elements. We present a comprehensive search over the Fe-Co-P ternary space for magnets, utilizing recently developed adaptive machine learning feedback to efficiently screen over 850 000 structures. We focus on machine learning acceleration as a paradigm for materials design. Further adaptive genetic algorithm searches and first-principles calculations aid in the identification of 16 new structures below the known convex hull. Five of them possess high magnetic polarization (J s > 1 T). The structures with desirable magnetic properties center on (Fe,Co) 2⁢ P. This supports conventional wisdom, which focuses on the mixture of the two known end compounds: Fe 2 ⁢P and Co 2 ⁢P. Our work provides guidance for synthesis. We find Fe 7 ⁢CoP 4 shows the most promise (J s = 1.03T and K 1 = 0.83MJ/m 3 ).

36 MATERIALS SCIENCE↗

Earthquake Ground Motion Simulations on the Sierra High-Performance Computing System

Physics-based three-dimensional (3D) numerical simulations of earthquake ground motions provide an alternative to reliance on limited and highly scattered empirical data and models for seismic hazard assessment. We performed fully 3D earthquake ground motion simulations on the Sierra and Lassen high-performance computing (HPC) platforms at Livermore Computing. These new HPC platforms feature graphics processing unit (GPU) acceleration for improved performance. Ground motions were computed with SW4, a summation-by-parts finite difference code for anelastic wave propagation. SW4 was recently enhanced to run efficiently on GPU-accelerated platforms. We found that SW4 runs almost fifty (50) times faster in terms of node-hours on new GPU-accelerated platforms compared to conventional central processing unit (CPU) platforms with agreement of computed ground motion time-series to the full single precision which they are written. We focused on moment magnitude, M W , 7.0 earthquakes on the Hayward Fault. We were able to run three simulations resolving ground motions from static displacements (0 Hz) to frequencies of 10 Hz (wavelengths as short as 50 meters) making these Sierra calculations the highest resolution regional-scale earthquake simulations ever made for California. Results show good agreement between simulated ground motion intensities (peak ground velocity and acceleration, spectral response) and empirical models, including frequencies above 10 Hz. On Lassen we were able to run a suite of 52 M W 7.0 Hayward Fault earthquake simulations resolved 0-5 Hz for both plane-layered, one-dimensional (1D) and 3D Earth models. These simulations were used to develop a methodology for characterizing and removing systematic path and site effects which bias ground motion intensities. These simulations using the newly enhanced SW4 code allow us to run higher resolution seismic simulations with shorter run times, providing a new capability for seismic hazard and risk studies. Results were presented at scientific conferences and workshops and two publications are being prepared based on work performed in this project.

58 GEOSCIENCES↗

Muon Neutrino Reconstruction at ICARUS with Machine Learning

The ICARUS T600 LArTPC detector successfully ran for three years at the underground LNGS laboratories, providing a first sensitive search for LSND-like anomalous electron neutrino appearance in the CNGS beam. After a significant overhauling at CERN, the T600 detector has been placed in its experimental hall at Fermilab, fully commissioned, and the first events observed with full detector readout. Regular data-taking began in May 2021 with neutrinos from the Booster Neutrino Beam (BNB) and neutrinos six degrees off-axis from the Neutrinos at the Main Injector (NuMI). Modern developments in machine learning have allowed for the development of an end-to-end machine learning-based event reconstruction for ICARUS data. This reconstruction folds in 3D voxel-level feature extraction using sparse convolutional neural networks and particle clustering using graph neural networks to produce outputs suitable for physics analyses. This poster will summarize the performance of a high-purity and high-efficiency end-to-end machine learning-based selection of muon neutrinos from the BNB and highlight studies of electromagnetic shower reconstruction from a neutral pion selection.

43 PARTICLE ACCELERATORS↗

Simple numerical method for predicting steady compressible flows

A numerical method for solving the isenthalpic form of the governing equations for compressible viscous and inviscid flows was developed. The method was based on the concept of flux vector splitting in its implicit form. The method was tested on several demanding inviscid and viscous configurations. Two different forms of the implicit operator were investigated. The time marching to steady state was accelerated by the implementation of the multigrid procedure. Its various forms very effectively increased the rate of convergence of the present scheme. High quality steady state results were obtained in most of the test cases; these required only short computational times due to the relative efficiency of the basic method.

Vonlavante, Ernst↗

Thermophysical and Thermomechanical Properties of Thermal Barrier Coating Systems

Thermal barrier coatings have been developed for advanced gas turbine and diesel engine applications to improve engine reliability and fuel efficiency. However, the issue of coating durability under high temperature cyclic conditions is still of major concern. The coating failure is closely related to thermal stresses and oxidation in the coating systems. Coating shrinkage cracking resulting from ceramic sintering and creep at high temperatures can further accelerate the coating failure process. The purpose of this paper is to address critical issues such as ceramic sintering and creep, thermal fatigue and their relevance to coating life prediction. Novel test approaches have been established to obtain critical thermophysical and thermomechanical properties of the coating systems under near-realistic temperature and stress gradients encountered in advanced engine systems. Emphasis is placed on the dynamic changes of the coating thermal conductivity and elastic modulus, fatigue and creep interactions, and resulting failure mechanisms during the simulated engine tests. Detailed experimental and modeling results describing processes occurring in the thermal barrier coating systems provide a framework for developing strategies to manage ceramic coating architecture, microstructure and properties.

Zhu, Dongming↗

Tradespace Exploration of Distributed Propulsors for Advanced On-Demand Mobility Concepts

Combustion-based sources of shaft power tend to significantly penalize distributed propulsion concepts, but electric motors represent an opportunity to advance the use of integrated distributed propulsion on an aircraft. This enables use of propellers in nontraditional, non-thrust-centric applications, including wing lift augmentation, through propeller slipstream acceleration from distributed leading edge propellers, as well as wingtip cruise propulsors. Developing propellers for these applications challenges long-held constraints within propeller design, such as the notion of optimizing for maximum propulsive efficiency, or the use of constant-speed propellers for high-performance aircraft. This paper explores the design space of fixed-pitch propellers for use as (1) lift augmentation when distributed about a wing's leading edge, and (2) as fixed-pitch cruise propellers with significant thrust at reduced tip speeds for takeoff. A methodology is developed for evaluating the high-level trades for these types of propellers and is applied to the exploration of a NASA Distributed Electric Propulsion concept. The results show that the leading edge propellers have very high solidity and pitch well outside of the empirical database, and that the cruise propellers can be operated over a wide RPM range to ensure that thrust can still be produced at takeoff without the need for a pitch change mechanism. To minimize noise exposure to observers on the ground, both the leading edge and cruise propellers are designed for low tip-speed operation during takeoff, climb, and approach.

Borer, Nicholas K.↗

Operational Results of Pulse Shaping Techniques for the High Voltage Convertor Modulator

The High Voltage Converter Modulators (HVCMs) are used to power the RF klystrons used throughout the accelerator systems at Spallation Neutron Source (SNS). The output voltage of the HVCM has significant droop and ripple which, combined with low level RF (LLRF) system limitations, affect performance and efficiency of accelerator cavities. In conjunction with the progress in development of the new HVCM controller, different pulse modulation techniques were implemented and studied on the test modulator. This paper discusses the results of implementation of frequency modulation and phase modulation on output voltage. Operational data, including full average power operation, of test modulator is also discussed. Future plans for the new modulation scheme will be presented.

Patel, Gunjan P↗

CO 2 storage site characterization using ensemble-based approaches with deep generative models

Estimating spatially distributed properties such as permeability from available sparse measurements is a great challenge in efficient subsurface CO 2 storage operations. In this paper, a deep generative model that can accurately capture complex subsurface structure is tested with an ensemble-based inversion method for accurate and accelerated characterization of CO 2 storage sites. We chose Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) for its realistic reservoir property representation and Ensemble Smoother with Multiple Data Assimilation (ES-MDA) for its robust data fitting and uncertainty quantification capability. WGAN-GP are trained to generate high-dimensional permeability fields from a low-dimensional latent space and ES-MDA then updates the latent variables by assimilating available measurements. Several subsurface site characterization examples including Gaussian, channelized, and fractured reservoirs are used to evaluate the accuracy and computational efficiency of the proposed method and the main features of the unknown permeability fields are characterized accurately with reliable uncertainty quantification. Furthermore, the estimation performance is compared with a widely-used variational, i.e., optimization-based, inversion approach, and the proposed approach outperforms the variational inversion method in several benchmark cases. We explain such superior performance by visualizing the objective function in the latent space: because of nonlinear and aggressive dimension reduction via generative modeling, the objective function surface becomes extremely complex while the ensemble approximation can smooth out the multi-modal surface during the minimization. This suggests that the ensemble-based approach works well over the variational approach when combined with deep generative models at the cost of forward model runs unless convergence-ensuring modifications are implemented in the variational inversion.

42 ENGINEERING↗